@ponythewhite/base-context 1.0.0 → 1.0.2

package/dist/modes/daemon/daemon-supervisor.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"daemon-supervisor.d.ts","sourceRoot":"","sources":["../../../src/modes/daemon/daemon-supervisor.ts"],"names":[],"mappings":"AAiCA,OAAO,EACN,KAAK,yBAAyB,EAI9B,MAAM,oCAAoC,CAAC;AAiB5C,OAAO,EAEN,KAAK,mBAAmB,EAExB,MAAM,oCAAoC,CAAC;AA+U5C,UAAU,uBAAuB;IAChC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oBAAoB,EAAE,yBAAyB,CAAC;IAChD,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAiQD,wBAAgB,2BAA2B,CAAC,mBAAmB,EAAE,mBAAmB,GAAG,MAAM,CAM5F;AA8BD,wBAAsB,uBAAuB,CAAC,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,KAAK,CAAC,CAK9F;AAED,qBAAa,gBAAgB;IAyD3B,OAAO,CAAC,QAAQ,CAAC,UAAU;IAxD5B,OAAO,CAAC,MAAM,CAAC,CAAS;IACxB,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAgB;IACtC,OAAO,CAAC,SAAS,CAAwB;IACzC,OAAO,CAAC,WAAW,CAAoC;IACvD,OAAO,CAAC,cAAc,CAAS;IAC/B,OAAO,CAAC,cAAc,CAAC,CAAuB;IAC9C,OAAO,CAAC,WAAW,CAAC,CAAwB;IAC5C,OAAO,CAAC,qBAAqB,CAAC,CAAQ;IACtC,OAAO,CAAC,SAAS,CAAC,CAA+D;IACjF,OAAO,CAAC,cAAc,CAAC,CAAgB;IACvC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,kBAAkB,CAAC,CAAsC;IACjE,OAAO,CAAC,QAAQ,CAAC,aAAa,CAA4B;IAC1D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,OAAO,CAAC,QAAQ,CAAC,aAAa,CAA6C;IAC3E,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAA6C;IACrF,OAAO,CAAC,QAAQ,CAAC,2BAA2B,CAAkD;IAC9F,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAA6C;IAC/E,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAqC;IAC7D,OAAO,CAAC,gBAAgB,CAAC,CAA8B;IACvD,OAAO,CAAC,QAAQ,CAAC,cAAc,CAA8C;IAC7E,0EAA0E;IAC1E,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAyE;IAC1G,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAkD;IACrF,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAAyB;IAC/D,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAS;IACvC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAgB;IAC3C,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAS;IAC9C,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAA4B;IACjE,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAS;IAC3C,OAAO,CAAC,cAAc,CAA0B;IAChD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAA6C;IACjF,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAqB;IAC9D,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IACzD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAsB;IAC9C,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAkB;IAClD,OAAO,CAAC,WAAW,CAAC,CAAc;IAClC,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;IAC1D,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;IAC1D,gGAAgG;IAChG,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAqB;IACxD,OAAO,CAAC,mBAAmB,CAAS;IACpC,OAAO,CAAC,mBAAmB,CAAC,CAAiC;IAC7D,OAAO,CAAC,sBAAsB,CAAC,CAAiB;IAChD,OAAO,CAAC,eAAe,CAAC,CAAkB;IAC1C,OAAO,CAAC,eAAe,CAAC,CAAgB;IACxC,OAAO,CAAC,iBAAiB,CAAC,CAAgC;IAC1D,OAAO,CAAC,iBAAiB,CAAC,CAAgB;IAC1C,OAAO,CAAC,iBAAiB,CAAC,CAAgB;IAC1C,OAAO,CAAC,kBAAkB,CAAC,CAAgC;IAC3D,OAAO,CAAC,sBAAsB,CAAC,CAAgB;IAC/C,OAAO,CAAC,4BAA4B,CAAS;IAC7C,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAA6B;IAEnE,YACkB,UAAU,EAAE,MAAM,EACnC,OAAO,EAAE,uBAAuB,EAoBhC;IAEK,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAiG3B;IAED,OAAO,CAAC,MAAM;IAgBd,OAAO,CAAC,GAAG;IAMX,OAAO,CAAC,sBAAsB;IAM9B,OAAO,CAAC,wBAAwB;IAMhC,OAAO,CAAC,uBAAuB;IAO/B,OAAO,CAAC,qCAAqC;YAkB/B,2BAA2B;YAoE3B,6BAA6B;YA8B7B,wBAAwB;IA8BtC,OAAO,CAAC,yBAAyB;IAgBjC,OAAO,CAAC,sBAAsB;IAwB9B,4HAA4H;IAC5H,OAAO,CAAC,0BAA0B;YAWpB,oBAAoB;YAgFpB,wBAAwB;YAcxB,iBAAiB;YAUjB,8BAA8B;YAW9B,6BAA6B;IAiC3C,OAAO,CAAC,8BAA8B;YAmBxB,sBAAsB;YAmBtB,6BAA6B;YAM7B,qBAAqB;IAOnC,OAAO,CAAC,6BAA6B;IAkBrC,OAAO,CAAC,qBAAqB;IAmC7B,4IAA4I;IAC5I,OAAO,CAAC,6BAA6B;IAyBrC,OAAO,CAAC,6BAA6B;IAqBrC,OAAO,CAAC,uBAAuB;IAY/B,OAAO,CAAC,6BAA6B;IAMrC,OAAO,CAAC,aAAa;IASrB,OAAO,CAAC,sBAAsB;IAY9B,OAAO,CAAC,gBAAgB;IAgFxB,OAAO,CAAC,wBAAwB;IAUhC,OAAO,CAAC,gBAAgB;YAIV,+BAA+B;IAsD7C,OAAO,CAAC,kCAAkC;IAY1C,OAAO,CAAC,mCAAmC;IAW3C,OAAO,CAAC,0BAA0B;IA2BlC,OAAO,CAAC,kBAAkB;IAI1B,OAAO,CAAC,mBAAmB;IAS3B,OAAO,CAAC,kBAAkB;IAQ1B,OAAO,CAAC,qBAAqB;IAQ7B,OAAO,CAAC,sCAAsC;IAgC9C,0FAA0F;IAC1F,OAAO,CAAC,sCAAsC;IAoD9C,OAAO,CAAC,sBAAsB;YAShB,UAAU;YAwLV,aAAa;YA8yBb,UAAU;YAqGV,iBAAiB;YASjB,kBAAkB;IAShC,OAAO,CAAC,qBAAqB;YAsBf,sBAAsB;YAqBtB,mBAAmB;YA8EnB,oBAAoB;YAmCpB,iBAAiB;IAU/B,OAAO,CAAC,uBAAuB;IAU/B,OAAO,CAAC,sBAAsB;IAS9B,OAAO,CAAC,mBAAmB;YAab,8BAA8B;YAmD9B,kBAAkB;IA4BhC,OAAO,CAAC,0BAA0B;YASpB,YAAY;YAyQZ,aAAa;IAuD3B,OAAO,CAAC,sBAAsB;YAiBhB,eAAe;YAoBf,oBAAoB;YAuFpB,sBAAsB;YAqCtB,iBAAiB;IAuD/B,OAAO,CAAC,wBAAwB;IAIhC,OAAO,CAAC,yBAAyB;IAUjC,OAAO,CAAC,mBAAmB;YAsBb,4BAA4B;IA+B1C,OAAO,CAAC,uBAAuB;IA+C/B,OAAO,CAAC,yBAAyB;IAyBjC,OAAO,CAAC,sBAAsB;IAa9B,OAAO,CAAC,kBAAkB;IAQ1B,OAAO,CAAC,yBAAyB;IA6BjC,OAAO,CAAC,wBAAwB;IAiBhC,OAAO,CAAC,sBAAsB;IAkB9B,OAAO,CAAC,iCAAiC;IAWzC,OAAO,CAAC,iCAAiC;YAa3B,aAAa;IAuI3B,OAAO,CAAC,yBAAyB;IASjC,OAAO,CAAC,wBAAwB;YAQlB,gCAAgC;YAoFhC,sBAAsB;YAyDtB,oBAAoB;YAiBpB,0BAA0B;YAgB1B,oCAAoC;IAalD,OAAO,CAAC,MAAM;IAKd,OAAO,CAAC,gBAAgB;IAWxB,OAAO,CAAC,kBAAkB;IAS1B,OAAO,CAAC,kBAAkB;IAiC1B,OAAO,CAAC,sBAAsB;IAM9B,OAAO,CAAC,6BAA6B;IAKrC,OAAO,CAAC,gBAAgB;IAexB,OAAO,CAAC,+BAA+B;IAMvC,OAAO,CAAC,mBAAmB;YAIb,gBAAgB;IAsB9B,OAAO,CAAC,6BAA6B;IAuBrC,OAAO,CAAC,wBAAwB;IAsBhC,OAAO,CAAC,sBAAsB;IAsB9B,OAAO,CAAC,0BAA0B;IAQlC,OAAO,CAAC,wBAAwB;IAYhC,OAAO,CAAC,sBAAsB;YAahB,yBAAyB;IAsEvC,OAAO,CAAC,iCAAiC;IAoBzC,OAAO,CAAC,6BAA6B;IASrC,OAAO,CAAC,uBAAuB;IAO/B,OAAO,CAAC,+BAA+B;IAkBvC,OAAO,CAAC,oBAAoB;IAkB5B,OAAO,CAAC,oBAAoB;IAQ5B,OAAO,CAAC,eAAe;IAOvB,OAAO,CAAC,yBAAyB;IAajC,OAAO,CAAC,oBAAoB;IAI5B,OAAO,CAAC,oBAAoB;IAa5B,sFAAsF;IACtF,OAAO,CAAC,cAAc;IAmBtB;;;;OAIG;IACH,OAAO,CAAC,iBAAiB;YAIX,gCAAgC;IAiB9C,OAAO,CAAC,2BAA2B;YAarB,yCAAyC;IAcvD,OAAO,CAAC,+BAA+B;IAyBvC,OAAO,CAAC,eAAe;IAIvB,gFAAgF;IAChF,OAAO,CAAC,gBAAgB;IAIxB,gEAAgE;IAChE,OAAO,CAAC,YAAY;IAIpB;;;OAGG;IACH,OAAO,CAAC,oBAAoB;IAU5B,OAAO,CAAC,4BAA4B;YAWtB,kBAAkB;IAkEhC,OAAO,CAAC,kBAAkB;IAe1B,OAAO,CAAC,gBAAgB;IAuBxB,OAAO,CAAC,mBAAmB;IAO3B,OAAO,CAAC,aAAa;YAUP,UAAU;IAsBxB,OAAO,CAAC,mBAAmB;IAI3B,OAAO,CAAC,0BAA0B;IAOlC,OAAO,CAAC,8BAA8B;IAMtC,OAAO,CAAC,YAAY;IAwBpB,OAAO,CAAC,mBAAmB;IAyB3B,OAAO,CAAC,uBAAuB;YA0BjB,eAAe;YAiBf,YAAY;IAiM1B,OAAO,CAAC,4BAA4B;IAQpC,OAAO,CAAC,mBAAmB;IA6D3B,OAAO,CAAC,0BAA0B;IAyClC,OAAO,CAAC,0BAA0B;YAgBpB,cAAc;IAiF5B,OAAO,CAAC,qBAAqB;IAkC7B,OAAO,CAAC,mBAAmB;YAIb,mBAAmB;IA2BjC,OAAO,CAAC,YAAY;YAgBN,qBAAqB;IAUnC,OAAO,CAAC,iBAAiB;IAmbzB,OAAO,CAAC,wBAAwB;IAahC,OAAO,CAAC,sBAAsB;IAiB9B,OAAO,CAAC,YAAY;IAepB,OAAO,CAAC,aAAa;YAgBP,uBAAuB;YAWvB,mBAAmB;YAsFnB,oBAAoB;YAmBpB,0BAA0B;IAsIxC,OAAO,CAAC,gCAAgC;IA0CxC;;;;;;OAMG;IACH,OAAO,CAAC,eAAe;IAiBvB;;;;OAIG;IACH,OAAO,CAAC,0BAA0B;YAcpB,UAAU;YAiBV,mBAAmB;IAoKjC;;;;OAIG;IACH,OAAO,CAAC,8BAA8B;YASxB,0BAA0B;YAqF1B,0BAA0B;YAgB1B,iCAAiC;YAgDjC,kCAAkC;IAyBhD,OAAO,CAAC,4BAA4B;IAQpC,OAAO,CAAC,4BAA4B;IAapC,OAAO,CAAC,0BAA0B;IAQlC,OAAO,CAAC,KAAK;IAIb,OAAO,CAAC,0BAA0B;IAOlC,OAAO,CAAC,eAAe;IAWvB,OAAO,CAAC,sBAAsB;IAe9B,OAAO,CAAC,qBAAqB;IAK7B,OAAO,CAAC,uBAAuB;IAS/B,OAAO,CAAC,qBAAqB;IAY7B,OAAO,CAAC,4BAA4B;IAkBpC,OAAO,CAAC,aAAa;YAUP,0BAA0B;YAQ1B,8BAA8B;YA4E9B,cAAc;IAQ5B,OAAO,CAAC,oBAAoB;YASd,QAAQ;CA4FtB","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport { createHash, randomBytes, randomUUID } from \"node:crypto\";\nimport {\n\tchmodSync,\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trenameSync,\n\trmSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { createServer, type Server, type Socket } from \"node:net\";\nimport { basename, dirname, join, resolve } from \"node:path\";\nimport { Writable } from \"node:stream\";\nimport { getLogger } from \"@ponythewhite/base-context-ai\";\nimport { createCliSubprocessEnv, createCliSubprocessLaunchSpec } from \"../../cli/subprocess-launch.js\";\nimport {\n\tappendRotatingLog,\n\tgetCronJobsPath,\n\tgetDaemonLogPath,\n\tgetDaemonUpdateRestartManifestPath,\n\tgetSessionsDir,\n\tVERSION,\n} from \"../../config.js\";\nimport {\n\ttype AgentFamilyCatalogEntry,\n\ttype AgentSessionMessageAgentSummary,\n\tassertAgentFamilyReach,\n\tassertAgentSessionNameAvailable,\n\tformatAgentSessionNameUnavailable,\n\tsessionNameReservationKey,\n} from \"../../core/agent-messages.js\";\nimport {\n\ttype AgentSessionRuntimeConfig,\n\ttype DurableAgentSessionRuntimeConfig,\n\tdurableAgentSessionRuntimeConfig,\n\tmergeAgentSessionRuntimeConfig,\n} from \"../../core/agent-session-config.js\";\nimport {\n\ttype AgentCronJob,\n\tAgentCronJobStore,\n\tisHeartbeatCronJob,\n\tmigrateLegacyCronJobsToSessionArtifacts,\n\tSESSION_SCHEDULED_JOBS_FILENAME,\n} from \"../../core/cron-jobs.js\";\nimport {\n\tclearOrphanProcessJournal,\n\tkillOrphanProcess,\n\tORPHAN_PROCESS_JOURNAL_ENV,\n\treadActiveOrphanProcesses,\n\tshouldReapOrphanProcess,\n} from \"../../core/orphan-process-journal.js\";\nimport { PromptAdmissionCancelledError, waitForPromptAdmission } from \"../../core/prompt-admission.js\";\nimport { RlmJournalOwner } from \"../../core/rlm-journal-owner.js\";\nimport {\n\tcanEvictWorker,\n\ttype IdleEvictionMinutes,\n\ttype WorkerEvictionSnapshot,\n} from \"../../core/session-action-store.js\";\nimport { canonicalSessionPath, getProcessStartId, SessionAlreadyActiveError } from \"../../core/session-lease.js\";\nimport { getSessionArtifactPathForFile, readSessionInfo, type SessionInfo } from \"../../core/session-manager.js\";\nimport { looksLikeSessionPath } from \"../../core/session-resolver.js\";\nimport { SettingsManager } from \"../../core/settings-manager.js\";\nimport { PRODUCT } from \"../../product-identity.js\";\nimport { isProcessAlive, processIdExists, signalProcessGroupOrProcess } from \"../../utils/child-process.js\";\nimport type { AgentConnectionHeartbeat } from \"../agent-connection/types.js\";\nimport { attachJsonlLineReader, serializeJsonLine } from \"../rpc/jsonl.js\";\nimport type { PrivateFrame } from \"../session-worker/private-framing.js\";\nimport { createActiveSessionId, type DaemonSocketClient } from \"./active-session-state.js\";\nimport {\n\tAgentRoster,\n\ttype AgentRosterEntry,\n\ttype AgentRosterMutation,\n\tpassivatedWorkerRosterEntry,\n\trosterAgentIdForSummary,\n\tsessionSummaryFromRosterEntry,\n\ttype WorkerRosterEntry,\n\tworkerRosterEntryFromSummary,\n} from \"./agent-roster.js\";\nimport {\n\ttype CommandJournalBeginResult,\n\tCommandRecoveryJournal,\n\tcreateCommandIdempotencyKey,\n} from \"./command-recovery-journal.js\";\nimport { CompactAssistantStreamReconstructor, isCompactAssistantDelta } from \"./compact-session-stream.js\";\nimport { DAEMON_CATALOG_ROLE_ENV, DaemonCatalogClient } from \"./daemon-catalog-process.js\";\nimport { deserializeDaemonError, serializeDaemonError } from \"./daemon-errors.js\";\nimport {\n\tCANONICAL_SESSION_OWNERSHIP_COMPATIBILITY,\n\tcollectDaemonClientEnv,\n\tcreateDaemonEventMeta,\n\tDAEMON_COMMAND_COMPATIBILITY,\n\tDAEMON_COMMAND_ENVELOPE_MIN_PROTOCOL_VERSION,\n\tDAEMON_DEFAULT_CLIENT_CAPABILITIES,\n\tDAEMON_DEFAULT_SERVER_CAPABILITIES,\n\tDAEMON_PROTOCOL_INFO,\n\tDAEMON_SCHEMA_ID,\n\tDAEMON_SCHEMA_REVISION,\n\tDAEMON_UPDATE_RESTART_FORMAT_VERSION,\n\ttype DaemonAttachResult,\n\ttype DaemonClientCapability,\n\ttype DaemonClosingReason,\n\ttype DaemonCommand,\n\ttype DaemonOutbound,\n\ttype DaemonPeerTransportTicket,\n\ttype DaemonResponse,\n\ttype DaemonServerCapability,\n\ttype DaemonUpdateRestartManifest,\n\tfailure,\n\tisDaemonCommandEnvelope,\n\tisDaemonMutatingCommand,\n\tisLegacyDaemonInspection,\n\tsalvageDaemonCommandId,\n\tsuccess,\n\tUPDATE_RESTART_DRAIN_COMMANDS,\n} from \"./daemon-protocol.js\";\nimport { getDaemonRuntimeIdentity } from \"./daemon-runtime-identity.js\";\nimport { matchesSessionIdSuffix } from \"./daemon-session-id.js\";\nimport {\n\tclassifySessionRosterStatus,\n\tisEvictableEmptySessionSummary,\n\tisSessionSummaryBusy,\n\ttype SessionSummary,\n\tsummaryForInactiveSession,\n} from \"./daemon-session-list.js\";\nimport {\n\tacquireDaemonSocketPathLease,\n\tcleanupDaemonSocketPath,\n\ttype DaemonSocketIdentity,\n\ttype DaemonSocketPathLease,\n\tdefaultDaemonSocketDir,\n\tdefaultDaemonSocketPath,\n\tgetDaemonSocketIdentity,\n\tnormalizeSocketPath,\n\tprepareDaemonSocketPath,\n\trestrictDaemonSocketPath,\n} from \"./daemon-socket.js\";\nimport {\n\tacquireDaemonSupervisorOwnership,\n\tisDaemonShutdownAdmissionActive,\n\twaitForDaemonStartupFence,\n} from \"./daemon-supervisor-ownership.js\";\nimport {\n\tDaemonWorkerAuthenticationError,\n\tDaemonWorkerClient,\n\tDaemonWorkerCompatibilityError,\n\tDaemonWorkerProbeTimeoutError,\n} from \"./daemon-worker-client.js\";\nimport {\n\tDAEMON_WORKER_ACTIVE_SESSION_ID_ENV,\n\tDAEMON_WORKER_INSTANCE_ID_ENV,\n\tDAEMON_WORKER_PEER_TRANSPORT_CAPABILITY,\n\tDAEMON_WORKER_RECOVERY_JOURNAL_ENV,\n\tDAEMON_WORKER_ROLE_ENV,\n\tDAEMON_WORKER_ROSTER_CAPABILITY,\n\tDAEMON_WORKER_STARTUP_GATE_COMMIT,\n\tDAEMON_WORKER_STARTUP_GATE_FD_ENV,\n\tDAEMON_WORKER_SUPERVISOR_SOCKET_ENV,\n\tDAEMON_WORKER_TOKEN_ENV,\n\ttype DaemonCreateCommand,\n\ttype DaemonWorkerDescriptor,\n\ttype DaemonWorkerFrameHeader,\n\ttype DaemonWorkerLifecycle,\n\ttype DaemonWorkerRosterOutbound,\n\tdurableDaemonCreateCommand,\n\tdurableDaemonWorkerDescriptor,\n\tROSTER_HEARTBEAT_INTERVAL_MS,\n\tSESSION_LEASE_OWNER_ID_ENV,\n\tSESSION_LEASES_ENABLED_ENV,\n} from \"./daemon-worker-protocol.js\";\nimport { MutationDrainLatch } from \"./mutation-drain-latch.js\";\nimport {\n\tcreateRlmLedgerRegistrySeedSource,\n\ttype RlmLedgerEdge,\n\tRlmSpawnLedger,\n\trlmLedgerPath,\n\ttombstoneSavedSessionDelete,\n} from \"./rlm-ledger.js\";\nimport { captureSavedSessionPageQuery } from \"./saved-session-page.js\";\nimport { SNAPSHOT_TARGET_CHUNK_BYTES, SnapshotTranscriptCache } from \"./snapshot-transcript-cache.js\";\nimport { WorkerRecoveryJournal } from \"./worker-recovery-journal.js\";\n\ntype DistributiveOmit<T, K extends keyof T> = T extends unknown ? Omit<T, K> : never;\ntype DaemonCommandBody = DistributiveOmit<DaemonCommand, \"id\">;\n\nconst structuredLog = getLogger(\"coding-agent.daemon-supervisor\");\nconst WORKER_CONNECT_TIMEOUT_MS = 30_000;\nconst ROSTER_WATCHDOG_INTERVAL_MS = 15_000;\nconst ROSTER_STALE_AFTER_MS = 3 * ROSTER_HEARTBEAT_INTERVAL_MS;\nconst SUPERVISOR_SERVER_CAPABILITIES: readonly DaemonServerCapability[] = [\n\t...DAEMON_DEFAULT_SERVER_CAPABILITIES,\n\t\"agent_roster\",\n\t\"direct_peer_transport\",\n\t\"rlm_ledger_mutation\",\n];\nconst PEER_TRANSPORT_GRANT_TTL_MS = 10_000;\nconst WORKER_REQUEST_TIMEOUT_MS = 24 * 60 * 60 * 1000;\nconst INPUT_PAUSE_CLEANUP_TIMEOUT_MS = 5_000;\nconst UPDATE_RESTART_MUTATION_DRAIN_TIMEOUT_MS = 80_000;\nconst UPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS = 90_000;\n// The whole pre-commit prepare (drain + worker fencing) must finish inside the\n// caller's 120s prepare_update_restart request timeout, or roll back; otherwise\n// an abandoned prepare leaves the daemon permanently fenced with workers stopped.\nconst UPDATE_RESTART_PREPARE_DEADLINE_MS = 100_000;\nconst WORKER_RETRY_DELAYS_MS = [250, 1000, 5000] as const;\nconst DEFERRED_RECOVERY_RECHECK_MS = 5000;\n// ~2.5 minutes of probing: each round is one 5s defer recheck plus a ~11s three-delay probe pass.\nconst MAX_DEFERRED_RECOVERY_ROUNDS = 10;\nconst STOP_FINALIZATION_RECHECK_MS = 250;\nconst STOP_FINALIZATION_SIGKILL_GRACE_MS = 5000;\nconst STOP_FINALIZATION_RETRY_MS = 5000;\nconst STALE_RECLAIM_WAIT_MS = 10_000;\n// Polling loops probe existence cheaply via kill(0); the ps-backed zombie and\n// identity checks are throttled so a wedged worker cannot saturate the\n// supervisor event loop with synchronous subprocess spawns.\nconst LIVENESS_IDENTITY_RECHECK_MS = 500;\nconst OWNED_WORKER_DISCONNECT_GRACE_MS = 30_000;\nconst IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS = 5 * 60_000;\nconst IDLE_EVICTION_MIN_SWEEP_INTERVAL_MS = 60_000;\nconst IDLE_EVICTION_DRAIN_TIMEOUT_MS = 5_000;\nconst CHILD_PASSIVATION_PER_WORKER_CAP = 2;\nconst SCHEDULED_WAKE_RETRY_MS = 60_000;\nconst SCHEDULED_WAKE_MAX_TIMEOUT_MS = 2_147_483_647;\nconst SCHEDULED_WAKE_CLIENT_ID = \"scheduled-wake\";\nconst SUPERVISOR_CONFIG_FILE_NAME = \"supervisor-config\";\nconst WORKER_STARTUP_GATE_FD = 3;\n\nconst DAEMON_COMMAND_TYPES: ReadonlySet<string> = new Set([\n\t\"ack_result\",\n\t\"list\",\n\t\"list_agent_peers\",\n\t\"rlm_ledger_mutate\",\n\t\"get_direct_worker_transport\",\n\t\"roster_subscribe\",\n\t\"roster_unsubscribe\",\n\t\"list_saved_sessions\",\n\t\"create\",\n\t\"attach\",\n\t\"reattach\",\n\t\"detach\",\n\t\"complete_owned_session\",\n\t\"promote_owned_session\",\n\t\"kill\",\n\t\"rename\",\n\t\"prompt\",\n\t\"cancel_prompt_admission\",\n\t\"prompt_and_wait\",\n\t\"steer\",\n\t\"follow_up\",\n\t\"restore_next_turn\",\n\t\"restore_actions\",\n\t\"append_custom_message\",\n\t\"resume_queue\",\n\t\"send_message\",\n\t\"send_result\",\n\t\"agent_messages_status\",\n\t\"agent_messages_pause\",\n\t\"agent_messages_resume\",\n\t\"agent_messages_clear\",\n\t\"abort\",\n\t\"start_side_question\",\n\t\"abort_side_question\",\n\t\"execute_bash\",\n\t\"execute_bash_and_wait\",\n\t\"abort_bash\",\n\t\"cancel_rlm_child\",\n\t\"delete_rlm_subagent\",\n\t\"wait_for_idle\",\n\t\"wait_for_headless_completion\",\n\t\"get_session_header\",\n\t\"get_state\",\n\t\"get_connection_state\",\n\t\"get_messages\",\n\t\"get_rlm_children\",\n\t\"get_session_stats\",\n\t\"get_context_tree\",\n\t\"get_commands\",\n\t\"get_resource_snapshot\",\n\t\"replace_acp_mcp_servers\",\n\t\"get_model_catalog\",\n\t\"get_available_models\",\n\t\"get_queue\",\n\t\"mutate_queued_message\",\n\t\"clear_queue\",\n\t\"abort_and_clear_queue\",\n\t\"acquire_session_input_pause\",\n\t\"release_session_input_pause\",\n\t\"cron_list\",\n\t\"heartbeats_list\",\n\t\"heartbeat_manage\",\n\t\"cron_add\",\n\t\"cron_cancel\",\n\t\"cron_resume\",\n\t\"heartbeat_get\",\n\t\"heartbeat_set\",\n\t\"heartbeat_update\",\n\t\"set_model\",\n\t\"cycle_model\",\n\t\"set_scoped_models\",\n\t\"set_thinking_level\",\n\t\"cycle_thinking_level\",\n\t\"set_service_tier\",\n\t\"set_transport\",\n\t\"set_steering_mode\",\n\t\"set_follow_up_mode\",\n\t\"set_auto_compaction\",\n\t\"set_auto_retry\",\n\t\"compact\",\n\t\"refine\",\n\t\"abort_compaction\",\n\t\"abort_branch_summary\",\n\t\"abort_retry\",\n\t\"reload\",\n\t\"new_session\",\n\t\"switch_session\",\n\t\"fork\",\n\t\"navigate_tree\",\n\t\"import_jsonl\",\n\t\"export_html\",\n\t\"export_jsonl\",\n\t\"set_session_name\",\n\t\"get_rlm_max_depth_status\",\n\t\"set_rlm_max_depth\",\n\t\"rename_saved_session\",\n\t\"delete_saved_session\",\n\t\"get_session_context\",\n\t\"get_session_tree\",\n\t\"get_user_messages_for_forking\",\n\t\"get_last_assistant_text\",\n\t\"get_system_prompt\",\n\t\"get_tool_definition\",\n\t\"set_session_entry_label\",\n\t\"extension_ui_response\",\n\t\"prepare_update_restart\",\n\t\"retry_worker\",\n\t\"restart\",\n\t\"shutdown\",\n]);\n\ninterface ResidentWorker {\n\tdescriptor: DaemonWorkerDescriptor;\n\tdescriptorPath: string;\n\tclient?: DaemonWorkerClient;\n\tcompatibilityError?: DaemonWorkerCompatibilityError;\n\theartbeatSnapshot?: AgentConnectionHeartbeat[];\n\theartbeatSnapshotStale?: boolean;\n\tsummaries: Map<string, SessionSummary>;\n\tsnapshotCache: Map<string, DaemonAttachResult>;\n\ttranscriptCaches: Map<string, SnapshotTranscriptCache>;\n\tsnapshotGenerations: Map<string, Map<string, SnapshotTranscriptGeneration>>;\n\tsnapshotLoads: Map<string, Promise<DaemonAttachResult>>;\n\trecovery?: Promise<void>;\n\tdeferredRecovery?: Promise<void>;\n\tintentionalStop: boolean;\n\tstopRevision: number;\n\tlaunchEnv?: Record<string, string>;\n\ttransientCreateCommand?: DaemonCreateCommand;\n\tstopFinalization?: Promise<void>;\n\townerCleanupTimer?: ReturnType<typeof setTimeout>;\n\tpromotedOwnerClientId?: string;\n\tupdateRestartPrepareClient?: DaemonWorkerClient;\n\tlastFrameAt?: number;\n\trosterStale?: boolean;\n\t/** worker_auth advertised peer-transport support; absent on workers from older builds. */\n\tpeerTransportCapable?: boolean;\n\t/** In-flight replacement connection during authentication; an allowed frame source alongside client. */\n\tpendingClient?: DaemonWorkerClient;\n\t/** Consecutive defer->probe rounds against a live-but-silent worker; bounded by MAX_DEFERRED_RECOVERY_ROUNDS. */\n\tdeferredRecoveryRounds?: number;\n\t/** Bumped per applied roster frame; a summaries pull that straddles a frame must not gap-fill. */\n\trosterEpoch?: number;\n\trosterApplyChain?: Promise<void>;\n\trosterRepairPull?: Promise<void>;\n}\n\ninterface SnapshotDuplicateValidation {\n\tpromise: Promise<void>;\n\tresolve: () => void;\n\treject: (error: Error) => void;\n}\n\ninterface SnapshotTranscriptGeneration {\n\ttranscript: SnapshotTranscriptCache;\n\tresult: DaemonAttachResult;\n\tbegin?: Buffer;\n\tend?: Buffer;\n\tincoming: boolean;\n\tretired: boolean;\n\tduplicateChunkIndex?: number;\n\tduplicateResult?: DaemonAttachResult;\n\tvalidation?: SnapshotDuplicateValidation;\n}\n\ninterface DaemonSupervisorOptions {\n\tsocketPath?: string;\n\tdefaultSessionConfig: AgentSessionRuntimeConfig;\n\tdescriptorDir?: string;\n}\n\ninterface PersistedSupervisorConfig {\n\tversion: 1;\n\tsocketPath: string;\n\tdefaultSessionConfig: DurableAgentSessionRuntimeConfig;\n}\n\ninterface WorkerMatch {\n\tworker: ResidentWorker;\n\tsummary: SessionSummary;\n}\n\ninterface WorkerAttachData {\n\tresult: DaemonAttachResult;\n\tworker: ResidentWorker;\n\ttranscript?: SnapshotTranscriptCache;\n\treleaseTranscript?: () => void;\n}\n\ninterface SupervisorPromptAdmission {\n\tclient: DaemonSocketClient;\n\tactiveSessionId: string;\n\tpublicAdmissionId: string;\n\tworkerAdmissionId: string;\n\tstatus: \"waiting\" | \"owned\" | \"cancelled\";\n\tcontroller: AbortController;\n\tworker?: ResidentWorker;\n\tworkerActiveSessionId?: string;\n}\n\ninterface SupervisorSessionInputPause {\n\towner: DaemonSocketClient;\n\tworker: ResidentWorker;\n\tactiveSessionId: string;\n\trequestedActiveSessionId: string;\n\tleaseKey: string;\n\tpauseId: string;\n\treleaseTask?: Promise<DaemonResponse>;\n}\n\nfunction throwIfAdmissionCancelled(admission: SupervisorPromptAdmission | undefined): void {\n\tif (admission?.status === \"cancelled\") throw new PromptAdmissionCancelledError();\n}\n\nclass SupervisorRecoveryCancelledError extends Error {\n\treadonly code = \"supervisor_recovery_cancelled\" as const;\n}\n\nclass SnapshotLoadInvalidatedError extends Error {}\n\nclass WorkerStopTimeoutError extends Error {}\n\nfunction isSupervisorGenerationStale(error: unknown): boolean {\n\treturn (\n\t\ttypeof error === \"object\" &&\n\t\terror !== null &&\n\t\t\"code\" in error &&\n\t\t(error as { code?: unknown }).code === \"supervisor_generation_stale\"\n\t);\n}\n\nfunction isSupervisorRecoveryCancelled(error: unknown): boolean {\n\treturn isSupervisorShutdownAdmissionCancelled(error) || isSupervisorGenerationStale(error);\n}\n\n// Workers can be registered mid-tree (a resumed subagent transcript), so descent is membership at\n// any step of the parent walk, never a comparison against the ultimate root alone.\nfunction rosterFamilyDescendsFrom(\n\tedges: readonly RlmLedgerEdge[],\n): (path: string, roots: ReadonlySet<string>) => boolean {\n\tconst parentByChild = new Map(\n\t\tedges.map((edge) => [canonicalSessionPath(edge.child), canonicalSessionPath(edge.parent)]),\n\t);\n\treturn (path, roots) => {\n\t\tconst visited = new Set<string>();\n\t\tlet current = path;\n\t\twhile (!visited.has(current)) {\n\t\t\tif (roots.has(current)) return true;\n\t\t\tvisited.add(current);\n\t\t\tconst parent = parentByChild.get(current);\n\t\t\tif (parent === undefined) return false;\n\t\t\tcurrent = parent;\n\t\t}\n\t\treturn false;\n\t};\n}\n\nfunction isDaemonWorkerProbeTimeout(error: unknown): boolean {\n\treturn error instanceof DaemonWorkerProbeTimeoutError;\n}\n\nfunction isSupervisorShutdownAdmissionCancelled(error: unknown): boolean {\n\treturn (\n\t\terror instanceof SupervisorRecoveryCancelledError ||\n\t\t(typeof error === \"object\" &&\n\t\t\terror !== null &&\n\t\t\t\"code\" in error &&\n\t\t\t(error as { code?: unknown }).code === \"supervisor_recovery_cancelled\")\n\t);\n}\n\nfunction delay(ms: number): Promise<void> {\n\treturn new Promise((resolveDelay) => setTimeout(resolveDelay, ms));\n}\n\nfunction unrefDelay(ms: number): Promise<void> {\n\treturn new Promise((resolveDelay) => setTimeout(resolveDelay, ms).unref());\n}\n\nfunction commitWorkerStartupGate(gate: Writable): Promise<void> {\n\treturn new Promise((resolveCommit, rejectCommit) => {\n\t\tlet settled = false;\n\t\tconst finish = (error?: Error | null) => {\n\t\t\tif (settled) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tsettled = true;\n\t\t\tif (error) {\n\t\t\t\trejectCommit(error);\n\t\t\t} else {\n\t\t\t\tresolveCommit();\n\t\t\t}\n\t\t};\n\t\tconst onError = (error: Error) => finish(error);\n\t\tgate.on(\"error\", onError);\n\t\tgate.once(\"close\", () => gate.off(\"error\", onError));\n\t\tgate.end(DAEMON_WORKER_STARTUP_GATE_COMMIT, (error?: Error | null) => finish(error));\n\t});\n}\n\nfunction withoutCommandId(command: DaemonCommand): DaemonCommandBody {\n\tconst { id: _id, ...body } = command;\n\treturn body as DaemonCommandBody;\n}\n\nfunction withoutSupervisorCreateFields(command: DaemonCreateCommand): DaemonCreateCommand {\n\tconst { launchEnv: _launchEnv, lifecycle: _lifecycle, ...workerCommand } = command;\n\treturn workerCommand;\n}\n\nfunction responseWithId(response: DaemonResponse, id: string | undefined): DaemonResponse {\n\treturn { ...response, id };\n}\n\nfunction isSessionSummary(value: unknown): value is SessionSummary {\n\tif (!value || typeof value !== \"object\") {\n\t\treturn false;\n\t}\n\tconst candidate = value as { id?: unknown; sessionId?: unknown; cwd?: unknown };\n\treturn (\n\t\ttypeof candidate.id === \"string\" && typeof candidate.sessionId === \"string\" && typeof candidate.cwd === \"string\"\n\t);\n}\n\nfunction isDaemonWorkerDescriptor(value: unknown, socketPath: string): value is DaemonWorkerDescriptor {\n\tif (!value || typeof value !== \"object\") {\n\t\treturn false;\n\t}\n\tconst descriptor = value as Partial<DaemonWorkerDescriptor>;\n\treturn (\n\t\t(descriptor.version === 1 || descriptor.version === 2) &&\n\t\ttypeof descriptor.supervisorSocketPath === \"string\" &&\n\t\tnormalizeSocketPath(descriptor.supervisorSocketPath) === socketPath &&\n\t\ttypeof descriptor.workerId === \"string\" &&\n\t\tNumber.isInteger(descriptor.pid) &&\n\t\t(descriptor.pid ?? 0) > 0 &&\n\t\t(descriptor.processStartId === undefined || typeof descriptor.processStartId === \"string\") &&\n\t\t(descriptor.ownerClientId === undefined || typeof descriptor.ownerClientId === \"string\") &&\n\t\ttypeof descriptor.socketPath === \"string\" &&\n\t\ttypeof descriptor.authenticationToken === \"string\" &&\n\t\t(descriptor.workerInstanceId === undefined || typeof descriptor.workerInstanceId === \"string\") &&\n\t\ttypeof descriptor.rootActiveSessionId === \"string\" &&\n\t\ttypeof descriptor.createdAt === \"string\" &&\n\t\ttypeof descriptor.updatedAt === \"string\" &&\n\t\tNumber.isInteger(descriptor.consecutiveFailures) &&\n\t\tdescriptor.createCommand !== undefined &&\n\t\ttypeof descriptor.createCommand === \"object\" &&\n\t\tdescriptor.createCommand.type === \"create\"\n\t);\n}\n\nclass PreRosterWorkerError extends Error {}\n\nfunction workerAuthAdvertisesRoster(data: unknown): boolean {\n\tif (typeof data !== \"object\" || data === null) return false;\n\tconst capabilities = (data as { capabilities?: unknown }).capabilities;\n\treturn Array.isArray(capabilities) && capabilities.includes(DAEMON_WORKER_ROSTER_CAPABILITY);\n}\n\nfunction workerAuthAdvertisesPeerTransport(data: unknown): boolean {\n\tif (typeof data !== \"object\" || data === null) return false;\n\tconst capabilities = (data as { capabilities?: unknown }).capabilities;\n\treturn Array.isArray(capabilities) && capabilities.includes(DAEMON_WORKER_PEER_TRANSPORT_CAPABILITY);\n}\n\nfunction sessionSummariesFromResponse(response: DaemonResponse): SessionSummary[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\" || !(\"sessions\" in response.data)) {\n\t\tthrow new Error(\"Session worker returned an invalid list response\");\n\t}\n\tconst sessions = (response.data as { sessions: unknown }).sessions;\n\tif (!Array.isArray(sessions) || !sessions.every(isSessionSummary)) {\n\t\tthrow new Error(\"Session worker returned an invalid list response\");\n\t}\n\treturn sessions;\n}\n\nfunction attachResultFromResponse(response: DaemonResponse): DaemonAttachResult {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\tthrow new Error(response.success ? \"Session worker returned an invalid attach response\" : response.error);\n\t}\n\tconst candidate = response.data as Partial<DaemonAttachResult>;\n\tif (typeof candidate.activeSessionId !== \"string\" || !candidate.snapshot || !candidate.client) {\n\t\tthrow new Error(\"Session worker returned an invalid attach response\");\n\t}\n\treturn candidate as DaemonAttachResult;\n}\n\nfunction cronJobsFromResponse(response: DaemonResponse): AgentCronJob[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\treturn [];\n\t}\n\tconst jobs = (response.data as { jobs?: unknown }).jobs;\n\treturn Array.isArray(jobs) ? (jobs as AgentCronJob[]) : [];\n}\n\nfunction heartbeatsFromResponse(response: DaemonResponse): AgentConnectionHeartbeat[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\treturn [];\n\t}\n\tconst heartbeats = (response.data as { heartbeats?: unknown }).heartbeats;\n\treturn Array.isArray(heartbeats) ? (heartbeats as AgentConnectionHeartbeat[]) : [];\n}\n\nfunction sortCronJobs(jobs: AgentCronJob[]): AgentCronJob[] {\n\treturn jobs.sort((left, right) => {\n\t\tif (left.nextRunAt === right.nextRunAt) {\n\t\t\treturn 0;\n\t\t}\n\t\tif (left.nextRunAt === undefined) {\n\t\t\treturn 1;\n\t\t}\n\t\tif (right.nextRunAt === undefined) {\n\t\t\treturn -1;\n\t\t}\n\t\treturn Date.parse(left.nextRunAt) - Date.parse(right.nextRunAt);\n\t});\n}\n\nfunction descriptorKey(socketPath: string): string {\n\treturn createHash(\"sha256\").update(normalizeSocketPath(socketPath)).digest(\"hex\").slice(0, 12);\n}\n\nfunction defaultWorkerDescriptorDir(agentDir: string, socketPath: string): string {\n\treturn join(agentDir, \"daemon-workers\", descriptorKey(socketPath));\n}\n\nexport function idleEvictionSweepIntervalMs(idleEvictionMinutes: IdleEvictionMinutes): number {\n\tif (idleEvictionMinutes === \"off\") return IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS;\n\treturn Math.max(\n\t\tIDLE_EVICTION_MIN_SWEEP_INTERVAL_MS,\n\t\tMath.min(IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS, (idleEvictionMinutes * 60_000) / 3),\n\t);\n}\n\nfunction workerSocketPath(supervisorSocketPath: string, workerId: string): string {\n\tconst key = descriptorKey(supervisorSocketPath);\n\tif (process.platform === \"win32\") {\n\t\treturn `\\\\\\\\.\\\\pipe\\\\${PRODUCT.command}-worker-${key}-${workerId.slice(0, 12)}`;\n\t}\n\treturn join(defaultDaemonSocketDir(), `worker-${key}-${workerId.slice(0, 12)}.sock`);\n}\n\nfunction isFinalizedTranscriptEvent(eventType: string | undefined): boolean {\n\treturn (\n\t\teventType === \"message_end\" ||\n\t\teventType === \"turn_end\" ||\n\t\teventType === \"compaction_end\" ||\n\t\teventType === \"bash_end\"\n\t);\n}\n\nfunction normalizeCapabilities(\n\tcapabilities: readonly DaemonClientCapability[] | undefined,\n\tsupportsExtensionUi: boolean | undefined,\n): Set<DaemonClientCapability> {\n\tconst normalized = new Set(capabilities ?? DAEMON_DEFAULT_CLIENT_CAPABILITIES);\n\tif (supportsExtensionUi) {\n\t\tnormalized.add(\"extension_ui\");\n\t}\n\treturn normalized;\n}\n\nexport async function runDaemonSupervisorMode(options: DaemonSupervisorOptions): Promise<never> {\n\tconst socketPath = normalizeSocketPath(options.socketPath ?? defaultDaemonSocketPath());\n\tconst supervisor = new DaemonSupervisor(socketPath, options);\n\tawait supervisor.start();\n\treturn new Promise(() => {});\n}\n\nexport class DaemonSupervisor {\n\tprivate server?: Server;\n\tprivate readonly ready: Promise<void>;\n\tprivate markReady: () => void = () => {};\n\tprivate rejectReady: (error: Error) => void = () => {};\n\tprivate ownsSocketPath = false;\n\tprivate socketIdentity?: DaemonSocketIdentity;\n\tprivate socketLease?: DaemonSocketPathLease;\n\tprivate socketLeaseCompromise?: Error;\n\tprivate ownership?: Awaited<ReturnType<typeof acquireDaemonSupervisorOwnership>>;\n\tprivate cleanupPromise?: Promise<void>;\n\tprivate shuttingDown = false;\n\tprivate startupComplete = false;\n\tprivate updateRestartPhase?: \"draining\" | \"fencing\" | \"prepared\";\n\tprivate readonly mutationDrain = new MutationDrainLatch();\n\tprivate readonly clients = new Set<DaemonSocketClient>();\n\tprivate readonly connectionIds = new WeakMap<DaemonSocketClient, string>();\n\tprivate readonly sessionInputPauseEpochs = new WeakMap<DaemonSocketClient, number>();\n\tprivate readonly detachingInputPauseSessions = new WeakMap<DaemonSocketClient, Set<string>>();\n\tprivate readonly protocolClientIds = new WeakMap<DaemonSocketClient, string>();\n\tprivate readonly workers = new Map<string, ResidentWorker>();\n\tprivate workerStopCounts?: Map<ResidentWorker, number>;\n\tprivate readonly openingWorkers = new Map<string, Promise<ResidentWorker>>();\n\t/** Public admission ids are scoped to the socket that registered them. */\n\tprivate readonly promptAdmissions = new Map<DaemonSocketClient, Map<string, SupervisorPromptAdmission>>();\n\tprivate readonly sessionInputPauses = new Map<string, SupervisorSessionInputPause>();\n\tprivate readonly signalCleanupHandlers: Array<() => void> = [];\n\tprivate readonly descriptorDir: string;\n\tprivate readonly generation = randomUUID();\n\tprivate readonly supervisorConfigPath: string;\n\tprivate readonly defaultSessionConfig: AgentSessionRuntimeConfig;\n\tprivate readonly snapshotCacheRoot: string;\n\tprivate commandJournal!: CommandRecoveryJournal;\n\tprivate readonly streamReconstructor = new CompactAssistantStreamReconstructor();\n\tprivate readonly compactCatchupInProgress = new Set<string>();\n\tprivate readonly pendingSessionNames = new Set<string>();\n\tprivate readonly catalog: DaemonCatalogClient;\n\tprivate readonly settingsManager: SettingsManager;\n\tprivate rosterStore?: AgentRoster;\n\tprivate readonly pendingRosterChanged = new Set<string>();\n\tprivate readonly pendingRosterRemoved = new Set<string>();\n\t/** Ids declared to subscribers: gates removals to once and keeps owned-only row ids private. */\n\tprivate readonly publishedRosterIds = new Set<string>();\n\tprivate rosterPushScheduled = false;\n\tprivate rosterWatchdogTimer?: ReturnType<typeof setInterval>;\n\tprivate rlmSpawnLedgerInstance?: RlmSpawnLedger;\n\tprivate rlmJournalOwner?: RlmJournalOwner;\n\tprivate rlmJournalClose?: Promise<void>;\n\tprivate idleEvictionTimer?: ReturnType<typeof setTimeout>;\n\tprivate idleEvictionSweep?: Promise<void>;\n\tprivate idleEvictionFence?: Promise<void>;\n\tprivate scheduledWakeTimer?: ReturnType<typeof setTimeout>;\n\tprivate scheduledWakeRecompute?: Promise<void>;\n\tprivate scheduledWakeRecomputeQueued = false;\n\tprivate readonly scheduledWakeFailures = new Map<string, number>();\n\n\tconstructor(\n\t\tprivate readonly socketPath: string,\n\t\toptions: DaemonSupervisorOptions,\n\t) {\n\t\tthis.ready = new Promise<void>((resolveReady, rejectReady) => {\n\t\t\tthis.markReady = resolveReady;\n\t\t\tthis.rejectReady = rejectReady;\n\t\t});\n\t\tvoid this.ready.catch(() => undefined);\n\t\tconst agentDir = options.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) {\n\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t}\n\t\tthis.descriptorDir = options.descriptorDir ?? defaultWorkerDescriptorDir(agentDir, socketPath);\n\t\tthis.supervisorConfigPath = join(this.descriptorDir, SUPERVISOR_CONFIG_FILE_NAME);\n\t\tthis.defaultSessionConfig = mergeAgentSessionRuntimeConfig(\n\t\t\toptions.defaultSessionConfig,\n\t\t\tthis.loadPersistedSupervisorConfig(),\n\t\t);\n\t\tthis.snapshotCacheRoot = join(this.descriptorDir, \"snapshot-cache\", this.generation);\n\t\tthis.catalog = new DaemonCatalogClient((message) => this.log(message));\n\t\tthis.settingsManager = SettingsManager.create(process.cwd(), this.defaultSessionConfig.agentDir ?? agentDir);\n\t}\n\n\tasync start(): Promise<void> {\n\t\ttry {\n\t\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\t\tif (!agentDir) {\n\t\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t\t}\n\t\t\tthis.socketLease = await acquireDaemonSocketPathLease(this.socketPath);\n\t\t\tthis.socketLease?.onCompromised((error) => this.handleSocketLeaseCompromised(error));\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait waitForDaemonStartupFence(this.socketPath);\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tconst journalScope = this.rlmJournalScope();\n\t\t\tthis.ownership = await acquireDaemonSupervisorOwnership({\n\t\t\t\tsocketPath: this.socketPath,\n\t\t\t\tdescriptorDir: this.descriptorDir,\n\t\t\t\tagentDir,\n\t\t\t\tjournalPath: journalScope.journalPath,\n\t\t\t\tgeneration: this.generation,\n\t\t\t\tappVersion: VERSION,\n\t\t\t});\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.rlmJournalOwner = await RlmJournalOwner.open(journalScope);\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait prepareDaemonSocketPath(this.socketPath, this.socketLease);\n\n\t\t\tmkdirSync(this.descriptorDir, { recursive: true, mode: 0o700 });\n\t\t\tchmodSync(this.descriptorDir, 0o700);\n\t\t\tthis.persistSupervisorConfig();\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t\tmkdirSync(this.snapshotCacheRoot, { recursive: true, mode: 0o700 });\n\t\t\tthis.commandJournal = new CommandRecoveryJournal(join(this.descriptorDir, \"command-journal.jsonl\"));\n\t\t\tthis.loadWorkerDescriptors();\n\t\t\tconst workersToAdopt = [...this.workers.values()];\n\n\t\t\tthis.server = createServer((socket) => this.handleConnection(socket));\n\t\t\tawait this.listen();\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.socketIdentity = getDaemonSocketIdentity(this.socketPath);\n\t\t\tif (process.platform !== \"win32\" && !this.socketIdentity) {\n\t\t\t\tthrow new Error(`Could not capture daemon socket identity: ${this.socketPath}`);\n\t\t\t}\n\t\t\tthis.ownsSocketPath = true;\n\t\t\trestrictDaemonSocketPath(this.socketPath);\n\n\t\t\tthis.registerSignalHandlers();\n\t\t\tconst ownedSessionFiles = new Set(\n\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t.flatMap((worker) => [worker.descriptor.sessionFile, worker.descriptor.createCommand.sessionPath])\n\t\t\t\t\t.filter((path): path is string => typeof path === \"string\")\n\t\t\t\t\t.map((path) => resolve(path)),\n\t\t\t);\n\t\t\tconst migratedJobs = migrateLegacyCronJobsToSessionArtifacts(getCronJobsPath(agentDir), {\n\t\t\t\tisSessionOwned: (job) => ownedSessionFiles.has(resolve(job.sessionFile)),\n\t\t\t});\n\t\t\tif (migratedJobs > 0) {\n\t\t\t\tthis.log(`Migrated ${migratedJobs} scheduled jobs into session artifacts`);\n\t\t\t}\n\t\t\tawait this.catalog.start().catch((error) => this.log(`Could not start daemon catalog: ${String(error)}`));\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait this.seedRosterLedger();\n\t\t\tlet adoptionFailure: unknown;\n\t\t\tlet adoptionFailed = false;\n\t\t\tawait Promise.all(\n\t\t\t\tworkersToAdopt.map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.adoptOrRecoverWorker(worker);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tif (!adoptionFailed) {\n\t\t\t\t\t\t\tadoptionFailed = true;\n\t\t\t\t\t\t\tadoptionFailure = error;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t\tif (adoptionFailed) {\n\t\t\t\tthrow adoptionFailure;\n\t\t\t}\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tthis.scheduleOwnedWorkerCleanup(worker);\n\t\t\t}\n\t\t\tthis.scheduleIdleEvictionSweep();\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\tthis.rosterWatchdogTimer = setInterval(() => this.sweepRosterStaleness(), ROSTER_WATCHDOG_INTERVAL_MS);\n\t\t\tthis.rosterWatchdogTimer.unref();\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait this.ownership.updatePhase(\"owner\");\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.startupComplete = true;\n\t\t\tthis.log(`${PRODUCT.name} daemon supervisor ${this.generation} listening on ${this.socketPath}`);\n\t\t\tthis.markReady();\n\t\t} catch (error) {\n\t\t\tconst startupError = error instanceof Error ? error : new Error(String(error));\n\t\t\tthis.log(`Daemon supervisor startup failed: ${startupError.stack ?? startupError.message}`);\n\t\t\tawait this.cleanupSupervisorResources();\n\t\t\tthis.rejectReady(startupError);\n\t\t\tthrow startupError;\n\t\t}\n\t}\n\n\tprivate listen(): Promise<void> {\n\t\treturn new Promise<void>((resolveListen, rejectListen) => {\n\t\t\tconst onError = (error: Error) => {\n\t\t\t\tthis.server?.off(\"listening\", onListening);\n\t\t\t\trejectListen(error);\n\t\t\t};\n\t\t\tconst onListening = () => {\n\t\t\t\tthis.server?.off(\"error\", onError);\n\t\t\t\tresolveListen();\n\t\t\t};\n\t\t\tthis.server?.once(\"error\", onError);\n\t\t\tthis.server?.once(\"listening\", onListening);\n\t\t\tthis.server?.listen(this.socketPath);\n\t\t});\n\t}\n\n\tprivate log(message: string): void {\n\t\tconsole.error(message);\n\t\tstructuredLog.warn(message, { socketPath: this.socketPath });\n\t\tappendRotatingLog(getDaemonLogPath(this.socketPath), `[${new Date().toISOString()}] supervisor: ${message}`);\n\t}\n\n\tprivate clearIdleEvictionTimer(): void {\n\t\tif (!this.idleEvictionTimer) return;\n\t\tclearTimeout(this.idleEvictionTimer);\n\t\tthis.idleEvictionTimer = undefined;\n\t}\n\n\tprivate clearRosterWatchdogTimer(): void {\n\t\tif (!this.rosterWatchdogTimer) return;\n\t\tclearInterval(this.rosterWatchdogTimer);\n\t\tthis.rosterWatchdogTimer = undefined;\n\t}\n\n\tprivate clearScheduledWakeTimer(): void {\n\t\tif (!this.scheduledWakeTimer) return;\n\t\tclearTimeout(this.scheduledWakeTimer);\n\t\tthis.scheduledWakeTimer = undefined;\n\t}\n\n\t// The supervisor only wakes non-resident trees; firing and delivery stay worker-owned.\n\tprivate scheduleScheduledSessionWakeRecompute(): void {\n\t\tif (this.shuttingDown) return;\n\t\tif (this.scheduledWakeRecompute) {\n\t\t\tthis.scheduledWakeRecomputeQueued = true;\n\t\t\treturn;\n\t\t}\n\t\tthis.scheduledWakeRecompute = this.recomputeScheduledSessionWake()\n\t\t\t.catch((error) => this.log(`Scheduled-session wake recompute failed: ${String(error)}`))\n\t\t\t.finally(() => {\n\t\t\t\tthis.scheduledWakeRecompute = undefined;\n\t\t\t\tif (this.scheduledWakeRecomputeQueued) {\n\t\t\t\t\tthis.scheduledWakeRecomputeQueued = false;\n\t\t\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\t\t}\n\t\t\t});\n\t}\n\n\t/** Durable truth: the ledger family (fork headers stripped) plus each session's scheduled-jobs artifact. */\n\tprivate async collectPassiveScheduledJobs(\n\t\tincludeInactive = false,\n\t): Promise<Array<{ rootSessionFile: string; job: AgentCronJob; info: SessionInfo }>> {\n\t\tconst pendingCancelRoots = new Set<string>();\n\t\tfor (const intent of this.collectEphemeralCancelIntents()) {\n\t\t\tconst context = this.workerSessionArtifactContext(intent);\n\t\t\tif (!context || !intent.descriptor.rootSessionId) continue;\n\t\t\ttry {\n\t\t\t\tawait this.cancelScheduledJobsForSessionTree(intent.descriptor.rootSessionId, context.sessionFile);\n\t\t\t\tthis.deleteWorkerDescriptor(intent);\n\t\t\t} catch {\n\t\t\t\t// Still owned until the cancel lands; the tree stays excluded below.\n\t\t\t\tpendingCancelRoots.add(canonicalSessionPath(context.sessionFile));\n\t\t\t}\n\t\t}\n\t\tconst infos = await this.rlmSpawnLedger().family();\n\t\tconst infoByPath = new Map(infos.map((info) => [canonicalSessionPath(info.path), info] as const));\n\t\tconst storeBySessionId = new Map<string, AgentCronJobStore>();\n\t\tconst infoBySessionId = new Map<string, SessionInfo>();\n\t\tfor (const info of infos) {\n\t\t\tif (info.state !== undefined && info.state.status !== \"active\") continue;\n\t\t\tconst artifactDir = getSessionArtifactPathForFile(resolve(info.path), info.id);\n\t\t\tif (!existsSync(join(artifactDir, SESSION_SCHEDULED_JOBS_FILENAME))) continue;\n\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\tstore.registerSessionArtifact(info.id, artifactDir);\n\t\t\tstoreBySessionId.set(info.id, store);\n\t\t\tinfoBySessionId.set(info.id, info);\n\t\t}\n\t\tif (infoBySessionId.size === 0) return [];\n\t\tconst uncoveredRootFor = (info: SessionInfo): string | undefined => {\n\t\t\tlet current = info;\n\t\t\tconst visited = new Set([canonicalSessionPath(current.path)]);\n\t\t\twhile (true) {\n\t\t\t\ttry {\n\t\t\t\t\tif (this.findWorkerBySessionFile(current.path)) return undefined;\n\t\t\t\t} catch {\n\t\t\t\t\treturn undefined;\n\t\t\t\t}\n\t\t\t\tif (!current.parentSessionPath) break;\n\t\t\t\tconst parent = infoByPath.get(canonicalSessionPath(current.parentSessionPath));\n\t\t\t\tif (!parent || visited.has(canonicalSessionPath(parent.path))) break;\n\t\t\t\tvisited.add(canonicalSessionPath(parent.path));\n\t\t\t\tcurrent = parent;\n\t\t\t}\n\t\t\treturn current.path;\n\t\t};\n\t\tconst results: Array<{ rootSessionFile: string; job: AgentCronJob; info: SessionInfo }> = [];\n\t\tfor (const [artifactSessionId, store] of storeBySessionId) {\n\t\t\tlet jobs: AgentCronJob[];\n\t\t\ttry {\n\t\t\t\tjobs = store.list();\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Skipping unreadable scheduled jobs for session ${artifactSessionId}: ${String(error)}`);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tfor (const job of jobs) {\n\t\t\t\tif (!includeInactive && job.status !== \"active\" && job.status !== \"paused\") continue;\n\t\t\t\tconst info = infoBySessionId.get(job.sessionId);\n\t\t\t\tif (!info) continue;\n\t\t\t\tconst rootSessionFile = uncoveredRootFor(info);\n\t\t\t\tif (rootSessionFile === undefined) continue;\n\t\t\t\tif (pendingCancelRoots.has(canonicalSessionPath(rootSessionFile))) continue;\n\t\t\t\tresults.push({ rootSessionFile, job, info });\n\t\t\t}\n\t\t}\n\t\treturn results;\n\t}\n\n\tprivate async recomputeScheduledSessionWake(): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst candidates = await this.collectPassiveScheduledJobs();\n\t\tthis.clearScheduledWakeTimer();\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst now = Date.now();\n\t\tconst wakeTimes: number[] = [];\n\t\tconst candidateRoots = new Set(candidates.map(({ rootSessionFile }) => canonicalSessionPath(rootSessionFile)));\n\t\tfor (const root of [...this.scheduledWakeFailures.keys()]) {\n\t\t\tif (!candidateRoots.has(root)) this.scheduledWakeFailures.delete(root);\n\t\t}\n\t\tfor (const { rootSessionFile, job } of candidates) {\n\t\t\tif (job.status !== \"active\" || job.nextRunAt === undefined) continue;\n\t\t\tconst runAt = Date.parse(job.nextRunAt);\n\t\t\tif (!Number.isFinite(runAt)) continue;\n\t\t\t// The failure floor keeps an overdue job off a hot retry loop.\n\t\t\tconst failedAt = this.scheduledWakeFailures.get(canonicalSessionPath(rootSessionFile));\n\t\t\twakeTimes.push(failedAt !== undefined ? Math.max(runAt, failedAt + SCHEDULED_WAKE_RETRY_MS) : runAt);\n\t\t}\n\t\tif (wakeTimes.length === 0) return;\n\t\tconst delay = Math.min(Math.max(0, Math.min(...wakeTimes) - now), SCHEDULED_WAKE_MAX_TIMEOUT_MS);\n\t\tthis.scheduledWakeTimer = setTimeout(() => {\n\t\t\tthis.scheduledWakeTimer = undefined;\n\t\t\tvoid this.wakeDueScheduledSessions().catch((error) =>\n\t\t\t\tthis.log(`Scheduled-session wake failed: ${String(error)}`),\n\t\t\t);\n\t\t}, delay);\n\t\tthis.scheduledWakeTimer.unref();\n\t}\n\n\tprivate async wakeDueScheduledSessions(now = Date.now()): Promise<void> {\n\t\t// Disarmed during update-restart preparation; the phase transition or next boot re-arms once.\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) {\n\t\t\tthis.clearScheduledWakeTimer();\n\t\t\treturn;\n\t\t}\n\t\ttry {\n\t\t\tconst due = new Map<string, string>();\n\t\t\tfor (const { rootSessionFile, job } of await this.collectPassiveScheduledJobs()) {\n\t\t\t\tif (job.status !== \"active\" || job.nextRunAt === undefined) continue;\n\t\t\t\tconst runAt = Date.parse(job.nextRunAt);\n\t\t\t\tif (!Number.isFinite(runAt) || runAt > now) continue;\n\t\t\t\tdue.set(canonicalSessionPath(rootSessionFile), rootSessionFile);\n\t\t\t}\n\t\t\tfor (const [rootKey, sessionPath] of due) {\n\t\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.createOrReuseWorker(SCHEDULED_WAKE_CLIENT_ID, { type: \"create\", sessionPath });\n\t\t\t\t\tthis.scheduledWakeFailures.delete(rootKey);\n\t\t\t\t\tthis.log(`Woke session worker for a due scheduled job: ${sessionPath}`);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tthis.scheduledWakeFailures.set(rootKey, Date.now());\n\t\t\t\t\tthis.log(`Scheduled wake failed for ${sessionPath}: ${String(error)}`);\n\t\t\t\t}\n\t\t\t}\n\t\t} finally {\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t}\n\t}\n\n\tprivate scheduleIdleEvictionSweep(): void {\n\t\tif (this.shuttingDown || this.idleEvictionTimer || this.idleEvictionSweep) return;\n\t\tconst delayMs = idleEvictionSweepIntervalMs(this.settingsManager.getIdleEvictionMinutes());\n\t\tthis.idleEvictionTimer = setTimeout(() => {\n\t\t\tthis.idleEvictionTimer = undefined;\n\t\t\tconst sweep = this.runIdleEvictionSweep()\n\t\t\t\t.catch((error) => this.log(`Idle eviction sweep failed: ${String(error)}`))\n\t\t\t\t.finally(() => {\n\t\t\t\t\tif (this.idleEvictionSweep === sweep) this.idleEvictionSweep = undefined;\n\t\t\t\t\tthis.scheduleIdleEvictionSweep();\n\t\t\t\t});\n\t\t\tthis.idleEvictionSweep = sweep;\n\t\t}, delayMs);\n\t\tthis.idleEvictionTimer.unref();\n\t}\n\n\tprivate workerEvictionSnapshot(worker: ResidentWorker): WorkerEvictionSnapshot {\n\t\treturn {\n\t\t\tlifecycle: worker.descriptor.lifecycle,\n\t\t\tisConnected: worker.client !== undefined,\n\t\t\tisStopping: this.isWorkerStopping(worker),\n\t\t\thasOwnerClient: worker.descriptor.ownerClientId !== undefined,\n\t\t\tisPreparingUpdateRestart:\n\t\t\t\tthis.updateRestartPhase !== undefined || worker.updateRestartPrepareClient !== undefined,\n\t\t\thasWakeBlindSchedule: this.isWakeBlindScheduledWorker(worker),\n\t\t\tsessions: this.workerRosterEntries(worker)\n\t\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t\t.map(sessionSummaryFromRosterEntry)\n\t\t\t\t.map((summary) => {\n\t\t\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\t\t\treturn {\n\t\t\t\t\t\tisSessionActive: isSessionSummaryBusy(summary),\n\t\t\t\t\t\tattachedClients: this.attachedClientCount(summary, activeSessionId),\n\t\t\t\t\t\thasRegisteredCronJob: summary.hasRegisteredCronJob === true,\n\t\t\t\t\t\tlastActivityAt: Date.parse(summary.lastActivityAt ?? \"\"),\n\t\t\t\t\t};\n\t\t\t\t}),\n\t\t};\n\t}\n\n\t/** A schedule whose root file sits outside the enumerable sessions root is invisible to the wake scan; keep it resident. */\n\tprivate isWakeBlindScheduledWorker(worker: ResidentWorker): boolean {\n\t\tconst sessionFile = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!sessionFile || !agentDir) return false;\n\t\tconst sessionsRoot = canonicalSessionPath(this.defaultSessionConfig.sessionDir ?? getSessionsDir(agentDir));\n\t\tif (dirname(canonicalSessionPath(sessionFile)) === sessionsRoot) return false;\n\t\treturn this.workerRosterEntries(worker).some(\n\t\t\t(entry) => entry.summary.hasRegisteredHeartbeat === true || entry.summary.hasRegisteredCronJob === true,\n\t\t);\n\t}\n\n\tprivate async runIdleEvictionSweep(now = Date.now()): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined || this.idleEvictionFence) return;\n\t\tawait this.settingsManager.reload();\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst idleEvictionMinutes = this.settingsManager.getIdleEvictionMinutes();\n\t\tif (idleEvictionMinutes === \"off\") return;\n\n\t\tconst refreshed = new Set<ResidentWorker>();\n\t\tawait Promise.all(\n\t\t\t[...this.workers.values()].map(async (worker) => {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.refreshWorkerSummaries(worker);\n\t\t\t\t\trefreshed.add(worker);\n\t\t\t\t} catch {\n\t\t\t\t\t// A disconnected or transitioning worker is never an eviction candidate.\n\t\t\t\t}\n\t\t\t}),\n\t\t);\n\t\tconst candidates = [...refreshed].filter((worker) =>\n\t\t\tcanEvictWorker(this.workerEvictionSnapshot(worker), idleEvictionMinutes, now),\n\t\t);\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t// Whole-tree candidates skip child work because stopWorker releases everything.\n\t\tawait Promise.all(\n\t\t\t[...refreshed]\n\t\t\t\t.filter((worker) => !candidates.includes(worker))\n\t\t\t\t.map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst response = await worker.client?.requestWorker(\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\ttype: \"worker_passivate_idle_children\",\n\t\t\t\t\t\t\t\tidleEvictionMinutes,\n\t\t\t\t\t\t\t\tnow,\n\t\t\t\t\t\t\t\tlimit: CHILD_PASSIVATION_PER_WORKER_CAP,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t30_000,\n\t\t\t\t\t\t);\n\t\t\t\t\t\tif (response && !response.success) throw new Error(response.error);\n\t\t\t\t\t\tawait this.refreshWorkerSummaries(worker);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\trefreshed.delete(worker);\n\t\t\t\t\t\tthis.log(`Child passivation sweep failed for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t);\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\n\t\tawait this.withEvictionFence(\"Timed out draining daemon mutations for idle eviction\", async () => {\n\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\tawait Promise.all(\n\t\t\t\tcandidates.map((worker) => this.refreshWorkerSummaries(worker).catch(() => refreshed.delete(worker))),\n\t\t\t);\n\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\tconst evictable = candidates.filter(\n\t\t\t\t(worker) =>\n\t\t\t\t\trefreshed.has(worker) &&\n\t\t\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\t\t\tcanEvictWorker(this.workerEvictionSnapshot(worker), idleEvictionMinutes, now),\n\t\t\t);\n\t\t\t// Promise.all may reject and release the fence while sibling stops are still\n\t\t\t// finishing. That is safe: a racing mutation either reaches a live worker or\n\t\t\t// gets a clean disconnected/unknown-session error.\n\t\t\tawait Promise.all(\n\t\t\t\tevictable.map(async (worker) => {\n\t\t\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\t\t\tconst snapshot = this.workerEvictionSnapshot(worker);\n\t\t\t\t\tconst idleMinutes = Math.floor(\n\t\t\t\t\t\tMath.min(...snapshot.sessions.map((session) => now - session.lastActivityAt)) / 60_000,\n\t\t\t\t\t);\n\t\t\t\t\tconst root = worker.summaries.get(worker.descriptor.rootActiveSessionId);\n\t\t\t\t\tawait this.stopWorker(worker, true);\n\t\t\t\t\tthis.log(\n\t\t\t\t\t\t`Evicted idle worker ${worker.descriptor.workerId} root=${root?.sessionId ?? worker.descriptor.rootSessionId ?? worker.descriptor.rootActiveSessionId} idleMinutes=${idleMinutes} sessions=${snapshot.sessions.length}`,\n\t\t\t\t\t);\n\t\t\t\t}),\n\t\t\t);\n\t\t});\n\t}\n\n\t/** Waits for any held eviction fence, then takes the slot; release clears it only if still ours. */\n\tprivate async acquireIdleEvictionFence(): Promise<() => void> {\n\t\twhile (this.idleEvictionFence) await this.idleEvictionFence;\n\t\tlet releaseFence: () => void = () => {};\n\t\tconst fence = new Promise<void>((resolveFence) => {\n\t\t\treleaseFence = resolveFence;\n\t\t});\n\t\tthis.idleEvictionFence = fence;\n\t\treturn () => {\n\t\t\tif (this.idleEvictionFence === fence) this.idleEvictionFence = undefined;\n\t\t\treleaseFence();\n\t\t};\n\t}\n\n\t/** Runs a passivation decision under the eviction fence after draining admitted mutations. */\n\tprivate async withEvictionFence(drainMessage: string, action: () => Promise<void>): Promise<void> {\n\t\tconst releaseFence = await this.acquireIdleEvictionFence();\n\t\ttry {\n\t\t\tawait this.mutationDrain.waitForDrain(0, AbortSignal.timeout(IDLE_EVICTION_DRAIN_TIMEOUT_MS), drainMessage);\n\t\t\tawait action();\n\t\t} finally {\n\t\t\treleaseFence();\n\t\t}\n\t}\n\n\tprivate async passivateWorkerIfStillEligible(\n\t\tworker: ResidentWorker,\n\t\tisStillEligible: () => boolean,\n\t\tdescribeEvicted: () => string,\n\t): Promise<void> {\n\t\tawait this.refreshWorkerSummaries(worker, false, true);\n\t\tif (!isStillEligible()) return;\n\t\tawait this.stopWorker(worker, true);\n\t\tthis.log(describeEvicted());\n\t}\n\n\tprivate async evictEmptySessionOnLastDetach(activeSessionId: string): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst worker = this.matchWorkers(activeSessionId)[0]?.worker;\n\t\tif (\n\t\t\t!worker ||\n\t\t\tworker.descriptor.lifecycle !== \"ready\" ||\n\t\t\t!worker.client ||\n\t\t\tworker.descriptor.ownerClientId !== undefined || // owned workers have their own cleanup path\n\t\t\tthis.isWorkerStopping(worker)\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\ttry {\n\t\t\tawait this.refreshWorkerSummaries(worker, false, true);\n\t\t} catch {\n\t\t\treturn;\n\t\t}\n\t\tif (!this.isEmptyDetachEvictionCandidate(worker)) return;\n\t\ttry {\n\t\t\t// Idle-sweep coordination: fence new mutations, drain admitted ones, re-read before deciding.\n\t\t\tawait this.withEvictionFence(\"Timed out draining daemon mutations for empty-session eviction\", () =>\n\t\t\t\tthis.passivateWorkerIfStillEligible(\n\t\t\t\t\tworker,\n\t\t\t\t\t() => this.isEmptyDetachEvictionCandidate(worker),\n\t\t\t\t\t() =>\n\t\t\t\t\t\t`Evicted empty session worker ${worker.descriptor.workerId} root=${worker.descriptor.rootSessionId ?? worker.descriptor.rootActiveSessionId} on last client detach`,\n\t\t\t\t),\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tthis.log(`Empty-session eviction failed for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate isEmptyDetachEvictionCandidate(worker: ResidentWorker): boolean {\n\t\tif (\n\t\t\tthis.shuttingDown ||\n\t\t\tthis.updateRestartPhase !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker ||\n\t\t\tthis.isWorkerStopping(worker) ||\n\t\t\tthis.isWakeBlindScheduledWorker(worker)\n\t\t) {\n\t\t\treturn false;\n\t\t}\n\t\tconst summaries = this.workerRosterEntries(worker)\n\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t.map(sessionSummaryFromRosterEntry);\n\t\tconst hasAttachedClient = summaries.some(\n\t\t\t(summary) => this.attachedClientCount(summary, summary.activeSessionId ?? summary.id) > 0,\n\t\t);\n\t\treturn summaries.length > 0 && !hasAttachedClient && summaries.every(isEvictableEmptySessionSummary);\n\t}\n\n\tprivate async assertCurrentOwnership(): Promise<void> {\n\t\tconst ownership = this.ownership;\n\t\tif (!ownership) {\n\t\t\tconst error = new Error(\n\t\t\t\t`Daemon supervisor generation ${this.generation} holds no registry ownership (never acquired or already released); ` +\n\t\t\t\t\t`socket: ${this.socketPath}; restart the daemon to recover — sessions are preserved`,\n\t\t\t);\n\t\t\tObject.assign(error, { code: \"supervisor_generation_stale\" as const });\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t\ttry {\n\t\t\tawait ownership.assertCurrent();\n\t\t} catch (error) {\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async assertServingCurrentOwnership(allowWorkerLedger = false): Promise<void> {\n\t\tthis.assertSupervisorServing(allowWorkerLedger);\n\t\tawait this.assertCurrentOwnership();\n\t\tthis.assertSupervisorServing(allowWorkerLedger);\n\t}\n\n\tprivate async assertRecoveryAllowed(): Promise<void> {\n\t\tawait this.assertServingCurrentOwnership();\n\t\tif (await isDaemonShutdownAdmissionActive()) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Daemon shutdown admission cancelled worker recovery\");\n\t\t}\n\t}\n\n\tprivate supervisorAuthenticationClaim(): {\n\t\tsupervisorGeneration: string;\n\t\tsupervisorPid: number;\n\t\tsupervisorProcessStartId?: string;\n\t\tsupervisorSocketPath: string;\n\t} {\n\t\tconst record = this.ownership?.record;\n\t\tif (!record) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Daemon supervisor ownership is unavailable\");\n\t\t}\n\t\treturn {\n\t\t\tsupervisorGeneration: this.generation,\n\t\t\tsupervisorPid: record.pid,\n\t\t\t...(record.processStartId ? { supervisorProcessStartId: record.processStartId } : {}),\n\t\t\tsupervisorSocketPath: record.socketPath,\n\t\t};\n\t}\n\n\tprivate loadWorkerDescriptors(): void {\n\t\tfor (const name of readdirSync(this.descriptorDir)) {\n\t\t\tif (name === SUPERVISOR_CONFIG_FILE_NAME || !name.endsWith(\".json\")) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst path = join(this.descriptorDir, name);\n\t\t\ttry {\n\t\t\t\tconst descriptor: unknown = JSON.parse(readFileSync(path, \"utf8\"));\n\t\t\t\tif (!isDaemonWorkerDescriptor(descriptor, this.socketPath)) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tdescriptor.supervisorSocketPath = normalizeSocketPath(descriptor.supervisorSocketPath);\n\t\t\t\tdescriptor.lifecycle = \"recovering\";\n\t\t\t\tdescriptor.recoveryJournalPath ??= join(this.descriptorDir, `${descriptor.workerId}.recovery.jsonl`);\n\t\t\t\tdescriptor.orphanProcessJournalPath ??= join(this.descriptorDir, `${descriptor.workerId}.orphans.jsonl`);\n\t\t\t\tconst durableDescriptor = durableDaemonWorkerDescriptor(descriptor);\n\t\t\t\tconst worker: ResidentWorker = {\n\t\t\t\t\tdescriptor: durableDescriptor,\n\t\t\t\t\tdescriptorPath: path,\n\t\t\t\t\tsummaries: new Map(),\n\t\t\t\t\tsnapshotCache: new Map(),\n\t\t\t\t\ttranscriptCaches: new Map(),\n\t\t\t\t\tsnapshotGenerations: new Map(),\n\t\t\t\t\tsnapshotLoads: new Map(),\n\t\t\t\t\tintentionalStop: durableDescriptor.stopRequestedAt !== undefined,\n\t\t\t\t\tstopRevision: 0,\n\t\t\t\t};\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.workers.set(durableDescriptor.workerId, worker);\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Ignoring invalid worker descriptor ${path}: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t}\n\n\t/** Trees whose ephemeral-stop cancel failed stay owned until a retry lands: tombstoned client-owned descriptors with no resident worker. */\n\tprivate collectEphemeralCancelIntents(): Array<{ descriptor: DaemonWorkerDescriptor; descriptorPath: string }> {\n\t\tconst intents: Array<{ descriptor: DaemonWorkerDescriptor; descriptorPath: string }> = [];\n\t\tlet names: string[];\n\t\ttry {\n\t\t\tnames = readdirSync(this.descriptorDir);\n\t\t} catch {\n\t\t\treturn intents;\n\t\t}\n\t\tfor (const name of names) {\n\t\t\tif (name === SUPERVISOR_CONFIG_FILE_NAME || !name.endsWith(\".json\")) continue;\n\t\t\tconst descriptorPath = join(this.descriptorDir, name);\n\t\t\tlet descriptor: unknown;\n\t\t\ttry {\n\t\t\t\tdescriptor = JSON.parse(readFileSync(descriptorPath, \"utf8\"));\n\t\t\t} catch {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (!isDaemonWorkerDescriptor(descriptor, this.socketPath)) continue;\n\t\t\tif (descriptor.stopRequestedAt === undefined || descriptor.ownerClientId === undefined) continue;\n\t\t\tif (this.workers.has(descriptor.workerId)) continue;\n\t\t\tintents.push({ descriptor, descriptorPath });\n\t\t}\n\t\treturn intents;\n\t}\n\n\tprivate loadPersistedSupervisorConfig(): AgentSessionRuntimeConfig | undefined {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(\n\t\t\t\treadFileSync(this.supervisorConfigPath, \"utf8\"),\n\t\t\t) as Partial<PersistedSupervisorConfig>;\n\t\t\tif (\n\t\t\t\tparsed.version !== 1 ||\n\t\t\t\ttypeof parsed.socketPath !== \"string\" ||\n\t\t\t\tnormalizeSocketPath(parsed.socketPath) !== this.socketPath ||\n\t\t\t\t!parsed.defaultSessionConfig ||\n\t\t\t\ttypeof parsed.defaultSessionConfig !== \"object\" ||\n\t\t\t\ttypeof parsed.defaultSessionConfig.agentDir !== \"string\"\n\t\t\t) {\n\t\t\t\treturn undefined;\n\t\t\t}\n\t\t\treturn durableAgentSessionRuntimeConfig(parsed.defaultSessionConfig);\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate persistSupervisorConfig(): void {\n\t\tconst persisted: PersistedSupervisorConfig = {\n\t\t\tversion: 1,\n\t\t\tsocketPath: this.socketPath,\n\t\t\tdefaultSessionConfig: durableAgentSessionRuntimeConfig(this.defaultSessionConfig),\n\t\t};\n\t\tconst tempPath = `${this.supervisorConfigPath}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(persisted, null, 2)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\trenameSync(tempPath, this.supervisorConfigPath);\n\t}\n\n\tprivate hasPersistedWorkerDescriptors(): boolean {\n\t\treturn readdirSync(this.descriptorDir).some(\n\t\t\t(name) => name !== SUPERVISOR_CONFIG_FILE_NAME && name.endsWith(\".json\"),\n\t\t);\n\t}\n\n\tprivate persistWorker(worker: ResidentWorker): void {\n\t\tworker.descriptor.updatedAt = new Date().toISOString();\n\t\tconst persisted = durableDaemonWorkerDescriptor(worker.descriptor);\n\t\tconst tempPath = `${worker.descriptorPath}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(persisted, null, 2)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\trenameSync(tempPath, worker.descriptorPath);\n\t}\n\n\tprivate deleteWorkerDescriptor(worker: { descriptorPath: string; descriptor: DaemonWorkerDescriptor }): void {\n\t\ttry {\n\t\t\trmSync(worker.descriptorPath, { force: true });\n\t\t\trmSync(worker.descriptor.recoveryJournalPath, { force: true });\n\t\t\tif (worker.descriptor.orphanProcessJournalPath) {\n\t\t\t\trmSync(worker.descriptor.orphanProcessJournalPath, { force: true });\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Failed to remove worker descriptor ${worker.descriptorPath}: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate handleConnection(socket: Socket): void {\n\t\tconst client: DaemonSocketClient = {\n\t\t\tid: createActiveSessionId(),\n\t\t\tsocket,\n\t\t\tattachedActiveSessionIds: new Set(),\n\t\t\tcatchupActiveSessionIds: new Set(),\n\t\t\tbackpressured: false,\n\t\t\tauthenticated: true,\n\t\t\tsnapshotActiveSessionIds: new Set(),\n\t\t\tdetachInput: () => {},\n\t\t\tsupportsExtensionUi: false,\n\t\t\tcapabilities: new Set(DAEMON_DEFAULT_CLIENT_CAPABILITIES),\n\t\t};\n\t\tthis.connectionIds.set(client, client.id);\n\t\tthis.sessionInputPauseEpochs.set(client, 0);\n\t\tthis.detachingInputPauseSessions.set(client, new Set());\n\t\tthis.clients.add(client);\n\t\tvoid this.ready.then(\n\t\t\t() => {\n\t\t\t\tif (!client.socket.destroyed && this.clients.has(client)) {\n\t\t\t\t\tthis.write(client, {\n\t\t\t\t\t\ttype: \"daemon_hello\",\n\t\t\t\t\t\tsocketPath: this.socketPath,\n\t\t\t\t\t\tprotocol: DAEMON_PROTOCOL_INFO,\n\t\t\t\t\t\tschemaId: DAEMON_SCHEMA_ID,\n\t\t\t\t\t\tschemaRevision: DAEMON_SCHEMA_REVISION,\n\t\t\t\t\t\tappVersion: VERSION,\n\t\t\t\t\t\truntime: getDaemonRuntimeIdentity(),\n\t\t\t\t\t\tsupervisorGeneration: this.generation,\n\t\t\t\t\t\tsupervisorOwnerToken: this.ownership?.record.token,\n\t\t\t\t\t\tsupervisorPid: process.pid,\n\t\t\t\t\t\tsupervisorProcessStartId: this.ownership?.record.processStartId,\n\t\t\t\t\t\tsupervisorSocketPath: this.ownership?.record.socketPath,\n\t\t\t\t\t\tclientId: client.id,\n\t\t\t\t\t\tserverCapabilities: SUPERVISOR_SERVER_CAPABILITIES,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t},\n\t\t\t() => client.socket.destroy(),\n\t\t);\n\n\t\tclient.detachInput = attachJsonlLineReader(socket, (line) => void this.handleLine(client, line));\n\t\tlet cleaned = false;\n\t\tconst cleanup = () => {\n\t\t\tif (cleaned) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tcleaned = true;\n\t\t\tclient.detachInput();\n\t\t\tthis.sessionInputPauseEpochs.set(client, (this.sessionInputPauseEpochs.get(client) ?? 0) + 1);\n\t\t\tconst ownerClientId = this.protocolClientId(client);\n\t\t\tvoid this.releaseClientSessionInputPauses(client, undefined, true).catch((error: unknown) =>\n\t\t\t\tthis.log(`Failed to release input pauses for disconnected client ${ownerClientId}: ${String(error)}`),\n\t\t\t);\n\t\t\tthis.clients.delete(client);\n\t\t\tthis.cancelWaitingPromptAdmissionsForClient(client);\n\t\t\tfor (const activeSessionId of [...client.attachedActiveSessionIds]) {\n\t\t\t\tclient.attachedActiveSessionIds.delete(activeSessionId);\n\t\t\t\tvoid this.syncWorkerExtensionUi(activeSessionId);\n\t\t\t\tvoid this.evictEmptySessionOnLastDetach(activeSessionId);\n\t\t\t}\n\t\t\tthis.scheduleOwnedWorkerCleanupForClient(this.protocolClientId(client));\n\t\t};\n\t\tsocket.on(\"close\", cleanup);\n\t\tsocket.on(\"error\", cleanup);\n\t\tsocket.on(\"drain\", () => {\n\t\t\tclient.backpressured = false;\n\t\t\tif (client.rosterResyncPending && client.rosterSubscribed === true) {\n\t\t\t\t// socket.write queues even when it reports backpressure: one resync per loss gap.\n\t\t\t\tclient.rosterResyncPending = false;\n\t\t\t\tthis.write(client, { type: \"roster_update\", changed: this.rosterEntriesForClient(), resync: true });\n\t\t\t}\n\t\t\tif (!client.snapshotStreaming) {\n\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t);\n\t\t\t}\n\t\t});\n\t}\n\n\tprivate cancelOwnedWorkerCleanup(clientId: string): void {\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.descriptor.ownerClientId !== clientId || !worker.ownerCleanupTimer) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t}\n\n\tprivate protocolClientId(client: DaemonSocketClient): string {\n\t\treturn this.protocolClientIds.get(client) ?? client.id;\n\t}\n\n\tprivate async releaseClientSessionInputPauses(\n\t\towner: DaemonSocketClient,\n\t\tactiveSessionId?: string,\n\t\tforceCleanupOnFailure = false,\n\t): Promise<void> {\n\t\tconst entries = [...this.sessionInputPauses.values()].filter(\n\t\t\t(entry) =>\n\t\t\t\tentry.owner === owner &&\n\t\t\t\t(activeSessionId === undefined ||\n\t\t\t\t\tentry.activeSessionId === activeSessionId ||\n\t\t\t\t\tentry.requestedActiveSessionId === activeSessionId),\n\t\t);\n\t\tawait Promise.all(\n\t\t\tentries.map(async (entry) => {\n\t\t\t\tif (this.workers.get(entry.worker.descriptor.workerId) !== entry.worker || !entry.worker.client) {\n\t\t\t\t\tif (this.sessionInputPauses.get(entry.pauseId) === entry) this.sessionInputPauses.delete(entry.pauseId);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tconst response = await this.forwardToWorker(\n\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tid: randomUUID(),\n\t\t\t\t\t\t\ttype: \"release_session_input_pause\",\n\t\t\t\t\t\t\tactiveSessionId: entry.activeSessionId,\n\t\t\t\t\t\t\tpauseId: entry.pauseId,\n\t\t\t\t\t\t},\n\t\t\t\t\t\tINPUT_PAUSE_CLEANUP_TIMEOUT_MS,\n\t\t\t\t\t);\n\t\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\t\tif (this.sessionInputPauses.get(entry.pauseId) === entry) this.sessionInputPauses.delete(entry.pauseId);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (forceCleanupOnFailure) {\n\t\t\t\t\t\tconst workerClient = entry.worker.client;\n\t\t\t\t\t\tif (workerClient) {\n\t\t\t\t\t\t\tworkerClient.close();\n\t\t\t\t\t\t\tawait this.handleWorkerClose(\n\t\t\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t\t\tworkerClient,\n\t\t\t\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\tthis.invalidateWorkerSessionInputPauses(\n\t\t\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t\t\t\"Session worker became unavailable while releasing an input pause\",\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t}),\n\t\t);\n\t}\n\n\tprivate invalidateWorkerSessionInputPauses(worker: ResidentWorker, reason: string): void {\n\t\tconst owners = new Set<DaemonSocketClient>();\n\t\tfor (const [pauseId, entry] of this.sessionInputPauses) {\n\t\t\tif (entry.worker !== worker) continue;\n\t\t\tthis.sessionInputPauses.delete(pauseId);\n\t\t\towners.add(entry.owner);\n\t\t}\n\t\tfor (const owner of owners) {\n\t\t\towner.socket.destroy(new Error(reason));\n\t\t}\n\t}\n\n\tprivate scheduleOwnedWorkerCleanupForClient(clientId: string): void {\n\t\tif ([...this.clients].some((client) => this.protocolClientId(client) === clientId)) {\n\t\t\treturn;\n\t\t}\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.descriptor.ownerClientId === clientId) {\n\t\t\t\tthis.scheduleOwnedWorkerCleanup(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate scheduleOwnedWorkerCleanup(worker: ResidentWorker): void {\n\t\tconst ownerClientId = worker.descriptor.ownerClientId;\n\t\tif (\n\t\t\t!ownerClientId ||\n\t\t\tworker.compatibilityError !== undefined ||\n\t\t\tworker.ownerCleanupTimer ||\n\t\t\t[...this.clients].some((client) => this.protocolClientId(client) === ownerClientId)\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tworker.ownerCleanupTimer = setTimeout(() => {\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t\tif (\n\t\t\t\tworker.compatibilityError !== undefined ||\n\t\t\t\tworker.descriptor.ownerClientId !== ownerClientId ||\n\t\t\t\t[...this.clients].some((client) => this.protocolClientId(client) === ownerClientId) ||\n\t\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t\t) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tvoid this.stopWorker(worker, true).catch((error) =>\n\t\t\t\tthis.log(`Could not clean up client-owned worker ${worker.descriptor.workerId}: ${String(error)}`),\n\t\t\t);\n\t\t}, OWNED_WORKER_DISCONNECT_GRACE_MS);\n\t\tworker.ownerCleanupTimer.unref();\n\t}\n\n\tprivate promptAdmissionKey(activeSessionId: string, publicAdmissionId: string): string {\n\t\treturn `${activeSessionId}\\0${publicAdmissionId}`;\n\t}\n\n\tprivate promptAdmissionsFor(client: DaemonSocketClient): Map<string, SupervisorPromptAdmission> {\n\t\tlet admissions = this.promptAdmissions.get(client);\n\t\tif (!admissions) {\n\t\t\tadmissions = new Map();\n\t\t\tthis.promptAdmissions.set(client, admissions);\n\t\t}\n\t\treturn admissions;\n\t}\n\n\tprivate getPromptAdmission(\n\t\tclient: DaemonSocketClient,\n\t\tactiveSessionId: string,\n\t\tpublicAdmissionId: string,\n\t): SupervisorPromptAdmission | undefined {\n\t\treturn this.promptAdmissions.get(client)?.get(this.promptAdmissionKey(activeSessionId, publicAdmissionId));\n\t}\n\n\tprivate deletePromptAdmission(admission: SupervisorPromptAdmission): void {\n\t\tconst admissions = this.promptAdmissions.get(admission.client);\n\t\tconst key = this.promptAdmissionKey(admission.activeSessionId, admission.publicAdmissionId);\n\t\tif (admissions?.get(key) !== admission) return;\n\t\tadmissions.delete(key);\n\t\tif (admissions.size === 0) this.promptAdmissions.delete(admission.client);\n\t}\n\n\tprivate cancelWaitingPromptAdmissionsForClient(client: DaemonSocketClient): void {\n\t\tfor (const admission of this.promptAdmissions.get(client)?.values() ?? []) {\n\t\t\tif (admission.status !== \"waiting\") continue;\n\t\t\tif (!admission.worker || !admission.workerActiveSessionId) {\n\t\t\t\tadmission.status = \"cancelled\";\n\t\t\t\tadmission.controller.abort();\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst worker = admission.worker;\n\t\t\tconst workerActiveSessionId = admission.workerActiveSessionId;\n\t\t\tvoid this.forwardToWorker(worker, {\n\t\t\t\ttype: \"cancel_prompt_admission\",\n\t\t\t\tactiveSessionId: workerActiveSessionId,\n\t\t\t\tadmissionId: admission.workerAdmissionId,\n\t\t\t})\n\t\t\t\t.then((response) => {\n\t\t\t\t\tif (admission.status !== \"waiting\") return;\n\t\t\t\t\tconst status =\n\t\t\t\t\t\tresponse.success && response.data && typeof response.data === \"object\" && \"status\" in response.data\n\t\t\t\t\t\t\t? (response.data as { status?: unknown }).status\n\t\t\t\t\t\t\t: undefined;\n\t\t\t\t\tif (status === \"owned\") admission.status = \"owned\";\n\t\t\t\t\telse if (status === \"cancelled\") admission.status = \"cancelled\";\n\t\t\t\t})\n\t\t\t\t.catch((error: unknown) => {\n\t\t\t\t\tthis.log(\n\t\t\t\t\t\t`Could not cancel prompt admission ${admission.workerAdmissionId} on disconnected client: ${String(error)}`,\n\t\t\t\t\t);\n\t\t\t\t});\n\t\t}\n\t}\n\n\t/** Non-async by design: prompt registration completes before handleLine's first await. */\n\tprivate parseCommandAndRegisterPromptAdmission(\n\t\tclient: DaemonSocketClient,\n\t\tline: string,\n\t): {\n\t\tcommand: DaemonCommand;\n\t\tenvelopeClientId?: string;\n\t\tprotocolVersion: number;\n\t\tadmission?: SupervisorPromptAdmission;\n\t} {\n\t\tconst parsed = JSON.parse(line) as unknown;\n\t\tconst envelope = isDaemonCommandEnvelope(parsed) ? parsed : undefined;\n\t\tif (!envelope) {\n\t\t\tthrow new Error(`Daemon commands require protocol ${DAEMON_COMMAND_ENVELOPE_MIN_PROTOCOL_VERSION} or newer`);\n\t\t}\n\t\tconst command = { ...envelope.command, id: envelope.id } as DaemonCommand;\n\t\tif (\n\t\t\tenvelope.protocol.version < CANONICAL_SESSION_OWNERSHIP_COMPATIBILITY.minProtocol &&\n\t\t\t!isLegacyDaemonInspection(command)\n\t\t) {\n\t\t\tthrow new Error(\n\t\t\t\t`Native daemon commands require protocol ${CANONICAL_SESSION_OWNERSHIP_COMPATIBILITY.minProtocol} canonical session ownership and bounded invocation output; only passive inspection is available`,\n\t\t\t);\n\t\t}\n\t\tlet admission: SupervisorPromptAdmission | undefined;\n\t\tif ((command.type === \"prompt\" || command.type === \"prompt_and_wait\") && command.admissionId !