@kolisachint/hoocode-agent 0.5.21 → 0.5.23

package/dist/core/lifeguard.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"lifeguard.d.ts","sourceRoot":"","sources":["../../src/core/lifeguard.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAqC3C,MAAM,WAAW,gBAAgB;IAChC,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,YAAY,CAAC;CACtB;AAED;;;;GAIG;AACH,qBAAa,iBAAkB,SAAQ,YAAY;IAClD,OAAO,CAAC,SAAS,CAAuC;IACxD,OAAO,CAAC,aAAa,CAA6B;IAClD,OAAO,CAAC,QAAQ,CAAqC;IACrD,4EAA4E;IAC5E,OAAO,CAAC,SAAS,CAA6B;IAC9C,gFAAgF;IAChF,OAAO,CAAC,aAAa,CAA6B;IAClD,OAAO,CAAC,aAAa,CAA+B;IACpD,+EAA+E;IAC/E,OAAO,CAAC,WAAW,CAAc;IACjC;;;;;OAKG;IACH,OAAO,CAAC,YAAY,CAAK;IACzB;;;;OAIG;IACH,OAAO,CAAC,OAAO,CAAqB;IACpC,OAAO,CAAC,QAAQ,CAAS;IACzB,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAS;IAC7B,OAAO,CAAC,qBAAqB,CAAC,CAAa;IAE3C,YAAY,GAAG,EAAE,MAAM,EAOtB;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAEnC;IAED;;;;;OAKG;IACH,OAAO,CAAC,cAAc;IAMtB;;;OAGG;IACH,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,IAAI,EAAE,YAAY,GAAG,IAAI,CAwBrE;IAED,+CAA+C;IAC/C,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAIrC;IAED,0DAA0D;IAC1D,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAE9C;IAED,qDAAqD;IACrD,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAErC;IAED,iDAAiD;IACjD,OAAO,IAAI,IAAI,CA4Bd;IAED,OAAO,CAAC,eAAe;IAqBvB,OAAO,CAAC,aAAa;IAsBrB,OAAO,CAAC,aAAa;IAiCrB;;;;;;;OAOG;IACH,OAAO,CAAC,QAAQ;IAQhB,OAAO,CAAC,OAAO;IAaf,OAAO,CAAC,uBAAuB;IAQ/B,OAAO,CAAC,gBAAgB;IAuBxB,OAAO,CAAC,cAAc;IAyBtB,OAAO,CAAC,aAAa;IAcrB,OAAO,CAAC,IAAI;CAoBZ","sourcesContent":["import type { ChildProcess } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, readdirSync, readFileSync, rmdirSync, statSync, unlinkSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { getDispatchRoot } from \"../config.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\n\nconst TIMEOUTS_MS: Record<string, number> = {\n\texplore: 5 * 60 * 1000,\n\tedit: 10 * 60 * 1000,\n\ttest: 10 * 60 * 1000,\n\treview: 8 * 60 * 1000,\n\tdoc: 5 * 60 * 1000,\n};\n\nconst HEARTBEAT_MISS_THRESHOLD_MS = 60000;\nconst HEARTBEAT_CHECK_INTERVAL_MS = 5000;\nconst PARENT_SHUTDOWN_GRACE_MS = 5000;\n\n/**\n * Concurrency tolerance. Each *additional* concurrently-monitored subagent adds\n * this fraction to the heartbeat-miss and hard-timeout budgets.\n *\n * When several subagents run at once (plus background MCP tools), they saturate\n * the CPU and starve the parent's event loop: its `setInterval` heartbeat check\n * fires late, and it cannot read a child's `{\"ping\":true}` line in time even\n * though the child is healthy and still working. Scaling the budgets by load\n * stops that contention from false-positive SIGKILLing healthy subagents — the\n * failure the demo hit when running many agents + MCP tools in the background.\n */\nconst LOAD_TOLERANCE_PER_PROCESS = 0.5;\n\n/**\n * Hard ceiling on the load tolerance multiplier, so a genuinely stuck subagent\n * is still eventually reaped no matter how busy the pool is.\n */\nconst MAX_LOAD_MULTIPLIER = 4;\n\nexport interface LifeguardProcess {\n\tpid: number;\n\ttask_id: string;\n\tagent_type: string;\n\tprocess: ChildProcess;\n}\n\n/**\n * Monitors running subagent processes for heartbeats, hard timeouts,\n * and parent-exit cleanup. Emits \"stalled\" and \"timeout\" events when\n * processes are terminated.\n */\nexport class SubagentLifeguard extends EventEmitter {\n\tprivate processes = new Map<string, LifeguardProcess>();\n\tprivate lastHeartbeat = new Map<string, number>();\n\tprivate timeouts = new Map<string, NodeJS.Timeout>();\n\t/** When each task started, used to compute the load-scaled hard timeout. */\n\tprivate startedAt = new Map<string, number>();\n\t/** Per-agent base hard timeout (before load scaling), captured at monitor(). */\n\tprivate baseTimeoutMs = new Map<string, number>();\n\tprivate checkInterval: NodeJS.Timeout | null = null;\n\t/** Wall-clock time the heartbeat check last ran, to measure event-loop lag. */\n\tprivate lastCheckAt = Date.now();\n\t/**\n\t * Count of external in-process tasks (e.g. background MCP tools) running in the\n\t * parent alongside the monitored subagents. These don't show up in `processes`\n\t * but still saturate the parent's CPU/event loop, so they're folded into the\n\t * load multiplier. Updated by the pool via setExternalLoad().\n\t */\n\tprivate externalLoad = 0;\n\t/**\n\t * Tasks already reaped (stalled/timeout kill sent) but whose `exit` has not\n\t * fired yet. The heartbeat check runs every 5s, so without this a stalled task\n\t * would re-emit \"stalled\" (and re-kill) on every tick until the process exits.