== undefined) {\n\t\t\tif (typeof command.activeSessionId !== \"string\" || typeof command.admissionId !== \"string\") {\n\t\t\t\tthrow new Error(\"Prompt admission requires string activeSessionId and admissionId\");\n\t\t\t}\n\t\t\tif (command.admissionId === \"\") throw new Error(\"admissionId must not be empty\");\n\t\t\tconst admissions = this.promptAdmissionsFor(client);\n\t\t\tconst key = this.promptAdmissionKey(command.activeSessionId, command.admissionId);\n\t\t\tif (admissions.has(key)) {\n\t\t\t\tthrow new Error(`Prompt admission id is already in use: ${command.admissionId}`);\n\t\t\t}\n\t\t\tadmission = {\n\t\t\t\tclient,\n\t\t\t\tactiveSessionId: command.activeSessionId,\n\t\t\t\tpublicAdmissionId: command.admissionId,\n\t\t\t\tworkerAdmissionId: `supervisor-admission:${randomUUID()}`,\n\t\t\t\tstatus: \"waiting\",\n\t\t\t\tcontroller: new AbortController(),\n\t\t\t};\n\t\t\tadmissions.set(key, admission);\n\t\t}\n\t\treturn {\n\t\t\tcommand,\n\t\t\tenvelopeClientId: envelope.clientId ?? client.id,\n\t\t\tprotocolVersion: envelope.protocol.version,\n\t\t\tadmission,\n\t\t};\n\t}\n\n\tprivate isWorkerLedgerEnvelope(line: string): boolean {\n\t\ttry {\n\t\t\tconst parsed: unknown = JSON.parse(line);\n\t\t\treturn isDaemonCommandEnvelope(parsed) && parsed.command.type === \"rlm_ledger_mutate\";\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate async handleLine(client: DaemonSocketClient, line: string): Promise<void> {\n\t\ttry {\n\t\t\tthis.assertSupervisorServing(this.shuttingDown && this.isWorkerLedgerEnvelope(line));\n\t\t} catch (error) {\n\t\t\tthis.write(client, failure(salvageDaemonCommandId(line), \"dispatch\", error, serializeDaemonError(error)));\n\t\t\treturn;\n\t\t}\n\t\tlet preParsed: ReturnType<DaemonSupervisor[\"parseCommandAndRegisterPromptAdmission\"]>;\n\t\ttry {\n\t\t\tpreParsed = this.parseCommandAndRegisterPromptAdmission(client, line);\n\t\t} catch (error) {\n\t\t\tthis.write(client, failure(salvageDaemonCommandId(line), \"parse\", error));\n\t\t\treturn;\n\t\t}\n\t\tconst command = preParsed.command;\n\t\tconst parsedAdmission = preParsed.admission;\n\t\tif (command.type === \"cancel_prompt_admission\" && this.updateRestartPhase !== undefined) {\n\t\t\tthis.write(client, failure(command.id, command.type, \"Daemon is preparing an update restart\"));\n\t\t\treturn;\n\t\t}\n\t\tconst cancellationAdmission =\n\t\t\tcommand.type === \"cancel_prompt_admission\"\n\t\t\t\t? this.getPromptAdmission(client, command.activeSessionId, command.admissionId)\n\t\t\t\t: undefined;\n\t\tif (cancellationAdmission?.status === \"waiting\" && !cancellationAdmission.worker) {\n\t\t\tcancellationAdmission.status = \"cancelled\";\n\t\t\tcancellationAdmission.controller.abort();\n\t\t}\n\t\ttry {\n\t\t\tawait waitForPromptAdmission(this.ready, parsedAdmission?.controller.signal);\n\t\t} catch (error) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, error));\n\t\t\treturn;\n\t\t}\n\t\tconst envelopeClientId = preParsed.envelopeClientId;\n\t\tif (envelopeClientId) {\n\t\t\tthis.protocolClientIds.set(client, envelopeClientId);\n\t\t\tclient.id = envelopeClientId;\n\t\t}\n\t\tthis.cancelOwnedWorkerCleanup(client.id);\n\t\tif (!DAEMON_COMMAND_TYPES.has(command.type)) {\n\t\t\tthis.write(client, failure(command.id, command.type, `Unknown daemon command: ${command.type}`));\n\t\t\treturn;\n\t\t}\n\t\tif (\n\t\t\tcommand.type === \"get_session_tree\" &&\n\t\t\tpreParsed.protocolVersion < DAEMON_COMMAND_COMPATIBILITY.get_session_tree.minProtocol\n\t\t) {\n\t\t\tthis.write(\n\t\t\t\tclient,\n\t\t\t\tfailure(\n\t\t\t\t\tcommand.id,\n\t\t\t\t\tcommand.type,\n\t\t\t\t\t`get_session_tree requires client protocol ${DAEMON_COMMAND_COMPATIBILITY.get_session_tree.minProtocol} or newer`,\n\t\t\t\t),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\ttry {\n\t\t\tawait waitForPromptAdmission(\n\t\t\t\tthis.assertServingCurrentOwnership(command.type === \"rlm_ledger_mutate\"),\n\t\t\t\tparsedAdmission?.controller.signal,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, error));\n\t\t\treturn;\n\t\t}\n\n\t\tconst mutation = isDaemonMutatingCommand(command);\n\t\tconst journalIdentity =\n\t\t\tenvelopeClientId && command.id && mutation ? { clientId: envelopeClientId, commandId: command.id } : undefined;\n\t\tconst existing = journalIdentity\n\t\t\t? this.commandJournal.lookup(journalIdentity.clientId, journalIdentity.commandId)\n\t\t\t: undefined;\n\t\tif (existing?.status === \"complete\") {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, existing.response);\n\t\t\treturn;\n\t\t}\n\t\tif (existing?.status === \"pending\" && journalIdentity) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(\n\t\t\t\tclient,\n\t\t\t\tfailure(command.id, command.type, \"The previous command result is uncertain and was not replayed\", {\n\t\t\t\t\tcode: \"command_result_uncertain\",\n\t\t\t\t\t...journalIdentity,\n\t\t\t\t}),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tconst phase = this.updateRestartPhase;\n\t\tconst restartRejected =\n\t\t\tphase === \"draining\"\n\t\t\t\t? !UPDATE_RESTART_DRAIN_COMMANDS.has(command.type)\n\t\t\t\t: phase !== undefined && !(phase === \"prepared\" && command.type === \"shutdown\");\n\t\tif (restartRejected && mutation && command.type !== \"rlm_ledger_mutate\") {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, \"Daemon is preparing an update restart\"));\n\t\t\treturn;\n\t\t}\n\t\tif (mutation && !UPDATE_RESTART_DRAIN_COMMANDS.has(command.type)) {\n\t\t\tconst idleEvictionFence = this.idleEvictionFence;\n\t\t\tif (idleEvictionFence) {\n\t\t\t\tawait idleEvictionFence;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertServingCurrentOwnership(command.type === \"rlm_ledger_mutate\");\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\t\tthis.write(client, failure(command.id, command.type, error, serializeDaemonError(error)));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif (journalIdentity) {\n\t\t\tlet admitted: CommandJournalBeginResult;\n\t\t\ttry {\n\t\t\t\tadmitted = this.commandJournal.begin(journalIdentity.clientId, journalIdentity.commandId, command.type);\n\t\t\t} catch (error) {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(client, failure(command.id, command.type, error, serializeDaemonError(error)));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (admitted.status === \"complete\") {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(client, admitted.response);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (admitted.status === \"pending\") {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(\n\t\t\t\t\tclient,\n\t\t\t\t\tfailure(command.id, command.type, \"The previous command result is uncertain and was not replayed\", {\n\t\t\t\t\t\tcode: \"command_result_uncertain\",\n\t\t\t\t\t\t...journalIdentity,\n\t\t\t\t\t}),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\n\t\t// Attach is intentionally read-only and is not fence-gated. If eviction wins\n\t\t// the race, attach fails cleanly with \"Session worker is not connected\" and\n\t\t// the client retries through the saved-session path instead of mutating state.\n\t\tif (mutation) this.mutationDrain.begin();\n\t\tlet commandCompleted = false;\n\t\ttry {\n\t\t\tconst response = await this.handleCommand(client, command, cancellationAdmission);\n\t\t\tcommandCompleted = true;\n\t\t\tif (response) {\n\t\t\t\tif (journalIdentity) {\n\t\t\t\t\tawait this.assertCurrentOwnership();\n\t\t\t\t\tthis.commandJournal.recordResult(journalIdentity.clientId, journalIdentity.commandId, response);\n\t\t\t\t}\n\t\t\t\tthis.write(client, response);\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Supervisor command ${command.type} failed: ${error instanceof Error ? error.stack : String(error)}`);\n\t\t\tlet response = failure(command.id, command.type, error, serializeDaemonError(error));\n\t\t\t// A completed command must not get a conflicting failure result after uncertain result persistence.\n\t\t\tif (journalIdentity && !commandCompleted && !isSupervisorGenerationStale(error)) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertCurrentOwnership();\n\t\t\t\t\tthis.commandJournal.recordResult(journalIdentity.clientId, journalIdentity.commandId, response);\n\t\t\t\t} catch (persistenceError) {\n\t\t\t\t\tconst failureError =\n\t\t\t\t\t\tpersistenceError === error\n\t\t\t\t\t\t\t? error\n\t\t\t\t\t\t\t: new AggregateError(\n\t\t\t\t\t\t\t\t\t[error, persistenceError],\n\t\t\t\t\t\t\t\t\t\"Supervisor command and result persistence failed\",\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\tresponse = failure(command.id, command.type, failureError, serializeDaemonError(failureError));\n\t\t\t\t}\n\t\t\t}\n\t\t\tthis.write(client, response);\n\t\t} finally {\n\t\t\tif (mutation) this.mutationDrain.end();\n\t\t}\n\t}\n\n\tprivate async handleCommand(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: DaemonCommand,\n\t\tcancellationAdmission?: SupervisorPromptAdmission,\n\t): Promise<DaemonResponse | undefined> {\n\t\tswitch (command.type) {\n\t\t\tcase \"cancel_prompt_admission\": {\n\t\t\t\tconst admission =\n\t\t\t\t\tcancellationAdmission ?? this.getPromptAdmission(client, command.activeSessionId, command.admissionId);\n\t\t\t\tif (!admission) return success(command.id, command.type, { status: \"unknown\" as const });\n\t\t\t\tif (admission.status === \"owned\") return success(command.id, command.type, { status: \"owned\" as const });\n\t\t\t\t// A definitive cancellation never downgrades to unknown/waiting.\n\t\t\t\tif (admission.status === \"cancelled\") {\n\t\t\t\t\treturn success(command.id, command.type, { status: \"cancelled\" as const });\n\t\t\t\t}\n\t\t\t\tif (!admission.worker || !admission.workerActiveSessionId) {\n\t\t\t\t\tadmission.status = \"cancelled\";\n\t\t\t\t\tadmission.controller.abort();\n\t\t\t\t\treturn success(command.id, command.type, { status: \"cancelled\" as const });\n\t\t\t\t}\n\t\t\t\tconst response = await this.forwardToWorker(admission.worker, {\n\t\t\t\t\t...command,\n\t\t\t\t\tactiveSessionId: admission.workerActiveSessionId,\n\t\t\t\t\tadmissionId: admission.workerAdmissionId,\n\t\t\t\t});\n\t\t\t\tconst status =\n\t\t\t\t\tresponse.success && response.data && typeof response.data === \"object\" && \"status\" in response.data\n\t\t\t\t\t\t? (response.data as { status: \"cancelled\" | \"owned\" | \"unknown\" }).status\n\t\t\t\t\t\t: \"unknown\";\n\t\t\t\t// Re-read: a socket close may have cancelled during the round-trip (cast widens TS's pre-await narrowing).\n\t\t\t\tconst current = (admission as SupervisorPromptAdmission).status;\n\t\t\t\tif (status === \"owned\") admission.status = \"owned\";\n\t\t\t\telse if (status === \"cancelled\") admission.status = \"cancelled\";\n\t\t\t\telse if (current !== \"cancelled\") admission.status = \"waiting\";\n\t\t\t\treturn { ...response, id: command.id };\n\t\t\t}\n\t\t\tcase \"ack_result\":\n\t\t\t\tthis.commandJournal.acknowledge(client.id, command.commandId);\n\t\t\t\treturn undefined;\n\t\t\tcase \"list\":\n\t\t\t\treturn this.handleList(client, command);\n\t\t\tcase \"roster_subscribe\":\n\t\t\t\tclient.rosterSubscribed = true;\n\t\t\t\treturn success(command.id, command.type, { roster: this.rosterEntriesForClient() });\n\t\t\tcase \"roster_unsubscribe\":\n\t\t\t\tclient.rosterSubscribed = false;\n\t\t\t\tclient.rosterResyncPending = false;\n\t\t\t\treturn success(command.id, command.type);\n\t\t\tcase \"list_agent_peers\": {\n\t\t\t\tconst requester = [...this.workers.values()].find(\n\t\t\t\t\t(worker) => worker.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t);\n\t\t\t\tif (!requester) throw new Error(\"Worker authentication failed\");\n\t\t\t\tconst peers = [...this.workers.values()]\n\t\t\t\t\t.filter(\n\t\t\t\t\t\t(worker) =>\n\t\t\t\t\t\t\tworker !== requester &&\n\t\t\t\t\t\t\tthis.isLiveWorker(worker) &&\n\t\t\t\t\t\t\tworker.descriptor.lifecycle === \"ready\" &&\n\t\t\t\t\t\t\tworker.client !== undefined,\n\t\t\t\t\t)\n\t\t\t\t\t.flatMap((worker) => {\n\t\t\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\t\t\treturn root ? [this.agentPeerSummary(sessionSummaryFromRosterEntry(root))] : [];\n\t\t\t\t\t});\n\t\t\t\treturn success(command.id, command.type, { peers });\n\t\t\t}\n\t\t\tcase \"rlm_ledger_mutate\": {\n\t\t\t\tconst worker = [...this.workers.values()].find(\n\t\t\t\t\t(candidate) => candidate.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t);\n\t\t\t\tif (\n\t\t\t\t\t!worker ||\n\t\t\t\t\t!command.workerInstanceId ||\n\t\t\t\t\tworker.descriptor.workerInstanceId !== command.workerInstanceId ||\n\t\t\t\t\tworker.compatibilityError ||\n\t\t\t\t\tthis.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId) !== \"current\"\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker authentication failed for RLM ledger mutation\");\n\t\t\t\t}\n\t\t\t\tconst owner = this.assertRlmJournalAdmission();\n\t\t\t\tconst scope = this.rlmJournalScope(worker.descriptor.sessionDir ?? this.defaultSessionConfig.sessionDir);\n\t\t\t\tif (canonicalSessionPath(scope.journalPath) !== owner.ledgerPath) {\n\t\t\t\t\tthrow new Error(\"Worker RLM ledger family is not owned by this supervisor\");\n\t\t\t\t}\n\t\t\t\tawait this.rlmSpawnLedger().mutate(command.mutation);\n\t\t\t\treturn success(command.id, command.type);\n\t\t\t}\n\t\t\tcase \"get_direct_worker_transport\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tif (match.worker.descriptor.ownerClientId !== undefined) {\n\t\t\t\t\tthrow new Error(\"Direct transport is unavailable for client-owned workers\");\n\t\t\t\t}\n\t\t\t\tconst ticket = await this.issuePeerTransport(match.worker, match.summary);\n\t\t\t\treturn success(command.id, command.type, ticket);\n\t\t\t}\n\t\t\tcase \"list_saved_sessions\":\n\t\t\t\treturn this.handleSavedSessionList(client, command);\n\t\t\tcase \"create\": {\n\t\t\t\tconst worker = await this.createOrReuseWorker(this.protocolClientId(client), command);\n\t\t\t\tconst requestedSummary = command.sessionPath\n\t\t\t\t\t? this.findSummaryInWorker(worker, command.sessionPath)\n\t\t\t\t\t: undefined;\n\t\t\t\tif (\n\t\t\t\t\trequestedSummary &&\n\t\t\t\t\t(requestedSummary.activeSessionId ?? requestedSummary.id) !== worker.descriptor.rootActiveSessionId\n\t\t\t\t) {\n\t\t\t\t\t// A create forwarded to a recovering worker still surfaces an opaque lifecycle error.\n\t\t\t\t\tconst response = await this.forwardToWorker(worker, withoutSupervisorCreateFields(command));\n\t\t\t\t\tif (response.success && isSessionSummary(response.data)) {\n\t\t\t\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(response.data), worker);\n\t\t\t\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t\t\t\t}\n\t\t\t\t\treturn responseWithId(response, command.id);\n\t\t\t\t}\n\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\tif (!root) {\n\t\t\t\t\tthrow new Error(\"Session worker started without a root session\");\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"create\", this.publicSummary(worker, sessionSummaryFromRosterEntry(root)));\n\t\t\t}\n\t\t\tcase \"attach\": {\n\t\t\t\tconst attached = await this.attachClient(client, command);\n\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\tconst transcript = attached.transcript;\n\t\t\t\t\tif (!transcript) {\n\t\t\t\t\t\tthrow new Error(\"Session worker did not provide a snapshot transcript\");\n\t\t\t\t\t}\n\t\t\t\t\tconst streamedResult = this.createStreamedAttachResult(attached.result, transcript);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tthis.write(client, success(command.id, \"attach\", streamedResult));\n\t\t\t\t\t\tvoid this.streamSnapshot(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\t\tstreamedResult,\n\t\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\t\t\"attach\",\n\t\t\t\t\t\t\tattached.releaseTranscript,\n\t\t\t\t\t\t).catch((error) =>\n\t\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t\t`Failed to stream attach snapshot for ${streamedResult.activeSessionId}: ${String(error)}`,\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tattached.releaseTranscript?.();\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\treturn undefined;\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"attach\", attached.result);\n\t\t\t}\n\t\t\tcase \"reattach\": {\n\t\t\t\tconst target = await this.findWorkerForClient(client, command.targetActiveSessionId);\n\t\t\t\tconst targetActiveSessionId = target.summary.activeSessionId ?? target.summary.id;\n\t\t\t\tif (targetActiveSessionId === command.activeSessionId) {\n\t\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\t\t\tdetachingSessions?.delete(command.targetActiveSessionId);\n\t\t\t\t\tdetachingSessions?.delete(targetActiveSessionId);\n\t\t\t\t\treturn success(command.id, command.type, { cancelled: false });\n\t\t\t\t}\n\t\t\t\tconst targetWasAttached = client.attachedActiveSessionIds.has(targetActiveSessionId);\n\t\t\t\tconst releaseSnapshotReservation = this.reserveSnapshotStream(client, targetActiveSessionId);\n\t\t\t\tlet releaseTranscript: (() => void) | undefined;\n\t\t\t\tclient.attachedActiveSessionIds.add(targetActiveSessionId);\n\t\t\t\ttry {\n\t\t\t\t\tconst attached = await this.attachClient(client, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\t\tactiveSessionId: targetActiveSessionId,\n\t\t\t\t\t});\n\t\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\t\t\tdetachingSessions?.delete(command.targetActiveSessionId);\n\t\t\t\t\tdetachingSessions?.delete(targetActiveSessionId);\n\t\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\t\tconst transcript =\n\t\t\t\t\t\t\tattached.transcript ?? this.getOrCreateTranscriptCache(attached.worker, attached.result);\n\t\t\t\t\t\treleaseTranscript = attached.releaseTranscript;\n\t\t\t\t\t\tconst streamedResult = this.createStreamedAttachResult(attached.result, transcript);\n\t\t\t\t\t\tthis.write(client, success(command.id, command.type, streamedResult));\n\t\t\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\t\t\tconst streaming = this.streamSnapshot(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\t\tstreamedResult,\n\t\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\t\t\"replacement\",\n\t\t\t\t\t\t\treleaseTranscript,\n\t\t\t\t\t\t\treleaseSnapshotReservation,\n\t\t\t\t\t\t);\n\t\t\t\t\t\treleaseTranscript = undefined;\n\t\t\t\t\t\tvoid streaming.catch((error) =>\n\t\t\t\t\t\t\tthis.log(`Failed to stream reattach snapshot for ${targetActiveSessionId}: ${String(error)}`),\n\t\t\t\t\t\t);\n\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t}\n\t\t\t\t\tthis.write(client, success(command.id, command.type, attached.result));\n\t\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\t\treleaseSnapshotReservation();\n\t\t\t\t\treturn undefined;\n\t\t\t\t} catch (error) {\n\t\t\t\t\treleaseTranscript?.();\n\t\t\t\t\tif (!targetWasAttached) {\n\t\t\t\t\t\tthis.detachClient(client, targetActiveSessionId);\n\t\t\t\t\t}\n\t\t\t\t\treleaseSnapshotReservation();\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t}\n\t\t\tcase \"acquire_session_input_pause\": {\n\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions.get(client);\n\t\t\t\tif (detachingSessions?.has(command.activeSessionId)) {\n\t\t\t\t\tthrow new Error(`Session is detaching: ${command.activeSessionId}`);\n\t\t\t\t}\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst activeSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\t\t\tif (detachingSessions?.has(command.activeSessionId) || detachingSessions?.has(activeSessionId)) {\n\t\t\t\t\tthrow new Error(`Session is detaching: ${command.activeSessionId}`);\n\t\t\t\t}\n\t\t\t\tconst ownerClientId = this.protocolClientId(client);\n\t\t\t\tconst connectionId = this.connectionIds.get(client);\n\t\t\t\tif (!connectionId) throw new Error(\"Daemon client connection identity is unavailable\");\n\t\t\t\tconst acquisitionEpoch = this.sessionInputPauseEpochs.get(client) ?? 0;\n\t\t\t\tconst existing = [...this.sessionInputPauses.values()].find(\n\t\t\t\t\t(entry) =>\n\t\t\t\t\t\tentry.owner === client &&\n\t\t\t\t\t\tentry.worker === match.worker &&\n\t\t\t\t\t\tentry.activeSessionId === activeSessionId &&\n\t\t\t\t\t\tentry.leaseKey === command.leaseKey &&\n\t\t\t\t\t\t!entry.releaseTask,\n\t\t\t\t);\n\t\t\t\tif (existing) {\n\t\t\t\t\treturn success(command.id, command.type, { pauseId: existing.pauseId });\n\t\t\t\t}\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, {\n\t\t\t\t\t...command,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tleaseKey: JSON.stringify([connectionId, ownerClientId, command.leaseKey]),\n\t\t\t\t});\n\t\t\t\tif (!response.success) return response;\n\t\t\t\tconst pauseId = (response.data as { pauseId?: unknown } | undefined)?.pauseId;\n\t\t\t\tif (typeof pauseId !== \"string\") throw new Error(\"Worker returned an invalid session input pause id\");\n\t\t\t\tif (!this.clients.has(client) || (this.sessionInputPauseEpochs.get(client) ?? 0) !== acquisitionEpoch) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst release = await this.forwardToWorker(match.worker, {\n\t\t\t\t\t\t\tid: randomUUID(),\n\t\t\t\t\t\t\ttype: \"release_session_input_pause\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tpauseId,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tif (!release.success) throw new Error(release.error);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tmatch.worker.client?.close();\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tthrow new Error(\"Session input pause acquisition was invalidated before completion\");\n\t\t\t\t}\n\t\t\t\tthis.sessionInputPauses.set(pauseId, {\n\t\t\t\t\towner: client,\n\t\t\t\t\tworker: match.worker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\trequestedActiveSessionId: command.activeSessionId,\n\t\t\t\t\tleaseKey: command.leaseKey,\n\t\t\t\t\tpauseId,\n\t\t\t\t});\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"release_session_input_pause\": {\n\t\t\t\tconst entry = this.sessionInputPauses.get(command.pauseId);\n\t\t\t\tif (!entry) return success(command.id, command.type);\n\t\t\t\tif (entry.owner !== client) {\n\t\t\t\t\tthrow new Error(`Session input pause is owned by another client: ${command.pauseId}`);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\tcommand.activeSessionId !== entry.activeSessionId &&\n\t\t\t\t\tcommand.activeSessionId !== entry.requestedActiveSessionId\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(`Session input pause belongs to another session: ${command.pauseId}`);\n\t\t\t\t}\n\t\t\t\tconst releaseTask =\n\t\t\t\t\tentry.releaseTask ??\n\t\t\t\t\tthis.forwardToWorker(entry.worker, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\tactiveSessionId: entry.activeSessionId,\n\t\t\t\t\t});\n\t\t\t\tentry.releaseTask = releaseTask;\n\t\t\t\ttry {\n\t\t\t\t\tconst response = await releaseTask;\n\t\t\t\t\tif (response.success && this.sessionInputPauses.get(command.pauseId) === entry) {\n\t\t\t\t\t\tthis.sessionInputPauses.delete(command.pauseId);\n\t\t\t\t\t}\n\t\t\t\t\treturn response;\n\t\t\t\t} finally {\n\t\t\t\t\tif (this.sessionInputPauses.get(command.pauseId) === entry && entry.releaseTask === releaseTask) {\n\t\t\t\t\t\tentry.releaseTask = undefined;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tcase \"detach\": {\n\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions.get(client) ?? new Set<string>();\n\t\t\t\tthis.detachingInputPauseSessions.set(client, detachingSessions);\n\t\t\t\tif (command.activeSessionId) detachingSessions.add(command.activeSessionId);\n\t\t\t\telse for (const activeSessionId of client.attachedActiveSessionIds) detachingSessions.add(activeSessionId);\n\t\t\t\tthis.sessionInputPauseEpochs.set(client, (this.sessionInputPauseEpochs.get(client) ?? 0) + 1);\n\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\tawait this.releaseClientSessionInputPauses(client, command.activeSessionId, true);\n\t\t\t\treturn success(command.id, \"detach\");\n\t\t\t}\n\t\t\tcase \"complete_owned_session\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tif (match.worker.descriptor.ownerClientId !== this.protocolClientId(client)) {\n\t\t\t\t\tthrow new Error(\"Session is not owned by this client\");\n\t\t\t\t}\n\t\t\t\tif (match.worker.ownerCleanupTimer) {\n\t\t\t\t\tclearTimeout(match.worker.ownerCleanupTimer);\n\t\t\t\t\tmatch.worker.ownerCleanupTimer = undefined;\n\t\t\t\t}\n\t\t\t\tawait this.stopWorker(match.worker, true);\n\t\t\t\treturn success(command.id, command.type);\n\t\t\t}\n\t\t\tcase \"promote_owned_session\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\treturn success(command.id, command.type, this.publicSummary(match.worker, match.summary));\n\t\t\t}\n\t\t\tcase \"retry_worker\": {\n\t\t\t\tconst direct = [...this.workers.values()].find(\n\t\t\t\t\t(worker) =>\n\t\t\t\t\t\tworker.descriptor.rootActiveSessionId === command.activeSessionId ||\n\t\t\t\t\t\tworker.descriptor.rootSessionId === command.activeSessionId,\n\t\t\t\t);\n\t\t\t\tconst worker = direct ?? (await this.findWorkerForClient(client, command.activeSessionId)).worker;\n\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\tif ((this.workerStopCounts?.get(worker) ?? 0) > 0) {\n\t\t\t\t\tthrow new Error(\"Session worker is stopping; retry after it finishes\");\n\t\t\t\t}\n\t\t\t\tworker.intentionalStop = false;\n\t\t\t\tworker.descriptor.stopRequestedAt = undefined;\n\t\t\t\tworker.descriptor.archiveOnStop = undefined;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tawait this.recoverWorker(worker);\n\t\t\t\tif (this.workers.get(worker.descriptor.workerId)?.descriptor.lifecycle !== \"ready\") {\n\t\t\t\t\tthrow new Error(worker.descriptor.lastError ?? \"Session worker recovery failed\");\n\t\t\t\t}\n\t\t\t\tconst summary = worker.summaries.get(worker.descriptor.rootActiveSessionId);\n\t\t\t\treturn success(command.id, command.type, summary ? this.publicSummary(worker, summary) : undefined);\n\t\t\t}\n\t\t\tcase \"restart\":\n\t\t\t\tsetImmediate(() => void this.shutdown(0, false, true, false, \"update\"));\n\t\t\t\treturn success(command.id, command.type);\n\t\t\tcase \"shutdown\":\n\t\t\t\tsetImmediate(() => void this.shutdown(0, true, false, command.force === true, \"shutdown\"));\n\t\t\t\treturn success(command.id, \"shutdown\");\n\t\t\tcase \"prepare_update_restart\": {\n\t\t\t\tconst manifest = await this.prepareUpdateRestart();\n\t\t\t\treturn success(command.id, \"prepare_update_restart\", manifest);\n\t\t\t}\n\t\t\tcase \"agent_messages_status\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst first = [...this.workers.values()].find((worker) => this.isLiveWorker(worker) && worker.client);\n\t\t\t\tif (!first) {\n\t\t\t\t\treturn success(command.id, command.type, { paused: false, limits: {} });\n\t\t\t\t}\n\t\t\t\treturn this.forwardToWorker(first, command);\n\t\t\t}\n\t\t\tcase \"agent_messages_pause\":\n\t\t\tcase \"agent_messages_resume\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst responses = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter((worker) => this.isLiveWorker(worker) && worker.client)\n\t\t\t\t\t\t.map((worker) => this.forwardToWorker(worker, command)),\n\t\t\t\t);\n\t\t\t\tconst failed = responses.find((response) => !response.success);\n\t\t\t\treturn failed ?? success(command.id, command.type, responses.find((response) => response.success)?.data);\n\t\t\t}\n\t\t\tcase \"cron_list\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst jobs = new Map<string, AgentCronJob>();\n\t\t\t\tconst responses = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map((worker) =>\n\t\t\t\t\t\t\tthis.forwardToWorker(worker, command, 5000).catch((error: unknown) =>\n\t\t\t\t\t\t\t\tfailure(command.id, command.type, error, serializeDaemonError(error)),\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t),\n\t\t\t\t);\n\t\t\t\tfor (const response of responses) {\n\t\t\t\t\tif (!response.success) {\n\t\t\t\t\t\tthis.log(`Could not list scheduled jobs from a worker: ${response.error}`);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tfor (const job of cronJobsFromResponse(response)) {\n\t\t\t\t\t\tjobs.set(job.id, job);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tfor (const { job } of await this.collectPassiveScheduledJobs(command.includeInactive === true)) {\n\t\t\t\t\tif (!jobs.has(job.id)) jobs.set(job.id, job);\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"cron_list\", { jobs: sortCronJobs([...jobs.values()]) });\n\t\t\t}\n\t\t\tcase \"heartbeats_list\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst workers = [...this.workers.values()].filter(\n\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.descriptor.lifecycle !== \"failed\",\n\t\t\t\t);\n\t\t\t\tconst heartbeats = new Map<string, AgentConnectionHeartbeat>();\n\t\t\t\tconst snapshots: Array<{ heartbeats?: AgentConnectionHeartbeat[]; response?: DaemonResponse }> =\n\t\t\t\t\tawait Promise.all(\n\t\t\t\t\t\tworkers.map(async (worker) => {\n\t\t\t\t\t\t\tif (worker.client && worker.descriptor.lifecycle === \"ready\") {\n\t\t\t\t\t\t\t\tconst response = await this.forwardToWorker(worker, command, 5000).catch((error: unknown) =>\n\t\t\t\t\t\t\t\t\tfailure(command.id, command.type, error, serializeDaemonError(error)),\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tif (response.success) {\n\t\t\t\t\t\t\t\t\tconst snapshot = heartbeatsFromResponse(response);\n\t\t\t\t\t\t\t\t\tworker.heartbeatSnapshot = snapshot;\n\t\t\t\t\t\t\t\t\tworker.heartbeatSnapshotStale = false;\n\t\t\t\t\t\t\t\t\treturn { heartbeats: snapshot };\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tthis.log(`Could not list heartbeats from a worker: ${response.error}`);\n\t\t\t\t\t\t\t\tif (worker.heartbeatSnapshot === undefined || worker.heartbeatSnapshotStale === true) {\n\t\t\t\t\t\t\t\t\treturn { response };\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (worker.heartbeatSnapshot !== undefined && worker.heartbeatSnapshotStale !== true) {\n\t\t\t\t\t\t\t\treturn { heartbeats: worker.heartbeatSnapshot };\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst state =\n\t\t\t\t\t\t\t\tworker.descriptor.lifecycle === \"ready\" ? \"disconnected\" : worker.descriptor.lifecycle;\n\t\t\t\t\t\t\tconst error = new Error(`Cannot list heartbeats while session worker is ${state}`);\n\t\t\t\t\t\t\treturn { response: failure(command.id, command.type, error, serializeDaemonError(error)) };\n\t\t\t\t\t\t}),\n\t\t\t\t\t);\n\t\t\t\tconst failed = snapshots.find((snapshot) => snapshot.response)?.response;\n\t\t\t\tif (failed) {\n\t\t\t\t\treturn failed;\n\t\t\t\t}\n\t\t\t\tfor (const snapshot of snapshots) {\n\t\t\t\t\tfor (const heartbeat of snapshot.heartbeats ?? []) {\n\t\t\t\t\t\theartbeats.set(heartbeat.job.id, heartbeat);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Passivated sessions keep their armed heartbeats; no worker can list them.\n\t\t\t\tfor (const { job, info } of await this.collectPassiveScheduledJobs()) {\n\t\t\t\t\tif (!isHeartbeatCronJob(job) || heartbeats.has(job.id)) continue;\n\t\t\t\t\theartbeats.set(job.id, {\n\t\t\t\t\t\tjob,\n\t\t\t\t\t\t...(info.name !== undefined ? { sessionName: info.name } : {}),\n\t\t\t\t\t\t...(info.firstMessage !== undefined ? { firstMessage: info.firstMessage } : {}),\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"heartbeats_list\", { heartbeats: [...heartbeats.values()] });\n\t\t\t}\n\t\t\tcase \"heartbeat_manage\": {\n\t\t\t\tconst cachedWorker = [...this.workers.values()].find((worker) =>\n\t\t\t\t\tworker.heartbeatSnapshot?.some(\n\t\t\t\t\t\t(heartbeat) =>\n\t\t\t\t\t\t\theartbeat.job.id === command.jobId && heartbeat.job.activeSessionId === command.activeSessionId,\n\t\t\t\t\t),\n\t\t\t\t);\n\t\t\t\tif (!cachedWorker) {\n\t\t\t\t\t// Passive jobs are managed against their durable store; no wake just to flip a status.\n\t\t\t\t\tconst passive = (await this.collectPassiveScheduledJobs()).find(\n\t\t\t\t\t\t({ job }) => job.id === command.jobId && job.activeSessionId === command.activeSessionId,\n\t\t\t\t\t);\n\t\t\t\t\tif (passive) {\n\t\t\t\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\t\t\t\tstore.registerSessionArtifact(\n\t\t\t\t\t\t\tpassive.info.id,\n\t\t\t\t\t\t\tgetSessionArtifactPathForFile(resolve(passive.info.path), passive.info.id),\n\t\t\t\t\t\t);\n\t\t\t\t\t\tconst heartbeat = store.manageHeartbeat(command.activeSessionId, command.jobId, command.action);\n\t\t\t\t\t\tif (heartbeat) {\n\t\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\t\treturn success(command.id, \"heartbeat_manage\", { heartbeat });\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tconst worker = cachedWorker ?? (await this.findWorkerForClient(client, command.activeSessionId)).worker;\n\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(worker, command);\n\t\t\t\tif (\n\t\t\t\t\tresponse.success &&\n\t\t\t\t\tresponse.data &&\n\t\t\t\t\ttypeof response.data === \"object\" &&\n\t\t\t\t\t\"heartbeat\" in response.data\n\t\t\t\t) {\n\t\t\t\t\tconst job = (response.data as { heartbeat?: AgentCronJob }).heartbeat;\n\t\t\t\t\tif (job && worker.heartbeatSnapshot) {\n\t\t\t\t\t\tconst existing = worker.heartbeatSnapshot.find((heartbeat) => heartbeat.job.id === job.id);\n\t\t\t\t\t\tconst remaining = worker.heartbeatSnapshot.filter((heartbeat) => heartbeat.job.id !== job.id);\n\t\t\t\t\t\tworker.heartbeatSnapshot =\n\t\t\t\t\t\t\tjob.status === \"active\" || job.status === \"paused\"\n\t\t\t\t\t\t\t\t? [...remaining, existing ? { ...existing, job } : { job }]\n\t\t\t\t\t\t\t\t: remaining;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"cron_add\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, command);\n\t\t\t\tif (response.success && command.promoteOwnedSession) {\n\t\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"cron_cancel\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst listed = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map(async (worker) => ({\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tresponse: await this.forwardToWorker(\n\t\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\t\t{ type: \"cron_list\", includeInactive: true },\n\t\t\t\t\t\t\t\t5000,\n\t\t\t\t\t\t\t).catch(() => undefined),\n\t\t\t\t\t\t})),\n\t\t\t\t);\n\t\t\t\tfor (const candidate of listed) {\n\t\t\t\t\tif (\n\t\t\t\t\t\tcandidate.response?.success &&\n\t\t\t\t\t\tcronJobsFromResponse(candidate.response).some((job) => job.id === command.jobId)\n\t\t\t\t\t) {\n\t\t\t\t\t\treturn this.forwardToWorker(candidate.worker, command);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tconst passive = (await this.collectPassiveScheduledJobs()).find(({ job }) => job.id === command.jobId);\n\t\t\t\tif (passive) {\n\t\t\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\t\t\tstore.registerSessionArtifact(\n\t\t\t\t\t\tpassive.info.id,\n\t\t\t\t\t\tgetSessionArtifactPathForFile(resolve(passive.info.path), passive.info.id),\n\t\t\t\t\t);\n\t\t\t\t\tconst job = store.cancel(command.jobId);\n\t\t\t\t\tif (job) {\n\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\treturn success(command.id, \"cron_cancel\", { job });\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tthrow new Error(`No cron job found: ${command.jobId}`);\n\t\t\t}\n\t\t\tcase \"cron_resume\": {\n\t\t\t\tif (command.activeSessionId !== undefined) {\n\t\t\t\t\tconst entry = this.roster().byActiveSessionId(command.activeSessionId);\n\t\t\t\t\tconst worker = entry?.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\t\t\tif (!worker || entry?.queuedChild) {\n\t\t\t\t\t\tthrow new Error(\"Cron resume requires an already bound destination runtime\");\n\t\t\t\t\t}\n\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(worker, command);\n\t\t\t\t}\n\t\t\t\tconst listed = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map(async (worker) => ({\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tresponse: await this.forwardToWorker(\n\t\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\t\t{ type: \"cron_list\", includeInactive: true },\n\t\t\t\t\t\t\t\t5000,\n\t\t\t\t\t\t\t).catch(() => undefined),\n\t\t\t\t\t\t})),\n\t\t\t\t);\n\t\t\t\tfor (const candidate of listed) {\n\t\t\t\t\tconst job = candidate.response?.success\n\t\t\t\t\t\t? cronJobsFromResponse(candidate.response).find((job) => job.id === command.jobId)\n\t\t\t\t\t\t: undefined;\n\t\t\t\t\tif (job) {\n\t\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, candidate.worker, job.activeSessionId);\n\t\t\t\t\t\treturn this.forwardToWorker(candidate.worker, command);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tthrow new Error(`Cron resume requires an already bound destination runtime for job: ${command.jobId}`);\n\t\t\t}\n\t\t\tcase \"heartbeat_get\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t}\n\t\t\tcase \"heartbeat_set\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, command);\n\t\t\t\tif (response.success && command.promoteOwnedSession) {\n\t\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"heartbeat_update\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t}\n\t\t\tcase \"rename_saved_session\": {\n\t\t\t\tconst target = await this.savedSessionNameReservationInput(command.sessionPath, command.name.trim());\n\t\t\t\treturn await this.withSessionNameReservation(target, async () => {\n\t\t\t\t\tawait this.assertSupervisorSavedSessionNameAvailable(command.sessionPath, target.name);\n\t\t\t\t\tif (!command.activeSessionId) {\n\t\t\t\t\t\tthis.assertRlmJournalAdmission();\n\t\t\t\t\t\tawait this.catalog.rename(command.sessionPath, command.name);\n\t\t\t\t\t\t// Third rename write point: an offline saved-session rename\n\t\t\t\t\t\t// changes the name the ledger carries for that child.\n\t\t\t\t\t\tawait this.rlmSpawnLedger().appendRenameByChildPath(command.sessionPath, target.name);\n\t\t\t\t\t\tconst entry = this.roster().bySessionFile(canonicalSessionPath(command.sessionPath));\n\t\t\t\t\t\tif (entry) {\n\t\t\t\t\t\t\tthis.writeRosterEntry({ ...entry, summary: { ...entry.summary, sessionName: target.name } });\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn success(command.id, command.type);\n\t\t\t\t\t}\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn await this.forwardToWorker(match.worker, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\tactiveSessionId: match.summary.activeSessionId ?? match.summary.id,\n\t\t\t\t\t});\n\t\t\t\t});\n\t\t\t}\n\t\t\tcase \"delete_saved_session\":\n\t\t\t\tif (!command.activeSessionId) {\n\t\t\t\t\tconst deletedPath = canonicalSessionPath(command.sessionPath);\n\t\t\t\t\tconst entry = this.roster().bySessionFile(deletedPath);\n\t\t\t\t\tif (entry?.summary.activeSessionId !== undefined) {\n\t\t\t\t\t\tthrow new Error(\"Cannot delete the currently active session\");\n\t\t\t\t\t}\n\t\t\t\t\tconst owner = this.findWorkerBySessionFile(command.sessionPath);\n\t\t\t\t\tif (owner) {\n\t\t\t\t\t\t// A client-owned worker's files are invisible to other clients: a foreign delete is an unknown target.\n\t\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, owner, command.sessionPath);\n\t\t\t\t\t\tif (owner.client && !this.isWorkerStopping(owner)) {\n\t\t\t\t\t\t\treturn this.forwardToWorker(owner, command);\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!(await this.reclaimStaleWorkerRegistration(owner))) {\n\t\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t\t`Session worker is ${this.effectiveWorkerState(owner)}; retry the delete once it is reachable`,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tthis.assertRlmJournalAdmission();\n\t\t\t\t\tawait tombstoneSavedSessionDelete(this.rlmSpawnLedger(), command.sessionPath, entry?.summary);\n\t\t\t\t\tconst result = await this.catalog.delete(command.sessionPath);\n\t\t\t\t\tif (result.ok && entry && this.roster().get(entry.agentId) === entry) {\n\t\t\t\t\t\tthis.roster().delete(entry.agentId);\n\t\t\t\t\t}\n\t\t\t\t\treturn success(command.id, command.type, result);\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t}\n\n\t\tif (command.type === \"send_message\" || command.type === \"send_result\") {\n\t\t\t// agentOrigin without fromActiveSessionId is trusted only at the direct socket-client boundary.\n\t\t\tconst source = command.fromActiveSessionId\n\t\t\t\t? await this.findWorkerForClient(client, command.fromActiveSessionId)\n\t\t\t\t: undefined;\n\t\t\tlet target: WorkerMatch;\n\t\t\ttry {\n\t\t\t\ttarget = await this.findWorkerForClient(client, command.targetActiveSessionId);\n\t\t\t} catch (error) {\n\t\t\t\tif (!(error instanceof Error) || !error.message.startsWith(\"Unknown active session:\")) throw error;\n\t\t\t\tconst cwd = source?.summary.cwd ?? this.defaultSessionConfig.cwd ?? process.cwd();\n\t\t\t\tlet sessionPath: string;\n\t\t\t\ttry {\n\t\t\t\t\tsessionPath = await this.catalog.resolve(\n\t\t\t\t\t\tcommand.targetActiveSessionId,\n\t\t\t\t\t\tcwd,\n\t\t\t\t\t\tsource?.worker.descriptor.sessionDir ?? this.defaultSessionConfig.sessionDir,\n\t\t\t\t\t);\n\t\t\t\t} catch (catalogError) {\n\t\t\t\t\t// Preserve selector ambiguity so a2a senders can distinguish it from\n\t\t\t\t\t// the original unknown-active-session lookup failure.\n\t\t\t\t\tif (catalogError instanceof Error && catalogError.message.startsWith(\"Ambiguous session selector\")) {\n\t\t\t\t\t\tthrow catalogError;\n\t\t\t\t\t}\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t\tif (source && command.agentOrigin === true) {\n\t\t\t\t\tconst targetInfo = await readSessionInfo(sessionPath);\n\t\t\t\t\tif (!targetInfo) throw new Error(`Unknown active session: ${command.targetActiveSessionId}`);\n\t\t\t\t\tassertAgentFamilyReach(\n\t\t\t\t\t\tthis.familyCatalogEntry(source.summary),\n\t\t\t\t\t\tthis.familyCatalogEntry(summaryForInactiveSession(targetInfo)),\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\tconst worker = await this.createOrReuseWorker(this.protocolClientId(client), {\n\t\t\t\t\ttype: \"create\",\n\t\t\t\t\tsessionPath,\n\t\t\t\t\tcontinueRecent: false,\n\t\t\t\t});\n\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\tconst summary =\n\t\t\t\t\tthis.findSummaryInWorker(worker, sessionPath) ??\n\t\t\t\t\t(root ? sessionSummaryFromRosterEntry(root) : undefined);\n\t\t\t\tif (!summary) throw new Error(\"Woken session worker has no target session\");\n\t\t\t\ttarget = { worker, summary };\n\t\t\t}\n\t\t\tconst targetActiveSessionId = target.summary.activeSessionId ?? target.summary.id;\n\t\t\tif (source && command.agentOrigin === true) {\n\t\t\t\tassertAgentFamilyReach(this.familyCatalogEntry(source.summary), this.familyCatalogEntry(target.summary));\n\t\t\t}\n\t\t\tif (source) {\n\t\t\t\tif ((source.summary.activeSessionId ?? source.summary.id) === targetActiveSessionId) {\n\t\t\t\t\tthrow new Error(\"Agent messaging cannot target the sending session\");\n\t\t\t\t}\n\t\t\t\tconst targetClient = this.requireAvailableWorkerClient(target.worker);\n\t\t\t\tconst response = await targetClient.requestWorker(\n\t\t\t\t\t{\n\t\t\t\t\t\t...(command.type === \"send_result\"\n\t\t\t\t\t\t\t? { type: \"worker_deliver_result\" as const, summary: command.summary, findings: command.findings }\n\t\t\t\t\t\t\t: { type: \"worker_deliver_message\" as const, message: command.message }),\n\t\t\t\t\t\ttargetActiveSessionId,\n\t\t\t\t\t\tsender: {\n\t\t\t\t\t\t\tactiveSessionId: source.summary.activeSessionId ?? source.summary.id,\n\t\t\t\t\t\t\tsessionId: source.summary.sessionId,\n\t\t\t\t\t\t\t...(source.summary.sessionName ? { sessionName: source.summary.sessionName } : {}),\n\t\t\t\t\t\t\truntimeKind: source.summary.runtimeKind ?? \"top-level\",\n\t\t\t\t\t\t\tclientId: client.id,\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\tWORKER_REQUEST_TIMEOUT_MS,\n\t\t\t\t);\n\t\t\t\treturn { ...response, id: command.id, command: command.type };\n\t\t\t}\n\t\t\treturn this.forwardToWorker(target.worker, { ...command, targetActiveSessionId });\n\t\t}\n\n\t\tif (!(\"activeSessionId\" in command) || typeof command.activeSessionId !== \"string\") {\n\t\t\tthrow new Error(`Supervisor cannot route daemon command: ${command.type}`);\n\t\t}\n\t\tconst admission =\n\t\t\t(command.type === \"prompt\" || command.type === \"prompt_and_wait\") && command.admissionId\n\t\t\t\t? this.getPromptAdmission(client, command.activeSessionId, command.admissionId)\n\t\t\t\t: undefined;\n\t\ttry {\n\t\t\tthrowIfAdmissionCancelled(admission);\n\t\t\tconst match = await waitForPromptAdmission(\n\t\t\t\tcommand.type === \"set_session_name\" && command.workerToken !== undefined\n\t\t\t\t\t? this.findWorker(\n\t\t\t\t\t\t\tcommand.activeSessionId,\n\t\t\t\t\t\t\t(worker) => worker.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t\t\t)\n\t\t\t\t\t: this.findWorkerForClient(client, command.activeSessionId),\n\t\t\t\tadmission?.controller.signal,\n\t\t\t);\n\t\t\tthrowIfAdmissionCancelled(admission);\n\t\t\tconst resolvedCommand = {\n\t\t\t\t...command,\n\t\t\t\tactiveSessionId: match.summary.activeSessionId ?? match.summary.id,\n\t\t\t\t...(admission ? { admissionId: admission.workerAdmissionId } : {}),\n\t\t\t} as DaemonCommand;\n\t\t\tif (admission) {\n\t\t\t\tadmission.worker = match.worker;\n\t\t\t\tadmission.workerActiveSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\t\t}\n\t\t\tconst isRootKill =\n\t\t\t\tcommand.type === \"kill\" &&\n\t\t\t\t(match.summary.activeSessionId ?? match.summary.id) === match.worker.descriptor.rootActiveSessionId;\n\t\t\tif (!isRootKill) {\n\t\t\t\tconst forward = async () => {\n\t\t\t\t\tconst response = await this.forwardToWorker(match.worker, resolvedCommand);\n\t\t\t\t\tif (admission && response.success) admission.status = \"owned\";\n\t\t\t\t\treturn response;\n\t\t\t\t};\n\t\t\t\tif (command.type === \"rename\" || command.type === \"set_session_name\") {\n\t\t\t\t\tconst reservation = this.summaryNameReservationInput(match.summary, command.name.trim());\n\t\t\t\t\treturn await this.withSessionNameReservation(reservation, async () => {\n\t\t\t\t\t\tawait this.assertSupervisorSessionNameAvailable(match.summary, reservation.name);\n\t\t\t\t\t\treturn forward();\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\treturn await forward();\n\t\t\t}\n\t\t\tthis.persistWorkerStopTombstone(match.worker, true);\n\t\t\tconst releaseStopOwnership = this.acquireWorkerStopOwnership(match.worker);\n\t\t\tlet response: DaemonResponse;\n\t\t\ttry {\n\t\t\t\tresponse = await this.forwardToWorker(match.worker, resolvedCommand);\n\t\t\t} finally {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.stopWorker(match.worker, true, false, true);\n\t\t\t\t} finally {\n\t\t\t\t\treleaseStopOwnership();\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn response;\n\t\t} finally {\n\t\t\tif (admission) this.deletePromptAdmission(admission);\n\t\t}\n\t}\n\n\tprivate async handleList(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"list\" }>,\n\t): Promise<DaemonResponse> {\n\t\tconst active: SessionSummary[] = [];\n\t\tconst activeByFile = new Map<string, SessionSummary>();\n\t\tlet busyClientOwnedSessionCount = 0;\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild) continue;\n\t\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\tif (worker === undefined) continue;\n\t\t\tconst summary = this.publicSummary(worker, sessionSummaryFromRosterEntry(entry));\n\t\t\tif (this.isVisibleWorker(worker)) {\n\t\t\t\tactive.push(summary);\n\t\t\t\tif (summary.sessionFile) activeByFile.set(canonicalSessionPath(summary.sessionFile), summary);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (summary.sessionFile) activeByFile.set(canonicalSessionPath(summary.sessionFile), summary);\n\t\t\tif (isSessionSummaryBusy(summary)) busyClientOwnedSessionCount += 1;\n\t\t\tif (command.includeClientOwned === true && this.isWorkerAccessibleToClient(client, worker)) {\n\t\t\t\tactive.push(summary);\n\t\t\t}\n\t\t}\n\t\tconst data = {\n\t\t\tsessions: active,\n\t\t\t...(command.includeClientOwned ? { busyClientOwnedSessionCount } : {}),\n\t\t};\n\t\tif (!command.all) {\n\t\t\treturn success(command.id, \"list\", data);\n\t\t}\n\t\tconst sessionDir = command.sessionDir ?? this.defaultSessionConfig.sessionDir;\n\t\tconst scanned = await this.catalog.list(command.cwd ? resolve(command.cwd) : undefined, sessionDir);\n\t\tconst cwd = command.cwd ? resolve(command.cwd) : undefined;\n\t\tconst merged: SessionSummary[] = [];\n\t\tconst servedRows = new Set(active);\n\t\tconst mergedActiveFiles = new Set<string>();\n\t\tconst scannedFiles = new Set<string>();\n\t\tfor (const info of scanned) {\n\t\t\tconst file = canonicalSessionPath(info.path);\n\t\t\tscannedFiles.add(file);\n\t\t\tconst workerRow = activeByFile.get(file);\n\t\t\tif (workerRow && servedRows.has(workerRow)) {\n\t\t\t\tmerged.push(workerRow);\n\t\t\t\tmergedActiveFiles.add(file);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// The on-disk scan is public: an unserved (client-owned) worker row hides its live metadata only.\n\t\t\tmerged.push(summaryForInactiveSession(info));\n\t\t}\n\t\t// The boot seed covers registered workers' families only, so dead families (pure on-disk\n\t\t// history, no registered worker anywhere in the tree) are not roster-resident. Their\n\t\t// subagent rows are read from the spawn ledger on demand instead.\n\t\tlet spawnEdges: RlmLedgerEdge[] = [];\n\t\ttry {\n\t\t\tspawnEdges = await this.rlmSpawnLedger().liveEdges();\n\t\t} catch (error) {\n\t\t\tthis.log(`Could not list spawn-ledger sessions: ${String(error)}`);\n\t\t}\n\t\tconst spawnParents = new Map(spawnEdges.map((edge) => [canonicalSessionPath(edge.child), edge.parent]));\n\t\tconst offlineRows: AgentRosterEntry[] = [];\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild || entry.summary.activeSessionId !== undefined) continue;\n\t\t\tif (entry.workerId !== undefined && this.workers.has(entry.workerId)) continue;\n\t\t\tconst file = entry.summary.sessionFile ? canonicalSessionPath(entry.summary.sessionFile) : undefined;\n\t\t\tif (file === undefined || scannedFiles.has(file) || activeByFile.has(file)) continue;\n\t\t\tofflineRows.push(entry);\n\t\t}\n\t\tfor (const hydrated of await Promise.all(offlineRows.map((entry) => this.hydrateSeededEntry(entry)))) {\n\t\t\tconst summary = sessionSummaryFromRosterEntry(hydrated);\n\t\t\tif (cwd !== undefined && resolve(summary.cwd) !== cwd) continue;\n\t\t\tif (!this.matchesListSessionDir(summary, sessionDir, spawnParents)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\tconst unseededFiles = new Set<string>();\n\t\tfor (const edge of spawnEdges) {\n\t\t\tconst childPath = canonicalSessionPath(edge.child);\n\t\t\tif (scannedFiles.has(childPath) || activeByFile.has(childPath) || unseededFiles.has(childPath)) continue;\n\t\t\tif (this.roster().hasSessionFile(childPath)) continue;\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\tunseededFiles.add(childPath);\n\t\t\t// Hydrated one at a time, like the boot seed: a large dead-family ledger must not fan\n\t\t\t// out into one concurrent transcript read per child.\n\t\t\tconst hydrated = await this.hydratedSeedEntry(entry);\n\t\t\t// The same classification a roster write would have applied: these rows read \"inactive\".\n\t\t\tconst summary = sessionSummaryFromRosterEntry({\n\t\t\t\t...hydrated,\n\t\t\t\tstatus: classifySessionRosterStatus(hydrated.summary),\n\t\t\t});\n\t\t\tif (cwd !== undefined && resolve(summary.cwd) !== cwd) continue;\n\t\t\tif (!this.matchesListSessionDir(summary, sessionDir, spawnParents)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\tfor (const summary of active) {\n\t\t\tconst file = summary.sessionFile ? canonicalSessionPath(summary.sessionFile) : undefined;\n\t\t\tif (file !== undefined && mergedActiveFiles.has(file)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\treturn success(command.id, \"list\", { ...data, sessions: merged });\n\t}\n\n\tprivate async hydratedSeedEntry<T extends WorkerRosterEntry>(entry: T): Promise<T> {\n\t\tconst info = entry.summary.sessionFile\n\t\t\t? await readSessionInfo(entry.summary.sessionFile).catch(() => undefined)\n\t\t\t: undefined;\n\t\tif (!info) return { ...entry, seededCwd: true as const };\n\t\tconst { seededCwd, ...rest } = entry;\n\t\treturn { ...rest, summary: { ...entry.summary, cwd: info.cwd } } as T;\n\t}\n\n\tprivate async hydrateSeededEntry(entry: AgentRosterEntry): Promise<AgentRosterEntry> {\n\t\tif (entry.seededCwd !== true || !entry.summary.sessionFile) return entry;\n\t\tconst hydrated = await this.hydratedSeedEntry(entry);\n\t\tif (hydrated.seededCwd === true) return entry;\n\t\tconst current = this.roster().get(entry.agentId);\n\t\tif (current !== entry) return current ?? entry;\n\t\treturn this.roster().write(hydrated, entry.workerId, entry.statusLabel);\n\t}\n\n\tprivate matchesListSessionDir(\n\t\tsummary: SessionSummary,\n\t\tsessionDir: string | undefined,\n\t\tspawnParents?: ReadonlyMap<string, string>,\n\t): boolean {\n\t\tif (sessionDir === undefined) return true;\n\t\tif (!summary.sessionFile) return false;\n\t\tlet file = resolve(summary.sessionFile);\n\t\tlet parentSessionPath = summary.parentSessionPath;\n\t\tconst visited = new Set<string>();\n\t\twhile (parentSessionPath !== undefined) {\n\t\t\tconst canonical = canonicalSessionPath(parentSessionPath);\n\t\t\tif (visited.has(canonical)) break;\n\t\t\tvisited.add(canonical);\n\t\t\tfile = resolve(parentSessionPath);\n\t\t\t// Dead-family ancestors are not roster-resident; the spawn ledger continues the walk.\n\t\t\tparentSessionPath =\n\t\t\t\tthis.roster().bySessionFile(canonical)?.summary.parentSessionPath ?? spawnParents?.get(canonical);\n\t\t}\n\t\treturn dirname(file) === resolve(sessionDir);\n\t}\n\n\tprivate async handleSavedSessionList(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"list_saved_sessions\" }>,\n\t): Promise<DaemonResponse> {\n\t\tconst pageQuery = captureSavedSessionPageQuery(command.page ?? {});\n\t\tconst scope = command.scope;\n\t\tlet cwd: string;\n\t\tlet sessionDir: string | undefined;\n\t\tif (\"activeSessionId\" in command) {\n\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\tcwd = match.summary.cwd;\n\t\t\tsessionDir = this.defaultSessionConfig.sessionDir;\n\t\t} else {\n\t\t\tcwd = resolve(command.cwd);\n\t\t\tsessionDir = command.sessionDir;\n\t\t}\n\t\tconst { agentDir, sessionsDir: ledgerSessionDir } = this.rlmJournalScope(sessionDir);\n\t\tconst page = await this.catalog.list(cwd, sessionDir, { page: pageQuery, scope, agentDir, ledgerSessionDir });\n\t\treturn success(command.id, \"list_saved_sessions\", page);\n\t}\n\n\tprivate async createOrReuseWorker(clientId: string, command: DaemonCreateCommand): Promise<ResidentWorker> {\n\t\tlet createCommand = command;\n\t\tif (command.name !== undefined) {\n\t\t\tconst normalizedName = command.name.trim();\n\t\t\tif (!normalizedName) {\n\t\t\t\tthrow new Error(\"Session name cannot be empty\");\n\t\t\t}\n\t\t\tcreateCommand = { ...command, name: normalizedName };\n\t\t}\n\t\tconst ownerClientId = command.lifecycle === \"client_owned\" ? clientId : undefined;\n\t\tif (command.sessionPath) {\n\t\t\tconst activeMatches = this.matchWorkers(command.sessionPath);\n\t\t\tif (\n\t\t\t\tactiveMatches.length === 1 &&\n\t\t\t\t!(await this.reclaimStaleWorkerRegistration(activeMatches[0]!.worker, command.launchEnv !== undefined))\n\t\t\t) {\n\t\t\t\treturn this.reuseWorkerForCreate(activeMatches[0]!.worker, ownerClientId, command.sessionPath);\n\t\t\t}\n\t\t\tif (activeMatches.length > 1) {\n\t\t\t\tthrow new Error(`Ambiguous active session \"${command.sessionPath}\"`);\n\t\t\t}\n\t\t\tconst config = mergeAgentSessionRuntimeConfig(this.defaultSessionConfig, command.config);\n\t\t\tconst sessionPath = looksLikeSessionPath(command.sessionPath)\n\t\t\t\t? resolve(command.sessionPath)\n\t\t\t\t: await this.catalog.resolve(command.sessionPath, config.cwd ?? process.cwd(), config.sessionDir);\n\t\t\tcreateCommand = { ...createCommand, sessionPath };\n\t\t}\n\t\tconst key = createCommand.sessionPath\n\t\t\t? canonicalSessionPath(createCommand.sessionPath)\n\t\t\t: `new:${command.id ? createCommandIdempotencyKey(clientId, command.id) : createActiveSessionId()}`;\n\t\tconst pending = this.openingWorkers.get(key);\n\t\tif (pending) {\n\t\t\treturn this.joinOpeningWorker(pending, ownerClientId, createCommand.sessionPath ?? key);\n\t\t}\n\t\tif (createCommand.sessionPath) {\n\t\t\tconst existing = this.findWorkerBySessionFile(createCommand.sessionPath);\n\t\t\tif (existing && !(await this.reclaimStaleWorkerRegistration(existing, command.launchEnv !== undefined))) {\n\t\t\t\treturn this.reuseWorkerForCreate(existing, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t\t// The reclaim await may have let a concurrent opener register; join it instead of double-launching.\n\t\t\tconst opened = this.openingWorkers.get(key);\n\t\t\tif (opened) {\n\t\t\t\treturn this.joinOpeningWorker(opened, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t}\n\t\tif (createCommand.sessionPath) {\n\t\t\tconst existing = this.findWorkerBySessionFile(createCommand.sessionPath);\n\t\t\tif (existing && !(await this.reclaimStaleWorkerRegistration(existing, command.launchEnv !== undefined))) {\n\t\t\t\treturn this.reuseWorkerForCreate(existing, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t}\n\t\tconst opening = (async () => {\n\t\t\tif (!createCommand.name) return this.launchWorker(createCommand, undefined, ownerClientId);\n\t\t\tconst savedSiblings = createCommand.sessionPath ? await this.rlmLedgerSiblings(createCommand.sessionPath) : [];\n\t\t\tconst target = savedSiblings.find(\n\t\t\t\t(session) => canonicalSessionPath(session.path) === canonicalSessionPath(createCommand.sessionPath!),\n\t\t\t);\n\t\t\tconst targetSummary = target ? summaryForInactiveSession(target) : { sessionId: \"new-root\", rlmDepth: 0 };\n\t\t\tconst reservation = this.summaryNameReservationInput(targetSummary, createCommand.name);\n\t\t\treturn this.withSessionNameReservation(reservation, async () => {\n\t\t\t\tif (target?.parentSessionPath && (target.rlmDepth ?? 0) > 0) {\n\t\t\t\t\tthis.assertSavedSiblingNameAvailable(savedSiblings, target, createCommand.name!);\n\t\t\t\t} else {\n\t\t\t\t\tawait this.assertSupervisorSessionNameAvailable(targetSummary, createCommand.name!);\n\t\t\t\t}\n\t\t\t\treturn this.launchWorker(createCommand, undefined, ownerClientId);\n\t\t\t});\n\t\t})();\n\t\tthis.openingWorkers.set(key, opening);\n\t\ttry {\n\t\t\treturn await opening;\n\t\t} finally {\n\t\t\tif (this.openingWorkers.get(key) === opening) {\n\t\t\t\tthis.openingWorkers.delete(key);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async reuseWorkerForCreate(\n\t\tworker: ResidentWorker,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): Promise<ResidentWorker> {\n\t\tif (worker.descriptor.lifecycle === \"failed\") {\n\t\t\tthrow new Error(\n\t\t\t\t`Session \"${sessionPath}\" is registered to a failed worker that could not be safely reclaimed`,\n\t\t\t);\n\t\t}\n\t\tthis.assertWorkerCreateOwner(worker, ownerClientId, sessionPath);\n\t\tif (!this.isWorkerReadyForCreate(worker)) {\n\t\t\tif (worker.recovery) {\n\t\t\t\tawait worker.recovery;\n\t\t\t} else if (this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tawait this.recoverWorker(worker);\n\t\t\t}\n\t\t}\n\t\tconst current = this.workers.get(worker.descriptor.workerId);\n\t\tif (!current) {\n\t\t\tthrow new Error(`Session \"${sessionPath}\" worker recovery was interrupted; retry opening the session`);\n\t\t}\n\t\tthis.assertWorkerCreateOwner(current, ownerClientId, sessionPath);\n\t\tif (!this.isWorkerReadyForCreate(current)) {\n\t\t\tif (!this.workerHasRosterRoot(current)) {\n\t\t\t\tthrow new Error(\n\t\t\t\t\t`Session \"${sessionPath}\" worker is unavailable for reuse: assigned root session is missing`,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst detail = current.descriptor.lastError ? `: ${current.descriptor.lastError}` : \"\";\n\t\t\tthrow new Error(`Session \"${sessionPath}\" worker is ${this.effectiveWorkerState(current)}${detail}`);\n\t\t}\n\t\treturn current;\n\t}\n\n\tprivate async joinOpeningWorker(\n\t\tpending: Promise<ResidentWorker>,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): Promise<ResidentWorker> {\n\t\tconst worker = await pending;\n\t\tthis.assertWorkerCreateOwner(worker, ownerClientId, sessionPath);\n\t\treturn worker;\n\t}\n\n\tprivate assertWorkerCreateOwner(\n\t\tworker: ResidentWorker,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): void {\n\t\tif (worker.descriptor.ownerClientId !== ownerClientId) {\n\t\t\tthrow new SessionAlreadyActiveError(sessionPath, worker.descriptor.rootActiveSessionId);\n\t\t}\n\t}\n\n\tprivate isWorkerReadyForCreate(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tworker.descriptor.lifecycle === \"ready\" &&\n\t\t\tworker.client !== undefined &&\n\t\t\tthis.workerHasRosterRoot(worker) &&\n\t\t\t!this.isWorkerStopping(worker)\n\t\t);\n\t}\n\n\tprivate workerHasRosterRoot(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId)?.workerId === worker.descriptor.workerId\n\t\t);\n\t}\n\n\t/**\n\t * A stopping worker whose process already died can strand its registration\n\t * (for example when the stop timed out and its finalization was interrupted\n\t * by a supervisor restart). Such a registration would block reopening the\n\t * saved transcript forever, so complete the interrupted stop and let the\n\t * caller launch a fresh worker for the saved session.\n\t */\n\tprivate async reclaimStaleWorkerRegistration(worker: ResidentWorker, freshCreate = false): Promise<boolean> {\n\t\tif (worker.client !== undefined || worker.recovery !== undefined) {\n\t\t\treturn false;\n\t\t}\n\t\tif (worker.descriptor.stopRequestedAt === undefined) {\n\t\t\tif (worker.descriptor.lifecycle !== \"failed\" || worker.descriptor.ownerClientId) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst identity = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\tif (identity === \"current\") {\n\t\t\t\tif (worker.compatibilityError || !freshCreate || !worker.descriptor.processStartId) return false;\n\t\t\t\tawait this.stopWorker(worker, true, true);\n\t\t\t\treturn true;\n\t\t\t}\n\t\t\tif (identity !== \"gone\" && identity !== \"replaced\") {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tworker.intentionalStop = true;\n\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t\treturn true;\n\t\t}\n\t\t// Fail fast before waiting on anything: only a confirmed-dead process is\n\t\t// reclaimable. A live, unknown, or still-stopping worker is left alone\n\t\t// and the caller reports the session as already active.\n\t\tconst identity = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\tif (identity !== \"gone\" && identity !== \"replaced\") {\n\t\t\treturn false;\n\t\t}\n\t\t// Single cleanup path: the background stop finalizer is identity-aware,\n\t\t// retrying, and single-flighted, so concurrent resumes share one stop.\n\t\t// The wait is bounded so a resume request always returns promptly.\n\t\tthis.scheduleWorkerStopFinalization(worker);\n\t\tconst finalization = worker.stopFinalization;\n\t\tif (finalization) {\n\t\t\tawait Promise.race([finalization.catch(() => undefined), unrefDelay(STALE_RECLAIM_WAIT_MS)]);\n\t\t}\n\t\tif (this.workers.get(worker.descriptor.workerId) === worker) {\n\t\t\t// The process is confirmed dead, so the registration must never be\n\t\t\t// reused; slow cleanup fails the resume honestly instead.\n\t\t\tthrow new Error(\n\t\t\t\t`Stopped session worker ${worker.descriptor.workerId} is still being cleaned up; retry shortly`,\n\t\t\t);\n\t\t}\n\t\tthis.log(`Reclaimed stale registration for stopped worker ${worker.descriptor.workerId}`);\n\t\treturn true;\n\t}\n\n\tprivate async promoteOwnedWorker(client: DaemonSocketClient, worker: ResidentWorker): Promise<void> {\n\t\tconst clientId = this.protocolClientId(client);\n\t\tif (worker.descriptor.ownerClientId === undefined && worker.promotedOwnerClientId === clientId) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.ownerClientId !== clientId) {\n\t\t\tthrow new Error(\"Session is not owned by this client\");\n\t\t}\n\t\tconst previousDescriptor = worker.descriptor;\n\t\tworker.descriptor = { ...previousDescriptor, ownerClientId: undefined };\n\t\ttry {\n\t\t\tthis.persistWorker(worker);\n\t\t} catch (error) {\n\t\t\tworker.descriptor = previousDescriptor;\n\t\t\tthrow error;\n\t\t}\n\t\tworker.promotedOwnerClientId = clientId;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tthis.roster().amend(entry.agentId, {});\n\t\t}\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tworker.launchEnv = undefined;\n\t\tworker.transientCreateCommand = undefined;\n\t}\n\n\tprivate describeWorkerSpawnFailure(error: Error): Error {\n\t\tconst errno = (error as NodeJS.ErrnoException).code;\n\t\tconst hint =\n\t\t\terrno === \"EMFILE\" || errno === \"ENFILE\"\n\t\t\t\t? ` (${this.workers.size} resident session workers are holding file descriptors; stop unused sessions or raise the open-file limit (ulimit -n))`\n\t\t\t\t: \"\";\n\t\treturn new Error(`Failed to spawn session worker: ${error.message}${hint}`);\n\t}\n\n\tprivate async launchWorker(\n\t\tcommand: DaemonCreateCommand,\n\t\texisting?: ResidentWorker,\n\t\townerClientId?: string,\n\t): Promise<ResidentWorker> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (existing && this.isWorkerRecoveryCancelled(existing)) {\n\t\t\tthrow new Error(`Session worker ${existing.descriptor.workerId} recovery was cancelled`);\n\t\t}\n\t\tconst recoveryStopRevision = existing?.stopRevision;\n\t\tconst launchEnv = command.launchEnv ?? existing?.launchEnv;\n\t\tconst createCommand: DaemonCreateCommand = {\n\t\t\t...withoutSupervisorCreateFields(command),\n\t\t\tconfig: mergeAgentSessionRuntimeConfig(this.defaultSessionConfig, command.config),\n\t\t};\n\t\tconst workerId = existing?.descriptor.workerId ?? createActiveSessionId();\n\t\tconst rootActiveSessionId = existing?.descriptor.rootActiveSessionId ?? createActiveSessionId();\n\t\tconst socketPath = existing?.descriptor.socketPath ?? workerSocketPath(this.socketPath, workerId);\n\t\tconst token = existing?.descriptor.authenticationToken ?? randomBytes(32).toString(\"base64url\");\n\t\t// Fresh per incarnation: peer transport grants must never survive a worker restart.\n\t\tconst workerInstanceId = randomUUID();\n\t\tconst now = new Date().toISOString();\n\t\tconst descriptorPath = existing?.descriptorPath ?? join(this.descriptorDir, `${workerId}.json`);\n\t\tconst recoveryJournalPath =\n\t\t\texisting?.descriptor.recoveryJournalPath ?? join(this.descriptorDir, `${workerId}.recovery.jsonl`);\n\t\tconst orphanProcessJournalPath =\n\t\t\texisting?.descriptor.orphanProcessJournalPath ?? join(this.descriptorDir, `${workerId}.orphans.jsonl`);\n\t\tconst launch = createCliSubprocessLaunchSpec([\"--mode\", \"daemon\", \"--daemon-socket\", socketPath]);\n\t\tconst workerEnvironment = createCliSubprocessEnv({\n\t\t\t...process.env,\n\t\t\t...launchEnv,\n\t\t\t[DAEMON_WORKER_ROLE_ENV]: \"1\",\n\t\t\t[DAEMON_WORKER_TOKEN_ENV]: token,\n\t\t\t[DAEMON_WORKER_INSTANCE_ID_ENV]: workerInstanceId,\n\t\t\t[DAEMON_WORKER_ACTIVE_SESSION_ID_ENV]: rootActiveSessionId,\n\t\t\t[DAEMON_WORKER_SUPERVISOR_SOCKET_ENV]: this.socketPath,\n\t\t\t[DAEMON_WORKER_RECOVERY_JOURNAL_ENV]: recoveryJournalPath,\n\t\t\t[DAEMON_WORKER_STARTUP_GATE_FD_ENV]: String(WORKER_STARTUP_GATE_FD),\n\t\t\t[ORPHAN_PROCESS_JOURNAL_ENV]: orphanProcessJournalPath,\n\t\t\t[SESSION_LEASES_ENABLED_ENV]: \"1\",\n\t\t\t[SESSION_LEASE_OWNER_ID_ENV]: rootActiveSessionId,\n\t\t});\n\t\tdelete workerEnvironment.BASE_CONTEXT_RLM_DEPTH;\n\t\tawait this.assertRecoveryAllowed();\n\t\tconst child: ChildProcess = spawn(launch.command, launch.args, {\n\t\t\tcwd: createCommand.config?.cwd ?? process.cwd(),\n\t\t\tdetached: true,\n\t\t\tenv: workerEnvironment,\n\t\t\tstdio: [\"ignore\", \"ignore\", \"pipe\", \"pipe\"],\n\t\t});\n\t\tconst detachWorkerStderr = child.stderr\n\t\t\t? attachJsonlLineReader(child.stderr, (line) => this.log(`Session worker ${workerId} stderr: ${line}`), {\n\t\t\t\t\tmaxLineLength: 64 * 1024,\n\t\t\t\t\tonLineOverflow: (prefix) => this.log(`Session worker ${workerId} stderr: ${prefix} [truncated]`),\n\t\t\t\t})\n\t\t\t: () => {};\n\t\tchild.once(\"close\", detachWorkerStderr);\n\t\tconst childClosed = new Promise<void>((resolveClose) => child.once(\"close\", () => resolveClose()));\n\t\tlet spawnFailure: Error | undefined;\n\t\tconst spawnSettled = new Promise<void>((resolveSpawn) => {\n\t\t\tchild.once(\"spawn\", () => resolveSpawn());\n\t\t\tchild.once(\"error\", (error) => {\n\t\t\t\tspawnFailure = error instanceof Error ? error : new Error(String(error));\n\t\t\t\tresolveSpawn();\n\t\t\t});\n\t\t});\n\t\tchild.on(\"error\", (error) => {\n\t\t\tthis.log(\n\t\t\t\t`Session worker ${workerId} process error: ${error instanceof Error ? error.message : String(error)}`,\n\t\t\t);\n\t\t});\n\t\t// A failed spawn (e.g. EMFILE) leaves child.stdio undefined.\n\t\tconst startupGate = child.stdio?.[WORKER_STARTUP_GATE_FD];\n\t\tconst previousDescriptor = existing?.descriptor;\n\t\tconst previousIntentionalStop = existing?.intentionalStop;\n\t\tlet descriptorAssigned = false;\n\t\tlet childPid: number;\n\t\tlet childProcessStartId: string | undefined;\n\t\tlet worker: ResidentWorker;\n\t\ttry {\n\t\t\tawait spawnSettled;\n\t\t\tif (spawnFailure) {\n\t\t\t\tthrow this.describeWorkerSpawnFailure(spawnFailure);\n\t\t\t}\n\t\t\tif (!child.pid) {\n\t\t\t\tthrow new Error(\"Failed to obtain daemon session worker pid\");\n\t\t\t}\n\t\t\tif (!(startupGate instanceof Writable)) {\n\t\t\t\tthrow new Error(\"Failed to create daemon session worker startup gate\");\n\t\t\t}\n\t\t\tchildPid = child.pid;\n\t\t\tchildProcessStartId = getProcessStartId(childPid);\n\t\t\tawait this.assertRecoveryAllowed();\n\n\t\t\tconst descriptor: DaemonWorkerDescriptor = {\n\t\t\t\tversion: 2,\n\t\t\t\tworkerId,\n\t\t\t\tpid: childPid,\n\t\t\t\t...(childProcessStartId ? { processStartId: childProcessStartId } : {}),\n\t\t\t\tsocketPath,\n\t\t\t\trecoveryJournalPath,\n\t\t\t\torphanProcessJournalPath,\n\t\t\t\tsupervisorSocketPath: this.socketPath,\n\t\t\t\tauthenticationToken: token,\n\t\t\t\tworkerInstanceId,\n\t\t\t\trootActiveSessionId,\n\t\t\t\townerClientId: existing?.descriptor.ownerClientId ?? ownerClientId,\n\t\t\t\tsessionDir: createCommand.config?.sessionDir,\n\t\t\t\ttelemetryDisabled: createCommand.config?.telemetryDisabled,\n\t\t\t\tcreatedAt: existing?.descriptor.createdAt ?? now,\n\t\t\t\tupdatedAt: now,\n\t\t\t\tlifecycle: \"starting\",\n\t\t\t\tcreateCommand: durableDaemonCreateCommand(createCommand),\n\t\t\t\tconsecutiveFailures: existing?.descriptor.consecutiveFailures ?? 0,\n\t\t\t};\n\t\t\tworker = existing ?? {\n\t\t\t\tdescriptor,\n\t\t\t\tdescriptorPath,\n\t\t\t\tsummaries: new Map(),\n\t\t\t\tsnapshotCache: new Map(),\n\t\t\t\ttranscriptCaches: new Map(),\n\t\t\t\tsnapshotGenerations: new Map(),\n\t\t\t\tsnapshotLoads: new Map(),\n\t\t\t\tintentionalStop: false,\n\t\t\t\tstopRevision: 0,\n\t\t\t\tlaunchEnv,\n\t\t\t\ttransientCreateCommand: ownerClientId ? createCommand : undefined,\n\t\t\t};\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor = descriptor;\n\t\t\tworker.launchEnv = launchEnv;\n\t\t\tworker.transientCreateCommand = descriptor.ownerClientId ? createCommand : undefined;\n\t\t\tdescriptorAssigned = true;\n\t\t\tthis.persistWorker(worker);\n\t\t\tworker.intentionalStop = false;\n\t\t\tthis.workers.set(workerId, worker);\n\t\t} catch (error) {\n\t\t\tif (startupGate instanceof Writable) {\n\t\t\t\tstartupGate.destroy();\n\t\t\t}\n\t\t\tawait childClosed;\n\t\t\tchild.unref();\n\t\t\ttry {\n\t\t\t\trmSync(`${descriptorPath}.${process.pid}.tmp`, { force: true });\n\t\t\t} catch (cleanupError) {\n\t\t\t\tthis.reportCleanupFailure(`worker launch temp ${workerId}`, cleanupError);\n\t\t\t}\n\t\t\tif (existing && descriptorAssigned && previousDescriptor) {\n\t\t\t\ttry {\n\t\t\t\t\texisting.descriptor = previousDescriptor;\n\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\tthis.reportCleanupFailure(`worker launch descriptor ${workerId}`, cleanupError);\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\n\t\ttry {\n\t\t\ttry {\n\t\t\t\tawait commitWorkerStartupGate(startupGate);\n\t\t\t} catch (error) {\n\t\t\t\tstartupGate.destroy();\n\t\t\t\tawait childClosed;\n\t\t\t\tthrow error;\n\t\t\t} finally {\n\t\t\t\tchild.unref();\n\t\t\t}\n\t\t\tconst client = await this.connectWorker(worker, WORKER_CONNECT_TIMEOUT_MS);\n\t\t\tconst response = await client.request(withoutCommandId(createCommand), WORKER_REQUEST_TIMEOUT_MS);\n\t\t\tif (!response.success) {\n\t\t\t\tthrow deserializeDaemonError(response);\n\t\t\t}\n\t\t\tif (!isSessionSummary(response.data)) {\n\t\t\t\tthrow new Error(\"Session worker returned an invalid create response\");\n\t\t\t}\n\t\t\tconst summary = response.data;\n\t\t\tif ((summary.activeSessionId ?? summary.id) !== rootActiveSessionId) {\n\t\t\t\tthrow new Error(\"Session worker did not preserve its assigned active session id\");\n\t\t\t}\n\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(summary), worker);\n\t\t\tworker.descriptor.rootSessionId = summary.sessionId;\n\t\t\tworker.descriptor.sessionFile = summary.sessionFile;\n\t\t\tawait this.subscribeWorker(worker, rootActiveSessionId);\n\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\tif (existing && (this.isWorkerRecoveryCancelled(worker) || worker.stopRevision !== recoveryStopRevision)) {\n\t\t\t\tthrow new Error(`Session worker ${workerId} recovery was cancelled`);\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\tworker.descriptor.lastError = undefined;\n\t\t\tthis.persistWorker(worker);\n\t\t\tif (!worker.descriptor.ownerClientId) {\n\t\t\t\tworker.launchEnv = undefined;\n\t\t\t\tworker.transientCreateCommand = undefined;\n\t\t\t}\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\treturn worker;\n\t\t} catch (error) {\n\t\t\tif (isSupervisorGenerationStale(error)) {\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tif (isSupervisorShutdownAdmissionCancelled(error)) {\n\t\t\t\tlet rolledBack = false;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.stopWorker(worker, existing === undefined, true, false, existing !== undefined, {\n\t\t\t\t\t\tchild,\n\t\t\t\t\t\tclosed: childClosed,\n\t\t\t\t\t});\n\t\t\t\t\trolledBack = true;\n\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\tthis.reportCleanupFailure(`cancelled worker launch ${workerId}`, cleanupError);\n\t\t\t\t}\n\t\t\t\tconst mappedWorker = this.workers.get(workerId);\n\t\t\t\tif (\n\t\t\t\t\trolledBack &&\n\t\t\t\t\texisting &&\n\t\t\t\t\tpreviousDescriptor &&\n\t\t\t\t\t!this.shuttingDown &&\n\t\t\t\t\texisting.stopRevision === recoveryStopRevision &&\n\t\t\t\t\texisting.descriptor.stopRequestedAt === undefined &&\n\t\t\t\t\t(mappedWorker === undefined || mappedWorker === existing)\n\t\t\t\t) {\n\t\t\t\t\texisting.descriptor = previousDescriptor;\n\t\t\t\t\texisting.intentionalStop = previousIntentionalStop ?? false;\n\t\t\t\t\tthis.workers.set(workerId, existing);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tthis.persistWorker(existing);\n\t\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\t\tthis.reportCleanupFailure(`cancelled worker recovery ${workerId}`, cleanupError);\n\t\t\t\t\t}\n\t\t\t\t\tthis.deferWorkerRecovery(existing, error instanceof Error ? error : new Error(String(error)));\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tconst shouldResumeRecovery =\n\t\t\t\texisting !== undefined &&\n\t\t\t\t!this.shuttingDown &&\n\t\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\t\tworker.stopRevision === recoveryStopRevision;\n\t\t\tawait this.stopWorker(worker, existing === undefined, true, false, existing !