\n\t */\n\tprivate reaping = new Set<string>();\n\tprivate disposed = false;\n\tprivate readonly cwd: string;\n\tprivate parentShutdownHandler?: () => void;\n\n\tconstructor(cwd: string) {\n\t\tsuper();\n\t\tthis.cwd = cwd;\n\t\tthis.setupParentExitHandlers();\n\t\tthis.sweepOldAgents();\n\t\tthis.lastCheckAt = Date.now();\n\t\tthis.checkInterval = setInterval(() => this.checkHeartbeats(), HEARTBEAT_CHECK_INTERVAL_MS);\n\t}\n\n\t/**\n\t * Set the count of external in-process tasks (background MCP tools) sharing the\n\t * parent's CPU/event loop. Folded into loadMultiplier() so concurrent MCP work\n\t * widens the heartbeat/timeout budgets just like extra monitored subagents do.\n\t * Negative values are clamped to 0.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.externalLoad = Math.max(0, Math.floor(count));\n\t}\n\n\t/**\n\t * Tolerance multiplier for the current load. 1 concurrent task → 1x; each extra\n\t * concurrent task adds LOAD_TOLERANCE_PER_PROCESS, capped at MAX_LOAD_MULTIPLIER.\n\t * Concurrency counts both monitored subagents and external load (background MCP\n\t * tools), since both contend for the same parent event loop.\n\t */\n\tprivate loadMultiplier(): number {\n\t\tconst concurrent = this.processes.size + this.externalLoad;\n\t\tconst mult = 1 + Math.max(0, concurrent - 1) * LOAD_TOLERANCE_PER_PROCESS;\n\t\treturn Math.min(mult, MAX_LOAD_MULTIPLIER);\n\t}\n\n\t/**\n\t * Begin monitoring a child process. The process must emit a\n\t * `{\"ping\":true}` JSON line on stdout every 30 seconds.\n\t */\n\tmonitor(task_id: string, agent_type: string, proc: ChildProcess): void {\n\t\tif (this.disposed) return;\n\n\t\tconst pid = proc.pid ?? 0;\n\t\tthis.processes.set(task_id, { pid, task_id, agent_type, process: proc });\n\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\n\t\tconst timeoutMs = TIMEOUTS_MS[agent_type] ?? TIMEOUTS_MS.explore;\n\t\tthis.startedAt.set(task_id, Date.now());\n\t\tthis.baseTimeoutMs.set(task_id, timeoutMs);\n\t\t// Arm the hard timeout scaled by current load. When it fires, handleTimeout\n\t\t// re-checks load and re-arms (up to MAX_LOAD_MULTIPLIER) if the pool is still\n\t\t// busy, so a slow-but-progressing subagent isn't killed for CPU contention.\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tthis.handleTimeout(task_id);\n\t\t\t},\n\t\t\tMath.round(timeoutMs * this.loadMultiplier()),\n\t\t);\n\t\tthis.timeouts.set(task_id, timeout);\n\n\t\tproc.once(\"exit\", () => {\n\t\t\tthis.untrack(task_id);\n\t\t});\n\t}\n\n\t/** Record a heartbeat for a monitored task. */\n\trecordHeartbeat(task_id: string): void {\n\t\tif (this.processes.has(task_id)) {\n\t\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\t\t}\n\t}\n\n\t/** Get the last recorded heartbeat timestamp, or null. */\n\tlastHeartbeatAt(task_id: string): number | null {\n\t\treturn this.lastHeartbeat.get(task_id) ?? null;\n\t}\n\n\t/** True if the task is currently being monitored. */\n\tisMonitoring(task_id: string): boolean {\n\t\treturn this.processes.has(task_id);\n\t}\n\n\t/** Kill all monitored processes and clean up. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tif (this.checkInterval) {\n\t\t\tclearInterval(this.checkInterval);\n\t\t\tthis.checkInterval = null;\n\t\t}\n\n\t\tfor (const timeout of this.timeouts.values()) {\n\t\t\tclearTimeout(timeout);\n\t\t}\n\t\tthis.timeouts.clear();\n\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\tthis.killTree(monitored);\n\t\t}\n\t\tthis.processes.clear();\n\t\tthis.lastHeartbeat.clear();\n\t\tthis.startedAt.clear();\n\t\tthis.baseTimeoutMs.clear();\n\t\tthis.reaping.clear();\n\t\tthis.removeAllListeners();\n\n\t\tif (this.parentShutdownHandler) {\n\t\t\tprocess.removeListener(\"SIGINT\", this.parentShutdownHandler);\n\t\t\tprocess.removeListener(\"SIGTERM\", this.parentShutdownHandler);\n\t\t}\n\t}\n\n\tprivate checkHeartbeats(): void {\n\t\tconst now = Date.now();\n\t\t// Event-loop lag: how much later than scheduled this check actually ran.\n\t\t// Heavy CPU load (many concurrent subagents + background MCP tools) starves\n\t\t// the loop, delaying both this check *and* our reading of child heartbeats.\n\t\t// Forgive that gap so the parent's own starvation isn't charged against the\n\t\t// children as a missed heartbeat.\n\t\tconst loopLag = Math.max(0, now - this.lastCheckAt - HEARTBEAT_CHECK_INTERVAL_MS);\n\t\tthis.lastCheckAt = now;\n\n\t\tconst threshold = HEARTBEAT_MISS_THRESHOLD_MS * this.loadMultiplier() + loopLag;\n\t\tfor (const [task_id] of this.processes) {\n\t\t\tif (this.reaping.has(task_id)) continue;\n\t\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\t\tif (last === undefined) continue;\n\t\t\tif (now - last > threshold) {\n\t\t\t\tthis.handleStalled(task_id);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate handleStalled(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\t\tthis.reaping.add(task_id);\n\n\t\t// Record why we reaped this child so a recurrence is diagnosable rather than\n\t\t// just \"stalled\": how long since the last heartbeat, and the load factors\n\t\t// (concurrent monitored subagents, threshold) that fed the decision.