== undefined).catch((stopError) =>\n\t\t\t\tthis.log(`Could not stop failed worker ${workerId}: ${String(stopError)}`),\n\t\t\t);\n\t\t\tif (\n\t\t\t\tshouldResumeRecovery &&\n\t\t\t\t!this.shuttingDown &&\n\t\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\t\tworker.stopRevision === recoveryStopRevision\n\t\t\t) {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\tworker.intentionalStop = false;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.workers.set(workerId, worker);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async connectWorker(worker: ResidentWorker, timeoutMs: number): Promise<DaemonWorkerClient> {\n\t\tconst deadline = Date.now() + timeoutMs;\n\t\tlet lastError: unknown;\n\t\twhile (Date.now() < deadline) {\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tconst client = new DaemonWorkerClient(worker.descriptor.socketPath);\n\t\t\ttry {\n\t\t\t\tawait client.connect(Math.min(500, Math.max(50, deadline - Date.now())));\n\t\t\t\tawait client.waitForHello(1000);\n\t\t\t\t// Listen before authenticating: the worker flushes its roster snapshot right after auth succeeds.\n\t\t\t\tclient.onFrame((frame) => this.handleWorkerFrame(worker, frame, client));\n\t\t\t\tclient.onClose((error) => void this.handleWorkerClose(worker, client, error));\n\t\t\t\tworker.pendingClient = client;\n\t\t\t\ttry {\n\t\t\t\t\tconst authResponse = await client.authenticateWorker(\n\t\t\t\t\t\tworker.descriptor.authenticationToken,\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\t...this.supervisorAuthenticationClaim(),\n\t\t\t\t\t\t\t...(worker.descriptor.workerInstanceId !== undefined\n\t\t\t\t\t\t\t\t? { workerInstanceId: worker.descriptor.workerInstanceId }\n\t\t\t\t\t\t\t\t: {}),\n\t\t\t\t\t\t},\n\t\t\t\t\t\t1000,\n\t\t\t\t\t);\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\tif (!workerAuthAdvertisesRoster(authResponse.data)) {\n\t\t\t\t\t\tthrow new PreRosterWorkerError(\"Session worker predates the roster protocol and must be restarted\");\n\t\t\t\t\t}\n\t\t\t\t\tworker.peerTransportCapable = workerAuthAdvertisesPeerTransport(authResponse.data);\n\t\t\t\t\tworker.lastFrameAt = Date.now();\n\t\t\t\t\tworker.client?.close();\n\t\t\t\t\tworker.client = client;\n\t\t\t\t\tworker.compatibilityError = undefined;\n\t\t\t\t\tworker.descriptor.lastError = undefined;\n\t\t\t\t\treturn client;\n\t\t\t\t} finally {\n\t\t\t\t\tif (worker.pendingClient === client) worker.pendingClient = undefined;\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tlastError = error;\n\t\t\t\tclient.close();\n\t\t\t\tif (\n\t\t\t\t\tisSupervisorRecoveryCancelled(error) ||\n\t\t\t\t\terror instanceof PreRosterWorkerError ||\n\t\t\t\t\terror instanceof DaemonWorkerAuthenticationError ||\n\t\t\t\t\terror instanceof DaemonWorkerCompatibilityError\n\t\t\t\t) {\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tthrow new DaemonWorkerProbeTimeoutError(`Timed out connecting to daemon session worker: ${String(lastError)}`);\n\t}\n\n\tprivate parkIncompatibleWorker(worker: ResidentWorker, error: DaemonWorkerCompatibilityError): void {\n\t\t// Compatibility refusal is not evidence that this process or its ownership is dead.\n\t\tworker.compatibilityError = error;\n\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\tworker.descriptor.lastError = error.message;\n\t\tconst client = worker.client;\n\t\tworker.client = undefined;\n\t\tclient?.close();\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tthis.persistWorker(worker);\n\t\tthis.markWorkerRosterEntries(worker, \"recovering\");\n\t\tthis.log(`Kept incompatible worker ${worker.descriptor.workerId}: ${error.message}`);\n\t}\n\n\tprivate async subscribeWorker(worker: ResidentWorker, activeSessionId: string): Promise<void> {\n\t\tif (!worker.client) {\n\t\t\tthrow new Error(\"Session worker is not connected\");\n\t\t}\n\t\tconst supportsExtensionUi = [...this.clients].some(\n\t\t\t(client) => client.attachedActiveSessionIds.has(activeSessionId) && client.supportsExtensionUi,\n\t\t);\n\t\tconst response = await worker.client.requestWorker({\n\t\t\ttype: \"worker_subscribe\",\n\t\t\tactiveSessionId,\n\t\t\tcapabilities: supportsExtensionUi\n\t\t\t\t? [\"attach_snapshot\", \"event_sequence\", \"extension_ui\", \"slim_attach\", \"chunked_snapshot\"]\n\t\t\t\t: [\"attach_snapshot\", \"event_sequence\", \"slim_attach\", \"chunked_snapshot\"],\n\t\t\tsupportsExtensionUi,\n\t\t});\n\t\tif (!response.success) {\n\t\t\tthrow new Error(response.error);\n\t\t}\n\t}\n\n\tprivate async adoptOrRecoverWorker(worker: ResidentWorker): Promise<void> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (worker.descriptor.stopRequestedAt) {\n\t\t\ttry {\n\t\t\t\t// A descriptor persisted before identity tracking has no\n\t\t\t\t// processStartId, so stopWorker could neither signal the live\n\t\t\t\t// process nor let the finalizer escalate. Authenticating on the\n\t\t\t\t// worker's socket proves the pid still belongs to our worker, so\n\t\t\t\t// the start id observed while it was alive can be persisted (and\n\t\t\t\t// the connected client gives stopWorker its graceful IPC path).\n\t\t\t\tif (worker.descriptor.processStartId === undefined && isProcessAlive(worker.descriptor.pid)) {\n\t\t\t\t\tconst observedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.connectWorker(worker, 2000);\n\t\t\t\t\t\tif (observedProcessStartId) {\n\t\t\t\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\t}\n\t\t\t\t\t} catch {\n\t\t\t\t\t\t// Unverifiable identity stays untrusted; the stop below\n\t\t\t\t\t\t// still runs its graceful path and the finalizer keeps\n\t\t\t\t\t\t// waiting rather than signalling a possibly-recycled pid.\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tawait this.stopWorker(worker, true, true, worker.descriptor.archiveOnStop === true);\n\t\t\t\tthis.log(`Completed intentional stop for worker ${worker.descriptor.workerId} during supervisor adoption`);\n\t\t\t} catch (error) {\n\t\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.log(`Could not complete intentional stop for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tlet observedProcessStartId: string | undefined;\n\t\ttry {\n\t\t\tif (!isProcessAlive(worker.descriptor.pid)) {\n\t\t\t\tthrow new Error(\"Session worker process is no longer running\");\n\t\t\t}\n\t\t\tobservedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\tawait this.connectWorker(worker, 2000);\n\t\t\tawait this.subscribeWorker(worker, worker.descriptor.rootActiveSessionId);\n\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\tif (worker.descriptor.processStartId === undefined && observedProcessStartId) {\n\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t} catch (error) {\n\t\t\tif (isSupervisorRecoveryCancelled(error)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tthis.log(`Could not adopt worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\t// A client-owned worker's launch env lives only with its owner; recoverWorker parks it instead.\n\t\t\tif (error instanceof PreRosterWorkerError && worker.descriptor.ownerClientId === undefined) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.restartPreRosterWorker(worker, observedProcessStartId);\n\t\t\t\t\treturn;\n\t\t\t\t} catch (restartError) {\n\t\t\t\t\tif (isSupervisorRecoveryCancelled(restartError)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tthis.log(`Could not restart pre-roster worker ${worker.descriptor.workerId}: ${String(restartError)}`);\n\t\t\t\t}\n\t\t\t}\n\t\t\tconst identityNow = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\tif (isDaemonWorkerProbeTimeout(error) && (identityNow === \"current\" || identityNow === \"unknown\")) {\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tvoid this.recoverWorker(worker).catch((recoveryError) =>\n\t\t\t\t\tthis.log(`Could not recover worker ${worker.descriptor.workerId}: ${String(recoveryError)}`),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tawait this.recoverWorker(worker);\n\t\t}\n\t}\n\n\tprivate async restartPreRosterWorker(\n\t\tworker: ResidentWorker,\n\t\tobservedProcessStartId: string | undefined,\n\t): Promise<void> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (worker.descriptor.processStartId === undefined && observedProcessStartId !== undefined) {\n\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t}\n\t\tconst identity = () => this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t// The one deliberate kill of a live worker: it authenticated as ours and predates the roster\n\t\t// protocol, so this identity-verified upgrade replaces it. No await between recheck and signal.\n\t\tif (identity() === \"current\") {\n\t\t\tsignalProcessGroupOrProcess(worker.descriptor.pid, \"SIGKILL\");\n\t\t\t// SIGKILL is uninterceptable; this wait only covers kernel teardown of the old process and socket.\n\t\t\tconst killDeadline = Date.now() + 1000;\n\t\t\twhile (identity() === \"current\" && Date.now() < killDeadline) {\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tconst finalIdentity = identity();\n\t\tif (finalIdentity !== \"gone\" && finalIdentity !== \"replaced\") {\n\t\t\t// A live or unverifiable survivor parks failed with no destructive cleanup: interruption\n\t\t\t// marking and orphan reaping must never run against a possibly-active worker.\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = `Pre-roster worker process ${worker.descriptor.pid} is still running and cannot be replaced safely`;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(`Kept pre-roster worker ${worker.descriptor.workerId} failed: ${worker.descriptor.lastError}`);\n\t\t\treturn;\n\t\t}\n\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tawait this.launchWorker(worker.descriptor.createCommand, worker, worker.descriptor.ownerClientId);\n\t}\n\n\tprivate async handleWorkerClose(worker: ResidentWorker, client: DaemonWorkerClient, error: Error): Promise<void> {\n\t\tif (worker.client !== client) {\n\t\t\treturn;\n\t\t}\n\t\tworker.client = undefined;\n\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker disconnected while input was paused\");\n\t\tconst interrupted = new Map<string, Set<string>>();\n\t\tfor (const [activeSessionId, generations] of worker.snapshotGenerations ?? []) {\n\t\t\tfor (const generation of generations.values()) {\n\t\t\t\tif (generation.incoming || !generation.transcript.complete) {\n\t\t\t\t\tconst snapshotIds = interrupted.get(activeSessionId) ?? new Set<string>();\n\t\t\t\t\tsnapshotIds.add(generation.transcript.snapshotId);\n\t\t\t\t\tinterrupted.set(activeSessionId, snapshotIds);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tfor (const [activeSessionId, transcript] of worker.transcriptCaches) {\n\t\t\tif (!transcript.complete) {\n\t\t\t\tconst snapshotIds = interrupted.get(activeSessionId) ?? new Set<string>();\n\t\t\t\tsnapshotIds.add(transcript.snapshotId);\n\t\t\t\tinterrupted.set(activeSessionId, snapshotIds);\n\t\t\t}\n\t\t}\n\t\tfor (const [activeSessionId, snapshotIds] of interrupted) {\n\t\t\tfor (const snapshotId of snapshotIds) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tnew Error(\"Session worker disconnected during snapshot transfer\"),\n\t\t\t\t\tfalse,\n\t\t\t\t\tsnapshotId,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t\tif (this.shuttingDown || worker.intentionalStop) {\n\t\t\treturn;\n\t\t}\n\t\tthis.markWorkerRosterEntries(worker, \"recovering\");\n\t\ttry {\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t} catch (recoveryError) {\n\t\t\tif (!isSupervisorGenerationStale(recoveryError)) {\n\t\t\t\tthis.deferWorkerRecovery(worker, error);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\tworker.descriptor.lastError = error.message;\n\t\tthis.persistWorker(worker);\n\t\tvoid this.recoverWorker(worker);\n\t}\n\n\tprivate isWorkerRecoveryEligible(worker: ResidentWorker): boolean {\n\t\treturn this.isWorkerRecoveryCandidate(worker) && worker.recovery === undefined;\n\t}\n\n\tprivate isWorkerRecoveryCandidate(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\t!this.shuttingDown &&\n\t\t\t!worker.intentionalStop &&\n\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.client === undefined\n\t\t);\n\t}\n\n\tprivate deferWorkerRecovery(worker: ResidentWorker, disconnectError: Error): void {\n\t\tif (worker.compatibilityError || worker.deferredRecovery) {\n\t\t\treturn;\n\t\t}\n\t\t// A live-but-silent worker must not probe forever: park it failed (user-visible through the\n\t\t// roster's failed status) and keep its process alive for a manual retry_worker.\n\t\tworker.deferredRecoveryRounds = (worker.deferredRecoveryRounds ?? 0) + 1;\n\t\tif (worker.deferredRecoveryRounds > MAX_DEFERRED_RECOVERY_ROUNDS) {\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = `Live session worker did not answer recovery probes for ${MAX_DEFERRED_RECOVERY_ROUNDS} rounds: ${disconnectError.message}`;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(\n\t\t\t\t`Worker ${worker.descriptor.workerId} is unresponsive; parked failed after ${MAX_DEFERRED_RECOVERY_ROUNDS} probe rounds`,\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\t\tworker.deferredRecovery = this.resumeDeferredWorkerRecovery(worker, disconnectError).finally(() => {\n\t\t\tworker.deferredRecovery = undefined;\n\t\t});\n\t}\n\n\tprivate async resumeDeferredWorkerRecovery(worker: ResidentWorker, disconnectError: Error): Promise<void> {\n\t\twhile (true) {\n\t\t\tawait unrefDelay(DEFERRED_RECOVERY_RECHECK_MS);\n\t\t\tif (worker.compatibilityError || !this.isWorkerRecoveryCandidate(worker)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t} catch (error) {\n\t\t\t\tif (isSupervisorGenerationStale(error)) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (worker.compatibilityError || !this.isWorkerRecoveryCandidate(worker)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\tworker.descriptor.lastError = disconnectError.message;\n\t\t\tthis.persistWorker(worker);\n\t\t\tvoid this.recoverWorker(worker);\n\t\t\treturn;\n\t\t}\n\t}\n\n\tprivate failWorkerSnapshotCache(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\terror: Error,\n\t\tcloseWorkerChannel = false,\n\t\texpectedSnapshotId?: string,\n\t): void {\n\t\tconst generations = worker.snapshotGenerations?.get(activeSessionId);\n\t\tif (expectedSnapshotId) {\n\t\t\tconst generation = generations?.get(expectedSnapshotId);\n\t\t\tif (generation) {\n\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, error);\n\t\t\t} else {\n\t\t\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\t\t\tif (transcript?.snapshotId !== expectedSnapshotId) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttranscript.markFailed(error);\n\t\t\t\ttranscript.dispose();\n\t\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === expectedSnapshotId) {\n\t\t\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tconst failedTranscripts = new Set<SnapshotTranscriptCache>();\n\t\t\tfor (const generation of [...(generations?.values() ?? [])]) {\n\t\t\t\tfailedTranscripts.add(generation.transcript);\n\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, error);\n\t\t\t}\n\t\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\t\tif (transcript && !failedTranscripts.has(transcript)) {\n\t\t\t\ttranscript.markFailed(error);\n\t\t\t\ttranscript.dispose();\n\t\t\t}\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t\tif (closeWorkerChannel) {\n\t\t\tconst client = worker.client;\n\t\t\tif (client) {\n\t\t\t\tthis.handleWorkerClose(worker, client, error);\n\t\t\t\tclient.close();\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate retireWorkerSnapshotCache(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\texpectedTranscript: SnapshotTranscriptCache,\n\t): void {\n\t\tif (worker.transcriptCaches.get(activeSessionId) === expectedTranscript) {\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t}\n\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === expectedTranscript.snapshotId) {\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, expectedTranscript.snapshotId);\n\t\tif (!generation) {\n\t\t\texpectedTranscript.dispose();\n\t\t\treturn;\n\t\t}\n\t\tgeneration.retired = true;\n\t\tthis.settleSnapshotDuplicateValidation(generation);\n\t\tif (generation.incoming) {\n\t\t\treturn;\n\t\t}\n\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\texpectedTranscript.dispose();\n\t}\n\n\tprivate snapshotGenerationsFor(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t): Map<string, SnapshotTranscriptGeneration> {\n\t\tworker.snapshotGenerations ??= new Map();\n\t\tlet generations = worker.snapshotGenerations.get(activeSessionId);\n\t\tif (!generations) {\n\t\t\tgenerations = new Map();\n\t\t\tworker.snapshotGenerations.set(activeSessionId, generations);\n\t\t}\n\t\treturn generations;\n\t}\n\n\tprivate snapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tsnapshotId: string,\n\t): SnapshotTranscriptGeneration | undefined {\n\t\treturn worker.snapshotGenerations?.get(activeSessionId)?.get(snapshotId);\n\t}\n\n\tprivate currentSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t): SnapshotTranscriptGeneration | undefined {\n\t\tworker.transcriptCaches ??= new Map();\n\t\tworker.snapshotCache ??= new Map();\n\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\tif (!transcript) {\n\t\t\treturn undefined;\n\t\t}\n\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\tlet generation = generations.get(transcript.snapshotId);\n\t\tif (generation) {\n\t\t\treturn generation;\n\t\t}\n\t\tconst result = worker.snapshotCache.get(activeSessionId);\n\t\tif (!result) {\n\t\t\treturn undefined;\n\t\t}\n\t\tgeneration = {\n\t\t\ttranscript,\n\t\t\tresult,\n\t\t\tincoming: false,\n\t\t\tretired: false,\n\t\t};\n\t\tgenerations.set(transcript.snapshotId, generation);\n\t\treturn generation;\n\t}\n\n\tprivate deleteSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tgeneration: SnapshotTranscriptGeneration,\n\t): void {\n\t\tconst generations = worker.snapshotGenerations?.get(activeSessionId);\n\t\tif (!generations) {\n\t\t\treturn;\n\t\t}\n\t\tif (generations.get(generation.transcript.snapshotId) === generation) {\n\t\t\tgenerations.delete(generation.transcript.snapshotId);\n\t\t}\n\t\tif (generations.size === 0) {\n\t\t\tworker.snapshotGenerations.delete(activeSessionId);\n\t\t}\n\t}\n\n\tprivate failSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tgeneration: SnapshotTranscriptGeneration,\n\t\terror: Error,\n\t): void {\n\t\tthis.settleSnapshotDuplicateValidation(generation, error);\n\t\tgeneration.transcript.markFailed(error);\n\t\tgeneration.transcript.dispose();\n\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\tif (worker.transcriptCaches.get(activeSessionId) === generation.transcript) {\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t}\n\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === generation.transcript.snapshotId) {\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t}\n\n\tprivate createSnapshotDuplicateValidation(): SnapshotDuplicateValidation {\n\t\tlet resolve!: () => void;\n\t\tlet reject!: (error: Error) => void;\n\t\tconst promise = new Promise<void>((resolvePromise, rejectPromise) => {\n\t\t\tresolve = resolvePromise;\n\t\t\treject = rejectPromise;\n\t\t});\n\t\tvoid promise.catch(() => undefined);\n\t\treturn { promise, resolve, reject };\n\t}\n\n\tprivate settleSnapshotDuplicateValidation(generation: SnapshotTranscriptGeneration, error?: Error): void {\n\t\tconst validation = generation.validation;\n\t\tif (!validation) {\n\t\t\treturn;\n\t\t}\n\t\tgeneration.validation = undefined;\n\t\tif (error) {\n\t\t\tvalidation.reject(error);\n\t\t} else {\n\t\t\tvalidation.resolve();\n\t\t}\n\t}\n\n\tprivate async recoverWorker(worker: ResidentWorker): Promise<void> {\n\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.ownerClientId && !worker.launchEnv && !isProcessAlive(worker.descriptor.pid)) {\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = \"Waiting for the owning client to reconnect\";\n\t\t\tthis.persistWorker(worker);\n\t\t\treturn;\n\t\t}\n\t\tif (worker.recovery) {\n\t\t\treturn worker.recovery;\n\t\t}\n\t\tworker.recovery = (async () => {\n\t\t\tlet keepProbingLiveWorker = false;\n\t\t\tfor (const retryDelay of WORKER_RETRY_DELAYS_MS) {\n\t\t\t\tawait delay(retryDelay);\n\t\t\t\tkeepProbingLiveWorker = false;\n\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\tconst identityNow = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\t\t\tconst identityCompatible =\n\t\t\t\t\t\tidentityNow === \"current\" ||\n\t\t\t\t\t\t(identityNow === \"unknown\" && worker.descriptor.processStartId === undefined);\n\t\t\t\t\tif (identityCompatible) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\tawait this.connectWorker(worker, 1500);\n\t\t\t\t\t\t\tawait this.subscribeWorker(worker, worker.descriptor.rootActiveSessionId);\n\t\t\t\t\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\t\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (worker.descriptor.processStartId === undefined) {\n\t\t\t\t\t\t\t\tconst observedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\t\t\t\t\t\tif (observedProcessStartId) {\n\t\t\t\t\t\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\t\t\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\t\t\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif (isSupervisorRecoveryCancelled(error) || error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t\t\tworker.client?.close();\n\t\t\t\t\t\t\tworker.client = undefined;\n\t\t\t\t\t\t\t// A worker with the same durable process identity may be load-slow.\n\t\t\t\t\t\t\t// Keep probing it instead of replacing live work after a timeout.\n\t\t\t\t\t\t\tkeepProbingLiveWorker = isDaemonWorkerProbeTimeout(error) || identityNow !== \"current\";\n\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif (identityNow === \"unknown\") {\n\t\t\t\t\t\tkeepProbingLiveWorker = true;\n\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t`Cannot safely replace live session worker ${worker.descriptor.workerId} without a verified process identity`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t\tconst recoveryCommand = worker.descriptor.ownerClientId ? worker.transientCreateCommand : undefined;\n\t\t\t\t\tif (!recoveryCommand || !worker.launchEnv) {\n\t\t\t\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\t\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\t\t\t\tworker.descriptor.lastError = \"Waiting for a client with fresh runtime context\";\n\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tawait this.launchWorker(recoveryCommand, worker, worker.descriptor.ownerClientId);\n\t\t\t\t\treturn;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (isSupervisorRecoveryCancelled(error) || this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t} catch {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tworker.client?.close();\n\t\t\t\t\tworker.client = undefined;\n\t\t\t\t\tworker.descriptor.consecutiveFailures++;\n\t\t\t\t\tworker.descriptor.lastFailureAt = new Date().toISOString();\n\t\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (keepProbingLiveWorker) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t} catch {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.deferWorkerRecovery(\n\t\t\t\t\tworker,\n\t\t\t\t\tnew Error(worker.descriptor.lastError ?? \"Live session worker did not answer recovery probes\"),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// Leak-over-kill: a live worker that keeps failing for non-timeout reasons parks failed with\n\t\t\t// its process intact. A verified-identity survivor is reclaimed by the next fresh create;\n\t\t\t// an unverifiable one waits for exit — killing a pid we cannot verify as ours is worse.\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(`Worker ${worker.descriptor.workerId} failed after three recovery attempts`);\n\t\t})().finally(() => {\n\t\t\tworker.recovery = undefined;\n\t\t});\n\t\treturn worker.recovery;\n\t}\n\n\tprivate isWorkerRecoveryCancelled(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.shuttingDown ||\n\t\t\tworker.intentionalStop ||\n\t\t\tworker.descriptor.stopRequestedAt !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t);\n\t}\n\n\tprivate isWorkerCleanupCancelled(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.shuttingDown ||\n\t\t\tworker.descriptor.stopRequestedAt !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t);\n\t}\n\n\tprivate async recoverUncertainWorkerOperations(worker: ResidentWorker): Promise<void> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tconst journal = new WorkerRecoveryJournal(worker.descriptor.recoveryJournalPath);\n\t\tconst latest = journal.getLatest();\n\t\tconst uncertain = latest.filter((record) => record.busy);\n\t\tif (uncertain.length > 0) await this.catalog.start();\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before destructive cleanup\");\n\t\t}\n\n\t\tconst interruptedSessions = new Map<\n\t\t\tstring,\n\t\t\t{ activeSessionId: string; sessionFile: string; operations: Set<string> }\n\t\t>();\n\t\tfor (const record of uncertain) {\n\t\t\tconst sessionFile =\n\t\t\t\trecord.sessionFile ??\n\t\t\t\t(record.activeSessionId === worker.descriptor.rootActiveSessionId\n\t\t\t\t\t? worker.descriptor.sessionFile\n\t\t\t\t\t: undefined);\n\t\t\tif (!sessionFile) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst key = `${record.activeSessionId}\\0${sessionFile}`;\n\t\t\tlet interrupted = interruptedSessions.get(key);\n\t\t\tif (!interrupted) {\n\t\t\t\tinterrupted = { activeSessionId: record.activeSessionId, sessionFile, operations: new Set() };\n\t\t\t\tinterruptedSessions.set(key, interrupted);\n\t\t\t}\n\t\t\tinterrupted.operations.add(record.operation);\n\t\t}\n\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before interruption was recorded\");\n\t\t}\n\t\tawait Promise.all(\n\t\t\t[...interruptedSessions.values()].map((interrupted) =>\n\t\t\t\tthis.catalog.markInterrupted(interrupted.sessionFile, interrupted.activeSessionId, [\n\t\t\t\t\t...interrupted.operations,\n\t\t\t\t]),\n\t\t\t),\n\t\t);\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before process cleanup\");\n\t\t}\n\t\tconst orphanProcessJournalPath = worker.descriptor.orphanProcessJournalPath;\n\t\tif (orphanProcessJournalPath) {\n\t\t\ttry {\n\t\t\t\tconst orphans = readActiveOrphanProcesses(orphanProcessJournalPath, worker.descriptor.pid);\n\t\t\t\tlet reapFailed = false;\n\t\t\t\tlet retryIsSafe = true;\n\t\t\t\tfor (const orphan of orphans) {\n\t\t\t\t\tif (orphan.processStartId === undefined) retryIsSafe = false;\n\t\t\t\t\tif (!shouldReapOrphanProcess(orphan)) {\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tif (!killOrphanProcess(orphan.pid)) reapFailed = true;\n\t\t\t\t}\n\t\t\t\tif (!reapFailed || !retryIsSafe) clearOrphanProcessJournal(orphanProcessJournalPath);\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Could not reap orphaned worker resources: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t\tif (uncertain.length === 0) {\n\t\t\treturn;\n\t\t}\n\t\tfor (const record of latest) {\n\t\t\tjournal.record({\n\t\t\t\tactiveSessionId: record.activeSessionId,\n\t\t\t\tsessionId: record.sessionId,\n\t\t\t\t...(record.sessionFile ? { sessionFile: record.sessionFile } : {}),\n\t\t\t\tbusy: false,\n\t\t\t\toperation: \"recovery_hold\",\n\t\t\t});\n\t\t}\n\t\tthis.log(\n\t\t\t`Recovered worker ${worker.descriptor.workerId} without replaying uncertain operations: ${uncertain\n\t\t\t\t.map((record) => record.operation)\n\t\t\t\t.join(\", \")}`,\n\t\t);\n\t}\n\tprivate async refreshWorkerSummaries(\n\t\tworker: ResidentWorker,\n\t\trecovery = false,\n\t\tfillGaps = recovery,\n\t\tretried = false,\n\t): Promise<void> {\n\t\tif (this.isWorkerStopping(worker)) {\n\t\t\tthrow new Error(\"Session worker is stopping\");\n\t\t}\n\t\tif (!worker.client) {\n\t\t\tthrow new Error(\"Session worker is not connected\");\n\t\t}\n\t\tconst pullSource = worker.client;\n\t\tconst epochAtStart = worker.rosterEpoch ?? 0;\n\t\tconst response = await pullSource.request({ type: \"list\" }, 5000);\n\t\t// A frame received mid-pull can remove rows this stale pull would resurrect; re-pull once, then skip the fill.\n\t\tif (fillGaps && (worker.rosterEpoch ?? 0) !== epochAtStart && !retried) {\n\t\t\treturn this.refreshWorkerSummaries(worker, recovery, fillGaps, true);\n\t\t}\n\t\tconst summaries = sessionSummariesFromResponse(response);\n\t\tconst nextSummaries = new Map(summaries.map((summary) => [summary.activeSessionId ?? summary.id, summary]));\n\t\tconst root = nextSummaries.get(worker.descriptor.rootActiveSessionId);\n\t\tif (recovery && !root) {\n\t\t\tthrow new Error(`Session worker omitted its root session during recovery`);\n\t\t}\n\t\tworker.summaries = nextSummaries;\n\t\tif (fillGaps) {\n\t\t\tawait this.chainWorkerRosterApply(worker, pullSource, () => {\n\t\t\t\tif ((worker.rosterEpoch ?? 0) === epochAtStart) this.syncRosterFromWorkerSummaries(worker);\n\t\t\t});\n\t\t}\n\t\tfor (const summary of summaries) {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\tif (summary.streamingMessage?.role === \"assistant\") {\n\t\t\t\tthis.streamReconstructor.seed(activeSessionId, summary.streamingMessage);\n\t\t\t} else if (!summary.isStreaming) {\n\t\t\t\tthis.streamReconstructor.clear(activeSessionId);\n\t\t\t}\n\t\t}\n\t\tif (root) {\n\t\t\tif (recovery) {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t}\n\t\t\tawait this.chainWorkerRosterApply(worker, pullSource, () => {\n\t\t\t\tif ((worker.rosterEpoch ?? 0) !== epochAtStart) return;\n\t\t\t\tworker.descriptor.rootSessionId = root.sessionId;\n\t\t\t\tworker.descriptor.sessionFile = root.sessionFile;\n\t\t\t\tworker.descriptor.createCommand = durableDaemonCreateCommand({\n\t\t\t\t\ttype: \"create\",\n\t\t\t\t\tsessionPath: root.sessionFile,\n\t\t\t\t\tnoSession: worker.descriptor.createCommand.noSession,\n\t\t\t\t});\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t});\n\t\t}\n\t}\n\n\tprivate async familyCatalogEntries(): Promise<AgentFamilyCatalogEntry[]> {\n\t\tconst rosterRows = [...this.roster().values()];\n\t\tconst entries = rosterRows.map((entry) => this.familyCatalogEntry(sessionSummaryFromRosterEntry(entry)));\n\t\tconst knownFiles = new Set(\n\t\t\trosterRows.flatMap((entry) =>\n\t\t\t\tentry.summary.sessionFile ? [canonicalSessionPath(entry.summary.sessionFile)] : [],\n\t\t\t),\n\t\t);\n\t\tconst scanned = await this.catalog.list(undefined, this.defaultSessionConfig.sessionDir);\n\t\tfor (const info of scanned) {\n\t\t\tif (knownFiles.has(canonicalSessionPath(info.path))) continue;\n\t\t\tif ((info.rlmDepth ?? (info.parentSessionPath ? -1 : 0)) !== 0) continue;\n\t\t\tentries.push(this.familyCatalogEntry(summaryForInactiveSession(info)));\n\t\t}\n\t\treturn entries;\n\t}\n\n\tprivate async withSessionNameReservation<T>(\n\t\tinput: { name: string; depth: number; parentSessionId?: string; parentSessionPath?: string },\n\t\taction: () => Promise<T>,\n\t): Promise<T> {\n\t\tconst key = sessionNameReservationKey(input);\n\t\tif (this.pendingSessionNames.has(key)) {\n\t\t\tthrow new Error(formatAgentSessionNameUnavailable(input.name, input.depth));\n\t\t}\n\t\tthis.pendingSessionNames.add(key);\n\t\ttry {\n\t\t\treturn await action();\n\t\t} finally {\n\t\t\tthis.pendingSessionNames.delete(key);\n\t\t}\n\t}\n\n\tprivate async assertSupervisorSessionNameAvailable(\n\t\ttarget: Pick<SessionSummary, \"sessionId\" | \"rlmDepth\" | \"parentSessionId\" | \"parentSessionPath\">,\n\t\tname: string,\n\t): Promise<void> {\n\t\tassertAgentSessionNameAvailable(await this.familyCatalogEntries(), {\n\t\t\tname,\n\t\t\tdepth: target.rlmDepth ?? 0,\n\t\t\tparentSessionId: target.parentSessionId,\n\t\t\tparentSessionPath: target.parentSessionPath ? canonicalSessionPath(target.parentSessionPath) : undefined,\n\t\t\tignoreSessionId: target.sessionId,\n\t\t});\n\t}\n\n\tprivate roster(): AgentRoster {\n\t\tthis.rosterStore ??= new AgentRoster(canonicalSessionPath, (mutation) => this.onRosterMutation(mutation));\n\t\treturn this.rosterStore;\n\t}\n\n\tprivate onRosterMutation(mutation: AgentRosterMutation): void {\n\t\tif (mutation.type === \"delete\") {\n\t\t\tthis.pendingRosterChanged.delete(mutation.agentId);\n\t\t\tthis.pendingRosterRemoved.add(mutation.agentId);\n\t\t} else {\n\t\t\tthis.pendingRosterRemoved.delete(mutation.agentId);\n\t\t\tthis.pendingRosterChanged.add(mutation.agentId);\n\t\t}\n\t\tthis.scheduleRosterPush();\n\t}\n\n\tprivate scheduleRosterPush(): void {\n\t\tif (this.rosterPushScheduled || this.shuttingDown) return;\n\t\tthis.rosterPushScheduled = true;\n\t\tsetImmediate(() => {\n\t\t\tthis.rosterPushScheduled = false;\n\t\t\tthis.flushRosterUpdates();\n\t\t});\n\t}\n\n\tprivate flushRosterUpdates(): void {\n\t\tconst changed: AgentRosterEntry[] = [];\n\t\tconst removed: string[] = [];\n\t\tfor (const agentId of this.pendingRosterRemoved) {\n\t\t\tif (this.publishedRosterIds.delete(agentId)) removed.push(agentId);\n\t\t}\n\t\tfor (const agentId of this.pendingRosterChanged) {\n\t\t\tconst entry = this.roster().get(agentId);\n\t\t\tif (!entry) continue;\n\t\t\tif (this.isRosterEntryVisibleToClients(entry)) {\n\t\t\t\tchanged.push(entry);\n\t\t\t\tthis.publishedRosterIds.add(agentId);\n\t\t\t} else if (this.publishedRosterIds.delete(agentId)) {\n\t\t\t\tremoved.push(agentId);\n\t\t\t}\n\t\t}\n\t\tthis.pendingRosterChanged.clear();\n\t\tthis.pendingRosterRemoved.clear();\n\t\tif (changed.length === 0 && removed.length === 0) return;\n\t\tfor (const client of this.clients) {\n\t\t\tif (client.rosterSubscribed !== true) continue;\n\t\t\tif (client.backpressured === true) {\n\t\t\t\tclient.rosterResyncPending = true;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.write(client, {\n\t\t\t\ttype: \"roster_update\",\n\t\t\t\tchanged,\n\t\t\t\t...(removed.length > 0 ? { removed } : {}),\n\t\t\t});\n\t\t}\n\t}\n\n\tprivate rosterEntriesForClient(): AgentRosterEntry[] {\n\t\tconst entries = [...this.roster().values()].filter((entry) => this.isRosterEntryVisibleToClients(entry));\n\t\tfor (const entry of entries) this.publishedRosterIds.add(entry.agentId);\n\t\treturn entries;\n\t}\n\n\tprivate isRosterEntryVisibleToClients(entry: AgentRosterEntry): boolean {\n\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\treturn worker === undefined || this.isVisibleWorker(worker);\n\t}\n\n\tprivate writeRosterEntry(\n\t\tentry: WorkerRosterEntry,\n\t\tworker?: ResidentWorker,\n\t\tstatusLabel?: AgentRosterEntry[\"statusLabel\"],\n\t): AgentRosterEntry {\n\t\tconst previousDirect = this.roster().get(entry.agentId)?.summary.directAttachedClients ?? 0;\n\t\tconst stored = this.roster().write(entry, worker?.descriptor.workerId, statusLabel);\n\t\t// Direct peers attach and detach on the worker socket, so their last detach arrives\n\t\t// here as roster truth instead of through a supervisor-socket close.\n\t\tif (worker !== undefined && previousDirect > 0 && (entry.summary.directAttachedClients ?? 0) === 0) {\n\t\t\tvoid this.evictEmptySessionOnLastDetach(entry.summary.activeSessionId ?? entry.summary.id);\n\t\t}\n\t\treturn stored;\n\t}\n\n\tprivate workerOwnedRosterSummaryForPath(canonicalPath: string): SessionSummary | undefined {\n\t\tconst entry = this.roster().bySessionFile(canonicalPath);\n\t\tif (!entry || entry.workerId === undefined || !this.workers.has(entry.workerId)) return undefined;\n\t\treturn sessionSummaryFromRosterEntry(entry);\n\t}\n\n\tprivate workerRosterEntries(worker: ResidentWorker): AgentRosterEntry[] {\n\t\treturn this.roster().entriesForWorker(worker.descriptor.workerId);\n\t}\n\n\tprivate async seedRosterLedger(): Promise<void> {\n\t\ttry {\n\t\t\tconst roots = new Set<string>();\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tconst root = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\t\t\tif (root !== undefined) roots.add(canonicalSessionPath(root));\n\t\t\t}\n\t\t\tif (roots.size === 0) return;\n\t\t\tconst edges = await this.rlmSpawnLedger().liveEdges();\n\t\t\tconst descendsFrom = rosterFamilyDescendsFrom(edges);\n\t\t\tfor (const edge of edges) {\n\t\t\t\tif (!descendsFrom(canonicalSessionPath(edge.parent), roots)) continue;\n\t\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\t\tif (this.roster().hasSessionFile(canonicalSessionPath(edge.child))) continue;\n\t\t\t\tthis.roster().write(await this.hydratedSeedEntry(entry));\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Could not seed the agent roster from the spawn ledger: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate rosterEntryForSpawnLedgerEdge(edge: RlmLedgerEdge): WorkerRosterEntry {\n\t\tconst persistedSessionId = basename(edge.child, \".jsonl\");\n\t\tconst summary: WorkerRosterEntry[\"summary\"] = {\n\t\t\tid: persistedSessionId,\n\t\t\tlifecycle: \"live\",\n\t\t\tactivity: \"idle\",\n\t\t\tisSessionActive: false,\n\t\t\truntimeKind: \"subagent\",\n\t\t\trlmDepth: edge.depth,\n\t\t\tsessionId: persistedSessionId,\n\t\t\tsessionFile: edge.child,\n\t\t\tsessionName: edge.name,\n\t\t\tcwd: dirname(edge.child),\n\t\t\tisStreaming: false,\n\t\t\tisCompacting: false,\n\t\t\tattachedClients: 0,\n\t\t\tmessageCount: 0,\n\t\t\tparentSessionPath: edge.parent,\n\t\t\trlmChildId: edge.childId,\n\t\t};\n\t\treturn { agentId: rosterAgentIdForSummary(summary), summary };\n\t}\n\n\tprivate consumeWorkerRosterDelta(worker: ResidentWorker, payload: Buffer, source?: DaemonWorkerClient): void {\n\t\tlet delta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>;\n\t\ttry {\n\t\t\tdelta = JSON.parse(payload.toString(\"utf8\")) as Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>;\n\t\t} catch {\n\t\t\treturn;\n\t\t}\n\t\tif (delta.type !== \"roster_delta\" || !Array.isArray(delta.entries)) return;\n\t\tworker.rosterEpoch = (worker.rosterEpoch ?? 0) + 1;\n\t\tconst applySource = source ?? worker.client ?? worker.pendingClient;\n\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\tif (delta.snapshot !== true && worker.rosterApplyChain === undefined) {\n\t\t\tthis.applyWorkerRosterDelta(worker, delta);\n\t\t\treturn;\n\t\t}\n\t\tthis.chainWorkerRosterApply(worker, applySource, () =>\n\t\t\tdelta.snapshot === true\n\t\t\t\t? this.applyWorkerRosterSnapshot(worker, delta, applySource)\n\t\t\t\t: this.applyWorkerRosterDelta(worker, delta),\n\t\t);\n\t}\n\n\tprivate chainWorkerRosterApply(\n\t\tworker: ResidentWorker,\n\t\tsource: DaemonWorkerClient | undefined,\n\t\tapply: () => void | Promise<void>,\n\t): Promise<void> {\n\t\tconst chained = (worker.rosterApplyChain ?? Promise.resolve())\n\t\t\t.then(() => {\n\t\t\t\tif (!this.isWorkerRosterApplyCurrent(worker, source)) return;\n\t\t\t\treturn apply();\n\t\t\t})\n\t\t\t.catch((error: unknown) => {\n\t\t\t\tthis.log(`could not apply a roster frame: ${String(error)}`);\n\t\t\t\tthis.scheduleRosterRepairPull(worker);\n\t\t\t});\n\t\tworker.rosterApplyChain = chained;\n\t\tvoid chained.finally(() => {\n\t\t\tif (worker.rosterApplyChain === chained) worker.rosterApplyChain = undefined;\n\t\t});\n\t\treturn chained;\n\t}\n\n\t// An apply is valid only while its own source connection is current: dead connections' parked applies abort.\n\tprivate isWorkerRosterApplyCurrent(worker: ResidentWorker, source: DaemonWorkerClient | undefined): boolean {\n\t\treturn (\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tsource !== undefined &&\n\t\t\t(source === worker.client || source === worker.pendingClient)\n\t\t);\n\t}\n\n\tprivate scheduleRosterRepairPull(worker: ResidentWorker): void {\n\t\tif (worker.rosterRepairPull || !this.isWorkerRosterApplyCurrent(worker, worker.client)) return;\n\t\t// The marker stays set while the repair's own fill applies, so a failing repair never respawns itself.\n\t\tworker.rosterRepairPull = this.refreshWorkerSummaries(worker, false, true)\n\t\t\t.catch((error: unknown) =>\n\t\t\t\tthis.log(`Roster repair pull failed for worker ${worker.descriptor.workerId}: ${String(error)}`),\n\t\t\t)\n\t\t\t.finally(() => {\n\t\t\t\tworker.rosterRepairPull = undefined;\n\t\t\t});\n\t}\n\n\tprivate applyWorkerRosterDelta(\n\t\tworker: ResidentWorker,\n\t\tdelta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>,\n\t): void {\n\t\tfor (const entry of delta.entries) {\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t\tthis.syncRootDescriptorFromRosterEntry(worker, entry);\n\t\t}\n\t\tfor (const agentId of delta.removedAgentIds ?? []) {\n\t\t\tthis.roster().delete(agentId);\n\t\t}\n\t}\n\n\tprivate async applyWorkerRosterSnapshot(\n\t\tworker: ResidentWorker,\n\t\tdelta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>,\n\t\tsource?: DaemonWorkerClient,\n\t): Promise<void> {\n\t\tlet edgesFailed = false;\n\t\tconst edges = await this.rlmSpawnLedger()\n\t\t\t.liveEdges()\n\t\t\t.catch((error: unknown) => {\n\t\t\t\tthis.log(`Could not read the spawn ledger during a snapshot apply: ${String(error)}`);\n\t\t\t\tedgesFailed = true;\n\t\t\t\treturn [] as RlmLedgerEdge[];\n\t\t\t});\n\t\tconst applySource = source ?? worker.client ?? worker.pendingClient;\n\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\tconst sent = new Set(delta.entries.map((entry) => entry.agentId));\n\t\tconst removed = new Set(delta.removedAgentIds ?? []);\n\t\tconst unclaimed = new Map<string, AgentRosterEntry>();\n\t\tif (!edgesFailed) {\n\t\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\t\tif (sent.has(entry.agentId)) continue;\n\t\t\t\tunclaimed.set(entry.agentId, entry);\n\t\t\t}\n\t\t}\n\t\t// Only this worker's family reseeds: anything wider can resurrect a client-owned worker's dropped children.\n\t\tconst workerRoot = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tconst rootPaths = new Set(workerRoot !== undefined ? [canonicalSessionPath(workerRoot)] : []);\n\t\tconst descendsFrom = rosterFamilyDescendsFrom(edges);\n\t\tconst familyEdges = edges.filter((edge) => descendsFrom(canonicalSessionPath(edge.parent), rootPaths));\n\t\t// \"Unclaimed\" rows survive: the sweep deletes them but the restore branch rewrites them.\n\t\tconst rowSurvivesWithoutReseed = (entry: WorkerRosterEntry, childPath: string): boolean => {\n\t\t\tif (unclaimed.has(entry.agentId)) return true;\n\t\t\tif (sent.has(entry.agentId) && !removed.has(entry.agentId)) return true;\n\t\t\tconst survives = (row: AgentRosterEntry | undefined): boolean =>\n\t\t\t\trow !== undefined && !unclaimed.has(row.agentId) && !removed.has(row.agentId);\n\t\t\treturn survives(this.roster().get(entry.agentId)) || survives(this.roster().bySessionFile(childPath));\n\t\t};\n\t\tconst seededEntries = new Map<string, WorkerRosterEntry>();\n\t\tfor (const edge of familyEdges) {\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (rowSurvivesWithoutReseed(entry, canonicalSessionPath(edge.child))) continue;\n\t\t\tseededEntries.set(entry.agentId, await this.hydratedSeedEntry(entry));\n\t\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\t}\n\n\t\t// Unreadable edges skip the absentee sweep: it cannot tell registry children from stale rows.\n\t\tfor (const entry of unclaimed.values()) this.roster().delete(entry.agentId);\n\t\tfor (const entry of delta.entries) {\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t\tthis.syncRootDescriptorFromRosterEntry(worker, entry);\n\t\t}\n\t\tfor (const agentId of removed) this.roster().delete(agentId);\n\t\tif (edgesFailed) {\n\t\t\tthis.scheduleRosterRepairPull(worker);\n\t\t\treturn;\n\t\t}\n\t\tfor (const edge of familyEdges) {\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\tif (this.roster().hasSessionFile(canonicalSessionPath(edge.child))) continue;\n\t\t\tconst previous = unclaimed.get(entry.agentId);\n\t\t\tif (previous) {\n\t\t\t\tconst { status, statusLabel, lastHeardFromAt, workerId, ...rest } = previous;\n\t\t\t\tthis.writeRosterEntry(rest, worker);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.roster().write(seededEntries.get(entry.agentId) ?? { ...entry, seededCwd: true });\n\t\t}\n\t}\n\n\tprivate syncRootDescriptorFromRosterEntry(worker: ResidentWorker, entry: WorkerRosterEntry): void {\n\t\tconst summary = entry.summary;\n\t\tif (summary.activeSessionId !== worker.descriptor.rootActiveSessionId) return;\n\t\tif (\n\t\t\tworker.descriptor.rootSessionId === summary.sessionId &&\n\t\t\tworker.descriptor.sessionFile === summary.sessionFile\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tworker.descriptor.rootSessionId = summary.sessionId;\n\t\tworker.descriptor.sessionFile = summary.sessionFile;\n\t\tworker.descriptor.createCommand = durableDaemonCreateCommand({\n\t\t\ttype: \"create\",\n\t\t\tsessionPath: summary.sessionFile,\n\t\t\tnoSession: worker.descriptor.createCommand.noSession,\n\t\t});\n\t\tthis.persistWorker(worker);\n\t}\n\n\t// Behind the pull-epoch guard the pull is never staler than the row it replaces; never steal another worker's claim.\n\tprivate syncRosterFromWorkerSummaries(worker: ResidentWorker): void {\n\t\tfor (const summary of worker.summaries.values()) {\n\t\t\tconst entry = workerRosterEntryFromSummary(summary);\n\t\t\tconst existing = this.roster().get(entry.agentId);\n\t\t\tif (existing?.workerId !== undefined && existing.workerId !== worker.descriptor.workerId) continue;\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t}\n\t}\n\n\tprivate markWorkerRosterEntries(worker: ResidentWorker, statusLabel: \"recovering\" | \"failed\" | undefined): void {\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tif (!entry.queuedChild && entry.summary.activeSessionId === undefined) continue;\n\t\t\tthis.roster().amend(entry.agentId, { statusLabel });\n\t\t}\n\t}\n\n\tprivate flipWorkerRosterEntriesInactive(worker: ResidentWorker): void {\n\t\t// Client-owned workers are ephemeral and private: their rows die with the registration.\n\t\tconst ephemeral = worker.descriptor.ownerClientId !== undefined;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tif (ephemeral || entry.queuedChild) {\n\t\t\t\tthis.roster().delete(entry.agentId);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// Registration marks survive eviction: passive rows still have schedules behind them.\n\t\t\tthis.writeRosterEntry(\n\t\t\t\tpassivatedWorkerRosterEntry(entry, {\n\t\t\t\t\thasRegisteredHeartbeat: entry.summary.hasRegisteredHeartbeat === true,\n\t\t\t\t\thasRegisteredCronJob: entry.summary.hasRegisteredCronJob === true,\n\t\t\t\t}),\n\t\t\t);\n\t\t}\n\t}\n\n\tprivate sweepRosterStaleness(now = Date.now()): void {\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.client === undefined || worker.lastFrameAt === undefined) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (now - worker.lastFrameAt > ROSTER_STALE_AFTER_MS) {\n\t\t\t\tconst lastHeardFromAt = new Date(worker.lastFrameAt).toISOString();\n\t\t\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\t\t\t// write() rebuilds rows without the mark; the sweep owns it and restamps only those.\n\t\t\t\t\tif (entry.lastHeardFromAt !== lastHeardFromAt) this.roster().amend(entry.agentId, { lastHeardFromAt });\n\t\t\t\t}\n\t\t\t\tworker.rosterStale = true;\n\t\t\t} else if (worker.rosterStale) {\n\t\t\t\tthis.clearRosterStaleness(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate clearRosterStaleness(worker: ResidentWorker): void {\n\t\tif (!worker.rosterStale) return;\n\t\tworker.rosterStale = false;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tthis.roster().amend(entry.agentId, { lastHeardFromAt: undefined });\n\t\t}\n\t}\n\n\tprivate rlmJournalScope(sessionDir?: string): { agentDir: string; sessionsDir: string; journalPath: string } {\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) throw new Error(\"Daemon supervisor config is missing agentDir\");\n\t\tconst sessionsDir = sessionDir ?? this.defaultSessionConfig.sessionDir ?? getSessionsDir(agentDir);\n\t\treturn { agentDir, sessionsDir, journalPath: rlmLedgerPath(agentDir, sessionsDir) };\n\t}\n\n\tprivate assertRlmJournalAdmission(): RlmJournalOwner {\n\t\tconst owner = this.rlmJournalOwner;\n\t\tif (!owner || this.rlmJournalClose) throw new Error(\"RLM journal mutation admission is closed\");\n\t\ttry {\n\t\t\tif (!this.ownership) throw new Error(\"RLM journal supervisor ownership is unavailable\");\n\t\t\tthis.ownership.assertJournalCurrent(owner.ledgerPath);\n\t\t} catch (error) {\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t\treturn owner;\n\t}\n\n\tprivate fenceRlmJournalOwner(): void {\n\t\tvoid this.closeRlmJournalOwner().catch((error: unknown) => this.reportCleanupFailure(\"RLM journal owner\", error));\n\t}\n\n\tprivate closeRlmJournalOwner(): Promise<void> {\n\t\tthis.rlmSpawnLedgerInstance?.stopAdmission();\n\t\tif (this.rlmJournalClose) return this.rlmJournalClose;\n\t\tconst owner = this.rlmJournalOwner;\n\t\tif (!owner) return Promise.resolve();\n\t\tthis.rlmJournalClose = (async () => {\n\t\t\tawait this.rlmSpawnLedgerInstance?.flush();\n\t\t\tawait owner.close();\n\t\t\tthis.rlmJournalOwner = undefined;\n\t\t})();\n\t\treturn this.rlmJournalClose;\n\t}\n\n\t/** Reads are local; every mutation is acknowledged by the external lifetime owner. */\n\tprivate rlmSpawnLedger(): RlmSpawnLedger {\n\t\tconst { agentDir, sessionsDir } = this.rlmJournalScope();\n\t\tthis.rlmSpawnLedgerInstance ??= new RlmSpawnLedger(\n\t\t\tagentDir,\n\t\t\tsessionsDir,\n\t\t\tcreateRlmLedgerRegistrySeedSource(),\n\t\t\t(message) => this.log(message),\n\t\t\t{\n\t\t\t\tmode: \"remote\",\n\t\t\t\tmutate: (mutation) => {\n\t\t\t\t\tconst owner = this.rlmJournalOwner;\n\t\t\t\t\tif (!owner) throw new Error(\"RLM journal owner is unavailable\");\n\t\t\t\t\treturn owner.mutate(mutation);\n\t\t\t\t},\n\t\t\t},\n\t\t);\n\t\treturn this.rlmSpawnLedgerInstance;\n\t}\n\n\t/**\n\t * Ledger-backed same-parent rows for name reservation and admission. Rows\n\t * carry ledger topology plus best-effort display fields; consumers here\n\t * only need id/path/name/depth/parent.\n\t */\n\tprivate rlmLedgerSiblings(sessionPath: string): Promise<SessionInfo[]> {\n\t\treturn this.rlmSpawnLedger().siblings(sessionPath);\n\t}\n\n\tprivate async savedSessionNameReservationInput(\n\t\tsessionPath: string,\n\t\tname: string,\n\t): Promise<{ name: string; depth: number; parentSessionId?: string; parentSessionPath?: string }> {\n\t\tconst targetPath = canonicalSessionPath(sessionPath);\n\t\tconst active = this.workerOwnedRosterSummaryForPath(targetPath);\n\t\tif (active) return this.summaryNameReservationInput(active, name);\n\t\tconst siblings = await this.rlmLedgerSiblings(sessionPath);\n\t\tconst saved = siblings.find((info) => canonicalSessionPath(info.path) === targetPath);\n\t\tif (!saved) throw new Error(`Session not found: ${sessionPath}`);\n\t\treturn {\n\t\t\tname,\n\t\t\tdepth: saved.rlmDepth ?? siblings.find((sibling) => sibling.rlmDepth !== undefined)?.rlmDepth ?? 0,\n\t\t\tparentSessionPath: saved.parentSessionPath,\n\t\t};\n\t}\n\n\tprivate summaryNameReservationInput(\n\t\ttarget: Pick<SessionSummary, \"rlmDepth\" | \"parentSessionId\" | \"parentSessionPath\">,\n\t\tname: string,\n\t): { name: string; depth: number; parentSessionId?: string; parentSessionPath?: string } {\n\t\tconst depth = target.rlmDepth ?? (target.parentSessionPath ? 1 : 0);\n\t\treturn {\n\t\t\tname,\n\t\t\tdepth,\n\t\t\t...(depth > 0 && target.parentSessionId ? { parentSessionId: target.parentSessionId } : {}),\n\t\t\t...(depth > 0 && target.parentSessionPath ? { parentSessionPath: target.parentSessionPath } : {}),\n\t\t};\n\t}\n\n\tprivate async assertSupervisorSavedSessionNameAvailable(sessionPath: string, name: string): Promise<void> {\n\t\tconst targetPath = canonicalSessionPath(sessionPath);\n\t\tconst active = this.workerOwnedRosterSummaryForPath(targetPath);\n\t\tif (active) return this.assertSupervisorSessionNameAvailable(active, name);\n\t\tconst siblings = await this.rlmLedgerSiblings(sessionPath);\n\t\tconst saved = siblings.find((info) => canonicalSessionPath(info.path) === targetPath);\n\t\tif (!saved) throw new Error(`Session not found: ${sessionPath}`);\n\t\tif (saved.parentSessionPath && (saved.rlmDepth ?? 0) > 0) {\n\t\t\tthis.assertSavedSiblingNameAvailable(siblings, saved, name);\n\t\t} else {\n\t\t\tawait this.assertSupervisorSessionNameAvailable(summaryForInactiveSession(saved), name);\n\t\t}\n\t}\n\n\tprivate assertSavedSiblingNameAvailable(siblings: SessionInfo[], target: SessionInfo, name: string): void {\n\t\tconst setDepth = target.rlmDepth ?? siblings.find((sibling) => sibling.rlmDepth !== undefined)?.rlmDepth ?? 0;\n\t\tassertAgentSessionNameAvailable(\n\t\t\tsiblings.map((info) => {\n\t\t\t\tconst summary = summaryForInactiveSession(info);\n\t\t\t\tconst ledgerRow = this.roster().bySessionFile(canonicalSessionPath(info.path));\n\t\t\t\treturn {\n\t\t\t\t\tid: summary.sessionId,\n\t\t\t\t\t...(summary.sessionName ? { name: summary.sessionName } : {}),\n\t\t\t\t\tdepth: setDepth,\n\t\t\t\t\tstatus: ledgerRow?.status ?? classifySessionRosterStatus(summary),\n\t\t\t\t\t...(summary.parentSessionPath\n\t\t\t\t\t\t? { parentSessionPath: canonicalSessionPath(summary.parentSessionPath) }\n\t\t\t\t\t\t: {}),\n\t\t\t\t};\n\t\t\t}),\n\t\t\t{\n\t\t\t\tname,\n\t\t\t\tdepth: setDepth,\n\t\t\t\tparentSessionPath: target.parentSessionPath ? canonicalSessionPath(target.parentSessionPath) : undefined,\n\t\t\t\tignoreSessionId: target.id,\n\t\t\t},\n\t\t);\n\t}\n\n\tprivate isVisibleWorker(worker: ResidentWorker): boolean {\n\t\treturn worker.descriptor.ownerClientId === undefined;\n\t}\n\n\t/** A worker with a durable or in-memory stop intent is stopping, never live. */\n\tprivate isWorkerStopping(worker: ResidentWorker): boolean {\n\t\treturn worker.intentionalStop || worker.descriptor.stopRequestedAt !== undefined;\n\t}\n\n\t/** Live workers are visible to all clients and not stopping. */\n\tprivate isLiveWorker(worker: ResidentWorker): boolean {\n\t\treturn this.isVisibleWorker(worker) && !this.isWorkerStopping(worker);\n\t}\n\n\t/**\n\t * The lifecycle reported to clients. A stop intent always wins, and a worker\n\t * whose process connection is gone is never reported as \"ready\".\n\t */\n\tprivate effectiveWorkerState(worker: ResidentWorker): DaemonWorkerLifecycle {\n\t\tif (this.isWorkerStopping(worker)) {\n\t\t\treturn \"stopping\";\n\t\t}\n\t\tif (worker.descriptor.lifecycle === \"ready\" && worker.client === undefined) {\n\t\t\treturn \"recovering\";\n\t\t}\n\t\treturn worker.descriptor.lifecycle;\n\t}\n\n\tprivate requireAvailableWorkerClient(worker: ResidentWorker, allowStopping = false): DaemonWorkerClient {\n\t\tif (\n\t\t\t!worker.client ||\n\t\t\tworker.descriptor.lifecycle !== \"ready\" ||\n\t\t\t(!allowStopping && this.isWorkerStopping(worker))\n\t\t) {\n\t\t\tthrow new Error(`Session worker is ${this.effectiveWorkerState(worker)}`);\n\t\t}\n\t\treturn worker.client;\n\t}\n\n\tprivate async issuePeerTransport(\n\t\tworker: ResidentWorker,\n\t\tsummary: SessionSummary,\n\t): Promise<DaemonPeerTransportTicket> {\n\t\tawait this.assertCurrentOwnership();\n\t\tif (!worker.peerTransportCapable) {\n\t\t\tthrow new Error(\"Session worker does not support direct peer transport\");\n\t\t}\n\t\tawait this.refreshWorkerSummaries(worker);\n\t\tconst workerClient = this.requireAvailableWorkerClient(worker);\n\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\tconst currentSummary = this.findSummaryInWorker(worker, activeSessionId);\n\t\tif (!currentSummary) {\n\t\t\tthrow new Error(\"Direct transport target changed during admission\");\n\t\t}\n\t\tconst workerInstanceId = worker.descriptor.workerInstanceId;\n\t\tconst workerProcessStartId = worker.descriptor.processStartId;\n\t\tif (!workerInstanceId || !workerProcessStartId) {\n\t\t\tthrow new Error(\"Direct transport requires an exact worker process identity\");\n\t\t}\n\t\tif (this.processIdentity(worker.descriptor.pid, workerProcessStartId) !== \"current\") {\n\t\t\tthrow new Error(\"Direct transport worker process identity is not current\");\n\t\t}\n\t\tlet socketIdentity: DaemonSocketIdentity | undefined;\n\t\ttry {\n\t\t\tsocketIdentity = getDaemonSocketIdentity(worker.descriptor.socketPath);\n\t\t} catch {\n\t\t\tsocketIdentity = undefined;\n\t\t}\n\t\tif (!socketIdentity) {\n\t\t\tthrow new Error(\"Direct transport requires an exact worker socket identity\");\n\t\t}\n\t\tconst grantId = randomUUID();\n\t\tconst token = randomBytes(32).toString(\"base64url\");\n\t\tconst expiresAt = new Date(Date.now() + PEER_TRANSPORT_GRANT_TTL_MS).toISOString();\n\t\tconst resolvedActiveSessionId = currentSummary.activeSessionId ?? currentSummary.id;\n\t\tconst registration = await workerClient.requestWorker(\n\t\t\t{\n\t\t\t\ttype: \"worker_register_peer_transport\",\n\t\t\t\tgrant: {\n\t\t\t\t\tgrantId,\n\t\t\t\t\ttoken,\n\t\t\t\t\texpiresAt,\n\t\t\t\t\tpurpose: \"session_client\",\n\t\t\t\t\tworkerInstanceId,\n\t\t\t\t\tactiveSessionId: resolvedActiveSessionId,\n\t\t\t\t\tissuerGeneration: this.generation,\n\t\t\t\t},\n\t\t\t},\n\t\t\t3000,\n\t\t);\n\t\tif (!registration.success) {\n\t\t\tthrow deserializeDaemonError(registration);\n\t\t}\n\t\treturn {\n\t\t\tpurpose: \"session_client\",\n\t\t\tsocketPath: worker.descriptor.socketPath,\n\t\t\tsocketIdentity,\n\t\t\tworkerInstanceId,\n\t\t\tactiveSessionId: resolvedActiveSessionId,\n\t\t\tgrantId,\n\t\t\ttoken,\n\t\t\texpiresAt,\n\t\t};\n\t}\n\n\tprivate familyCatalogEntry(summary: SessionSummary): AgentFamilyCatalogEntry {\n\t\tconst depth = summary.rlmDepth ?? (summary.parentSessionPath ? 1 : 0);\n\t\treturn {\n\t\t\tid: summary.sessionId,\n\t\t\t...(summary.sessionName ? { name: summary.sessionName } : {}),\n\t\t\tdepth,\n\t\t\tstatus: summary.rosterStatus ?? classifySessionRosterStatus(summary),\n\t\t\t...(depth > 0 && summary.parentSessionId ? { parentSessionId: summary.parentSessionId } : {}),\n\t\t\t...(depth > 0 && summary.parentSessionPath\n\t\t\t\t? { parentSessionPath: canonicalSessionPath(summary.parentSessionPath) }\n\t\t\t\t: {}),\n\t\t\t...(summary.sessionFile ? { sessionPath: canonicalSessionPath(summary.sessionFile) } : {}),\n\t\t};\n\t}\n\n\tprivate agentPeerSummary(summary: SessionSummary): AgentSessionMessageAgentSummary {\n\t\treturn {\n\t\t\tactiveSessionId: summary.activeSessionId ?? summary.id,\n\t\t\tsessionId: summary.sessionId,\n\t\t\t...(summary.sessionName ? { sessionName: summary.sessionName } : {}),\n\t\t\truntimeKind: summary.runtimeKind ?? \"top-level\",\n\t\t\tcwd: summary.cwd,\n\t\t\tisStreaming: summary.isStreaming,\n\t\t\tunfinishedActionCount:\n\t\t\t\tsummary.unfinishedActionCount ??\n\t\t\t\t(summary.sessionActions.active\n\t\t\t\t\t? 1 + summary.sessionActions.queuedCount\n\t\t\t\t\t: summary.sessionActions.queuedCount),\n\t\t\t...(summary.parentActiveSessionId ? { parentActiveSessionId: summary.parentActiveSessionId } : {}),\n\t\t\t...(summary.parentSessionId ? { parentSessionId: summary.parentSessionId } : {}),\n\t\t\t...(summary.parentSessionPath ? { parentSessionPath: summary.parentSessionPath } : {}),\n\t\t\t...(summary.sessionFile ? { sessionPath: summary.sessionFile } : {}),\n\t\t\t...(summary.rlmDepth !== undefined ? { rlmDepth: summary.rlmDepth } : {}),\n\t\t\tstatus: summary.rosterStatus ?? classifySessionRosterStatus(summary),\n\t\t\t...(summary.rlmChildId ? { rlmChildId: summary.rlmChildId } : {}),\n\t\t};\n\t}\n\n\tprivate attachedClientCount(summary: SessionSummary, activeSessionId: string): number {\n\t\treturn (\n\t\t\t(summary.directAttachedClients ?? 0) +\n\t\t\t[...this.clients].filter((client) => client.attachedActiveSessionIds.has(activeSessionId)).length\n\t\t);\n\t}\n\n\tprivate publicSummary(worker: ResidentWorker, summary: SessionSummary): SessionSummary {\n\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\treturn {\n\t\t\t...summary,\n\t\t\tattachedClients: this.attachedClientCount(summary, activeSessionId),\n\t\t\tworkerState: this.effectiveWorkerState(worker),\n\t\t\tworkerPid: worker.descriptor.pid,\n\t\t};\n\t}\n\n\tprivate async findWorker(\n\t\tselector: string,\n\t\tincludeWorker?: (worker: ResidentWorker) => boolean,\n\t): Promise<WorkerMatch> {\n\t\tlet matches = this.matchWorkers(selector, includeWorker);\n\t\tif (matches.length === 0) {\n\t\t\tawait Promise.all(\n\t\t\t\t[...this.workers.values()].map((worker) =>\n\t\t\t\t\tthis.refreshWorkerSummaries(worker, false, true).catch(() => undefined),\n\t\t\t\t),\n\t\t\t);\n\t\t\tmatches = this.matchWorkers(selector, includeWorker);\n\t\t}\n\t\tif (matches.length === 1) {\n\t\t\treturn matches[0]!;\n\t\t}\n\t\tif (matches.length > 1) {\n\t\t\tthrow new Error(`Ambiguous active session \"${selector}\"`);\n\t\t}\n\t\tthrow new Error(`Unknown active session: ${selector}`);\n\t}\n\n\tprivate findWorkerForClient(client: DaemonSocketClient, selector: string): Promise<WorkerMatch> {\n\t\treturn this.findWorker(selector, (worker) => this.isWorkerAccessibleToClient(client, worker));\n\t}\n\n\tprivate isWorkerAccessibleToClient(client: DaemonSocketClient, worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tworker.descriptor.ownerClientId === undefined ||\n\t\t\tworker.descriptor.ownerClientId === this.protocolClientId(client)\n\t\t);\n\t}\n\n\tprivate assertWorkerAccessibleToClient(client: DaemonSocketClient, worker: ResidentWorker, selector: string): void {\n\t\tif (!this.isWorkerAccessibleToClient(client, worker)) {\n\t\t\tthrow new Error(`Unknown active session: ${selector}`);\n\t\t}\n\t}\n\n\tprivate matchWorkers(selector: string, includeWorker?: (worker: ResidentWorker) => boolean): WorkerMatch[] {\n\t\tconst exact: WorkerMatch[] = [];\n\t\tconst suffix: WorkerMatch[] = [];\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild) continue;\n\t\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\tif (!worker || (includeWorker && !includeWorker(worker))) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst summary = sessionSummaryFromRosterEntry(entry);\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\tconst match = { worker, summary };\n\t\t\tif (activeSessionId === selector || summary.sessionId === selector || summary.sessionName === selector) {\n\t\t\t\texact.push(match);\n\t\t\t} else if (\n\t\t\t\tmatchesSessionIdSuffix(activeSessionId, selector) ||\n\t\t\t\tmatchesSessionIdSuffix(summary.sessionId, selector)\n\t\t\t) {\n\t\t\t\tsuffix.push(match);\n\t\t\t}\n\t\t}\n\t\treturn exact.length > 0 ? exact : suffix;\n\t}\n\n\tprivate findSummaryInWorker(worker: ResidentWorker, selector: string): SessionSummary | undefined {\n\t\tconst pathSelector = looksLikeSessionPath(selector) ? canonicalSessionPath(selector) : undefined;\n\t\tconst summaries = this.workerRosterEntries(worker)\n\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t.map(sessionSummaryFromRosterEntry);\n\t\tconst exact = summaries.find((summary) => {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\treturn (\n\t\t\t\tactiveSessionId === selector ||\n\t\t\t\tsummary.sessionId === selector ||\n\t\t\t\tsummary.sessionName === selector ||\n\t\t\t\t(pathSelector !== undefined &&\n\t\t\t\t\tsummary.sessionFile !== undefined &&\n\t\t\t\t\tcanonicalSessionPath(summary.sessionFile) === pathSelector)\n\t\t\t);\n\t\t});\n\t\tif (exact) return exact;\n\t\treturn summaries.find((summary) => {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\treturn (\n\t\t\t\tmatchesSessionIdSuffix(activeSessionId, selector) || matchesSessionIdSuffix(summary.sessionId, selector)\n\t\t\t);\n\t\t});\n\t}\n\n\tprivate findWorkerBySessionFile(sessionFile: string, exclude?: ResidentWorker): ResidentWorker | undefined {\n\t\tconst target = canonicalSessionPath(sessionFile);\n\t\tconst targetEntry = this.roster().bySessionFile(target);\n\t\tconst matches = new Set<ResidentWorker>();\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker === exclude) continue;\n\t\t\t// The roster is the one live ownership source; the stale pull cache must not resurrect a match.\n\t\t\tconst summaryMatches = targetEntry?.workerId === worker.descriptor.workerId;\n\t\t\tconst descriptorPath = worker.descriptor.sessionFile\n\t\t\t\t? canonicalSessionPath(worker.descriptor.sessionFile)\n\t\t\t\t: undefined;\n\t\t\tconst configuredPath = worker.descriptor.createCommand.sessionPath\n\t\t\t\t? canonicalSessionPath(worker.descriptor.createCommand.sessionPath)\n\t\t\t\t: undefined;\n\t\t\tif (!summaryMatches && descriptorPath !== target && configuredPath !== target) continue;\n\t\t\tif (descriptorPath && configuredPath && descriptorPath !== configuredPath) {\n\t\t\t\tthrow new Error(`Conflicting resident session paths for worker ${worker.descriptor.workerId}`);\n\t\t\t}\n\t\t\tmatches.add(worker);\n\t\t}\n\t\tif (matches.size > 1) {\n\t\t\tthrow new Error(`Ambiguous resident session path \"${sessionFile}\"`);\n\t\t}\n\t\treturn matches.values().next().value;\n\t}\n\n\tprivate async forwardToWorker(\n\t\tworker: ResidentWorker,\n\t\tcommand: DaemonCommand,\n\t\ttimeoutMs = WORKER_REQUEST_TIMEOUT_MS,\n\t): Promise<DaemonResponse> {\n\t\tconst client = this.requireAvailableWorkerClient(worker, command.type === \"kill\");\n\t\tconst response = await client.request(withoutCommandId(command), timeoutMs);\n\t\tif (command.type === \"get_state\" && response.success && isSessionSummary(response.data)) {\n\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t}\n\t\tif (command.type === \"rename\" && response.success && isSessionSummary(response.data)) {\n\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(response.data), worker);\n\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t}\n\t\treturn responseWithId(response, command.id);\n\t}\n\n\tprivate async attachClient(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"attach\" }>,\n\t): Promise<WorkerAttachData> {\n\t\tconst ownedWorker = [...this.workers.values()].find(\n\t\t\t(worker) =>\n\t\t\t\tworker.descriptor.ownerClientId !== undefined &&\n\t\t\t\t(worker.descriptor.rootActiveSessionId === command.activeSessionId ||\n\t\t\t\t\tworker.descriptor.rootSessionId === command.activeSessionId),\n\t\t);\n\t\tif (ownedWorker) {\n\t\t\tif (ownedWorker.descriptor.ownerClientId !== this.protocolClientId(client)) {\n\t\t\t\tthrow new Error(`Unknown active session: ${command.activeSessionId}`);\n\t\t\t}\n\t\t\tthis.assertTelemetryAttachAllowed(ownedWorker, command.telemetryDisabled);\n\t\t\townedWorker.launchEnv = command.launchEnv ?? ownedWorker.launchEnv;\n\t\t\tif (!ownedWorker.client || ownedWorker.descriptor.lifecycle !== \"ready\") {\n\t\t\t\tif (command.recoveryConfig) {\n\t\t\t\t\townedWorker.transientCreateCommand = {\n\t\t\t\t\t\t...ownedWorker.descriptor.createCommand,\n\t\t\t\t\t\tconfig: {\n\t\t\t\t\t\t\t...command.recoveryConfig,\n\t\t\t\t\t\t\t...(ownedWorker.descriptor.telemetryDisabled === true ? { telemetryDisabled: true } : {}),\n\t\t\t\t\t\t},\n\t\t\t\t\t\tenv: command.env,\n\t\t\t\t\t\tlaunchEnv: command.launchEnv,\n\t\t\t\t\t\tlifecycle: \"client_owned\",\n\t\t\t\t\t};\n\t\t\t\t}\n\t\t\t\tif (!ownedWorker.launchEnv) {\n\t\t\t\t\tthrow new Error(\"Client-owned session recovery requires the owning client environment\");\n\t\t\t\t}\n\t\t\t\townedWorker.intentionalStop = false;\n\t\t\t\townedWorker.descriptor.stopRequestedAt = undefined;\n\t\t\t\townedWorker.descriptor.archiveOnStop = undefined;\n\t\t\t\townedWorker.descriptor.lifecycle = \"recovering\";\n\t\t\t\townedWorker.descriptor.consecutiveFailures = 0;\n\t\t\t\townedWorker.deferredRecoveryRounds = 0;\n\t\t\t\tthis.persistWorker(ownedWorker);\n\t\t\t\tawait this.recoverWorker(ownedWorker);\n\t\t\t}\n\t\t}\n\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\tthis.assertTelemetryAttachAllowed(match.worker, command.telemetryDisabled);\n\t\tthis.requireAvailableWorkerClient(match.worker);\n\t\tconst activeSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\tconst duplicateValidation = this.currentSnapshotGeneration(match.worker, activeSessionId)?.validation;\n\t\tif (duplicateValidation) {\n\t\t\tawait duplicateValidation.promise;\n\t\t}\n\t\tif (command.clientId) {\n\t\t\tclient.id = command.clientId;\n\t\t}\n\t\tclient.capabilities = normalizeCapabilities(command.capabilities, command.supportsExtensionUi);\n\t\tclient.supportsExtensionUi = client.capabilities.has(\"extension_ui\");\n\n\t\tlet result = match.worker.snapshotCache.get(activeSessionId);\n\t\tif (\n\t\t\tresult &&\n\t\t\t!client.capabilities.has(\"chunked_snapshot\") &&\n\t\t\tresult.snapshot.messages.length < result.snapshot.summary.messageCount\n\t\t) {\n\t\t\tresult = undefined;\n\t\t}\n\t\tif (!result) {\n\t\t\tconst snapshotLoadKey = `${activeSessionId}:${client.capabilities.has(\"chunked_snapshot\") ? \"chunked\" : \"full\"}`;\n\t\t\tlet retryInvalidatedLoad = true;\n\t\t\twhile (!result) {\n\t\t\t\tlet loading = match.worker.snapshotLoads.get(snapshotLoadKey);\n\t\t\t\tif (!loading) {\n\t\t\t\t\tconst observedSnapshotId =\n\t\t\t\t\t\tmatch.worker.transcriptCaches.get(activeSessionId)?.snapshotId ??\n\t\t\t\t\t\tmatch.worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id;\n\t\t\t\t\tloading = (async () => {\n\t\t\t\t\t\tconst workerClient = this.requireAvailableWorkerClient(match.worker);\n\t\t\t\t\t\tconst response = await workerClient.request({\n\t\t\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tcapabilities: client.capabilities.has(\"chunked_snapshot\")\n\t\t\t\t\t\t\t\t? [\"attach_snapshot\", \"event_sequence\", \"slim_attach\", \"chunked_snapshot\"]\n\t\t\t\t\t\t\t\t: [\"attach_snapshot\", \"event_sequence\", \"slim_attach\"],\n\t\t\t\t\t\t\tsupportsExtensionUi: false,\n\t\t\t\t\t\t\tenv: command.env ?? collectDaemonClientEnv(),\n\t\t\t\t\t\t});\n\t\t\t\t\t\tconst loaded = attachResultFromResponse(response);\n\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) !== loading) {\n\t\t\t\t\t\t\tthrow new SnapshotLoadInvalidatedError(\"Session snapshot changed during attach\");\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn this.cacheLoadedSnapshot(match.worker, activeSessionId, loaded, observedSnapshotId);\n\t\t\t\t\t})();\n\t\t\t\t\tmatch.worker.snapshotLoads.set(snapshotLoadKey, loading);\n\t\t\t\t\tvoid loading.then(\n\t\t\t\t\t\tasync (loaded) => {\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\tconst snapshotId = loaded.snapshotStream?.id;\n\t\t\t\t\t\t\t\tconst transcript = snapshotId\n\t\t\t\t\t\t\t\t\t? this.snapshotGeneration(match.worker, loaded.activeSessionId, snapshotId)?.transcript\n\t\t\t\t\t\t\t\t\t: undefined;\n\t\t\t\t\t\t\t\tif (transcript && !transcript.complete) {\n\t\t\t\t\t\t\t\t\tlet chunkIndex = 0;\n\t\t\t\t\t\t\t\t\twhile (await transcript.waitForChunk(chunkIndex)) {\n\t\t\t\t\t\t\t\t\t\tchunkIndex++;\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\t\t// Failed transfers must allow a fresh snapshot request.\n\t\t\t\t\t\t\t} finally {\n\t\t\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) === loading) {\n\t\t\t\t\t\t\t\t\tmatch.worker.snapshotLoads.delete(snapshotLoadKey);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t},\n\t\t\t\t\t\t() => {\n\t\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) === loading) {\n\t\t\t\t\t\t\t\tmatch.worker.snapshotLoads.delete(snapshotLoadKey);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t},\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tresult = await loading;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (!(error instanceof SnapshotLoadInvalidatedError)) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tif (!retryInvalidatedLoad) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tretryInvalidatedLoad = false;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tthis.requireAvailableWorkerClient(match.worker);\n\t\tconst wasAttached = client.attachedActiveSessionIds.has(activeSessionId);\n\t\tlet transcript: SnapshotTranscriptCache | undefined;\n\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\twhile (true) {\n\t\t\t\tconst validation = this.currentSnapshotGeneration(match.worker, activeSessionId)?.validation;\n\t\t\t\tif (validation) {\n\t\t\t\t\tawait validation.promise;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tresult = match.worker.snapshotCache.get(activeSessionId) ?? result;\n\t\t\t\ttranscript = this.getOrCreateTranscriptCache(match.worker, result);\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t\tconst releaseTranscript = transcript?.retain();\n\t\tclient.attachedActiveSessionIds.add(activeSessionId);\n\t\ttry {\n\t\t\tconst publicSummary = this.publicSummary(match.worker, result.snapshot.summary);\n\t\t\tif (publicSummary.streamingMessage?.role === \"assistant\") {\n\t\t\t\tthis.streamReconstructor.seed(activeSessionId, publicSummary.streamingMessage);\n\t\t\t} else {\n\t\t\t\tfor (let index = result.snapshot.messages.length - 1; index >= 0; index--) {\n\t\t\t\t\tconst latestMessage = result.snapshot.messages[index];\n\t\t\t\t\tif (latestMessage?.role === \"assistant\") {\n\t\t\t\t\t\tthis.streamReconstructor.seed(activeSessionId, latestMessage);\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tconst publicResult: DaemonAttachResult = {\n\t\t\t\t...result,\n\t\t\t\tstate: result.state ? publicSummary : undefined,\n\t\t\t\tsnapshot: { ...result.snapshot, summary: publicSummary },\n\t\t\t\tclient: { id: client.id, capabilities: [...client.capabilities] },\n\t\t\t};\n\t\t\tif (publicResult.state && publicResult.messages) {\n\t\t\t\tthis.write(client, {\n\t\t\t\t\ttype: \"session_attached\",\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tstate: publicResult.state,\n\t\t\t\t\tmessages: publicResult.messages,\n\t\t\t\t\tsnapshot: publicResult.snapshot,\n\t\t\t\t\treplay: publicResult.replay,\n\t\t\t\t\tlastEventSequence: publicResult.lastEventSequence,\n\t\t\t\t});\n\t\t\t}\n\t\t\tvoid this.syncWorkerExtensionUi(activeSessionId);\n\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\tdetachingSessions?.delete(activeSessionId);\n\t\t\treturn { result: publicResult, worker: match.worker, transcript, releaseTranscript };\n\t\t} catch (error) {\n\t\t\treleaseTranscript?.();\n\t\t\tif (!wasAttached) {\n\t\t\t\tclient.attachedActiveSessionIds.delete(activeSessionId);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate assertTelemetryAttachAllowed(worker: ResidentWorker, telemetryDisabled: true | undefined): void {\n\t\tif (telemetryDisabled && worker.descriptor.telemetryDisabled !== true) {\n\t\t\tthrow new Error(\n\t\t\t\t\"Cannot attach to this active agent while telemetry is disabled for the current invocation. Stop the agent and retry so it can restart without telemetry.\",\n\t\t\t);\n\t\t}\n\t}\n\n\tprivate cacheLoadedSnapshot(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tloaded: DaemonAttachResult,\n\t\tobservedSnapshotId: string | undefined,\n\t): DaemonAttachResult {\n\t\tconst currentTranscript = worker.transcriptCaches.get(activeSessionId);\n\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\tconst currentResult = currentGeneration?.result ?? worker.snapshotCache.get(activeSessionId);\n\t\tconst currentSnapshotId = currentTranscript?.snapshotId ?? currentResult?.snapshotStream?.id;\n\t\tconst loadedSnapshotId = loaded.snapshotStream?.id;\n\t\tif (!loaded.snapshotStream) {\n\t\t\tif (currentSnapshotId && currentSnapshotId !== observedSnapshotId) {\n\t\t\t\treturn loaded;\n\t\t\t}\n\t\t\tworker.snapshotCache.set(activeSessionId, loaded);\n\t\t\treturn loaded;\n\t\t}\n\t\tif (\n\t\t\tcurrentSnapshotId &&\n\t\t\tcurrentSnapshotId !== loadedSnapshotId &&\n\t\t\t(currentSnapshotId !== observedSnapshotId ||\n\t\t\t\t(currentResult?.lastEventSequence ?? -1) > loaded.lastEventSequence)\n\t\t) {\n\t\t\treturn currentResult ?? loaded;\n\t\t}\n\t\tlet transcript = currentTranscript;\n\t\tif (transcript && transcript.snapshotId !== loaded.snapshotStream.id) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, transcript);\n\t\t\ttranscript = undefined;\n\t\t}\n\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\tlet generation = generations.get(loaded.snapshotStream.id);\n\t\tif (!transcript) {\n\t\t\ttranscript = generation?.transcript;\n\t\t}\n\t\tif (!transcript) {\n\t\t\ttranscript = new SnapshotTranscriptCache({\n\t\t\t\tactiveSessionId,\n\t\t\t\tsnapshotId: loaded.snapshotStream.id,\n\t\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\t\ttargetChunkBytes: loaded.snapshotStream.targetChunkBytes,\n\t\t\t});\n\t\t}\n\t\tif (!generation) {\n\t\t\tgeneration = {\n\t\t\t\ttranscript,\n\t\t\t\tresult: loaded,\n\t\t\t\tincoming: false,\n\t\t\t\tretired: false,\n\t\t\t};\n\t\t\tgenerations.set(loaded.snapshotStream.id, generation);\n\t\t} else {\n\t\t\tgeneration.result = loaded;\n\t\t\tgeneration.retired = false;\n\t\t}\n\t\tworker.transcriptCaches.set(activeSessionId, transcript);\n\t\tworker.snapshotCache.set(activeSessionId, loaded);\n\t\treturn loaded;\n\t}\n\n\tprivate getOrCreateTranscriptCache(worker: ResidentWorker, result: DaemonAttachResult): SnapshotTranscriptCache {\n\t\tconst activeSessionId = result.activeSessionId;\n\t\tconst existing = worker.transcriptCaches.get(activeSessionId);\n\t\tif (existing && (!result.snapshotStream || existing.snapshotId === result.snapshotStream.id)) {\n\t\t\treturn existing;\n\t\t}\n\t\tif (result.snapshot.messages.length < result.snapshot.summary.messageCount) {\n\t\t\tthrow new Error(\"Session snapshot generation changed before its transcript could be selected\");\n\t\t}\n\t\tif (existing) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, existing);\n\t\t}\n\t\tconst revision = createHash(\"sha256\")\n\t\t\t.update(\n\t\t\t\t`${activeSessionId}:${result.snapshot.summary.sessionId}:${result.lastEventSequence}:${result.snapshot.messages.length}`,\n\t\t\t)\n\t\t\t.digest(\"hex\")\n\t\t\t.slice(0, 16);\n\t\tconst transcript = new SnapshotTranscriptCache({\n\t\t\tactiveSessionId,\n\t\t\tsnapshotId: `${activeSessionId}-${revision}`,\n\t\t\tmessages: result.snapshot.messages,\n\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\ttargetChunkBytes: SNAPSHOT_TARGET_CHUNK_BYTES,\n\t\t});\n\t\tworker.transcriptCaches.set(activeSessionId, transcript);\n\t\tconst cachedResult = {\n\t\t\t...result,\n\t\t\tmessages: result.messages ? [] : undefined,\n\t\t\tsnapshot: { ...result.snapshot, messages: [] },\n\t\t};\n\t\tworker.snapshotCache.set(activeSessionId, cachedResult);\n\t\tthis.snapshotGenerationsFor(worker, activeSessionId).set(transcript.snapshotId, {\n\t\t\ttranscript,\n\t\t\tresult: cachedResult,\n\t\t\tincoming: false,\n\t\t\tretired: false,\n\t\t});\n\t\treturn transcript;\n\t}\n\n\tprivate createStreamedAttachResult(\n\t\tresult: DaemonAttachResult,\n\t\ttranscript: SnapshotTranscriptCache,\n\t): DaemonAttachResult {\n\t\treturn {\n\t\t\t...result,\n\t\t\tmessages: result.messages ? [] : undefined,\n\t\t\tsnapshot: { ...result.snapshot, messages: [] },\n\t\t\tsnapshotStream: {\n\t\t\t\tid: transcript.snapshotId,\n\t\t\t\tmessageCount: result.snapshot.summary.messageCount,\n\t\t\t\ttargetChunkBytes: transcript.targetChunkBytes,\n\t\t\t},\n\t\t};\n\t}\n\n\tprivate async streamSnapshot(\n\t\tclient: DaemonSocketClient,\n\t\tworker: ResidentWorker,\n\t\tresult: DaemonAttachResult,\n\t\ttranscript: SnapshotTranscriptCache,\n\t\tpurpose: \"attach\" | \"replacement\" | \"resync\" = \"attach\",\n\t\tretainedTranscriptRelease?: () => void,\n\t\treleaseSnapshotReservation = this.reserveSnapshotStream(client, result.activeSessionId),\n\t): Promise<void> {\n\t\tconst stream = result.snapshotStream;\n\t\tconst releaseTranscript = retainedTranscriptRelease ?? transcript.retain();\n\t\tif (!stream || client.socket.destroyed) {\n\t\t\treleaseSnapshotReservation();\n\t\t\treleaseTranscript();\n\t\t\treturn;\n\t\t}\n\t\tconst { messages: _messages, ...snapshotHeader } = result.snapshot;\n\t\ttry {\n\t\t\tif (\n\t\t\t\t!(await this.writeSnapshotRecord(client, {\n\t\t\t\t\ttype: \"session_snapshot_begin\",\n\t\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\t\tsnapshotId: stream.id,\n\t\t\t\t\tsnapshot: snapshotHeader,\n\t\t\t\t\tmessageCount: stream.messageCount,\n\t\t\t\t\ttargetChunkBytes: stream.targetChunkBytes,\n\t\t\t\t\tpurpose,\n\t\t\t\t}))\n\t\t\t) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tlet chunkCount = 0;\n\t\t\twhile (true) {\n\t\t\t\tlet chunk: Buffer | undefined;\n\t\t\t\ttry {\n\t\t\t\t\tchunk = await transcript.waitForChunk(chunkCount);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tconst streamError = error instanceof Error ? error : new Error(String(error));\n\t\t\t\t\tthis.failWorkerSnapshotCache(worker, result.activeSessionId, streamError, false, stream.id);\n\t\t\t\t\tthrow streamError;\n\t\t\t\t}\n\t\t\t\tif (!chunk) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tif (!(await this.writeSnapshotBuffer(client, chunk))) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tchunkCount++;\n\t\t\t}\n\t\t\tawait this.writeSnapshotRecord(client, {\n\t\t\t\ttype: \"session_snapshot_end\",\n\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\tsnapshotId: stream.id,\n\t\t\t\tchunkCount,\n\t\t\t\tlastEventSequence: result.lastEventSequence,\n\t\t\t\tlastEventCursor: result.lastEventCursor,\n\t\t\t});\n\t\t} catch (error) {\n\t\t\tconst streamError = error instanceof Error ? error : new Error(String(error));\n\t\t\tif (!client.socket.destroyed) {\n\t\t\t\ttry {\n\t\t\t\t\tconst delivered = await this.writeSnapshotRecord(client, {\n\t\t\t\t\t\ttype: \"session_snapshot_failed\",\n\t\t\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\t\t\tsnapshotId: stream.id,\n\t\t\t\t\t\terror: streamError.message,\n\t\t\t\t\t});\n\t\t\t\t\tif (!delivered && !client.socket.destroyed) {\n\t\t\t\t\t\tclient.socket.destroy(streamError);\n\t\t\t\t\t}\n\t\t\t\t} catch (deliveryError) {\n\t\t\t\t\tclient.socket.destroy(deliveryError instanceof Error ? deliveryError : new Error(String(deliveryError)));\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow streamError;\n\t\t} finally {\n\t\t\treleaseSnapshotReservation();\n\t\t\treleaseTranscript();\n\t\t}\n\t}\n\n\tprivate reserveSnapshotStream(client: DaemonSocketClient, activeSessionId: string): () => void {\n\t\tclient.snapshotStreaming = true;\n\t\tclient.snapshotActiveSessionIds ??= new Set();\n\t\tclient.snapshotActiveSessionIds.add(activeSessionId);\n\t\tclient.snapshotActiveSessionCounts ??= new Map();\n\t\tclient.snapshotActiveSessionCounts.set(\n\t\t\tactiveSessionId,\n\t\t\t(client.snapshotActiveSessionCounts.get(activeSessionId) ?? 0) + 1,\n\t\t);\n\t\tlet released = false;\n\t\treturn () => {\n\t\t\tif (released) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\treleased = true;\n\t\t\tconst streamCount = client.snapshotActiveSessionCounts?.get(activeSessionId) ?? 1;\n\t\t\tif (streamCount > 1) {\n\t\t\t\tclient.snapshotActiveSessionCounts?.set(activeSessionId, streamCount - 1);\n\t\t\t} else {\n\t\t\t\tclient.snapshotActiveSessionCounts?.delete(activeSessionId);\n\t\t\t\tclient.snapshotActiveSessionIds?.delete(activeSessionId);\n\t\t\t}\n\t\t\tclient.snapshotStreaming = (client.snapshotActiveSessionIds?.size ?? 0) > 0;\n\t\t\tif (!client.snapshotStreaming) {\n\t\t\t\tclient.backpressured = false;\n\t\t\t}\n\t\t\tif (!client.snapshotStreaming && client.catchupActiveSessionIds?.size) {\n\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t);\n\t\t\t}\n\t\t};\n\t}\n\n\tprivate writeSnapshotRecord(client: DaemonSocketClient, message: DaemonOutbound): Promise<boolean> {\n\t\treturn this.writeSnapshotBuffer(client, Buffer.from(serializeJsonLine(message)));\n\t}\n\n\tprivate async writeSnapshotBuffer(client: DaemonSocketClient, buffer: Uint8Array): Promise<boolean> {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn false;\n\t\t}\n\t\tif (this.writeSerialized(client, buffer)) {\n\t\t\treturn true;\n\t\t}\n\t\treturn new Promise<boolean>((resolveDrain) => {\n\t\t\tlet settled = false;\n\t\t\tconst finish = (value: boolean) => {\n\t\t\t\tif (settled) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tsettled = true;\n\t\t\t\tclient.socket.off(\"drain\", onDrain);\n\t\t\t\tclient.socket.off(\"close\", onClose);\n\t\t\t\tclient.socket.off(\"error\", onClose);\n\t\t\t\tresolveDrain(value);\n\t\t\t};\n\t\t\tconst onDrain = () => finish(true);\n\t\t\tconst onClose = () => finish(false);\n\t\t\tclient.socket.once(\"drain\", onDrain);\n\t\t\tclient.socket.once(\"close\", onClose);\n\t\t\tclient.socket.once(\"error\", onClose);\n\t\t});\n\t}\n\n\tprivate detachClient(client: DaemonSocketClient, activeSessionId?: string): void {\n\t\tconst targets = activeSessionId ? [activeSessionId] : [...client.attachedActiveSessionIds];\n\t\tfor (const selector of targets) {\n\t\t\tconst match = this.matchWorkers(selector)[0];\n\t\t\tconst resolvedId = match ? (match.summary.activeSessionId ?? match.summary.id) : selector;\n\t\t\tif (!client.attachedActiveSessionIds.delete(resolvedId)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tclient.catchupActiveSessionIds?.delete(resolvedId);\n\t\t\tclient.catchupPurposes?.delete(resolvedId);\n\t\t\tthis.write(client, { type: \"session_detached\", activeSessionId: resolvedId });\n\t\t\tvoid this.syncWorkerExtensionUi(resolvedId);\n\t\t\tvoid this.evictEmptySessionOnLastDetach(resolvedId);\n\t\t}\n\t}\n\n\tprivate async syncWorkerExtensionUi(activeSessionId: string): Promise<void> {\n\t\tconst match = this.matchWorkers(activeSessionId)[0];\n\t\tif (!match?.worker.client) {\n\t\t\treturn;\n\t\t}\n\t\tawait this.subscribeWorker(match.worker, match.summary.activeSessionId ?? match.summary.id).catch(\n\t\t\t() => undefined,\n\t\t);\n\t}\n\n\tprivate handleWorkerFrame(\n\t\tworker: ResidentWorker,\n\t\tframe: PrivateFrame<DaemonWorkerFrameHeader>,\n\t\tsource?: DaemonWorkerClient,\n\t): void {\n\t\tif (frame.header.kind !== \"outbound\") {\n\t\t\treturn;\n\t\t}\n\t\tif (source !== undefined && source !== worker.client && source !== worker.pendingClient) {\n\t\t\treturn;\n\t\t}\n\t\tworker.lastFrameAt = Date.now();\n\t\tthis.clearRosterStaleness(worker);\n\t\tconst {\n\t\t\toutboundType,\n\t\t\tactiveSessionId,\n\t\t\tsnapshotId: frameSnapshotId,\n\t\t\tsessionEventType,\n\t\t\tpayloadEncoding,\n\t\t\tsnapshotPurpose,\n\t\t} = frame.header;\n\t\tif (outboundType === \"roster_delta\") {\n\t\t\tthis.consumeWorkerRosterDelta(worker, frame.payload, source);\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"roster_heartbeat\") {\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"heartbeats_changed\") {\n\t\t\tworker.heartbeatSnapshotStale = true;\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_begin\" && activeSessionId) {\n\t\t\ttry {\n\t\t\t\tconst begin = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t{ type: \"session_snapshot_begin\" }\n\t\t\t\t>;\n\t\t\t\tif (\n\t\t\t\t\tbegin.type !== \"session_snapshot_begin\" ||\n\t\t\t\t\tbegin.activeSessionId !== activeSessionId ||\n\t\t\t\t\ttypeof begin.snapshotId !== \"string\" ||\n\t\t\t\t\t(frameSnapshotId !== undefined && frameSnapshotId !== begin.snapshotId) ||\n\t\t\t\t\ttypeof begin.targetChunkBytes !== \"number\" ||\n\t\t\t\t\t!begin.snapshot ||\n\t\t\t\t\t!isSessionSummary(begin.snapshot.summary)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid snapshot begin frame\");\n\t\t\t\t}\n\t\t\t\tconst publicSummary = this.publicSummary(worker, begin.snapshot.summary);\n\t\t\t\tconst snapshot = {\n\t\t\t\t\t...begin.snapshot,\n\t\t\t\t\tsummary: publicSummary,\n\t\t\t\t\tmessages: [],\n\t\t\t\t};\n\t\t\t\tconst result: DaemonAttachResult = {\n\t\t\t\t\tprotocol: DAEMON_PROTOCOL_INFO,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tsnapshot,\n\t\t\t\t\treplay: {\n\t\t\t\t\t\tstatus: \"complete\",\n\t\t\t\t\t\ttoSequence: snapshot.lastEventSequence,\n\t\t\t\t\t\t...(snapshot.lastEventCursor ? { toCursor: snapshot.lastEventCursor } : {}),\n\t\t\t\t\t},\n\t\t\t\t\tlastEventSequence: snapshot.lastEventSequence,\n\t\t\t\t\t...(snapshot.lastEventCursor ? { lastEventCursor: snapshot.lastEventCursor } : {}),\n\t\t\t\t\tsnapshotStream: {\n\t\t\t\t\t\tid: begin.snapshotId,\n\t\t\t\t\t\tmessageCount: begin.messageCount,\n\t\t\t\t\t\ttargetChunkBytes: begin.targetChunkBytes,\n\t\t\t\t\t},\n\t\t\t\t\tclient: { id: \"supervisor\", capabilities: [\"chunked_snapshot\"] },\n\t\t\t\t};\n\t\t\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\t\t\tlet generation = generations.get(begin.snapshotId);\n\t\t\t\tif (generation?.incoming) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tnew Error(`Snapshot ${begin.snapshotId} restarted before completion`),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tbegin.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\t// Snapshot summaries/state include live fields (for example activity and attached client\n\t\t\t\t// counts) that can change without advancing the transcript sequence. Treat the stable\n\t\t\t\t// transfer envelope as identity; duplicate chunks and end metadata are still byte-checked.\n\t\t\t\tconst duplicate =\n\t\t\t\t\tgeneration?.transcript.complete === true &&\n\t\t\t\t\tgeneration.end !== undefined &&\n\t\t\t\t\tgeneration.result.snapshotStream?.messageCount === begin.messageCount &&\n\t\t\t\t\tgeneration.result.snapshotStream?.targetChunkBytes === begin.targetChunkBytes &&\n\t\t\t\t\tgeneration.result.lastEventSequence === result.lastEventSequence &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventSequence === result.snapshot.lastEventSequence &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventCursor?.generation === result.snapshot.lastEventCursor?.generation &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventCursor?.sequence === result.snapshot.lastEventCursor?.sequence;\n\t\t\t\tif (generation?.transcript.complete && !duplicate) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tnew Error(`Snapshot ${begin.snapshotId} did not match the cached transfer`),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tbegin.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\t\t\tconst currentResult = currentGeneration?.result ?? worker.snapshotCache.get(activeSessionId);\n\t\t\t\tconst isOlderThanCurrent =\n\t\t\t\t\tcurrentGeneration !== undefined &&\n\t\t\t\t\tcurrentGeneration.transcript.snapshotId !== begin.snapshotId &&\n\t\t\t\t\tcurrentResult !== undefined &&\n\t\t\t\t\tresult.lastEventSequence < currentResult.lastEventSequence;\n\t\t\t\tif (isOlderThanCurrent && !generation) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (duplicate && generation) {\n\t\t\t\t\tgeneration.incoming = true;\n\t\t\t\t\tgeneration.duplicateChunkIndex = 0;\n\t\t\t\t\tgeneration.duplicateResult = result;\n\t\t\t\t\tgeneration.validation = this.createSnapshotDuplicateValidation();\n\t\t\t\t\tif (currentGeneration === generation) {\n\t\t\t\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t\t\t\t}\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\tcurrentGeneration &&\n\t\t\t\t\tcurrentGeneration.transcript.snapshotId !== begin.snapshotId &&\n\t\t\t\t\t!isOlderThanCurrent\n\t\t\t\t) {\n\t\t\t\t\tif (!currentGeneration.transcript.complete && !currentGeneration.incoming) {\n\t\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tnew Error(`Snapshot ${currentGeneration.transcript.snapshotId} was superseded`),\n\t\t\t\t\t\t\tfalse,\n\t\t\t\t\t\t\tcurrentGeneration.transcript.snapshotId,\n\t\t\t\t\t\t);\n\t\t\t\t\t} else {\n\t\t\t\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, currentGeneration.transcript);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tif (!generation) {\n\t\t\t\t\tconst transcript = new SnapshotTranscriptCache({\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tsnapshotId: begin.snapshotId,\n\t\t\t\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\t\t\t\ttargetChunkBytes: begin.targetChunkBytes,\n\t\t\t\t\t});\n\t\t\t\t\tgeneration = {\n\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\tresult,\n\t\t\t\t\t\tincoming: false,\n\t\t\t\t\t\tretired: isOlderThanCurrent,\n\t\t\t\t\t};\n\t\t\t\t\tthis.snapshotGenerationsFor(worker, activeSessionId).set(begin.snapshotId, generation);\n\t\t\t\t}\n\t\t\t\tgeneration.result = result;\n\t\t\t\tgeneration.begin = Buffer.from(frame.payload);\n\t\t\t\tgeneration.end = undefined;\n\t\t\t\tgeneration.incoming = true;\n\t\t\t\tgeneration.duplicateChunkIndex = undefined;\n\t\t\t\tgeneration.duplicateResult = undefined;\n\t\t\t\tgeneration.validation = undefined;\n\t\t\t\tif (!isOlderThanCurrent) {\n\t\t\t\t\tgeneration.retired = false;\n\t\t\t\t\tworker.transcriptCaches.set(activeSessionId, generation.transcript);\n\t\t\t\t\tworker.snapshotCache.set(activeSessionId, result);\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Invalid worker snapshot begin frame: ${String(error)}`);\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_chunk\" && activeSessionId) {\n\t\t\tconst snapshotId = frameSnapshotId ?? worker.transcriptCaches.get(activeSessionId)?.snapshotId;\n\t\t\tif (!snapshotId) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, snapshotId);\n\t\t\tif (generation?.incoming) {\n\t\t\t\ttry {\n\t\t\t\t\tconst duplicateIndex = generation.duplicateChunkIndex;\n\t\t\t\t\tif (duplicateIndex === undefined) {\n\t\t\t\t\t\tgeneration.transcript.appendEncodedChunk(Buffer.from(frame.payload));\n\t\t\t\t\t} else {\n\t\t\t\t\t\tconst chunk = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t\t\t{ type: \"session_snapshot_chunk\" }\n\t\t\t\t\t\t>;\n\t\t\t\t\t\tif (\n\t\t\t\t\t\t\tchunk.type !== \"session_snapshot_chunk\" ||\n\t\t\t\t\t\t\tchunk.activeSessionId !== activeSessionId ||\n\t\t\t\t\t\t\tchunk.snapshotId !== generation.transcript.snapshotId ||\n\t\t\t\t\t\t\tchunk.index !== duplicateIndex ||\n\t\t\t\t\t\t\t!generation.transcript.readChunk(duplicateIndex).equals(Buffer.from(frame.payload))\n\t\t\t\t\t\t) {\n\t\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t\t`Duplicate snapshot ${generation.transcript.snapshotId} did not match cached bytes`,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t\tgeneration.duplicateChunkIndex = duplicateIndex + 1;\n\t\t\t\t\t}\n\t\t\t\t} catch (error) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tgeneration.transcript.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_end\" && activeSessionId) {\n\t\t\tconst snapshotId = frameSnapshotId ?? worker.transcriptCaches.get(activeSessionId)?.snapshotId;\n\t\t\tif (!snapshotId) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, snapshotId);\n\t\t\tif (!generation?.incoming) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst transcript = generation.transcript;\n\t\t\ttry {\n\t\t\t\tconst duplicateChunkCount = generation.duplicateChunkIndex;\n\t\t\t\tif (duplicateChunkCount === undefined) {\n\t\t\t\t\ttranscript.markComplete();\n\t\t\t\t\tif (!generation.begin) {\n\t\t\t\t\t\tthrow new Error(`Snapshot ${transcript.snapshotId} has no begin frame`);\n\t\t\t\t\t}\n\t\t\t\t\tgeneration.end = Buffer.from(frame.payload);\n\t\t\t\t} else {\n\t\t\t\t\tconst end = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t\t{ type: \"session_snapshot_end\" }\n\t\t\t\t\t>;\n\t\t\t\t\tif (\n\t\t\t\t\t\tend.type !== \"session_snapshot_end\" ||\n\t\t\t\t\t\tend.activeSessionId !== activeSessionId ||\n\t\t\t\t\t\tend.snapshotId !== transcript.snapshotId ||\n\t\t\t\t\t\tend.chunkCount !== duplicateChunkCount ||\n\t\t\t\t\t\tend.chunkCount !== transcript.chunkCount ||\n\t\t\t\t\t\t!generation.end?.equals(frame.payload)\n\t\t\t\t\t) {\n\t\t\t\t\t\tthrow new Error(`Duplicate snapshot ${transcript.snapshotId} ended with different metadata`);\n\t\t\t\t\t}\n\t\t\t\t\tif (!generation.duplicateResult) {\n\t\t\t\t\t\tthrow new Error(`Duplicate snapshot ${transcript.snapshotId} has no result`);\n\t\t\t\t\t}\n\t\t\t\t\tgeneration.result = generation.duplicateResult;\n\t\t\t\t\tif (worker.transcriptCaches.get(activeSessionId) === transcript) {\n\t\t\t\t\t\tworker.snapshotCache.set(activeSessionId, generation.duplicateResult);\n\t\t\t\t\t}\n\t\t\t\t\tthis.settleSnapshotDuplicateValidation(generation);\n\t\t\t\t}\n\t\t\t\tgeneration.incoming = false;\n\t\t\t\tgeneration.duplicateChunkIndex = undefined;\n\t\t\t\tgeneration.duplicateResult = undefined;\n\t\t\t} catch (error) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t\ttranscript.snapshotId,\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst published = worker.transcriptCaches.get(activeSessionId) === transcript;\n\t\t\tif (generation.retired) {\n\t\t\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\t\t\ttranscript.dispose();\n\t\t\t}\n\t\t\tif (published && (snapshotPurpose === \"replacement\" || snapshotPurpose === \"catchup\")) {\n\t\t\t\tfor (const client of this.clients) {\n\t\t\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) continue;\n\t\t\t\t\tthis.queueCatchup(client, activeSessionId, snapshotPurpose === \"replacement\" ? \"replacement\" : \"resync\");\n\t\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_failed\" && activeSessionId) {\n\t\t\ttry {\n\t\t\t\tconst failed = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t{ type: \"session_snapshot_failed\" }\n\t\t\t\t>;\n\t\t\t\tif (\n\t\t\t\t\tfailed.type !== \"session_snapshot_failed\" ||\n\t\t\t\t\tfailed.activeSessionId !== activeSessionId ||\n\t\t\t\t\ttypeof failed.snapshotId !== \"string\" ||\n\t\t\t\t\ttypeof failed.error !== \"string\" ||\n\t\t\t\t\t(frameSnapshotId !== undefined && frameSnapshotId !== failed.snapshotId)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid snapshot failure frame\");\n\t\t\t\t}\n\t\t\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\t\t\tconst generation =\n\t\t\t\t\tthis.snapshotGeneration(worker, activeSessionId, failed.snapshotId) ??\n\t\t\t\t\t(currentGeneration?.transcript.snapshotId === failed.snapshotId ? currentGeneration : undefined);\n\t\t\t\tif (!generation) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst published = worker.transcriptCaches.get(activeSessionId) === generation.transcript;\n\t\t\t\tthis.failWorkerSnapshotCache(worker, activeSessionId, new Error(failed.error), false, failed.snapshotId);\n\t\t\t\tif (published && (snapshotPurpose === \"replacement\" || snapshotPurpose === \"catchup\")) {\n\t\t\t\t\tfor (const client of this.clients) {\n\t\t\t\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) continue;\n\t\t\t\t\t\tthis.queueCatchup(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tsnapshotPurpose === \"replacement\" ? \"replacement\" : \"resync\",\n\t\t\t\t\t\t);\n\t\t\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (\n\t\t\toutboundType === \"daemon_hello\" ||\n\t\t\toutboundType === \"response\" ||\n\t\t\toutboundType === \"session_list_progress\" ||\n\t\t\toutboundType === \"session_list_item\" ||\n\t\t\toutboundType === \"session_attached\" ||\n\t\t\toutboundType === \"session_detached\" ||\n\t\t\t!activeSessionId\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tlet publicPayload = frame.payload;\n\t\tlet decodedOutbound: DaemonOutbound | undefined;\n\t\tif (payloadEncoding === \"assistant-delta\") {\n\t\t\tlet compactValue: unknown;\n\t\t\ttry {\n\t\t\t\tcompactValue = JSON.parse(frame.payload.toString(\"utf8\"));\n\t\t\t} catch {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!isCompactAssistantDelta(compactValue)) {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst reconstructed = this.streamReconstructor.reconstruct(compactValue);\n\t\t\tif (!reconstructed) {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tpublicPayload = Buffer.from(serializeJsonLine(reconstructed));\n\t\t} else if (\n\t\t\tsessionEventType === \"message_start\" ||\n\t\t\tsessionEventType === \"message_end\" ||\n\t\t\toutboundType === \"session_replaced\" ||\n\t\t\toutboundType === \"session_resynced\" ||\n\t\t\toutboundType === \"session_closed\"\n\t\t) {\n\t\t\ttry {\n\t\t\t\tdecodedOutbound = JSON.parse(frame.payload.toString(\"utf8\")) as DaemonOutbound;\n\t\t\t\tthis.streamReconstructor.observe(decodedOutbound);\n\t\t\t} catch {\n\t\t\t\t// A malformed worker event is still isolated to this worker connection.\n\t\t\t}\n\t\t}\n\t\tconst replacementSnapshotFollows =\n\t\t\tdecodedOutbound?.type === \"session_replaced\" && decodedOutbound.snapshotFollows === true;\n\t\tthis.invalidateWorkerSnapshot(\n\t\t\tworker,\n\t\t\tactiveSessionId,\n\t\t\toutboundType === \"session_replaced\" ||\n\t\t\t\toutboundType === \"session_closed\" ||\n\t\t\t\tisFinalizedTranscriptEvent(sessionEventType),\n\t\t);\n\t\tfor (const client of this.clients) {\n\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (replacementSnapshotFollows && !client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (outboundType === \"extension_ui_request\" && !client.supportsExtensionUi) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (client.snapshotActiveSessionIds?.has(activeSessionId)) {\n\t\t\t\tthis.queueCatchup(client, activeSessionId, outboundType === \"session_replaced\" ? \"replacement\" : \"resync\");\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (client.backpressured === true) {\n\t\t\t\tthis.queueCatchup(client, activeSessionId, outboundType === \"session_replaced\" ? \"replacement\" : \"resync\");\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.writeSerialized(client, publicPayload);\n\t\t}\n\t\tif (\n\t\t\tdecodedOutbound?.type === \"session_closed\" &&\n\t\t\tdecodedOutbound.reason === \"shutdown\" &&\n\t\t\tactiveSessionId === worker.descriptor.rootActiveSessionId &&\n\t\t\t!this.shuttingDown\n\t\t) {\n\t\t\tworker.intentionalStop = true;\n\t\t\t// An exact stop owns its registration and descriptor cleanup until its\n\t\t\t// tuple assertions complete. A synchronous root shutdown event can arrive\n\t\t\t// before its request resolves, so leave both intact while it is active.\n\t\t\tif ((this.workerStopCounts?.get(worker) ?? 0) === 0) {\n\t\t\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\t\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate invalidateWorkerSnapshot(worker: ResidentWorker, activeSessionId: string, transcriptChanged = true): void {\n\t\tworker.snapshotCache.delete(activeSessionId);\n\t\tif (!transcriptChanged) {\n\t\t\treturn;\n\t\t}\n\t\tworker.snapshotLoads.delete(`${activeSessionId}:chunked`);\n\t\tworker.snapshotLoads.delete(`${activeSessionId}:full`);\n\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\tif (transcript) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, transcript);\n\t\t}\n\t}\n\n\tprivate scheduleCompactCatchup(worker: ResidentWorker, activeSessionId: string): void {\n\t\tif (this.compactCatchupInProgress.has(activeSessionId)) {\n\t\t\treturn;\n\t\t}\n\t\tthis.compactCatchupInProgress.add(activeSessionId);\n\t\tthis.invalidateWorkerSnapshot(worker, activeSessionId);\n\t\tconst clients = [...this.clients].filter((client) => client.attachedActiveSessionIds.has(activeSessionId));\n\t\tfor (const client of clients) {\n\t\t\tthis.queueCatchup(client, activeSessionId);\n\t\t}\n\t\tvoid Promise.all(clients.map((client) => this.catchUpClient(client)))\n\t\t\t.catch((error) => this.log(`Failed compact catch-up for ${activeSessionId}: ${String(error)}`))\n\t\t\t.finally(() => {\n\t\t\t\tthis.compactCatchupInProgress.delete(activeSessionId);\n\t\t\t});\n\t}\n\n\tprivate queueCatchup(\n\t\tclient: DaemonSocketClient,\n\t\tactiveSessionId: string,\n\t\tpurpose: \"replacement\" | \"resync\" = \"resync\",\n\t): void {\n\t\tif (!client.catchupActiveSessionIds) {\n\t\t\tclient.catchupActiveSessionIds = new Set();\n\t\t}\n\t\tclient.catchupActiveSessionIds.add(activeSessionId);\n\t\tclient.catchupPurposes ??= new Map();\n\t\tif (purpose === \"replacement\" || !client.catchupPurposes.has(activeSessionId)) {\n\t\t\tclient.catchupPurposes.set(activeSessionId, purpose);\n\t\t}\n\t}\n\n\tprivate catchUpClient(client: DaemonSocketClient): Promise<void> {\n\t\tif (client.catchupPromise) {\n\t\t\treturn client.catchupPromise;\n\t\t}\n\t\tif (client.snapshotStreaming || client.backpressured) {\n\t\t\treturn Promise.resolve();\n\t\t}\n\t\tconst catchup = this.drainClientCatchupQueue(client).finally(() => {\n\t\t\tif (client.catchupPromise === catchup) {\n\t\t\t\tclient.catchupPromise = undefined;\n\t\t\t}\n\t\t});\n\t\tclient.catchupPromise = catchup;\n\t\treturn catchup;\n\t}\n\n\tprivate async drainClientCatchupQueue(client: DaemonSocketClient): Promise<void> {\n\t\twhile (\n\t\t\t!client.socket.destroyed &&\n\t\t\t!client.snapshotStreaming &&\n\t\t\t!client.backpressured &&\n\t\t\tclient.catchupActiveSessionIds?.size\n\t\t) {\n\t\t\tawait this.drainClientCatchups(client);\n\t\t}\n\t}\n\n\tprivate async drainClientCatchups(client: DaemonSocketClient): Promise<void> {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn;\n\t\t}\n\t\tconst pending = [...(client.catchupActiveSessionIds ?? [])].map((activeSessionId) => ({\n\t\t\tactiveSessionId,\n\t\t\tpurpose: client.catchupPurposes?.get(activeSessionId) ?? (\"resync\" as const),\n\t\t}));\n\t\tclient.catchupActiveSessionIds?.clear();\n\t\tclient.catchupPurposes?.clear();\n\t\tfor (let index = 0; index < pending.length; index++) {\n\t\t\tconst { activeSessionId, purpose } = pending[index]!;\n\t\t\tlet releaseTranscript: (() => void) | undefined;\n\t\t\ttry {\n\t\t\t\tconst attached = await this.attachClient(client, {\n\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tcapabilities: [...client.capabilities],\n\t\t\t\t\tsupportsExtensionUi: client.supportsExtensionUi,\n\t\t\t\t});\n\t\t\t\treleaseTranscript = attached.releaseTranscript;\n\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\tconst transcript = attached.transcript;\n\t\t\t\t\tif (!transcript) {\n\t\t\t\t\t\tthrow new Error(\"Session worker did not provide a snapshot transcript\");\n\t\t\t\t\t}\n\t\t\t\t\tif (purpose === \"replacement\") {\n\t\t\t\t\t\tthis.write(client, {\n\t\t\t\t\t\t\ttype: \"session_replaced\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tstate: attached.result.snapshot.state,\n\t\t\t\t\t\t\tmessages: [],\n\t\t\t\t\t\t\tsnapshotFollows: true,\n\t\t\t\t\t\t\tmeta: createDaemonEventMeta(\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tattached.result.lastEventSequence,\n\t\t\t\t\t\t\t\tundefined,\n\t\t\t\t\t\t\t\tattached.result.lastEventCursor?.generation,\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t\tawait this.streamSnapshot(\n\t\t\t\t\t\tclient,\n\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\tthis.createStreamedAttachResult(attached.result, transcript),\n\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\tpurpose,\n\t\t\t\t\t\treleaseTranscript,\n\t\t\t\t\t);\n\t\t\t\t\treleaseTranscript = undefined;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tconst meta = createDaemonEventMeta(\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tattached.result.lastEventSequence,\n\t\t\t\t\tundefined,\n\t\t\t\t\tattached.result.lastEventCursor?.generation,\n\t\t\t\t);\n\t\t\t\tconst catchup: DaemonOutbound =\n\t\t\t\t\tpurpose === \"replacement\"\n\t\t\t\t\t\t? {\n\t\t\t\t\t\t\t\ttype: \"session_replaced\",\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tstate: attached.result.snapshot.state,\n\t\t\t\t\t\t\t\tmessages: attached.result.snapshot.messages,\n\t\t\t\t\t\t\t\tmeta,\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t: {\n\t\t\t\t\t\t\t\ttype: \"session_resynced\",\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tsnapshot: attached.result.snapshot,\n\t\t\t\t\t\t\t\tmeta,\n\t\t\t\t\t\t\t};\n\t\t\t\tif (!this.write(client, catchup)) {\n\t\t\t\t\tfor (const remaining of pending.slice(index + 1)) {\n\t\t\t\t\t\tthis.queueCatchup(client, remaining.activeSessionId, remaining.purpose);\n\t\t\t\t\t}\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\treleaseTranscript?.();\n\t\t\t\tthis.log(`Failed to catch up client ${client.id} for ${activeSessionId}: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async prepareUpdateRestart(): Promise<DaemonUpdateRestartManifest> {\n\t\tif (this.updateRestartPhase !== undefined) throw new Error(\"Daemon is already preparing an update restart\");\n\t\tthis.updateRestartPhase = \"draining\";\n\t\ttry {\n\t\t\tconst deadline = Date.now() + UPDATE_RESTART_PREPARE_DEADLINE_MS;\n\t\t\tconst abort = AbortSignal.timeout(Math.min(UPDATE_RESTART_MUTATION_DRAIN_TIMEOUT_MS, deadline - Date.now()));\n\t\t\tawait this.mutationDrain.waitForDrain(1, abort, \"Timed out draining daemon mutations for update restart\");\n\t\t\tthis.updateRestartPhase = \"fencing\";\n\t\t\tawait this.mutationDrain.waitForDrain(1, abort, \"Timed out draining daemon mutations for update restart\");\n\t\t\tconst manifest = await this.prepareUpdateRestartFenced(deadline);\n\t\t\tthis.updateRestartPhase = \"prepared\";\n\t\t\treturn manifest;\n\t\t} catch (error) {\n\t\t\tthis.updateRestartPhase = undefined;\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async prepareUpdateRestartFenced(deadline: number): Promise<DaemonUpdateRestartManifest> {\n\t\tconst residents = [...this.workers.values()];\n\t\tconst unavailable = residents.find(\n\t\t\t(worker) =>\n\t\t\t\tthis.isWorkerStopping(worker) || worker.descriptor.lifecycle !== \"ready\" || worker.client === undefined,\n\t\t);\n\t\tif (unavailable) {\n\t\t\tthrow new Error(\n\t\t\t\t`Cannot prepare update restart while resident worker ${unavailable.descriptor.workerId} is ${this.effectiveWorkerState(unavailable)}${unavailable.client ? \"\" : \" and disconnected\"}`,\n\t\t\t);\n\t\t}\n\t\tconst workers = residents as Array<ResidentWorker & { client: DaemonWorkerClient }>;\n\t\tconst acknowledged: ResidentWorker[] = [];\n\t\tconst preparationResults = await Promise.allSettled(\n\t\t\tworkers.map(async (worker) => {\n\t\t\t\tconst client = worker.client;\n\t\t\t\tconst response = await client.requestWorker(\n\t\t\t\t\t{ type: \"worker_prepare_update\" },\n\t\t\t\t\tMath.max(1, Math.min(UPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS, deadline - Date.now())),\n\t\t\t\t);\n\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\tworker.updateRestartPrepareClient = client;\n\t\t\t\tacknowledged.push(worker);\n\t\t\t\tif (!response.data || typeof response.data !== \"object\") {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid update manifest\");\n\t\t\t\t}\n\t\t\t\tif (worker.client !== client || worker.descriptor.lifecycle !== \"ready\") {\n\t\t\t\t\tthrow new Error(`Worker ${worker.descriptor.workerId} disconnected during update preparation`);\n\t\t\t\t}\n\t\t\t\tconst manifest = response.data as DaemonUpdateRestartManifest;\n\t\t\t\tif (manifest.formatVersion !== DAEMON_UPDATE_RESTART_FORMAT_VERSION) {\n\t\t\t\t\tthrow new Error(`Worker returned unsupported update manifest version ${manifest.formatVersion}`);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\t!manifest.sessions.some(\n\t\t\t\t\t\t(session) => session.activeSessionId === worker.descriptor.rootActiveSessionId,\n\t\t\t\t\t) &&\n\t\t\t\t\t!manifest.discardedActiveSessionIds?.includes(worker.descriptor.rootActiveSessionId)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t`Worker ${worker.descriptor.workerId} omitted its root disposition from the update manifest`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\treturn { worker, manifest };\n\t\t\t}),\n\t\t);\n\t\tconst cancelAcknowledged = async () => {\n\t\t\tawait Promise.all(\n\t\t\t\tacknowledged.map(async (worker) => {\n\t\t\t\t\tconst prepareClient = worker.updateRestartPrepareClient;\n\t\t\t\t\tworker.updateRestartPrepareClient = undefined;\n\t\t\t\t\tif (!prepareClient) return;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst response = await prepareClient.requestWorker({ type: \"worker_cancel_update\" }, 5000);\n\t\t\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t`Could not cancel prepared worker ${worker.descriptor.workerId}; reconnecting it: ${String(error)}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t\tprepareClient.close();\n\t\t\t\t\t\tif (worker.client && worker.client !== prepareClient) worker.client.close();\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t};\n\t\tconst preparationFailure = preparationResults.find(\n\t\t\t(result): result is PromiseRejectedResult => result.status === \"rejected\",\n\t\t);\n\t\tif (preparationFailure) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow preparationFailure.reason;\n\t\t}\n\t\tconst prepared = preparationResults.flatMap((result) =>\n\t\t\tresult.status === \"fulfilled\" ? [result.value.worker] : [],\n\t\t);\n\t\tconst responses = preparationResults.flatMap((result) =>\n\t\t\tresult.status === \"fulfilled\" ? [result.value.manifest] : [],\n\t\t);\n\t\tconst discardedActiveSessionIds = responses.flatMap((manifest) => manifest.discardedActiveSessionIds ?? []);\n\t\tconst manifest: DaemonUpdateRestartManifest = {\n\t\t\tformatVersion: DAEMON_UPDATE_RESTART_FORMAT_VERSION,\n\t\t\tcreatedAt: new Date().toISOString(),\n\t\t\tsessions: responses.flatMap((manifest) => manifest.sessions),\n\t\t\t...(discardedActiveSessionIds.length > 0 ? { discardedActiveSessionIds } : {}),\n\t\t};\n\t\t// A worker that disconnected after preparing cancelled its checkpoint with\n\t\t// the old client; a recovered replacement may have admitted inputs past the\n\t\t// captured manifest. Abort before the manifest is persisted so the caller's\n\t\t// fallback cannot restore from the stale checkpoint.\n\t\tconst staleWorker = prepared.find((worker) => worker.client !== worker.updateRestartPrepareClient);\n\t\tif (staleWorker) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow new Error(\n\t\t\t\t`Worker ${staleWorker.descriptor.workerId} reconnected during update preparation; its checkpoint is stale`,\n\t\t\t);\n\t\t}\n\t\ttry {\n\t\t\tthis.validateAndPersistUpdateManifest(manifest);\n\t\t} catch (error) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow error;\n\t\t}\n\t\t// Commit through the connection that owns the prepared transaction; a client\n\t\t// swapped in after the check above must fail the commit rather than reach a\n\t\t// worker that no longer holds the checkpoint.\n\t\tconst commitClients = new Map(prepared.map((worker) => [worker, worker.updateRestartPrepareClient]));\n\t\tfor (const worker of prepared) worker.updateRestartPrepareClient = undefined;\n\t\tconst commitResults = await Promise.allSettled(\n\t\t\tprepared.map(async (worker) => {\n\t\t\t\tconst client = commitClients.get(worker);\n\t\t\t\tif (!client) throw new Error(`Worker ${worker.descriptor.workerId} disconnected before update commit`);\n\t\t\t\tconst response = await client.requestWorker(\n\t\t\t\t\t{ type: \"worker_commit_update\" },\n\t\t\t\t\tUPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS,\n\t\t\t\t);\n\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t}),\n\t\t);\n\t\tconst commitFailure = commitResults.find(\n\t\t\t(result): result is PromiseRejectedResult => result.status === \"rejected\",\n\t\t);\n\t\tif (commitFailure) {\n\t\t\tthis.log(`Update restart commit response failed; forcing restart completion: ${String(commitFailure.reason)}`);\n\t\t\tawait Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false, true)));\n\t\t\treturn manifest;\n\t\t}\n\t\tconst stopResults = await Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false)));\n\t\tif (stopResults.some((result) => result.status === \"rejected\")) {\n\t\t\tthis.log(\"A committed update worker did not stop gracefully; forcing restart completion\");\n\t\t\tawait Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false, true)));\n\t\t}\n\t\treturn manifest;\n\t}\n\n\tprivate validateAndPersistUpdateManifest(manifest: DaemonUpdateRestartManifest): void {\n\t\tif (manifest.formatVersion !== DAEMON_UPDATE_RESTART_FORMAT_VERSION) {\n\t\t\tthrow new Error(`Unsupported update manifest version ${manifest.formatVersion}`);\n\t\t}\n\t\tconst activeSessionIds = new Set<string>();\n\t\tconst sessionFiles = new Set<string>();\n\t\tfor (const discardedActiveSessionId of manifest.discardedActiveSessionIds ?? []) {\n\t\t\tif (!discardedActiveSessionId || activeSessionIds.has(discardedActiveSessionId)) {\n\t\t\t\tthrow new Error(\"Update manifest contains an invalid discarded session disposition\");\n\t\t\t}\n\t\t\tactiveSessionIds.add(discardedActiveSessionId);\n\t\t}\n\t\tfor (const session of manifest.sessions) {\n\t\t\tif (!session.activeSessionId || !session.sessionFile) {\n\t\t\t\tthrow new Error(\"Update manifest contains an incomplete session checkpoint\");\n\t\t\t}\n\t\t\tif (activeSessionIds.has(session.activeSessionId)) {\n\t\t\t\tthrow new Error(`Update manifest contains duplicate active session ${session.activeSessionId}`);\n\t\t\t}\n\t\t\tconst sessionFile = canonicalSessionPath(session.sessionFile);\n\t\t\tif (sessionFiles.has(sessionFile)) {\n\t\t\t\tthrow new Error(`Update manifest contains duplicate session file ${sessionFile}`);\n\t\t\t}\n\t\t\tactiveSessionIds.add(session.activeSessionId);\n\t\t\tsessionFiles.add(sessionFile);\n\t\t}\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) {\n\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t}\n\t\tconst path = getDaemonUpdateRestartManifestPath(this.socketPath, agentDir);\n\t\tmkdirSync(dirname(path), { recursive: true, mode: 0o700 });\n\t\tconst tempPath = `${path}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(manifest)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\tconst validated = JSON.parse(readFileSync(tempPath, \"utf8\")) as DaemonUpdateRestartManifest;\n\t\tif (!Array.isArray(validated.sessions) || validated.sessions.length !== manifest.sessions.length) {\n\t\t\tthrow new Error(\"Could not validate aggregate update manifest\");\n\t\t}\n\t\trenameSync(tempPath, path);\n\t}\n\n\t/**\n\t * Verdict on whether a pid is still the process we launched. Callers must\n\t * be conservative in both directions: signal a pid only on \"current\"\n\t * (never SIGKILL a recycled pid), and clean up a registration only on\n\t * \"gone\"/\"replaced\" (never orphan a live worker because a transient\n\t * identity lookup failed).\n\t */\n\tprivate processIdentity(\n\t\tpid: number,\n\t\tprocessStartId: string | undefined,\n\t): \"current\" | \"replaced\" | \"gone\" | \"unknown\" {\n\t\tif (!isProcessAlive(pid)) {\n\t\t\treturn \"gone\";\n\t\t}\n\t\tif (processStartId === undefined) {\n\t\t\treturn \"unknown\";\n\t\t}\n\t\tconst observed = getProcessStartId(pid);\n\t\tif (observed === undefined) {\n\t\t\treturn \"unknown\";\n\t\t}\n\t\treturn observed === processStartId ? \"current\" : \"replaced\";\n\t}\n\n\t/**\n\t * Keep an exact stop's registration and descriptor authoritative while any\n\t * part of its cleanup is in flight. Root kills acquire this before forwarding\n\t * because a synchronous shutdown event may arrive before the worker replies.\n\t */\n\tprivate acquireWorkerStopOwnership(worker: ResidentWorker): () => void {\n\t\tif (!this.workerStopCounts) this.workerStopCounts = new Map();\n\t\tconst stopCounts = this.workerStopCounts;\n\t\tstopCounts.set(worker, (stopCounts.get(worker) ?? 0) + 1);\n\t\tlet released = false;\n\t\treturn () => {\n\t\t\tif (released) return;\n\t\t\treleased = true;\n\t\t\tconst remaining = (stopCounts.get(worker) ?? 1) - 1;\n\t\t\tif (remaining === 0) stopCounts.delete(worker);\n\t\t\telse stopCounts.set(worker, remaining);\n\t\t};\n\t}\n\n\tprivate async stopWorker(\n\t\tworker: ResidentWorker,\n\t\tremoveDescriptor: boolean,\n\t\tforce = false,\n\t\tarchiveSession = false,\n\t\trecoveryCleanup = false,\n\t\tdirectChild?: { child: ChildProcess; closed: Promise<void> },\n\t): Promise<void> {\n\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\tconst releaseStopOwnership = this.acquireWorkerStopOwnership(worker);\n\t\ttry {\n\t\t\tawait this.stopWorkerUntracked(worker, removeDescriptor, force, archiveSession, recoveryCleanup, directChild);\n\t\t} finally {\n\t\t\treleaseStopOwnership();\n\t\t}\n\t}\n\n\tprivate async stopWorkerUntracked(\n\t\tworker: ResidentWorker,\n\t\tremoveDescriptor: boolean,\n\t\tforce = false,\n\t\tarchiveSession = false,\n\t\trecoveryCleanup = false,\n\t\tdirectChild?: { child: ChildProcess; closed: Promise<void> },\n\t): Promise<void> {\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tif (!recoveryCleanup) {\n\t\t\tworker.stopRevision++;\n\t\t}\n\t\t// A retry can rescind this stop and relaunch the worker while we await\n\t\t// below. Bind every liveness check and signal to the process this stop\n\t\t// entered with, and abort cleanup once the stop no longer applies: the\n\t\t// pid changed (relaunched) or a removeDescriptor stop lost its tombstone\n\t\t// (rescinded, even before the successor pid lands).\n\t\tconst entryPid = worker.descriptor.pid;\n\t\tconst entryStartId = worker.descriptor.processStartId;\n\t\tconst assertStopStillApplies = () => {\n\t\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\t\tif (directChild) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (\n\t\t\t\tworker.descriptor.pid !== entryPid ||\n\t\t\t\t(removeDescriptor && worker.descriptor.stopRequestedAt === undefined)\n\t\t\t) {\n\t\t\t\tthrow new Error(`Session worker ${worker.descriptor.workerId} was relaunched during stop`);\n\t\t\t}\n\t\t};\n\t\ttry {\n\t\t\tif (removeDescriptor) {\n\t\t\t\tthis.persistWorkerStopTombstone(worker, archiveSession);\n\t\t\t} else {\n\t\t\t\tworker.intentionalStop = true;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tif (!directChild) {\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tthis.reportCleanupFailure(`worker rollback state ${worker.descriptor.workerId}`, error);\n\t\t}\n\t\tconst transferError = new Error(\"Session worker stopped during snapshot transfer\");\n\t\tconst generationTranscripts = new Set<SnapshotTranscriptCache>();\n\t\tfor (const [activeSessionId, generations] of [...(worker.snapshotGenerations ?? new Map())]) {\n\t\t\tfor (const generation of [...generations.values()]) {\n\t\t\t\tgenerationTranscripts.add(generation.transcript);\n\t\t\t\tif (generation.incoming || !generation.transcript.complete || generation.validation) {\n\t\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, transferError);\n\t\t\t\t} else {\n\t\t\t\t\tgeneration.transcript.dispose();\n\t\t\t\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tfor (const transcript of worker.transcriptCaches.values()) {\n\t\t\tif (!generationTranscripts.has(transcript) && !transcript.complete) {\n\t\t\t\ttranscript.markFailed(transferError);\n\t\t\t}\n\t\t\ttranscript.dispose();\n\t\t}\n\t\tworker.transcriptCaches.clear();\n\t\tworker.snapshotCache.clear();\n\t\tworker.snapshotGenerations?.clear();\n\t\tif (worker.client) {\n\t\t\tif (archiveSession) {\n\t\t\t\tawait worker.client\n\t\t\t\t\t.requestWorker({ type: \"worker_archive_and_shutdown\" }, force ? 1000 : 5000)\n\t\t\t\t\t.catch(() => undefined);\n\t\t\t} else {\n\t\t\t\tawait worker.client.request({ type: \"shutdown\" }, force ? 1000 : 5000).catch(() => undefined);\n\t\t\t}\n\t\t\tassertStopStillApplies();\n\t\t\tworker.client.close();\n\t\t\tworker.client = undefined;\n\t\t} else if (directChild) {\n\t\t\tdirectChild.child.kill(\"SIGTERM\");\n\t\t} else if (this.processIdentity(entryPid, entryStartId) === \"current\") {\n\t\t\tsignalProcessGroupOrProcess(entryPid, \"SIGTERM\");\n\t\t}\n\t\t// Identity-aware in both directions: a replaced pid counts as gone (never\n\t\t// signal a recycled pid) while an unknown identity counts as alive (never\n\t\t// clean up a possibly-live worker on a transient lookup failure). kill(0)\n\t\t// runs on every poll; the expensive identity check is throttled.\n\t\tlet identityVerdict: \"current\" | \"replaced\" | \"gone\" | \"unknown\" = \"current\";\n\t\tlet identityCheckedAt = 0;\n\t\tconst isWorkerProcessAlive = () => {\n\t\t\tif (directChild) {\n\t\t\t\treturn directChild.child.exitCode === null && directChild.child.signalCode === null;\n\t\t\t}\n\t\t\tif (!processIdExists(entryPid)) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst now = Date.now();\n\t\t\tif (now - identityCheckedAt >= LIVENESS_IDENTITY_RECHECK_MS) {\n\t\t\t\tidentityCheckedAt = now;\n\t\t\t\tidentityVerdict = this.processIdentity(entryPid, entryStartId);\n\t\t\t}\n\t\t\treturn identityVerdict !== \"replaced\" && identityVerdict !== \"gone\";\n\t\t};\n\t\tconst gracefulDeadline = Date.now() + (force ? 500 : 2000);\n\t\twhile (isWorkerProcessAlive() && Date.now() < gracefulDeadline) {\n\t\t\tawait delay(25);\n\t\t}\n\t\tlet sigkillSent = false;\n\t\tif (force && isWorkerProcessAlive()) {\n\t\t\tassertStopStillApplies();\n\t\t\tif (directChild) {\n\t\t\t\tsigkillSent = directChild.child.kill(\"SIGKILL\");\n\t\t\t} else if (this.processIdentity(entryPid, entryStartId) === \"current\") {\n\t\t\t\t// Fresh, unthrottled check: the cached verdict may be up to 500ms\n\t\t\t\t// old, long enough for the pid to be recycled.\n\t\t\t\tsignalProcessGroupOrProcess(entryPid, \"SIGKILL\");\n\t\t\t\tsigkillSent = true;\n\t\t\t}\n\t\t\tconst forceDeadline = Date.now() + 1000;\n\t\t\twhile (isWorkerProcessAlive() && Date.now() < forceDeadline) {\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tif (isWorkerProcessAlive()) {\n\t\t\tworker.intentionalStop = worker.descriptor.stopRequestedAt !== undefined;\n\t\t\tif (removeDescriptor) {\n\t\t\t\tthis.scheduleWorkerStopFinalization(worker);\n\t\t\t}\n\t\t\tthrow new WorkerStopTimeoutError(\n\t\t\t\t`Session worker ${worker.descriptor.workerId} did not stop${sigkillSent ? \" after SIGKILL\" : \"\"}`,\n\t\t\t);\n\t\t}\n\t\tif (directChild) {\n\t\t\tawait directChild.closed;\n\t\t}\n\t\tassertStopStillApplies();\n\t\tif (removeDescriptor && worker.descriptor.archiveOnStop) {\n\t\t\tif (force) {\n\t\t\t\tthis.reclaimStoppedWorkerCronLock(worker);\n\t\t\t}\n\t\t\tawait this.finalizeArchivedWorkerStop(worker);\n\t\t\tassertStopStillApplies();\n\t\t}\n\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t// Client-owned schedules die with the registration, like their roster rows. Cancel\n\t\t// before the worker leaves the map, so no recompute sees the tree uncovered mid-stop.\n\t\tlet ephemeralCancelSettled = true;\n\t\tif (removeDescriptor && worker.descriptor.ownerClientId !== undefined) {\n\t\t\tephemeralCancelSettled = await this.cancelEphemeralWorkerScheduledJobs(worker);\n\t\t}\n\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t// A failed cancel keeps the stop tombstone as the durable intent; the enumeration retry or the next boot finishes it.\n\t\tif (removeDescriptor && ephemeralCancelSettled) {\n\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t}\n\t\tif (!this.shuttingDown) {\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t}\n\t}\n\n\t/**\n\t * A stop that timed out leaves a tombstoned registration behind. Keep\n\t * escalating in the background until the process is gone, then finish the\n\t * interrupted cleanup instead of leaving a dead worker registered forever.\n\t */\n\tprivate scheduleWorkerStopFinalization(worker: ResidentWorker): void {\n\t\tif (worker.stopFinalization) {\n\t\t\treturn;\n\t\t}\n\t\tworker.stopFinalization = this.finalizeTimedOutWorkerStop(worker).finally(() => {\n\t\t\tworker.stopFinalization = undefined;\n\t\t});\n\t}\n\n\tprivate async finalizeTimedOutWorkerStop(worker: ResidentWorker): Promise<void> {\n\t\t// Bind to the exact process generation being stopped: a retry can rescind\n\t\t// the stop and relaunch with a new pid, and the OS can recycle the old\n\t\t// pid. The finalizer must never follow either successor.\n\t\tconst pid = worker.descriptor.pid;\n\t\tconst processStartId = worker.descriptor.processStartId;\n\t\tconst stopRevision = worker.stopRevision;\n\t\tconst isStopGenerationCurrent = () =>\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.stopRevision === stopRevision &&\n\t\t\tworker.descriptor.stopRequestedAt !== undefined &&\n\t\t\tworker.descriptor.pid === pid;\n\t\t// A replaced pid counts as gone (never SIGKILL a recycled pid); an\n\t\t// unobservable identity counts as alive (never clean up a possibly-live\n\t\t// worker). kill(0) probes every poll; ps-backed checks are throttled.\n\t\tlet stoppedVerdict = true;\n\t\tlet stoppedCanSignal = processStartId !== undefined;\n\t\tlet stoppedCheckedAt = 0;\n\t\tconst isStoppedProcessAlive = () => {\n\t\t\tif (!processIdExists(pid)) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst now = Date.now();\n\t\t\tif (now - stoppedCheckedAt < LIVENESS_IDENTITY_RECHECK_MS) {\n\t\t\t\treturn stoppedVerdict;\n\t\t\t}\n\t\t\tstoppedCheckedAt = now;\n\t\t\tif (!isProcessAlive(pid)) {\n\t\t\t\tstoppedVerdict = false;\n\t\t\t} else if (processStartId === undefined) {\n\t\t\t\tstoppedVerdict = true;\n\t\t\t\t// Without an identity captured while the original worker was known\n\t\t\t\t// alive, this pid may now belong to an unrelated process. Keep\n\t\t\t\t// waiting for it to disappear, but never escalate by pid alone.\n\t\t\t\tstoppedCanSignal = false;\n\t\t\t} else {\n\t\t\t\tconst observed = getProcessStartId(pid);\n\t\t\t\tstoppedVerdict = observed !== processStartId ? observed === undefined : true;\n\t\t\t\tstoppedCanSignal = observed === processStartId;\n\t\t\t}\n\t\t\treturn stoppedVerdict;\n\t\t};\n\t\tconst sigkillDeadline = Date.now() + STOP_FINALIZATION_SIGKILL_GRACE_MS;\n\t\tlet killed = false;\n\t\twhile (!this.shuttingDown) {\n\t\t\tif (!isStopGenerationCurrent()) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!isStoppedProcessAlive()) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tif (!killed && stoppedCanSignal && Date.now() >= sigkillDeadline) {\n\t\t\t\t// Fresh, unthrottled identity check right before signalling: the\n\t\t\t\t// cached verdict may be up to 500ms old, long enough for the pid\n\t\t\t\t// to be recycled by an unrelated process. A transiently\n\t\t\t\t// unobservable identity skips this attempt but keeps escalation\n\t\t\t\t// armed so a wedged worker is still killed on a later pass.\n\t\t\t\tconst observedNow = processStartId === undefined ? undefined : getProcessStartId(pid);\n\t\t\t\tif (processStartId === undefined || observedNow === processStartId) {\n\t\t\t\t\tsignalProcessGroupOrProcess(pid, \"SIGKILL\");\n\t\t\t\t\tkilled = true;\n\t\t\t\t}\n\t\t\t}\n\t\t\tawait unrefDelay(STOP_FINALIZATION_RECHECK_MS);\n\t\t}\n\t\t// Retry transient cleanup failures (for example catalog archival) so a\n\t\t// dead worker's registration is never stranded permanently. Each attempt\n\t\t// bumps the worker's stopRevision, so rescission is detected through the\n\t\t// registration and tombstone instead of the waiting-phase snapshot.\n\t\tconst isCleanupStillWanted = () =>\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.descriptor.stopRequestedAt !== undefined &&\n\t\t\tworker.descriptor.pid === pid;\n\t\twhile (!this.shuttingDown && isCleanupStillWanted()) {\n\t\t\ttry {\n\t\t\t\tawait this.stopWorker(worker, true, true, worker.descriptor.archiveOnStop === true);\n\t\t\t\tthis.log(`Finalized timed-out stop for worker ${worker.descriptor.workerId}`);\n\t\t\t\treturn;\n\t\t\t} catch (error) {\n\t\t\t\tthis.reportCleanupFailure(`timed-out worker stop ${worker.descriptor.workerId}`, error);\n\t\t\t\tawait unrefDelay(STOP_FINALIZATION_RETRY_MS);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async finalizeArchivedWorkerStop(worker: ResidentWorker): Promise<void> {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.rootSessionId) {\n\t\t\tconst cronStore = AgentCronJobStore.forSessionArtifacts();\n\t\t\tcronStore.registerSessionArtifact(worker.descriptor.rootSessionId, context.artifactDir);\n\t\t\tcronStore.cancelJobsForSession({\n\t\t\t\tsessionId: worker.descriptor.rootSessionId,\n\t\t\t\tsessionFile: context.sessionFile,\n\t\t\t});\n\t\t\tawait this.catalog.archive(context.sessionFile, worker.descriptor.rootSessionId);\n\t\t}\n\t}\n\n\tprivate async cancelScheduledJobsForSessionTree(\n\t\trootSessionId: string,\n\t\trootSessionFile: string,\n\t\tstillWanted?: () => boolean,\n\t\texclude?: ResidentWorker,\n\t): Promise<void> {\n\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\tconst sessions = [{ sessionId: rootSessionId, sessionFile: rootSessionFile }];\n\t\tconst childrenByParent = new Map<string, SessionInfo[]>();\n\t\tfor (const info of await this.rlmSpawnLedger().family()) {\n\t\t\tif (!info.parentSessionPath) continue;\n\t\t\tconst key = canonicalSessionPath(info.parentSessionPath);\n\t\t\tchildrenByParent.set(key, [...(childrenByParent.get(key) ?? []), info]);\n\t\t}\n\t\tconst queue = [canonicalSessionPath(rootSessionFile)];\n\t\tconst visited = new Set(queue);\n\t\twhile (queue.length > 0) {\n\t\t\tfor (const info of childrenByParent.get(queue.shift()!) ?? []) {\n\t\t\t\tconst child = canonicalSessionPath(info.path);\n\t\t\t\tif (visited.has(child)) continue;\n\t\t\t\tvisited.add(child);\n\t\t\t\tsessions.push({ sessionId: info.id, sessionFile: info.path });\n\t\t\t\tqueue.push(child);\n\t\t\t}\n\t\t}\n\t\t// A worker covering any tree member owns its stores again; a stale intent must not kill new schedules.\n\t\tfor (const { sessionFile } of sessions) {\n\t\t\ttry {\n\t\t\t\tif (this.findWorkerBySessionFile(sessionFile, exclude)) return;\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\t\tlet registered = false;\n\t\tfor (const { sessionId, sessionFile } of sessions) {\n\t\t\tconst artifactDir = getSessionArtifactPathForFile(sessionFile, sessionId);\n\t\t\tif (!existsSync(join(artifactDir, SESSION_SCHEDULED_JOBS_FILENAME))) continue;\n\t\t\tstore.registerSessionArtifact(sessionId, artifactDir);\n\t\t\tregistered = true;\n\t\t}\n\t\tif (!registered) return;\n\t\t// Re-checked in the same synchronous turn as the walk: a promotion committed during the family read keeps its schedules.\n\t\tif (stillWanted && !stillWanted()) return;\n\t\tfor (const { sessionFile } of sessions) {\n\t\t\tstore.cancelJobsForSession({ sessionFile });\n\t\t}\n\t}\n\n\tprivate async cancelEphemeralWorkerScheduledJobs(worker: ResidentWorker): Promise<boolean> {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context || !worker.descriptor.rootSessionId) {\n\t\t\treturn true;\n\t\t}\n\t\ttry {\n\t\t\tawait this.cancelScheduledJobsForSessionTree(\n\t\t\t\tworker.descriptor.rootSessionId,\n\t\t\t\tcontext.sessionFile,\n\t\t\t\t() => worker.descriptor.ownerClientId !== undefined,\n\t\t\t\tworker,\n\t\t\t);\n\t\t\treturn true;\n\t\t} catch (error) {\n\t\t\t// A promotion that landed during the failed read means the cancel is no longer wanted.\n\t\t\tif (worker.descriptor.ownerClientId === undefined) {\n\t\t\t\treturn true;\n\t\t\t}\n\t\t\tthis.log(\n\t\t\t\t`Could not cancel scheduled jobs for client-owned worker ${worker.descriptor.workerId}: ${String(error)}`,\n\t\t\t);\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate reclaimStoppedWorkerCronLock(worker: ResidentWorker): void {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context) {\n\t\t\treturn;\n\t\t}\n\t\trmSync(join(context.artifactDir, `${SESSION_SCHEDULED_JOBS_FILENAME}.lock`), { recursive: true, force: true });\n\t}\n\n\tprivate workerSessionArtifactContext(worker: {\n\t\tdescriptor: DaemonWorkerDescriptor;\n\t}): { sessionFile: string; artifactDir: string } | undefined {\n\t\tconst sessionFile = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tif (!sessionFile || !worker.descriptor.rootSessionId) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn {\n\t\t\tsessionFile,\n\t\t\tartifactDir: getSessionArtifactPathForFile(sessionFile, worker.descriptor.rootSessionId),\n\t\t};\n\t}\n\n\tprivate persistWorkerStopTombstone(worker: ResidentWorker, archiveSession = false): void {\n\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\tworker.intentionalStop = true;\n\t\tworker.descriptor.stopRequestedAt ??= new Date().toISOString();\n\t\tworker.descriptor.archiveOnStop ||= archiveSession;\n\t\tthis.persistWorker(worker);\n\t}\n\n\tprivate write(client: DaemonSocketClient, message: DaemonOutbound): boolean {\n\t\treturn this.writeSerialized(client, serializeJsonLine(message));\n\t}\n\n\tprivate broadcastHeartbeatsChanged(): void {\n\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\tfor (const client of this.clients) {\n\t\t\tthis.write(client, { type: \"heartbeats_changed\" });\n\t\t}\n\t}\n\n\tprivate writeSerialized(client: DaemonSocketClient, line: string | Uint8Array): boolean {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn false;\n\t\t}\n\t\tconst accepted = client.socket.write(line);\n\t\tif (!accepted) {\n\t\t\tclient.backpressured = true;\n\t\t}\n\t\treturn accepted;\n\t}\n\n\tprivate registerSignalHandlers(): void {\n\t\tconst signals: NodeJS.Signals[] = [\"SIGINT\", \"SIGTERM\"];\n\t\tif (process.platform !== \"win32\") {\n\t\t\tsignals.push(\"SIGHUP\");\n\t\t}\n\t\tfor (const signal of signals) {\n\t\t\tconst handler = () => void this.shutdown(signal === \"SIGINT\" ? 130 : signal === \"SIGHUP\" ? 129 : 143, false);\n\t\t\tprocess.on(signal, handler);\n\t\t\tthis.signalCleanupHandlers.push(() => process.off(signal, handler));\n\t\t}\n\t\tconst exitHandler = () => this.cleanupSocket();\n\t\tprocess.on(\"exit\", exitHandler);\n\t\tthis.signalCleanupHandlers.push(() => process.off(\"exit\", exitHandler));\n\t}\n\n\tprivate assertSocketLeaseHeld(): void {\n\t\tconst compromise = this.socketLeaseCompromise ?? this.socketLease?.compromise;\n\t\tif (compromise) throw new Error(`Daemon socket lease was compromised: ${compromise.message}`);\n\t}\n\n\tprivate assertSupervisorServing(allowWorkerLedger = false): void {\n\t\tthis.assertSocketLeaseHeld();\n\t\tif (this.shuttingDown && (!allowWorkerLedger || this.rlmJournalClose !== undefined)) {\n\t\t\tconst error = new Error(`Daemon supervisor generation ${this.generation} is shutting down; retry the command`);\n\t\t\tObject.assign(error, { code: \"supervisor_generation_stale\" as const });\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate fenceSupervisorSocket(): void {\n\t\ttry {\n\t\t\tthis.server?.close();\n\t\t} catch {\n\t\t\t// The server may already be closed by a concurrent shutdown.\n\t\t}\n\t\tfor (const client of this.clients) {\n\t\t\tclient.detachInput();\n\t\t\tclient.socket.destroy();\n\t\t}\n\t}\n\n\tprivate handleSocketLeaseCompromised(error: Error): void {\n\t\tif (this.socketLeaseCompromise) return;\n\t\tthis.socketLeaseCompromise = error;\n\t\tthis.shuttingDown = true;\n\t\tthis.fenceRlmJournalOwner();\n\t\tthis.fenceSupervisorSocket();\n\t\tconst message = `Daemon socket lease was compromised; relinquishing supervisor ownership: ${error.message}`;\n\t\ttry {\n\t\t\tthis.log(message);\n\t\t} catch {\n\t\t\tconsole.error(message);\n\t\t}\n\t\tif (!this.startupComplete) return;\n\t\tvoid this.cleanupSupervisorResources().catch((cleanupError) =>\n\t\t\tthis.reportCleanupFailure(\"compromised daemon socket lease\", cleanupError),\n\t\t);\n\t}\n\n\tprivate cleanupSocket(): void {\n\t\tif (!this.ownsSocketPath) {\n\t\t\treturn;\n\t\t}\n\t\tthis.ownsSocketPath = false;\n\t\tconst identity = this.socketIdentity;\n\t\tthis.socketIdentity = undefined;\n\t\tcleanupDaemonSocketPath(this.socketPath, identity, this.socketLease);\n\t}\n\n\tprivate async cleanupSupervisorResources(): Promise<void> {\n\t\tif (this.cleanupPromise) {\n\t\t\treturn this.cleanupPromise;\n\t\t}\n\t\tthis.cleanupPromise = this.cleanupSupervisorResourcesOnce();\n\t\treturn this.cleanupPromise;\n\t}\n\n\tprivate async cleanupSupervisorResourcesOnce(): Promise<void> {\n\t\tthis.shuttingDown = true;\n\t\tthis.clearIdleEvictionTimer();\n\t\tthis.clearScheduledWakeTimer();\n\t\tthis.clearRosterWatchdogTimer();\n\t\tawait this.idleEvictionSweep?.catch(() => undefined);\n\t\tfor (const cleanup of this.signalCleanupHandlers.splice(0)) {\n\t\t\tawait this.runCleanupStep(\"signal handler\", cleanup);\n\t\t}\n\t\tconst server = this.server;\n\t\tthis.server = undefined;\n\t\tconst serverClosed = new Promise<void>((resolveClose) => {\n\t\t\tif (!server?.listening) {\n\t\t\t\tresolveClose();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tserver.close(() => resolveClose());\n\t\t\t} catch (error) {\n\t\t\t\tthis.reportCleanupFailure(\"daemon server\", error);\n\t\t\t\tresolveClose();\n\t\t\t}\n\t\t});\n\t\tfor (const client of this.clients) {\n\t\t\tclient.attachedActiveSessionIds.clear();\n\t\t\tawait this.runCleanupStep(`daemon client input ${client.id}`, () => client.detachInput());\n\t\t\tawait this.runCleanupStep(`daemon client socket ${client.id}`, () => {\n\t\t\t\tclient.socket.destroy();\n\t\t\t});\n\t\t}\n\t\tthis.clients.clear();\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.ownerCleanupTimer) {\n\t\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t\t}\n\t\t\tawait this.runCleanupStep(`worker client ${worker.descriptor.workerId}`, () => worker.client?.close());\n\t\t\tworker.client = undefined;\n\t\t\tconst transcripts = new Set(worker.transcriptCaches.values());\n\t\t\tfor (const generations of worker.snapshotGenerations?.values() ?? []) {\n\t\t\t\tfor (const generation of generations.values()) {\n\t\t\t\t\ttranscripts.add(generation.transcript);\n\t\t\t\t\tthis.settleSnapshotDuplicateValidation(\n\t\t\t\t\t\tgeneration,\n\t\t\t\t\t\tnew Error(\"Daemon supervisor stopped during snapshot transfer\"),\n\t\t\t\t\t);\n\t\t\t\t\tif (!generation.transcript.complete) {\n\t\t\t\t\t\tgeneration.transcript.markFailed(new Error(\"Daemon supervisor stopped during snapshot transfer\"));\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tfor (const transcript of transcripts) {\n\t\t\t\tawait this.runCleanupStep(`worker transcript ${worker.descriptor.workerId}`, () => transcript.dispose());\n\t\t\t}\n\t\t\tworker.transcriptCaches.clear();\n\t\t\tworker.snapshotGenerations?.clear();\n\t\t\tworker.snapshotCache.clear();\n\t\t\tworker.snapshotLoads.clear();\n\t\t}\n\t\tthis.workers.clear();\n\t\tthis.openingWorkers.clear();\n\t\tawait this.runCleanupStep(\"daemon catalog\", () => this.catalog.stop());\n\t\tawait this.runCleanupStep(\"daemon server\", () => serverClosed);\n\t\tawait this.runCleanupStep(\"daemon socket\", () => this.cleanupSocket());\n\t\tawait this.runCleanupStep(\"supervisor cache\", () => {\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t});\n\t\tawait this.closeRlmJournalOwner();\n\t\tconst lease = this.socketLease;\n\t\tthis.socketLease = undefined;\n\t\tawait this.runCleanupStep(\"daemon socket lock\", async () => lease?.release());\n\t\tconst ownership = this.ownership;\n\t\tthis.ownership = undefined;\n\t\tawait this.runCleanupStep(\"daemon ownership\", async () => ownership?.release());\n\t}\n\n\tprivate async runCleanupStep(label: string, action: () => void | Promise<void>): Promise<void> {\n\t\ttry {\n\t\t\tawait action();\n\t\t} catch (error) {\n\t\t\tthis.reportCleanupFailure(label, error);\n\t\t}\n\t}\n\n\tprivate reportCleanupFailure(label: string, error: unknown): void {\n\t\tconst message = `Failed to clean up ${label}: ${String(error)}`;\n\t\ttry {\n\t\t\tthis.log(message);\n\t\t} catch {\n\t\t\tconsole.error(message);\n\t\t}\n\t}\n\n\tprivate async shutdown(\n\t\texitCode: number,\n\t\tstopWorkers: boolean,\n\t\trelaunch = false,\n\t\tforceWorkers = false,\n\t\tclosingReason?: DaemonClosingReason,\n\t): Promise<never> {\n\t\tif (this.shuttingDown) {\n\t\t\tprocess.exit(exitCode);\n\t\t}\n\t\tthis.shuttingDown = true;\n\t\tthis.clearIdleEvictionTimer();\n\t\tthis.clearScheduledWakeTimer();\n\t\tthis.clearRosterWatchdogTimer();\n\t\tawait this.idleEvictionSweep?.catch(() => undefined);\n\t\tif (closingReason) {\n\t\t\tfor (const client of this.clients) {\n\t\t\t\tthis.write(client, { type: \"daemon_closing\", reason: closingReason });\n\t\t\t}\n\t\t}\n\t\tfor (const cleanup of this.signalCleanupHandlers) {\n\t\t\tcleanup();\n\t\t}\n\t\tif (stopWorkers) {\n\t\t\tawait Promise.all(\n\t\t\t\t[...this.workers.values()].map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.stopWorker(worker, true, forceWorkers, true);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\t\tthis.log(`Keeping live incompatible worker ${worker.descriptor.workerId}: ${error.message}`);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!(error instanceof WorkerStopTimeoutError)) {\n\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t`Worker ${worker.descriptor.workerId} remains tombstoned for recovery after shutdown: ${error.message}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t\tif (!this.hasPersistedWorkerDescriptors()) {\n\t\t\t\trmSync(this.supervisorConfigPath, { force: true });\n\t\t\t}\n\t\t} else {\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tif (worker.compatibilityError) continue;\n\t\t\t\tworker.intentionalStop = true;\n\t\t\t\tworker.client?.close();\n\t\t\t\tworker.client = undefined;\n\t\t\t}\n\t\t}\n\t\tawait this.catalog.stop();\n\t\tfor (const client of this.clients) {\n\t\t\tclient.detachInput();\n\t\t\tclient.socket.end();\n\t\t}\n\t\tawait new Promise<void>((resolveClose) => this.server?.close(() => resolveClose()) ?? resolveClose());\n\t\tawait this.runCleanupStep(\"daemon socket\", () => this.cleanupSocket());\n\t\tawait this.runCleanupStep(\"supervisor cache\", () => {\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t});\n\t\tawait this.closeRlmJournalOwner();\n\t\tconst lease = this.socketLease;\n\t\tthis.socketLease = undefined;\n\t\tawait this.runCleanupStep(\"daemon socket lock\", async () => lease?.release());\n\t\tconst ownership = this.ownership;\n\t\tthis.ownership = undefined;\n\t\tawait this.runCleanupStep(\"daemon ownership\", async () => ownership?.release());\n\t\tif (relaunch) {\n\t\t\tconst launch = createCliSubprocessLaunchSpec([\"--mode\", \"daemon\", \"--daemon-socket\", this.socketPath]);\n\t\t\tconst environment = createCliSubprocessEnv();\n\t\t\tdelete environment[DAEMON_CATALOG_ROLE_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_ROLE_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_TOKEN_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_ACTIVE_SESSION_ID_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_SUPERVISOR_SOCKET_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_RECOVERY_JOURNAL_ENV];\n\t\t\tdelete environment[ORPHAN_PROCESS_JOURNAL_ENV];\n\t\t\tdelete environment[SESSION_LEASES_ENABLED_ENV];\n\t\t\tdelete environment[SESSION_LEASE_OWNER_ID_ENV];\n\t\t\tconst replacement = spawn(launch.command, launch.args, {\n\t\t\t\tcwd: this.defaultSessionConfig.cwd ?? process.cwd(),\n\t\t\t\tdetached: true,\n\t\t\t\tenv: environment,\n\t\t\t\tstdio: \"ignore\",\n\t\t\t});\n\t\t\treplacement.unref();\n\t\t}\n\t\tprocess.exit(exitCode);\n\t}\n}\n"]}