\n\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\tconst silentMs = last === undefined ? -1 : Date.now() - last;\n\t\tconsole.error(\n\t\t\t`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +\n\t\t\t\t`silent_ms=${silentMs} concurrent=${this.processes.size} ` +\n\t\t\t\t`load_mult=${this.loadMultiplier().toFixed(2)} base_threshold_ms=${HEARTBEAT_MISS_THRESHOLD_MS}`,\n\t\t);\n\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"stalled\", { task_id, pid: monitored.pid });\n\t\t// Process exit handler will call untrack()\n\t}\n\n\tprivate handleTimeout(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\n\t\t// Under load the wall-clock timer can fire while the subagent is still doing\n\t\t// real work — its turns are just slow because the CPU is shared. Re-arm rather\n\t\t// than kill, up to a hard ceiling (base * MAX_LOAD_MULTIPLIER) so a genuinely\n\t\t// stuck agent still terminates.\n\t\tconst started = this.startedAt.get(task_id) ?? Date.now();\n\t\tconst base = this.baseTimeoutMs.get(task_id) ?? TIMEOUTS_MS[monitored.agent_type] ?? TIMEOUTS_MS.explore;\n\t\tconst elapsed = Date.now() - started;\n\t\tconst ceiling = base * MAX_LOAD_MULTIPLIER;\n\t\tif (this.loadMultiplier() > 1 && elapsed < ceiling) {\n\t\t\tconst remaining = ceiling - elapsed;\n\t\t\tconst next = Math.min(\n\t\t\t\tMath.round(base * this.loadMultiplier()),\n\t\t\t\tMath.max(HEARTBEAT_CHECK_INTERVAL_MS, remaining),\n\t\t\t);\n\t\t\tthis.timeouts.set(\n\t\t\t\ttask_id,\n\t\t\t\tsetTimeout(() => this.handleTimeout(task_id), next),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tthis.reaping.add(task_id);\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"timeout\", { task_id, pid: monitored.pid });\n\t\tthis.timeouts.delete(task_id);\n\t\t// Process exit handler will call untrack()\n\t}\n\n\t/**\n\t * Kill a monitored subagent and everything it spawned. The pool spawns\n\t * children detached on POSIX (each leads its own process group), so the\n\t * group/tree kill reaches nested grandchildren (the subagent's bash commands,\n\t * its own subagents) that a single-PID SIGKILL would orphan. A pid of 0 means\n\t * the spawn never produced a process — nothing to kill (and `kill(-0)` would\n\t * signal our own process group).\n\t */\n\tprivate killTree(monitored: LifeguardProcess): void {\n\t\tif (monitored.pid > 0) {\n\t\t\tkillProcessTree(monitored.pid);\n\t\t} else if (!monitored.process.killed) {\n\t\t\tmonitored.process.kill(\"SIGKILL\");\n\t\t}\n\t}\n\n\tprivate untrack(task_id: string): void {\n\t\tconst timeout = this.timeouts.get(task_id);\n\t\tif (timeout) {\n\t\t\tclearTimeout(timeout);\n\t\t\tthis.timeouts.delete(task_id);\n\t\t}\n\t\tthis.processes.delete(task_id);\n\t\tthis.lastHeartbeat.delete(task_id);\n\t\tthis.startedAt.delete(task_id);\n\t\tthis.baseTimeoutMs.delete(task_id);\n\t\tthis.reaping.delete(task_id);\n\t}\n\n\tprivate setupParentExitHandlers(): void {\n\t\tconst shutdown = () => this.gracefulShutdown();\n\t\tthis.parentShutdownHandler = shutdown;\n\t\tprocess.setMaxListeners(Math.max(process.getMaxListeners(), 20));\n\t\tprocess.once(\"SIGINT\", shutdown);\n\t\tprocess.once(\"SIGTERM\", shutdown);\n\t}\n\n\tprivate gracefulShutdown(): void {\n\t\t// SIGTERM each child's process group (POSIX; the pool spawns them detached\n\t\t// as group leaders) so their own descendants get the graceful signal too.\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\ttry {\n\t\t\t\tif (process.platform !== \"win32\" && monitored.pid > 0) {\n\t\t\t\t\tprocess.kill(-monitored.pid, \"SIGTERM\");\n\t\t\t\t} else if (!monitored.process.killed) {\n\t\t\t\t\tmonitored.process.kill(\"SIGTERM\");\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Already gone.\n\t\t\t}\n\t\t}\n\n\t\t// Kill whole trees after the grace period\n\t\tsetTimeout(() => {\n\t\t\tfor (const monitored of this.processes.values()) {\n\t\t\t\tthis.killTree(monitored);\n\t\t\t}\n\t\t}, PARENT_SHUTDOWN_GRACE_MS).unref();\n\t}\n\n\tprivate sweepOldAgents(): void {\n\t\tconst dispatchDir = getDispatchRoot(this.cwd);\n\t\tif (!existsSync(dispatchDir)) return;\n\n\t\tconst now = Date.now();\n\t\tconst cutoff = 24 * 60 * 60 * 1000; // 24 hours\n\n\t\tfor (const entry of readdirSync(dispatchDir)) {\n\t\t\tconst entryPath = join(dispatchDir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (!stats.isDirectory()) continue;\n\n\t\t\t\tif (now - stats.mtimeMs > cutoff) {\n\t\t\t\t\tconst hasRunningPid = this.hasRunningPid(entryPath);\n\t\t\t\t\tif (!hasRunningPid) {\n\t\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore errors for individual entries\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate hasRunningPid(dir: string): boolean {\n\t\tconst pidFile = join(dir, \"pid\");\n\t\tif (!existsSync(pidFile)) return false;\n\n\t\ttry {\n\t\t\tconst pid = Number.parseInt(readFileSync(pidFile, \"utf-8\"), 10);\n\t\t\tif (Number.isNaN(pid)) return false;\n\t\t\tprocess.kill(pid, 0); // Check if process exists\n\t\t\treturn true;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate rmrf(dir: string): void {\n\t\tfor (const entry of readdirSync(dir)) {\n\t\t\tconst entryPath = join(dir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (stats.isDirectory()) {\n\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t} else {\n\t\t\t\t\tunlinkSync(entryPath);\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore\n\t\t\t}\n\t\t}\n\t\ttry {\n\t\t\trmdirSync(dir);\n\t\t} catch {\n\t\t\t// Ignore\n\t\t}\n\t}\n}\n"]}
+{"version":3,"file":"lifeguard.d.ts","sourceRoot":"","sources":["../../src/core/lifeguard.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAsC3C,MAAM,WAAW,gBAAgB;IAChC,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,YAAY,CAAC;CACtB;AAED;;;;GAIG;AACH,qBAAa,iBAAkB,SAAQ,YAAY;IAClD,OAAO,CAAC,SAAS,CAAuC;IACxD,OAAO,CAAC,aAAa,CAA6B;IAClD,OAAO,CAAC,QAAQ,CAAqC;IACrD,4EAA4E;IAC5E,OAAO,CAAC,SAAS,CAA6B;IAC9C,gFAAgF;IAChF,OAAO,CAAC,aAAa,CAA6B;IAClD,OAAO,CAAC,aAAa,CAA+B;IACpD,+EAA+E;IAC/E,OAAO,CAAC,WAAW,CAAc;IACjC;;;;;OAKG;IACH,OAAO,CAAC,YAAY,CAAK;IACzB;;;;OAIG;IACH,OAAO,CAAC,OAAO,CAAqB;IACpC,OAAO,CAAC,QAAQ,CAAS;IACzB,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAS;IAC7B,OAAO,CAAC,qBAAqB,CAAC,CAAa;IAE3C,YAAY,GAAG,EAAE,MAAM,EAOtB;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAEnC;IAED;;;;;OAKG;IACH,OAAO,CAAC,cAAc;IAMtB;;;OAGG;IACH,OAAO,CAAC,OAAO,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,IAAI,EAAE,YAAY,GAAG,IAAI,CAwBrE;IAED,+CAA+C;IAC/C,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,IAAI,CAIrC;IAED,0DAA0D;IAC1D,eAAe,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAE9C;IAED,qDAAqD;IACrD,YAAY,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAErC;IAED,iDAAiD;IACjD,OAAO,IAAI,IAAI,CA4Bd;IAED,OAAO,CAAC,eAAe;IAqBvB,OAAO,CAAC,aAAa;IAsBrB,OAAO,CAAC,aAAa;IAiCrB;;;;;;;OAOG;IACH,OAAO,CAAC,QAAQ;IAQhB,OAAO,CAAC,OAAO;IAaf,OAAO,CAAC,uBAAuB;IAQ/B,OAAO,CAAC,gBAAgB;IAuBxB,OAAO,CAAC,cAAc;IAyBtB,OAAO,CAAC,aAAa;IAcrB,OAAO,CAAC,IAAI;CAoBZ","sourcesContent":["import type { ChildProcess } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, readdirSync, readFileSync, rmdirSync, statSync, unlinkSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { getDispatchRoot } from \"../config.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\nimport { agentLog } from \"./agent-log.js\";\n\nconst TIMEOUTS_MS: Record<string, number> = {\n\texplore: 5 * 60 * 1000,\n\tedit: 10 * 60 * 1000,\n\ttest: 10 * 60 * 1000,\n\treview: 8 * 60 * 1000,\n\tdoc: 5 * 60 * 1000,\n};\n\nconst HEARTBEAT_MISS_THRESHOLD_MS = 60000;\nconst HEARTBEAT_CHECK_INTERVAL_MS = 5000;\nconst PARENT_SHUTDOWN_GRACE_MS = 5000;\n\n/**\n * Concurrency tolerance. Each *additional* concurrently-monitored subagent adds\n * this fraction to the heartbeat-miss and hard-timeout budgets.\n *\n * When several subagents run at once (plus background MCP tools), they saturate\n * the CPU and starve the parent's event loop: its `setInterval` heartbeat check\n * fires late, and it cannot read a child's `{\"ping\":true}` line in time even\n * though the child is healthy and still working. Scaling the budgets by load\n * stops that contention from false-positive SIGKILLing healthy subagents — the\n * failure the demo hit when running many agents + MCP tools in the background.\n */\nconst LOAD_TOLERANCE_PER_PROCESS = 0.5;\n\n/**\n * Hard ceiling on the load tolerance multiplier, so a genuinely stuck subagent\n * is still eventually reaped no matter how busy the pool is.\n */\nconst MAX_LOAD_MULTIPLIER = 4;\n\nexport interface LifeguardProcess {\n\tpid: number;\n\ttask_id: string;\n\tagent_type: string;\n\tprocess: ChildProcess;\n}\n\n/**\n * Monitors running subagent processes for heartbeats, hard timeouts,\n * and parent-exit cleanup. Emits \"stalled\" and \"timeout\" events when\n * processes are terminated.\n */\nexport class SubagentLifeguard extends EventEmitter {\n\tprivate processes = new Map<string, LifeguardProcess>();\n\tprivate lastHeartbeat = new Map<string, number>();\n\tprivate timeouts = new Map<string, NodeJS.Timeout>();\n\t/** When each task started, used to compute the load-scaled hard timeout. */\n\tprivate startedAt = new Map<string, number>();\n\t/** Per-agent base hard timeout (before load scaling), captured at monitor(). */\n\tprivate baseTimeoutMs = new Map<string, number>();\n\tprivate checkInterval: NodeJS.Timeout | null = null;\n\t/** Wall-clock time the heartbeat check last ran, to measure event-loop lag. */\n\tprivate lastCheckAt = Date.now();\n\t/**\n\t * Count of external in-process tasks (e.g. background MCP tools) running in the\n\t * parent alongside the monitored subagents. These don't show up in `processes`\n\t * but still saturate the parent's CPU/event loop, so they're folded into the\n\t * load multiplier. Updated by the pool via setExternalLoad().