+{"version":3,"file":"daemon-supervisor.d.ts","sourceRoot":"","sources":["../../../src/modes/daemon/daemon-supervisor.ts"],"names":[],"mappings":"AAiCA,OAAO,EACN,KAAK,yBAAyB,EAI9B,MAAM,oCAAoC,CAAC;AAiB5C,OAAO,EAEN,KAAK,mBAAmB,EAExB,MAAM,oCAAoC,CAAC;AAoV5C,UAAU,uBAAuB;IAChC,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,oBAAoB,EAAE,yBAAyB,CAAC;IAChD,aAAa,CAAC,EAAE,MAAM,CAAC;CACvB;AAiQD,wBAAgB,2BAA2B,CAAC,mBAAmB,EAAE,mBAAmB,GAAG,MAAM,CAM5F;AA8BD,wBAAsB,uBAAuB,CAAC,OAAO,EAAE,uBAAuB,GAAG,OAAO,CAAC,KAAK,CAAC,CAK9F;AAED,qBAAa,gBAAgB;IA4D3B,OAAO,CAAC,QAAQ,CAAC,UAAU;IA3D5B,OAAO,CAAC,MAAM,CAAC,CAAS;IACxB,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAgB;IACtC,OAAO,CAAC,SAAS,CAAwB;IACzC,OAAO,CAAC,WAAW,CAAoC;IACvD,OAAO,CAAC,cAAc,CAAS;IAC/B,OAAO,CAAC,cAAc,CAAC,CAAuB;IAC9C,OAAO,CAAC,WAAW,CAAC,CAAwB;IAC5C,OAAO,CAAC,qBAAqB,CAAC,CAAQ;IACtC,OAAO,CAAC,SAAS,CAAC,CAA+D;IACjF,OAAO,CAAC,cAAc,CAAC,CAAgB;IACvC,OAAO,CAAC,YAAY,CAAS;IAC7B,OAAO,CAAC,eAAe,CAAS;IAChC,OAAO,CAAC,kBAAkB,CAAC,CAAsC;IACjE,OAAO,CAAC,QAAQ,CAAC,aAAa,CAA4B;IAC1D,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAiC;IACzD,OAAO,CAAC,QAAQ,CAAC,aAAa,CAA6C;IAC3E,OAAO,CAAC,QAAQ,CAAC,uBAAuB,CAA6C;IACrF,OAAO,CAAC,QAAQ,CAAC,2BAA2B,CAAkD;IAC9F,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAA6C;IAC/E,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAqC;IAC7D,OAAO,CAAC,gBAAgB,CAAC,CAA8B;IACvD,OAAO,CAAC,QAAQ,CAAC,cAAc,CAA8C;IAC7E,0EAA0E;IAC1E,OAAO,CAAC,QAAQ,CAAC,gBAAgB,CAAyE;IAC1G,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAkD;IACrF,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAAyB;IAC/D,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAS;IACvC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAgB;IAC3C,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAS;IAC9C,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAA4B;IACjE,OAAO,CAAC,QAAQ,CAAC,iBAAiB,CAAS;IAC3C,OAAO,CAAC,cAAc,CAA0B;IAChD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAA6C;IACjF,OAAO,CAAC,QAAQ,CAAC,wBAAwB,CAAqB;IAC9D,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IACzD,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAsB;IAC9C,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAkB;IAClD,OAAO,CAAC,QAAQ,CAAC,WAAW,CAA+B;IAC3D,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAA6B;IAChE,OAAO,CAAC,QAAQ,CAAC,8BAA8B,CAA0C;IACzF,OAAO,CAAC,WAAW,CAAC,CAAc;IAClC,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;IAC1D,OAAO,CAAC,QAAQ,CAAC,oBAAoB,CAAqB;IAC1D,gGAAgG;IAChG,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAqB;IACxD,OAAO,CAAC,mBAAmB,CAAS;IACpC,OAAO,CAAC,mBAAmB,CAAC,CAAiC;IAC7D,OAAO,CAAC,sBAAsB,CAAC,CAAiB;IAChD,OAAO,CAAC,eAAe,CAAC,CAAkB;IAC1C,OAAO,CAAC,eAAe,CAAC,CAAgB;IACxC,OAAO,CAAC,iBAAiB,CAAC,CAAgC;IAC1D,OAAO,CAAC,iBAAiB,CAAC,CAAgB;IAC1C,OAAO,CAAC,iBAAiB,CAAC,CAAgB;IAC1C,OAAO,CAAC,kBAAkB,CAAC,CAAgC;IAC3D,OAAO,CAAC,sBAAsB,CAAC,CAAgB;IAC/C,OAAO,CAAC,4BAA4B,CAAS;IAC7C,OAAO,CAAC,QAAQ,CAAC,qBAAqB,CAA6B;IAEnE,YACkB,UAAU,EAAE,MAAM,EACnC,OAAO,EAAE,uBAAuB,EAqBhC;IAEK,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAkG3B;IAED,OAAO,CAAC,MAAM;IAgBd,OAAO,CAAC,GAAG;IAMX,OAAO,CAAC,sBAAsB;IAM9B,OAAO,CAAC,wBAAwB;IAMhC,OAAO,CAAC,uBAAuB;IAO/B,OAAO,CAAC,qCAAqC;YAkB/B,2BAA2B;YAoE3B,6BAA6B;YA8B7B,wBAAwB;IA8BtC,OAAO,CAAC,yBAAyB;IAgBjC,OAAO,CAAC,sBAAsB;IAwB9B,4HAA4H;IAC5H,OAAO,CAAC,0BAA0B;YAWpB,oBAAoB;YAgFpB,wBAAwB;YAcxB,iBAAiB;YAUjB,8BAA8B;YAW9B,6BAA6B;IAiC3C,OAAO,CAAC,8BAA8B;YAmBxB,sBAAsB;YAmBtB,6BAA6B;YAM7B,qBAAqB;IAOnC,OAAO,CAAC,6BAA6B;IAkBrC,OAAO,CAAC,qBAAqB;IAmC7B,4IAA4I;IAC5I,OAAO,CAAC,6BAA6B;IAyBrC,OAAO,CAAC,6BAA6B;IAqBrC,OAAO,CAAC,uBAAuB;IAY/B,OAAO,CAAC,6BAA6B;IAMrC,OAAO,CAAC,aAAa;IASrB,OAAO,CAAC,sBAAsB;IAY9B,OAAO,CAAC,gBAAgB;IAgFxB,OAAO,CAAC,wBAAwB;IAUhC,OAAO,CAAC,gBAAgB;YAIV,+BAA+B;IAsD7C,OAAO,CAAC,kCAAkC;IAY1C,OAAO,CAAC,mCAAmC;IAW3C,OAAO,CAAC,0BAA0B;IA2BlC,OAAO,CAAC,kBAAkB;IAI1B,OAAO,CAAC,mBAAmB;IAS3B,OAAO,CAAC,kBAAkB;IAQ1B,OAAO,CAAC,qBAAqB;IAQ7B,OAAO,CAAC,sCAAsC;IAgC9C,0FAA0F;IAC1F,OAAO,CAAC,sCAAsC;IAoD9C,OAAO,CAAC,sBAAsB;YAYhB,UAAU;IA0LxB,OAAO,CAAC,gBAAgB;YAIV,iBAAiB;IAmB/B,OAAO,CAAC,aAAa;IAIrB,OAAO,CAAC,wBAAwB;YAMlB,sBAAsB;YAqBtB,wBAAwB;YAwBxB,iBAAiB;YAoFjB,aAAa;YAgzBb,UAAU;YAqGV,iBAAiB;YASjB,kBAAkB;IAShC,OAAO,CAAC,qBAAqB;YAsBf,sBAAsB;YAqBtB,mBAAmB;YA8EnB,oBAAoB;YAmCpB,iBAAiB;IAU/B,OAAO,CAAC,uBAAuB;IAU/B,OAAO,CAAC,sBAAsB;IAS9B,OAAO,CAAC,mBAAmB;YAab,8BAA8B;YAmD9B,kBAAkB;IA4BhC,OAAO,CAAC,0BAA0B;YASpB,YAAY;YAwQZ,aAAa;IAwD3B,OAAO,CAAC,sBAAsB;YAiBhB,eAAe;YAoBf,oBAAoB;YAoGpB,sBAAsB;YA6CtB,iBAAiB;IAuD/B,OAAO,CAAC,wBAAwB;IAIhC,OAAO,CAAC,yBAAyB;IAUjC,OAAO,CAAC,mBAAmB;YAsBb,4BAA4B;IA+B1C,OAAO,CAAC,uBAAuB;IA+C/B,OAAO,CAAC,yBAAyB;IAyBjC,OAAO,CAAC,sBAAsB;IAa9B,OAAO,CAAC,kBAAkB;IAQ1B,OAAO,CAAC,yBAAyB;IA6BjC,OAAO,CAAC,wBAAwB;IAiBhC,OAAO,CAAC,sBAAsB;IAkB9B,OAAO,CAAC,iCAAiC;IAWzC,OAAO,CAAC,iCAAiC;YAa3B,aAAa;IAkK3B,OAAO,CAAC,yBAAyB;IASjC,OAAO,CAAC,wBAAwB;YAQlB,gCAAgC;YAoFhC,sBAAsB;YAyDtB,oBAAoB;YAiBpB,0BAA0B;YAgB1B,oCAAoC;IAalD,OAAO,CAAC,MAAM;IAKd,OAAO,CAAC,gBAAgB;IAWxB,OAAO,CAAC,kBAAkB;IAS1B,OAAO,CAAC,kBAAkB;IAiC1B,OAAO,CAAC,sBAAsB;IAM9B,OAAO,CAAC,6BAA6B;IAKrC,OAAO,CAAC,gBAAgB;IA2BxB,OAAO,CAAC,+BAA+B;IAMvC,OAAO,CAAC,mBAAmB;YAIb,gBAAgB;IAsB9B,OAAO,CAAC,6BAA6B;IAuBrC,OAAO,CAAC,wBAAwB;IAsBhC,OAAO,CAAC,sBAAsB;IAsB9B,OAAO,CAAC,0BAA0B;IAQlC,OAAO,CAAC,wBAAwB;IAYhC,OAAO,CAAC,sBAAsB;YAahB,yBAAyB;IAsEvC,OAAO,CAAC,iCAAiC;IAoBzC,OAAO,CAAC,6BAA6B;IASrC,OAAO,CAAC,uBAAuB;IAO/B,OAAO,CAAC,+BAA+B;IAkBvC,OAAO,CAAC,oBAAoB;IAkB5B,OAAO,CAAC,oBAAoB;IAQ5B,OAAO,CAAC,eAAe;IAOvB,OAAO,CAAC,yBAAyB;IAajC,OAAO,CAAC,oBAAoB;IAI5B,OAAO,CAAC,oBAAoB;IAa5B,sFAAsF;IACtF,OAAO,CAAC,cAAc;IAmBtB;;;;OAIG;IACH,OAAO,CAAC,iBAAiB;YAIX,gCAAgC;IAiB9C,OAAO,CAAC,2BAA2B;YAarB,yCAAyC;IAcvD,OAAO,CAAC,+BAA+B;IAyBvC,OAAO,CAAC,eAAe;IAIvB,gFAAgF;IAChF,OAAO,CAAC,gBAAgB;IAIxB,gEAAgE;IAChE,OAAO,CAAC,YAAY;IAIpB;;;OAGG;IACH,OAAO,CAAC,oBAAoB;IAU5B,OAAO,CAAC,4BAA4B;YAWtB,kBAAkB;IAkEhC,OAAO,CAAC,kBAAkB;IAe1B,OAAO,CAAC,gBAAgB;IAuBxB,OAAO,CAAC,mBAAmB;IAO3B,OAAO,CAAC,aAAa;YAUP,UAAU;IAsBxB,OAAO,CAAC,mBAAmB;IAI3B,OAAO,CAAC,0BAA0B;IAOlC,OAAO,CAAC,8BAA8B;IAMtC,OAAO,CAAC,YAAY;IAwBpB,OAAO,CAAC,mBAAmB;IAyB3B,OAAO,CAAC,uBAAuB;YA0BjB,eAAe;YAiBf,YAAY;IA4L1B,OAAO,CAAC,mBAAmB;IA6D3B,OAAO,CAAC,0BAA0B;IAyClC,OAAO,CAAC,0BAA0B;YAgBpB,cAAc;IAiF5B,OAAO,CAAC,qBAAqB;IAkC7B,OAAO,CAAC,mBAAmB;YAIb,mBAAmB;IA2BjC,OAAO,CAAC,YAAY;YAgBN,qBAAqB;IAUnC,OAAO,CAAC,iBAAiB;IAmbzB,OAAO,CAAC,wBAAwB;IAahC,OAAO,CAAC,sBAAsB;IAiB9B,OAAO,CAAC,YAAY;IAepB,OAAO,CAAC,aAAa;YAgBP,uBAAuB;YAWvB,mBAAmB;YAsFnB,oBAAoB;YAmBpB,0BAA0B;IAsIxC,OAAO,CAAC,gCAAgC;IA0CxC;;;;;;OAMG;IACH,OAAO,CAAC,eAAe;IAiBvB;;;;OAIG;IACH,OAAO,CAAC,0BAA0B;YAcpB,UAAU;YAiBV,mBAAmB;IAqKjC;;;;OAIG;IACH,OAAO,CAAC,8BAA8B;YASxB,0BAA0B;YAqF1B,0BAA0B;YAgB1B,iCAAiC;YAgDjC,kCAAkC;IAyBhD,OAAO,CAAC,4BAA4B;IAQpC,OAAO,CAAC,4BAA4B;IAapC,OAAO,CAAC,0BAA0B;IAQlC,OAAO,CAAC,KAAK;IAIb,OAAO,CAAC,0BAA0B;IAOlC,OAAO,CAAC,eAAe;IAWvB,OAAO,CAAC,sBAAsB;IAe9B,OAAO,CAAC,qBAAqB;IAK7B,OAAO,CAAC,uBAAuB;IAS/B,OAAO,CAAC,qBAAqB;IAY7B,OAAO,CAAC,4BAA4B;IAkBpC,OAAO,CAAC,aAAa;YAUP,0BAA0B;YAQ1B,8BAA8B;YA4E9B,cAAc;IAQ5B,OAAO,CAAC,oBAAoB;YASd,QAAQ;CA4FtB","sourcesContent":["import { type ChildProcess, spawn } from \"node:child_process\";\nimport { createHash, randomBytes, randomUUID } from \"node:crypto\";\nimport {\n\tchmodSync,\n\texistsSync,\n\tmkdirSync,\n\treaddirSync,\n\treadFileSync,\n\trenameSync,\n\trmSync,\n\twriteFileSync,\n} from \"node:fs\";\nimport { createServer, type Server, type Socket } from \"node:net\";\nimport { basename, dirname, join, resolve } from \"node:path\";\nimport { Writable } from \"node:stream\";\nimport { getLogger } from \"@ponythewhite/base-context-ai\";\nimport { createCliSubprocessEnv, createCliSubprocessLaunchSpec } from \"../../cli/subprocess-launch.js\";\nimport {\n\tappendRotatingLog,\n\tgetCronJobsPath,\n\tgetDaemonLogPath,\n\tgetDaemonUpdateRestartManifestPath,\n\tgetSessionsDir,\n\tVERSION,\n} from \"../../config.js\";\nimport {\n\ttype AgentFamilyCatalogEntry,\n\ttype AgentSessionMessageAgentSummary,\n\tassertAgentFamilyReach,\n\tassertAgentSessionNameAvailable,\n\tformatAgentSessionNameUnavailable,\n\tsessionNameReservationKey,\n} from \"../../core/agent-messages.js\";\nimport {\n\ttype AgentSessionRuntimeConfig,\n\ttype DurableAgentSessionRuntimeConfig,\n\tdurableAgentSessionRuntimeConfig,\n\tmergeAgentSessionRuntimeConfig,\n} from \"../../core/agent-session-config.js\";\nimport {\n\ttype AgentCronJob,\n\tAgentCronJobStore,\n\tisHeartbeatCronJob,\n\tmigrateLegacyCronJobsToSessionArtifacts,\n\tSESSION_SCHEDULED_JOBS_FILENAME,\n} from \"../../core/cron-jobs.js\";\nimport {\n\tclearOrphanProcessJournal,\n\tkillOrphanProcess,\n\tORPHAN_PROCESS_JOURNAL_ENV,\n\treadActiveOrphanProcesses,\n\tshouldReapOrphanProcess,\n} from \"../../core/orphan-process-journal.js\";\nimport { PromptAdmissionCancelledError, waitForPromptAdmission } from \"../../core/prompt-admission.js\";\nimport { RlmJournalOwner } from \"../../core/rlm-journal-owner.js\";\nimport {\n\tcanEvictWorker,\n\ttype IdleEvictionMinutes,\n\ttype WorkerEvictionSnapshot,\n} from \"../../core/session-action-store.js\";\nimport { canonicalSessionPath, getProcessStartId, SessionAlreadyActiveError } from \"../../core/session-lease.js\";\nimport { getSessionArtifactPathForFile, readSessionInfo, type SessionInfo } from \"../../core/session-manager.js\";\nimport { looksLikeSessionPath } from \"../../core/session-resolver.js\";\nimport { SettingsManager } from \"../../core/settings-manager.js\";\nimport { PRODUCT } from \"../../product-identity.js\";\nimport { isProcessAlive, processIdExists, signalProcessGroupOrProcess } from \"../../utils/child-process.js\";\nimport type { AgentConnectionHeartbeat } from \"../agent-connection/types.js\";\nimport { attachJsonlLineReader, serializeJsonLine } from \"../rpc/jsonl.js\";\nimport type { PrivateFrame } from \"../session-worker/private-framing.js\";\nimport { createActiveSessionId, type DaemonSocketClient } from \"./active-session-state.js\";\nimport {\n\tAgentRoster,\n\ttype AgentRosterEntry,\n\ttype AgentRosterMutation,\n\tpassivatedWorkerRosterEntry,\n\trosterAgentIdForSummary,\n\tsessionSummaryFromRosterEntry,\n\ttype WorkerRosterEntry,\n\tworkerRosterEntryFromSummary,\n} from \"./agent-roster.js\";\nimport {\n\ttype CommandJournalBeginResult,\n\tCommandRecoveryJournal,\n\tcreateCommandIdempotencyKey,\n} from \"./command-recovery-journal.js\";\nimport { CompactAssistantStreamReconstructor, isCompactAssistantDelta } from \"./compact-session-stream.js\";\nimport { DAEMON_CATALOG_ROLE_ENV, DaemonCatalogClient } from \"./daemon-catalog-process.js\";\nimport { deserializeDaemonError, serializeDaemonError } from \"./daemon-errors.js\";\nimport {\n\tCANONICAL_SESSION_OWNERSHIP_COMPATIBILITY,\n\tcollectDaemonClientEnv,\n\tcreateDaemonEventMeta,\n\tDAEMON_COMMAND_COMPATIBILITY,\n\tDAEMON_COMMAND_ENVELOPE_MIN_PROTOCOL_VERSION,\n\tDAEMON_DEFAULT_CLIENT_CAPABILITIES,\n\tDAEMON_DEFAULT_SERVER_CAPABILITIES,\n\tDAEMON_PROTOCOL_INFO,\n\tDAEMON_SCHEMA_ID,\n\tDAEMON_SCHEMA_REVISION,\n\tDAEMON_UPDATE_RESTART_FORMAT_VERSION,\n\ttype DaemonAttachResult,\n\ttype DaemonClientCapability,\n\ttype DaemonClosingReason,\n\ttype DaemonCommand,\n\ttype DaemonOutbound,\n\ttype DaemonPeerTransportTicket,\n\ttype DaemonResponse,\n\ttype DaemonServerCapability,\n\ttype DaemonUpdateRestartManifest,\n\tfailure,\n\tisDaemonCommandEnvelope,\n\tisDaemonMutatingCommand,\n\tisLegacyDaemonInspection,\n\tsalvageDaemonCommandId,\n\tsuccess,\n\tUPDATE_RESTART_DRAIN_COMMANDS,\n} from \"./daemon-protocol.js\";\nimport { DaemonRlmCapacityCoordinator } from \"./daemon-rlm-capacity-coordinator.js\";\nimport { getDaemonRuntimeIdentity } from \"./daemon-runtime-identity.js\";\nimport { matchesSessionIdSuffix } from \"./daemon-session-id.js\";\nimport {\n\tclassifySessionRosterStatus,\n\tisEvictableEmptySessionSummary,\n\tisSessionSummaryBusy,\n\ttype SessionSummary,\n\tsummaryForInactiveSession,\n} from \"./daemon-session-list.js\";\nimport {\n\tacquireDaemonSocketPathLease,\n\tcleanupDaemonSocketPath,\n\ttype DaemonSocketIdentity,\n\ttype DaemonSocketPathLease,\n\tdefaultDaemonSocketDir,\n\tdefaultDaemonSocketPath,\n\tgetDaemonSocketIdentity,\n\tnormalizeSocketPath,\n\tprepareDaemonSocketPath,\n\trestrictDaemonSocketPath,\n} from \"./daemon-socket.js\";\nimport {\n\tacquireDaemonSupervisorOwnership,\n\tisDaemonShutdownAdmissionActive,\n\twaitForDaemonStartupFence,\n} from \"./daemon-supervisor-ownership.js\";\nimport {\n\tDaemonWorkerAuthenticationError,\n\tDaemonWorkerClient,\n\tDaemonWorkerCompatibilityError,\n\tDaemonWorkerProbeTimeoutError,\n} from \"./daemon-worker-client.js\";\nimport {\n\tDAEMON_WORKER_ACTIVE_SESSION_ID_ENV,\n\tDAEMON_WORKER_INSTANCE_ID_ENV,\n\tDAEMON_WORKER_PEER_TRANSPORT_CAPABILITY,\n\tDAEMON_WORKER_RECOVERY_JOURNAL_ENV,\n\tDAEMON_WORKER_ROLE_ENV,\n\tDAEMON_WORKER_ROSTER_CAPABILITY,\n\tDAEMON_WORKER_STARTUP_GATE_COMMIT,\n\tDAEMON_WORKER_STARTUP_GATE_FD_ENV,\n\tDAEMON_WORKER_SUPERVISOR_SOCKET_ENV,\n\tDAEMON_WORKER_TOKEN_ENV,\n\ttype DaemonCreateCommand,\n\ttype DaemonWorkerDescriptor,\n\ttype DaemonWorkerFrameHeader,\n\ttype DaemonWorkerLifecycle,\n\ttype DaemonWorkerRosterOutbound,\n\tdurableDaemonCreateCommand,\n\tdurableDaemonWorkerDescriptor,\n\tisDaemonRlmCapacitySnapshot,\n\tROSTER_HEARTBEAT_INTERVAL_MS,\n\tSESSION_LEASE_OWNER_ID_ENV,\n\tSESSION_LEASES_ENABLED_ENV,\n} from \"./daemon-worker-protocol.js\";\nimport { MutationDrainLatch } from \"./mutation-drain-latch.js\";\nimport {\n\tcreateRlmLedgerRegistrySeedSource,\n\ttype RlmLedgerEdge,\n\tRlmSpawnLedger,\n\trlmLedgerPath,\n\ttombstoneSavedSessionDelete,\n} from \"./rlm-ledger.js\";\nimport { captureSavedSessionPageQuery } from \"./saved-session-page.js\";\nimport { SNAPSHOT_TARGET_CHUNK_BYTES, SnapshotTranscriptCache } from \"./snapshot-transcript-cache.js\";\nimport { WorkerRecoveryJournal } from \"./worker-recovery-journal.js\";\n\ntype DistributiveOmit<T, K extends keyof T> = T extends unknown ? Omit<T, K> : never;\ntype DaemonCommandBody = DistributiveOmit<DaemonCommand, \"id\">;\n\nconst structuredLog = getLogger(\"coding-agent.daemon-supervisor\");\nconst WORKER_CONNECT_TIMEOUT_MS = 30_000;\nconst ROSTER_WATCHDOG_INTERVAL_MS = 15_000;\nconst ROSTER_STALE_AFTER_MS = 3 * ROSTER_HEARTBEAT_INTERVAL_MS;\nconst SUPERVISOR_SERVER_CAPABILITIES: readonly DaemonServerCapability[] = [\n\t...DAEMON_DEFAULT_SERVER_CAPABILITIES,\n\t\"agent_roster\",\n\t\"direct_peer_transport\",\n\t\"rlm_ledger_mutation\",\n];\nconst PEER_TRANSPORT_GRANT_TTL_MS = 10_000;\nconst WORKER_REQUEST_TIMEOUT_MS = 24 * 60 * 60 * 1000;\nconst INPUT_PAUSE_CLEANUP_TIMEOUT_MS = 5_000;\nconst UPDATE_RESTART_MUTATION_DRAIN_TIMEOUT_MS = 80_000;\nconst UPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS = 90_000;\n// The whole pre-commit prepare (drain + worker fencing) must finish inside the\n// caller's 120s prepare_update_restart request timeout, or roll back; otherwise\n// an abandoned prepare leaves the daemon permanently fenced with workers stopped.\nconst UPDATE_RESTART_PREPARE_DEADLINE_MS = 100_000;\nconst WORKER_RETRY_DELAYS_MS = [250, 1000, 5000] as const;\nconst DEFERRED_RECOVERY_RECHECK_MS = 5000;\n// ~2.5 minutes of probing: each round is one 5s defer recheck plus a ~11s three-delay probe pass.\nconst MAX_DEFERRED_RECOVERY_ROUNDS = 10;\nconst STOP_FINALIZATION_RECHECK_MS = 250;\nconst STOP_FINALIZATION_SIGKILL_GRACE_MS = 5000;\nconst STOP_FINALIZATION_RETRY_MS = 5000;\nconst STALE_RECLAIM_WAIT_MS = 10_000;\n// Polling loops probe existence cheaply via kill(0); the ps-backed zombie and\n// identity checks are throttled so a wedged worker cannot saturate the\n// supervisor event loop with synchronous subprocess spawns.\nconst LIVENESS_IDENTITY_RECHECK_MS = 500;\nconst OWNED_WORKER_DISCONNECT_GRACE_MS = 30_000;\nconst IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS = 5 * 60_000;\nconst IDLE_EVICTION_MIN_SWEEP_INTERVAL_MS = 60_000;\nconst IDLE_EVICTION_DRAIN_TIMEOUT_MS = 5_000;\nconst CHILD_PASSIVATION_PER_WORKER_CAP = 2;\nconst SCHEDULED_WAKE_RETRY_MS = 60_000;\nconst SCHEDULED_WAKE_MAX_TIMEOUT_MS = 2_147_483_647;\nconst SCHEDULED_WAKE_CLIENT_ID = \"scheduled-wake\";\nconst SUPERVISOR_CONFIG_FILE_NAME = \"supervisor-config\";\nconst WORKER_STARTUP_GATE_FD = 3;\n\nconst DAEMON_COMMAND_TYPES: ReadonlySet<string> = new Set([\n\t\"ack_result\",\n\t\"list\",\n\t\"list_agent_peers\",\n\t\"rlm_ledger_mutate\",\n\t\"rlm_capacity\",\n\t\"get_direct_worker_transport\",\n\t\"roster_subscribe\",\n\t\"roster_unsubscribe\",\n\t\"list_saved_sessions\",\n\t\"create\",\n\t\"attach\",\n\t\"reattach\",\n\t\"detach\",\n\t\"complete_owned_session\",\n\t\"promote_owned_session\",\n\t\"kill\",\n\t\"rename\",\n\t\"prompt\",\n\t\"cancel_prompt_admission\",\n\t\"prompt_and_wait\",\n\t\"steer\",\n\t\"follow_up\",\n\t\"restore_next_turn\",\n\t\"restore_actions\",\n\t\"append_custom_message\",\n\t\"resume_queue\",\n\t\"send_message\",\n\t\"send_result\",\n\t\"agent_messages_status\",\n\t\"agent_messages_pause\",\n\t\"agent_messages_resume\",\n\t\"agent_messages_clear\",\n\t\"abort\",\n\t\"start_side_question\",\n\t\"abort_side_question\",\n\t\"execute_bash\",\n\t\"execute_bash_and_wait\",\n\t\"abort_bash\",\n\t\"cancel_rlm_child\",\n\t\"delete_rlm_subagent\",\n\t\"wait_for_idle\",\n\t\"wait_for_headless_completion\",\n\t\"get_session_header\",\n\t\"get_state\",\n\t\"get_connection_state\",\n\t\"get_messages\",\n\t\"get_rlm_children\",\n\t\"get_session_stats\",\n\t\"get_context_tree\",\n\t\"get_commands\",\n\t\"get_resource_snapshot\",\n\t\"replace_acp_mcp_servers\",\n\t\"get_model_catalog\",\n\t\"get_available_models\",\n\t\"get_queue\",\n\t\"mutate_queued_message\",\n\t\"clear_queue\",\n\t\"abort_and_clear_queue\",\n\t\"acquire_session_input_pause\",\n\t\"release_session_input_pause\",\n\t\"cron_list\",\n\t\"heartbeats_list\",\n\t\"heartbeat_manage\",\n\t\"cron_add\",\n\t\"cron_cancel\",\n\t\"cron_resume\",\n\t\"heartbeat_get\",\n\t\"heartbeat_set\",\n\t\"heartbeat_update\",\n\t\"set_model\",\n\t\"cycle_model\",\n\t\"set_scoped_models\",\n\t\"set_thinking_level\",\n\t\"cycle_thinking_level\",\n\t\"set_service_tier\",\n\t\"set_transport\",\n\t\"set_steering_mode\",\n\t\"set_follow_up_mode\",\n\t\"set_auto_compaction\",\n\t\"set_auto_retry\",\n\t\"compact\",\n\t\"refine\",\n\t\"abort_compaction\",\n\t\"abort_branch_summary\",\n\t\"abort_retry\",\n\t\"reload\",\n\t\"new_session\",\n\t\"switch_session\",\n\t\"fork\",\n\t\"navigate_tree\",\n\t\"import_jsonl\",\n\t\"export_html\",\n\t\"export_jsonl\",\n\t\"set_session_name\",\n\t\"get_rlm_max_depth_status\",\n\t\"set_rlm_max_depth\",\n\t\"get_rlm_max_subagents_status\",\n\t\"set_rlm_max_subagents\",\n\t\"rename_saved_session\",\n\t\"delete_saved_session\",\n\t\"get_session_context\",\n\t\"get_session_tree\",\n\t\"get_user_messages_for_forking\",\n\t\"get_last_assistant_text\",\n\t\"get_system_prompt\",\n\t\"get_tool_definition\",\n\t\"set_session_entry_label\",\n\t\"extension_ui_response\",\n\t\"prepare_update_restart\",\n\t\"retry_worker\",\n\t\"restart\",\n\t\"shutdown\",\n]);\n\ninterface ResidentWorker {\n\tdescriptor: DaemonWorkerDescriptor;\n\tdescriptorPath: string;\n\tclient?: DaemonWorkerClient;\n\tcompatibilityError?: DaemonWorkerCompatibilityError;\n\theartbeatSnapshot?: AgentConnectionHeartbeat[];\n\theartbeatSnapshotStale?: boolean;\n\tsummaries: Map<string, SessionSummary>;\n\tsnapshotCache: Map<string, DaemonAttachResult>;\n\ttranscriptCaches: Map<string, SnapshotTranscriptCache>;\n\tsnapshotGenerations: Map<string, Map<string, SnapshotTranscriptGeneration>>;\n\tsnapshotLoads: Map<string, Promise<DaemonAttachResult>>;\n\trecovery?: Promise<void>;\n\tdeferredRecovery?: Promise<void>;\n\tintentionalStop: boolean;\n\tstopRevision: number;\n\tlaunchEnv?: Record<string, string>;\n\ttransientCreateCommand?: DaemonCreateCommand;\n\tstopFinalization?: Promise<void>;\n\townerCleanupTimer?: ReturnType<typeof setTimeout>;\n\tpromotedOwnerClientId?: string;\n\tupdateRestartPrepareClient?: DaemonWorkerClient;\n\tlastFrameAt?: number;\n\trosterStale?: boolean;\n\t/** worker_auth advertised peer-transport support; absent on workers from older builds. */\n\tpeerTransportCapable?: boolean;\n\t/** In-flight replacement connection during authentication; an allowed frame source alongside client. */\n\tpendingClient?: DaemonWorkerClient;\n\t/** Consecutive defer->probe rounds against a live-but-silent worker; bounded by MAX_DEFERRED_RECOVERY_ROUNDS. */\n\tdeferredRecoveryRounds?: number;\n\t/** Bumped per applied roster frame; a summaries pull that straddles a frame must not gap-fill. */\n\trosterEpoch?: number;\n\trosterApplyChain?: Promise<void>;\n\trosterRepairPull?: Promise<void>;\n}\n\ninterface SnapshotDuplicateValidation {\n\tpromise: Promise<void>;\n\tresolve: () => void;\n\treject: (error: Error) => void;\n}\n\ninterface SnapshotTranscriptGeneration {\n\ttranscript: SnapshotTranscriptCache;\n\tresult: DaemonAttachResult;\n\tbegin?: Buffer;\n\tend?: Buffer;\n\tincoming: boolean;\n\tretired: boolean;\n\tduplicateChunkIndex?: number;\n\tduplicateResult?: DaemonAttachResult;\n\tvalidation?: SnapshotDuplicateValidation;\n}\n\ninterface DaemonSupervisorOptions {\n\tsocketPath?: string;\n\tdefaultSessionConfig: AgentSessionRuntimeConfig;\n\tdescriptorDir?: string;\n}\n\ninterface PersistedSupervisorConfig {\n\tversion: 1;\n\tsocketPath: string;\n\tdefaultSessionConfig: DurableAgentSessionRuntimeConfig;\n}\n\ninterface WorkerMatch {\n\tworker: ResidentWorker;\n\tsummary: SessionSummary;\n}\n\ninterface WorkerAttachData {\n\tresult: DaemonAttachResult;\n\tworker: ResidentWorker;\n\ttranscript?: SnapshotTranscriptCache;\n\treleaseTranscript?: () => void;\n}\n\ninterface SupervisorPromptAdmission {\n\tclient: DaemonSocketClient;\n\tactiveSessionId: string;\n\tpublicAdmissionId: string;\n\tworkerAdmissionId: string;\n\tstatus: \"waiting\" | \"owned\" | \"cancelled\";\n\tcontroller: AbortController;\n\tworker?: ResidentWorker;\n\tworkerActiveSessionId?: string;\n}\n\ninterface SupervisorSessionInputPause {\n\towner: DaemonSocketClient;\n\tworker: ResidentWorker;\n\tactiveSessionId: string;\n\trequestedActiveSessionId: string;\n\tleaseKey: string;\n\tpauseId: string;\n\treleaseTask?: Promise<DaemonResponse>;\n}\n\nfunction throwIfAdmissionCancelled(admission: SupervisorPromptAdmission | undefined): void {\n\tif (admission?.status === \"cancelled\") throw new PromptAdmissionCancelledError();\n}\n\nclass SupervisorRecoveryCancelledError extends Error {\n\treadonly code = \"supervisor_recovery_cancelled\" as const;\n}\n\nclass SnapshotLoadInvalidatedError extends Error {}\n\nclass WorkerStopTimeoutError extends Error {}\n\nfunction isSupervisorGenerationStale(error: unknown): boolean {\n\treturn (\n\t\ttypeof error === \"object\" &&\n\t\terror !== null &&\n\t\t\"code\" in error &&\n\t\t(error as { code?: unknown }).code === \"supervisor_generation_stale\"\n\t);\n}\n\nfunction isSupervisorRecoveryCancelled(error: unknown): boolean {\n\treturn isSupervisorShutdownAdmissionCancelled(error) || isSupervisorGenerationStale(error);\n}\n\n// Workers can be registered mid-tree (a resumed subagent transcript), so descent is membership at\n// any step of the parent walk, never a comparison against the ultimate root alone.\nfunction rosterFamilyDescendsFrom(\n\tedges: readonly RlmLedgerEdge[],\n): (path: string, roots: ReadonlySet<string>) => boolean {\n\tconst parentByChild = new Map(\n\t\tedges.map((edge) => [canonicalSessionPath(edge.child), canonicalSessionPath(edge.parent)]),\n\t);\n\treturn (path, roots) => {\n\t\tconst visited = new Set<string>();\n\t\tlet current = path;\n\t\twhile (!visited.has(current)) {\n\t\t\tif (roots.has(current)) return true;\n\t\t\tvisited.add(current);\n\t\t\tconst parent = parentByChild.get(current);\n\t\t\tif (parent === undefined) return false;\n\t\t\tcurrent = parent;\n\t\t}\n\t\treturn false;\n\t};\n}\n\nfunction isDaemonWorkerProbeTimeout(error: unknown): boolean {\n\treturn error instanceof DaemonWorkerProbeTimeoutError;\n}\n\nfunction isSupervisorShutdownAdmissionCancelled(error: unknown): boolean {\n\treturn (\n\t\terror instanceof SupervisorRecoveryCancelledError ||\n\t\t(typeof error === \"object\" &&\n\t\t\terror !== null &&\n\t\t\t\"code\" in error &&\n\t\t\t(error as { code?: unknown }).code === \"supervisor_recovery_cancelled\")\n\t);\n}\n\nfunction delay(ms: number): Promise<void> {\n\treturn new Promise((resolveDelay) => setTimeout(resolveDelay, ms));\n}\n\nfunction unrefDelay(ms: number): Promise<void> {\n\treturn new Promise((resolveDelay) => setTimeout(resolveDelay, ms).unref());\n}\n\nfunction commitWorkerStartupGate(gate: Writable): Promise<void> {\n\treturn new Promise((resolveCommit, rejectCommit) => {\n\t\tlet settled = false;\n\t\tconst finish = (error?: Error | null) => {\n\t\t\tif (settled) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tsettled = true;\n\t\t\tif (error) {\n\t\t\t\trejectCommit(error);\n\t\t\t} else {\n\t\t\t\tresolveCommit();\n\t\t\t}\n\t\t};\n\t\tconst onError = (error: Error) => finish(error);\n\t\tgate.on(\"error\", onError);\n\t\tgate.once(\"close\", () => gate.off(\"error\", onError));\n\t\tgate.end(DAEMON_WORKER_STARTUP_GATE_COMMIT, (error?: Error | null) => finish(error));\n\t});\n}\n\nfunction withoutCommandId(command: DaemonCommand): DaemonCommandBody {\n\tconst { id: _id, ...body } = command;\n\treturn body as DaemonCommandBody;\n}\n\nfunction withoutSupervisorCreateFields(command: DaemonCreateCommand): DaemonCreateCommand {\n\tconst { launchEnv: _launchEnv, lifecycle: _lifecycle, ...workerCommand } = command;\n\treturn workerCommand;\n}\n\nfunction responseWithId(response: DaemonResponse, id: string | undefined): DaemonResponse {\n\treturn { ...response, id };\n}\n\nfunction isSessionSummary(value: unknown): value is SessionSummary {\n\tif (!value || typeof value !== \"object\") {\n\t\treturn false;\n\t}\n\tconst candidate = value as { id?: unknown; sessionId?: unknown; cwd?: unknown };\n\treturn (\n\t\ttypeof candidate.id === \"string\" && typeof candidate.sessionId === \"string\" && typeof candidate.cwd === \"string\"\n\t);\n}\n\nfunction isDaemonWorkerDescriptor(value: unknown, socketPath: string): value is DaemonWorkerDescriptor {\n\tif (!value || typeof value !== \"object\") {\n\t\treturn false;\n\t}\n\tconst descriptor = value as Partial<DaemonWorkerDescriptor>;\n\treturn (\n\t\t(descriptor.version === 1 || descriptor.version === 2) &&\n\t\ttypeof descriptor.supervisorSocketPath === \"string\" &&\n\t\tnormalizeSocketPath(descriptor.supervisorSocketPath) === socketPath &&\n\t\ttypeof descriptor.workerId === \"string\" &&\n\t\tNumber.isInteger(descriptor.pid) &&\n\t\t(descriptor.pid ?? 0) > 0 &&\n\t\t(descriptor.processStartId === undefined || typeof descriptor.processStartId === \"string\") &&\n\t\t(descriptor.ownerClientId === undefined || typeof descriptor.ownerClientId === \"string\") &&\n\t\ttypeof descriptor.socketPath === \"string\" &&\n\t\ttypeof descriptor.authenticationToken === \"string\" &&\n\t\t(descriptor.workerInstanceId === undefined || typeof descriptor.workerInstanceId === \"string\") &&\n\t\ttypeof descriptor.rootActiveSessionId === \"string\" &&\n\t\ttypeof descriptor.createdAt === \"string\" &&\n\t\ttypeof descriptor.updatedAt === \"string\" &&\n\t\tNumber.isInteger(descriptor.consecutiveFailures) &&\n\t\tdescriptor.createCommand !== undefined &&\n\t\ttypeof descriptor.createCommand === \"object\" &&\n\t\tdescriptor.createCommand.type === \"create\"\n\t);\n}\n\nclass PreRosterWorkerError extends Error {}\n\nfunction workerAuthAdvertisesRoster(data: unknown): boolean {\n\tif (typeof data !== \"object\" || data === null) return false;\n\tconst capabilities = (data as { capabilities?: unknown }).capabilities;\n\treturn Array.isArray(capabilities) && capabilities.includes(DAEMON_WORKER_ROSTER_CAPABILITY);\n}\n\nfunction workerAuthAdvertisesPeerTransport(data: unknown): boolean {\n\tif (typeof data !== \"object\" || data === null) return false;\n\tconst capabilities = (data as { capabilities?: unknown }).capabilities;\n\treturn Array.isArray(capabilities) && capabilities.includes(DAEMON_WORKER_PEER_TRANSPORT_CAPABILITY);\n}\n\nfunction sessionSummariesFromResponse(response: DaemonResponse): SessionSummary[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\" || !(\"sessions\" in response.data)) {\n\t\tthrow new Error(\"Session worker returned an invalid list response\");\n\t}\n\tconst sessions = (response.data as { sessions: unknown }).sessions;\n\tif (!Array.isArray(sessions) || !sessions.every(isSessionSummary)) {\n\t\tthrow new Error(\"Session worker returned an invalid list response\");\n\t}\n\treturn sessions;\n}\n\nfunction attachResultFromResponse(response: DaemonResponse): DaemonAttachResult {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\tthrow new Error(response.success ? \"Session worker returned an invalid attach response\" : response.error);\n\t}\n\tconst candidate = response.data as Partial<DaemonAttachResult>;\n\tif (typeof candidate.activeSessionId !== \"string\" || !candidate.snapshot || !candidate.client) {\n\t\tthrow new Error(\"Session worker returned an invalid attach response\");\n\t}\n\treturn candidate as DaemonAttachResult;\n}\n\nfunction cronJobsFromResponse(response: DaemonResponse): AgentCronJob[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\treturn [];\n\t}\n\tconst jobs = (response.data as { jobs?: unknown }).jobs;\n\treturn Array.isArray(jobs) ? (jobs as AgentCronJob[]) : [];\n}\n\nfunction heartbeatsFromResponse(response: DaemonResponse): AgentConnectionHeartbeat[] {\n\tif (!response.success || !response.data || typeof response.data !== \"object\") {\n\t\treturn [];\n\t}\n\tconst heartbeats = (response.data as { heartbeats?: unknown }).heartbeats;\n\treturn Array.isArray(heartbeats) ? (heartbeats as AgentConnectionHeartbeat[]) : [];\n}\n\nfunction sortCronJobs(jobs: AgentCronJob[]): AgentCronJob[] {\n\treturn jobs.sort((left, right) => {\n\t\tif (left.nextRunAt === right.nextRunAt) {\n\t\t\treturn 0;\n\t\t}\n\t\tif (left.nextRunAt === undefined) {\n\t\t\treturn 1;\n\t\t}\n\t\tif (right.nextRunAt === undefined) {\n\t\t\treturn -1;\n\t\t}\n\t\treturn Date.parse(left.nextRunAt) - Date.parse(right.nextRunAt);\n\t});\n}\n\nfunction descriptorKey(socketPath: string): string {\n\treturn createHash(\"sha256\").update(normalizeSocketPath(socketPath)).digest(\"hex\").slice(0, 12);\n}\n\nfunction defaultWorkerDescriptorDir(agentDir: string, socketPath: string): string {\n\treturn join(agentDir, \"daemon-workers\", descriptorKey(socketPath));\n}\n\nexport function idleEvictionSweepIntervalMs(idleEvictionMinutes: IdleEvictionMinutes): number {\n\tif (idleEvictionMinutes === \"off\") return IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS;\n\treturn Math.max(\n\t\tIDLE_EVICTION_MIN_SWEEP_INTERVAL_MS,\n\t\tMath.min(IDLE_EVICTION_MAX_SWEEP_INTERVAL_MS, (idleEvictionMinutes * 60_000) / 3),\n\t);\n}\n\nfunction workerSocketPath(supervisorSocketPath: string, workerId: string): string {\n\tconst key = descriptorKey(supervisorSocketPath);\n\tif (process.platform === \"win32\") {\n\t\treturn `\\\\\\\\.\\\\pipe\\\\${PRODUCT.command}-worker-${key}-${workerId.slice(0, 12)}`;\n\t}\n\treturn join(defaultDaemonSocketDir(), `worker-${key}-${workerId.slice(0, 12)}.sock`);\n}\n\nfunction isFinalizedTranscriptEvent(eventType: string | undefined): boolean {\n\treturn (\n\t\teventType === \"message_end\" ||\n\t\teventType === \"turn_end\" ||\n\t\teventType === \"compaction_end\" ||\n\t\teventType === \"bash_end\"\n\t);\n}\n\nfunction normalizeCapabilities(\n\tcapabilities: readonly DaemonClientCapability[] | undefined,\n\tsupportsExtensionUi: boolean | undefined,\n): Set<DaemonClientCapability> {\n\tconst normalized = new Set(capabilities ?? DAEMON_DEFAULT_CLIENT_CAPABILITIES);\n\tif (supportsExtensionUi) {\n\t\tnormalized.add(\"extension_ui\");\n\t}\n\treturn normalized;\n}\n\nexport async function runDaemonSupervisorMode(options: DaemonSupervisorOptions): Promise<never> {\n\tconst socketPath = normalizeSocketPath(options.socketPath ?? defaultDaemonSocketPath());\n\tconst supervisor = new DaemonSupervisor(socketPath, options);\n\tawait supervisor.start();\n\treturn new Promise(() => {});\n}\n\nexport class DaemonSupervisor {\n\tprivate server?: Server;\n\tprivate readonly ready: Promise<void>;\n\tprivate markReady: () => void = () => {};\n\tprivate rejectReady: (error: Error) => void = () => {};\n\tprivate ownsSocketPath = false;\n\tprivate socketIdentity?: DaemonSocketIdentity;\n\tprivate socketLease?: DaemonSocketPathLease;\n\tprivate socketLeaseCompromise?: Error;\n\tprivate ownership?: Awaited<ReturnType<typeof acquireDaemonSupervisorOwnership>>;\n\tprivate cleanupPromise?: Promise<void>;\n\tprivate shuttingDown = false;\n\tprivate startupComplete = false;\n\tprivate updateRestartPhase?: \"draining\" | \"fencing\" | \"prepared\";\n\tprivate readonly mutationDrain = new MutationDrainLatch();\n\tprivate readonly clients = new Set<DaemonSocketClient>();\n\tprivate readonly connectionIds = new WeakMap<DaemonSocketClient, string>();\n\tprivate readonly sessionInputPauseEpochs = new WeakMap<DaemonSocketClient, number>();\n\tprivate readonly detachingInputPauseSessions = new WeakMap<DaemonSocketClient, Set<string>>();\n\tprivate readonly protocolClientIds = new WeakMap<DaemonSocketClient, string>();\n\tprivate readonly workers = new Map<string, ResidentWorker>();\n\tprivate workerStopCounts?: Map<ResidentWorker, number>;\n\tprivate readonly openingWorkers = new Map<string, Promise<ResidentWorker>>();\n\t/** Public admission ids are scoped to the socket that registered them. */\n\tprivate readonly promptAdmissions = new Map<DaemonSocketClient, Map<string, SupervisorPromptAdmission>>();\n\tprivate readonly sessionInputPauses = new Map<string, SupervisorSessionInputPause>();\n\tprivate readonly signalCleanupHandlers: Array<() => void> = [];\n\tprivate readonly descriptorDir: string;\n\tprivate readonly generation = randomUUID();\n\tprivate readonly supervisorConfigPath: string;\n\tprivate readonly defaultSessionConfig: AgentSessionRuntimeConfig;\n\tprivate readonly snapshotCacheRoot: string;\n\tprivate commandJournal!: CommandRecoveryJournal;\n\tprivate readonly streamReconstructor = new CompactAssistantStreamReconstructor();\n\tprivate readonly compactCatchupInProgress = new Set<string>();\n\tprivate readonly pendingSessionNames = new Set<string>();\n\tprivate readonly catalog: DaemonCatalogClient;\n\tprivate readonly settingsManager: SettingsManager;\n\tprivate readonly rlmCapacity: DaemonRlmCapacityCoordinator;\n\tprivate readonly pendingRlmCapacity = new Set<ResidentWorker>();\n\tprivate readonly releasedDuringCapacityRecovery = new Map<ResidentWorker, Set<string>>();\n\tprivate rosterStore?: AgentRoster;\n\tprivate readonly pendingRosterChanged = new Set<string>();\n\tprivate readonly pendingRosterRemoved = new Set<string>();\n\t/** Ids declared to subscribers: gates removals to once and keeps owned-only row ids private. */\n\tprivate readonly publishedRosterIds = new Set<string>();\n\tprivate rosterPushScheduled = false;\n\tprivate rosterWatchdogTimer?: ReturnType<typeof setInterval>;\n\tprivate rlmSpawnLedgerInstance?: RlmSpawnLedger;\n\tprivate rlmJournalOwner?: RlmJournalOwner;\n\tprivate rlmJournalClose?: Promise<void>;\n\tprivate idleEvictionTimer?: ReturnType<typeof setTimeout>;\n\tprivate idleEvictionSweep?: Promise<void>;\n\tprivate idleEvictionFence?: Promise<void>;\n\tprivate scheduledWakeTimer?: ReturnType<typeof setTimeout>;\n\tprivate scheduledWakeRecompute?: Promise<void>;\n\tprivate scheduledWakeRecomputeQueued = false;\n\tprivate readonly scheduledWakeFailures = new Map<string, number>();\n\n\tconstructor(\n\t\tprivate readonly socketPath: string,\n\t\toptions: DaemonSupervisorOptions,\n\t) {\n\t\tthis.ready = new Promise<void>((resolveReady, rejectReady) => {\n\t\t\tthis.markReady = resolveReady;\n\t\t\tthis.rejectReady = rejectReady;\n\t\t});\n\t\tvoid this.ready.catch(() => undefined);\n\t\tconst agentDir = options.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) {\n\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t}\n\t\tthis.descriptorDir = options.descriptorDir ?? defaultWorkerDescriptorDir(agentDir, socketPath);\n\t\tthis.supervisorConfigPath = join(this.descriptorDir, SUPERVISOR_CONFIG_FILE_NAME);\n\t\tthis.defaultSessionConfig = mergeAgentSessionRuntimeConfig(\n\t\t\toptions.defaultSessionConfig,\n\t\t\tthis.loadPersistedSupervisorConfig(),\n\t\t);\n\t\tthis.snapshotCacheRoot = join(this.descriptorDir, \"snapshot-cache\", this.generation);\n\t\tthis.catalog = new DaemonCatalogClient((message) => this.log(message));\n\t\tthis.settingsManager = SettingsManager.create(process.cwd(), this.defaultSessionConfig.agentDir ?? agentDir);\n\t\tthis.rlmCapacity = new DaemonRlmCapacityCoordinator(this.settingsManager);\n\t}\n\n\tasync start(): Promise<void> {\n\t\ttry {\n\t\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\t\tif (!agentDir) {\n\t\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t\t}\n\t\t\tthis.socketLease = await acquireDaemonSocketPathLease(this.socketPath);\n\t\t\tthis.socketLease?.onCompromised((error) => this.handleSocketLeaseCompromised(error));\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait waitForDaemonStartupFence(this.socketPath);\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tconst journalScope = this.rlmJournalScope();\n\t\t\tthis.ownership = await acquireDaemonSupervisorOwnership({\n\t\t\t\tsocketPath: this.socketPath,\n\t\t\t\tdescriptorDir: this.descriptorDir,\n\t\t\t\tagentDir,\n\t\t\t\tjournalPath: journalScope.journalPath,\n\t\t\t\tgeneration: this.generation,\n\t\t\t\tappVersion: VERSION,\n\t\t\t});\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.rlmJournalOwner = await RlmJournalOwner.open(journalScope);\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait prepareDaemonSocketPath(this.socketPath, this.socketLease);\n\n\t\t\tmkdirSync(this.descriptorDir, { recursive: true, mode: 0o700 });\n\t\t\tchmodSync(this.descriptorDir, 0o700);\n\t\t\tthis.persistSupervisorConfig();\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t\tmkdirSync(this.snapshotCacheRoot, { recursive: true, mode: 0o700 });\n\t\t\tthis.commandJournal = new CommandRecoveryJournal(join(this.descriptorDir, \"command-journal.jsonl\"));\n\t\t\tthis.loadWorkerDescriptors();\n\t\t\tconst workersToAdopt = [...this.workers.values()];\n\t\t\tfor (const worker of workersToAdopt) this.pendingRlmCapacity.add(worker);\n\n\t\t\tthis.server = createServer((socket) => this.handleConnection(socket));\n\t\t\tawait this.listen();\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.socketIdentity = getDaemonSocketIdentity(this.socketPath);\n\t\t\tif (process.platform !== \"win32\" && !this.socketIdentity) {\n\t\t\t\tthrow new Error(`Could not capture daemon socket identity: ${this.socketPath}`);\n\t\t\t}\n\t\t\tthis.ownsSocketPath = true;\n\t\t\trestrictDaemonSocketPath(this.socketPath);\n\n\t\t\tthis.registerSignalHandlers();\n\t\t\tconst ownedSessionFiles = new Set(\n\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t.flatMap((worker) => [worker.descriptor.sessionFile, worker.descriptor.createCommand.sessionPath])\n\t\t\t\t\t.filter((path): path is string => typeof path === \"string\")\n\t\t\t\t\t.map((path) => resolve(path)),\n\t\t\t);\n\t\t\tconst migratedJobs = migrateLegacyCronJobsToSessionArtifacts(getCronJobsPath(agentDir), {\n\t\t\t\tisSessionOwned: (job) => ownedSessionFiles.has(resolve(job.sessionFile)),\n\t\t\t});\n\t\t\tif (migratedJobs > 0) {\n\t\t\t\tthis.log(`Migrated ${migratedJobs} scheduled jobs into session artifacts`);\n\t\t\t}\n\t\t\tawait this.catalog.start().catch((error) => this.log(`Could not start daemon catalog: ${String(error)}`));\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait this.seedRosterLedger();\n\t\t\tlet adoptionFailure: unknown;\n\t\t\tlet adoptionFailed = false;\n\t\t\tawait Promise.all(\n\t\t\t\tworkersToAdopt.map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.adoptOrRecoverWorker(worker, this.ready);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tif (!adoptionFailed) {\n\t\t\t\t\t\t\tadoptionFailed = true;\n\t\t\t\t\t\t\tadoptionFailure = error;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t\tif (adoptionFailed) {\n\t\t\t\tthrow adoptionFailure;\n\t\t\t}\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tthis.scheduleOwnedWorkerCleanup(worker);\n\t\t\t}\n\t\t\tthis.scheduleIdleEvictionSweep();\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\tthis.rosterWatchdogTimer = setInterval(() => this.sweepRosterStaleness(), ROSTER_WATCHDOG_INTERVAL_MS);\n\t\t\tthis.rosterWatchdogTimer.unref();\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tawait this.ownership.updatePhase(\"owner\");\n\t\t\tthis.assertSocketLeaseHeld();\n\t\t\tthis.startupComplete = true;\n\t\t\tthis.log(`${PRODUCT.name} daemon supervisor ${this.generation} listening on ${this.socketPath}`);\n\t\t\tthis.markReady();\n\t\t} catch (error) {\n\t\t\tconst startupError = error instanceof Error ? error : new Error(String(error));\n\t\t\tthis.log(`Daemon supervisor startup failed: ${startupError.stack ?? startupError.message}`);\n\t\t\tawait this.cleanupSupervisorResources();\n\t\t\tthis.rejectReady(startupError);\n\t\t\tthrow startupError;\n\t\t}\n\t}\n\n\tprivate listen(): Promise<void> {\n\t\treturn new Promise<void>((resolveListen, rejectListen) => {\n\t\t\tconst onError = (error: Error) => {\n\t\t\t\tthis.server?.off(\"listening\", onListening);\n\t\t\t\trejectListen(error);\n\t\t\t};\n\t\t\tconst onListening = () => {\n\t\t\t\tthis.server?.off(\"error\", onError);\n\t\t\t\tresolveListen();\n\t\t\t};\n\t\t\tthis.server?.once(\"error\", onError);\n\t\t\tthis.server?.once(\"listening\", onListening);\n\t\t\tthis.server?.listen(this.socketPath);\n\t\t});\n\t}\n\n\tprivate log(message: string): void {\n\t\tconsole.error(message);\n\t\tstructuredLog.warn(message, { socketPath: this.socketPath });\n\t\tappendRotatingLog(getDaemonLogPath(this.socketPath), `[${new Date().toISOString()}] supervisor: ${message}`);\n\t}\n\n\tprivate clearIdleEvictionTimer(): void {\n\t\tif (!this.idleEvictionTimer) return;\n\t\tclearTimeout(this.idleEvictionTimer);\n\t\tthis.idleEvictionTimer = undefined;\n\t}\n\n\tprivate clearRosterWatchdogTimer(): void {\n\t\tif (!this.rosterWatchdogTimer) return;\n\t\tclearInterval(this.rosterWatchdogTimer);\n\t\tthis.rosterWatchdogTimer = undefined;\n\t}\n\n\tprivate clearScheduledWakeTimer(): void {\n\t\tif (!this.scheduledWakeTimer) return;\n\t\tclearTimeout(this.scheduledWakeTimer);\n\t\tthis.scheduledWakeTimer = undefined;\n\t}\n\n\t// The supervisor only wakes non-resident trees; firing and delivery stay worker-owned.\n\tprivate scheduleScheduledSessionWakeRecompute(): void {\n\t\tif (this.shuttingDown) return;\n\t\tif (this.scheduledWakeRecompute) {\n\t\t\tthis.scheduledWakeRecomputeQueued = true;\n\t\t\treturn;\n\t\t}\n\t\tthis.scheduledWakeRecompute = this.recomputeScheduledSessionWake()\n\t\t\t.catch((error) => this.log(`Scheduled-session wake recompute failed: ${String(error)}`))\n\t\t\t.finally(() => {\n\t\t\t\tthis.scheduledWakeRecompute = undefined;\n\t\t\t\tif (this.scheduledWakeRecomputeQueued) {\n\t\t\t\t\tthis.scheduledWakeRecomputeQueued = false;\n\t\t\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\t\t}\n\t\t\t});\n\t}\n\n\t/** Durable truth: the ledger family (fork headers stripped) plus each session's scheduled-jobs artifact. */\n\tprivate async collectPassiveScheduledJobs(\n\t\tincludeInactive = false,\n\t): Promise<Array<{ rootSessionFile: string; job: AgentCronJob; info: SessionInfo }>> {\n\t\tconst pendingCancelRoots = new Set<string>();\n\t\tfor (const intent of this.collectEphemeralCancelIntents()) {\n\t\t\tconst context = this.workerSessionArtifactContext(intent);\n\t\t\tif (!context || !intent.descriptor.rootSessionId) continue;\n\t\t\ttry {\n\t\t\t\tawait this.cancelScheduledJobsForSessionTree(intent.descriptor.rootSessionId, context.sessionFile);\n\t\t\t\tthis.deleteWorkerDescriptor(intent);\n\t\t\t} catch {\n\t\t\t\t// Still owned until the cancel lands; the tree stays excluded below.\n\t\t\t\tpendingCancelRoots.add(canonicalSessionPath(context.sessionFile));\n\t\t\t}\n\t\t}\n\t\tconst infos = await this.rlmSpawnLedger().family();\n\t\tconst infoByPath = new Map(infos.map((info) => [canonicalSessionPath(info.path), info] as const));\n\t\tconst storeBySessionId = new Map<string, AgentCronJobStore>();\n\t\tconst infoBySessionId = new Map<string, SessionInfo>();\n\t\tfor (const info of infos) {\n\t\t\tif (info.state !== undefined && info.state.status !== \"active\") continue;\n\t\t\tconst artifactDir = getSessionArtifactPathForFile(resolve(info.path), info.id);\n\t\t\tif (!existsSync(join(artifactDir, SESSION_SCHEDULED_JOBS_FILENAME))) continue;\n\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\tstore.registerSessionArtifact(info.id, artifactDir);\n\t\t\tstoreBySessionId.set(info.id, store);\n\t\t\tinfoBySessionId.set(info.id, info);\n\t\t}\n\t\tif (infoBySessionId.size === 0) return [];\n\t\tconst uncoveredRootFor = (info: SessionInfo): string | undefined => {\n\t\t\tlet current = info;\n\t\t\tconst visited = new Set([canonicalSessionPath(current.path)]);\n\t\t\twhile (true) {\n\t\t\t\ttry {\n\t\t\t\t\tif (this.findWorkerBySessionFile(current.path)) return undefined;\n\t\t\t\t} catch {\n\t\t\t\t\treturn undefined;\n\t\t\t\t}\n\t\t\t\tif (!current.parentSessionPath) break;\n\t\t\t\tconst parent = infoByPath.get(canonicalSessionPath(current.parentSessionPath));\n\t\t\t\tif (!parent || visited.has(canonicalSessionPath(parent.path))) break;\n\t\t\t\tvisited.add(canonicalSessionPath(parent.path));\n\t\t\t\tcurrent = parent;\n\t\t\t}\n\t\t\treturn current.path;\n\t\t};\n\t\tconst results: Array<{ rootSessionFile: string; job: AgentCronJob; info: SessionInfo }> = [];\n\t\tfor (const [artifactSessionId, store] of storeBySessionId) {\n\t\t\tlet jobs: AgentCronJob[];\n\t\t\ttry {\n\t\t\t\tjobs = store.list();\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Skipping unreadable scheduled jobs for session ${artifactSessionId}: ${String(error)}`);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tfor (const job of jobs) {\n\t\t\t\tif (!includeInactive && job.status !== \"active\" && job.status !== \"paused\") continue;\n\t\t\t\tconst info = infoBySessionId.get(job.sessionId);\n\t\t\t\tif (!info) continue;\n\t\t\t\tconst rootSessionFile = uncoveredRootFor(info);\n\t\t\t\tif (rootSessionFile === undefined) continue;\n\t\t\t\tif (pendingCancelRoots.has(canonicalSessionPath(rootSessionFile))) continue;\n\t\t\t\tresults.push({ rootSessionFile, job, info });\n\t\t\t}\n\t\t}\n\t\treturn results;\n\t}\n\n\tprivate async recomputeScheduledSessionWake(): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst candidates = await this.collectPassiveScheduledJobs();\n\t\tthis.clearScheduledWakeTimer();\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst now = Date.now();\n\t\tconst wakeTimes: number[] = [];\n\t\tconst candidateRoots = new Set(candidates.map(({ rootSessionFile }) => canonicalSessionPath(rootSessionFile)));\n\t\tfor (const root of [...this.scheduledWakeFailures.keys()]) {\n\t\t\tif (!candidateRoots.has(root)) this.scheduledWakeFailures.delete(root);\n\t\t}\n\t\tfor (const { rootSessionFile, job } of candidates) {\n\t\t\tif (job.status !== \"active\" || job.nextRunAt === undefined) continue;\n\t\t\tconst runAt = Date.parse(job.nextRunAt);\n\t\t\tif (!Number.isFinite(runAt)) continue;\n\t\t\t// The failure floor keeps an overdue job off a hot retry loop.\n\t\t\tconst failedAt = this.scheduledWakeFailures.get(canonicalSessionPath(rootSessionFile));\n\t\t\twakeTimes.push(failedAt !== undefined ? Math.max(runAt, failedAt + SCHEDULED_WAKE_RETRY_MS) : runAt);\n\t\t}\n\t\tif (wakeTimes.length === 0) return;\n\t\tconst delay = Math.min(Math.max(0, Math.min(...wakeTimes) - now), SCHEDULED_WAKE_MAX_TIMEOUT_MS);\n\t\tthis.scheduledWakeTimer = setTimeout(() => {\n\t\t\tthis.scheduledWakeTimer = undefined;\n\t\t\tvoid this.wakeDueScheduledSessions().catch((error) =>\n\t\t\t\tthis.log(`Scheduled-session wake failed: ${String(error)}`),\n\t\t\t);\n\t\t}, delay);\n\t\tthis.scheduledWakeTimer.unref();\n\t}\n\n\tprivate async wakeDueScheduledSessions(now = Date.now()): Promise<void> {\n\t\t// Disarmed during update-restart preparation; the phase transition or next boot re-arms once.\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) {\n\t\t\tthis.clearScheduledWakeTimer();\n\t\t\treturn;\n\t\t}\n\t\ttry {\n\t\t\tconst due = new Map<string, string>();\n\t\t\tfor (const { rootSessionFile, job } of await this.collectPassiveScheduledJobs()) {\n\t\t\t\tif (job.status !== \"active\" || job.nextRunAt === undefined) continue;\n\t\t\t\tconst runAt = Date.parse(job.nextRunAt);\n\t\t\t\tif (!Number.isFinite(runAt) || runAt > now) continue;\n\t\t\t\tdue.set(canonicalSessionPath(rootSessionFile), rootSessionFile);\n\t\t\t}\n\t\t\tfor (const [rootKey, sessionPath] of due) {\n\t\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.createOrReuseWorker(SCHEDULED_WAKE_CLIENT_ID, { type: \"create\", sessionPath });\n\t\t\t\t\tthis.scheduledWakeFailures.delete(rootKey);\n\t\t\t\t\tthis.log(`Woke session worker for a due scheduled job: ${sessionPath}`);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tthis.scheduledWakeFailures.set(rootKey, Date.now());\n\t\t\t\t\tthis.log(`Scheduled wake failed for ${sessionPath}: ${String(error)}`);\n\t\t\t\t}\n\t\t\t}\n\t\t} finally {\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t}\n\t}\n\n\tprivate scheduleIdleEvictionSweep(): void {\n\t\tif (this.shuttingDown || this.idleEvictionTimer || this.idleEvictionSweep) return;\n\t\tconst delayMs = idleEvictionSweepIntervalMs(this.settingsManager.getIdleEvictionMinutes());\n\t\tthis.idleEvictionTimer = setTimeout(() => {\n\t\t\tthis.idleEvictionTimer = undefined;\n\t\t\tconst sweep = this.runIdleEvictionSweep()\n\t\t\t\t.catch((error) => this.log(`Idle eviction sweep failed: ${String(error)}`))\n\t\t\t\t.finally(() => {\n\t\t\t\t\tif (this.idleEvictionSweep === sweep) this.idleEvictionSweep = undefined;\n\t\t\t\t\tthis.scheduleIdleEvictionSweep();\n\t\t\t\t});\n\t\t\tthis.idleEvictionSweep = sweep;\n\t\t}, delayMs);\n\t\tthis.idleEvictionTimer.unref();\n\t}\n\n\tprivate workerEvictionSnapshot(worker: ResidentWorker): WorkerEvictionSnapshot {\n\t\treturn {\n\t\t\tlifecycle: worker.descriptor.lifecycle,\n\t\t\tisConnected: worker.client !== undefined,\n\t\t\tisStopping: this.isWorkerStopping(worker),\n\t\t\thasOwnerClient: worker.descriptor.ownerClientId !== undefined,\n\t\t\tisPreparingUpdateRestart:\n\t\t\t\tthis.updateRestartPhase !== undefined || worker.updateRestartPrepareClient !== undefined,\n\t\t\thasWakeBlindSchedule: this.isWakeBlindScheduledWorker(worker),\n\t\t\tsessions: this.workerRosterEntries(worker)\n\t\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t\t.map(sessionSummaryFromRosterEntry)\n\t\t\t\t.map((summary) => {\n\t\t\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\t\t\treturn {\n\t\t\t\t\t\tisSessionActive: isSessionSummaryBusy(summary),\n\t\t\t\t\t\tattachedClients: this.attachedClientCount(summary, activeSessionId),\n\t\t\t\t\t\thasRegisteredCronJob: summary.hasRegisteredCronJob === true,\n\t\t\t\t\t\tlastActivityAt: Date.parse(summary.lastActivityAt ?? \"\"),\n\t\t\t\t\t};\n\t\t\t\t}),\n\t\t};\n\t}\n\n\t/** A schedule whose root file sits outside the enumerable sessions root is invisible to the wake scan; keep it resident. */\n\tprivate isWakeBlindScheduledWorker(worker: ResidentWorker): boolean {\n\t\tconst sessionFile = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!sessionFile || !agentDir) return false;\n\t\tconst sessionsRoot = canonicalSessionPath(this.defaultSessionConfig.sessionDir ?? getSessionsDir(agentDir));\n\t\tif (dirname(canonicalSessionPath(sessionFile)) === sessionsRoot) return false;\n\t\treturn this.workerRosterEntries(worker).some(\n\t\t\t(entry) => entry.summary.hasRegisteredHeartbeat === true || entry.summary.hasRegisteredCronJob === true,\n\t\t);\n\t}\n\n\tprivate async runIdleEvictionSweep(now = Date.now()): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined || this.idleEvictionFence) return;\n\t\tawait this.settingsManager.reload();\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst idleEvictionMinutes = this.settingsManager.getIdleEvictionMinutes();\n\t\tif (idleEvictionMinutes === \"off\") return;\n\n\t\tconst refreshed = new Set<ResidentWorker>();\n\t\tawait Promise.all(\n\t\t\t[...this.workers.values()].map(async (worker) => {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.refreshWorkerSummaries(worker);\n\t\t\t\t\trefreshed.add(worker);\n\t\t\t\t} catch {\n\t\t\t\t\t// A disconnected or transitioning worker is never an eviction candidate.\n\t\t\t\t}\n\t\t\t}),\n\t\t);\n\t\tconst candidates = [...refreshed].filter((worker) =>\n\t\t\tcanEvictWorker(this.workerEvictionSnapshot(worker), idleEvictionMinutes, now),\n\t\t);\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t// Whole-tree candidates skip child work because stopWorker releases everything.\n\t\tawait Promise.all(\n\t\t\t[...refreshed]\n\t\t\t\t.filter((worker) => !candidates.includes(worker))\n\t\t\t\t.map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst response = await worker.client?.requestWorker(\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\ttype: \"worker_passivate_idle_children\",\n\t\t\t\t\t\t\t\tidleEvictionMinutes,\n\t\t\t\t\t\t\t\tnow,\n\t\t\t\t\t\t\t\tlimit: CHILD_PASSIVATION_PER_WORKER_CAP,\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t\t30_000,\n\t\t\t\t\t\t);\n\t\t\t\t\t\tif (response && !response.success) throw new Error(response.error);\n\t\t\t\t\t\tawait this.refreshWorkerSummaries(worker);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\trefreshed.delete(worker);\n\t\t\t\t\t\tthis.log(`Child passivation sweep failed for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t);\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\n\t\tawait this.withEvictionFence(\"Timed out draining daemon mutations for idle eviction\", async () => {\n\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\tawait Promise.all(\n\t\t\t\tcandidates.map((worker) => this.refreshWorkerSummaries(worker).catch(() => refreshed.delete(worker))),\n\t\t\t);\n\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\tconst evictable = candidates.filter(\n\t\t\t\t(worker) =>\n\t\t\t\t\trefreshed.has(worker) &&\n\t\t\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\t\t\tcanEvictWorker(this.workerEvictionSnapshot(worker), idleEvictionMinutes, now),\n\t\t\t);\n\t\t\t// Promise.all may reject and release the fence while sibling stops are still\n\t\t\t// finishing. That is safe: a racing mutation either reaches a live worker or\n\t\t\t// gets a clean disconnected/unknown-session error.\n\t\t\tawait Promise.all(\n\t\t\t\tevictable.map(async (worker) => {\n\t\t\t\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\t\t\t\tconst snapshot = this.workerEvictionSnapshot(worker);\n\t\t\t\t\tconst idleMinutes = Math.floor(\n\t\t\t\t\t\tMath.min(...snapshot.sessions.map((session) => now - session.lastActivityAt)) / 60_000,\n\t\t\t\t\t);\n\t\t\t\t\tconst root = worker.summaries.get(worker.descriptor.rootActiveSessionId);\n\t\t\t\t\tawait this.stopWorker(worker, true);\n\t\t\t\t\tthis.log(\n\t\t\t\t\t\t`Evicted idle worker ${worker.descriptor.workerId} root=${root?.sessionId ?? worker.descriptor.rootSessionId ?? worker.descriptor.rootActiveSessionId} idleMinutes=${idleMinutes} sessions=${snapshot.sessions.length}`,\n\t\t\t\t\t);\n\t\t\t\t}),\n\t\t\t);\n\t\t});\n\t}\n\n\t/** Waits for any held eviction fence, then takes the slot; release clears it only if still ours. */\n\tprivate async acquireIdleEvictionFence(): Promise<() => void> {\n\t\twhile (this.idleEvictionFence) await this.idleEvictionFence;\n\t\tlet releaseFence: () => void = () => {};\n\t\tconst fence = new Promise<void>((resolveFence) => {\n\t\t\treleaseFence = resolveFence;\n\t\t});\n\t\tthis.idleEvictionFence = fence;\n\t\treturn () => {\n\t\t\tif (this.idleEvictionFence === fence) this.idleEvictionFence = undefined;\n\t\t\treleaseFence();\n\t\t};\n\t}\n\n\t/** Runs a passivation decision under the eviction fence after draining admitted mutations. */\n\tprivate async withEvictionFence(drainMessage: string, action: () => Promise<void>): Promise<void> {\n\t\tconst releaseFence = await this.acquireIdleEvictionFence();\n\t\ttry {\n\t\t\tawait this.mutationDrain.waitForDrain(0, AbortSignal.timeout(IDLE_EVICTION_DRAIN_TIMEOUT_MS), drainMessage);\n\t\t\tawait action();\n\t\t} finally {\n\t\t\treleaseFence();\n\t\t}\n\t}\n\n\tprivate async passivateWorkerIfStillEligible(\n\t\tworker: ResidentWorker,\n\t\tisStillEligible: () => boolean,\n\t\tdescribeEvicted: () => string,\n\t): Promise<void> {\n\t\tawait this.refreshWorkerSummaries(worker, false, true);\n\t\tif (!isStillEligible()) return;\n\t\tawait this.stopWorker(worker, true);\n\t\tthis.log(describeEvicted());\n\t}\n\n\tprivate async evictEmptySessionOnLastDetach(activeSessionId: string): Promise<void> {\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined) return;\n\t\tconst worker = this.matchWorkers(activeSessionId)[0]?.worker;\n\t\tif (\n\t\t\t!worker ||\n\t\t\tworker.descriptor.lifecycle !== \"ready\" ||\n\t\t\t!worker.client ||\n\t\t\tworker.descriptor.ownerClientId !== undefined || // owned workers have their own cleanup path\n\t\t\tthis.isWorkerStopping(worker)\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\ttry {\n\t\t\tawait this.refreshWorkerSummaries(worker, false, true);\n\t\t} catch {\n\t\t\treturn;\n\t\t}\n\t\tif (!this.isEmptyDetachEvictionCandidate(worker)) return;\n\t\ttry {\n\t\t\t// Idle-sweep coordination: fence new mutations, drain admitted ones, re-read before deciding.\n\t\t\tawait this.withEvictionFence(\"Timed out draining daemon mutations for empty-session eviction\", () =>\n\t\t\t\tthis.passivateWorkerIfStillEligible(\n\t\t\t\t\tworker,\n\t\t\t\t\t() => this.isEmptyDetachEvictionCandidate(worker),\n\t\t\t\t\t() =>\n\t\t\t\t\t\t`Evicted empty session worker ${worker.descriptor.workerId} root=${worker.descriptor.rootSessionId ?? worker.descriptor.rootActiveSessionId} on last client detach`,\n\t\t\t\t),\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tthis.log(`Empty-session eviction failed for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate isEmptyDetachEvictionCandidate(worker: ResidentWorker): boolean {\n\t\tif (\n\t\t\tthis.shuttingDown ||\n\t\t\tthis.updateRestartPhase !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker ||\n\t\t\tthis.isWorkerStopping(worker) ||\n\t\t\tthis.isWakeBlindScheduledWorker(worker)\n\t\t) {\n\t\t\treturn false;\n\t\t}\n\t\tconst summaries = this.workerRosterEntries(worker)\n\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t.map(sessionSummaryFromRosterEntry);\n\t\tconst hasAttachedClient = summaries.some(\n\t\t\t(summary) => this.attachedClientCount(summary, summary.activeSessionId ?? summary.id) > 0,\n\t\t);\n\t\treturn summaries.length > 0 && !hasAttachedClient && summaries.every(isEvictableEmptySessionSummary);\n\t}\n\n\tprivate async assertCurrentOwnership(): Promise<void> {\n\t\tconst ownership = this.ownership;\n\t\tif (!ownership) {\n\t\t\tconst error = new Error(\n\t\t\t\t`Daemon supervisor generation ${this.generation} holds no registry ownership (never acquired or already released); ` +\n\t\t\t\t\t`socket: ${this.socketPath}; restart the daemon to recover — sessions are preserved`,\n\t\t\t);\n\t\t\tObject.assign(error, { code: \"supervisor_generation_stale\" as const });\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t\ttry {\n\t\t\tawait ownership.assertCurrent();\n\t\t} catch (error) {\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async assertServingCurrentOwnership(allowWorkerLedger = false): Promise<void> {\n\t\tthis.assertSupervisorServing(allowWorkerLedger);\n\t\tawait this.assertCurrentOwnership();\n\t\tthis.assertSupervisorServing(allowWorkerLedger);\n\t}\n\n\tprivate async assertRecoveryAllowed(): Promise<void> {\n\t\tawait this.assertServingCurrentOwnership();\n\t\tif (await isDaemonShutdownAdmissionActive()) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Daemon shutdown admission cancelled worker recovery\");\n\t\t}\n\t}\n\n\tprivate supervisorAuthenticationClaim(): {\n\t\tsupervisorGeneration: string;\n\t\tsupervisorPid: number;\n\t\tsupervisorProcessStartId?: string;\n\t\tsupervisorSocketPath: string;\n\t} {\n\t\tconst record = this.ownership?.record;\n\t\tif (!record) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Daemon supervisor ownership is unavailable\");\n\t\t}\n\t\treturn {\n\t\t\tsupervisorGeneration: this.generation,\n\t\t\tsupervisorPid: record.pid,\n\t\t\t...(record.processStartId ? { supervisorProcessStartId: record.processStartId } : {}),\n\t\t\tsupervisorSocketPath: record.socketPath,\n\t\t};\n\t}\n\n\tprivate loadWorkerDescriptors(): void {\n\t\tfor (const name of readdirSync(this.descriptorDir)) {\n\t\t\tif (name === SUPERVISOR_CONFIG_FILE_NAME || !name.endsWith(\".json\")) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst path = join(this.descriptorDir, name);\n\t\t\ttry {\n\t\t\t\tconst descriptor: unknown = JSON.parse(readFileSync(path, \"utf8\"));\n\t\t\t\tif (!isDaemonWorkerDescriptor(descriptor, this.socketPath)) {\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tdescriptor.supervisorSocketPath = normalizeSocketPath(descriptor.supervisorSocketPath);\n\t\t\t\tdescriptor.lifecycle = \"recovering\";\n\t\t\t\tdescriptor.recoveryJournalPath ??= join(this.descriptorDir, `${descriptor.workerId}.recovery.jsonl`);\n\t\t\t\tdescriptor.orphanProcessJournalPath ??= join(this.descriptorDir, `${descriptor.workerId}.orphans.jsonl`);\n\t\t\t\tconst durableDescriptor = durableDaemonWorkerDescriptor(descriptor);\n\t\t\t\tconst worker: ResidentWorker = {\n\t\t\t\t\tdescriptor: durableDescriptor,\n\t\t\t\t\tdescriptorPath: path,\n\t\t\t\t\tsummaries: new Map(),\n\t\t\t\t\tsnapshotCache: new Map(),\n\t\t\t\t\ttranscriptCaches: new Map(),\n\t\t\t\t\tsnapshotGenerations: new Map(),\n\t\t\t\t\tsnapshotLoads: new Map(),\n\t\t\t\t\tintentionalStop: durableDescriptor.stopRequestedAt !== undefined,\n\t\t\t\t\tstopRevision: 0,\n\t\t\t\t};\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.workers.set(durableDescriptor.workerId, worker);\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Ignoring invalid worker descriptor ${path}: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t}\n\n\t/** Trees whose ephemeral-stop cancel failed stay owned until a retry lands: tombstoned client-owned descriptors with no resident worker. */\n\tprivate collectEphemeralCancelIntents(): Array<{ descriptor: DaemonWorkerDescriptor; descriptorPath: string }> {\n\t\tconst intents: Array<{ descriptor: DaemonWorkerDescriptor; descriptorPath: string }> = [];\n\t\tlet names: string[];\n\t\ttry {\n\t\t\tnames = readdirSync(this.descriptorDir);\n\t\t} catch {\n\t\t\treturn intents;\n\t\t}\n\t\tfor (const name of names) {\n\t\t\tif (name === SUPERVISOR_CONFIG_FILE_NAME || !name.endsWith(\".json\")) continue;\n\t\t\tconst descriptorPath = join(this.descriptorDir, name);\n\t\t\tlet descriptor: unknown;\n\t\t\ttry {\n\t\t\t\tdescriptor = JSON.parse(readFileSync(descriptorPath, \"utf8\"));\n\t\t\t} catch {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (!isDaemonWorkerDescriptor(descriptor, this.socketPath)) continue;\n\t\t\tif (descriptor.stopRequestedAt === undefined || descriptor.ownerClientId === undefined) continue;\n\t\t\tif (this.workers.has(descriptor.workerId)) continue;\n\t\t\tintents.push({ descriptor, descriptorPath });\n\t\t}\n\t\treturn intents;\n\t}\n\n\tprivate loadPersistedSupervisorConfig(): AgentSessionRuntimeConfig | undefined {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(\n\t\t\t\treadFileSync(this.supervisorConfigPath, \"utf8\"),\n\t\t\t) as Partial<PersistedSupervisorConfig>;\n\t\t\tif (\n\t\t\t\tparsed.version !== 1 ||\n\t\t\t\ttypeof parsed.socketPath !== \"string\" ||\n\t\t\t\tnormalizeSocketPath(parsed.socketPath) !== this.socketPath ||\n\t\t\t\t!parsed.defaultSessionConfig ||\n\t\t\t\ttypeof parsed.defaultSessionConfig !== \"object\" ||\n\t\t\t\ttypeof parsed.defaultSessionConfig.agentDir !== \"string\"\n\t\t\t) {\n\t\t\t\treturn undefined;\n\t\t\t}\n\t\t\treturn durableAgentSessionRuntimeConfig(parsed.defaultSessionConfig);\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate persistSupervisorConfig(): void {\n\t\tconst persisted: PersistedSupervisorConfig = {\n\t\t\tversion: 1,\n\t\t\tsocketPath: this.socketPath,\n\t\t\tdefaultSessionConfig: durableAgentSessionRuntimeConfig(this.defaultSessionConfig),\n\t\t};\n\t\tconst tempPath = `${this.supervisorConfigPath}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(persisted, null, 2)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\trenameSync(tempPath, this.supervisorConfigPath);\n\t}\n\n\tprivate hasPersistedWorkerDescriptors(): boolean {\n\t\treturn readdirSync(this.descriptorDir).some(\n\t\t\t(name) => name !== SUPERVISOR_CONFIG_FILE_NAME && name.endsWith(\".json\"),\n\t\t);\n\t}\n\n\tprivate persistWorker(worker: ResidentWorker): void {\n\t\tworker.descriptor.updatedAt = new Date().toISOString();\n\t\tconst persisted = durableDaemonWorkerDescriptor(worker.descriptor);\n\t\tconst tempPath = `${worker.descriptorPath}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(persisted, null, 2)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\trenameSync(tempPath, worker.descriptorPath);\n\t}\n\n\tprivate deleteWorkerDescriptor(worker: { descriptorPath: string; descriptor: DaemonWorkerDescriptor }): void {\n\t\ttry {\n\t\t\trmSync(worker.descriptorPath, { force: true });\n\t\t\trmSync(worker.descriptor.recoveryJournalPath, { force: true });\n\t\t\tif (worker.descriptor.orphanProcessJournalPath) {\n\t\t\t\trmSync(worker.descriptor.orphanProcessJournalPath, { force: true });\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Failed to remove worker descriptor ${worker.descriptorPath}: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate handleConnection(socket: Socket): void {\n\t\tconst client: DaemonSocketClient = {\n\t\t\tid: createActiveSessionId(),\n\t\t\tsocket,\n\t\t\tattachedActiveSessionIds: new Set(),\n\t\t\tcatchupActiveSessionIds: new Set(),\n\t\t\tbackpressured: false,\n\t\t\tauthenticated: true,\n\t\t\tsnapshotActiveSessionIds: new Set(),\n\t\t\tdetachInput: () => {},\n\t\t\tsupportsExtensionUi: false,\n\t\t\tcapabilities: new Set(DAEMON_DEFAULT_CLIENT_CAPABILITIES),\n\t\t};\n\t\tthis.connectionIds.set(client, client.id);\n\t\tthis.sessionInputPauseEpochs.set(client, 0);\n\t\tthis.detachingInputPauseSessions.set(client, new Set());\n\t\tthis.clients.add(client);\n\t\tvoid this.ready.then(\n\t\t\t() => {\n\t\t\t\tif (!client.socket.destroyed && this.clients.has(client)) {\n\t\t\t\t\tthis.write(client, {\n\t\t\t\t\t\ttype: \"daemon_hello\",\n\t\t\t\t\t\tsocketPath: this.socketPath,\n\t\t\t\t\t\tprotocol: DAEMON_PROTOCOL_INFO,\n\t\t\t\t\t\tschemaId: DAEMON_SCHEMA_ID,\n\t\t\t\t\t\tschemaRevision: DAEMON_SCHEMA_REVISION,\n\t\t\t\t\t\tappVersion: VERSION,\n\t\t\t\t\t\truntime: getDaemonRuntimeIdentity(),\n\t\t\t\t\t\tsupervisorGeneration: this.generation,\n\t\t\t\t\t\tsupervisorOwnerToken: this.ownership?.record.token,\n\t\t\t\t\t\tsupervisorPid: process.pid,\n\t\t\t\t\t\tsupervisorProcessStartId: this.ownership?.record.processStartId,\n\t\t\t\t\t\tsupervisorSocketPath: this.ownership?.record.socketPath,\n\t\t\t\t\t\tclientId: client.id,\n\t\t\t\t\t\tserverCapabilities: SUPERVISOR_SERVER_CAPABILITIES,\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t},\n\t\t\t() => client.socket.destroy(),\n\t\t);\n\n\t\tclient.detachInput = attachJsonlLineReader(socket, (line) => void this.handleLine(client, line));\n\t\tlet cleaned = false;\n\t\tconst cleanup = () => {\n\t\t\tif (cleaned) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tcleaned = true;\n\t\t\tclient.detachInput();\n\t\t\tthis.sessionInputPauseEpochs.set(client, (this.sessionInputPauseEpochs.get(client) ?? 0) + 1);\n\t\t\tconst ownerClientId = this.protocolClientId(client);\n\t\t\tvoid this.releaseClientSessionInputPauses(client, undefined, true).catch((error: unknown) =>\n\t\t\t\tthis.log(`Failed to release input pauses for disconnected client ${ownerClientId}: ${String(error)}`),\n\t\t\t);\n\t\t\tthis.clients.delete(client);\n\t\t\tthis.cancelWaitingPromptAdmissionsForClient(client);\n\t\t\tfor (const activeSessionId of [...client.attachedActiveSessionIds]) {\n\t\t\t\tclient.attachedActiveSessionIds.delete(activeSessionId);\n\t\t\t\tvoid this.syncWorkerExtensionUi(activeSessionId);\n\t\t\t\tvoid this.evictEmptySessionOnLastDetach(activeSessionId);\n\t\t\t}\n\t\t\tthis.scheduleOwnedWorkerCleanupForClient(this.protocolClientId(client));\n\t\t};\n\t\tsocket.on(\"close\", cleanup);\n\t\tsocket.on(\"error\", cleanup);\n\t\tsocket.on(\"drain\", () => {\n\t\t\tclient.backpressured = false;\n\t\t\tif (client.rosterResyncPending && client.rosterSubscribed === true) {\n\t\t\t\t// socket.write queues even when it reports backpressure: one resync per loss gap.\n\t\t\t\tclient.rosterResyncPending = false;\n\t\t\t\tthis.write(client, { type: \"roster_update\", changed: this.rosterEntriesForClient(), resync: true });\n\t\t\t}\n\t\t\tif (!client.snapshotStreaming) {\n\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t);\n\t\t\t}\n\t\t});\n\t}\n\n\tprivate cancelOwnedWorkerCleanup(clientId: string): void {\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.descriptor.ownerClientId !== clientId || !worker.ownerCleanupTimer) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t}\n\n\tprivate protocolClientId(client: DaemonSocketClient): string {\n\t\treturn this.protocolClientIds.get(client) ?? client.id;\n\t}\n\n\tprivate async releaseClientSessionInputPauses(\n\t\towner: DaemonSocketClient,\n\t\tactiveSessionId?: string,\n\t\tforceCleanupOnFailure = false,\n\t): Promise<void> {\n\t\tconst entries = [...this.sessionInputPauses.values()].filter(\n\t\t\t(entry) =>\n\t\t\t\tentry.owner === owner &&\n\t\t\t\t(activeSessionId === undefined ||\n\t\t\t\t\tentry.activeSessionId === activeSessionId ||\n\t\t\t\t\tentry.requestedActiveSessionId === activeSessionId),\n\t\t);\n\t\tawait Promise.all(\n\t\t\tentries.map(async (entry) => {\n\t\t\t\tif (this.workers.get(entry.worker.descriptor.workerId) !== entry.worker || !entry.worker.client) {\n\t\t\t\t\tif (this.sessionInputPauses.get(entry.pauseId) === entry) this.sessionInputPauses.delete(entry.pauseId);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tconst response = await this.forwardToWorker(\n\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\tid: randomUUID(),\n\t\t\t\t\t\t\ttype: \"release_session_input_pause\",\n\t\t\t\t\t\t\tactiveSessionId: entry.activeSessionId,\n\t\t\t\t\t\t\tpauseId: entry.pauseId,\n\t\t\t\t\t\t},\n\t\t\t\t\t\tINPUT_PAUSE_CLEANUP_TIMEOUT_MS,\n\t\t\t\t\t);\n\t\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\t\tif (this.sessionInputPauses.get(entry.pauseId) === entry) this.sessionInputPauses.delete(entry.pauseId);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (forceCleanupOnFailure) {\n\t\t\t\t\t\tconst workerClient = entry.worker.client;\n\t\t\t\t\t\tif (workerClient) {\n\t\t\t\t\t\t\tworkerClient.close();\n\t\t\t\t\t\t\tawait this.handleWorkerClose(\n\t\t\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t\t\tworkerClient,\n\t\t\t\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t} else {\n\t\t\t\t\t\t\tthis.invalidateWorkerSessionInputPauses(\n\t\t\t\t\t\t\t\tentry.worker,\n\t\t\t\t\t\t\t\t\"Session worker became unavailable while releasing an input pause\",\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t}),\n\t\t);\n\t}\n\n\tprivate invalidateWorkerSessionInputPauses(worker: ResidentWorker, reason: string): void {\n\t\tconst owners = new Set<DaemonSocketClient>();\n\t\tfor (const [pauseId, entry] of this.sessionInputPauses) {\n\t\t\tif (entry.worker !== worker) continue;\n\t\t\tthis.sessionInputPauses.delete(pauseId);\n\t\t\towners.add(entry.owner);\n\t\t}\n\t\tfor (const owner of owners) {\n\t\t\towner.socket.destroy(new Error(reason));\n\t\t}\n\t}\n\n\tprivate scheduleOwnedWorkerCleanupForClient(clientId: string): void {\n\t\tif ([...this.clients].some((client) => this.protocolClientId(client) === clientId)) {\n\t\t\treturn;\n\t\t}\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.descriptor.ownerClientId === clientId) {\n\t\t\t\tthis.scheduleOwnedWorkerCleanup(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate scheduleOwnedWorkerCleanup(worker: ResidentWorker): void {\n\t\tconst ownerClientId = worker.descriptor.ownerClientId;\n\t\tif (\n\t\t\t!ownerClientId ||\n\t\t\tworker.compatibilityError !== undefined ||\n\t\t\tworker.ownerCleanupTimer ||\n\t\t\t[...this.clients].some((client) => this.protocolClientId(client) === ownerClientId)\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tworker.ownerCleanupTimer = setTimeout(() => {\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t\tif (\n\t\t\t\tworker.compatibilityError !== undefined ||\n\t\t\t\tworker.descriptor.ownerClientId !== ownerClientId ||\n\t\t\t\t[...this.clients].some((client) => this.protocolClientId(client) === ownerClientId) ||\n\t\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t\t) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tvoid this.stopWorker(worker, true).catch((error) =>\n\t\t\t\tthis.log(`Could not clean up client-owned worker ${worker.descriptor.workerId}: ${String(error)}`),\n\t\t\t);\n\t\t}, OWNED_WORKER_DISCONNECT_GRACE_MS);\n\t\tworker.ownerCleanupTimer.unref();\n\t}\n\n\tprivate promptAdmissionKey(activeSessionId: string, publicAdmissionId: string): string {\n\t\treturn `${activeSessionId}\\0${publicAdmissionId}`;\n\t}\n\n\tprivate promptAdmissionsFor(client: DaemonSocketClient): Map<string, SupervisorPromptAdmission> {\n\t\tlet admissions = this.promptAdmissions.get(client);\n\t\tif (!admissions) {\n\t\t\tadmissions = new Map();\n\t\t\tthis.promptAdmissions.set(client, admissions);\n\t\t}\n\t\treturn admissions;\n\t}\n\n\tprivate getPromptAdmission(\n\t\tclient: DaemonSocketClient,\n\t\tactiveSessionId: string,\n\t\tpublicAdmissionId: string,\n\t): SupervisorPromptAdmission | undefined {\n\t\treturn this.promptAdmissions.get(client)?.get(this.promptAdmissionKey(activeSessionId, publicAdmissionId));\n\t}\n\n\tprivate deletePromptAdmission(admission: SupervisorPromptAdmission): void {\n\t\tconst admissions = this.promptAdmissions.get(admission.client);\n\t\tconst key = this.promptAdmissionKey(admission.activeSessionId, admission.publicAdmissionId);\n\t\tif (admissions?.get(key) !== admission) return;\n\t\tadmissions.delete(key);\n\t\tif (admissions.size === 0) this.promptAdmissions.delete(admission.client);\n\t}\n\n\tprivate cancelWaitingPromptAdmissionsForClient(client: DaemonSocketClient): void {\n\t\tfor (const admission of this.promptAdmissions.get(client)?.values() ?? []) {\n\t\t\tif (admission.status !== \"waiting\") continue;\n\t\t\tif (!admission.worker || !admission.workerActiveSessionId) {\n\t\t\t\tadmission.status = \"cancelled\";\n\t\t\t\tadmission.controller.abort();\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst worker = admission.worker;\n\t\t\tconst workerActiveSessionId = admission.workerActiveSessionId;\n\t\t\tvoid this.forwardToWorker(worker, {\n\t\t\t\ttype: \"cancel_prompt_admission\",\n\t\t\t\tactiveSessionId: workerActiveSessionId,\n\t\t\t\tadmissionId: admission.workerAdmissionId,\n\t\t\t})\n\t\t\t\t.then((response) => {\n\t\t\t\t\tif (admission.status !== \"waiting\") return;\n\t\t\t\t\tconst status =\n\t\t\t\t\t\tresponse.success && response.data && typeof response.data === \"object\" && \"status\" in response.data\n\t\t\t\t\t\t\t? (response.data as { status?: unknown }).status\n\t\t\t\t\t\t\t: undefined;\n\t\t\t\t\tif (status === \"owned\") admission.status = \"owned\";\n\t\t\t\t\telse if (status === \"cancelled\") admission.status = \"cancelled\";\n\t\t\t\t})\n\t\t\t\t.catch((error: unknown) => {\n\t\t\t\t\tthis.log(\n\t\t\t\t\t\t`Could not cancel prompt admission ${admission.workerAdmissionId} on disconnected client: ${String(error)}`,\n\t\t\t\t\t);\n\t\t\t\t});\n\t\t}\n\t}\n\n\t/** Non-async by design: prompt registration completes before handleLine's first await. */\n\tprivate parseCommandAndRegisterPromptAdmission(\n\t\tclient: DaemonSocketClient,\n\t\tline: string,\n\t): {\n\t\tcommand: DaemonCommand;\n\t\tenvelopeClientId?: string;\n\t\tprotocolVersion: number;\n\t\tadmission?: SupervisorPromptAdmission;\n\t} {\n\t\tconst parsed = JSON.parse(line) as unknown;\n\t\tconst envelope = isDaemonCommandEnvelope(parsed) ? parsed : undefined;\n\t\tif (!envelope) {\n\t\t\tthrow new Error(`Daemon commands require protocol ${DAEMON_COMMAND_ENVELOPE_MIN_PROTOCOL_VERSION} or newer`);\n\t\t}\n\t\tconst command = { ...envelope.command, id: envelope.id } as DaemonCommand;\n\t\tif (\n\t\t\tenvelope.protocol.version < CANONICAL_SESSION_OWNERSHIP_COMPATIBILITY.minProtocol &&\n\t\t\t!isLegacyDaemonInspection(command)\n\t\t) {\n\t\t\tthrow new Error(\n\t\t\t\t`Native daemon commands require protocol ${CANONICAL_SESSION_OWNERSHIP_COMPATIBILITY.minProtocol} canonical session ownership and bounded invocation output; only passive inspection is available`,\n\t\t\t);\n\t\t}\n\t\tlet admission: SupervisorPromptAdmission | undefined;\n\t\tif ((command.type === \"prompt\" || command.type === \"prompt_and_wait\") && command.admissionId !== undefined) {\n\t\t\tif (typeof command.activeSessionId !== \"string\" || typeof command.admissionId !== \"string\") {\n\t\t\t\tthrow new Error(\"Prompt admission requires string activeSessionId and admissionId\");\n\t\t\t}\n\t\t\tif (command.admissionId === \"\") throw new Error(\"admissionId must not be empty\");\n\t\t\tconst admissions = this.promptAdmissionsFor(client);\n\t\t\tconst key = this.promptAdmissionKey(command.activeSessionId, command.admissionId);\n\t\t\tif (admissions.has(key)) {\n\t\t\t\tthrow new Error(`Prompt admission id is already in use: ${command.admissionId}`);\n\t\t\t}\n\t\t\tadmission = {\n\t\t\t\tclient,\n\t\t\t\tactiveSessionId: command.activeSessionId,\n\t\t\t\tpublicAdmissionId: command.admissionId,\n\t\t\t\tworkerAdmissionId: `supervisor-admission:${randomUUID()}`,\n\t\t\t\tstatus: \"waiting\",\n\t\t\t\tcontroller: new AbortController(),\n\t\t\t};\n\t\t\tadmissions.set(key, admission);\n\t\t}\n\t\treturn {\n\t\t\tcommand,\n\t\t\tenvelopeClientId: envelope.clientId ?? client.id,\n\t\t\tprotocolVersion: envelope.protocol.version,\n\t\t\tadmission,\n\t\t};\n\t}\n\n\tprivate isWorkerLedgerEnvelope(line: string): boolean {\n\t\ttry {\n\t\t\tconst parsed: unknown = JSON.parse(line);\n\t\t\treturn (\n\t\t\t\tisDaemonCommandEnvelope(parsed) &&\n\t\t\t\t(parsed.command.type === \"rlm_ledger_mutate\" || parsed.command.type === \"rlm_capacity\")\n\t\t\t);\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate async handleLine(client: DaemonSocketClient, line: string): Promise<void> {\n\t\ttry {\n\t\t\tthis.assertSupervisorServing(this.shuttingDown && this.isWorkerLedgerEnvelope(line));\n\t\t} catch (error) {\n\t\t\tthis.write(client, failure(salvageDaemonCommandId(line), \"dispatch\", error, serializeDaemonError(error)));\n\t\t\treturn;\n\t\t}\n\t\tlet preParsed: ReturnType<DaemonSupervisor[\"parseCommandAndRegisterPromptAdmission\"]>;\n\t\ttry {\n\t\t\tpreParsed = this.parseCommandAndRegisterPromptAdmission(client, line);\n\t\t} catch (error) {\n\t\t\tthis.write(client, failure(salvageDaemonCommandId(line), \"parse\", error));\n\t\t\treturn;\n\t\t}\n\t\tconst command = preParsed.command;\n\t\tconst parsedAdmission = preParsed.admission;\n\t\tif (command.type === \"cancel_prompt_admission\" && this.updateRestartPhase !== undefined) {\n\t\t\tthis.write(client, failure(command.id, command.type, \"Daemon is preparing an update restart\"));\n\t\t\treturn;\n\t\t}\n\t\tconst cancellationAdmission =\n\t\t\tcommand.type === \"cancel_prompt_admission\"\n\t\t\t\t? this.getPromptAdmission(client, command.activeSessionId, command.admissionId)\n\t\t\t\t: undefined;\n\t\tif (cancellationAdmission?.status === \"waiting\" && !cancellationAdmission.worker) {\n\t\t\tcancellationAdmission.status = \"cancelled\";\n\t\t\tcancellationAdmission.controller.abort();\n\t\t}\n\t\ttry {\n\t\t\tawait waitForPromptAdmission(this.ready, parsedAdmission?.controller.signal);\n\t\t} catch (error) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, error));\n\t\t\treturn;\n\t\t}\n\t\tconst envelopeClientId = preParsed.envelopeClientId;\n\t\tif (envelopeClientId) {\n\t\t\tthis.protocolClientIds.set(client, envelopeClientId);\n\t\t\tclient.id = envelopeClientId;\n\t\t}\n\t\tthis.cancelOwnedWorkerCleanup(client.id);\n\t\tif (!DAEMON_COMMAND_TYPES.has(command.type)) {\n\t\t\tthis.write(client, failure(command.id, command.type, `Unknown daemon command: ${command.type}`));\n\t\t\treturn;\n\t\t}\n\t\tif (\n\t\t\tcommand.type === \"get_session_tree\" &&\n\t\t\tpreParsed.protocolVersion < DAEMON_COMMAND_COMPATIBILITY.get_session_tree.minProtocol\n\t\t) {\n\t\t\tthis.write(\n\t\t\t\tclient,\n\t\t\t\tfailure(\n\t\t\t\t\tcommand.id,\n\t\t\t\t\tcommand.type,\n\t\t\t\t\t`get_session_tree requires client protocol ${DAEMON_COMMAND_COMPATIBILITY.get_session_tree.minProtocol} or newer`,\n\t\t\t\t),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\ttry {\n\t\t\tawait waitForPromptAdmission(\n\t\t\t\tthis.assertServingCurrentOwnership(command.type === \"rlm_ledger_mutate\" || command.type === \"rlm_capacity\"),\n\t\t\t\tparsedAdmission?.controller.signal,\n\t\t\t);\n\t\t} catch (error) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, error));\n\t\t\treturn;\n\t\t}\n\n\t\tconst mutation = isDaemonMutatingCommand(command);\n\t\tconst journalIdentity =\n\t\t\tenvelopeClientId && command.id && mutation ? { clientId: envelopeClientId, commandId: command.id } : undefined;\n\t\tconst existing = journalIdentity\n\t\t\t? this.commandJournal.lookup(journalIdentity.clientId, journalIdentity.commandId)\n\t\t\t: undefined;\n\t\tif (existing?.status === \"complete\") {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, existing.response);\n\t\t\treturn;\n\t\t}\n\t\tif (existing?.status === \"pending\" && journalIdentity) {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(\n\t\t\t\tclient,\n\t\t\t\tfailure(command.id, command.type, \"The previous command result is uncertain and was not replayed\", {\n\t\t\t\t\tcode: \"command_result_uncertain\",\n\t\t\t\t\t...journalIdentity,\n\t\t\t\t}),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tconst phase = this.updateRestartPhase;\n\t\tconst restartRejected =\n\t\t\tphase === \"draining\"\n\t\t\t\t? !UPDATE_RESTART_DRAIN_COMMANDS.has(command.type)\n\t\t\t\t: phase !== undefined && !(phase === \"prepared\" && command.type === \"shutdown\");\n\t\tif (restartRejected && mutation && command.type !== \"rlm_ledger_mutate\" && command.type !== \"rlm_capacity\") {\n\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\tthis.write(client, failure(command.id, command.type, \"Daemon is preparing an update restart\"));\n\t\t\treturn;\n\t\t}\n\t\tif (mutation && !UPDATE_RESTART_DRAIN_COMMANDS.has(command.type)) {\n\t\t\tconst idleEvictionFence = this.idleEvictionFence;\n\t\t\tif (idleEvictionFence) {\n\t\t\t\tawait idleEvictionFence;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertServingCurrentOwnership(\n\t\t\t\t\t\tcommand.type === \"rlm_ledger_mutate\" || command.type === \"rlm_capacity\",\n\t\t\t\t\t);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\t\tthis.write(client, failure(command.id, command.type, error, serializeDaemonError(error)));\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tif (journalIdentity) {\n\t\t\tlet admitted: CommandJournalBeginResult;\n\t\t\ttry {\n\t\t\t\tadmitted = this.commandJournal.begin(journalIdentity.clientId, journalIdentity.commandId, command.type);\n\t\t\t} catch (error) {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(client, failure(command.id, command.type, error, serializeDaemonError(error)));\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (admitted.status === \"complete\") {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(client, admitted.response);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (admitted.status === \"pending\") {\n\t\t\t\tif (parsedAdmission) this.deletePromptAdmission(parsedAdmission);\n\t\t\t\tthis.write(\n\t\t\t\t\tclient,\n\t\t\t\t\tfailure(command.id, command.type, \"The previous command result is uncertain and was not replayed\", {\n\t\t\t\t\t\tcode: \"command_result_uncertain\",\n\t\t\t\t\t\t...journalIdentity,\n\t\t\t\t\t}),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\n\t\t// Attach is intentionally read-only and is not fence-gated. If eviction wins\n\t\t// the race, attach fails cleanly with \"Session worker is not connected\" and\n\t\t// the client retries through the saved-session path instead of mutating state.\n\t\tif (mutation) this.mutationDrain.begin();\n\t\tlet commandCompleted = false;\n\t\ttry {\n\t\t\tconst response = await this.handleCommand(client, command, cancellationAdmission);\n\t\t\tcommandCompleted = true;\n\t\t\tif (response) {\n\t\t\t\tif (journalIdentity) {\n\t\t\t\t\tawait this.assertCurrentOwnership();\n\t\t\t\t\tthis.commandJournal.recordResult(journalIdentity.clientId, journalIdentity.commandId, response);\n\t\t\t\t}\n\t\t\t\tthis.write(client, response);\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Supervisor command ${command.type} failed: ${error instanceof Error ? error.stack : String(error)}`);\n\t\t\tlet response = failure(command.id, command.type, error, serializeDaemonError(error));\n\t\t\t// A completed command must not get a conflicting failure result after uncertain result persistence.\n\t\t\tif (journalIdentity && !commandCompleted && !isSupervisorGenerationStale(error)) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertCurrentOwnership();\n\t\t\t\t\tthis.commandJournal.recordResult(journalIdentity.clientId, journalIdentity.commandId, response);\n\t\t\t\t} catch (persistenceError) {\n\t\t\t\t\tconst failureError =\n\t\t\t\t\t\tpersistenceError === error\n\t\t\t\t\t\t\t? error\n\t\t\t\t\t\t\t: new AggregateError(\n\t\t\t\t\t\t\t\t\t[error, persistenceError],\n\t\t\t\t\t\t\t\t\t\"Supervisor command and result persistence failed\",\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\tresponse = failure(command.id, command.type, failureError, serializeDaemonError(failureError));\n\t\t\t\t}\n\t\t\t}\n\t\t\tthis.write(client, response);\n\t\t} finally {\n\t\t\tif (mutation) this.mutationDrain.end();\n\t\t}\n\t}\n\n\tprivate rlmCapacityOwner(worker: ResidentWorker): string {\n\t\treturn `${worker.descriptor.workerId}:${worker.descriptor.workerInstanceId ?? \"unavailable\"}`;\n\t}\n\n\tprivate async rlmCapacityFamily(sessionFile: string | undefined, owner: string, sessionId: string): Promise<string> {\n\t\tif (!sessionFile) return `memory:${owner}:${sessionId}`;\n\t\tconst parents = new Map(\n\t\t\t(await this.rlmSpawnLedger().liveEdges({ strict: true })).map(\n\t\t\t\t(edge) => [canonicalSessionPath(edge.child), canonicalSessionPath(edge.parent)] as const,\n\t\t\t),\n\t\t);\n\t\tlet root = canonicalSessionPath(sessionFile);\n\t\tconst visited = new Set<string>();\n\t\twhile (parents.has(root)) {\n\t\t\tif (visited.has(root)) throw new Error(\"RLM family ancestry contains a cycle\");\n\t\t\tvisited.add(root);\n\t\t\tconst parent = parents.get(root);\n\t\t\tif (parent === undefined) break;\n\t\t\troot = parent;\n\t\t}\n\t\treturn `file:${root}`;\n\t}\n\n\tprivate isRlmMainRoot(sessionFile: string | undefined, family: string): boolean {\n\t\treturn sessionFile === undefined || family === `file:${canonicalSessionPath(sessionFile)}`;\n\t}\n\n\tprivate releaseWorkerRlmCapacity(worker: ResidentWorker): void {\n\t\tthis.rlmCapacity.releaseOwner(this.rlmCapacityOwner(worker));\n\t\tthis.pendingRlmCapacity.delete(worker);\n\t\tthis.releasedDuringCapacityRecovery.delete(worker);\n\t}\n\n\tprivate async assertRlmCapacityKnown(family: string): Promise<void> {\n\t\tfor (const worker of [...this.pendingRlmCapacity]) {\n\t\t\tconst identity = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\tif (identity === \"gone\" || identity === \"replaced\") {\n\t\t\t\tthis.releaseWorkerRlmCapacity(worker);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst path = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\t\tconst pendingFamily = path\n\t\t\t\t? await this.rlmCapacityFamily(\n\t\t\t\t\t\tpath,\n\t\t\t\t\t\tthis.rlmCapacityOwner(worker),\n\t\t\t\t\t\tworker.descriptor.rootSessionId ?? \"pending\",\n\t\t\t\t\t)\n\t\t\t\t: undefined;\n\t\t\tif (this.pendingRlmCapacity.has(worker) && (pendingFamily === undefined || pendingFamily === family)) {\n\t\t\t\tthrow new Error(\"RLM family capacity is waiting for a live worker to reconnect\");\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async restoreWorkerRlmCapacity(worker: ResidentWorker, client: DaemonWorkerClient): Promise<void> {\n\t\tif (!this.pendingRlmCapacity.has(worker)) return;\n\t\tconst owner = this.rlmCapacityOwner(worker);\n\t\tconst response = await client.requestWorker({ type: \"worker_get_rlm_capacity\" }, 2000);\n\t\tif (!response.success) throw deserializeDaemonError(response);\n\t\tif (!isDaemonRlmCapacitySnapshot(response.data))\n\t\t\tthrow new Error(\"Worker returned an invalid RLM capacity snapshot\");\n\t\tconst recovered = await Promise.all(\n\t\t\tresponse.data.reservations.map(async (reservation) => ({\n\t\t\t\treservation,\n\t\t\t\tfamily: await this.rlmCapacityFamily(reservation.sessionFile, owner, reservation.sessionId),\n\t\t\t})),\n\t\t);\n\t\tif (this.rlmCapacityOwner(worker) !== owner || !this.pendingRlmCapacity.has(worker)) return;\n\t\tconst released = this.releasedDuringCapacityRecovery.get(worker);\n\t\tfor (const { reservation, family } of recovered) {\n\t\t\tif (released?.has(reservation.reservationId)) continue;\n\t\t\tif (reservation.residentRoot && this.isRlmMainRoot(reservation.sessionFile, family)) continue;\n\t\t\tthis.rlmCapacity.restore(owner, reservation.reservationId, family);\n\t\t}\n\t\tthis.pendingRlmCapacity.delete(worker);\n\t\tthis.releasedDuringCapacityRecovery.delete(worker);\n\t}\n\n\tprivate async handleRlmCapacity(command: Extract<DaemonCommand, { type: \"rlm_capacity\" }>): Promise<DaemonResponse> {\n\t\tconst worker = [...this.workers.values()].find(\n\t\t\t(candidate) =>\n\t\t\t\tcandidate.descriptor.authenticationToken === command.workerToken &&\n\t\t\t\tcandidate.descriptor.workerInstanceId === command.workerInstanceId,\n\t\t);\n\t\tif (\n\t\t\t!worker ||\n\t\t\t!command.workerInstanceId ||\n\t\t\tworker.compatibilityError ||\n\t\t\tthis.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId) !== \"current\"\n\t\t) {\n\t\t\tthrow new Error(\"Worker authentication failed for RLM capacity\");\n\t\t}\n\t\tconst owner = this.rlmCapacityOwner(worker);\n\t\tconst journal = this.assertRlmJournalAdmission();\n\t\tconst scope = this.rlmJournalScope(worker.descriptor.sessionDir ?? this.defaultSessionConfig.sessionDir);\n\t\tif (canonicalSessionPath(scope.journalPath) !== journal.ledgerPath) {\n\t\t\tthrow new Error(\"Worker RLM family is not owned by this supervisor\");\n\t\t}\n\t\tconst operation = command.operation;\n\t\tif (!operation || typeof operation !== \"object\") throw new Error(\"Invalid RLM capacity operation\");\n\t\tif (operation.op === \"release\") {\n\t\t\tif (typeof operation.reservationId !== \"string\" || !operation.reservationId)\n\t\t\t\tthrow new Error(\"Invalid RLM reservation id\");\n\t\t\tthis.rlmCapacity.release(owner, operation.reservationId);\n\t\t\tif (this.pendingRlmCapacity.has(worker)) {\n\t\t\t\tlet released = this.releasedDuringCapacityRecovery.get(worker);\n\t\t\t\tif (!released) {\n\t\t\t\t\treleased = new Set();\n\t\t\t\t\tthis.releasedDuringCapacityRecovery.set(worker, released);\n\t\t\t\t}\n\t\t\t\treleased.add(operation.reservationId);\n\t\t\t}\n\t\t\treturn success(command.id, command.type);\n\t\t}\n\t\tif (operation.op !== \"status\" && operation.op !== \"set\" && operation.op !== \"reserve\")\n\t\t\tthrow new Error(\"Unknown RLM capacity operation\");\n\t\tif (\n\t\t\ttypeof operation.sessionId !== \"string\" ||\n\t\t\t!operation.sessionId ||\n\t\t\t(operation.sessionFile !== undefined && (typeof operation.sessionFile !== \"string\" || !operation.sessionFile))\n\t\t) {\n\t\t\tthrow new Error(\"Invalid RLM capacity origin\");\n\t\t}\n\t\tconst family = await this.rlmCapacityFamily(operation.sessionFile, owner, operation.sessionId);\n\t\tawait this.assertServingCurrentOwnership(true);\n\t\tif (this.rlmCapacityOwner(worker) !== owner || worker.descriptor.authenticationToken !== command.workerToken) {\n\t\t\tthrow new Error(\"RLM capacity worker incarnation changed\");\n\t\t}\n\t\tif (operation.op === \"status\") return success(command.id, command.type, this.rlmCapacity.getStatus(family));\n\t\tif (this.shuttingDown || this.updateRestartPhase !== undefined || worker.descriptor.stopRequestedAt) {\n\t\t\tthrow new Error(\"RLM capacity admission is closed while the worker or daemon is stopping\");\n\t\t}\n\t\tif (operation.op === \"set\") {\n\t\t\treturn success(\n\t\t\t\tcommand.id,\n\t\t\t\tcommand.type,\n\t\t\t\tawait this.rlmCapacity.setMaxSubagents(family, operation.maxSubagents),\n\t\t\t);\n\t\t}\n\t\tif (\n\t\t\ttypeof operation.reservationId !== \"string\" ||\n\t\t\t!operation.reservationId ||\n\t\t\t(operation.residentRoot !== undefined && operation.residentRoot !== true)\n\t\t)\n\t\t\tthrow new Error(\"Invalid RLM reservation\");\n\t\tif (operation.residentRoot && this.isRlmMainRoot(operation.sessionFile, family)) {\n\t\t\treturn success(command.id, command.type, this.rlmCapacity.getStatus(family));\n\t\t}\n\t\tawait this.assertRlmCapacityKnown(family);\n\t\tawait this.assertServingCurrentOwnership();\n\t\tif (\n\t\t\tthis.updateRestartPhase !== undefined ||\n\t\t\tworker.descriptor.stopRequestedAt ||\n\t\t\tthis.rlmCapacityOwner(worker) !== owner ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker ||\n\t\t\tthis.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId) !== \"current\"\n\t\t) {\n\t\t\tthrow new Error(\"RLM capacity admission changed while resolving its family\");\n\t\t}\n\t\treturn success(command.id, command.type, this.rlmCapacity.reserve(owner, operation.reservationId, family));\n\t}\n\n\tprivate async handleCommand(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: DaemonCommand,\n\t\tcancellationAdmission?: SupervisorPromptAdmission,\n\t): Promise<DaemonResponse | undefined> {\n\t\tswitch (command.type) {\n\t\t\tcase \"cancel_prompt_admission\": {\n\t\t\t\tconst admission =\n\t\t\t\t\tcancellationAdmission ?? this.getPromptAdmission(client, command.activeSessionId, command.admissionId);\n\t\t\t\tif (!admission) return success(command.id, command.type, { status: \"unknown\" as const });\n\t\t\t\tif (admission.status === \"owned\") return success(command.id, command.type, { status: \"owned\" as const });\n\t\t\t\t// A definitive cancellation never downgrades to unknown/waiting.\n\t\t\t\tif (admission.status === \"cancelled\") {\n\t\t\t\t\treturn success(command.id, command.type, { status: \"cancelled\" as const });\n\t\t\t\t}\n\t\t\t\tif (!admission.worker || !admission.workerActiveSessionId) {\n\t\t\t\t\tadmission.status = \"cancelled\";\n\t\t\t\t\tadmission.controller.abort();\n\t\t\t\t\treturn success(command.id, command.type, { status: \"cancelled\" as const });\n\t\t\t\t}\n\t\t\t\tconst response = await this.forwardToWorker(admission.worker, {\n\t\t\t\t\t...command,\n\t\t\t\t\tactiveSessionId: admission.workerActiveSessionId,\n\t\t\t\t\tadmissionId: admission.workerAdmissionId,\n\t\t\t\t});\n\t\t\t\tconst status =\n\t\t\t\t\tresponse.success && response.data && typeof response.data === \"object\" && \"status\" in response.data\n\t\t\t\t\t\t? (response.data as { status: \"cancelled\" | \"owned\" | \"unknown\" }).status\n\t\t\t\t\t\t: \"unknown\";\n\t\t\t\t// Re-read: a socket close may have cancelled during the round-trip (cast widens TS's pre-await narrowing).\n\t\t\t\tconst current = (admission as SupervisorPromptAdmission).status;\n\t\t\t\tif (status === \"owned\") admission.status = \"owned\";\n\t\t\t\telse if (status === \"cancelled\") admission.status = \"cancelled\";\n\t\t\t\telse if (current !== \"cancelled\") admission.status = \"waiting\";\n\t\t\t\treturn { ...response, id: command.id };\n\t\t\t}\n\t\t\tcase \"ack_result\":\n\t\t\t\tthis.commandJournal.acknowledge(client.id, command.commandId);\n\t\t\t\treturn undefined;\n\t\t\tcase \"list\":\n\t\t\t\treturn this.handleList(client, command);\n\t\t\tcase \"roster_subscribe\":\n\t\t\t\tclient.rosterSubscribed = true;\n\t\t\t\treturn success(command.id, command.type, { roster: this.rosterEntriesForClient() });\n\t\t\tcase \"roster_unsubscribe\":\n\t\t\t\tclient.rosterSubscribed = false;\n\t\t\t\tclient.rosterResyncPending = false;\n\t\t\t\treturn success(command.id, command.type);\n\t\t\tcase \"list_agent_peers\": {\n\t\t\t\tconst requester = [...this.workers.values()].find(\n\t\t\t\t\t(worker) => worker.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t);\n\t\t\t\tif (!requester) throw new Error(\"Worker authentication failed\");\n\t\t\t\tconst peers = [...this.workers.values()]\n\t\t\t\t\t.filter(\n\t\t\t\t\t\t(worker) =>\n\t\t\t\t\t\t\tworker !== requester &&\n\t\t\t\t\t\t\tthis.isLiveWorker(worker) &&\n\t\t\t\t\t\t\tworker.descriptor.lifecycle === \"ready\" &&\n\t\t\t\t\t\t\tworker.client !== undefined,\n\t\t\t\t\t)\n\t\t\t\t\t.flatMap((worker) => {\n\t\t\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\t\t\treturn root ? [this.agentPeerSummary(sessionSummaryFromRosterEntry(root))] : [];\n\t\t\t\t\t});\n\t\t\t\treturn success(command.id, command.type, { peers });\n\t\t\t}\n\t\t\tcase \"rlm_capacity\":\n\t\t\t\treturn this.handleRlmCapacity(command);\n\t\t\tcase \"rlm_ledger_mutate\": {\n\t\t\t\tconst worker = [...this.workers.values()].find(\n\t\t\t\t\t(candidate) => candidate.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t);\n\t\t\t\tif (\n\t\t\t\t\t!worker ||\n\t\t\t\t\t!command.workerInstanceId ||\n\t\t\t\t\tworker.descriptor.workerInstanceId !== command.workerInstanceId ||\n\t\t\t\t\tworker.compatibilityError ||\n\t\t\t\t\tthis.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId) !== \"current\"\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker authentication failed for RLM ledger mutation\");\n\t\t\t\t}\n\t\t\t\tconst owner = this.assertRlmJournalAdmission();\n\t\t\t\tconst scope = this.rlmJournalScope(worker.descriptor.sessionDir ?? this.defaultSessionConfig.sessionDir);\n\t\t\t\tif (canonicalSessionPath(scope.journalPath) !== owner.ledgerPath) {\n\t\t\t\t\tthrow new Error(\"Worker RLM ledger family is not owned by this supervisor\");\n\t\t\t\t}\n\t\t\t\tawait this.rlmSpawnLedger().mutate(command.mutation);\n\t\t\t\treturn success(command.id, command.type);\n\t\t\t}\n\t\t\tcase \"get_direct_worker_transport\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tif (match.worker.descriptor.ownerClientId !== undefined) {\n\t\t\t\t\tthrow new Error(\"Direct transport is unavailable for client-owned workers\");\n\t\t\t\t}\n\t\t\t\tconst ticket = await this.issuePeerTransport(match.worker, match.summary);\n\t\t\t\treturn success(command.id, command.type, ticket);\n\t\t\t}\n\t\t\tcase \"list_saved_sessions\":\n\t\t\t\treturn this.handleSavedSessionList(client, command);\n\t\t\tcase \"create\": {\n\t\t\t\tconst worker = await this.createOrReuseWorker(this.protocolClientId(client), command);\n\t\t\t\tconst requestedSummary = command.sessionPath\n\t\t\t\t\t? this.findSummaryInWorker(worker, command.sessionPath)\n\t\t\t\t\t: undefined;\n\t\t\t\tif (\n\t\t\t\t\trequestedSummary &&\n\t\t\t\t\t(requestedSummary.activeSessionId ?? requestedSummary.id) !== worker.descriptor.rootActiveSessionId\n\t\t\t\t) {\n\t\t\t\t\t// A create forwarded to a recovering worker still surfaces an opaque lifecycle error.\n\t\t\t\t\tconst response = await this.forwardToWorker(worker, withoutSupervisorCreateFields(command));\n\t\t\t\t\tif (response.success && isSessionSummary(response.data)) {\n\t\t\t\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(response.data), worker);\n\t\t\t\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t\t\t\t}\n\t\t\t\t\treturn responseWithId(response, command.id);\n\t\t\t\t}\n\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\tif (!root) {\n\t\t\t\t\tthrow new Error(\"Session worker started without a root session\");\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"create\", this.publicSummary(worker, sessionSummaryFromRosterEntry(root)));\n\t\t\t}\n\t\t\tcase \"attach\": {\n\t\t\t\tconst attached = await this.attachClient(client, command);\n\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\tconst transcript = attached.transcript;\n\t\t\t\t\tif (!transcript) {\n\t\t\t\t\t\tthrow new Error(\"Session worker did not provide a snapshot transcript\");\n\t\t\t\t\t}\n\t\t\t\t\tconst streamedResult = this.createStreamedAttachResult(attached.result, transcript);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tthis.write(client, success(command.id, \"attach\", streamedResult));\n\t\t\t\t\t\tvoid this.streamSnapshot(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\t\tstreamedResult,\n\t\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\t\t\"attach\",\n\t\t\t\t\t\t\tattached.releaseTranscript,\n\t\t\t\t\t\t).catch((error) =>\n\t\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t\t`Failed to stream attach snapshot for ${streamedResult.activeSessionId}: ${String(error)}`,\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tattached.releaseTranscript?.();\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\treturn undefined;\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"attach\", attached.result);\n\t\t\t}\n\t\t\tcase \"reattach\": {\n\t\t\t\tconst target = await this.findWorkerForClient(client, command.targetActiveSessionId);\n\t\t\t\tconst targetActiveSessionId = target.summary.activeSessionId ?? target.summary.id;\n\t\t\t\tif (targetActiveSessionId === command.activeSessionId) {\n\t\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\t\t\tdetachingSessions?.delete(command.targetActiveSessionId);\n\t\t\t\t\tdetachingSessions?.delete(targetActiveSessionId);\n\t\t\t\t\treturn success(command.id, command.type, { cancelled: false });\n\t\t\t\t}\n\t\t\t\tconst targetWasAttached = client.attachedActiveSessionIds.has(targetActiveSessionId);\n\t\t\t\tconst releaseSnapshotReservation = this.reserveSnapshotStream(client, targetActiveSessionId);\n\t\t\t\tlet releaseTranscript: (() => void) | undefined;\n\t\t\t\tclient.attachedActiveSessionIds.add(targetActiveSessionId);\n\t\t\t\ttry {\n\t\t\t\t\tconst attached = await this.attachClient(client, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\t\tactiveSessionId: targetActiveSessionId,\n\t\t\t\t\t});\n\t\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\t\t\tdetachingSessions?.delete(command.targetActiveSessionId);\n\t\t\t\t\tdetachingSessions?.delete(targetActiveSessionId);\n\t\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\t\tconst transcript =\n\t\t\t\t\t\t\tattached.transcript ?? this.getOrCreateTranscriptCache(attached.worker, attached.result);\n\t\t\t\t\t\treleaseTranscript = attached.releaseTranscript;\n\t\t\t\t\t\tconst streamedResult = this.createStreamedAttachResult(attached.result, transcript);\n\t\t\t\t\t\tthis.write(client, success(command.id, command.type, streamedResult));\n\t\t\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\t\t\tconst streaming = this.streamSnapshot(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\t\tstreamedResult,\n\t\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\t\t\"replacement\",\n\t\t\t\t\t\t\treleaseTranscript,\n\t\t\t\t\t\t\treleaseSnapshotReservation,\n\t\t\t\t\t\t);\n\t\t\t\t\t\treleaseTranscript = undefined;\n\t\t\t\t\t\tvoid streaming.catch((error) =>\n\t\t\t\t\t\t\tthis.log(`Failed to stream reattach snapshot for ${targetActiveSessionId}: ${String(error)}`),\n\t\t\t\t\t\t);\n\t\t\t\t\t\treturn undefined;\n\t\t\t\t\t}\n\t\t\t\t\tthis.write(client, success(command.id, command.type, attached.result));\n\t\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\t\treleaseSnapshotReservation();\n\t\t\t\t\treturn undefined;\n\t\t\t\t} catch (error) {\n\t\t\t\t\treleaseTranscript?.();\n\t\t\t\t\tif (!targetWasAttached) {\n\t\t\t\t\t\tthis.detachClient(client, targetActiveSessionId);\n\t\t\t\t\t}\n\t\t\t\t\treleaseSnapshotReservation();\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t}\n\t\t\tcase \"acquire_session_input_pause\": {\n\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions.get(client);\n\t\t\t\tif (detachingSessions?.has(command.activeSessionId)) {\n\t\t\t\t\tthrow new Error(`Session is detaching: ${command.activeSessionId}`);\n\t\t\t\t}\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst activeSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\t\t\tif (detachingSessions?.has(command.activeSessionId) || detachingSessions?.has(activeSessionId)) {\n\t\t\t\t\tthrow new Error(`Session is detaching: ${command.activeSessionId}`);\n\t\t\t\t}\n\t\t\t\tconst ownerClientId = this.protocolClientId(client);\n\t\t\t\tconst connectionId = this.connectionIds.get(client);\n\t\t\t\tif (!connectionId) throw new Error(\"Daemon client connection identity is unavailable\");\n\t\t\t\tconst acquisitionEpoch = this.sessionInputPauseEpochs.get(client) ?? 0;\n\t\t\t\tconst existing = [...this.sessionInputPauses.values()].find(\n\t\t\t\t\t(entry) =>\n\t\t\t\t\t\tentry.owner === client &&\n\t\t\t\t\t\tentry.worker === match.worker &&\n\t\t\t\t\t\tentry.activeSessionId === activeSessionId &&\n\t\t\t\t\t\tentry.leaseKey === command.leaseKey &&\n\t\t\t\t\t\t!entry.releaseTask,\n\t\t\t\t);\n\t\t\t\tif (existing) {\n\t\t\t\t\treturn success(command.id, command.type, { pauseId: existing.pauseId });\n\t\t\t\t}\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, {\n\t\t\t\t\t...command,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tleaseKey: JSON.stringify([connectionId, ownerClientId, command.leaseKey]),\n\t\t\t\t});\n\t\t\t\tif (!response.success) return response;\n\t\t\t\tconst pauseId = (response.data as { pauseId?: unknown } | undefined)?.pauseId;\n\t\t\t\tif (typeof pauseId !== \"string\") throw new Error(\"Worker returned an invalid session input pause id\");\n\t\t\t\tif (!this.clients.has(client) || (this.sessionInputPauseEpochs.get(client) ?? 0) !== acquisitionEpoch) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst release = await this.forwardToWorker(match.worker, {\n\t\t\t\t\t\t\tid: randomUUID(),\n\t\t\t\t\t\t\ttype: \"release_session_input_pause\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tpauseId,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tif (!release.success) throw new Error(release.error);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tmatch.worker.client?.close();\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tthrow new Error(\"Session input pause acquisition was invalidated before completion\");\n\t\t\t\t}\n\t\t\t\tthis.sessionInputPauses.set(pauseId, {\n\t\t\t\t\towner: client,\n\t\t\t\t\tworker: match.worker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\trequestedActiveSessionId: command.activeSessionId,\n\t\t\t\t\tleaseKey: command.leaseKey,\n\t\t\t\t\tpauseId,\n\t\t\t\t});\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"release_session_input_pause\": {\n\t\t\t\tconst entry = this.sessionInputPauses.get(command.pauseId);\n\t\t\t\tif (!entry) return success(command.id, command.type);\n\t\t\t\tif (entry.owner !== client) {\n\t\t\t\t\tthrow new Error(`Session input pause is owned by another client: ${command.pauseId}`);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\tcommand.activeSessionId !== entry.activeSessionId &&\n\t\t\t\t\tcommand.activeSessionId !== entry.requestedActiveSessionId\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(`Session input pause belongs to another session: ${command.pauseId}`);\n\t\t\t\t}\n\t\t\t\tconst releaseTask =\n\t\t\t\t\tentry.releaseTask ??\n\t\t\t\t\tthis.forwardToWorker(entry.worker, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\tactiveSessionId: entry.activeSessionId,\n\t\t\t\t\t});\n\t\t\t\tentry.releaseTask = releaseTask;\n\t\t\t\ttry {\n\t\t\t\t\tconst response = await releaseTask;\n\t\t\t\t\tif (response.success && this.sessionInputPauses.get(command.pauseId) === entry) {\n\t\t\t\t\t\tthis.sessionInputPauses.delete(command.pauseId);\n\t\t\t\t\t}\n\t\t\t\t\treturn response;\n\t\t\t\t} finally {\n\t\t\t\t\tif (this.sessionInputPauses.get(command.pauseId) === entry && entry.releaseTask === releaseTask) {\n\t\t\t\t\t\tentry.releaseTask = undefined;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tcase \"detach\": {\n\t\t\t\tconst detachingSessions = this.detachingInputPauseSessions.get(client) ?? new Set<string>();\n\t\t\t\tthis.detachingInputPauseSessions.set(client, detachingSessions);\n\t\t\t\tif (command.activeSessionId) detachingSessions.add(command.activeSessionId);\n\t\t\t\telse for (const activeSessionId of client.attachedActiveSessionIds) detachingSessions.add(activeSessionId);\n\t\t\t\tthis.sessionInputPauseEpochs.set(client, (this.sessionInputPauseEpochs.get(client) ?? 0) + 1);\n\t\t\t\tthis.detachClient(client, command.activeSessionId);\n\t\t\t\tawait this.releaseClientSessionInputPauses(client, command.activeSessionId, true);\n\t\t\t\treturn success(command.id, \"detach\");\n\t\t\t}\n\t\t\tcase \"complete_owned_session\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tif (match.worker.descriptor.ownerClientId !== this.protocolClientId(client)) {\n\t\t\t\t\tthrow new Error(\"Session is not owned by this client\");\n\t\t\t\t}\n\t\t\t\tif (match.worker.ownerCleanupTimer) {\n\t\t\t\t\tclearTimeout(match.worker.ownerCleanupTimer);\n\t\t\t\t\tmatch.worker.ownerCleanupTimer = undefined;\n\t\t\t\t}\n\t\t\t\tawait this.stopWorker(match.worker, true);\n\t\t\t\treturn success(command.id, command.type);\n\t\t\t}\n\t\t\tcase \"promote_owned_session\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\treturn success(command.id, command.type, this.publicSummary(match.worker, match.summary));\n\t\t\t}\n\t\t\tcase \"retry_worker\": {\n\t\t\t\tconst direct = [...this.workers.values()].find(\n\t\t\t\t\t(worker) =>\n\t\t\t\t\t\tworker.descriptor.rootActiveSessionId === command.activeSessionId ||\n\t\t\t\t\t\tworker.descriptor.rootSessionId === command.activeSessionId,\n\t\t\t\t);\n\t\t\t\tconst worker = direct ?? (await this.findWorkerForClient(client, command.activeSessionId)).worker;\n\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\tif ((this.workerStopCounts?.get(worker) ?? 0) > 0) {\n\t\t\t\t\tthrow new Error(\"Session worker is stopping; retry after it finishes\");\n\t\t\t\t}\n\t\t\t\tworker.intentionalStop = false;\n\t\t\t\tworker.descriptor.stopRequestedAt = undefined;\n\t\t\t\tworker.descriptor.archiveOnStop = undefined;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tawait this.recoverWorker(worker);\n\t\t\t\tif (this.workers.get(worker.descriptor.workerId)?.descriptor.lifecycle !== \"ready\") {\n\t\t\t\t\tthrow new Error(worker.descriptor.lastError ?? \"Session worker recovery failed\");\n\t\t\t\t}\n\t\t\t\tconst summary = worker.summaries.get(worker.descriptor.rootActiveSessionId);\n\t\t\t\treturn success(command.id, command.type, summary ? this.publicSummary(worker, summary) : undefined);\n\t\t\t}\n\t\t\tcase \"restart\":\n\t\t\t\tsetImmediate(() => void this.shutdown(0, false, true, false, \"update\"));\n\t\t\t\treturn success(command.id, command.type);\n\t\t\tcase \"shutdown\":\n\t\t\t\tsetImmediate(() => void this.shutdown(0, true, false, command.force === true, \"shutdown\"));\n\t\t\t\treturn success(command.id, \"shutdown\");\n\t\t\tcase \"prepare_update_restart\": {\n\t\t\t\tconst manifest = await this.prepareUpdateRestart();\n\t\t\t\treturn success(command.id, \"prepare_update_restart\", manifest);\n\t\t\t}\n\t\t\tcase \"agent_messages_status\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst first = [...this.workers.values()].find((worker) => this.isLiveWorker(worker) && worker.client);\n\t\t\t\tif (!first) {\n\t\t\t\t\treturn success(command.id, command.type, { paused: false, limits: {} });\n\t\t\t\t}\n\t\t\t\treturn this.forwardToWorker(first, command);\n\t\t\t}\n\t\t\tcase \"agent_messages_pause\":\n\t\t\tcase \"agent_messages_resume\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst responses = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter((worker) => this.isLiveWorker(worker) && worker.client)\n\t\t\t\t\t\t.map((worker) => this.forwardToWorker(worker, command)),\n\t\t\t\t);\n\t\t\t\tconst failed = responses.find((response) => !response.success);\n\t\t\t\treturn failed ?? success(command.id, command.type, responses.find((response) => response.success)?.data);\n\t\t\t}\n\t\t\tcase \"cron_list\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst jobs = new Map<string, AgentCronJob>();\n\t\t\t\tconst responses = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map((worker) =>\n\t\t\t\t\t\t\tthis.forwardToWorker(worker, command, 5000).catch((error: unknown) =>\n\t\t\t\t\t\t\t\tfailure(command.id, command.type, error, serializeDaemonError(error)),\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t),\n\t\t\t\t);\n\t\t\t\tfor (const response of responses) {\n\t\t\t\t\tif (!response.success) {\n\t\t\t\t\t\tthis.log(`Could not list scheduled jobs from a worker: ${response.error}`);\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tfor (const job of cronJobsFromResponse(response)) {\n\t\t\t\t\t\tjobs.set(job.id, job);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tfor (const { job } of await this.collectPassiveScheduledJobs(command.includeInactive === true)) {\n\t\t\t\t\tif (!jobs.has(job.id)) jobs.set(job.id, job);\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"cron_list\", { jobs: sortCronJobs([...jobs.values()]) });\n\t\t\t}\n\t\t\tcase \"heartbeats_list\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst workers = [...this.workers.values()].filter(\n\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.descriptor.lifecycle !== \"failed\",\n\t\t\t\t);\n\t\t\t\tconst heartbeats = new Map<string, AgentConnectionHeartbeat>();\n\t\t\t\tconst snapshots: Array<{ heartbeats?: AgentConnectionHeartbeat[]; response?: DaemonResponse }> =\n\t\t\t\t\tawait Promise.all(\n\t\t\t\t\t\tworkers.map(async (worker) => {\n\t\t\t\t\t\t\tif (worker.client && worker.descriptor.lifecycle === \"ready\") {\n\t\t\t\t\t\t\t\tconst response = await this.forwardToWorker(worker, command, 5000).catch((error: unknown) =>\n\t\t\t\t\t\t\t\t\tfailure(command.id, command.type, error, serializeDaemonError(error)),\n\t\t\t\t\t\t\t\t);\n\t\t\t\t\t\t\t\tif (response.success) {\n\t\t\t\t\t\t\t\t\tconst snapshot = heartbeatsFromResponse(response);\n\t\t\t\t\t\t\t\t\tworker.heartbeatSnapshot = snapshot;\n\t\t\t\t\t\t\t\t\tworker.heartbeatSnapshotStale = false;\n\t\t\t\t\t\t\t\t\treturn { heartbeats: snapshot };\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\tthis.log(`Could not list heartbeats from a worker: ${response.error}`);\n\t\t\t\t\t\t\t\tif (worker.heartbeatSnapshot === undefined || worker.heartbeatSnapshotStale === true) {\n\t\t\t\t\t\t\t\t\treturn { response };\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (worker.heartbeatSnapshot !== undefined && worker.heartbeatSnapshotStale !== true) {\n\t\t\t\t\t\t\t\treturn { heartbeats: worker.heartbeatSnapshot };\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tconst state =\n\t\t\t\t\t\t\t\tworker.descriptor.lifecycle === \"ready\" ? \"disconnected\" : worker.descriptor.lifecycle;\n\t\t\t\t\t\t\tconst error = new Error(`Cannot list heartbeats while session worker is ${state}`);\n\t\t\t\t\t\t\treturn { response: failure(command.id, command.type, error, serializeDaemonError(error)) };\n\t\t\t\t\t\t}),\n\t\t\t\t\t);\n\t\t\t\tconst failed = snapshots.find((snapshot) => snapshot.response)?.response;\n\t\t\t\tif (failed) {\n\t\t\t\t\treturn failed;\n\t\t\t\t}\n\t\t\t\tfor (const snapshot of snapshots) {\n\t\t\t\t\tfor (const heartbeat of snapshot.heartbeats ?? []) {\n\t\t\t\t\t\theartbeats.set(heartbeat.job.id, heartbeat);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\t// Passivated sessions keep their armed heartbeats; no worker can list them.\n\t\t\t\tfor (const { job, info } of await this.collectPassiveScheduledJobs()) {\n\t\t\t\t\tif (!isHeartbeatCronJob(job) || heartbeats.has(job.id)) continue;\n\t\t\t\t\theartbeats.set(job.id, {\n\t\t\t\t\t\tjob,\n\t\t\t\t\t\t...(info.name !== undefined ? { sessionName: info.name } : {}),\n\t\t\t\t\t\t...(info.firstMessage !== undefined ? { firstMessage: info.firstMessage } : {}),\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\treturn success(command.id, \"heartbeats_list\", { heartbeats: [...heartbeats.values()] });\n\t\t\t}\n\t\t\tcase \"heartbeat_manage\": {\n\t\t\t\tconst cachedWorker = [...this.workers.values()].find((worker) =>\n\t\t\t\t\tworker.heartbeatSnapshot?.some(\n\t\t\t\t\t\t(heartbeat) =>\n\t\t\t\t\t\t\theartbeat.job.id === command.jobId && heartbeat.job.activeSessionId === command.activeSessionId,\n\t\t\t\t\t),\n\t\t\t\t);\n\t\t\t\tif (!cachedWorker) {\n\t\t\t\t\t// Passive jobs are managed against their durable store; no wake just to flip a status.\n\t\t\t\t\tconst passive = (await this.collectPassiveScheduledJobs()).find(\n\t\t\t\t\t\t({ job }) => job.id === command.jobId && job.activeSessionId === command.activeSessionId,\n\t\t\t\t\t);\n\t\t\t\t\tif (passive) {\n\t\t\t\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\t\t\t\tstore.registerSessionArtifact(\n\t\t\t\t\t\t\tpassive.info.id,\n\t\t\t\t\t\t\tgetSessionArtifactPathForFile(resolve(passive.info.path), passive.info.id),\n\t\t\t\t\t\t);\n\t\t\t\t\t\tconst heartbeat = store.manageHeartbeat(command.activeSessionId, command.jobId, command.action);\n\t\t\t\t\t\tif (heartbeat) {\n\t\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\t\treturn success(command.id, \"heartbeat_manage\", { heartbeat });\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tconst worker = cachedWorker ?? (await this.findWorkerForClient(client, command.activeSessionId)).worker;\n\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(worker, command);\n\t\t\t\tif (\n\t\t\t\t\tresponse.success &&\n\t\t\t\t\tresponse.data &&\n\t\t\t\t\ttypeof response.data === \"object\" &&\n\t\t\t\t\t\"heartbeat\" in response.data\n\t\t\t\t) {\n\t\t\t\t\tconst job = (response.data as { heartbeat?: AgentCronJob }).heartbeat;\n\t\t\t\t\tif (job && worker.heartbeatSnapshot) {\n\t\t\t\t\t\tconst existing = worker.heartbeatSnapshot.find((heartbeat) => heartbeat.job.id === job.id);\n\t\t\t\t\t\tconst remaining = worker.heartbeatSnapshot.filter((heartbeat) => heartbeat.job.id !== job.id);\n\t\t\t\t\t\tworker.heartbeatSnapshot =\n\t\t\t\t\t\t\tjob.status === \"active\" || job.status === \"paused\"\n\t\t\t\t\t\t\t\t? [...remaining, existing ? { ...existing, job } : { job }]\n\t\t\t\t\t\t\t\t: remaining;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"cron_add\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, command);\n\t\t\t\tif (response.success && command.promoteOwnedSession) {\n\t\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"cron_cancel\": {\n\t\t\t\tif (command.activeSessionId) {\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t\t}\n\t\t\t\tconst listed = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map(async (worker) => ({\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tresponse: await this.forwardToWorker(\n\t\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\t\t{ type: \"cron_list\", includeInactive: true },\n\t\t\t\t\t\t\t\t5000,\n\t\t\t\t\t\t\t).catch(() => undefined),\n\t\t\t\t\t\t})),\n\t\t\t\t);\n\t\t\t\tfor (const candidate of listed) {\n\t\t\t\t\tif (\n\t\t\t\t\t\tcandidate.response?.success &&\n\t\t\t\t\t\tcronJobsFromResponse(candidate.response).some((job) => job.id === command.jobId)\n\t\t\t\t\t) {\n\t\t\t\t\t\treturn this.forwardToWorker(candidate.worker, command);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tconst passive = (await this.collectPassiveScheduledJobs()).find(({ job }) => job.id === command.jobId);\n\t\t\t\tif (passive) {\n\t\t\t\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\t\t\t\tstore.registerSessionArtifact(\n\t\t\t\t\t\tpassive.info.id,\n\t\t\t\t\t\tgetSessionArtifactPathForFile(resolve(passive.info.path), passive.info.id),\n\t\t\t\t\t);\n\t\t\t\t\tconst job = store.cancel(command.jobId);\n\t\t\t\t\tif (job) {\n\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\treturn success(command.id, \"cron_cancel\", { job });\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tthrow new Error(`No cron job found: ${command.jobId}`);\n\t\t\t}\n\t\t\tcase \"cron_resume\": {\n\t\t\t\tif (command.activeSessionId !== undefined) {\n\t\t\t\t\tconst entry = this.roster().byActiveSessionId(command.activeSessionId);\n\t\t\t\t\tconst worker = entry?.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\t\t\tif (!worker || entry?.queuedChild) {\n\t\t\t\t\t\tthrow new Error(\"Cron resume requires an already bound destination runtime\");\n\t\t\t\t\t}\n\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, worker, command.activeSessionId);\n\t\t\t\t\treturn this.forwardToWorker(worker, command);\n\t\t\t\t}\n\t\t\t\tconst listed = await Promise.all(\n\t\t\t\t\t[...this.workers.values()]\n\t\t\t\t\t\t.filter(\n\t\t\t\t\t\t\t(worker) => this.isLiveWorker(worker) && worker.client && worker.descriptor.lifecycle === \"ready\",\n\t\t\t\t\t\t)\n\t\t\t\t\t\t.map(async (worker) => ({\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tresponse: await this.forwardToWorker(\n\t\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\t\t{ type: \"cron_list\", includeInactive: true },\n\t\t\t\t\t\t\t\t5000,\n\t\t\t\t\t\t\t).catch(() => undefined),\n\t\t\t\t\t\t})),\n\t\t\t\t);\n\t\t\t\tfor (const candidate of listed) {\n\t\t\t\t\tconst job = candidate.response?.success\n\t\t\t\t\t\t? cronJobsFromResponse(candidate.response).find((job) => job.id === command.jobId)\n\t\t\t\t\t\t: undefined;\n\t\t\t\t\tif (job) {\n\t\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, candidate.worker, job.activeSessionId);\n\t\t\t\t\t\treturn this.forwardToWorker(candidate.worker, command);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tthrow new Error(`Cron resume requires an already bound destination runtime for job: ${command.jobId}`);\n\t\t\t}\n\t\t\tcase \"heartbeat_get\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t}\n\t\t\tcase \"heartbeat_set\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\tconst response = await this.forwardToWorker(match.worker, command);\n\t\t\t\tif (response.success && command.promoteOwnedSession) {\n\t\t\t\t\tawait this.promoteOwnedWorker(client, match.worker);\n\t\t\t\t}\n\t\t\t\treturn response;\n\t\t\t}\n\t\t\tcase \"heartbeat_update\": {\n\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\treturn this.forwardToWorker(match.worker, command);\n\t\t\t}\n\t\t\tcase \"rename_saved_session\": {\n\t\t\t\tconst target = await this.savedSessionNameReservationInput(command.sessionPath, command.name.trim());\n\t\t\t\treturn await this.withSessionNameReservation(target, async () => {\n\t\t\t\t\tawait this.assertSupervisorSavedSessionNameAvailable(command.sessionPath, target.name);\n\t\t\t\t\tif (!command.activeSessionId) {\n\t\t\t\t\t\tthis.assertRlmJournalAdmission();\n\t\t\t\t\t\tawait this.catalog.rename(command.sessionPath, command.name);\n\t\t\t\t\t\t// Third rename write point: an offline saved-session rename\n\t\t\t\t\t\t// changes the name the ledger carries for that child.\n\t\t\t\t\t\tawait this.rlmSpawnLedger().appendRenameByChildPath(command.sessionPath, target.name);\n\t\t\t\t\t\tconst entry = this.roster().bySessionFile(canonicalSessionPath(command.sessionPath));\n\t\t\t\t\t\tif (entry) {\n\t\t\t\t\t\t\tthis.writeRosterEntry({ ...entry, summary: { ...entry.summary, sessionName: target.name } });\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn success(command.id, command.type);\n\t\t\t\t\t}\n\t\t\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\t\t\treturn await this.forwardToWorker(match.worker, {\n\t\t\t\t\t\t...command,\n\t\t\t\t\t\tactiveSessionId: match.summary.activeSessionId ?? match.summary.id,\n\t\t\t\t\t});\n\t\t\t\t});\n\t\t\t}\n\t\t\tcase \"delete_saved_session\":\n\t\t\t\tif (!command.activeSessionId) {\n\t\t\t\t\tconst deletedPath = canonicalSessionPath(command.sessionPath);\n\t\t\t\t\tconst entry = this.roster().bySessionFile(deletedPath);\n\t\t\t\t\tif (entry?.summary.activeSessionId !== undefined) {\n\t\t\t\t\t\tthrow new Error(\"Cannot delete the currently active session\");\n\t\t\t\t\t}\n\t\t\t\t\tconst owner = this.findWorkerBySessionFile(command.sessionPath);\n\t\t\t\t\tif (owner) {\n\t\t\t\t\t\t// A client-owned worker's files are invisible to other clients: a foreign delete is an unknown target.\n\t\t\t\t\t\tthis.assertWorkerAccessibleToClient(client, owner, command.sessionPath);\n\t\t\t\t\t\tif (owner.client && !this.isWorkerStopping(owner)) {\n\t\t\t\t\t\t\treturn this.forwardToWorker(owner, command);\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!(await this.reclaimStaleWorkerRegistration(owner))) {\n\t\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t\t`Session worker is ${this.effectiveWorkerState(owner)}; retry the delete once it is reachable`,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tthis.assertRlmJournalAdmission();\n\t\t\t\t\tawait tombstoneSavedSessionDelete(this.rlmSpawnLedger(), command.sessionPath, entry?.summary);\n\t\t\t\t\tconst result = await this.catalog.delete(command.sessionPath);\n\t\t\t\t\tif (result.ok && entry && this.roster().get(entry.agentId) === entry) {\n\t\t\t\t\t\tthis.roster().delete(entry.agentId);\n\t\t\t\t\t}\n\t\t\t\t\treturn success(command.id, command.type, result);\n\t\t\t\t}\n\t\t\t\tbreak;\n\t\t}\n\n\t\tif (command.type === \"send_message\" || command.type === \"send_result\") {\n\t\t\t// agentOrigin without fromActiveSessionId is trusted only at the direct socket-client boundary.\n\t\t\tconst source = command.fromActiveSessionId\n\t\t\t\t? await this.findWorkerForClient(client, command.fromActiveSessionId)\n\t\t\t\t: undefined;\n\t\t\tlet target: WorkerMatch;\n\t\t\ttry {\n\t\t\t\ttarget = await this.findWorkerForClient(client, command.targetActiveSessionId);\n\t\t\t} catch (error) {\n\t\t\t\tif (!(error instanceof Error) || !error.message.startsWith(\"Unknown active session:\")) throw error;\n\t\t\t\tconst cwd = source?.summary.cwd ?? this.defaultSessionConfig.cwd ?? process.cwd();\n\t\t\t\tlet sessionPath: string;\n\t\t\t\ttry {\n\t\t\t\t\tsessionPath = await this.catalog.resolve(\n\t\t\t\t\t\tcommand.targetActiveSessionId,\n\t\t\t\t\t\tcwd,\n\t\t\t\t\t\tsource?.worker.descriptor.sessionDir ?? this.defaultSessionConfig.sessionDir,\n\t\t\t\t\t);\n\t\t\t\t} catch (catalogError) {\n\t\t\t\t\t// Preserve selector ambiguity so a2a senders can distinguish it from\n\t\t\t\t\t// the original unknown-active-session lookup failure.\n\t\t\t\t\tif (catalogError instanceof Error && catalogError.message.startsWith(\"Ambiguous session selector\")) {\n\t\t\t\t\t\tthrow catalogError;\n\t\t\t\t\t}\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t\tif (source && command.agentOrigin === true) {\n\t\t\t\t\tconst targetInfo = await readSessionInfo(sessionPath);\n\t\t\t\t\tif (!targetInfo) throw new Error(`Unknown active session: ${command.targetActiveSessionId}`);\n\t\t\t\t\tassertAgentFamilyReach(\n\t\t\t\t\t\tthis.familyCatalogEntry(source.summary),\n\t\t\t\t\t\tthis.familyCatalogEntry(summaryForInactiveSession(targetInfo)),\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\tconst worker = await this.createOrReuseWorker(this.protocolClientId(client), {\n\t\t\t\t\ttype: \"create\",\n\t\t\t\t\tsessionPath,\n\t\t\t\t\tcontinueRecent: false,\n\t\t\t\t});\n\t\t\t\tconst root = this.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId);\n\t\t\t\tconst summary =\n\t\t\t\t\tthis.findSummaryInWorker(worker, sessionPath) ??\n\t\t\t\t\t(root ? sessionSummaryFromRosterEntry(root) : undefined);\n\t\t\t\tif (!summary) throw new Error(\"Woken session worker has no target session\");\n\t\t\t\ttarget = { worker, summary };\n\t\t\t}\n\t\t\tconst targetActiveSessionId = target.summary.activeSessionId ?? target.summary.id;\n\t\t\tif (source && command.agentOrigin === true) {\n\t\t\t\tassertAgentFamilyReach(this.familyCatalogEntry(source.summary), this.familyCatalogEntry(target.summary));\n\t\t\t}\n\t\t\tif (source) {\n\t\t\t\tif ((source.summary.activeSessionId ?? source.summary.id) === targetActiveSessionId) {\n\t\t\t\t\tthrow new Error(\"Agent messaging cannot target the sending session\");\n\t\t\t\t}\n\t\t\t\tconst targetClient = this.requireAvailableWorkerClient(target.worker);\n\t\t\t\tconst response = await targetClient.requestWorker(\n\t\t\t\t\t{\n\t\t\t\t\t\t...(command.type === \"send_result\"\n\t\t\t\t\t\t\t? { type: \"worker_deliver_result\" as const, summary: command.summary, findings: command.findings }\n\t\t\t\t\t\t\t: { type: \"worker_deliver_message\" as const, message: command.message }),\n\t\t\t\t\t\ttargetActiveSessionId,\n\t\t\t\t\t\tsender: {\n\t\t\t\t\t\t\tactiveSessionId: source.summary.activeSessionId ?? source.summary.id,\n\t\t\t\t\t\t\tsessionId: source.summary.sessionId,\n\t\t\t\t\t\t\t...(source.summary.sessionName ? { sessionName: source.summary.sessionName } : {}),\n\t\t\t\t\t\t\truntimeKind: source.summary.runtimeKind ?? \"top-level\",\n\t\t\t\t\t\t\tclientId: client.id,\n\t\t\t\t\t\t},\n\t\t\t\t\t},\n\t\t\t\t\tWORKER_REQUEST_TIMEOUT_MS,\n\t\t\t\t);\n\t\t\t\treturn { ...response, id: command.id, command: command.type };\n\t\t\t}\n\t\t\treturn this.forwardToWorker(target.worker, { ...command, targetActiveSessionId });\n\t\t}\n\n\t\tif (!(\"activeSessionId\" in command) || typeof command.activeSessionId !== \"string\") {\n\t\t\tthrow new Error(`Supervisor cannot route daemon command: ${command.type}`);\n\t\t}\n\t\tconst admission =\n\t\t\t(command.type === \"prompt\" || command.type === \"prompt_and_wait\") && command.admissionId\n\t\t\t\t? this.getPromptAdmission(client, command.activeSessionId, command.admissionId)\n\t\t\t\t: undefined;\n\t\ttry {\n\t\t\tthrowIfAdmissionCancelled(admission);\n\t\t\tconst match = await waitForPromptAdmission(\n\t\t\t\tcommand.type === \"set_session_name\" && command.workerToken !== undefined\n\t\t\t\t\t? this.findWorker(\n\t\t\t\t\t\t\tcommand.activeSessionId,\n\t\t\t\t\t\t\t(worker) => worker.descriptor.authenticationToken === command.workerToken,\n\t\t\t\t\t\t)\n\t\t\t\t\t: this.findWorkerForClient(client, command.activeSessionId),\n\t\t\t\tadmission?.controller.signal,\n\t\t\t);\n\t\t\tthrowIfAdmissionCancelled(admission);\n\t\t\tconst resolvedCommand = {\n\t\t\t\t...command,\n\t\t\t\tactiveSessionId: match.summary.activeSessionId ?? match.summary.id,\n\t\t\t\t...(admission ? { admissionId: admission.workerAdmissionId } : {}),\n\t\t\t} as DaemonCommand;\n\t\t\tif (admission) {\n\t\t\t\tadmission.worker = match.worker;\n\t\t\t\tadmission.workerActiveSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\t\t}\n\t\t\tconst isRootKill =\n\t\t\t\tcommand.type === \"kill\" &&\n\t\t\t\t(match.summary.activeSessionId ?? match.summary.id) === match.worker.descriptor.rootActiveSessionId;\n\t\t\tif (!isRootKill) {\n\t\t\t\tconst forward = async () => {\n\t\t\t\t\tconst response = await this.forwardToWorker(match.worker, resolvedCommand);\n\t\t\t\t\tif (admission && response.success) admission.status = \"owned\";\n\t\t\t\t\treturn response;\n\t\t\t\t};\n\t\t\t\tif (command.type === \"rename\" || command.type === \"set_session_name\") {\n\t\t\t\t\tconst reservation = this.summaryNameReservationInput(match.summary, command.name.trim());\n\t\t\t\t\treturn await this.withSessionNameReservation(reservation, async () => {\n\t\t\t\t\t\tawait this.assertSupervisorSessionNameAvailable(match.summary, reservation.name);\n\t\t\t\t\t\treturn forward();\n\t\t\t\t\t});\n\t\t\t\t}\n\t\t\t\treturn await forward();\n\t\t\t}\n\t\t\tthis.persistWorkerStopTombstone(match.worker, true);\n\t\t\tconst releaseStopOwnership = this.acquireWorkerStopOwnership(match.worker);\n\t\t\tlet response: DaemonResponse;\n\t\t\ttry {\n\t\t\t\tresponse = await this.forwardToWorker(match.worker, resolvedCommand);\n\t\t\t} finally {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.stopWorker(match.worker, true, false, true);\n\t\t\t\t} finally {\n\t\t\t\t\treleaseStopOwnership();\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn response;\n\t\t} finally {\n\t\t\tif (admission) this.deletePromptAdmission(admission);\n\t\t}\n\t}\n\n\tprivate async handleList(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"list\" }>,\n\t): Promise<DaemonResponse> {\n\t\tconst active: SessionSummary[] = [];\n\t\tconst activeByFile = new Map<string, SessionSummary>();\n\t\tlet busyClientOwnedSessionCount = 0;\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild) continue;\n\t\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\tif (worker === undefined) continue;\n\t\t\tconst summary = this.publicSummary(worker, sessionSummaryFromRosterEntry(entry));\n\t\t\tif (this.isVisibleWorker(worker)) {\n\t\t\t\tactive.push(summary);\n\t\t\t\tif (summary.sessionFile) activeByFile.set(canonicalSessionPath(summary.sessionFile), summary);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (summary.sessionFile) activeByFile.set(canonicalSessionPath(summary.sessionFile), summary);\n\t\t\tif (isSessionSummaryBusy(summary)) busyClientOwnedSessionCount += 1;\n\t\t\tif (command.includeClientOwned === true && this.isWorkerAccessibleToClient(client, worker)) {\n\t\t\t\tactive.push(summary);\n\t\t\t}\n\t\t}\n\t\tconst data = {\n\t\t\tsessions: active,\n\t\t\t...(command.includeClientOwned ? { busyClientOwnedSessionCount } : {}),\n\t\t};\n\t\tif (!command.all) {\n\t\t\treturn success(command.id, \"list\", data);\n\t\t}\n\t\tconst sessionDir = command.sessionDir ?? this.defaultSessionConfig.sessionDir;\n\t\tconst scanned = await this.catalog.list(command.cwd ? resolve(command.cwd) : undefined, sessionDir);\n\t\tconst cwd = command.cwd ? resolve(command.cwd) : undefined;\n\t\tconst merged: SessionSummary[] = [];\n\t\tconst servedRows = new Set(active);\n\t\tconst mergedActiveFiles = new Set<string>();\n\t\tconst scannedFiles = new Set<string>();\n\t\tfor (const info of scanned) {\n\t\t\tconst file = canonicalSessionPath(info.path);\n\t\t\tscannedFiles.add(file);\n\t\t\tconst workerRow = activeByFile.get(file);\n\t\t\tif (workerRow && servedRows.has(workerRow)) {\n\t\t\t\tmerged.push(workerRow);\n\t\t\t\tmergedActiveFiles.add(file);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// The on-disk scan is public: an unserved (client-owned) worker row hides its live metadata only.\n\t\t\tmerged.push(summaryForInactiveSession(info));\n\t\t}\n\t\t// The boot seed covers registered workers' families only, so dead families (pure on-disk\n\t\t// history, no registered worker anywhere in the tree) are not roster-resident. Their\n\t\t// subagent rows are read from the spawn ledger on demand instead.\n\t\tlet spawnEdges: RlmLedgerEdge[] = [];\n\t\ttry {\n\t\t\tspawnEdges = await this.rlmSpawnLedger().liveEdges();\n\t\t} catch (error) {\n\t\t\tthis.log(`Could not list spawn-ledger sessions: ${String(error)}`);\n\t\t}\n\t\tconst spawnParents = new Map(spawnEdges.map((edge) => [canonicalSessionPath(edge.child), edge.parent]));\n\t\tconst offlineRows: AgentRosterEntry[] = [];\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild || entry.summary.activeSessionId !== undefined) continue;\n\t\t\tif (entry.workerId !== undefined && this.workers.has(entry.workerId)) continue;\n\t\t\tconst file = entry.summary.sessionFile ? canonicalSessionPath(entry.summary.sessionFile) : undefined;\n\t\t\tif (file === undefined || scannedFiles.has(file) || activeByFile.has(file)) continue;\n\t\t\tofflineRows.push(entry);\n\t\t}\n\t\tfor (const hydrated of await Promise.all(offlineRows.map((entry) => this.hydrateSeededEntry(entry)))) {\n\t\t\tconst summary = sessionSummaryFromRosterEntry(hydrated);\n\t\t\tif (cwd !== undefined && resolve(summary.cwd) !== cwd) continue;\n\t\t\tif (!this.matchesListSessionDir(summary, sessionDir, spawnParents)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\tconst unseededFiles = new Set<string>();\n\t\tfor (const edge of spawnEdges) {\n\t\t\tconst childPath = canonicalSessionPath(edge.child);\n\t\t\tif (scannedFiles.has(childPath) || activeByFile.has(childPath) || unseededFiles.has(childPath)) continue;\n\t\t\tif (this.roster().hasSessionFile(childPath)) continue;\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\tunseededFiles.add(childPath);\n\t\t\t// Hydrated one at a time, like the boot seed: a large dead-family ledger must not fan\n\t\t\t// out into one concurrent transcript read per child.\n\t\t\tconst hydrated = await this.hydratedSeedEntry(entry);\n\t\t\t// The same classification a roster write would have applied: these rows read \"inactive\".\n\t\t\tconst summary = sessionSummaryFromRosterEntry({\n\t\t\t\t...hydrated,\n\t\t\t\tstatus: classifySessionRosterStatus(hydrated.summary),\n\t\t\t});\n\t\t\tif (cwd !== undefined && resolve(summary.cwd) !== cwd) continue;\n\t\t\tif (!this.matchesListSessionDir(summary, sessionDir, spawnParents)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\tfor (const summary of active) {\n\t\t\tconst file = summary.sessionFile ? canonicalSessionPath(summary.sessionFile) : undefined;\n\t\t\tif (file !== undefined && mergedActiveFiles.has(file)) continue;\n\t\t\tmerged.push(summary);\n\t\t}\n\t\treturn success(command.id, \"list\", { ...data, sessions: merged });\n\t}\n\n\tprivate async hydratedSeedEntry<T extends WorkerRosterEntry>(entry: T): Promise<T> {\n\t\tconst info = entry.summary.sessionFile\n\t\t\t? await readSessionInfo(entry.summary.sessionFile).catch(() => undefined)\n\t\t\t: undefined;\n\t\tif (!info) return { ...entry, seededCwd: true as const };\n\t\tconst { seededCwd, ...rest } = entry;\n\t\treturn { ...rest, summary: { ...entry.summary, cwd: info.cwd } } as T;\n\t}\n\n\tprivate async hydrateSeededEntry(entry: AgentRosterEntry): Promise<AgentRosterEntry> {\n\t\tif (entry.seededCwd !== true || !entry.summary.sessionFile) return entry;\n\t\tconst hydrated = await this.hydratedSeedEntry(entry);\n\t\tif (hydrated.seededCwd === true) return entry;\n\t\tconst current = this.roster().get(entry.agentId);\n\t\tif (current !== entry) return current ?? entry;\n\t\treturn this.roster().write(hydrated, entry.workerId, entry.statusLabel);\n\t}\n\n\tprivate matchesListSessionDir(\n\t\tsummary: SessionSummary,\n\t\tsessionDir: string | undefined,\n\t\tspawnParents?: ReadonlyMap<string, string>,\n\t): boolean {\n\t\tif (sessionDir === undefined) return true;\n\t\tif (!summary.sessionFile) return false;\n\t\tlet file = resolve(summary.sessionFile);\n\t\tlet parentSessionPath = summary.parentSessionPath;\n\t\tconst visited = new Set<string>();\n\t\twhile (parentSessionPath !== undefined) {\n\t\t\tconst canonical = canonicalSessionPath(parentSessionPath);\n\t\t\tif (visited.has(canonical)) break;\n\t\t\tvisited.add(canonical);\n\t\t\tfile = resolve(parentSessionPath);\n\t\t\t// Dead-family ancestors are not roster-resident; the spawn ledger continues the walk.\n\t\t\tparentSessionPath =\n\t\t\t\tthis.roster().bySessionFile(canonical)?.summary.parentSessionPath ?? spawnParents?.get(canonical);\n\t\t}\n\t\treturn dirname(file) === resolve(sessionDir);\n\t}\n\n\tprivate async handleSavedSessionList(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"list_saved_sessions\" }>,\n\t): Promise<DaemonResponse> {\n\t\tconst pageQuery = captureSavedSessionPageQuery(command.page ?? {});\n\t\tconst scope = command.scope;\n\t\tlet cwd: string;\n\t\tlet sessionDir: string | undefined;\n\t\tif (\"activeSessionId\" in command) {\n\t\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\t\tcwd = match.summary.cwd;\n\t\t\tsessionDir = this.defaultSessionConfig.sessionDir;\n\t\t} else {\n\t\t\tcwd = resolve(command.cwd);\n\t\t\tsessionDir = command.sessionDir;\n\t\t}\n\t\tconst { agentDir, sessionsDir: ledgerSessionDir } = this.rlmJournalScope(sessionDir);\n\t\tconst page = await this.catalog.list(cwd, sessionDir, { page: pageQuery, scope, agentDir, ledgerSessionDir });\n\t\treturn success(command.id, \"list_saved_sessions\", page);\n\t}\n\n\tprivate async createOrReuseWorker(clientId: string, command: DaemonCreateCommand): Promise<ResidentWorker> {\n\t\tlet createCommand = command;\n\t\tif (command.name !== undefined) {\n\t\t\tconst normalizedName = command.name.trim();\n\t\t\tif (!normalizedName) {\n\t\t\t\tthrow new Error(\"Session name cannot be empty\");\n\t\t\t}\n\t\t\tcreateCommand = { ...command, name: normalizedName };\n\t\t}\n\t\tconst ownerClientId = command.lifecycle === \"client_owned\" ? clientId : undefined;\n\t\tif (command.sessionPath) {\n\t\t\tconst activeMatches = this.matchWorkers(command.sessionPath);\n\t\t\tif (\n\t\t\t\tactiveMatches.length === 1 &&\n\t\t\t\t!(await this.reclaimStaleWorkerRegistration(activeMatches[0]!.worker, command.launchEnv !== undefined))\n\t\t\t) {\n\t\t\t\treturn this.reuseWorkerForCreate(activeMatches[0]!.worker, ownerClientId, command.sessionPath);\n\t\t\t}\n\t\t\tif (activeMatches.length > 1) {\n\t\t\t\tthrow new Error(`Ambiguous active session \"${command.sessionPath}\"`);\n\t\t\t}\n\t\t\tconst config = mergeAgentSessionRuntimeConfig(this.defaultSessionConfig, command.config);\n\t\t\tconst sessionPath = looksLikeSessionPath(command.sessionPath)\n\t\t\t\t? resolve(command.sessionPath)\n\t\t\t\t: await this.catalog.resolve(command.sessionPath, config.cwd ?? process.cwd(), config.sessionDir);\n\t\t\tcreateCommand = { ...createCommand, sessionPath };\n\t\t}\n\t\tconst key = createCommand.sessionPath\n\t\t\t? canonicalSessionPath(createCommand.sessionPath)\n\t\t\t: `new:${command.id ? createCommandIdempotencyKey(clientId, command.id) : createActiveSessionId()}`;\n\t\tconst pending = this.openingWorkers.get(key);\n\t\tif (pending) {\n\t\t\treturn this.joinOpeningWorker(pending, ownerClientId, createCommand.sessionPath ?? key);\n\t\t}\n\t\tif (createCommand.sessionPath) {\n\t\t\tconst existing = this.findWorkerBySessionFile(createCommand.sessionPath);\n\t\t\tif (existing && !(await this.reclaimStaleWorkerRegistration(existing, command.launchEnv !== undefined))) {\n\t\t\t\treturn this.reuseWorkerForCreate(existing, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t\t// The reclaim await may have let a concurrent opener register; join it instead of double-launching.\n\t\t\tconst opened = this.openingWorkers.get(key);\n\t\t\tif (opened) {\n\t\t\t\treturn this.joinOpeningWorker(opened, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t}\n\t\tif (createCommand.sessionPath) {\n\t\t\tconst existing = this.findWorkerBySessionFile(createCommand.sessionPath);\n\t\t\tif (existing && !(await this.reclaimStaleWorkerRegistration(existing, command.launchEnv !== undefined))) {\n\t\t\t\treturn this.reuseWorkerForCreate(existing, ownerClientId, createCommand.sessionPath);\n\t\t\t}\n\t\t}\n\t\tconst opening = (async () => {\n\t\t\tif (!createCommand.name) return this.launchWorker(createCommand, undefined, ownerClientId);\n\t\t\tconst savedSiblings = createCommand.sessionPath ? await this.rlmLedgerSiblings(createCommand.sessionPath) : [];\n\t\t\tconst target = savedSiblings.find(\n\t\t\t\t(session) => canonicalSessionPath(session.path) === canonicalSessionPath(createCommand.sessionPath!),\n\t\t\t);\n\t\t\tconst targetSummary = target ? summaryForInactiveSession(target) : { sessionId: \"new-root\", rlmDepth: 0 };\n\t\t\tconst reservation = this.summaryNameReservationInput(targetSummary, createCommand.name);\n\t\t\treturn this.withSessionNameReservation(reservation, async () => {\n\t\t\t\tif (target?.parentSessionPath && (target.rlmDepth ?? 0) > 0) {\n\t\t\t\t\tthis.assertSavedSiblingNameAvailable(savedSiblings, target, createCommand.name!);\n\t\t\t\t} else {\n\t\t\t\t\tawait this.assertSupervisorSessionNameAvailable(targetSummary, createCommand.name!);\n\t\t\t\t}\n\t\t\t\treturn this.launchWorker(createCommand, undefined, ownerClientId);\n\t\t\t});\n\t\t})();\n\t\tthis.openingWorkers.set(key, opening);\n\t\ttry {\n\t\t\treturn await opening;\n\t\t} finally {\n\t\t\tif (this.openingWorkers.get(key) === opening) {\n\t\t\t\tthis.openingWorkers.delete(key);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async reuseWorkerForCreate(\n\t\tworker: ResidentWorker,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): Promise<ResidentWorker> {\n\t\tif (worker.descriptor.lifecycle === \"failed\") {\n\t\t\tthrow new Error(\n\t\t\t\t`Session \"${sessionPath}\" is registered to a failed worker that could not be safely reclaimed`,\n\t\t\t);\n\t\t}\n\t\tthis.assertWorkerCreateOwner(worker, ownerClientId, sessionPath);\n\t\tif (!this.isWorkerReadyForCreate(worker)) {\n\t\t\tif (worker.recovery) {\n\t\t\t\tawait worker.recovery;\n\t\t\t} else if (this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tawait this.recoverWorker(worker);\n\t\t\t}\n\t\t}\n\t\tconst current = this.workers.get(worker.descriptor.workerId);\n\t\tif (!current) {\n\t\t\tthrow new Error(`Session \"${sessionPath}\" worker recovery was interrupted; retry opening the session`);\n\t\t}\n\t\tthis.assertWorkerCreateOwner(current, ownerClientId, sessionPath);\n\t\tif (!this.isWorkerReadyForCreate(current)) {\n\t\t\tif (!this.workerHasRosterRoot(current)) {\n\t\t\t\tthrow new Error(\n\t\t\t\t\t`Session \"${sessionPath}\" worker is unavailable for reuse: assigned root session is missing`,\n\t\t\t\t);\n\t\t\t}\n\t\t\tconst detail = current.descriptor.lastError ? `: ${current.descriptor.lastError}` : \"\";\n\t\t\tthrow new Error(`Session \"${sessionPath}\" worker is ${this.effectiveWorkerState(current)}${detail}`);\n\t\t}\n\t\treturn current;\n\t}\n\n\tprivate async joinOpeningWorker(\n\t\tpending: Promise<ResidentWorker>,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): Promise<ResidentWorker> {\n\t\tconst worker = await pending;\n\t\tthis.assertWorkerCreateOwner(worker, ownerClientId, sessionPath);\n\t\treturn worker;\n\t}\n\n\tprivate assertWorkerCreateOwner(\n\t\tworker: ResidentWorker,\n\t\townerClientId: string | undefined,\n\t\tsessionPath: string,\n\t): void {\n\t\tif (worker.descriptor.ownerClientId !== ownerClientId) {\n\t\t\tthrow new SessionAlreadyActiveError(sessionPath, worker.descriptor.rootActiveSessionId);\n\t\t}\n\t}\n\n\tprivate isWorkerReadyForCreate(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tworker.descriptor.lifecycle === \"ready\" &&\n\t\t\tworker.client !== undefined &&\n\t\t\tthis.workerHasRosterRoot(worker) &&\n\t\t\t!this.isWorkerStopping(worker)\n\t\t);\n\t}\n\n\tprivate workerHasRosterRoot(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.roster().byActiveSessionId(worker.descriptor.rootActiveSessionId)?.workerId === worker.descriptor.workerId\n\t\t);\n\t}\n\n\t/**\n\t * A stopping worker whose process already died can strand its registration\n\t * (for example when the stop timed out and its finalization was interrupted\n\t * by a supervisor restart). Such a registration would block reopening the\n\t * saved transcript forever, so complete the interrupted stop and let the\n\t * caller launch a fresh worker for the saved session.\n\t */\n\tprivate async reclaimStaleWorkerRegistration(worker: ResidentWorker, freshCreate = false): Promise<boolean> {\n\t\tif (worker.client !== undefined || worker.recovery !== undefined) {\n\t\t\treturn false;\n\t\t}\n\t\tif (worker.descriptor.stopRequestedAt === undefined) {\n\t\t\tif (worker.descriptor.lifecycle !== \"failed\" || worker.descriptor.ownerClientId) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst identity = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\tif (identity === \"current\") {\n\t\t\t\tif (worker.compatibilityError || !freshCreate || !worker.descriptor.processStartId) return false;\n\t\t\t\tawait this.stopWorker(worker, true, true);\n\t\t\t\treturn true;\n\t\t\t}\n\t\t\tif (identity !== \"gone\" && identity !== \"replaced\") {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tworker.intentionalStop = true;\n\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t\treturn true;\n\t\t}\n\t\t// Fail fast before waiting on anything: only a confirmed-dead process is\n\t\t// reclaimable. A live, unknown, or still-stopping worker is left alone\n\t\t// and the caller reports the session as already active.\n\t\tconst identity = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\tif (identity !== \"gone\" && identity !== \"replaced\") {\n\t\t\treturn false;\n\t\t}\n\t\t// Single cleanup path: the background stop finalizer is identity-aware,\n\t\t// retrying, and single-flighted, so concurrent resumes share one stop.\n\t\t// The wait is bounded so a resume request always returns promptly.\n\t\tthis.scheduleWorkerStopFinalization(worker);\n\t\tconst finalization = worker.stopFinalization;\n\t\tif (finalization) {\n\t\t\tawait Promise.race([finalization.catch(() => undefined), unrefDelay(STALE_RECLAIM_WAIT_MS)]);\n\t\t}\n\t\tif (this.workers.get(worker.descriptor.workerId) === worker) {\n\t\t\t// The process is confirmed dead, so the registration must never be\n\t\t\t// reused; slow cleanup fails the resume honestly instead.\n\t\t\tthrow new Error(\n\t\t\t\t`Stopped session worker ${worker.descriptor.workerId} is still being cleaned up; retry shortly`,\n\t\t\t);\n\t\t}\n\t\tthis.log(`Reclaimed stale registration for stopped worker ${worker.descriptor.workerId}`);\n\t\treturn true;\n\t}\n\n\tprivate async promoteOwnedWorker(client: DaemonSocketClient, worker: ResidentWorker): Promise<void> {\n\t\tconst clientId = this.protocolClientId(client);\n\t\tif (worker.descriptor.ownerClientId === undefined && worker.promotedOwnerClientId === clientId) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.ownerClientId !== clientId) {\n\t\t\tthrow new Error(\"Session is not owned by this client\");\n\t\t}\n\t\tconst previousDescriptor = worker.descriptor;\n\t\tworker.descriptor = { ...previousDescriptor, ownerClientId: undefined };\n\t\ttry {\n\t\t\tthis.persistWorker(worker);\n\t\t} catch (error) {\n\t\t\tworker.descriptor = previousDescriptor;\n\t\t\tthrow error;\n\t\t}\n\t\tworker.promotedOwnerClientId = clientId;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tthis.roster().amend(entry.agentId, {});\n\t\t}\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tworker.launchEnv = undefined;\n\t\tworker.transientCreateCommand = undefined;\n\t}\n\n\tprivate describeWorkerSpawnFailure(error: Error): Error {\n\t\tconst errno = (error as NodeJS.ErrnoException).code;\n\t\tconst hint =\n\t\t\terrno === \"EMFILE\" || errno === \"ENFILE\"\n\t\t\t\t? ` (${this.workers.size} resident session workers are holding file descriptors; stop unused sessions or raise the open-file limit (ulimit -n))`\n\t\t\t\t: \"\";\n\t\treturn new Error(`Failed to spawn session worker: ${error.message}${hint}`);\n\t}\n\n\tprivate async launchWorker(\n\t\tcommand: DaemonCreateCommand,\n\t\texisting?: ResidentWorker,\n\t\townerClientId?: string,\n\t): Promise<ResidentWorker> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (existing && this.isWorkerRecoveryCancelled(existing)) {\n\t\t\tthrow new Error(`Session worker ${existing.descriptor.workerId} recovery was cancelled`);\n\t\t}\n\t\tconst recoveryStopRevision = existing?.stopRevision;\n\t\tconst launchEnv = command.launchEnv ?? existing?.launchEnv;\n\t\tconst createCommand: DaemonCreateCommand = {\n\t\t\t...withoutSupervisorCreateFields(command),\n\t\t\tconfig: mergeAgentSessionRuntimeConfig(this.defaultSessionConfig, command.config),\n\t\t};\n\t\tconst workerId = existing?.descriptor.workerId ?? createActiveSessionId();\n\t\tconst rootActiveSessionId = existing?.descriptor.rootActiveSessionId ?? createActiveSessionId();\n\t\tconst socketPath = existing?.descriptor.socketPath ?? workerSocketPath(this.socketPath, workerId);\n\t\tconst token = existing?.descriptor.authenticationToken ?? randomBytes(32).toString(\"base64url\");\n\t\t// Fresh per incarnation: peer transport grants must never survive a worker restart.\n\t\tconst workerInstanceId = randomUUID();\n\t\tconst now = new Date().toISOString();\n\t\tconst descriptorPath = existing?.descriptorPath ?? join(this.descriptorDir, `${workerId}.json`);\n\t\tconst recoveryJournalPath =\n\t\t\texisting?.descriptor.recoveryJournalPath ?? join(this.descriptorDir, `${workerId}.recovery.jsonl`);\n\t\tconst orphanProcessJournalPath =\n\t\t\texisting?.descriptor.orphanProcessJournalPath ?? join(this.descriptorDir, `${workerId}.orphans.jsonl`);\n\t\tconst launch = createCliSubprocessLaunchSpec([\"--mode\", \"daemon\", \"--daemon-socket\", socketPath]);\n\t\tconst workerEnvironment = createCliSubprocessEnv({\n\t\t\t...process.env,\n\t\t\t...launchEnv,\n\t\t\t[DAEMON_WORKER_ROLE_ENV]: \"1\",\n\t\t\t[DAEMON_WORKER_TOKEN_ENV]: token,\n\t\t\t[DAEMON_WORKER_INSTANCE_ID_ENV]: workerInstanceId,\n\t\t\t[DAEMON_WORKER_ACTIVE_SESSION_ID_ENV]: rootActiveSessionId,\n\t\t\t[DAEMON_WORKER_SUPERVISOR_SOCKET_ENV]: this.socketPath,\n\t\t\t[DAEMON_WORKER_RECOVERY_JOURNAL_ENV]: recoveryJournalPath,\n\t\t\t[DAEMON_WORKER_STARTUP_GATE_FD_ENV]: String(WORKER_STARTUP_GATE_FD),\n\t\t\t[ORPHAN_PROCESS_JOURNAL_ENV]: orphanProcessJournalPath,\n\t\t\t[SESSION_LEASES_ENABLED_ENV]: \"1\",\n\t\t\t[SESSION_LEASE_OWNER_ID_ENV]: rootActiveSessionId,\n\t\t});\n\t\tdelete workerEnvironment.BASE_CONTEXT_RLM_DEPTH;\n\t\tawait this.assertRecoveryAllowed();\n\t\tconst child: ChildProcess = spawn(launch.command, launch.args, {\n\t\t\tcwd: createCommand.config?.cwd ?? process.cwd(),\n\t\t\tdetached: true,\n\t\t\tenv: workerEnvironment,\n\t\t\tstdio: [\"ignore\", \"ignore\", \"pipe\", \"pipe\"],\n\t\t});\n\t\tconst detachWorkerStderr = child.stderr\n\t\t\t? attachJsonlLineReader(child.stderr, (line) => this.log(`Session worker ${workerId} stderr: ${line}`), {\n\t\t\t\t\tmaxLineLength: 64 * 1024,\n\t\t\t\t\tonLineOverflow: (prefix) => this.log(`Session worker ${workerId} stderr: ${prefix} [truncated]`),\n\t\t\t\t})\n\t\t\t: () => {};\n\t\tchild.once(\"close\", detachWorkerStderr);\n\t\tconst childClosed = new Promise<void>((resolveClose) => child.once(\"close\", () => resolveClose()));\n\t\tlet spawnFailure: Error | undefined;\n\t\tconst spawnSettled = new Promise<void>((resolveSpawn) => {\n\t\t\tchild.once(\"spawn\", () => resolveSpawn());\n\t\t\tchild.once(\"error\", (error) => {\n\t\t\t\tspawnFailure = error instanceof Error ? error : new Error(String(error));\n\t\t\t\tresolveSpawn();\n\t\t\t});\n\t\t});\n\t\tchild.on(\"error\", (error) => {\n\t\t\tthis.log(\n\t\t\t\t`Session worker ${workerId} process error: ${error instanceof Error ? error.message : String(error)}`,\n\t\t\t);\n\t\t});\n\t\t// A failed spawn (e.g. EMFILE) leaves child.stdio undefined.\n\t\tconst startupGate = child.stdio?.[WORKER_STARTUP_GATE_FD];\n\t\tconst previousDescriptor = existing?.descriptor;\n\t\tconst previousIntentionalStop = existing?.intentionalStop;\n\t\tlet descriptorAssigned = false;\n\t\tlet childPid: number;\n\t\tlet childProcessStartId: string | undefined;\n\t\tlet worker: ResidentWorker;\n\t\ttry {\n\t\t\tawait spawnSettled;\n\t\t\tif (spawnFailure) {\n\t\t\t\tthrow this.describeWorkerSpawnFailure(spawnFailure);\n\t\t\t}\n\t\t\tif (!child.pid) {\n\t\t\t\tthrow new Error(\"Failed to obtain daemon session worker pid\");\n\t\t\t}\n\t\t\tif (!(startupGate instanceof Writable)) {\n\t\t\t\tthrow new Error(\"Failed to create daemon session worker startup gate\");\n\t\t\t}\n\t\t\tchildPid = child.pid;\n\t\t\tchildProcessStartId = getProcessStartId(childPid);\n\t\t\tawait this.assertRecoveryAllowed();\n\n\t\t\tconst descriptor: DaemonWorkerDescriptor = {\n\t\t\t\tversion: 2,\n\t\t\t\tworkerId,\n\t\t\t\tpid: childPid,\n\t\t\t\t...(childProcessStartId ? { processStartId: childProcessStartId } : {}),\n\t\t\t\tsocketPath,\n\t\t\t\trecoveryJournalPath,\n\t\t\t\torphanProcessJournalPath,\n\t\t\t\tsupervisorSocketPath: this.socketPath,\n\t\t\t\tauthenticationToken: token,\n\t\t\t\tworkerInstanceId,\n\t\t\t\trootActiveSessionId,\n\t\t\t\townerClientId: existing?.descriptor.ownerClientId ?? ownerClientId,\n\t\t\t\tsessionDir: createCommand.config?.sessionDir,\n\t\t\t\tcreatedAt: existing?.descriptor.createdAt ?? now,\n\t\t\t\tupdatedAt: now,\n\t\t\t\tlifecycle: \"starting\",\n\t\t\t\tcreateCommand: durableDaemonCreateCommand(createCommand),\n\t\t\t\tconsecutiveFailures: existing?.descriptor.consecutiveFailures ?? 0,\n\t\t\t};\n\t\t\tworker = existing ?? {\n\t\t\t\tdescriptor,\n\t\t\t\tdescriptorPath,\n\t\t\t\tsummaries: new Map(),\n\t\t\t\tsnapshotCache: new Map(),\n\t\t\t\ttranscriptCaches: new Map(),\n\t\t\t\tsnapshotGenerations: new Map(),\n\t\t\t\tsnapshotLoads: new Map(),\n\t\t\t\tintentionalStop: false,\n\t\t\t\tstopRevision: 0,\n\t\t\t\tlaunchEnv,\n\t\t\t\ttransientCreateCommand: ownerClientId ? createCommand : undefined,\n\t\t\t};\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor = descriptor;\n\t\t\tworker.launchEnv = launchEnv;\n\t\t\tworker.transientCreateCommand = descriptor.ownerClientId ? createCommand : undefined;\n\t\t\tdescriptorAssigned = true;\n\t\t\tthis.persistWorker(worker);\n\t\t\tworker.intentionalStop = false;\n\t\t\tthis.workers.set(workerId, worker);\n\t\t} catch (error) {\n\t\t\tif (startupGate instanceof Writable) {\n\t\t\t\tstartupGate.destroy();\n\t\t\t}\n\t\t\tawait childClosed;\n\t\t\tchild.unref();\n\t\t\ttry {\n\t\t\t\trmSync(`${descriptorPath}.${process.pid}.tmp`, { force: true });\n\t\t\t} catch (cleanupError) {\n\t\t\t\tthis.reportCleanupFailure(`worker launch temp ${workerId}`, cleanupError);\n\t\t\t}\n\t\t\tif (existing && descriptorAssigned && previousDescriptor) {\n\t\t\t\ttry {\n\t\t\t\t\texisting.descriptor = previousDescriptor;\n\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\tthis.reportCleanupFailure(`worker launch descriptor ${workerId}`, cleanupError);\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\n\t\ttry {\n\t\t\ttry {\n\t\t\t\tawait commitWorkerStartupGate(startupGate);\n\t\t\t} catch (error) {\n\t\t\t\tstartupGate.destroy();\n\t\t\t\tawait childClosed;\n\t\t\t\tthrow error;\n\t\t\t} finally {\n\t\t\t\tchild.unref();\n\t\t\t}\n\t\t\tconst client = await this.connectWorker(worker, WORKER_CONNECT_TIMEOUT_MS);\n\t\t\tconst response = await client.request(withoutCommandId(createCommand), WORKER_REQUEST_TIMEOUT_MS);\n\t\t\tif (!response.success) {\n\t\t\t\tthrow deserializeDaemonError(response);\n\t\t\t}\n\t\t\tif (!isSessionSummary(response.data)) {\n\t\t\t\tthrow new Error(\"Session worker returned an invalid create response\");\n\t\t\t}\n\t\t\tconst summary = response.data;\n\t\t\tif ((summary.activeSessionId ?? summary.id) !== rootActiveSessionId) {\n\t\t\t\tthrow new Error(\"Session worker did not preserve its assigned active session id\");\n\t\t\t}\n\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(summary), worker);\n\t\t\tworker.descriptor.rootSessionId = summary.sessionId;\n\t\t\tworker.descriptor.sessionFile = summary.sessionFile;\n\t\t\tawait this.subscribeWorker(worker, rootActiveSessionId);\n\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\tif (existing && (this.isWorkerRecoveryCancelled(worker) || worker.stopRevision !== recoveryStopRevision)) {\n\t\t\t\tthrow new Error(`Session worker ${workerId} recovery was cancelled`);\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\tworker.descriptor.lastError = undefined;\n\t\t\tthis.persistWorker(worker);\n\t\t\tif (!worker.descriptor.ownerClientId) {\n\t\t\t\tworker.launchEnv = undefined;\n\t\t\t\tworker.transientCreateCommand = undefined;\n\t\t\t}\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\treturn worker;\n\t\t} catch (error) {\n\t\t\tif (isSupervisorGenerationStale(error)) {\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tif (isSupervisorShutdownAdmissionCancelled(error)) {\n\t\t\t\tlet rolledBack = false;\n\t\t\t\ttry {\n\t\t\t\t\tawait this.stopWorker(worker, existing === undefined, true, false, existing !