\n\t */\n\tprivate externalLoad = 0;\n\t/**\n\t * Tasks already reaped (stalled/timeout kill sent) but whose `exit` has not\n\t * fired yet. The heartbeat check runs every 5s, so without this a stalled task\n\t * would re-emit \"stalled\" (and re-kill) on every tick until the process exits.\n\t */\n\tprivate reaping = new Set<string>();\n\tprivate disposed = false;\n\tprivate readonly cwd: string;\n\tprivate parentShutdownHandler?: () => void;\n\n\tconstructor(cwd: string) {\n\t\tsuper();\n\t\tthis.cwd = cwd;\n\t\tthis.setupParentExitHandlers();\n\t\tthis.sweepOldAgents();\n\t\tthis.lastCheckAt = Date.now();\n\t\tthis.checkInterval = setInterval(() => this.checkHeartbeats(), HEARTBEAT_CHECK_INTERVAL_MS);\n\t}\n\n\t/**\n\t * Set the count of external in-process tasks (background MCP tools) sharing the\n\t * parent's CPU/event loop. Folded into loadMultiplier() so concurrent MCP work\n\t * widens the heartbeat/timeout budgets just like extra monitored subagents do.\n\t * Negative values are clamped to 0.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.externalLoad = Math.max(0, Math.floor(count));\n\t}\n\n\t/**\n\t * Tolerance multiplier for the current load. 1 concurrent task → 1x; each extra\n\t * concurrent task adds LOAD_TOLERANCE_PER_PROCESS, capped at MAX_LOAD_MULTIPLIER.\n\t * Concurrency counts both monitored subagents and external load (background MCP\n\t * tools), since both contend for the same parent event loop.\n\t */\n\tprivate loadMultiplier(): number {\n\t\tconst concurrent = this.processes.size + this.externalLoad;\n\t\tconst mult = 1 + Math.max(0, concurrent - 1) * LOAD_TOLERANCE_PER_PROCESS;\n\t\treturn Math.min(mult, MAX_LOAD_MULTIPLIER);\n\t}\n\n\t/**\n\t * Begin monitoring a child process. The process must emit a\n\t * `{\"ping\":true}` JSON line on stdout every 30 seconds.\n\t */\n\tmonitor(task_id: string, agent_type: string, proc: ChildProcess): void {\n\t\tif (this.disposed) return;\n\n\t\tconst pid = proc.pid ?? 0;\n\t\tthis.processes.set(task_id, { pid, task_id, agent_type, process: proc });\n\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\n\t\tconst timeoutMs = TIMEOUTS_MS[agent_type] ?? TIMEOUTS_MS.explore;\n\t\tthis.startedAt.set(task_id, Date.now());\n\t\tthis.baseTimeoutMs.set(task_id, timeoutMs);\n\t\t// Arm the hard timeout scaled by current load. When it fires, handleTimeout\n\t\t// re-checks load and re-arms (up to MAX_LOAD_MULTIPLIER) if the pool is still\n\t\t// busy, so a slow-but-progressing subagent isn't killed for CPU contention.\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tthis.handleTimeout(task_id);\n\t\t\t},\n\t\t\tMath.round(timeoutMs * this.loadMultiplier()),\n\t\t);\n\t\tthis.timeouts.set(task_id, timeout);\n\n\t\tproc.once(\"exit\", () => {\n\t\t\tthis.untrack(task_id);\n\t\t});\n\t}\n\n\t/** Record a heartbeat for a monitored task. */\n\trecordHeartbeat(task_id: string): void {\n\t\tif (this.processes.has(task_id)) {\n\t\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\t\t}\n\t}\n\n\t/** Get the last recorded heartbeat timestamp, or null. */\n\tlastHeartbeatAt(task_id: string): number | null {\n\t\treturn this.lastHeartbeat.get(task_id) ?? null;\n\t}\n\n\t/** True if the task is currently being monitored. */\n\tisMonitoring(task_id: string): boolean {\n\t\treturn this.processes.has(task_id);\n\t}\n\n\t/** Kill all monitored processes and clean up. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tif (this.checkInterval) {\n\t\t\tclearInterval(this.checkInterval);\n\t\t\tthis.checkInterval = null;\n\t\t}\n\n\t\tfor (const timeout of this.timeouts.values()) {\n\t\t\tclearTimeout(timeout);\n\t\t}\n\t\tthis.timeouts.clear();\n\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\tthis.killTree(monitored);\n\t\t}\n\t\tthis.processes.clear();\n\t\tthis.lastHeartbeat.clear();\n\t\tthis.startedAt.clear();\n\t\tthis.baseTimeoutMs.clear();\n\t\tthis.reaping.clear();\n\t\tthis.removeAllListeners();\n\n\t\tif (this.parentShutdownHandler) {\n\t\t\tprocess.removeListener(\"SIGINT\", this.parentShutdownHandler);\n\t\t\tprocess.removeListener(\"SIGTERM\", this.parentShutdownHandler);\n\t\t}\n\t}\n\n\tprivate checkHeartbeats(): void {\n\t\tconst now = Date.now();\n\t\t// Event-loop lag: how much later than scheduled this check actually ran.\n\t\t// Heavy CPU load (many concurrent subagents + background MCP tools) starves\n\t\t// the loop, delaying both this check *and* our reading of child heartbeats.\n\t\t// Forgive that gap so the parent's own starvation isn't charged against the\n\t\t// children as a missed heartbeat.\n\t\tconst loopLag = Math.max(0, now - this.lastCheckAt - HEARTBEAT_CHECK_INTERVAL_MS);\n\t\tthis.lastCheckAt = now;\n\n\t\tconst threshold = HEARTBEAT_MISS_THRESHOLD_MS * this.loadMultiplier() + loopLag;\n\t\tfor (const [task_id] of this.processes) {\n\t\t\tif (this.reaping.has(task_id)) continue;\n\t\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\t\tif (last === undefined) continue;\n\t\t\tif (now - last > threshold) {\n\t\t\t\tthis.handleStalled(task_id);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate handleStalled(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\t\tthis.reaping.add(task_id);\n\n\t\t// Record why we reaped this child so a recurrence is diagnosable rather than\n\t\t// just \"stalled\": how long since the last heartbeat, and the load factors\n\t\t// (concurrent monitored subagents, threshold) that fed the decision.\n\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\tconst silentMs = last === undefined ? -1 : Date.now() - last;\n\t\tagentLog(\n\t\t\t`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +\n\t\t\t\t`silent_ms=${silentMs} concurrent=${this.processes.size} ` +\n\t\t\t\t`load_mult=${this.loadMultiplier().toFixed(2)} base_threshold_ms=${HEARTBEAT_MISS_THRESHOLD_MS}`,\n\t\t);\n\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"stalled\", { task_id, pid: monitored.pid });\n\t\t// Process exit handler will call untrack()\n\t}\n\n\tprivate handleTimeout(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\n\t\t// Under load the wall-clock timer can fire while the subagent is still doing\n\t\t// real work — its turns are just slow because the CPU is shared. Re-arm rather\n\t\t// than kill, up to a hard ceiling (base * MAX_LOAD_MULTIPLIER) so a genuinely\n\t\t// stuck agent still terminates.\n\t\tconst started = this.startedAt.get(task_id) ?? Date.now();\n\t\tconst base = this.baseTimeoutMs.get(task_id) ?? TIMEOUTS_MS[monitored.agent_type] ?? TIMEOUTS_MS.explore;\n\t\tconst elapsed = Date.now() - started;\n\t\tconst ceiling = base * MAX_LOAD_MULTIPLIER;\n\t\tif (this.loadMultiplier() > 1 && elapsed < ceiling) {\n\t\t\tconst remaining = ceiling - elapsed;\n\t\t\tconst next = Math.min(\n\t\t\t\tMath.round(base * this.loadMultiplier()),\n\t\t\t\tMath.max(HEARTBEAT_CHECK_INTERVAL_MS, remaining),\n\t\t\t);\n\t\t\tthis.timeouts.set(\n\t\t\t\ttask_id,\n\t\t\t\tsetTimeout(() => this.handleTimeout(task_id), next),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tthis.reaping.add(task_id);\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"timeout\", { task_id, pid: monitored.pid });\n\t\tthis.timeouts.delete(task_id);\n\t\t// Process exit handler will call untrack()\n\t}\n\n\t/**\n\t * Kill a monitored subagent and everything it spawned. The pool spawns\n\t * children detached on POSIX (each leads its own process group), so the\n\t * group/tree kill reaches nested grandchildren (the subagent's bash commands,\n\t * its own subagents) that a single-PID SIGKILL would orphan. A pid of 0 means\n\t * the spawn never produced a process — nothing to kill (and `kill(-0)` would\n\t * signal our own process group).\n\t */\n\tprivate killTree(monitored: LifeguardProcess): void {\n\t\tif (monitored.pid > 0) {\n\t\t\tkillProcessTree(monitored.pid);\n\t\t} else if (!monitored.process.killed) {\n\t\t\tmonitored.process.kill(\"SIGKILL\");\n\t\t}\n\t}\n\n\tprivate untrack(task_id: string): void {\n\t\tconst timeout = this.timeouts.get(task_id);\n\t\tif (timeout) {\n\t\t\tclearTimeout(timeout);\n\t\t\tthis.timeouts.delete(task_id);\n\t\t}\n\t\tthis.processes.delete(task_id);\n\t\tthis.lastHeartbeat.delete(task_id);\n\t\tthis.startedAt.delete(task_id);\n\t\tthis.baseTimeoutMs.delete(task_id);\n\t\tthis.reaping.delete(task_id);\n\t}\n\n\tprivate setupParentExitHandlers(): void {\n\t\tconst shutdown = () => this.gracefulShutdown();\n\t\tthis.parentShutdownHandler = shutdown;\n\t\tprocess.setMaxListeners(Math.max(process.getMaxListeners(), 20));\n\t\tprocess.once(\"SIGINT\", shutdown);\n\t\tprocess.once(\"SIGTERM\", shutdown);\n\t}\n\n\tprivate gracefulShutdown(): void {\n\t\t// SIGTERM each child's process group (POSIX; the pool spawns them detached\n\t\t// as group leaders) so their own descendants get the graceful signal too.\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\ttry {\n\t\t\t\tif (process.platform !== \"win32\" && monitored.pid > 0) {\n\t\t\t\t\tprocess.kill(-monitored.pid, \"SIGTERM\");\n\t\t\t\t} else if (!monitored.process.killed) {\n\t\t\t\t\tmonitored.process.kill(\"SIGTERM\");\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Already gone.