== undefined, {\n\t\t\t\t\t\tchild,\n\t\t\t\t\t\tclosed: childClosed,\n\t\t\t\t\t});\n\t\t\t\t\trolledBack = true;\n\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\tthis.reportCleanupFailure(`cancelled worker launch ${workerId}`, cleanupError);\n\t\t\t\t}\n\t\t\t\tconst mappedWorker = this.workers.get(workerId);\n\t\t\t\tif (\n\t\t\t\t\trolledBack &&\n\t\t\t\t\texisting &&\n\t\t\t\t\tpreviousDescriptor &&\n\t\t\t\t\t!this.shuttingDown &&\n\t\t\t\t\texisting.stopRevision === recoveryStopRevision &&\n\t\t\t\t\texisting.descriptor.stopRequestedAt === undefined &&\n\t\t\t\t\t(mappedWorker === undefined || mappedWorker === existing)\n\t\t\t\t) {\n\t\t\t\t\texisting.descriptor = previousDescriptor;\n\t\t\t\t\texisting.intentionalStop = previousIntentionalStop ?? false;\n\t\t\t\t\tthis.workers.set(workerId, existing);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tthis.persistWorker(existing);\n\t\t\t\t\t} catch (cleanupError) {\n\t\t\t\t\t\tthis.reportCleanupFailure(`cancelled worker recovery ${workerId}`, cleanupError);\n\t\t\t\t\t}\n\t\t\t\t\tthis.deferWorkerRecovery(existing, error instanceof Error ? error : new Error(String(error)));\n\t\t\t\t}\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tconst shouldResumeRecovery =\n\t\t\t\texisting !== undefined &&\n\t\t\t\t!this.shuttingDown &&\n\t\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\t\tworker.stopRevision === recoveryStopRevision;\n\t\t\tawait this.stopWorker(worker, existing === undefined, true, false, existing !== undefined).catch((stopError) =>\n\t\t\t\tthis.log(`Could not stop failed worker ${workerId}: ${String(stopError)}`),\n\t\t\t);\n\t\t\tif (\n\t\t\t\tshouldResumeRecovery &&\n\t\t\t\t!this.shuttingDown &&\n\t\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\t\tworker.stopRevision === recoveryStopRevision\n\t\t\t) {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\tworker.intentionalStop = false;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.workers.set(workerId, worker);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async connectWorker(worker: ResidentWorker, timeoutMs: number): Promise<DaemonWorkerClient> {\n\t\tconst deadline = Date.now() + timeoutMs;\n\t\tlet lastError: unknown;\n\t\twhile (Date.now() < deadline) {\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tconst client = new DaemonWorkerClient(worker.descriptor.socketPath);\n\t\t\ttry {\n\t\t\t\tawait client.connect(Math.min(500, Math.max(50, deadline - Date.now())));\n\t\t\t\tawait client.waitForHello(1000);\n\t\t\t\t// Listen before authenticating: the worker flushes its roster snapshot right after auth succeeds.\n\t\t\t\tclient.onFrame((frame) => this.handleWorkerFrame(worker, frame, client));\n\t\t\t\tclient.onClose((error) => void this.handleWorkerClose(worker, client, error));\n\t\t\t\tworker.pendingClient = client;\n\t\t\t\ttry {\n\t\t\t\t\tconst authResponse = await client.authenticateWorker(\n\t\t\t\t\t\tworker.descriptor.authenticationToken,\n\t\t\t\t\t\t{\n\t\t\t\t\t\t\t...this.supervisorAuthenticationClaim(),\n\t\t\t\t\t\t\t...(worker.descriptor.workerInstanceId !== undefined\n\t\t\t\t\t\t\t\t? { workerInstanceId: worker.descriptor.workerInstanceId }\n\t\t\t\t\t\t\t\t: {}),\n\t\t\t\t\t\t},\n\t\t\t\t\t\t1000,\n\t\t\t\t\t);\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\tif (!workerAuthAdvertisesRoster(authResponse.data)) {\n\t\t\t\t\t\tthrow new PreRosterWorkerError(\"Session worker predates the roster protocol and must be restarted\");\n\t\t\t\t\t}\n\t\t\t\t\tawait this.restoreWorkerRlmCapacity(worker, client);\n\t\t\t\t\tworker.peerTransportCapable = workerAuthAdvertisesPeerTransport(authResponse.data);\n\t\t\t\t\tworker.lastFrameAt = Date.now();\n\t\t\t\t\tworker.client?.close();\n\t\t\t\t\tworker.client = client;\n\t\t\t\t\tworker.compatibilityError = undefined;\n\t\t\t\t\tworker.descriptor.lastError = undefined;\n\t\t\t\t\treturn client;\n\t\t\t\t} finally {\n\t\t\t\t\tif (worker.pendingClient === client) worker.pendingClient = undefined;\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tlastError = error;\n\t\t\t\tclient.close();\n\t\t\t\tif (\n\t\t\t\t\tisSupervisorRecoveryCancelled(error) ||\n\t\t\t\t\terror instanceof PreRosterWorkerError ||\n\t\t\t\t\terror instanceof DaemonWorkerAuthenticationError ||\n\t\t\t\t\terror instanceof DaemonWorkerCompatibilityError\n\t\t\t\t) {\n\t\t\t\t\tthrow error;\n\t\t\t\t}\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tthrow new DaemonWorkerProbeTimeoutError(`Timed out connecting to daemon session worker: ${String(lastError)}`);\n\t}\n\n\tprivate parkIncompatibleWorker(worker: ResidentWorker, error: DaemonWorkerCompatibilityError): void {\n\t\t// Compatibility refusal is not evidence that this process or its ownership is dead.\n\t\tworker.compatibilityError = error;\n\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\tworker.descriptor.lastError = error.message;\n\t\tconst client = worker.client;\n\t\tworker.client = undefined;\n\t\tclient?.close();\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tthis.persistWorker(worker);\n\t\tthis.markWorkerRosterEntries(worker, \"recovering\");\n\t\tthis.log(`Kept incompatible worker ${worker.descriptor.workerId}: ${error.message}`);\n\t}\n\n\tprivate async subscribeWorker(worker: ResidentWorker, activeSessionId: string): Promise<void> {\n\t\tif (!worker.client) {\n\t\t\tthrow new Error(\"Session worker is not connected\");\n\t\t}\n\t\tconst supportsExtensionUi = [...this.clients].some(\n\t\t\t(client) => client.attachedActiveSessionIds.has(activeSessionId) && client.supportsExtensionUi,\n\t\t);\n\t\tconst response = await worker.client.requestWorker({\n\t\t\ttype: \"worker_subscribe\",\n\t\t\tactiveSessionId,\n\t\t\tcapabilities: supportsExtensionUi\n\t\t\t\t? [\"attach_snapshot\", \"event_sequence\", \"extension_ui\", \"slim_attach\", \"chunked_snapshot\"]\n\t\t\t\t: [\"attach_snapshot\", \"event_sequence\", \"slim_attach\", \"chunked_snapshot\"],\n\t\t\tsupportsExtensionUi,\n\t\t});\n\t\tif (!response.success) {\n\t\t\tthrow new Error(response.error);\n\t\t}\n\t}\n\n\tprivate async adoptOrRecoverWorker(worker: ResidentWorker, recoveryReady?: Promise<void>): Promise<void> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (worker.descriptor.stopRequestedAt) {\n\t\t\ttry {\n\t\t\t\t// A descriptor persisted before identity tracking has no\n\t\t\t\t// processStartId, so stopWorker could neither signal the live\n\t\t\t\t// process nor let the finalizer escalate. Authenticating on the\n\t\t\t\t// worker's socket proves the pid still belongs to our worker, so\n\t\t\t\t// the start id observed while it was alive can be persisted (and\n\t\t\t\t// the connected client gives stopWorker its graceful IPC path).\n\t\t\t\tif (worker.descriptor.processStartId === undefined && isProcessAlive(worker.descriptor.pid)) {\n\t\t\t\t\tconst observedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.connectWorker(worker, 2000);\n\t\t\t\t\t\tif (observedProcessStartId) {\n\t\t\t\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\t}\n\t\t\t\t\t} catch {\n\t\t\t\t\t\t// Unverifiable identity stays untrusted; the stop below\n\t\t\t\t\t\t// still runs its graceful path and the finalizer keeps\n\t\t\t\t\t\t// waiting rather than signalling a possibly-recycled pid.\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tawait this.stopWorker(worker, true, true, worker.descriptor.archiveOnStop === true);\n\t\t\t\tthis.log(`Completed intentional stop for worker ${worker.descriptor.workerId} during supervisor adoption`);\n\t\t\t} catch (error) {\n\t\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.log(`Could not complete intentional stop for worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tlet observedProcessStartId: string | undefined;\n\t\ttry {\n\t\t\tif (!isProcessAlive(worker.descriptor.pid)) {\n\t\t\t\tthrow new Error(\"Session worker process is no longer running\");\n\t\t\t}\n\t\t\tobservedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\tawait this.connectWorker(worker, 2000);\n\t\t\tawait this.subscribeWorker(worker, worker.descriptor.rootActiveSessionId);\n\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\tif (worker.descriptor.processStartId === undefined && observedProcessStartId) {\n\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t}\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t} catch (error) {\n\t\t\tif (isSupervisorRecoveryCancelled(error)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tthis.log(`Could not adopt worker ${worker.descriptor.workerId}: ${String(error)}`);\n\t\t\tif (recoveryReady) {\n\t\t\t\t// Recovery can create workers that call back for capacity. Keep it single-flight,\n\t\t\t\t// but let adoption finish before those callbacks need the normal ready gate.\n\t\t\t\tvoid this.recoverWorker(worker, {\n\t\t\t\t\tready: recoveryReady,\n\t\t\t\t\t...(error instanceof PreRosterWorkerError && worker.descriptor.ownerClientId === undefined\n\t\t\t\t\t\t? { preRoster: { observedProcessStartId } }\n\t\t\t\t\t\t: {}),\n\t\t\t\t}).catch((recoveryError) =>\n\t\t\t\t\tthis.log(`Could not recover worker ${worker.descriptor.workerId}: ${String(recoveryError)}`),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// A client-owned worker's launch env lives only with its owner; recoverWorker parks it instead.\n\t\t\tif (error instanceof PreRosterWorkerError && worker.descriptor.ownerClientId === undefined) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.restartPreRosterWorker(worker, observedProcessStartId);\n\t\t\t\t\treturn;\n\t\t\t\t} catch (restartError) {\n\t\t\t\t\tif (isSupervisorRecoveryCancelled(restartError)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tthis.log(`Could not restart pre-roster worker ${worker.descriptor.workerId}: ${String(restartError)}`);\n\t\t\t\t}\n\t\t\t}\n\t\t\tconst identityNow = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\tif (isDaemonWorkerProbeTimeout(error) && (identityNow === \"current\" || identityNow === \"unknown\")) {\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tvoid this.recoverWorker(worker).catch((recoveryError) =>\n\t\t\t\t\tthis.log(`Could not recover worker ${worker.descriptor.workerId}: ${String(recoveryError)}`),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tawait this.recoverWorker(worker);\n\t\t}\n\t}\n\n\tprivate async restartPreRosterWorker(\n\t\tworker: ResidentWorker,\n\t\tobservedProcessStartId: string | undefined,\n\t): Promise<void> {\n\t\tconst stopRevision = worker.stopRevision;\n\t\tconst workerInstanceId = worker.descriptor.workerInstanceId;\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (\n\t\t\tthis.isWorkerRecoveryCancelled(worker) ||\n\t\t\tworker.stopRevision !== stopRevision ||\n\t\t\tworker.descriptor.workerInstanceId !== workerInstanceId\n\t\t)\n\t\t\treturn;\n\t\tif (worker.descriptor.processStartId === undefined && observedProcessStartId !== undefined) {\n\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t}\n\t\tconst identity = () => this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t// The one deliberate kill of a live worker: it authenticated as ours and predates the roster\n\t\t// protocol, so this identity-verified upgrade replaces it. No await between recheck and signal.\n\t\tif (identity() === \"current\") {\n\t\t\tsignalProcessGroupOrProcess(worker.descriptor.pid, \"SIGKILL\");\n\t\t\t// SIGKILL is uninterceptable; this wait only covers kernel teardown of the old process and socket.\n\t\t\tconst killDeadline = Date.now() + 1000;\n\t\t\twhile (identity() === \"current\" && Date.now() < killDeadline) {\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tconst finalIdentity = identity();\n\t\tif (finalIdentity !== \"gone\" && finalIdentity !== \"replaced\") {\n\t\t\t// A live or unverifiable survivor parks failed with no destructive cleanup: interruption\n\t\t\t// marking and orphan reaping must never run against a possibly-active worker.\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = `Pre-roster worker process ${worker.descriptor.pid} is still running and cannot be replaced safely`;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(`Kept pre-roster worker ${worker.descriptor.workerId} failed: ${worker.descriptor.lastError}`);\n\t\t\treturn;\n\t\t}\n\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tawait this.launchWorker(worker.descriptor.createCommand, worker, worker.descriptor.ownerClientId);\n\t}\n\n\tprivate async handleWorkerClose(worker: ResidentWorker, client: DaemonWorkerClient, error: Error): Promise<void> {\n\t\tif (worker.client !== client) {\n\t\t\treturn;\n\t\t}\n\t\tworker.client = undefined;\n\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker disconnected while input was paused\");\n\t\tconst interrupted = new Map<string, Set<string>>();\n\t\tfor (const [activeSessionId, generations] of worker.snapshotGenerations ?? []) {\n\t\t\tfor (const generation of generations.values()) {\n\t\t\t\tif (generation.incoming || !generation.transcript.complete) {\n\t\t\t\t\tconst snapshotIds = interrupted.get(activeSessionId) ?? new Set<string>();\n\t\t\t\t\tsnapshotIds.add(generation.transcript.snapshotId);\n\t\t\t\t\tinterrupted.set(activeSessionId, snapshotIds);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tfor (const [activeSessionId, transcript] of worker.transcriptCaches) {\n\t\t\tif (!transcript.complete) {\n\t\t\t\tconst snapshotIds = interrupted.get(activeSessionId) ?? new Set<string>();\n\t\t\t\tsnapshotIds.add(transcript.snapshotId);\n\t\t\t\tinterrupted.set(activeSessionId, snapshotIds);\n\t\t\t}\n\t\t}\n\t\tfor (const [activeSessionId, snapshotIds] of interrupted) {\n\t\t\tfor (const snapshotId of snapshotIds) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tnew Error(\"Session worker disconnected during snapshot transfer\"),\n\t\t\t\t\tfalse,\n\t\t\t\t\tsnapshotId,\n\t\t\t\t);\n\t\t\t}\n\t\t}\n\t\tif (this.shuttingDown || worker.intentionalStop) {\n\t\t\treturn;\n\t\t}\n\t\tthis.markWorkerRosterEntries(worker, \"recovering\");\n\t\ttry {\n\t\t\tawait this.assertRecoveryAllowed();\n\t\t} catch (recoveryError) {\n\t\t\tif (!isSupervisorGenerationStale(recoveryError)) {\n\t\t\t\tthis.deferWorkerRecovery(worker, error);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\tworker.descriptor.lastError = error.message;\n\t\tthis.persistWorker(worker);\n\t\tvoid this.recoverWorker(worker);\n\t}\n\n\tprivate isWorkerRecoveryEligible(worker: ResidentWorker): boolean {\n\t\treturn this.isWorkerRecoveryCandidate(worker) && worker.recovery === undefined;\n\t}\n\n\tprivate isWorkerRecoveryCandidate(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\t!this.shuttingDown &&\n\t\t\t!worker.intentionalStop &&\n\t\t\tworker.descriptor.stopRequestedAt === undefined &&\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.client === undefined\n\t\t);\n\t}\n\n\tprivate deferWorkerRecovery(worker: ResidentWorker, disconnectError: Error): void {\n\t\tif (worker.compatibilityError || worker.deferredRecovery) {\n\t\t\treturn;\n\t\t}\n\t\t// A live-but-silent worker must not probe forever: park it failed (user-visible through the\n\t\t// roster's failed status) and keep its process alive for a manual retry_worker.\n\t\tworker.deferredRecoveryRounds = (worker.deferredRecoveryRounds ?? 0) + 1;\n\t\tif (worker.deferredRecoveryRounds > MAX_DEFERRED_RECOVERY_ROUNDS) {\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = `Live session worker did not answer recovery probes for ${MAX_DEFERRED_RECOVERY_ROUNDS} rounds: ${disconnectError.message}`;\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(\n\t\t\t\t`Worker ${worker.descriptor.workerId} is unresponsive; parked failed after ${MAX_DEFERRED_RECOVERY_ROUNDS} probe rounds`,\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\t\tworker.deferredRecovery = this.resumeDeferredWorkerRecovery(worker, disconnectError).finally(() => {\n\t\t\tworker.deferredRecovery = undefined;\n\t\t});\n\t}\n\n\tprivate async resumeDeferredWorkerRecovery(worker: ResidentWorker, disconnectError: Error): Promise<void> {\n\t\twhile (true) {\n\t\t\tawait unrefDelay(DEFERRED_RECOVERY_RECHECK_MS);\n\t\t\tif (worker.compatibilityError || !this.isWorkerRecoveryCandidate(worker)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t} catch (error) {\n\t\t\t\tif (isSupervisorGenerationStale(error)) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (worker.compatibilityError || !this.isWorkerRecoveryCandidate(worker)) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!this.isWorkerRecoveryEligible(worker)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\tworker.descriptor.lastError = disconnectError.message;\n\t\t\tthis.persistWorker(worker);\n\t\t\tvoid this.recoverWorker(worker);\n\t\t\treturn;\n\t\t}\n\t}\n\n\tprivate failWorkerSnapshotCache(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\terror: Error,\n\t\tcloseWorkerChannel = false,\n\t\texpectedSnapshotId?: string,\n\t): void {\n\t\tconst generations = worker.snapshotGenerations?.get(activeSessionId);\n\t\tif (expectedSnapshotId) {\n\t\t\tconst generation = generations?.get(expectedSnapshotId);\n\t\t\tif (generation) {\n\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, error);\n\t\t\t} else {\n\t\t\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\t\t\tif (transcript?.snapshotId !== expectedSnapshotId) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttranscript.markFailed(error);\n\t\t\t\ttranscript.dispose();\n\t\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === expectedSnapshotId) {\n\t\t\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t\t\t}\n\t\t\t}\n\t\t} else {\n\t\t\tconst failedTranscripts = new Set<SnapshotTranscriptCache>();\n\t\t\tfor (const generation of [...(generations?.values() ?? [])]) {\n\t\t\t\tfailedTranscripts.add(generation.transcript);\n\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, error);\n\t\t\t}\n\t\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\t\tif (transcript && !failedTranscripts.has(transcript)) {\n\t\t\t\ttranscript.markFailed(error);\n\t\t\t\ttranscript.dispose();\n\t\t\t}\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t\tif (closeWorkerChannel) {\n\t\t\tconst client = worker.client;\n\t\t\tif (client) {\n\t\t\t\tthis.handleWorkerClose(worker, client, error);\n\t\t\t\tclient.close();\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate retireWorkerSnapshotCache(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\texpectedTranscript: SnapshotTranscriptCache,\n\t): void {\n\t\tif (worker.transcriptCaches.get(activeSessionId) === expectedTranscript) {\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t}\n\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === expectedTranscript.snapshotId) {\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, expectedTranscript.snapshotId);\n\t\tif (!generation) {\n\t\t\texpectedTranscript.dispose();\n\t\t\treturn;\n\t\t}\n\t\tgeneration.retired = true;\n\t\tthis.settleSnapshotDuplicateValidation(generation);\n\t\tif (generation.incoming) {\n\t\t\treturn;\n\t\t}\n\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\texpectedTranscript.dispose();\n\t}\n\n\tprivate snapshotGenerationsFor(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t): Map<string, SnapshotTranscriptGeneration> {\n\t\tworker.snapshotGenerations ??= new Map();\n\t\tlet generations = worker.snapshotGenerations.get(activeSessionId);\n\t\tif (!generations) {\n\t\t\tgenerations = new Map();\n\t\t\tworker.snapshotGenerations.set(activeSessionId, generations);\n\t\t}\n\t\treturn generations;\n\t}\n\n\tprivate snapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tsnapshotId: string,\n\t): SnapshotTranscriptGeneration | undefined {\n\t\treturn worker.snapshotGenerations?.get(activeSessionId)?.get(snapshotId);\n\t}\n\n\tprivate currentSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t): SnapshotTranscriptGeneration | undefined {\n\t\tworker.transcriptCaches ??= new Map();\n\t\tworker.snapshotCache ??= new Map();\n\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\tif (!transcript) {\n\t\t\treturn undefined;\n\t\t}\n\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\tlet generation = generations.get(transcript.snapshotId);\n\t\tif (generation) {\n\t\t\treturn generation;\n\t\t}\n\t\tconst result = worker.snapshotCache.get(activeSessionId);\n\t\tif (!result) {\n\t\t\treturn undefined;\n\t\t}\n\t\tgeneration = {\n\t\t\ttranscript,\n\t\t\tresult,\n\t\t\tincoming: false,\n\t\t\tretired: false,\n\t\t};\n\t\tgenerations.set(transcript.snapshotId, generation);\n\t\treturn generation;\n\t}\n\n\tprivate deleteSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tgeneration: SnapshotTranscriptGeneration,\n\t): void {\n\t\tconst generations = worker.snapshotGenerations?.get(activeSessionId);\n\t\tif (!generations) {\n\t\t\treturn;\n\t\t}\n\t\tif (generations.get(generation.transcript.snapshotId) === generation) {\n\t\t\tgenerations.delete(generation.transcript.snapshotId);\n\t\t}\n\t\tif (generations.size === 0) {\n\t\t\tworker.snapshotGenerations.delete(activeSessionId);\n\t\t}\n\t}\n\n\tprivate failSnapshotGeneration(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tgeneration: SnapshotTranscriptGeneration,\n\t\terror: Error,\n\t): void {\n\t\tthis.settleSnapshotDuplicateValidation(generation, error);\n\t\tgeneration.transcript.markFailed(error);\n\t\tgeneration.transcript.dispose();\n\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\tif (worker.transcriptCaches.get(activeSessionId) === generation.transcript) {\n\t\t\tworker.transcriptCaches.delete(activeSessionId);\n\t\t}\n\t\tif (worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id === generation.transcript.snapshotId) {\n\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t}\n\t}\n\n\tprivate createSnapshotDuplicateValidation(): SnapshotDuplicateValidation {\n\t\tlet resolve!: () => void;\n\t\tlet reject!: (error: Error) => void;\n\t\tconst promise = new Promise<void>((resolvePromise, rejectPromise) => {\n\t\t\tresolve = resolvePromise;\n\t\t\treject = rejectPromise;\n\t\t});\n\t\tvoid promise.catch(() => undefined);\n\t\treturn { promise, resolve, reject };\n\t}\n\n\tprivate settleSnapshotDuplicateValidation(generation: SnapshotTranscriptGeneration, error?: Error): void {\n\t\tconst validation = generation.validation;\n\t\tif (!validation) {\n\t\t\treturn;\n\t\t}\n\t\tgeneration.validation = undefined;\n\t\tif (error) {\n\t\t\tvalidation.reject(error);\n\t\t} else {\n\t\t\tvalidation.resolve();\n\t\t}\n\t}\n\n\tprivate async recoverWorker(\n\t\tworker: ResidentWorker,\n\t\tstartup?: { ready: Promise<void>; preRoster?: { observedProcessStartId?: string } },\n\t): Promise<void> {\n\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.ownerClientId && !worker.launchEnv && !isProcessAlive(worker.descriptor.pid)) {\n\t\t\tthis.releaseWorkerRlmCapacity(worker);\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tworker.descriptor.lastError = \"Waiting for the owning client to reconnect\";\n\t\t\tthis.persistWorker(worker);\n\t\t\treturn;\n\t\t}\n\t\tif (worker.recovery) {\n\t\t\treturn worker.recovery;\n\t\t}\n\t\tconst stopRevision = worker.stopRevision;\n\t\tconst workerInstanceId = worker.descriptor.workerInstanceId;\n\t\tworker.recovery = (async () => {\n\t\t\tif (startup) {\n\t\t\t\tawait startup.ready;\n\t\t\t\tif (\n\t\t\t\t\tthis.isWorkerRecoveryCancelled(worker) ||\n\t\t\t\t\tworker.stopRevision !== stopRevision ||\n\t\t\t\t\tworker.descriptor.workerInstanceId !== workerInstanceId\n\t\t\t\t)\n\t\t\t\t\treturn;\n\t\t\t\tif (startup.preRoster) {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.restartPreRosterWorker(worker, startup.preRoster.observedProcessStartId);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t} catch (restartError) {\n\t\t\t\t\t\tif (isSupervisorRecoveryCancelled(restartError)) return;\n\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t`Could not restart pre-roster worker ${worker.descriptor.workerId}: ${String(restartError)}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tlet keepProbingLiveWorker = false;\n\t\t\tfor (const retryDelay of WORKER_RETRY_DELAYS_MS) {\n\t\t\t\tawait delay(retryDelay);\n\t\t\t\tkeepProbingLiveWorker = false;\n\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\tconst identityNow = this.processIdentity(worker.descriptor.pid, worker.descriptor.processStartId);\n\t\t\t\t\tconst identityCompatible =\n\t\t\t\t\t\tidentityNow === \"current\" ||\n\t\t\t\t\t\t(identityNow === \"unknown\" && worker.descriptor.processStartId === undefined);\n\t\t\t\t\tif (identityCompatible) {\n\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\tawait this.connectWorker(worker, 1500);\n\t\t\t\t\t\t\tawait this.subscribeWorker(worker, worker.descriptor.rootActiveSessionId);\n\t\t\t\t\t\t\tawait this.refreshWorkerSummaries(worker, true);\n\t\t\t\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tif (worker.descriptor.processStartId === undefined) {\n\t\t\t\t\t\t\t\tconst observedProcessStartId = getProcessStartId(worker.descriptor.pid);\n\t\t\t\t\t\t\t\tif (observedProcessStartId) {\n\t\t\t\t\t\t\t\t\tworker.descriptor.processStartId = observedProcessStartId;\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t\t\tworker.descriptor.lifecycle = \"ready\";\n\t\t\t\t\t\t\tworker.descriptor.consecutiveFailures = 0;\n\t\t\t\t\t\t\tworker.deferredRecoveryRounds = 0;\n\t\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\t\tif (isSupervisorRecoveryCancelled(error) || error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t\t\tworker.client?.close();\n\t\t\t\t\t\t\tworker.client = undefined;\n\t\t\t\t\t\t\t// A worker with the same durable process identity may be load-slow.\n\t\t\t\t\t\t\t// Keep probing it instead of replacing live work after a timeout.\n\t\t\t\t\t\t\tkeepProbingLiveWorker = isDaemonWorkerProbeTimeout(error) || identityNow !== \"current\";\n\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t}\n\t\t\t\t\t}\n\t\t\t\t\tif (identityNow === \"unknown\") {\n\t\t\t\t\t\tkeepProbingLiveWorker = true;\n\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t`Cannot safely replace live session worker ${worker.descriptor.workerId} without a verified process identity`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t\tthis.releaseWorkerRlmCapacity(worker);\n\t\t\t\t\tconst recoveryCommand = worker.descriptor.ownerClientId ? worker.transientCreateCommand : undefined;\n\t\t\t\t\tif (!recoveryCommand || !worker.launchEnv) {\n\t\t\t\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\t\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\t\t\t\tworker.descriptor.lastError = \"Waiting for a client with fresh runtime context\";\n\t\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tawait this.recoverUncertainWorkerOperations(worker);\n\t\t\t\t\tif (this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tawait this.launchWorker(recoveryCommand, worker, worker.descriptor.ownerClientId);\n\t\t\t\t\treturn;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (isSupervisorRecoveryCancelled(error) || this.isWorkerRecoveryCancelled(worker)) {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t\t} catch {\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\tthis.parkIncompatibleWorker(worker, error);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\tworker.client?.close();\n\t\t\t\t\tworker.client = undefined;\n\t\t\t\t\tworker.descriptor.consecutiveFailures++;\n\t\t\t\t\tworker.descriptor.lastFailureAt = new Date().toISOString();\n\t\t\t\t\tworker.descriptor.lastError = error instanceof Error ? error.message : String(error);\n\t\t\t\t\tthis.persistWorker(worker);\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (keepProbingLiveWorker) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t\t} catch {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t\tthis.deferWorkerRecovery(\n\t\t\t\t\tworker,\n\t\t\t\t\tnew Error(worker.descriptor.lastError ?? \"Live session worker did not answer recovery probes\"),\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\t// Leak-over-kill: a live worker that keeps failing for non-timeout reasons parks failed with\n\t\t\t// its process intact. A verified-identity survivor is reclaimed by the next fresh create;\n\t\t\t// an unverifiable one waits for exit — killing a pid we cannot verify as ours is worse.\n\t\t\tworker.descriptor.lifecycle = \"failed\";\n\t\t\tthis.persistWorker(worker);\n\t\t\tthis.markWorkerRosterEntries(worker, \"failed\");\n\t\t\tthis.log(`Worker ${worker.descriptor.workerId} failed after three recovery attempts`);\n\t\t})().finally(() => {\n\t\t\tworker.recovery = undefined;\n\t\t});\n\t\treturn worker.recovery;\n\t}\n\n\tprivate isWorkerRecoveryCancelled(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.shuttingDown ||\n\t\t\tworker.intentionalStop ||\n\t\t\tworker.descriptor.stopRequestedAt !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t);\n\t}\n\n\tprivate isWorkerCleanupCancelled(worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tthis.shuttingDown ||\n\t\t\tworker.descriptor.stopRequestedAt !== undefined ||\n\t\t\tthis.workers.get(worker.descriptor.workerId) !== worker\n\t\t);\n\t}\n\n\tprivate async recoverUncertainWorkerOperations(worker: ResidentWorker): Promise<void> {\n\t\tawait this.assertRecoveryAllowed();\n\t\tconst journal = new WorkerRecoveryJournal(worker.descriptor.recoveryJournalPath);\n\t\tconst latest = journal.getLatest();\n\t\tconst uncertain = latest.filter((record) => record.busy);\n\t\tif (uncertain.length > 0) await this.catalog.start();\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before destructive cleanup\");\n\t\t}\n\n\t\tconst interruptedSessions = new Map<\n\t\t\tstring,\n\t\t\t{ activeSessionId: string; sessionFile: string; operations: Set<string> }\n\t\t>();\n\t\tfor (const record of uncertain) {\n\t\t\tconst sessionFile =\n\t\t\t\trecord.sessionFile ??\n\t\t\t\t(record.activeSessionId === worker.descriptor.rootActiveSessionId\n\t\t\t\t\t? worker.descriptor.sessionFile\n\t\t\t\t\t: undefined);\n\t\t\tif (!sessionFile) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst key = `${record.activeSessionId}\\0${sessionFile}`;\n\t\t\tlet interrupted = interruptedSessions.get(key);\n\t\t\tif (!interrupted) {\n\t\t\t\tinterrupted = { activeSessionId: record.activeSessionId, sessionFile, operations: new Set() };\n\t\t\t\tinterruptedSessions.set(key, interrupted);\n\t\t\t}\n\t\t\tinterrupted.operations.add(record.operation);\n\t\t}\n\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before interruption was recorded\");\n\t\t}\n\t\tawait Promise.all(\n\t\t\t[...interruptedSessions.values()].map((interrupted) =>\n\t\t\t\tthis.catalog.markInterrupted(interrupted.sessionFile, interrupted.activeSessionId, [\n\t\t\t\t\t...interrupted.operations,\n\t\t\t\t]),\n\t\t\t),\n\t\t);\n\t\tawait this.assertRecoveryAllowed();\n\t\tif (this.isWorkerCleanupCancelled(worker)) {\n\t\t\tthrow new SupervisorRecoveryCancelledError(\"Worker recovery was cancelled before process cleanup\");\n\t\t}\n\t\tconst orphanProcessJournalPath = worker.descriptor.orphanProcessJournalPath;\n\t\tif (orphanProcessJournalPath) {\n\t\t\ttry {\n\t\t\t\tconst orphans = readActiveOrphanProcesses(orphanProcessJournalPath, worker.descriptor.pid);\n\t\t\t\tlet reapFailed = false;\n\t\t\t\tlet retryIsSafe = true;\n\t\t\t\tfor (const orphan of orphans) {\n\t\t\t\t\tif (orphan.processStartId === undefined) retryIsSafe = false;\n\t\t\t\t\tif (!shouldReapOrphanProcess(orphan)) {\n\t\t\t\t\t\tcontinue;\n\t\t\t\t\t}\n\t\t\t\t\tif (!killOrphanProcess(orphan.pid)) reapFailed = true;\n\t\t\t\t}\n\t\t\t\tif (!reapFailed || !retryIsSafe) clearOrphanProcessJournal(orphanProcessJournalPath);\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Could not reap orphaned worker resources: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t\tif (uncertain.length === 0) {\n\t\t\treturn;\n\t\t}\n\t\tfor (const record of latest) {\n\t\t\tjournal.record({\n\t\t\t\tactiveSessionId: record.activeSessionId,\n\t\t\t\tsessionId: record.sessionId,\n\t\t\t\t...(record.sessionFile ? { sessionFile: record.sessionFile } : {}),\n\t\t\t\tbusy: false,\n\t\t\t\toperation: \"recovery_hold\",\n\t\t\t});\n\t\t}\n\t\tthis.log(\n\t\t\t`Recovered worker ${worker.descriptor.workerId} without replaying uncertain operations: ${uncertain\n\t\t\t\t.map((record) => record.operation)\n\t\t\t\t.join(\", \")}`,\n\t\t);\n\t}\n\tprivate async refreshWorkerSummaries(\n\t\tworker: ResidentWorker,\n\t\trecovery = false,\n\t\tfillGaps = recovery,\n\t\tretried = false,\n\t): Promise<void> {\n\t\tif (this.isWorkerStopping(worker)) {\n\t\t\tthrow new Error(\"Session worker is stopping\");\n\t\t}\n\t\tif (!worker.client) {\n\t\t\tthrow new Error(\"Session worker is not connected\");\n\t\t}\n\t\tconst pullSource = worker.client;\n\t\tconst epochAtStart = worker.rosterEpoch ?? 0;\n\t\tconst response = await pullSource.request({ type: \"list\" }, 5000);\n\t\t// A frame received mid-pull can remove rows this stale pull would resurrect; re-pull once, then skip the fill.\n\t\tif (fillGaps && (worker.rosterEpoch ?? 0) !== epochAtStart && !retried) {\n\t\t\treturn this.refreshWorkerSummaries(worker, recovery, fillGaps, true);\n\t\t}\n\t\tconst summaries = sessionSummariesFromResponse(response);\n\t\tconst nextSummaries = new Map(summaries.map((summary) => [summary.activeSessionId ?? summary.id, summary]));\n\t\tconst root = nextSummaries.get(worker.descriptor.rootActiveSessionId);\n\t\tif (recovery && !root) {\n\t\t\tthrow new Error(`Session worker omitted its root session during recovery`);\n\t\t}\n\t\tworker.summaries = nextSummaries;\n\t\tif (fillGaps) {\n\t\t\tawait this.chainWorkerRosterApply(worker, pullSource, () => {\n\t\t\t\tif ((worker.rosterEpoch ?? 0) === epochAtStart) this.syncRosterFromWorkerSummaries(worker);\n\t\t\t});\n\t\t}\n\t\tfor (const summary of summaries) {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\tif (summary.streamingMessage?.role === \"assistant\") {\n\t\t\t\tthis.streamReconstructor.seed(activeSessionId, summary.streamingMessage);\n\t\t\t} else if (!summary.isStreaming) {\n\t\t\t\tthis.streamReconstructor.clear(activeSessionId);\n\t\t\t}\n\t\t}\n\t\tif (root) {\n\t\t\tif (recovery) {\n\t\t\t\tawait this.assertRecoveryAllowed();\n\t\t\t}\n\t\t\tawait this.chainWorkerRosterApply(worker, pullSource, () => {\n\t\t\t\tif ((worker.rosterEpoch ?? 0) !== epochAtStart) return;\n\t\t\t\tworker.descriptor.rootSessionId = root.sessionId;\n\t\t\t\tworker.descriptor.sessionFile = root.sessionFile;\n\t\t\t\tworker.descriptor.createCommand = durableDaemonCreateCommand({\n\t\t\t\t\ttype: \"create\",\n\t\t\t\t\tsessionPath: root.sessionFile,\n\t\t\t\t\tnoSession: worker.descriptor.createCommand.noSession,\n\t\t\t\t});\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t});\n\t\t}\n\t}\n\n\tprivate async familyCatalogEntries(): Promise<AgentFamilyCatalogEntry[]> {\n\t\tconst rosterRows = [...this.roster().values()];\n\t\tconst entries = rosterRows.map((entry) => this.familyCatalogEntry(sessionSummaryFromRosterEntry(entry)));\n\t\tconst knownFiles = new Set(\n\t\t\trosterRows.flatMap((entry) =>\n\t\t\t\tentry.summary.sessionFile ? [canonicalSessionPath(entry.summary.sessionFile)] : [],\n\t\t\t),\n\t\t);\n\t\tconst scanned = await this.catalog.list(undefined, this.defaultSessionConfig.sessionDir);\n\t\tfor (const info of scanned) {\n\t\t\tif (knownFiles.has(canonicalSessionPath(info.path))) continue;\n\t\t\tif ((info.rlmDepth ?? (info.parentSessionPath ? -1 : 0)) !== 0) continue;\n\t\t\tentries.push(this.familyCatalogEntry(summaryForInactiveSession(info)));\n\t\t}\n\t\treturn entries;\n\t}\n\n\tprivate async withSessionNameReservation<T>(\n\t\tinput: { name: string; depth: number; parentSessionId?: string; parentSessionPath?: string },\n\t\taction: () => Promise<T>,\n\t): Promise<T> {\n\t\tconst key = sessionNameReservationKey(input);\n\t\tif (this.pendingSessionNames.has(key)) {\n\t\t\tthrow new Error(formatAgentSessionNameUnavailable(input.name, input.depth));\n\t\t}\n\t\tthis.pendingSessionNames.add(key);\n\t\ttry {\n\t\t\treturn await action();\n\t\t} finally {\n\t\t\tthis.pendingSessionNames.delete(key);\n\t\t}\n\t}\n\n\tprivate async assertSupervisorSessionNameAvailable(\n\t\ttarget: Pick<SessionSummary, \"sessionId\" | \"rlmDepth\" | \"parentSessionId\" | \"parentSessionPath\">,\n\t\tname: string,\n\t): Promise<void> {\n\t\tassertAgentSessionNameAvailable(await this.familyCatalogEntries(), {\n\t\t\tname,\n\t\t\tdepth: target.rlmDepth ?? 0,\n\t\t\tparentSessionId: target.parentSessionId,\n\t\t\tparentSessionPath: target.parentSessionPath ? canonicalSessionPath(target.parentSessionPath) : undefined,\n\t\t\tignoreSessionId: target.sessionId,\n\t\t});\n\t}\n\n\tprivate roster(): AgentRoster {\n\t\tthis.rosterStore ??= new AgentRoster(canonicalSessionPath, (mutation) => this.onRosterMutation(mutation));\n\t\treturn this.rosterStore;\n\t}\n\n\tprivate onRosterMutation(mutation: AgentRosterMutation): void {\n\t\tif (mutation.type === \"delete\") {\n\t\t\tthis.pendingRosterChanged.delete(mutation.agentId);\n\t\t\tthis.pendingRosterRemoved.add(mutation.agentId);\n\t\t} else {\n\t\t\tthis.pendingRosterRemoved.delete(mutation.agentId);\n\t\t\tthis.pendingRosterChanged.add(mutation.agentId);\n\t\t}\n\t\tthis.scheduleRosterPush();\n\t}\n\n\tprivate scheduleRosterPush(): void {\n\t\tif (this.rosterPushScheduled || this.shuttingDown) return;\n\t\tthis.rosterPushScheduled = true;\n\t\tsetImmediate(() => {\n\t\t\tthis.rosterPushScheduled = false;\n\t\t\tthis.flushRosterUpdates();\n\t\t});\n\t}\n\n\tprivate flushRosterUpdates(): void {\n\t\tconst changed: AgentRosterEntry[] = [];\n\t\tconst removed: string[] = [];\n\t\tfor (const agentId of this.pendingRosterRemoved) {\n\t\t\tif (this.publishedRosterIds.delete(agentId)) removed.push(agentId);\n\t\t}\n\t\tfor (const agentId of this.pendingRosterChanged) {\n\t\t\tconst entry = this.roster().get(agentId);\n\t\t\tif (!entry) continue;\n\t\t\tif (this.isRosterEntryVisibleToClients(entry)) {\n\t\t\t\tchanged.push(entry);\n\t\t\t\tthis.publishedRosterIds.add(agentId);\n\t\t\t} else if (this.publishedRosterIds.delete(agentId)) {\n\t\t\t\tremoved.push(agentId);\n\t\t\t}\n\t\t}\n\t\tthis.pendingRosterChanged.clear();\n\t\tthis.pendingRosterRemoved.clear();\n\t\tif (changed.length === 0 && removed.length === 0) return;\n\t\tfor (const client of this.clients) {\n\t\t\tif (client.rosterSubscribed !== true) continue;\n\t\t\tif (client.backpressured === true) {\n\t\t\t\tclient.rosterResyncPending = true;\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.write(client, {\n\t\t\t\ttype: \"roster_update\",\n\t\t\t\tchanged,\n\t\t\t\t...(removed.length > 0 ? { removed } : {}),\n\t\t\t});\n\t\t}\n\t}\n\n\tprivate rosterEntriesForClient(): AgentRosterEntry[] {\n\t\tconst entries = [...this.roster().values()].filter((entry) => this.isRosterEntryVisibleToClients(entry));\n\t\tfor (const entry of entries) this.publishedRosterIds.add(entry.agentId);\n\t\treturn entries;\n\t}\n\n\tprivate isRosterEntryVisibleToClients(entry: AgentRosterEntry): boolean {\n\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\treturn worker === undefined || this.isVisibleWorker(worker);\n\t}\n\n\tprivate writeRosterEntry(\n\t\tentry: WorkerRosterEntry,\n\t\tworker?: ResidentWorker,\n\t\tstatusLabel?: AgentRosterEntry[\"statusLabel\"],\n\t): AgentRosterEntry {\n\t\tconst existing = entry.summary.sessionFile\n\t\t\t? this.roster().bySessionFile(entry.summary.sessionFile)\n\t\t\t: this.roster().get(entry.agentId);\n\t\t// A parent's saved-child row cannot replace that child's independent live worker.\n\t\tif (\n\t\t\tworker &&\n\t\t\t!entry.summary.activeSessionId &&\n\t\t\texisting?.summary.activeSessionId &&\n\t\t\texisting.workerId !== undefined &&\n\t\t\texisting.workerId !== worker.descriptor.workerId\n\t\t)\n\t\t\treturn existing;\n\t\tconst previousDirect = this.roster().get(entry.agentId)?.summary.directAttachedClients ?? 0;\n\t\tconst stored = this.roster().write(entry, worker?.descriptor.workerId, statusLabel);\n\t\t// Direct peers attach and detach on the worker socket, so their last detach arrives\n\t\t// here as roster truth instead of through a supervisor-socket close.\n\t\tif (worker !== undefined && previousDirect > 0 && (entry.summary.directAttachedClients ?? 0) === 0) {\n\t\t\tvoid this.evictEmptySessionOnLastDetach(entry.summary.activeSessionId ?? entry.summary.id);\n\t\t}\n\t\treturn stored;\n\t}\n\n\tprivate workerOwnedRosterSummaryForPath(canonicalPath: string): SessionSummary | undefined {\n\t\tconst entry = this.roster().bySessionFile(canonicalPath);\n\t\tif (!entry || entry.workerId === undefined || !this.workers.has(entry.workerId)) return undefined;\n\t\treturn sessionSummaryFromRosterEntry(entry);\n\t}\n\n\tprivate workerRosterEntries(worker: ResidentWorker): AgentRosterEntry[] {\n\t\treturn this.roster().entriesForWorker(worker.descriptor.workerId);\n\t}\n\n\tprivate async seedRosterLedger(): Promise<void> {\n\t\ttry {\n\t\t\tconst roots = new Set<string>();\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tconst root = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\t\t\tif (root !== undefined) roots.add(canonicalSessionPath(root));\n\t\t\t}\n\t\t\tif (roots.size === 0) return;\n\t\t\tconst edges = await this.rlmSpawnLedger().liveEdges();\n\t\t\tconst descendsFrom = rosterFamilyDescendsFrom(edges);\n\t\t\tfor (const edge of edges) {\n\t\t\t\tif (!descendsFrom(canonicalSessionPath(edge.parent), roots)) continue;\n\t\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\t\tif (this.roster().hasSessionFile(canonicalSessionPath(edge.child))) continue;\n\t\t\t\tthis.roster().write(await this.hydratedSeedEntry(entry));\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tthis.log(`Could not seed the agent roster from the spawn ledger: ${String(error)}`);\n\t\t}\n\t}\n\n\tprivate rosterEntryForSpawnLedgerEdge(edge: RlmLedgerEdge): WorkerRosterEntry {\n\t\tconst persistedSessionId = basename(edge.child, \".jsonl\");\n\t\tconst summary: WorkerRosterEntry[\"summary\"] = {\n\t\t\tid: persistedSessionId,\n\t\t\tlifecycle: \"live\",\n\t\t\tactivity: \"idle\",\n\t\t\tisSessionActive: false,\n\t\t\truntimeKind: \"subagent\",\n\t\t\trlmDepth: edge.depth,\n\t\t\tsessionId: persistedSessionId,\n\t\t\tsessionFile: edge.child,\n\t\t\tsessionName: edge.name,\n\t\t\tcwd: dirname(edge.child),\n\t\t\tisStreaming: false,\n\t\t\tisCompacting: false,\n\t\t\tattachedClients: 0,\n\t\t\tmessageCount: 0,\n\t\t\tparentSessionPath: edge.parent,\n\t\t\trlmChildId: edge.childId,\n\t\t};\n\t\treturn { agentId: rosterAgentIdForSummary(summary), summary };\n\t}\n\n\tprivate consumeWorkerRosterDelta(worker: ResidentWorker, payload: Buffer, source?: DaemonWorkerClient): void {\n\t\tlet delta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>;\n\t\ttry {\n\t\t\tdelta = JSON.parse(payload.toString(\"utf8\")) as Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>;\n\t\t} catch {\n\t\t\treturn;\n\t\t}\n\t\tif (delta.type !== \"roster_delta\" || !Array.isArray(delta.entries)) return;\n\t\tworker.rosterEpoch = (worker.rosterEpoch ?? 0) + 1;\n\t\tconst applySource = source ?? worker.client ?? worker.pendingClient;\n\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\tif (delta.snapshot !== true && worker.rosterApplyChain === undefined) {\n\t\t\tthis.applyWorkerRosterDelta(worker, delta);\n\t\t\treturn;\n\t\t}\n\t\tthis.chainWorkerRosterApply(worker, applySource, () =>\n\t\t\tdelta.snapshot === true\n\t\t\t\t? this.applyWorkerRosterSnapshot(worker, delta, applySource)\n\t\t\t\t: this.applyWorkerRosterDelta(worker, delta),\n\t\t);\n\t}\n\n\tprivate chainWorkerRosterApply(\n\t\tworker: ResidentWorker,\n\t\tsource: DaemonWorkerClient | undefined,\n\t\tapply: () => void | Promise<void>,\n\t): Promise<void> {\n\t\tconst chained = (worker.rosterApplyChain ?? Promise.resolve())\n\t\t\t.then(() => {\n\t\t\t\tif (!this.isWorkerRosterApplyCurrent(worker, source)) return;\n\t\t\t\treturn apply();\n\t\t\t})\n\t\t\t.catch((error: unknown) => {\n\t\t\t\tthis.log(`could not apply a roster frame: ${String(error)}`);\n\t\t\t\tthis.scheduleRosterRepairPull(worker);\n\t\t\t});\n\t\tworker.rosterApplyChain = chained;\n\t\tvoid chained.finally(() => {\n\t\t\tif (worker.rosterApplyChain === chained) worker.rosterApplyChain = undefined;\n\t\t});\n\t\treturn chained;\n\t}\n\n\t// An apply is valid only while its own source connection is current: dead connections' parked applies abort.\n\tprivate isWorkerRosterApplyCurrent(worker: ResidentWorker, source: DaemonWorkerClient | undefined): boolean {\n\t\treturn (\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tsource !== undefined &&\n\t\t\t(source === worker.client || source === worker.pendingClient)\n\t\t);\n\t}\n\n\tprivate scheduleRosterRepairPull(worker: ResidentWorker): void {\n\t\tif (worker.rosterRepairPull || !this.isWorkerRosterApplyCurrent(worker, worker.client)) return;\n\t\t// The marker stays set while the repair's own fill applies, so a failing repair never respawns itself.\n\t\tworker.rosterRepairPull = this.refreshWorkerSummaries(worker, false, true)\n\t\t\t.catch((error: unknown) =>\n\t\t\t\tthis.log(`Roster repair pull failed for worker ${worker.descriptor.workerId}: ${String(error)}`),\n\t\t\t)\n\t\t\t.finally(() => {\n\t\t\t\tworker.rosterRepairPull = undefined;\n\t\t\t});\n\t}\n\n\tprivate applyWorkerRosterDelta(\n\t\tworker: ResidentWorker,\n\t\tdelta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>,\n\t): void {\n\t\tfor (const entry of delta.entries) {\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t\tthis.syncRootDescriptorFromRosterEntry(worker, entry);\n\t\t}\n\t\tfor (const agentId of delta.removedAgentIds ?? []) {\n\t\t\tthis.roster().delete(agentId);\n\t\t}\n\t}\n\n\tprivate async applyWorkerRosterSnapshot(\n\t\tworker: ResidentWorker,\n\t\tdelta: Extract<DaemonWorkerRosterOutbound, { type: \"roster_delta\" }>,\n\t\tsource?: DaemonWorkerClient,\n\t): Promise<void> {\n\t\tlet edgesFailed = false;\n\t\tconst edges = await this.rlmSpawnLedger()\n\t\t\t.liveEdges()\n\t\t\t.catch((error: unknown) => {\n\t\t\t\tthis.log(`Could not read the spawn ledger during a snapshot apply: ${String(error)}`);\n\t\t\t\tedgesFailed = true;\n\t\t\t\treturn [] as RlmLedgerEdge[];\n\t\t\t});\n\t\tconst applySource = source ?? worker.client ?? worker.pendingClient;\n\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\tconst sent = new Set(delta.entries.map((entry) => entry.agentId));\n\t\tconst removed = new Set(delta.removedAgentIds ?? []);\n\t\tconst unclaimed = new Map<string, AgentRosterEntry>();\n\t\tif (!edgesFailed) {\n\t\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\t\tif (sent.has(entry.agentId)) continue;\n\t\t\t\tunclaimed.set(entry.agentId, entry);\n\t\t\t}\n\t\t}\n\t\t// Only this worker's family reseeds: anything wider can resurrect a client-owned worker's dropped children.\n\t\tconst workerRoot = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tconst rootPaths = new Set(workerRoot !== undefined ? [canonicalSessionPath(workerRoot)] : []);\n\t\tconst descendsFrom = rosterFamilyDescendsFrom(edges);\n\t\tconst familyEdges = edges.filter((edge) => descendsFrom(canonicalSessionPath(edge.parent), rootPaths));\n\t\t// \"Unclaimed\" rows survive: the sweep deletes them but the restore branch rewrites them.\n\t\tconst rowSurvivesWithoutReseed = (entry: WorkerRosterEntry, childPath: string): boolean => {\n\t\t\tif (unclaimed.has(entry.agentId)) return true;\n\t\t\tif (sent.has(entry.agentId) && !removed.has(entry.agentId)) return true;\n\t\t\tconst survives = (row: AgentRosterEntry | undefined): boolean =>\n\t\t\t\trow !== undefined && !unclaimed.has(row.agentId) && !removed.has(row.agentId);\n\t\t\treturn survives(this.roster().get(entry.agentId)) || survives(this.roster().bySessionFile(childPath));\n\t\t};\n\t\tconst seededEntries = new Map<string, WorkerRosterEntry>();\n\t\tfor (const edge of familyEdges) {\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (rowSurvivesWithoutReseed(entry, canonicalSessionPath(edge.child))) continue;\n\t\t\tseededEntries.set(entry.agentId, await this.hydratedSeedEntry(entry));\n\t\t\tif (!this.isWorkerRosterApplyCurrent(worker, applySource)) return;\n\t\t}\n\n\t\t// Unreadable edges skip the absentee sweep: it cannot tell registry children from stale rows.\n\t\tfor (const entry of unclaimed.values()) this.roster().delete(entry.agentId);\n\t\tfor (const entry of delta.entries) {\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t\tthis.syncRootDescriptorFromRosterEntry(worker, entry);\n\t\t}\n\t\tfor (const agentId of removed) this.roster().delete(agentId);\n\t\tif (edgesFailed) {\n\t\t\tthis.scheduleRosterRepairPull(worker);\n\t\t\treturn;\n\t\t}\n\t\tfor (const edge of familyEdges) {\n\t\t\tconst entry = this.rosterEntryForSpawnLedgerEdge(edge);\n\t\t\tif (this.roster().has(entry.agentId)) continue;\n\t\t\tif (this.roster().hasSessionFile(canonicalSessionPath(edge.child))) continue;\n\t\t\tconst previous = unclaimed.get(entry.agentId);\n\t\t\tif (previous) {\n\t\t\t\tconst { status, statusLabel, lastHeardFromAt, workerId, ...rest } = previous;\n\t\t\t\tthis.writeRosterEntry(rest, worker);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.roster().write(seededEntries.get(entry.agentId) ?? { ...entry, seededCwd: true });\n\t\t}\n\t}\n\n\tprivate syncRootDescriptorFromRosterEntry(worker: ResidentWorker, entry: WorkerRosterEntry): void {\n\t\tconst summary = entry.summary;\n\t\tif (summary.activeSessionId !== worker.descriptor.rootActiveSessionId) return;\n\t\tif (\n\t\t\tworker.descriptor.rootSessionId === summary.sessionId &&\n\t\t\tworker.descriptor.sessionFile === summary.sessionFile\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tworker.descriptor.rootSessionId = summary.sessionId;\n\t\tworker.descriptor.sessionFile = summary.sessionFile;\n\t\tworker.descriptor.createCommand = durableDaemonCreateCommand({\n\t\t\ttype: \"create\",\n\t\t\tsessionPath: summary.sessionFile,\n\t\t\tnoSession: worker.descriptor.createCommand.noSession,\n\t\t});\n\t\tthis.persistWorker(worker);\n\t}\n\n\t// Behind the pull-epoch guard the pull is never staler than the row it replaces; never steal another worker's claim.\n\tprivate syncRosterFromWorkerSummaries(worker: ResidentWorker): void {\n\t\tfor (const summary of worker.summaries.values()) {\n\t\t\tconst entry = workerRosterEntryFromSummary(summary);\n\t\t\tconst existing = this.roster().get(entry.agentId);\n\t\t\tif (existing?.workerId !== undefined && existing.workerId !== worker.descriptor.workerId) continue;\n\t\t\tthis.writeRosterEntry(entry, worker);\n\t\t}\n\t}\n\n\tprivate markWorkerRosterEntries(worker: ResidentWorker, statusLabel: \"recovering\" | \"failed\" | undefined): void {\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tif (!entry.queuedChild && entry.summary.activeSessionId === undefined) continue;\n\t\t\tthis.roster().amend(entry.agentId, { statusLabel });\n\t\t}\n\t}\n\n\tprivate flipWorkerRosterEntriesInactive(worker: ResidentWorker): void {\n\t\t// Client-owned workers are ephemeral and private: their rows die with the registration.\n\t\tconst ephemeral = worker.descriptor.ownerClientId !== undefined;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tif (ephemeral || entry.queuedChild) {\n\t\t\t\tthis.roster().delete(entry.agentId);\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\t// Registration marks survive eviction: passive rows still have schedules behind them.\n\t\t\tthis.writeRosterEntry(\n\t\t\t\tpassivatedWorkerRosterEntry(entry, {\n\t\t\t\t\thasRegisteredHeartbeat: entry.summary.hasRegisteredHeartbeat === true,\n\t\t\t\t\thasRegisteredCronJob: entry.summary.hasRegisteredCronJob === true,\n\t\t\t\t}),\n\t\t\t);\n\t\t}\n\t}\n\n\tprivate sweepRosterStaleness(now = Date.now()): void {\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.client === undefined || worker.lastFrameAt === undefined) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (now - worker.lastFrameAt > ROSTER_STALE_AFTER_MS) {\n\t\t\t\tconst lastHeardFromAt = new Date(worker.lastFrameAt).toISOString();\n\t\t\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\t\t\t// write() rebuilds rows without the mark; the sweep owns it and restamps only those.\n\t\t\t\t\tif (entry.lastHeardFromAt !== lastHeardFromAt) this.roster().amend(entry.agentId, { lastHeardFromAt });\n\t\t\t\t}\n\t\t\t\tworker.rosterStale = true;\n\t\t\t} else if (worker.rosterStale) {\n\t\t\t\tthis.clearRosterStaleness(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate clearRosterStaleness(worker: ResidentWorker): void {\n\t\tif (!worker.rosterStale) return;\n\t\tworker.rosterStale = false;\n\t\tfor (const entry of this.workerRosterEntries(worker)) {\n\t\t\tthis.roster().amend(entry.agentId, { lastHeardFromAt: undefined });\n\t\t}\n\t}\n\n\tprivate rlmJournalScope(sessionDir?: string): { agentDir: string; sessionsDir: string; journalPath: string } {\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) throw new Error(\"Daemon supervisor config is missing agentDir\");\n\t\tconst sessionsDir = sessionDir ?? this.defaultSessionConfig.sessionDir ?? getSessionsDir(agentDir);\n\t\treturn { agentDir, sessionsDir, journalPath: rlmLedgerPath(agentDir, sessionsDir) };\n\t}\n\n\tprivate assertRlmJournalAdmission(): RlmJournalOwner {\n\t\tconst owner = this.rlmJournalOwner;\n\t\tif (!owner || this.rlmJournalClose) throw new Error(\"RLM journal mutation admission is closed\");\n\t\ttry {\n\t\t\tif (!this.ownership) throw new Error(\"RLM journal supervisor ownership is unavailable\");\n\t\t\tthis.ownership.assertJournalCurrent(owner.ledgerPath);\n\t\t} catch (error) {\n\t\t\tthis.fenceRlmJournalOwner();\n\t\t\tthrow error;\n\t\t}\n\t\treturn owner;\n\t}\n\n\tprivate fenceRlmJournalOwner(): void {\n\t\tvoid this.closeRlmJournalOwner().catch((error: unknown) => this.reportCleanupFailure(\"RLM journal owner\", error));\n\t}\n\n\tprivate closeRlmJournalOwner(): Promise<void> {\n\t\tthis.rlmSpawnLedgerInstance?.stopAdmission();\n\t\tif (this.rlmJournalClose) return this.rlmJournalClose;\n\t\tconst owner = this.rlmJournalOwner;\n\t\tif (!owner) return Promise.resolve();\n\t\tthis.rlmJournalClose = (async () => {\n\t\t\tawait this.rlmSpawnLedgerInstance?.flush();\n\t\t\tawait owner.close();\n\t\t\tthis.rlmJournalOwner = undefined;\n\t\t})();\n\t\treturn this.rlmJournalClose;\n\t}\n\n\t/** Reads are local; every mutation is acknowledged by the external lifetime owner. */\n\tprivate rlmSpawnLedger(): RlmSpawnLedger {\n\t\tconst { agentDir, sessionsDir } = this.rlmJournalScope();\n\t\tthis.rlmSpawnLedgerInstance ??= new RlmSpawnLedger(\n\t\t\tagentDir,\n\t\t\tsessionsDir,\n\t\t\tcreateRlmLedgerRegistrySeedSource(),\n\t\t\t(message) => this.log(message),\n\t\t\t{\n\t\t\t\tmode: \"remote\",\n\t\t\t\tmutate: (mutation) => {\n\t\t\t\t\tconst owner = this.rlmJournalOwner;\n\t\t\t\t\tif (!owner) throw new Error(\"RLM journal owner is unavailable\");\n\t\t\t\t\treturn owner.mutate(mutation);\n\t\t\t\t},\n\t\t\t},\n\t\t);\n\t\treturn this.rlmSpawnLedgerInstance;\n\t}\n\n\t/**\n\t * Ledger-backed same-parent rows for name reservation and admission. Rows\n\t * carry ledger topology plus best-effort display fields; consumers here\n\t * only need id/path/name/depth/parent.\n\t */\n\tprivate rlmLedgerSiblings(sessionPath: string): Promise<SessionInfo[]> {\n\t\treturn this.rlmSpawnLedger().siblings(sessionPath);\n\t}\n\n\tprivate async savedSessionNameReservationInput(\n\t\tsessionPath: string,\n\t\tname: string,\n\t): Promise<{ name: string; depth: number; parentSessionId?: string; parentSessionPath?: string }> {\n\t\tconst targetPath = canonicalSessionPath(sessionPath);\n\t\tconst active = this.workerOwnedRosterSummaryForPath(targetPath);\n\t\tif (active) return this.summaryNameReservationInput(active, name);\n\t\tconst siblings = await this.rlmLedgerSiblings(sessionPath);\n\t\tconst saved = siblings.find((info) => canonicalSessionPath(info.path) === targetPath);\n\t\tif (!saved) throw new Error(`Session not found: ${sessionPath}`);\n\t\treturn {\n\t\t\tname,\n\t\t\tdepth: saved.rlmDepth ?? siblings.find((sibling) => sibling.rlmDepth !== undefined)?.rlmDepth ?? 0,\n\t\t\tparentSessionPath: saved.parentSessionPath,\n\t\t};\n\t}\n\n\tprivate summaryNameReservationInput(\n\t\ttarget: Pick<SessionSummary, \"rlmDepth\" | \"parentSessionId\" | \"parentSessionPath\">,\n\t\tname: string,\n\t): { name: string; depth: number; parentSessionId?: string; parentSessionPath?: string } {\n\t\tconst depth = target.rlmDepth ?? (target.parentSessionPath ? 1 : 0);\n\t\treturn {\n\t\t\tname,\n\t\t\tdepth,\n\t\t\t...(depth > 0 && target.parentSessionId ? { parentSessionId: target.parentSessionId } : {}),\n\t\t\t...(depth > 0 && target.parentSessionPath ? { parentSessionPath: target.parentSessionPath } : {}),\n\t\t};\n\t}\n\n\tprivate async assertSupervisorSavedSessionNameAvailable(sessionPath: string, name: string): Promise<void> {\n\t\tconst targetPath = canonicalSessionPath(sessionPath);\n\t\tconst active = this.workerOwnedRosterSummaryForPath(targetPath);\n\t\tif (active) return this.assertSupervisorSessionNameAvailable(active, name);\n\t\tconst siblings = await this.rlmLedgerSiblings(sessionPath);\n\t\tconst saved = siblings.find((info) => canonicalSessionPath(info.path) === targetPath);\n\t\tif (!saved) throw new Error(`Session not found: ${sessionPath}`);\n\t\tif (saved.parentSessionPath && (saved.rlmDepth ?? 0) > 0) {\n\t\t\tthis.assertSavedSiblingNameAvailable(siblings, saved, name);\n\t\t} else {\n\t\t\tawait this.assertSupervisorSessionNameAvailable(summaryForInactiveSession(saved), name);\n\t\t}\n\t}\n\n\tprivate assertSavedSiblingNameAvailable(siblings: SessionInfo[], target: SessionInfo, name: string): void {\n\t\tconst setDepth = target.rlmDepth ?? siblings.find((sibling) => sibling.rlmDepth !== undefined)?.rlmDepth ?? 0;\n\t\tassertAgentSessionNameAvailable(\n\t\t\tsiblings.map((info) => {\n\t\t\t\tconst summary = summaryForInactiveSession(info);\n\t\t\t\tconst ledgerRow = this.roster().bySessionFile(canonicalSessionPath(info.path));\n\t\t\t\treturn {\n\t\t\t\t\tid: summary.sessionId,\n\t\t\t\t\t...(summary.sessionName ? { name: summary.sessionName } : {}),\n\t\t\t\t\tdepth: setDepth,\n\t\t\t\t\tstatus: ledgerRow?.status ?? classifySessionRosterStatus(summary),\n\t\t\t\t\t...(summary.parentSessionPath\n\t\t\t\t\t\t? { parentSessionPath: canonicalSessionPath(summary.parentSessionPath) }\n\t\t\t\t\t\t: {}),\n\t\t\t\t};\n\t\t\t}),\n\t\t\t{\n\t\t\t\tname,\n\t\t\t\tdepth: setDepth,\n\t\t\t\tparentSessionPath: target.parentSessionPath ? canonicalSessionPath(target.parentSessionPath) : undefined,\n\t\t\t\tignoreSessionId: target.id,\n\t\t\t},\n\t\t);\n\t}\n\n\tprivate isVisibleWorker(worker: ResidentWorker): boolean {\n\t\treturn worker.descriptor.ownerClientId === undefined;\n\t}\n\n\t/** A worker with a durable or in-memory stop intent is stopping, never live. */\n\tprivate isWorkerStopping(worker: ResidentWorker): boolean {\n\t\treturn worker.intentionalStop || worker.descriptor.stopRequestedAt !== undefined;\n\t}\n\n\t/** Live workers are visible to all clients and not stopping. */\n\tprivate isLiveWorker(worker: ResidentWorker): boolean {\n\t\treturn this.isVisibleWorker(worker) && !this.isWorkerStopping(worker);\n\t}\n\n\t/**\n\t * The lifecycle reported to clients. A stop intent always wins, and a worker\n\t * whose process connection is gone is never reported as \"ready\".\n\t */\n\tprivate effectiveWorkerState(worker: ResidentWorker): DaemonWorkerLifecycle {\n\t\tif (this.isWorkerStopping(worker)) {\n\t\t\treturn \"stopping\";\n\t\t}\n\t\tif (worker.descriptor.lifecycle === \"ready\" && worker.client === undefined) {\n\t\t\treturn \"recovering\";\n\t\t}\n\t\treturn worker.descriptor.lifecycle;\n\t}\n\n\tprivate requireAvailableWorkerClient(worker: ResidentWorker, allowStopping = false): DaemonWorkerClient {\n\t\tif (\n\t\t\t!worker.client ||\n\t\t\tworker.descriptor.lifecycle !== \"ready\" ||\n\t\t\t(!allowStopping && this.isWorkerStopping(worker))\n\t\t) {\n\t\t\tthrow new Error(`Session worker is ${this.effectiveWorkerState(worker)}`);\n\t\t}\n\t\treturn worker.client;\n\t}\n\n\tprivate async issuePeerTransport(\n\t\tworker: ResidentWorker,\n\t\tsummary: SessionSummary,\n\t): Promise<DaemonPeerTransportTicket> {\n\t\tawait this.assertCurrentOwnership();\n\t\tif (!worker.peerTransportCapable) {\n\t\t\tthrow new Error(\"Session worker does not support direct peer transport\");\n\t\t}\n\t\tawait this.refreshWorkerSummaries(worker);\n\t\tconst workerClient = this.requireAvailableWorkerClient(worker);\n\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\tconst currentSummary = this.findSummaryInWorker(worker, activeSessionId);\n\t\tif (!currentSummary) {\n\t\t\tthrow new Error(\"Direct transport target changed during admission\");\n\t\t}\n\t\tconst workerInstanceId = worker.descriptor.workerInstanceId;\n\t\tconst workerProcessStartId = worker.descriptor.processStartId;\n\t\tif (!workerInstanceId || !workerProcessStartId) {\n\t\t\tthrow new Error(\"Direct transport requires an exact worker process identity\");\n\t\t}\n\t\tif (this.processIdentity(worker.descriptor.pid, workerProcessStartId) !== \"current\") {\n\t\t\tthrow new Error(\"Direct transport worker process identity is not current\");\n\t\t}\n\t\tlet socketIdentity: DaemonSocketIdentity | undefined;\n\t\ttry {\n\t\t\tsocketIdentity = getDaemonSocketIdentity(worker.descriptor.socketPath);\n\t\t} catch {\n\t\t\tsocketIdentity = undefined;\n\t\t}\n\t\tif (!socketIdentity) {\n\t\t\tthrow new Error(\"Direct transport requires an exact worker socket identity\");\n\t\t}\n\t\tconst grantId = randomUUID();\n\t\tconst token = randomBytes(32).toString(\"base64url\");\n\t\tconst expiresAt = new Date(Date.now() + PEER_TRANSPORT_GRANT_TTL_MS).toISOString();\n\t\tconst resolvedActiveSessionId = currentSummary.activeSessionId ?? currentSummary.id;\n\t\tconst registration = await workerClient.requestWorker(\n\t\t\t{\n\t\t\t\ttype: \"worker_register_peer_transport\",\n\t\t\t\tgrant: {\n\t\t\t\t\tgrantId,\n\t\t\t\t\ttoken,\n\t\t\t\t\texpiresAt,\n\t\t\t\t\tpurpose: \"session_client\",\n\t\t\t\t\tworkerInstanceId,\n\t\t\t\t\tactiveSessionId: resolvedActiveSessionId,\n\t\t\t\t\tissuerGeneration: this.generation,\n\t\t\t\t},\n\t\t\t},\n\t\t\t3000,\n\t\t);\n\t\tif (!registration.success) {\n\t\t\tthrow deserializeDaemonError(registration);\n\t\t}\n\t\treturn {\n\t\t\tpurpose: \"session_client\",\n\t\t\tsocketPath: worker.descriptor.socketPath,\n\t\t\tsocketIdentity,\n\t\t\tworkerInstanceId,\n\t\t\tactiveSessionId: resolvedActiveSessionId,\n\t\t\tgrantId,\n\t\t\ttoken,\n\t\t\texpiresAt,\n\t\t};\n\t}\n\n\tprivate familyCatalogEntry(summary: SessionSummary): AgentFamilyCatalogEntry {\n\t\tconst depth = summary.rlmDepth ?? (summary.parentSessionPath ? 1 : 0);\n\t\treturn {\n\t\t\tid: summary.sessionId,\n\t\t\t...(summary.sessionName ? { name: summary.sessionName } : {}),\n\t\t\tdepth,\n\t\t\tstatus: summary.rosterStatus ?? classifySessionRosterStatus(summary),\n\t\t\t...(depth > 0 && summary.parentSessionId ? { parentSessionId: summary.parentSessionId } : {}),\n\t\t\t...(depth > 0 && summary.parentSessionPath\n\t\t\t\t? { parentSessionPath: canonicalSessionPath(summary.parentSessionPath) }\n\t\t\t\t: {}),\n\t\t\t...(summary.sessionFile ? { sessionPath: canonicalSessionPath(summary.sessionFile) } : {}),\n\t\t};\n\t}\n\n\tprivate agentPeerSummary(summary: SessionSummary): AgentSessionMessageAgentSummary {\n\t\treturn {\n\t\t\tactiveSessionId: summary.activeSessionId ?? summary.id,\n\t\t\tsessionId: summary.sessionId,\n\t\t\t...(summary.sessionName ? { sessionName: summary.sessionName } : {}),\n\t\t\truntimeKind: summary.runtimeKind ?? \"top-level\",\n\t\t\tcwd: summary.cwd,\n\t\t\tisStreaming: summary.isStreaming,\n\t\t\tunfinishedActionCount:\n\t\t\t\tsummary.unfinishedActionCount ??\n\t\t\t\t(summary.sessionActions.active\n\t\t\t\t\t? 1 + summary.sessionActions.queuedCount\n\t\t\t\t\t: summary.sessionActions.queuedCount),\n\t\t\t...(summary.parentActiveSessionId ? { parentActiveSessionId: summary.parentActiveSessionId } : {}),\n\t\t\t...(summary.parentSessionId ? { parentSessionId: summary.parentSessionId } : {}),\n\t\t\t...(summary.parentSessionPath ? { parentSessionPath: summary.parentSessionPath } : {}),\n\t\t\t...(summary.sessionFile ? { sessionPath: summary.sessionFile } : {}),\n\t\t\t...(summary.rlmDepth !== undefined ? { rlmDepth: summary.rlmDepth } : {}),\n\t\t\tstatus: summary.rosterStatus ?? classifySessionRosterStatus(summary),\n\t\t\t...(summary.rlmChildId ? { rlmChildId: summary.rlmChildId } : {}),\n\t\t};\n\t}\n\n\tprivate attachedClientCount(summary: SessionSummary, activeSessionId: string): number {\n\t\treturn (\n\t\t\t(summary.directAttachedClients ?? 0) +\n\t\t\t[...this.clients].filter((client) => client.attachedActiveSessionIds.has(activeSessionId)).length\n\t\t);\n\t}\n\n\tprivate publicSummary(worker: ResidentWorker, summary: SessionSummary): SessionSummary {\n\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\treturn {\n\t\t\t...summary,\n\t\t\tattachedClients: this.attachedClientCount(summary, activeSessionId),\n\t\t\tworkerState: this.effectiveWorkerState(worker),\n\t\t\tworkerPid: worker.descriptor.pid,\n\t\t};\n\t}\n\n\tprivate async findWorker(\n\t\tselector: string,\n\t\tincludeWorker?: (worker: ResidentWorker) => boolean,\n\t): Promise<WorkerMatch> {\n\t\tlet matches = this.matchWorkers(selector, includeWorker);\n\t\tif (matches.length === 0) {\n\t\t\tawait Promise.all(\n\t\t\t\t[...this.workers.values()].map((worker) =>\n\t\t\t\t\tthis.refreshWorkerSummaries(worker, false, true).catch(() => undefined),\n\t\t\t\t),\n\t\t\t);\n\t\t\tmatches = this.matchWorkers(selector, includeWorker);\n\t\t}\n\t\tif (matches.length === 1) {\n\t\t\treturn matches[0]!;\n\t\t}\n\t\tif (matches.length > 1) {\n\t\t\tthrow new Error(`Ambiguous active session \"${selector}\"`);\n\t\t}\n\t\tthrow new Error(`Unknown active session: ${selector}`);\n\t}\n\n\tprivate findWorkerForClient(client: DaemonSocketClient, selector: string): Promise<WorkerMatch> {\n\t\treturn this.findWorker(selector, (worker) => this.isWorkerAccessibleToClient(client, worker));\n\t}\n\n\tprivate isWorkerAccessibleToClient(client: DaemonSocketClient, worker: ResidentWorker): boolean {\n\t\treturn (\n\t\t\tworker.descriptor.ownerClientId === undefined ||\n\t\t\tworker.descriptor.ownerClientId === this.protocolClientId(client)\n\t\t);\n\t}\n\n\tprivate assertWorkerAccessibleToClient(client: DaemonSocketClient, worker: ResidentWorker, selector: string): void {\n\t\tif (!this.isWorkerAccessibleToClient(client, worker)) {\n\t\t\tthrow new Error(`Unknown active session: ${selector}`);\n\t\t}\n\t}\n\n\tprivate matchWorkers(selector: string, includeWorker?: (worker: ResidentWorker) => boolean): WorkerMatch[] {\n\t\tconst exact: WorkerMatch[] = [];\n\t\tconst suffix: WorkerMatch[] = [];\n\t\tfor (const entry of this.roster().values()) {\n\t\t\tif (entry.queuedChild) continue;\n\t\t\tconst worker = entry.workerId !== undefined ? this.workers.get(entry.workerId) : undefined;\n\t\t\tif (!worker || (includeWorker && !includeWorker(worker))) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tconst summary = sessionSummaryFromRosterEntry(entry);\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\tconst match = { worker, summary };\n\t\t\tif (activeSessionId === selector || summary.sessionId === selector || summary.sessionName === selector) {\n\t\t\t\texact.push(match);\n\t\t\t} else if (\n\t\t\t\tmatchesSessionIdSuffix(activeSessionId, selector) ||\n\t\t\t\tmatchesSessionIdSuffix(summary.sessionId, selector)\n\t\t\t) {\n\t\t\t\tsuffix.push(match);\n\t\t\t}\n\t\t}\n\t\treturn exact.length > 0 ? exact : suffix;\n\t}\n\n\tprivate findSummaryInWorker(worker: ResidentWorker, selector: string): SessionSummary | undefined {\n\t\tconst pathSelector = looksLikeSessionPath(selector) ? canonicalSessionPath(selector) : undefined;\n\t\tconst summaries = this.workerRosterEntries(worker)\n\t\t\t.filter((entry) => !entry.queuedChild)\n\t\t\t.map(sessionSummaryFromRosterEntry);\n\t\tconst exact = summaries.find((summary) => {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\treturn (\n\t\t\t\tactiveSessionId === selector ||\n\t\t\t\tsummary.sessionId === selector ||\n\t\t\t\tsummary.sessionName === selector ||\n\t\t\t\t(pathSelector !== undefined &&\n\t\t\t\t\tsummary.sessionFile !== undefined &&\n\t\t\t\t\tcanonicalSessionPath(summary.sessionFile) === pathSelector)\n\t\t\t);\n\t\t});\n\t\tif (exact) return exact;\n\t\treturn summaries.find((summary) => {\n\t\t\tconst activeSessionId = summary.activeSessionId ?? summary.id;\n\t\t\treturn (\n\t\t\t\tmatchesSessionIdSuffix(activeSessionId, selector) || matchesSessionIdSuffix(summary.sessionId, selector)\n\t\t\t);\n\t\t});\n\t}\n\n\tprivate findWorkerBySessionFile(sessionFile: string, exclude?: ResidentWorker): ResidentWorker | undefined {\n\t\tconst target = canonicalSessionPath(sessionFile);\n\t\tconst targetEntry = this.roster().bySessionFile(target);\n\t\tconst matches = new Set<ResidentWorker>();\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker === exclude) continue;\n\t\t\t// The roster is the one live ownership source; the stale pull cache must not resurrect a match.\n\t\t\tconst summaryMatches = targetEntry?.workerId === worker.descriptor.workerId;\n\t\t\tconst descriptorPath = worker.descriptor.sessionFile\n\t\t\t\t? canonicalSessionPath(worker.descriptor.sessionFile)\n\t\t\t\t: undefined;\n\t\t\tconst configuredPath = worker.descriptor.createCommand.sessionPath\n\t\t\t\t? canonicalSessionPath(worker.descriptor.createCommand.sessionPath)\n\t\t\t\t: undefined;\n\t\t\tif (!summaryMatches && descriptorPath !== target && configuredPath !== target) continue;\n\t\t\tif (descriptorPath && configuredPath && descriptorPath !== configuredPath) {\n\t\t\t\tthrow new Error(`Conflicting resident session paths for worker ${worker.descriptor.workerId}`);\n\t\t\t}\n\t\t\tmatches.add(worker);\n\t\t}\n\t\tif (matches.size > 1) {\n\t\t\tthrow new Error(`Ambiguous resident session path \"${sessionFile}\"`);\n\t\t}\n\t\treturn matches.values().next().value;\n\t}\n\n\tprivate async forwardToWorker(\n\t\tworker: ResidentWorker,\n\t\tcommand: DaemonCommand,\n\t\ttimeoutMs = WORKER_REQUEST_TIMEOUT_MS,\n\t): Promise<DaemonResponse> {\n\t\tconst client = this.requireAvailableWorkerClient(worker, command.type === \"kill\");\n\t\tconst response = await client.request(withoutCommandId(command), timeoutMs);\n\t\tif (command.type === \"get_state\" && response.success && isSessionSummary(response.data)) {\n\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t}\n\t\tif (command.type === \"rename\" && response.success && isSessionSummary(response.data)) {\n\t\t\tthis.writeRosterEntry(workerRosterEntryFromSummary(response.data), worker);\n\t\t\treturn { ...response, id: command.id, data: this.publicSummary(worker, response.data) };\n\t\t}\n\t\treturn responseWithId(response, command.id);\n\t}\n\n\tprivate async attachClient(\n\t\tclient: DaemonSocketClient,\n\t\tcommand: Extract<DaemonCommand, { type: \"attach\" }>,\n\t): Promise<WorkerAttachData> {\n\t\tconst ownedWorker = [...this.workers.values()].find(\n\t\t\t(worker) =>\n\t\t\t\tworker.descriptor.ownerClientId !== undefined &&\n\t\t\t\t(worker.descriptor.rootActiveSessionId === command.activeSessionId ||\n\t\t\t\t\tworker.descriptor.rootSessionId === command.activeSessionId),\n\t\t);\n\t\tif (ownedWorker) {\n\t\t\tif (ownedWorker.descriptor.ownerClientId !== this.protocolClientId(client)) {\n\t\t\t\tthrow new Error(`Unknown active session: ${command.activeSessionId}`);\n\t\t\t}\n\t\t\townedWorker.launchEnv = command.launchEnv ?? ownedWorker.launchEnv;\n\t\t\tif (!ownedWorker.client || ownedWorker.descriptor.lifecycle !== \"ready\") {\n\t\t\t\tif (command.recoveryConfig) {\n\t\t\t\t\townedWorker.transientCreateCommand = {\n\t\t\t\t\t\t...ownedWorker.descriptor.createCommand,\n\t\t\t\t\t\tconfig: command.recoveryConfig,\n\t\t\t\t\t\tenv: command.env,\n\t\t\t\t\t\tlaunchEnv: command.launchEnv,\n\t\t\t\t\t\tlifecycle: \"client_owned\",\n\t\t\t\t\t};\n\t\t\t\t}\n\t\t\t\tif (!ownedWorker.launchEnv) {\n\t\t\t\t\tthrow new Error(\"Client-owned session recovery requires the owning client environment\");\n\t\t\t\t}\n\t\t\t\townedWorker.intentionalStop = false;\n\t\t\t\townedWorker.descriptor.stopRequestedAt = undefined;\n\t\t\t\townedWorker.descriptor.archiveOnStop = undefined;\n\t\t\t\townedWorker.descriptor.lifecycle = \"recovering\";\n\t\t\t\townedWorker.descriptor.consecutiveFailures = 0;\n\t\t\t\townedWorker.deferredRecoveryRounds = 0;\n\t\t\t\tthis.persistWorker(ownedWorker);\n\t\t\t\tawait this.recoverWorker(ownedWorker);\n\t\t\t}\n\t\t}\n\t\tconst match = await this.findWorkerForClient(client, command.activeSessionId);\n\t\tthis.requireAvailableWorkerClient(match.worker);\n\t\tconst activeSessionId = match.summary.activeSessionId ?? match.summary.id;\n\t\tconst duplicateValidation = this.currentSnapshotGeneration(match.worker, activeSessionId)?.validation;\n\t\tif (duplicateValidation) {\n\t\t\tawait duplicateValidation.promise;\n\t\t}\n\t\tif (command.clientId) {\n\t\t\tclient.id = command.clientId;\n\t\t}\n\t\tclient.capabilities = normalizeCapabilities(command.capabilities, command.supportsExtensionUi);\n\t\tclient.supportsExtensionUi = client.capabilities.has(\"extension_ui\");\n\n\t\tlet result = match.worker.snapshotCache.get(activeSessionId);\n\t\tif (\n\t\t\tresult &&\n\t\t\t!client.capabilities.has(\"chunked_snapshot\") &&\n\t\t\tresult.snapshot.messages.length < result.snapshot.summary.messageCount\n\t\t) {\n\t\t\tresult = undefined;\n\t\t}\n\t\tif (!result) {\n\t\t\tconst snapshotLoadKey = `${activeSessionId}:${client.capabilities.has(\"chunked_snapshot\") ? \"chunked\" : \"full\"}`;\n\t\t\tlet retryInvalidatedLoad = true;\n\t\t\twhile (!result) {\n\t\t\t\tlet loading = match.worker.snapshotLoads.get(snapshotLoadKey);\n\t\t\t\tif (!loading) {\n\t\t\t\t\tconst observedSnapshotId =\n\t\t\t\t\t\tmatch.worker.transcriptCaches.get(activeSessionId)?.snapshotId ??\n\t\t\t\t\t\tmatch.worker.snapshotCache.get(activeSessionId)?.snapshotStream?.id;\n\t\t\t\t\tloading = (async () => {\n\t\t\t\t\t\tconst workerClient = this.requireAvailableWorkerClient(match.worker);\n\t\t\t\t\t\tconst response = await workerClient.request({\n\t\t\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tcapabilities: client.capabilities.has(\"chunked_snapshot\")\n\t\t\t\t\t\t\t\t? [\"attach_snapshot\", \"event_sequence\", \"slim_attach\", \"chunked_snapshot\"]\n\t\t\t\t\t\t\t\t: [\"attach_snapshot\", \"event_sequence\", \"slim_attach\"],\n\t\t\t\t\t\t\tsupportsExtensionUi: false,\n\t\t\t\t\t\t\tenv: command.env ?? collectDaemonClientEnv(),\n\t\t\t\t\t\t});\n\t\t\t\t\t\tconst loaded = attachResultFromResponse(response);\n\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) !== loading) {\n\t\t\t\t\t\t\tthrow new SnapshotLoadInvalidatedError(\"Session snapshot changed during attach\");\n\t\t\t\t\t\t}\n\t\t\t\t\t\treturn this.cacheLoadedSnapshot(match.worker, activeSessionId, loaded, observedSnapshotId);\n\t\t\t\t\t})();\n\t\t\t\t\tmatch.worker.snapshotLoads.set(snapshotLoadKey, loading);\n\t\t\t\t\tvoid loading.then(\n\t\t\t\t\t\tasync (loaded) => {\n\t\t\t\t\t\t\ttry {\n\t\t\t\t\t\t\t\tconst snapshotId = loaded.snapshotStream?.id;\n\t\t\t\t\t\t\t\tconst transcript = snapshotId\n\t\t\t\t\t\t\t\t\t? this.snapshotGeneration(match.worker, loaded.activeSessionId, snapshotId)?.transcript\n\t\t\t\t\t\t\t\t\t: undefined;\n\t\t\t\t\t\t\t\tif (transcript && !transcript.complete) {\n\t\t\t\t\t\t\t\t\tlet chunkIndex = 0;\n\t\t\t\t\t\t\t\t\twhile (await transcript.waitForChunk(chunkIndex)) {\n\t\t\t\t\t\t\t\t\t\tchunkIndex++;\n\t\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t} catch {\n\t\t\t\t\t\t\t\t// Failed transfers must allow a fresh snapshot request.\n\t\t\t\t\t\t\t} finally {\n\t\t\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) === loading) {\n\t\t\t\t\t\t\t\t\tmatch.worker.snapshotLoads.delete(snapshotLoadKey);\n\t\t\t\t\t\t\t\t}\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t},\n\t\t\t\t\t\t() => {\n\t\t\t\t\t\t\tif (match.worker.snapshotLoads.get(snapshotLoadKey) === loading) {\n\t\t\t\t\t\t\t\tmatch.worker.snapshotLoads.delete(snapshotLoadKey);\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t},\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\ttry {\n\t\t\t\t\tresult = await loading;\n\t\t\t\t} catch (error) {\n\t\t\t\t\tif (!(error instanceof SnapshotLoadInvalidatedError)) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tif (!retryInvalidatedLoad) {\n\t\t\t\t\t\tthrow error;\n\t\t\t\t\t}\n\t\t\t\t\tretryInvalidatedLoad = false;\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tthis.requireAvailableWorkerClient(match.worker);\n\t\tconst wasAttached = client.attachedActiveSessionIds.has(activeSessionId);\n\t\tlet transcript: SnapshotTranscriptCache | undefined;\n\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\twhile (true) {\n\t\t\t\tconst validation = this.currentSnapshotGeneration(match.worker, activeSessionId)?.validation;\n\t\t\t\tif (validation) {\n\t\t\t\t\tawait validation.promise;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tresult = match.worker.snapshotCache.get(activeSessionId) ?? result;\n\t\t\t\ttranscript = this.getOrCreateTranscriptCache(match.worker, result);\n\t\t\t\tbreak;\n\t\t\t}\n\t\t}\n\t\tconst releaseTranscript = transcript?.retain();\n\t\tclient.attachedActiveSessionIds.add(activeSessionId);\n\t\ttry {\n\t\t\tconst publicSummary = this.publicSummary(match.worker, result.snapshot.summary);\n\t\t\tif (publicSummary.streamingMessage?.role === \"assistant\") {\n\t\t\t\tthis.streamReconstructor.seed(activeSessionId, publicSummary.streamingMessage);\n\t\t\t} else {\n\t\t\t\tfor (let index = result.snapshot.messages.length - 1; index >= 0; index--) {\n\t\t\t\t\tconst latestMessage = result.snapshot.messages[index];\n\t\t\t\t\tif (latestMessage?.role === \"assistant\") {\n\t\t\t\t\t\tthis.streamReconstructor.seed(activeSessionId, latestMessage);\n\t\t\t\t\t\tbreak;\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tconst publicResult: DaemonAttachResult = {\n\t\t\t\t...result,\n\t\t\t\tstate: result.state ? publicSummary : undefined,\n\t\t\t\tsnapshot: { ...result.snapshot, summary: publicSummary },\n\t\t\t\tclient: { id: client.id, capabilities: [...client.capabilities] },\n\t\t\t};\n\t\t\tif (publicResult.state && publicResult.messages) {\n\t\t\t\tthis.write(client, {\n\t\t\t\t\ttype: \"session_attached\",\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tstate: publicResult.state,\n\t\t\t\t\tmessages: publicResult.messages,\n\t\t\t\t\tsnapshot: publicResult.snapshot,\n\t\t\t\t\treplay: publicResult.replay,\n\t\t\t\t\tlastEventSequence: publicResult.lastEventSequence,\n\t\t\t\t});\n\t\t\t}\n\t\t\tvoid this.syncWorkerExtensionUi(activeSessionId);\n\t\t\tconst detachingSessions = this.detachingInputPauseSessions?.get(client);\n\t\t\tdetachingSessions?.delete(command.activeSessionId);\n\t\t\tdetachingSessions?.delete(activeSessionId);\n\t\t\treturn { result: publicResult, worker: match.worker, transcript, releaseTranscript };\n\t\t} catch (error) {\n\t\t\treleaseTranscript?.();\n\t\t\tif (!wasAttached) {\n\t\t\t\tclient.attachedActiveSessionIds.delete(activeSessionId);\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate cacheLoadedSnapshot(\n\t\tworker: ResidentWorker,\n\t\tactiveSessionId: string,\n\t\tloaded: DaemonAttachResult,\n\t\tobservedSnapshotId: string | undefined,\n\t): DaemonAttachResult {\n\t\tconst currentTranscript = worker.transcriptCaches.get(activeSessionId);\n\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\tconst currentResult = currentGeneration?.result ?? worker.snapshotCache.get(activeSessionId);\n\t\tconst currentSnapshotId = currentTranscript?.snapshotId ?? currentResult?.snapshotStream?.id;\n\t\tconst loadedSnapshotId = loaded.snapshotStream?.id;\n\t\tif (!loaded.snapshotStream) {\n\t\t\tif (currentSnapshotId && currentSnapshotId !== observedSnapshotId) {\n\t\t\t\treturn loaded;\n\t\t\t}\n\t\t\tworker.snapshotCache.set(activeSessionId, loaded);\n\t\t\treturn loaded;\n\t\t}\n\t\tif (\n\t\t\tcurrentSnapshotId &&\n\t\t\tcurrentSnapshotId !== loadedSnapshotId &&\n\t\t\t(currentSnapshotId !== observedSnapshotId ||\n\t\t\t\t(currentResult?.lastEventSequence ?? -1) > loaded.lastEventSequence)\n\t\t) {\n\t\t\treturn currentResult ?? loaded;\n\t\t}\n\t\tlet transcript = currentTranscript;\n\t\tif (transcript && transcript.snapshotId !== loaded.snapshotStream.id) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, transcript);\n\t\t\ttranscript = undefined;\n\t\t}\n\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\tlet generation = generations.get(loaded.snapshotStream.id);\n\t\tif (!transcript) {\n\t\t\ttranscript = generation?.transcript;\n\t\t}\n\t\tif (!transcript) {\n\t\t\ttranscript = new SnapshotTranscriptCache({\n\t\t\t\tactiveSessionId,\n\t\t\t\tsnapshotId: loaded.snapshotStream.id,\n\t\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\t\ttargetChunkBytes: loaded.snapshotStream.targetChunkBytes,\n\t\t\t});\n\t\t}\n\t\tif (!generation) {\n\t\t\tgeneration = {\n\t\t\t\ttranscript,\n\t\t\t\tresult: loaded,\n\t\t\t\tincoming: false,\n\t\t\t\tretired: false,\n\t\t\t};\n\t\t\tgenerations.set(loaded.snapshotStream.id, generation);\n\t\t} else {\n\t\t\tgeneration.result = loaded;\n\t\t\tgeneration.retired = false;\n\t\t}\n\t\tworker.transcriptCaches.set(activeSessionId, transcript);\n\t\tworker.snapshotCache.set(activeSessionId, loaded);\n\t\treturn loaded;\n\t}\n\n\tprivate getOrCreateTranscriptCache(worker: ResidentWorker, result: DaemonAttachResult): SnapshotTranscriptCache {\n\t\tconst activeSessionId = result.activeSessionId;\n\t\tconst existing = worker.transcriptCaches.get(activeSessionId);\n\t\tif (existing && (!result.snapshotStream || existing.snapshotId === result.snapshotStream.id)) {\n\t\t\treturn existing;\n\t\t}\n\t\tif (result.snapshot.messages.length < result.snapshot.summary.messageCount) {\n\t\t\tthrow new Error(\"Session snapshot generation changed before its transcript could be selected\");\n\t\t}\n\t\tif (existing) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, existing);\n\t\t}\n\t\tconst revision = createHash(\"sha256\")\n\t\t\t.update(\n\t\t\t\t`${activeSessionId}:${result.snapshot.summary.sessionId}:${result.lastEventSequence}:${result.snapshot.messages.length}`,\n\t\t\t)\n\t\t\t.digest(\"hex\")\n\t\t\t.slice(0, 16);\n\t\tconst transcript = new SnapshotTranscriptCache({\n\t\t\tactiveSessionId,\n\t\t\tsnapshotId: `${activeSessionId}-${revision}`,\n\t\t\tmessages: result.snapshot.messages,\n\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\ttargetChunkBytes: SNAPSHOT_TARGET_CHUNK_BYTES,\n\t\t});\n\t\tworker.transcriptCaches.set(activeSessionId, transcript);\n\t\tconst cachedResult = {\n\t\t\t...result,\n\t\t\tmessages: result.messages ? [] : undefined,\n\t\t\tsnapshot: { ...result.snapshot, messages: [] },\n\t\t};\n\t\tworker.snapshotCache.set(activeSessionId, cachedResult);\n\t\tthis.snapshotGenerationsFor(worker, activeSessionId).set(transcript.snapshotId, {\n\t\t\ttranscript,\n\t\t\tresult: cachedResult,\n\t\t\tincoming: false,\n\t\t\tretired: false,\n\t\t});\n\t\treturn transcript;\n\t}\n\n\tprivate createStreamedAttachResult(\n\t\tresult: DaemonAttachResult,\n\t\ttranscript: SnapshotTranscriptCache,\n\t): DaemonAttachResult {\n\t\treturn {\n\t\t\t...result,\n\t\t\tmessages: result.messages ? [] : undefined,\n\t\t\tsnapshot: { ...result.snapshot, messages: [] },\n\t\t\tsnapshotStream: {\n\t\t\t\tid: transcript.snapshotId,\n\t\t\t\tmessageCount: result.snapshot.summary.messageCount,\n\t\t\t\ttargetChunkBytes: transcript.targetChunkBytes,\n\t\t\t},\n\t\t};\n\t}\n\n\tprivate async streamSnapshot(\n\t\tclient: DaemonSocketClient,\n\t\tworker: ResidentWorker,\n\t\tresult: DaemonAttachResult,\n\t\ttranscript: SnapshotTranscriptCache,\n\t\tpurpose: \"attach\" | \"replacement\" | \"resync\" = \"attach\",\n\t\tretainedTranscriptRelease?: () => void,\n\t\treleaseSnapshotReservation = this.reserveSnapshotStream(client, result.activeSessionId),\n\t): Promise<void> {\n\t\tconst stream = result.snapshotStream;\n\t\tconst releaseTranscript = retainedTranscriptRelease ?? transcript.retain();\n\t\tif (!stream || client.socket.destroyed) {\n\t\t\treleaseSnapshotReservation();\n\t\t\treleaseTranscript();\n\t\t\treturn;\n\t\t}\n\t\tconst { messages: _messages, ...snapshotHeader } = result.snapshot;\n\t\ttry {\n\t\t\tif (\n\t\t\t\t!(await this.writeSnapshotRecord(client, {\n\t\t\t\t\ttype: \"session_snapshot_begin\",\n\t\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\t\tsnapshotId: stream.id,\n\t\t\t\t\tsnapshot: snapshotHeader,\n\t\t\t\t\tmessageCount: stream.messageCount,\n\t\t\t\t\ttargetChunkBytes: stream.targetChunkBytes,\n\t\t\t\t\tpurpose,\n\t\t\t\t}))\n\t\t\t) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tlet chunkCount = 0;\n\t\t\twhile (true) {\n\t\t\t\tlet chunk: Buffer | undefined;\n\t\t\t\ttry {\n\t\t\t\t\tchunk = await transcript.waitForChunk(chunkCount);\n\t\t\t\t} catch (error) {\n\t\t\t\t\tconst streamError = error instanceof Error ? error : new Error(String(error));\n\t\t\t\t\tthis.failWorkerSnapshotCache(worker, result.activeSessionId, streamError, false, stream.id);\n\t\t\t\t\tthrow streamError;\n\t\t\t\t}\n\t\t\t\tif (!chunk) {\n\t\t\t\t\tbreak;\n\t\t\t\t}\n\t\t\t\tif (!(await this.writeSnapshotBuffer(client, chunk))) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tchunkCount++;\n\t\t\t}\n\t\t\tawait this.writeSnapshotRecord(client, {\n\t\t\t\ttype: \"session_snapshot_end\",\n\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\tsnapshotId: stream.id,\n\t\t\t\tchunkCount,\n\t\t\t\tlastEventSequence: result.lastEventSequence,\n\t\t\t\tlastEventCursor: result.lastEventCursor,\n\t\t\t});\n\t\t} catch (error) {\n\t\t\tconst streamError = error instanceof Error ? error : new Error(String(error));\n\t\t\tif (!client.socket.destroyed) {\n\t\t\t\ttry {\n\t\t\t\t\tconst delivered = await this.writeSnapshotRecord(client, {\n\t\t\t\t\t\ttype: \"session_snapshot_failed\",\n\t\t\t\t\t\tactiveSessionId: result.activeSessionId,\n\t\t\t\t\t\tsnapshotId: stream.id,\n\t\t\t\t\t\terror: streamError.message,\n\t\t\t\t\t});\n\t\t\t\t\tif (!delivered && !client.socket.destroyed) {\n\t\t\t\t\t\tclient.socket.destroy(streamError);\n\t\t\t\t\t}\n\t\t\t\t} catch (deliveryError) {\n\t\t\t\t\tclient.socket.destroy(deliveryError instanceof Error ? deliveryError : new Error(String(deliveryError)));\n\t\t\t\t}\n\t\t\t}\n\t\t\tthrow streamError;\n\t\t} finally {\n\t\t\treleaseSnapshotReservation();\n\t\t\treleaseTranscript();\n\t\t}\n\t}\n\n\tprivate reserveSnapshotStream(client: DaemonSocketClient, activeSessionId: string): () => void {\n\t\tclient.snapshotStreaming = true;\n\t\tclient.snapshotActiveSessionIds ??= new Set();\n\t\tclient.snapshotActiveSessionIds.add(activeSessionId);\n\t\tclient.snapshotActiveSessionCounts ??= new Map();\n\t\tclient.snapshotActiveSessionCounts.set(\n\t\t\tactiveSessionId,\n\t\t\t(client.snapshotActiveSessionCounts.get(activeSessionId) ?? 0) + 1,\n\t\t);\n\t\tlet released = false;\n\t\treturn () => {\n\t\t\tif (released) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\treleased = true;\n\t\t\tconst streamCount = client.snapshotActiveSessionCounts?.get(activeSessionId) ?? 1;\n\t\t\tif (streamCount > 1) {\n\t\t\t\tclient.snapshotActiveSessionCounts?.set(activeSessionId, streamCount - 1);\n\t\t\t} else {\n\t\t\t\tclient.snapshotActiveSessionCounts?.delete(activeSessionId);\n\t\t\t\tclient.snapshotActiveSessionIds?.delete(activeSessionId);\n\t\t\t}\n\t\t\tclient.snapshotStreaming = (client.snapshotActiveSessionIds?.size ?? 0) > 0;\n\t\t\tif (!client.snapshotStreaming) {\n\t\t\t\tclient.backpressured = false;\n\t\t\t}\n\t\t\tif (!client.snapshotStreaming && client.catchupActiveSessionIds?.size) {\n\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t);\n\t\t\t}\n\t\t};\n\t}\n\n\tprivate writeSnapshotRecord(client: DaemonSocketClient, message: DaemonOutbound): Promise<boolean> {\n\t\treturn this.writeSnapshotBuffer(client, Buffer.from(serializeJsonLine(message)));\n\t}\n\n\tprivate async writeSnapshotBuffer(client: DaemonSocketClient, buffer: Uint8Array): Promise<boolean> {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn false;\n\t\t}\n\t\tif (this.writeSerialized(client, buffer)) {\n\t\t\treturn true;\n\t\t}\n\t\treturn new Promise<boolean>((resolveDrain) => {\n\t\t\tlet settled = false;\n\t\t\tconst finish = (value: boolean) => {\n\t\t\t\tif (settled) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tsettled = true;\n\t\t\t\tclient.socket.off(\"drain\", onDrain);\n\t\t\t\tclient.socket.off(\"close\", onClose);\n\t\t\t\tclient.socket.off(\"error\", onClose);\n\t\t\t\tresolveDrain(value);\n\t\t\t};\n\t\t\tconst onDrain = () => finish(true);\n\t\t\tconst onClose = () => finish(false);\n\t\t\tclient.socket.once(\"drain\", onDrain);\n\t\t\tclient.socket.once(\"close\", onClose);\n\t\t\tclient.socket.once(\"error\", onClose);\n\t\t});\n\t}\n\n\tprivate detachClient(client: DaemonSocketClient, activeSessionId?: string): void {\n\t\tconst targets = activeSessionId ? [activeSessionId] : [...client.attachedActiveSessionIds];\n\t\tfor (const selector of targets) {\n\t\t\tconst match = this.matchWorkers(selector)[0];\n\t\t\tconst resolvedId = match ? (match.summary.activeSessionId ?? match.summary.id) : selector;\n\t\t\tif (!client.attachedActiveSessionIds.delete(resolvedId)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tclient.catchupActiveSessionIds?.delete(resolvedId);\n\t\t\tclient.catchupPurposes?.delete(resolvedId);\n\t\t\tthis.write(client, { type: \"session_detached\", activeSessionId: resolvedId });\n\t\t\tvoid this.syncWorkerExtensionUi(resolvedId);\n\t\t\tvoid this.evictEmptySessionOnLastDetach(resolvedId);\n\t\t}\n\t}\n\n\tprivate async syncWorkerExtensionUi(activeSessionId: string): Promise<void> {\n\t\tconst match = this.matchWorkers(activeSessionId)[0];\n\t\tif (!match?.worker.client) {\n\t\t\treturn;\n\t\t}\n\t\tawait this.subscribeWorker(match.worker, match.summary.activeSessionId ?? match.summary.id).catch(\n\t\t\t() => undefined,\n\t\t);\n\t}\n\n\tprivate handleWorkerFrame(\n\t\tworker: ResidentWorker,\n\t\tframe: PrivateFrame<DaemonWorkerFrameHeader>,\n\t\tsource?: DaemonWorkerClient,\n\t): void {\n\t\tif (frame.header.kind !== \"outbound\") {\n\t\t\treturn;\n\t\t}\n\t\tif (source !== undefined && source !== worker.client && source !== worker.pendingClient) {\n\t\t\treturn;\n\t\t}\n\t\tworker.lastFrameAt = Date.now();\n\t\tthis.clearRosterStaleness(worker);\n\t\tconst {\n\t\t\toutboundType,\n\t\t\tactiveSessionId,\n\t\t\tsnapshotId: frameSnapshotId,\n\t\t\tsessionEventType,\n\t\t\tpayloadEncoding,\n\t\t\tsnapshotPurpose,\n\t\t} = frame.header;\n\t\tif (outboundType === \"roster_delta\") {\n\t\t\tthis.consumeWorkerRosterDelta(worker, frame.payload, source);\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"roster_heartbeat\") {\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"heartbeats_changed\") {\n\t\t\tworker.heartbeatSnapshotStale = true;\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_begin\" && activeSessionId) {\n\t\t\ttry {\n\t\t\t\tconst begin = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t{ type: \"session_snapshot_begin\" }\n\t\t\t\t>;\n\t\t\t\tif (\n\t\t\t\t\tbegin.type !== \"session_snapshot_begin\" ||\n\t\t\t\t\tbegin.activeSessionId !== activeSessionId ||\n\t\t\t\t\ttypeof begin.snapshotId !== \"string\" ||\n\t\t\t\t\t(frameSnapshotId !== undefined && frameSnapshotId !== begin.snapshotId) ||\n\t\t\t\t\ttypeof begin.targetChunkBytes !== \"number\" ||\n\t\t\t\t\t!begin.snapshot ||\n\t\t\t\t\t!isSessionSummary(begin.snapshot.summary)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid snapshot begin frame\");\n\t\t\t\t}\n\t\t\t\tconst publicSummary = this.publicSummary(worker, begin.snapshot.summary);\n\t\t\t\tconst snapshot = {\n\t\t\t\t\t...begin.snapshot,\n\t\t\t\t\tsummary: publicSummary,\n\t\t\t\t\tmessages: [],\n\t\t\t\t};\n\t\t\t\tconst result: DaemonAttachResult = {\n\t\t\t\t\tprotocol: DAEMON_PROTOCOL_INFO,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tsnapshot,\n\t\t\t\t\treplay: {\n\t\t\t\t\t\tstatus: \"complete\",\n\t\t\t\t\t\ttoSequence: snapshot.lastEventSequence,\n\t\t\t\t\t\t...(snapshot.lastEventCursor ? { toCursor: snapshot.lastEventCursor } : {}),\n\t\t\t\t\t},\n\t\t\t\t\tlastEventSequence: snapshot.lastEventSequence,\n\t\t\t\t\t...(snapshot.lastEventCursor ? { lastEventCursor: snapshot.lastEventCursor } : {}),\n\t\t\t\t\tsnapshotStream: {\n\t\t\t\t\t\tid: begin.snapshotId,\n\t\t\t\t\t\tmessageCount: begin.messageCount,\n\t\t\t\t\t\ttargetChunkBytes: begin.targetChunkBytes,\n\t\t\t\t\t},\n\t\t\t\t\tclient: { id: \"supervisor\", capabilities: [\"chunked_snapshot\"] },\n\t\t\t\t};\n\t\t\t\tconst generations = this.snapshotGenerationsFor(worker, activeSessionId);\n\t\t\t\tlet generation = generations.get(begin.snapshotId);\n\t\t\t\tif (generation?.incoming) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tnew Error(`Snapshot ${begin.snapshotId} restarted before completion`),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tbegin.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\t// Snapshot summaries/state include live fields (for example activity and attached client\n\t\t\t\t// counts) that can change without advancing the transcript sequence. Treat the stable\n\t\t\t\t// transfer envelope as identity; duplicate chunks and end metadata are still byte-checked.\n\t\t\t\tconst duplicate =\n\t\t\t\t\tgeneration?.transcript.complete === true &&\n\t\t\t\t\tgeneration.end !== undefined &&\n\t\t\t\t\tgeneration.result.snapshotStream?.messageCount === begin.messageCount &&\n\t\t\t\t\tgeneration.result.snapshotStream?.targetChunkBytes === begin.targetChunkBytes &&\n\t\t\t\t\tgeneration.result.lastEventSequence === result.lastEventSequence &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventSequence === result.snapshot.lastEventSequence &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventCursor?.generation === result.snapshot.lastEventCursor?.generation &&\n\t\t\t\t\tgeneration.result.snapshot.lastEventCursor?.sequence === result.snapshot.lastEventCursor?.sequence;\n\t\t\t\tif (generation?.transcript.complete && !duplicate) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tnew Error(`Snapshot ${begin.snapshotId} did not match the cached transfer`),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tbegin.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\t\t\tconst currentResult = currentGeneration?.result ?? worker.snapshotCache.get(activeSessionId);\n\t\t\t\tconst isOlderThanCurrent =\n\t\t\t\t\tcurrentGeneration !== undefined &&\n\t\t\t\t\tcurrentGeneration.transcript.snapshotId !== begin.snapshotId &&\n\t\t\t\t\tcurrentResult !== undefined &&\n\t\t\t\t\tresult.lastEventSequence < currentResult.lastEventSequence;\n\t\t\t\tif (isOlderThanCurrent && !generation) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (duplicate && generation) {\n\t\t\t\t\tgeneration.incoming = true;\n\t\t\t\t\tgeneration.duplicateChunkIndex = 0;\n\t\t\t\t\tgeneration.duplicateResult = result;\n\t\t\t\t\tgeneration.validation = this.createSnapshotDuplicateValidation();\n\t\t\t\t\tif (currentGeneration === generation) {\n\t\t\t\t\t\tworker.snapshotCache.delete(activeSessionId);\n\t\t\t\t\t}\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\tcurrentGeneration &&\n\t\t\t\t\tcurrentGeneration.transcript.snapshotId !== begin.snapshotId &&\n\t\t\t\t\t!isOlderThanCurrent\n\t\t\t\t) {\n\t\t\t\t\tif (!currentGeneration.transcript.complete && !currentGeneration.incoming) {\n\t\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\t\tworker,\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tnew Error(`Snapshot ${currentGeneration.transcript.snapshotId} was superseded`),\n\t\t\t\t\t\t\tfalse,\n\t\t\t\t\t\t\tcurrentGeneration.transcript.snapshotId,\n\t\t\t\t\t\t);\n\t\t\t\t\t} else {\n\t\t\t\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, currentGeneration.transcript);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t\tif (!generation) {\n\t\t\t\t\tconst transcript = new SnapshotTranscriptCache({\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\tsnapshotId: begin.snapshotId,\n\t\t\t\t\t\tcacheRoot: this.snapshotCacheRoot,\n\t\t\t\t\t\ttargetChunkBytes: begin.targetChunkBytes,\n\t\t\t\t\t});\n\t\t\t\t\tgeneration = {\n\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\tresult,\n\t\t\t\t\t\tincoming: false,\n\t\t\t\t\t\tretired: isOlderThanCurrent,\n\t\t\t\t\t};\n\t\t\t\t\tthis.snapshotGenerationsFor(worker, activeSessionId).set(begin.snapshotId, generation);\n\t\t\t\t}\n\t\t\t\tgeneration.result = result;\n\t\t\t\tgeneration.begin = Buffer.from(frame.payload);\n\t\t\t\tgeneration.end = undefined;\n\t\t\t\tgeneration.incoming = true;\n\t\t\t\tgeneration.duplicateChunkIndex = undefined;\n\t\t\t\tgeneration.duplicateResult = undefined;\n\t\t\t\tgeneration.validation = undefined;\n\t\t\t\tif (!isOlderThanCurrent) {\n\t\t\t\t\tgeneration.retired = false;\n\t\t\t\t\tworker.transcriptCaches.set(activeSessionId, generation.transcript);\n\t\t\t\t\tworker.snapshotCache.set(activeSessionId, result);\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tthis.log(`Invalid worker snapshot begin frame: ${String(error)}`);\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_chunk\" && activeSessionId) {\n\t\t\tconst snapshotId = frameSnapshotId ?? worker.transcriptCaches.get(activeSessionId)?.snapshotId;\n\t\t\tif (!snapshotId) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, snapshotId);\n\t\t\tif (generation?.incoming) {\n\t\t\t\ttry {\n\t\t\t\t\tconst duplicateIndex = generation.duplicateChunkIndex;\n\t\t\t\t\tif (duplicateIndex === undefined) {\n\t\t\t\t\t\tgeneration.transcript.appendEncodedChunk(Buffer.from(frame.payload));\n\t\t\t\t\t} else {\n\t\t\t\t\t\tconst chunk = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t\t\t{ type: \"session_snapshot_chunk\" }\n\t\t\t\t\t\t>;\n\t\t\t\t\t\tif (\n\t\t\t\t\t\t\tchunk.type !== \"session_snapshot_chunk\" ||\n\t\t\t\t\t\t\tchunk.activeSessionId !== activeSessionId ||\n\t\t\t\t\t\t\tchunk.snapshotId !== generation.transcript.snapshotId ||\n\t\t\t\t\t\t\tchunk.index !== duplicateIndex ||\n\t\t\t\t\t\t\t!generation.transcript.readChunk(duplicateIndex).equals(Buffer.from(frame.payload))\n\t\t\t\t\t\t) {\n\t\t\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t\t\t`Duplicate snapshot ${generation.transcript.snapshotId} did not match cached bytes`,\n\t\t\t\t\t\t\t);\n\t\t\t\t\t\t}\n\t\t\t\t\t\tgeneration.duplicateChunkIndex = duplicateIndex + 1;\n\t\t\t\t\t}\n\t\t\t\t} catch (error) {\n\t\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\t\tworker,\n\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\t\ttrue,\n\t\t\t\t\t\tgeneration.transcript.snapshotId,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_end\" && activeSessionId) {\n\t\t\tconst snapshotId = frameSnapshotId ?? worker.transcriptCaches.get(activeSessionId)?.snapshotId;\n\t\t\tif (!snapshotId) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst generation = this.snapshotGeneration(worker, activeSessionId, snapshotId);\n\t\t\tif (!generation?.incoming) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst transcript = generation.transcript;\n\t\t\ttry {\n\t\t\t\tconst duplicateChunkCount = generation.duplicateChunkIndex;\n\t\t\t\tif (duplicateChunkCount === undefined) {\n\t\t\t\t\ttranscript.markComplete();\n\t\t\t\t\tif (!generation.begin) {\n\t\t\t\t\t\tthrow new Error(`Snapshot ${transcript.snapshotId} has no begin frame`);\n\t\t\t\t\t}\n\t\t\t\t\tgeneration.end = Buffer.from(frame.payload);\n\t\t\t\t} else {\n\t\t\t\t\tconst end = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t\t{ type: \"session_snapshot_end\" }\n\t\t\t\t\t>;\n\t\t\t\t\tif (\n\t\t\t\t\t\tend.type !== \"session_snapshot_end\" ||\n\t\t\t\t\t\tend.activeSessionId !== activeSessionId ||\n\t\t\t\t\t\tend.snapshotId !== transcript.snapshotId ||\n\t\t\t\t\t\tend.chunkCount !== duplicateChunkCount ||\n\t\t\t\t\t\tend.chunkCount !== transcript.chunkCount ||\n\t\t\t\t\t\t!generation.end?.equals(frame.payload)\n\t\t\t\t\t) {\n\t\t\t\t\t\tthrow new Error(`Duplicate snapshot ${transcript.snapshotId} ended with different metadata`);\n\t\t\t\t\t}\n\t\t\t\t\tif (!generation.duplicateResult) {\n\t\t\t\t\t\tthrow new Error(`Duplicate snapshot ${transcript.snapshotId} has no result`);\n\t\t\t\t\t}\n\t\t\t\t\tgeneration.result = generation.duplicateResult;\n\t\t\t\t\tif (worker.transcriptCaches.get(activeSessionId) === transcript) {\n\t\t\t\t\t\tworker.snapshotCache.set(activeSessionId, generation.duplicateResult);\n\t\t\t\t\t}\n\t\t\t\t\tthis.settleSnapshotDuplicateValidation(generation);\n\t\t\t\t}\n\t\t\t\tgeneration.incoming = false;\n\t\t\t\tgeneration.duplicateChunkIndex = undefined;\n\t\t\t\tgeneration.duplicateResult = undefined;\n\t\t\t} catch (error) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t\ttranscript.snapshotId,\n\t\t\t\t);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst published = worker.transcriptCaches.get(activeSessionId) === transcript;\n\t\t\tif (generation.retired) {\n\t\t\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\t\t\ttranscript.dispose();\n\t\t\t}\n\t\t\tif (published && (snapshotPurpose === \"replacement\" || snapshotPurpose === \"catchup\")) {\n\t\t\t\tfor (const client of this.clients) {\n\t\t\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) continue;\n\t\t\t\t\tthis.queueCatchup(client, activeSessionId, snapshotPurpose === \"replacement\" ? \"replacement\" : \"resync\");\n\t\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (outboundType === \"session_snapshot_failed\" && activeSessionId) {\n\t\t\ttry {\n\t\t\t\tconst failed = JSON.parse(frame.payload.toString(\"utf8\")) as Extract<\n\t\t\t\t\tDaemonOutbound,\n\t\t\t\t\t{ type: \"session_snapshot_failed\" }\n\t\t\t\t>;\n\t\t\t\tif (\n\t\t\t\t\tfailed.type !== \"session_snapshot_failed\" ||\n\t\t\t\t\tfailed.activeSessionId !== activeSessionId ||\n\t\t\t\t\ttypeof failed.snapshotId !== \"string\" ||\n\t\t\t\t\ttypeof failed.error !== \"string\" ||\n\t\t\t\t\t(frameSnapshotId !== undefined && frameSnapshotId !== failed.snapshotId)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid snapshot failure frame\");\n\t\t\t\t}\n\t\t\t\tconst currentGeneration = this.currentSnapshotGeneration(worker, activeSessionId);\n\t\t\t\tconst generation =\n\t\t\t\t\tthis.snapshotGeneration(worker, activeSessionId, failed.snapshotId) ??\n\t\t\t\t\t(currentGeneration?.transcript.snapshotId === failed.snapshotId ? currentGeneration : undefined);\n\t\t\t\tif (!generation) {\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst published = worker.transcriptCaches.get(activeSessionId) === generation.transcript;\n\t\t\t\tthis.failWorkerSnapshotCache(worker, activeSessionId, new Error(failed.error), false, failed.snapshotId);\n\t\t\t\tif (published && (snapshotPurpose === \"replacement\" || snapshotPurpose === \"catchup\")) {\n\t\t\t\t\tfor (const client of this.clients) {\n\t\t\t\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) continue;\n\t\t\t\t\t\tthis.queueCatchup(\n\t\t\t\t\t\t\tclient,\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tsnapshotPurpose === \"replacement\" ? \"replacement\" : \"resync\",\n\t\t\t\t\t\t);\n\t\t\t\t\t\tvoid this.catchUpClient(client).catch((error) =>\n\t\t\t\t\t\t\tthis.log(`Failed to catch up client ${client.id}: ${String(error)}`),\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\tthis.failWorkerSnapshotCache(\n\t\t\t\t\tworker,\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\terror instanceof Error ? error : new Error(String(error)),\n\t\t\t\t\ttrue,\n\t\t\t\t);\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\t\tif (\n\t\t\toutboundType === \"daemon_hello\" ||\n\t\t\toutboundType === \"response\" ||\n\t\t\toutboundType === \"session_list_progress\" ||\n\t\t\toutboundType === \"session_list_item\" ||\n\t\t\toutboundType === \"session_attached\" ||\n\t\t\toutboundType === \"session_detached\" ||\n\t\t\t!activeSessionId\n\t\t) {\n\t\t\treturn;\n\t\t}\n\t\tlet publicPayload = frame.payload;\n\t\tlet decodedOutbound: DaemonOutbound | undefined;\n\t\tif (payloadEncoding === \"assistant-delta\") {\n\t\t\tlet compactValue: unknown;\n\t\t\ttry {\n\t\t\t\tcompactValue = JSON.parse(frame.payload.toString(\"utf8\"));\n\t\t\t} catch {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!isCompactAssistantDelta(compactValue)) {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tconst reconstructed = this.streamReconstructor.reconstruct(compactValue);\n\t\t\tif (!reconstructed) {\n\t\t\t\tthis.scheduleCompactCatchup(worker, activeSessionId);\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tpublicPayload = Buffer.from(serializeJsonLine(reconstructed));\n\t\t} else if (\n\t\t\tsessionEventType === \"message_start\" ||\n\t\t\tsessionEventType === \"message_end\" ||\n\t\t\toutboundType === \"session_replaced\" ||\n\t\t\toutboundType === \"session_resynced\" ||\n\t\t\toutboundType === \"session_closed\"\n\t\t) {\n\t\t\ttry {\n\t\t\t\tdecodedOutbound = JSON.parse(frame.payload.toString(\"utf8\")) as DaemonOutbound;\n\t\t\t\tthis.streamReconstructor.observe(decodedOutbound);\n\t\t\t} catch {\n\t\t\t\t// A malformed worker event is still isolated to this worker connection.\n\t\t\t}\n\t\t}\n\t\tconst replacementSnapshotFollows =\n\t\t\tdecodedOutbound?.type === \"session_replaced\" && decodedOutbound.snapshotFollows === true;\n\t\tthis.invalidateWorkerSnapshot(\n\t\t\tworker,\n\t\t\tactiveSessionId,\n\t\t\toutboundType === \"session_replaced\" ||\n\t\t\t\toutboundType === \"session_closed\" ||\n\t\t\t\tisFinalizedTranscriptEvent(sessionEventType),\n\t\t);\n\t\tfor (const client of this.clients) {\n\t\t\tif (!client.attachedActiveSessionIds.has(activeSessionId)) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (replacementSnapshotFollows && !client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (outboundType === \"extension_ui_request\" && !client.supportsExtensionUi) {\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (client.snapshotActiveSessionIds?.has(activeSessionId)) {\n\t\t\t\tthis.queueCatchup(client, activeSessionId, outboundType === \"session_replaced\" ? \"replacement\" : \"resync\");\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tif (client.backpressured === true) {\n\t\t\t\tthis.queueCatchup(client, activeSessionId, outboundType === \"session_replaced\" ? \"replacement\" : \"resync\");\n\t\t\t\tcontinue;\n\t\t\t}\n\t\t\tthis.writeSerialized(client, publicPayload);\n\t\t}\n\t\tif (\n\t\t\tdecodedOutbound?.type === \"session_closed\" &&\n\t\t\tdecodedOutbound.reason === \"shutdown\" &&\n\t\t\tactiveSessionId === worker.descriptor.rootActiveSessionId &&\n\t\t\t!this.shuttingDown\n\t\t) {\n\t\t\tworker.intentionalStop = true;\n\t\t\t// An exact stop owns its registration and descriptor cleanup until its\n\t\t\t// tuple assertions complete. A synchronous root shutdown event can arrive\n\t\t\t// before its request resolves, so leave both intact while it is active.\n\t\t\tif ((this.workerStopCounts?.get(worker) ?? 0) === 0) {\n\t\t\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\t\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate invalidateWorkerSnapshot(worker: ResidentWorker, activeSessionId: string, transcriptChanged = true): void {\n\t\tworker.snapshotCache.delete(activeSessionId);\n\t\tif (!transcriptChanged) {\n\t\t\treturn;\n\t\t}\n\t\tworker.snapshotLoads.delete(`${activeSessionId}:chunked`);\n\t\tworker.snapshotLoads.delete(`${activeSessionId}:full`);\n\t\tconst transcript = worker.transcriptCaches.get(activeSessionId);\n\t\tif (transcript) {\n\t\t\tthis.retireWorkerSnapshotCache(worker, activeSessionId, transcript);\n\t\t}\n\t}\n\n\tprivate scheduleCompactCatchup(worker: ResidentWorker, activeSessionId: string): void {\n\t\tif (this.compactCatchupInProgress.has(activeSessionId)) {\n\t\t\treturn;\n\t\t}\n\t\tthis.compactCatchupInProgress.add(activeSessionId);\n\t\tthis.invalidateWorkerSnapshot(worker, activeSessionId);\n\t\tconst clients = [...this.clients].filter((client) => client.attachedActiveSessionIds.has(activeSessionId));\n\t\tfor (const client of clients) {\n\t\t\tthis.queueCatchup(client, activeSessionId);\n\t\t}\n\t\tvoid Promise.all(clients.map((client) => this.catchUpClient(client)))\n\t\t\t.catch((error) => this.log(`Failed compact catch-up for ${activeSessionId}: ${String(error)}`))\n\t\t\t.finally(() => {\n\t\t\t\tthis.compactCatchupInProgress.delete(activeSessionId);\n\t\t\t});\n\t}\n\n\tprivate queueCatchup(\n\t\tclient: DaemonSocketClient,\n\t\tactiveSessionId: string,\n\t\tpurpose: \"replacement\" | \"resync\" = \"resync\",\n\t): void {\n\t\tif (!client.catchupActiveSessionIds) {\n\t\t\tclient.catchupActiveSessionIds = new Set();\n\t\t}\n\t\tclient.catchupActiveSessionIds.add(activeSessionId);\n\t\tclient.catchupPurposes ??= new Map();\n\t\tif (purpose === \"replacement\" || !client.catchupPurposes.has(activeSessionId)) {\n\t\t\tclient.catchupPurposes.set(activeSessionId, purpose);\n\t\t}\n\t}\n\n\tprivate catchUpClient(client: DaemonSocketClient): Promise<void> {\n\t\tif (client.catchupPromise) {\n\t\t\treturn client.catchupPromise;\n\t\t}\n\t\tif (client.snapshotStreaming || client.backpressured) {\n\t\t\treturn Promise.resolve();\n\t\t}\n\t\tconst catchup = this.drainClientCatchupQueue(client).finally(() => {\n\t\t\tif (client.catchupPromise === catchup) {\n\t\t\t\tclient.catchupPromise = undefined;\n\t\t\t}\n\t\t});\n\t\tclient.catchupPromise = catchup;\n\t\treturn catchup;\n\t}\n\n\tprivate async drainClientCatchupQueue(client: DaemonSocketClient): Promise<void> {\n\t\twhile (\n\t\t\t!client.socket.destroyed &&\n\t\t\t!client.snapshotStreaming &&\n\t\t\t!client.backpressured &&\n\t\t\tclient.catchupActiveSessionIds?.size\n\t\t) {\n\t\t\tawait this.drainClientCatchups(client);\n\t\t}\n\t}\n\n\tprivate async drainClientCatchups(client: DaemonSocketClient): Promise<void> {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn;\n\t\t}\n\t\tconst pending = [...(client.catchupActiveSessionIds ?? [])].map((activeSessionId) => ({\n\t\t\tactiveSessionId,\n\t\t\tpurpose: client.catchupPurposes?.get(activeSessionId) ?? (\"resync\" as const),\n\t\t}));\n\t\tclient.catchupActiveSessionIds?.clear();\n\t\tclient.catchupPurposes?.clear();\n\t\tfor (let index = 0; index < pending.length; index++) {\n\t\t\tconst { activeSessionId, purpose } = pending[index]!;\n\t\t\tlet releaseTranscript: (() => void) | undefined;\n\t\t\ttry {\n\t\t\t\tconst attached = await this.attachClient(client, {\n\t\t\t\t\ttype: \"attach\",\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tcapabilities: [...client.capabilities],\n\t\t\t\t\tsupportsExtensionUi: client.supportsExtensionUi,\n\t\t\t\t});\n\t\t\t\treleaseTranscript = attached.releaseTranscript;\n\t\t\t\tif (client.capabilities.has(\"chunked_snapshot\")) {\n\t\t\t\t\tconst transcript = attached.transcript;\n\t\t\t\t\tif (!transcript) {\n\t\t\t\t\t\tthrow new Error(\"Session worker did not provide a snapshot transcript\");\n\t\t\t\t\t}\n\t\t\t\t\tif (purpose === \"replacement\") {\n\t\t\t\t\t\tthis.write(client, {\n\t\t\t\t\t\t\ttype: \"session_replaced\",\n\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\tstate: attached.result.snapshot.state,\n\t\t\t\t\t\t\tmessages: [],\n\t\t\t\t\t\t\tsnapshotFollows: true,\n\t\t\t\t\t\t\tmeta: createDaemonEventMeta(\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tattached.result.lastEventSequence,\n\t\t\t\t\t\t\t\tundefined,\n\t\t\t\t\t\t\t\tattached.result.lastEventCursor?.generation,\n\t\t\t\t\t\t\t),\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t\tawait this.streamSnapshot(\n\t\t\t\t\t\tclient,\n\t\t\t\t\t\tattached.worker,\n\t\t\t\t\t\tthis.createStreamedAttachResult(attached.result, transcript),\n\t\t\t\t\t\ttranscript,\n\t\t\t\t\t\tpurpose,\n\t\t\t\t\t\treleaseTranscript,\n\t\t\t\t\t);\n\t\t\t\t\treleaseTranscript = undefined;\n\t\t\t\t\tcontinue;\n\t\t\t\t}\n\t\t\t\tconst meta = createDaemonEventMeta(\n\t\t\t\t\tactiveSessionId,\n\t\t\t\t\tattached.result.lastEventSequence,\n\t\t\t\t\tundefined,\n\t\t\t\t\tattached.result.lastEventCursor?.generation,\n\t\t\t\t);\n\t\t\t\tconst catchup: DaemonOutbound =\n\t\t\t\t\tpurpose === \"replacement\"\n\t\t\t\t\t\t? {\n\t\t\t\t\t\t\t\ttype: \"session_replaced\",\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tstate: attached.result.snapshot.state,\n\t\t\t\t\t\t\t\tmessages: attached.result.snapshot.messages,\n\t\t\t\t\t\t\t\tmeta,\n\t\t\t\t\t\t\t}\n\t\t\t\t\t\t: {\n\t\t\t\t\t\t\t\ttype: \"session_resynced\",\n\t\t\t\t\t\t\t\tactiveSessionId,\n\t\t\t\t\t\t\t\tsnapshot: attached.result.snapshot,\n\t\t\t\t\t\t\t\tmeta,\n\t\t\t\t\t\t\t};\n\t\t\t\tif (!this.write(client, catchup)) {\n\t\t\t\t\tfor (const remaining of pending.slice(index + 1)) {\n\t\t\t\t\t\tthis.queueCatchup(client, remaining.activeSessionId, remaining.purpose);\n\t\t\t\t\t}\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t} catch (error) {\n\t\t\t\treleaseTranscript?.();\n\t\t\t\tthis.log(`Failed to catch up client ${client.id} for ${activeSessionId}: ${String(error)}`);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async prepareUpdateRestart(): Promise<DaemonUpdateRestartManifest> {\n\t\tif (this.updateRestartPhase !== undefined) throw new Error(\"Daemon is already preparing an update restart\");\n\t\tthis.updateRestartPhase = \"draining\";\n\t\ttry {\n\t\t\tconst deadline = Date.now() + UPDATE_RESTART_PREPARE_DEADLINE_MS;\n\t\t\tconst abort = AbortSignal.timeout(Math.min(UPDATE_RESTART_MUTATION_DRAIN_TIMEOUT_MS, deadline - Date.now()));\n\t\t\tawait this.mutationDrain.waitForDrain(1, abort, \"Timed out draining daemon mutations for update restart\");\n\t\t\tthis.updateRestartPhase = \"fencing\";\n\t\t\tawait this.mutationDrain.waitForDrain(1, abort, \"Timed out draining daemon mutations for update restart\");\n\t\t\tconst manifest = await this.prepareUpdateRestartFenced(deadline);\n\t\t\tthis.updateRestartPhase = \"prepared\";\n\t\t\treturn manifest;\n\t\t} catch (error) {\n\t\t\tthis.updateRestartPhase = undefined;\n\t\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate async prepareUpdateRestartFenced(deadline: number): Promise<DaemonUpdateRestartManifest> {\n\t\tconst residents = [...this.workers.values()];\n\t\tconst unavailable = residents.find(\n\t\t\t(worker) =>\n\t\t\t\tthis.isWorkerStopping(worker) || worker.descriptor.lifecycle !== \"ready\" || worker.client === undefined,\n\t\t);\n\t\tif (unavailable) {\n\t\t\tthrow new Error(\n\t\t\t\t`Cannot prepare update restart while resident worker ${unavailable.descriptor.workerId} is ${this.effectiveWorkerState(unavailable)}${unavailable.client ? \"\" : \" and disconnected\"}`,\n\t\t\t);\n\t\t}\n\t\tconst workers = residents as Array<ResidentWorker & { client: DaemonWorkerClient }>;\n\t\tconst acknowledged: ResidentWorker[] = [];\n\t\tconst preparationResults = await Promise.allSettled(\n\t\t\tworkers.map(async (worker) => {\n\t\t\t\tconst client = worker.client;\n\t\t\t\tconst response = await client.requestWorker(\n\t\t\t\t\t{ type: \"worker_prepare_update\" },\n\t\t\t\t\tMath.max(1, Math.min(UPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS, deadline - Date.now())),\n\t\t\t\t);\n\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\tworker.updateRestartPrepareClient = client;\n\t\t\t\tacknowledged.push(worker);\n\t\t\t\tif (!response.data || typeof response.data !== \"object\") {\n\t\t\t\t\tthrow new Error(\"Worker returned an invalid update manifest\");\n\t\t\t\t}\n\t\t\t\tif (worker.client !== client || worker.descriptor.lifecycle !== \"ready\") {\n\t\t\t\t\tthrow new Error(`Worker ${worker.descriptor.workerId} disconnected during update preparation`);\n\t\t\t\t}\n\t\t\t\tconst manifest = response.data as DaemonUpdateRestartManifest;\n\t\t\t\tif (manifest.formatVersion !== DAEMON_UPDATE_RESTART_FORMAT_VERSION) {\n\t\t\t\t\tthrow new Error(`Worker returned unsupported update manifest version ${manifest.formatVersion}`);\n\t\t\t\t}\n\t\t\t\tif (\n\t\t\t\t\t!manifest.sessions.some(\n\t\t\t\t\t\t(session) => session.activeSessionId === worker.descriptor.rootActiveSessionId,\n\t\t\t\t\t) &&\n\t\t\t\t\t!manifest.discardedActiveSessionIds?.includes(worker.descriptor.rootActiveSessionId)\n\t\t\t\t) {\n\t\t\t\t\tthrow new Error(\n\t\t\t\t\t\t`Worker ${worker.descriptor.workerId} omitted its root disposition from the update manifest`,\n\t\t\t\t\t);\n\t\t\t\t}\n\t\t\t\treturn { worker, manifest };\n\t\t\t}),\n\t\t);\n\t\tconst cancelAcknowledged = async () => {\n\t\t\tawait Promise.all(\n\t\t\t\tacknowledged.map(async (worker) => {\n\t\t\t\t\tconst prepareClient = worker.updateRestartPrepareClient;\n\t\t\t\t\tworker.updateRestartPrepareClient = undefined;\n\t\t\t\t\tif (!prepareClient) return;\n\t\t\t\t\ttry {\n\t\t\t\t\t\tconst response = await prepareClient.requestWorker({ type: \"worker_cancel_update\" }, 5000);\n\t\t\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t`Could not cancel prepared worker ${worker.descriptor.workerId}; reconnecting it: ${String(error)}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t\tprepareClient.close();\n\t\t\t\t\t\tif (worker.client && worker.client !== prepareClient) worker.client.close();\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t};\n\t\tconst preparationFailure = preparationResults.find(\n\t\t\t(result): result is PromiseRejectedResult => result.status === \"rejected\",\n\t\t);\n\t\tif (preparationFailure) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow preparationFailure.reason;\n\t\t}\n\t\tconst prepared = preparationResults.flatMap((result) =>\n\t\t\tresult.status === \"fulfilled\" ? [result.value.worker] : [],\n\t\t);\n\t\tconst responses = preparationResults.flatMap((result) =>\n\t\t\tresult.status === \"fulfilled\" ? [result.value.manifest] : [],\n\t\t);\n\t\tconst discardedActiveSessionIds = responses.flatMap((manifest) => manifest.discardedActiveSessionIds ?? []);\n\t\tconst manifest: DaemonUpdateRestartManifest = {\n\t\t\tformatVersion: DAEMON_UPDATE_RESTART_FORMAT_VERSION,\n\t\t\tcreatedAt: new Date().toISOString(),\n\t\t\tsessions: responses.flatMap((manifest) => manifest.sessions),\n\t\t\t...(discardedActiveSessionIds.length > 0 ? { discardedActiveSessionIds } : {}),\n\t\t};\n\t\t// A worker that disconnected after preparing cancelled its checkpoint with\n\t\t// the old client; a recovered replacement may have admitted inputs past the\n\t\t// captured manifest. Abort before the manifest is persisted so the caller's\n\t\t// fallback cannot restore from the stale checkpoint.\n\t\tconst staleWorker = prepared.find((worker) => worker.client !== worker.updateRestartPrepareClient);\n\t\tif (staleWorker) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow new Error(\n\t\t\t\t`Worker ${staleWorker.descriptor.workerId} reconnected during update preparation; its checkpoint is stale`,\n\t\t\t);\n\t\t}\n\t\ttry {\n\t\t\tthis.validateAndPersistUpdateManifest(manifest);\n\t\t} catch (error) {\n\t\t\tawait cancelAcknowledged();\n\t\t\tthrow error;\n\t\t}\n\t\t// Commit through the connection that owns the prepared transaction; a client\n\t\t// swapped in after the check above must fail the commit rather than reach a\n\t\t// worker that no longer holds the checkpoint.\n\t\tconst commitClients = new Map(prepared.map((worker) => [worker, worker.updateRestartPrepareClient]));\n\t\tfor (const worker of prepared) worker.updateRestartPrepareClient = undefined;\n\t\tconst commitResults = await Promise.allSettled(\n\t\t\tprepared.map(async (worker) => {\n\t\t\t\tconst client = commitClients.get(worker);\n\t\t\t\tif (!client) throw new Error(`Worker ${worker.descriptor.workerId} disconnected before update commit`);\n\t\t\t\tconst response = await client.requestWorker(\n\t\t\t\t\t{ type: \"worker_commit_update\" },\n\t\t\t\t\tUPDATE_RESTART_WORKER_REQUEST_TIMEOUT_MS,\n\t\t\t\t);\n\t\t\t\tif (!response.success) throw new Error(response.error);\n\t\t\t}),\n\t\t);\n\t\tconst commitFailure = commitResults.find(\n\t\t\t(result): result is PromiseRejectedResult => result.status === \"rejected\",\n\t\t);\n\t\tif (commitFailure) {\n\t\t\tthis.log(`Update restart commit response failed; forcing restart completion: ${String(commitFailure.reason)}`);\n\t\t\tawait Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false, true)));\n\t\t\treturn manifest;\n\t\t}\n\t\tconst stopResults = await Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false)));\n\t\tif (stopResults.some((result) => result.status === \"rejected\")) {\n\t\t\tthis.log(\"A committed update worker did not stop gracefully; forcing restart completion\");\n\t\t\tawait Promise.allSettled(prepared.map((worker) => this.stopWorker(worker, false, true)));\n\t\t}\n\t\treturn manifest;\n\t}\n\n\tprivate validateAndPersistUpdateManifest(manifest: DaemonUpdateRestartManifest): void {\n\t\tif (manifest.formatVersion !== DAEMON_UPDATE_RESTART_FORMAT_VERSION) {\n\t\t\tthrow new Error(`Unsupported update manifest version ${manifest.formatVersion}`);\n\t\t}\n\t\tconst activeSessionIds = new Set<string>();\n\t\tconst sessionFiles = new Set<string>();\n\t\tfor (const discardedActiveSessionId of manifest.discardedActiveSessionIds ?? []) {\n\t\t\tif (!discardedActiveSessionId || activeSessionIds.has(discardedActiveSessionId)) {\n\t\t\t\tthrow new Error(\"Update manifest contains an invalid discarded session disposition\");\n\t\t\t}\n\t\t\tactiveSessionIds.add(discardedActiveSessionId);\n\t\t}\n\t\tfor (const session of manifest.sessions) {\n\t\t\tif (!session.activeSessionId || !session.sessionFile) {\n\t\t\t\tthrow new Error(\"Update manifest contains an incomplete session checkpoint\");\n\t\t\t}\n\t\t\tif (activeSessionIds.has(session.activeSessionId)) {\n\t\t\t\tthrow new Error(`Update manifest contains duplicate active session ${session.activeSessionId}`);\n\t\t\t}\n\t\t\tconst sessionFile = canonicalSessionPath(session.sessionFile);\n\t\t\tif (sessionFiles.has(sessionFile)) {\n\t\t\t\tthrow new Error(`Update manifest contains duplicate session file ${sessionFile}`);\n\t\t\t}\n\t\t\tactiveSessionIds.add(session.activeSessionId);\n\t\t\tsessionFiles.add(sessionFile);\n\t\t}\n\t\tconst agentDir = this.defaultSessionConfig.agentDir;\n\t\tif (!agentDir) {\n\t\t\tthrow new Error(\"Daemon supervisor config is missing agentDir\");\n\t\t}\n\t\tconst path = getDaemonUpdateRestartManifestPath(this.socketPath, agentDir);\n\t\tmkdirSync(dirname(path), { recursive: true, mode: 0o700 });\n\t\tconst tempPath = `${path}.${process.pid}.tmp`;\n\t\twriteFileSync(tempPath, `${JSON.stringify(manifest)}\\n`, { mode: 0o600 });\n\t\tchmodSync(tempPath, 0o600);\n\t\tconst validated = JSON.parse(readFileSync(tempPath, \"utf8\")) as DaemonUpdateRestartManifest;\n\t\tif (!Array.isArray(validated.sessions) || validated.sessions.length !== manifest.sessions.length) {\n\t\t\tthrow new Error(\"Could not validate aggregate update manifest\");\n\t\t}\n\t\trenameSync(tempPath, path);\n\t}\n\n\t/**\n\t * Verdict on whether a pid is still the process we launched. Callers must\n\t * be conservative in both directions: signal a pid only on \"current\"\n\t * (never SIGKILL a recycled pid), and clean up a registration only on\n\t * \"gone\"/\"replaced\" (never orphan a live worker because a transient\n\t * identity lookup failed).\n\t */\n\tprivate processIdentity(\n\t\tpid: number,\n\t\tprocessStartId: string | undefined,\n\t): \"current\" | \"replaced\" | \"gone\" | \"unknown\" {\n\t\tif (!isProcessAlive(pid)) {\n\t\t\treturn \"gone\";\n\t\t}\n\t\tif (processStartId === undefined) {\n\t\t\treturn \"unknown\";\n\t\t}\n\t\tconst observed = getProcessStartId(pid);\n\t\tif (observed === undefined) {\n\t\t\treturn \"unknown\";\n\t\t}\n\t\treturn observed === processStartId ? \"current\" : \"replaced\";\n\t}\n\n\t/**\n\t * Keep an exact stop's registration and descriptor authoritative while any\n\t * part of its cleanup is in flight. Root kills acquire this before forwarding\n\t * because a synchronous shutdown event may arrive before the worker replies.\n\t */\n\tprivate acquireWorkerStopOwnership(worker: ResidentWorker): () => void {\n\t\tif (!this.workerStopCounts) this.workerStopCounts = new Map();\n\t\tconst stopCounts = this.workerStopCounts;\n\t\tstopCounts.set(worker, (stopCounts.get(worker) ?? 0) + 1);\n\t\tlet released = false;\n\t\treturn () => {\n\t\t\tif (released) return;\n\t\t\treleased = true;\n\t\t\tconst remaining = (stopCounts.get(worker) ?? 1) - 1;\n\t\t\tif (remaining === 0) stopCounts.delete(worker);\n\t\t\telse stopCounts.set(worker, remaining);\n\t\t};\n\t}\n\n\tprivate async stopWorker(\n\t\tworker: ResidentWorker,\n\t\tremoveDescriptor: boolean,\n\t\tforce = false,\n\t\tarchiveSession = false,\n\t\trecoveryCleanup = false,\n\t\tdirectChild?: { child: ChildProcess; closed: Promise<void> },\n\t): Promise<void> {\n\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\tconst releaseStopOwnership = this.acquireWorkerStopOwnership(worker);\n\t\ttry {\n\t\t\tawait this.stopWorkerUntracked(worker, removeDescriptor, force, archiveSession, recoveryCleanup, directChild);\n\t\t} finally {\n\t\t\treleaseStopOwnership();\n\t\t}\n\t}\n\n\tprivate async stopWorkerUntracked(\n\t\tworker: ResidentWorker,\n\t\tremoveDescriptor: boolean,\n\t\tforce = false,\n\t\tarchiveSession = false,\n\t\trecoveryCleanup = false,\n\t\tdirectChild?: { child: ChildProcess; closed: Promise<void> },\n\t): Promise<void> {\n\t\tif (worker.ownerCleanupTimer) {\n\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t}\n\t\tif (!recoveryCleanup) {\n\t\t\tworker.stopRevision++;\n\t\t}\n\t\t// A retry can rescind this stop and relaunch the worker while we await\n\t\t// below. Bind every liveness check and signal to the process this stop\n\t\t// entered with, and abort cleanup once the stop no longer applies: the\n\t\t// pid changed (relaunched) or a removeDescriptor stop lost its tombstone\n\t\t// (rescinded, even before the successor pid lands).\n\t\tconst entryPid = worker.descriptor.pid;\n\t\tconst entryStartId = worker.descriptor.processStartId;\n\t\tconst assertStopStillApplies = () => {\n\t\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\t\tif (directChild) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (\n\t\t\t\tworker.descriptor.pid !== entryPid ||\n\t\t\t\t(removeDescriptor && worker.descriptor.stopRequestedAt === undefined)\n\t\t\t) {\n\t\t\t\tthrow new Error(`Session worker ${worker.descriptor.workerId} was relaunched during stop`);\n\t\t\t}\n\t\t};\n\t\ttry {\n\t\t\tif (removeDescriptor) {\n\t\t\t\tthis.persistWorkerStopTombstone(worker, archiveSession);\n\t\t\t} else {\n\t\t\t\tworker.intentionalStop = true;\n\t\t\t\tworker.descriptor.lifecycle = \"recovering\";\n\t\t\t\tthis.persistWorker(worker);\n\t\t\t}\n\t\t} catch (error) {\n\t\t\tif (!directChild) {\n\t\t\t\tthrow error;\n\t\t\t}\n\t\t\tthis.reportCleanupFailure(`worker rollback state ${worker.descriptor.workerId}`, error);\n\t\t}\n\t\tconst transferError = new Error(\"Session worker stopped during snapshot transfer\");\n\t\tconst generationTranscripts = new Set<SnapshotTranscriptCache>();\n\t\tfor (const [activeSessionId, generations] of [...(worker.snapshotGenerations ?? new Map())]) {\n\t\t\tfor (const generation of [...generations.values()]) {\n\t\t\t\tgenerationTranscripts.add(generation.transcript);\n\t\t\t\tif (generation.incoming || !generation.transcript.complete || generation.validation) {\n\t\t\t\t\tthis.failSnapshotGeneration(worker, activeSessionId, generation, transferError);\n\t\t\t\t} else {\n\t\t\t\t\tgeneration.transcript.dispose();\n\t\t\t\t\tthis.deleteSnapshotGeneration(worker, activeSessionId, generation);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\t\tfor (const transcript of worker.transcriptCaches.values()) {\n\t\t\tif (!generationTranscripts.has(transcript) && !transcript.complete) {\n\t\t\t\ttranscript.markFailed(transferError);\n\t\t\t}\n\t\t\ttranscript.dispose();\n\t\t}\n\t\tworker.transcriptCaches.clear();\n\t\tworker.snapshotCache.clear();\n\t\tworker.snapshotGenerations?.clear();\n\t\tif (worker.client) {\n\t\t\tif (archiveSession) {\n\t\t\t\tawait worker.client\n\t\t\t\t\t.requestWorker({ type: \"worker_archive_and_shutdown\" }, force ? 1000 : 5000)\n\t\t\t\t\t.catch(() => undefined);\n\t\t\t} else {\n\t\t\t\tawait worker.client.request({ type: \"shutdown\" }, force ? 1000 : 5000).catch(() => undefined);\n\t\t\t}\n\t\t\tassertStopStillApplies();\n\t\t\tworker.client.close();\n\t\t\tworker.client = undefined;\n\t\t} else if (directChild) {\n\t\t\tdirectChild.child.kill(\"SIGTERM\");\n\t\t} else if (this.processIdentity(entryPid, entryStartId) === \"current\") {\n\t\t\tsignalProcessGroupOrProcess(entryPid, \"SIGTERM\");\n\t\t}\n\t\t// Identity-aware in both directions: a replaced pid counts as gone (never\n\t\t// signal a recycled pid) while an unknown identity counts as alive (never\n\t\t// clean up a possibly-live worker on a transient lookup failure). kill(0)\n\t\t// runs on every poll; the expensive identity check is throttled.\n\t\tlet identityVerdict: \"current\" | \"replaced\" | \"gone\" | \"unknown\" = \"current\";\n\t\tlet identityCheckedAt = 0;\n\t\tconst isWorkerProcessAlive = () => {\n\t\t\tif (directChild) {\n\t\t\t\treturn directChild.child.exitCode === null && directChild.child.signalCode === null;\n\t\t\t}\n\t\t\tif (!processIdExists(entryPid)) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst now = Date.now();\n\t\t\tif (now - identityCheckedAt >= LIVENESS_IDENTITY_RECHECK_MS) {\n\t\t\t\tidentityCheckedAt = now;\n\t\t\t\tidentityVerdict = this.processIdentity(entryPid, entryStartId);\n\t\t\t}\n\t\t\treturn identityVerdict !== \"replaced\" && identityVerdict !== \"gone\";\n\t\t};\n\t\tconst gracefulDeadline = Date.now() + (force ? 500 : 2000);\n\t\twhile (isWorkerProcessAlive() && Date.now() < gracefulDeadline) {\n\t\t\tawait delay(25);\n\t\t}\n\t\tlet sigkillSent = false;\n\t\tif (force && isWorkerProcessAlive()) {\n\t\t\tassertStopStillApplies();\n\t\t\tif (directChild) {\n\t\t\t\tsigkillSent = directChild.child.kill(\"SIGKILL\");\n\t\t\t} else if (this.processIdentity(entryPid, entryStartId) === \"current\") {\n\t\t\t\t// Fresh, unthrottled check: the cached verdict may be up to 500ms\n\t\t\t\t// old, long enough for the pid to be recycled.\n\t\t\t\tsignalProcessGroupOrProcess(entryPid, \"SIGKILL\");\n\t\t\t\tsigkillSent = true;\n\t\t\t}\n\t\t\tconst forceDeadline = Date.now() + 1000;\n\t\t\twhile (isWorkerProcessAlive() && Date.now() < forceDeadline) {\n\t\t\t\tawait delay(25);\n\t\t\t}\n\t\t}\n\t\tif (isWorkerProcessAlive()) {\n\t\t\tworker.intentionalStop = worker.descriptor.stopRequestedAt !== undefined;\n\t\t\tif (removeDescriptor) {\n\t\t\t\tthis.scheduleWorkerStopFinalization(worker);\n\t\t\t}\n\t\t\tthrow new WorkerStopTimeoutError(\n\t\t\t\t`Session worker ${worker.descriptor.workerId} did not stop${sigkillSent ? \" after SIGKILL\" : \"\"}`,\n\t\t\t);\n\t\t}\n\t\tif (directChild) {\n\t\t\tawait directChild.closed;\n\t\t}\n\t\tassertStopStillApplies();\n\t\tif (removeDescriptor && worker.descriptor.archiveOnStop) {\n\t\t\tif (force) {\n\t\t\t\tthis.reclaimStoppedWorkerCronLock(worker);\n\t\t\t}\n\t\t\tawait this.finalizeArchivedWorkerStop(worker);\n\t\t\tassertStopStillApplies();\n\t\t}\n\t\tthis.invalidateWorkerSessionInputPauses(worker, \"Session worker stopped while input was paused\");\n\t\t// Client-owned schedules die with the registration, like their roster rows. Cancel\n\t\t// before the worker leaves the map, so no recompute sees the tree uncovered mid-stop.\n\t\tlet ephemeralCancelSettled = true;\n\t\tif (removeDescriptor && worker.descriptor.ownerClientId !== undefined) {\n\t\t\tephemeralCancelSettled = await this.cancelEphemeralWorkerScheduledJobs(worker);\n\t\t}\n\t\tthis.releaseWorkerRlmCapacity(worker);\n\t\tthis.workers.delete(worker.descriptor.workerId);\n\t\tthis.flipWorkerRosterEntriesInactive(worker);\n\t\t// A failed cancel keeps the stop tombstone as the durable intent; the enumeration retry or the next boot finishes it.\n\t\tif (removeDescriptor && ephemeralCancelSettled) {\n\t\t\tthis.deleteWorkerDescriptor(worker);\n\t\t}\n\t\tif (!this.shuttingDown) {\n\t\t\tthis.broadcastHeartbeatsChanged();\n\t\t}\n\t}\n\n\t/**\n\t * A stop that timed out leaves a tombstoned registration behind. Keep\n\t * escalating in the background until the process is gone, then finish the\n\t * interrupted cleanup instead of leaving a dead worker registered forever.\n\t */\n\tprivate scheduleWorkerStopFinalization(worker: ResidentWorker): void {\n\t\tif (worker.stopFinalization) {\n\t\t\treturn;\n\t\t}\n\t\tworker.stopFinalization = this.finalizeTimedOutWorkerStop(worker).finally(() => {\n\t\t\tworker.stopFinalization = undefined;\n\t\t});\n\t}\n\n\tprivate async finalizeTimedOutWorkerStop(worker: ResidentWorker): Promise<void> {\n\t\t// Bind to the exact process generation being stopped: a retry can rescind\n\t\t// the stop and relaunch with a new pid, and the OS can recycle the old\n\t\t// pid. The finalizer must never follow either successor.\n\t\tconst pid = worker.descriptor.pid;\n\t\tconst processStartId = worker.descriptor.processStartId;\n\t\tconst stopRevision = worker.stopRevision;\n\t\tconst isStopGenerationCurrent = () =>\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.stopRevision === stopRevision &&\n\t\t\tworker.descriptor.stopRequestedAt !== undefined &&\n\t\t\tworker.descriptor.pid === pid;\n\t\t// A replaced pid counts as gone (never SIGKILL a recycled pid); an\n\t\t// unobservable identity counts as alive (never clean up a possibly-live\n\t\t// worker). kill(0) probes every poll; ps-backed checks are throttled.\n\t\tlet stoppedVerdict = true;\n\t\tlet stoppedCanSignal = processStartId !== undefined;\n\t\tlet stoppedCheckedAt = 0;\n\t\tconst isStoppedProcessAlive = () => {\n\t\t\tif (!processIdExists(pid)) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tconst now = Date.now();\n\t\t\tif (now - stoppedCheckedAt < LIVENESS_IDENTITY_RECHECK_MS) {\n\t\t\t\treturn stoppedVerdict;\n\t\t\t}\n\t\t\tstoppedCheckedAt = now;\n\t\t\tif (!isProcessAlive(pid)) {\n\t\t\t\tstoppedVerdict = false;\n\t\t\t} else if (processStartId === undefined) {\n\t\t\t\tstoppedVerdict = true;\n\t\t\t\t// Without an identity captured while the original worker was known\n\t\t\t\t// alive, this pid may now belong to an unrelated process. Keep\n\t\t\t\t// waiting for it to disappear, but never escalate by pid alone.\n\t\t\t\tstoppedCanSignal = false;\n\t\t\t} else {\n\t\t\t\tconst observed = getProcessStartId(pid);\n\t\t\t\tstoppedVerdict = observed !== processStartId ? observed === undefined : true;\n\t\t\t\tstoppedCanSignal = observed === processStartId;\n\t\t\t}\n\t\t\treturn stoppedVerdict;\n\t\t};\n\t\tconst sigkillDeadline = Date.now() + STOP_FINALIZATION_SIGKILL_GRACE_MS;\n\t\tlet killed = false;\n\t\twhile (!this.shuttingDown) {\n\t\t\tif (!isStopGenerationCurrent()) {\n\t\t\t\treturn;\n\t\t\t}\n\t\t\tif (!isStoppedProcessAlive()) {\n\t\t\t\tbreak;\n\t\t\t}\n\t\t\tif (!killed && stoppedCanSignal && Date.now() >= sigkillDeadline) {\n\t\t\t\t// Fresh, unthrottled identity check right before signalling: the\n\t\t\t\t// cached verdict may be up to 500ms old, long enough for the pid\n\t\t\t\t// to be recycled by an unrelated process. A transiently\n\t\t\t\t// unobservable identity skips this attempt but keeps escalation\n\t\t\t\t// armed so a wedged worker is still killed on a later pass.\n\t\t\t\tconst observedNow = processStartId === undefined ? undefined : getProcessStartId(pid);\n\t\t\t\tif (processStartId === undefined || observedNow === processStartId) {\n\t\t\t\t\tsignalProcessGroupOrProcess(pid, \"SIGKILL\");\n\t\t\t\t\tkilled = true;\n\t\t\t\t}\n\t\t\t}\n\t\t\tawait unrefDelay(STOP_FINALIZATION_RECHECK_MS);\n\t\t}\n\t\t// Retry transient cleanup failures (for example catalog archival) so a\n\t\t// dead worker's registration is never stranded permanently. Each attempt\n\t\t// bumps the worker's stopRevision, so rescission is detected through the\n\t\t// registration and tombstone instead of the waiting-phase snapshot.\n\t\tconst isCleanupStillWanted = () =>\n\t\t\tthis.workers.get(worker.descriptor.workerId) === worker &&\n\t\t\tworker.descriptor.stopRequestedAt !== undefined &&\n\t\t\tworker.descriptor.pid === pid;\n\t\twhile (!this.shuttingDown && isCleanupStillWanted()) {\n\t\t\ttry {\n\t\t\t\tawait this.stopWorker(worker, true, true, worker.descriptor.archiveOnStop === true);\n\t\t\t\tthis.log(`Finalized timed-out stop for worker ${worker.descriptor.workerId}`);\n\t\t\t\treturn;\n\t\t\t} catch (error) {\n\t\t\t\tthis.reportCleanupFailure(`timed-out worker stop ${worker.descriptor.workerId}`, error);\n\t\t\t\tawait unrefDelay(STOP_FINALIZATION_RETRY_MS);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate async finalizeArchivedWorkerStop(worker: ResidentWorker): Promise<void> {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context) {\n\t\t\treturn;\n\t\t}\n\t\tif (worker.descriptor.rootSessionId) {\n\t\t\tconst cronStore = AgentCronJobStore.forSessionArtifacts();\n\t\t\tcronStore.registerSessionArtifact(worker.descriptor.rootSessionId, context.artifactDir);\n\t\t\tcronStore.cancelJobsForSession({\n\t\t\t\tsessionId: worker.descriptor.rootSessionId,\n\t\t\t\tsessionFile: context.sessionFile,\n\t\t\t});\n\t\t\tawait this.catalog.archive(context.sessionFile, worker.descriptor.rootSessionId);\n\t\t}\n\t}\n\n\tprivate async cancelScheduledJobsForSessionTree(\n\t\trootSessionId: string,\n\t\trootSessionFile: string,\n\t\tstillWanted?: () => boolean,\n\t\texclude?: ResidentWorker,\n\t): Promise<void> {\n\t\tconst store = AgentCronJobStore.forSessionArtifacts();\n\t\tconst sessions = [{ sessionId: rootSessionId, sessionFile: rootSessionFile }];\n\t\tconst childrenByParent = new Map<string, SessionInfo[]>();\n\t\tfor (const info of await this.rlmSpawnLedger().family()) {\n\t\t\tif (!info.parentSessionPath) continue;\n\t\t\tconst key = canonicalSessionPath(info.parentSessionPath);\n\t\t\tchildrenByParent.set(key, [...(childrenByParent.get(key) ?? []), info]);\n\t\t}\n\t\tconst queue = [canonicalSessionPath(rootSessionFile)];\n\t\tconst visited = new Set(queue);\n\t\twhile (queue.length > 0) {\n\t\t\tfor (const info of childrenByParent.get(queue.shift()!) ?? []) {\n\t\t\t\tconst child = canonicalSessionPath(info.path);\n\t\t\t\tif (visited.has(child)) continue;\n\t\t\t\tvisited.add(child);\n\t\t\t\tsessions.push({ sessionId: info.id, sessionFile: info.path });\n\t\t\t\tqueue.push(child);\n\t\t\t}\n\t\t}\n\t\t// A worker covering any tree member owns its stores again; a stale intent must not kill new schedules.\n\t\tfor (const { sessionFile } of sessions) {\n\t\t\ttry {\n\t\t\t\tif (this.findWorkerBySessionFile(sessionFile, exclude)) return;\n\t\t\t} catch {\n\t\t\t\treturn;\n\t\t\t}\n\t\t}\n\t\tlet registered = false;\n\t\tfor (const { sessionId, sessionFile } of sessions) {\n\t\t\tconst artifactDir = getSessionArtifactPathForFile(sessionFile, sessionId);\n\t\t\tif (!existsSync(join(artifactDir, SESSION_SCHEDULED_JOBS_FILENAME))) continue;\n\t\t\tstore.registerSessionArtifact(sessionId, artifactDir);\n\t\t\tregistered = true;\n\t\t}\n\t\tif (!registered) return;\n\t\t// Re-checked in the same synchronous turn as the walk: a promotion committed during the family read keeps its schedules.\n\t\tif (stillWanted && !stillWanted()) return;\n\t\tfor (const { sessionFile } of sessions) {\n\t\t\tstore.cancelJobsForSession({ sessionFile });\n\t\t}\n\t}\n\n\tprivate async cancelEphemeralWorkerScheduledJobs(worker: ResidentWorker): Promise<boolean> {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context || !worker.descriptor.rootSessionId) {\n\t\t\treturn true;\n\t\t}\n\t\ttry {\n\t\t\tawait this.cancelScheduledJobsForSessionTree(\n\t\t\t\tworker.descriptor.rootSessionId,\n\t\t\t\tcontext.sessionFile,\n\t\t\t\t() => worker.descriptor.ownerClientId !== undefined,\n\t\t\t\tworker,\n\t\t\t);\n\t\t\treturn true;\n\t\t} catch (error) {\n\t\t\t// A promotion that landed during the failed read means the cancel is no longer wanted.\n\t\t\tif (worker.descriptor.ownerClientId === undefined) {\n\t\t\t\treturn true;\n\t\t\t}\n\t\t\tthis.log(\n\t\t\t\t`Could not cancel scheduled jobs for client-owned worker ${worker.descriptor.workerId}: ${String(error)}`,\n\t\t\t);\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate reclaimStoppedWorkerCronLock(worker: ResidentWorker): void {\n\t\tconst context = this.workerSessionArtifactContext(worker);\n\t\tif (!context) {\n\t\t\treturn;\n\t\t}\n\t\trmSync(join(context.artifactDir, `${SESSION_SCHEDULED_JOBS_FILENAME}.lock`), { recursive: true, force: true });\n\t}\n\n\tprivate workerSessionArtifactContext(worker: {\n\t\tdescriptor: DaemonWorkerDescriptor;\n\t}): { sessionFile: string; artifactDir: string } | undefined {\n\t\tconst sessionFile = worker.descriptor.sessionFile ?? worker.descriptor.createCommand.sessionPath;\n\t\tif (!sessionFile || !worker.descriptor.rootSessionId) {\n\t\t\treturn undefined;\n\t\t}\n\t\treturn {\n\t\t\tsessionFile,\n\t\t\tartifactDir: getSessionArtifactPathForFile(sessionFile, worker.descriptor.rootSessionId),\n\t\t};\n\t}\n\n\tprivate persistWorkerStopTombstone(worker: ResidentWorker, archiveSession = false): void {\n\t\tif (worker.compatibilityError) throw worker.compatibilityError;\n\t\tworker.intentionalStop = true;\n\t\tworker.descriptor.stopRequestedAt ??= new Date().toISOString();\n\t\tworker.descriptor.archiveOnStop ||= archiveSession;\n\t\tthis.persistWorker(worker);\n\t}\n\n\tprivate write(client: DaemonSocketClient, message: DaemonOutbound): boolean {\n\t\treturn this.writeSerialized(client, serializeJsonLine(message));\n\t}\n\n\tprivate broadcastHeartbeatsChanged(): void {\n\t\tthis.scheduleScheduledSessionWakeRecompute();\n\t\tfor (const client of this.clients) {\n\t\t\tthis.write(client, { type: \"heartbeats_changed\" });\n\t\t}\n\t}\n\n\tprivate writeSerialized(client: DaemonSocketClient, line: string | Uint8Array): boolean {\n\t\tif (client.socket.destroyed) {\n\t\t\treturn false;\n\t\t}\n\t\tconst accepted = client.socket.write(line);\n\t\tif (!accepted) {\n\t\t\tclient.backpressured = true;\n\t\t}\n\t\treturn accepted;\n\t}\n\n\tprivate registerSignalHandlers(): void {\n\t\tconst signals: NodeJS.Signals[] = [\"SIGINT\", \"SIGTERM\"];\n\t\tif (process.platform !== \"win32\") {\n\t\t\tsignals.push(\"SIGHUP\");\n\t\t}\n\t\tfor (const signal of signals) {\n\t\t\tconst handler = () => void this.shutdown(signal === \"SIGINT\" ? 130 : signal === \"SIGHUP\" ? 129 : 143, false);\n\t\t\tprocess.on(signal, handler);\n\t\t\tthis.signalCleanupHandlers.push(() => process.off(signal, handler));\n\t\t}\n\t\tconst exitHandler = () => this.cleanupSocket();\n\t\tprocess.on(\"exit\", exitHandler);\n\t\tthis.signalCleanupHandlers.push(() => process.off(\"exit\", exitHandler));\n\t}\n\n\tprivate assertSocketLeaseHeld(): void {\n\t\tconst compromise = this.socketLeaseCompromise ?? this.socketLease?.compromise;\n\t\tif (compromise) throw new Error(`Daemon socket lease was compromised: ${compromise.message}`);\n\t}\n\n\tprivate assertSupervisorServing(allowWorkerLedger = false): void {\n\t\tthis.assertSocketLeaseHeld();\n\t\tif (this.shuttingDown && (!allowWorkerLedger || this.rlmJournalClose !== undefined)) {\n\t\t\tconst error = new Error(`Daemon supervisor generation ${this.generation} is shutting down; retry the command`);\n\t\t\tObject.assign(error, { code: \"supervisor_generation_stale\" as const });\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\tprivate fenceSupervisorSocket(): void {\n\t\ttry {\n\t\t\tthis.server?.close();\n\t\t} catch {\n\t\t\t// The server may already be closed by a concurrent shutdown.\n\t\t}\n\t\tfor (const client of this.clients) {\n\t\t\tclient.detachInput();\n\t\t\tclient.socket.destroy();\n\t\t}\n\t}\n\n\tprivate handleSocketLeaseCompromised(error: Error): void {\n\t\tif (this.socketLeaseCompromise) return;\n\t\tthis.socketLeaseCompromise = error;\n\t\tthis.shuttingDown = true;\n\t\tthis.fenceRlmJournalOwner();\n\t\tthis.fenceSupervisorSocket();\n\t\tconst message = `Daemon socket lease was compromised; relinquishing supervisor ownership: ${error.message}`;\n\t\ttry {\n\t\t\tthis.log(message);\n\t\t} catch {\n\t\t\tconsole.error(message);\n\t\t}\n\t\tif (!this.startupComplete) return;\n\t\tvoid this.cleanupSupervisorResources().catch((cleanupError) =>\n\t\t\tthis.reportCleanupFailure(\"compromised daemon socket lease\", cleanupError),\n\t\t);\n\t}\n\n\tprivate cleanupSocket(): void {\n\t\tif (!this.ownsSocketPath) {\n\t\t\treturn;\n\t\t}\n\t\tthis.ownsSocketPath = false;\n\t\tconst identity = this.socketIdentity;\n\t\tthis.socketIdentity = undefined;\n\t\tcleanupDaemonSocketPath(this.socketPath, identity, this.socketLease);\n\t}\n\n\tprivate async cleanupSupervisorResources(): Promise<void> {\n\t\tif (this.cleanupPromise) {\n\t\t\treturn this.cleanupPromise;\n\t\t}\n\t\tthis.cleanupPromise = this.cleanupSupervisorResourcesOnce();\n\t\treturn this.cleanupPromise;\n\t}\n\n\tprivate async cleanupSupervisorResourcesOnce(): Promise<void> {\n\t\tthis.shuttingDown = true;\n\t\tthis.clearIdleEvictionTimer();\n\t\tthis.clearScheduledWakeTimer();\n\t\tthis.clearRosterWatchdogTimer();\n\t\tawait this.idleEvictionSweep?.catch(() => undefined);\n\t\tfor (const cleanup of this.signalCleanupHandlers.splice(0)) {\n\t\t\tawait this.runCleanupStep(\"signal handler\", cleanup);\n\t\t}\n\t\tconst server = this.server;\n\t\tthis.server = undefined;\n\t\tconst serverClosed = new Promise<void>((resolveClose) => {\n\t\t\tif (!server?.listening) {\n\t\t\t\tresolveClose();\n\t\t\t\treturn;\n\t\t\t}\n\t\t\ttry {\n\t\t\t\tserver.close(() => resolveClose());\n\t\t\t} catch (error) {\n\t\t\t\tthis.reportCleanupFailure(\"daemon server\", error);\n\t\t\t\tresolveClose();\n\t\t\t}\n\t\t});\n\t\tfor (const client of this.clients) {\n\t\t\tclient.attachedActiveSessionIds.clear();\n\t\t\tawait this.runCleanupStep(`daemon client input ${client.id}`, () => client.detachInput());\n\t\t\tawait this.runCleanupStep(`daemon client socket ${client.id}`, () => {\n\t\t\t\tclient.socket.destroy();\n\t\t\t});\n\t\t}\n\t\tthis.clients.clear();\n\t\tfor (const worker of this.workers.values()) {\n\t\t\tif (worker.ownerCleanupTimer) {\n\t\t\t\tclearTimeout(worker.ownerCleanupTimer);\n\t\t\t\tworker.ownerCleanupTimer = undefined;\n\t\t\t}\n\t\t\tawait this.runCleanupStep(`worker client ${worker.descriptor.workerId}`, () => worker.client?.close());\n\t\t\tworker.client = undefined;\n\t\t\tconst transcripts = new Set(worker.transcriptCaches.values());\n\t\t\tfor (const generations of worker.snapshotGenerations?.values() ?? []) {\n\t\t\t\tfor (const generation of generations.values()) {\n\t\t\t\t\ttranscripts.add(generation.transcript);\n\t\t\t\t\tthis.settleSnapshotDuplicateValidation(\n\t\t\t\t\t\tgeneration,\n\t\t\t\t\t\tnew Error(\"Daemon supervisor stopped during snapshot transfer\"),\n\t\t\t\t\t);\n\t\t\t\t\tif (!generation.transcript.complete) {\n\t\t\t\t\t\tgeneration.transcript.markFailed(new Error(\"Daemon supervisor stopped during snapshot transfer\"));\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t}\n\t\t\tfor (const transcript of transcripts) {\n\t\t\t\tawait this.runCleanupStep(`worker transcript ${worker.descriptor.workerId}`, () => transcript.dispose());\n\t\t\t}\n\t\t\tworker.transcriptCaches.clear();\n\t\t\tworker.snapshotGenerations?.clear();\n\t\t\tworker.snapshotCache.clear();\n\t\t\tworker.snapshotLoads.clear();\n\t\t}\n\t\tthis.workers.clear();\n\t\tthis.openingWorkers.clear();\n\t\tawait this.runCleanupStep(\"daemon catalog\", () => this.catalog.stop());\n\t\tawait this.runCleanupStep(\"daemon server\", () => serverClosed);\n\t\tawait this.runCleanupStep(\"daemon socket\", () => this.cleanupSocket());\n\t\tawait this.runCleanupStep(\"supervisor cache\", () => {\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t});\n\t\tawait this.closeRlmJournalOwner();\n\t\tconst lease = this.socketLease;\n\t\tthis.socketLease = undefined;\n\t\tawait this.runCleanupStep(\"daemon socket lock\", async () => lease?.release());\n\t\tconst ownership = this.ownership;\n\t\tthis.ownership = undefined;\n\t\tawait this.runCleanupStep(\"daemon ownership\", async () => ownership?.release());\n\t}\n\n\tprivate async runCleanupStep(label: string, action: () => void | Promise<void>): Promise<void> {\n\t\ttry {\n\t\t\tawait action();\n\t\t} catch (error) {\n\t\t\tthis.reportCleanupFailure(label, error);\n\t\t}\n\t}\n\n\tprivate reportCleanupFailure(label: string, error: unknown): void {\n\t\tconst message = `Failed to clean up ${label}: ${String(error)}`;\n\t\ttry {\n\t\t\tthis.log(message);\n\t\t} catch {\n\t\t\tconsole.error(message);\n\t\t}\n\t}\n\n\tprivate async shutdown(\n\t\texitCode: number,\n\t\tstopWorkers: boolean,\n\t\trelaunch = false,\n\t\tforceWorkers = false,\n\t\tclosingReason?: DaemonClosingReason,\n\t): Promise<never> {\n\t\tif (this.shuttingDown) {\n\t\t\tprocess.exit(exitCode);\n\t\t}\n\t\tthis.shuttingDown = true;\n\t\tthis.clearIdleEvictionTimer();\n\t\tthis.clearScheduledWakeTimer();\n\t\tthis.clearRosterWatchdogTimer();\n\t\tawait this.idleEvictionSweep?.catch(() => undefined);\n\t\tif (closingReason) {\n\t\t\tfor (const client of this.clients) {\n\t\t\t\tthis.write(client, { type: \"daemon_closing\", reason: closingReason });\n\t\t\t}\n\t\t}\n\t\tfor (const cleanup of this.signalCleanupHandlers) {\n\t\t\tcleanup();\n\t\t}\n\t\tif (stopWorkers) {\n\t\t\tawait Promise.all(\n\t\t\t\t[...this.workers.values()].map(async (worker) => {\n\t\t\t\t\ttry {\n\t\t\t\t\t\tawait this.stopWorker(worker, true, forceWorkers, true);\n\t\t\t\t\t} catch (error) {\n\t\t\t\t\t\tif (error instanceof DaemonWorkerCompatibilityError) {\n\t\t\t\t\t\t\tthis.log(`Keeping live incompatible worker ${worker.descriptor.workerId}: ${error.message}`);\n\t\t\t\t\t\t\treturn;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tif (!(error instanceof WorkerStopTimeoutError)) {\n\t\t\t\t\t\t\tthrow error;\n\t\t\t\t\t\t}\n\t\t\t\t\t\tthis.log(\n\t\t\t\t\t\t\t`Worker ${worker.descriptor.workerId} remains tombstoned for recovery after shutdown: ${error.message}`,\n\t\t\t\t\t\t);\n\t\t\t\t\t}\n\t\t\t\t}),\n\t\t\t);\n\t\t\tif (!this.hasPersistedWorkerDescriptors()) {\n\t\t\t\trmSync(this.supervisorConfigPath, { force: true });\n\t\t\t}\n\t\t} else {\n\t\t\tfor (const worker of this.workers.values()) {\n\t\t\t\tif (worker.compatibilityError) continue;\n\t\t\t\tworker.intentionalStop = true;\n\t\t\t\tworker.client?.close();\n\t\t\t\tworker.client = undefined;\n\t\t\t}\n\t\t}\n\t\tawait this.catalog.stop();\n\t\tfor (const client of this.clients) {\n\t\t\tclient.detachInput();\n\t\t\tclient.socket.end();\n\t\t}\n\t\tawait new Promise<void>((resolveClose) => this.server?.close(() => resolveClose()) ?? resolveClose());\n\t\tawait this.runCleanupStep(\"daemon socket\", () => this.cleanupSocket());\n\t\tawait this.runCleanupStep(\"supervisor cache\", () => {\n\t\t\trmSync(this.snapshotCacheRoot, { recursive: true, force: true });\n\t\t});\n\t\tawait this.closeRlmJournalOwner();\n\t\tconst lease = this.socketLease;\n\t\tthis.socketLease = undefined;\n\t\tawait this.runCleanupStep(\"daemon socket lock\", async () => lease?.release());\n\t\tconst ownership = this.ownership;\n\t\tthis.ownership = undefined;\n\t\tawait this.runCleanupStep(\"daemon ownership\", async () => ownership?.release());\n\t\tif (relaunch) {\n\t\t\tconst launch = createCliSubprocessLaunchSpec([\"--mode\", \"daemon\", \"--daemon-socket\", this.socketPath]);\n\t\t\tconst environment = createCliSubprocessEnv();\n\t\t\tdelete environment[DAEMON_CATALOG_ROLE_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_ROLE_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_TOKEN_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_ACTIVE_SESSION_ID_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_SUPERVISOR_SOCKET_ENV];\n\t\t\tdelete environment[DAEMON_WORKER_RECOVERY_JOURNAL_ENV];\n\t\t\tdelete environment[ORPHAN_PROCESS_JOURNAL_ENV];\n\t\t\tdelete environment[SESSION_LEASES_ENABLED_ENV];\n\t\t\tdelete environment[SESSION_LEASE_OWNER_ID_ENV];\n\t\t\tconst replacement = spawn(launch.command, launch.args, {\n\t\t\t\tcwd: this.defaultSessionConfig.cwd ?? process.cwd(),\n\t\t\t\tdetached: true,\n\t\t\t\tenv: environment,\n\t\t\t\tstdio: \"ignore\",\n\t\t\t});\n\t\t\treplacement.unref();\n\t\t}\n\t\tprocess.exit(exitCode);\n\t}\n}\n"]}