\n\t\t\t}\n\t\t}\n\n\t\t// Kill whole trees after the grace period\n\t\tsetTimeout(() => {\n\t\t\tfor (const monitored of this.processes.values()) {\n\t\t\t\tthis.killTree(monitored);\n\t\t\t}\n\t\t}, PARENT_SHUTDOWN_GRACE_MS).unref();\n\t}\n\n\tprivate sweepOldAgents(): void {\n\t\tconst dispatchDir = getDispatchRoot(this.cwd);\n\t\tif (!existsSync(dispatchDir)) return;\n\n\t\tconst now = Date.now();\n\t\tconst cutoff = 24 * 60 * 60 * 1000; // 24 hours\n\n\t\tfor (const entry of readdirSync(dispatchDir)) {\n\t\t\tconst entryPath = join(dispatchDir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (!stats.isDirectory()) continue;\n\n\t\t\t\tif (now - stats.mtimeMs > cutoff) {\n\t\t\t\t\tconst hasRunningPid = this.hasRunningPid(entryPath);\n\t\t\t\t\tif (!hasRunningPid) {\n\t\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore errors for individual entries\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate hasRunningPid(dir: string): boolean {\n\t\tconst pidFile = join(dir, \"pid\");\n\t\tif (!existsSync(pidFile)) return false;\n\n\t\ttry {\n\t\t\tconst pid = Number.parseInt(readFileSync(pidFile, \"utf-8\"), 10);\n\t\t\tif (Number.isNaN(pid)) return false;\n\t\t\tprocess.kill(pid, 0); // Check if process exists\n\t\t\treturn true;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate rmrf(dir: string): void {\n\t\tfor (const entry of readdirSync(dir)) {\n\t\t\tconst entryPath = join(dir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (stats.isDirectory()) {\n\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t} else {\n\t\t\t\t\tunlinkSync(entryPath);\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore\n\t\t\t}\n\t\t}\n\t\ttry {\n\t\t\trmdirSync(dir);\n\t\t} catch {\n\t\t\t// Ignore\n\t\t}\n\t}\n}\n"]}

package/dist/core/lifeguard.js

@@ -3,6 +3,7 @@ import { existsSync, readdirSync, readFileSync, rmdirSync, statSync, unlinkSync
 import { join } from "node:path";
 import { getDispatchRoot } from "../config.js";
 import { killProcessTree } from "../utils/shell.js";
+import { agentLog } from "./agent-log.js";
 const TIMEOUTS_MS = {
     explore: 5 * 60 * 1000,
     edit: 10 * 60 * 1000,
@@ -186,7 +187,7 @@ export class SubagentLifeguard extends EventEmitter {
         // (concurrent monitored subagents, threshold) that fed the decision.
         const last = this.lastHeartbeat.get(task_id);
         const silentMs = last === undefined ? -1 : Date.now() - last;
-        console.error(`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +
+        agentLog(`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +
             `silent_ms=${silentMs} concurrent=${this.processes.size} ` +
             `load_mult=${this.loadMultiplier().toFixed(2)} base_threshold_ms=${HEARTBEAT_MISS_THRESHOLD_MS}`);
         this.killTree(monitored);

package/dist/core/lifeguard.js.map

@@ -1 +1 @@
-{"version":3,"file":"lifeguard.js","sourceRoot":"","sources":["../../src/core/lifeguard.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC3C,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,YAAY,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACjG,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAC/C,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAEpD,MAAM,WAAW,GAA2B;IAC3C,OAAO,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;IACtB,IAAI,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;IACpB,IAAI,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;IACpB,MAAM,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;IACrB,GAAG,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;CAClB,CAAC;AAEF,MAAM,2BAA2B,GAAG,KAAK,CAAC;AAC1C,MAAM,2BAA2B,GAAG,IAAI,CAAC;AACzC,MAAM,wBAAwB,GAAG,IAAI,CAAC;AAEtC;;;;;;;;;;GAUG;AACH,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAEvC;;;GAGG;AACH,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAS9B;;;;GAIG;AACH,MAAM,OAAO,iBAAkB,SAAQ,YAAY;IAC1C,SAAS,GAAG,IAAI,GAAG,EAA4B,CAAC;IAChD,aAAa,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,QAAQ,GAAG,IAAI,GAAG,EAA0B,CAAC;IACrD,4EAA4E;IACpE,SAAS,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC9C,gFAAgF;IACxE,aAAa,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,aAAa,GAA0B,IAAI,CAAC;IACpD,+EAA+E;IACvE,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACjC;;;;;OAKG;IACK,YAAY,GAAG,CAAC,CAAC;IACzB;;;;OAIG;IACK,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAC5B,QAAQ,GAAG,KAAK,CAAC;IACR,GAAG,CAAS;IACrB,qBAAqB,CAAc;IAE3C,YAAY,GAAW,EAAE;QACxB,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,GAAG,GAAG,GAAG,CAAC;QACf,IAAI,CAAC,uBAAuB,EAAE,CAAC;QAC/B,IAAI,CAAC,cAAc,EAAE,CAAC;QACtB,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC9B,IAAI,CAAC,aAAa,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,eAAe,EAAE,EAAE,2BAA2B,CAAC,CAAC;IAAA,CAC5F;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAa,EAAQ;QACpC,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC;IAAA,CACnD;IAED;;;;;OAKG;IACK,cAAc,GAAW;QAChC,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,GAAG,IAAI,CAAC,YAAY,CAAC;QAC3D,MAAM,IAAI,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,CAAC,CAAC,GAAG,0BAA0B,CAAC;QAC1E,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,mBAAmB,CAAC,CAAC;IAAA,CAC3C;IAED;;;OAGG;IACH,OAAO,CAAC,OAAe,EAAE,UAAkB,EAAE,IAAkB,EAAQ;QACtE,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAE1B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;QAC1B,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzE,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAE5C,MAAM,SAAS,GAAG,WAAW,CAAC,UAAU,CAAC,IAAI,WAAW,CAAC,OAAO,CAAC;QACjE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QACxC,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;QAC3C,4EAA4E;QAC5E,8EAA8E;QAC9E,4EAA4E;QAC5E,MAAM,OAAO,GAAG,UAAU,CACzB,GAAG,EAAE,CAAC;YACL,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QAAA,CAC5B,EACD,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC,CAC7C,CAAC;QACF,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAEpC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE,CAAC;YACvB,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAAA,CACtB,CAAC,CAAC;IAAA,CACH;IAED,+CAA+C;IAC/C,eAAe,CAAC,OAAe,EAAQ;QACtC,IAAI,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YACjC,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAC7C,CAAC;IAAA,CACD;IAED,0DAA0D;IAC1D,eAAe,CAAC,OAAe,EAAiB;QAC/C,OAAO,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC;IAAA,CAC/C;IAED,qDAAqD;IACrD,YAAY,CAAC,OAAe,EAAW;QACtC,OAAO,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAAA,CACnC;IAED,iDAAiD;IACjD,OAAO,GAAS;QACf,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAC1B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QAErB,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACxB,aAAa,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;YAClC,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC;QAC3B,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,EAAE,CAAC;YAC9C,YAAY,CAAC,OAAO,CAAC,CAAC;QACvB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,CAAC;QAEtB,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;YACjD,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAC1B,CAAC;QACD,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;QACvB,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC;QAC3B,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;QACvB,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC;QAC3B,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QACrB,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAE1B,IAAI,IAAI,CAAC,qBAAqB,EAAE,CAAC;YAChC,OAAO,CAAC,cAAc,CAAC,QAAQ,EAAE,IAAI,CAAC,qBAAqB,CAAC,CAAC;YAC7D,OAAO,CAAC,cAAc,CAAC,SAAS,EAAE,IAAI,CAAC,qBAAqB,CAAC,CAAC;QAC/D,CAAC;IAAA,CACD;IAEO,eAAe,GAAS;QAC/B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,yEAAyE;QACzE,4EAA4E;QAC5E,4EAA4E;QAC5E,4EAA4E;QAC5E,kCAAkC;QAClC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,IAAI,CAAC,WAAW,GAAG,2BAA2B,CAAC,CAAC;QAClF,IAAI,CAAC,WAAW,GAAG,GAAG,CAAC;QAEvB,MAAM,SAAS,GAAG,2BAA2B,GAAG,IAAI,CAAC,cAAc,EAAE,GAAG,OAAO,CAAC;QAChF,KAAK,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACxC,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;gBAAE,SAAS;YACxC,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;YAC7C,IAAI,IAAI,KAAK,SAAS;gBAAE,SAAS;YACjC,IAAI,GAAG,GAAG,IAAI,GAAG,SAAS,EAAE,CAAC;gBAC5B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;YAC7B,CAAC;QACF,CAAC;IAAA,CACD;IAEO,aAAa,CAAC,OAAe,EAAQ;QAC5C,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO;QACpD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAE1B,6EAA6E;QAC7E,0EAA0E;QAC1E,qEAAqE;QACrE,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC7C,MAAM,QAAQ,GAAG,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;QAC7D,OAAO,CAAC,KAAK,CACZ,+BAA+B,OAAO,UAAU,SAAS,CAAC,UAAU,GAAG;YACtE,aAAa,QAAQ,eAAe,IAAI,CAAC,SAAS,CAAC,IAAI,GAAG;YAC1D,aAAa,IAAI,CAAC,cAAc,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,sBAAsB,2BAA2B,EAAE,CACjG,CAAC;QAEF,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAEzB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC;QACtD,2CAA2C;IADW,CAEtD;IAEO,aAAa,CAAC,OAAe,EAAQ;QAC5C,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO;QAEpD,6EAA6E;QAC7E,iFAA+E;QAC/E,8EAA8E;QAC9E,gCAAgC;QAChC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1D,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,WAAW,CAAC,SAAS,CAAC,UAAU,CAAC,IAAI,WAAW,CAAC,OAAO,CAAC;QACzG,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC;QACrC,MAAM,OAAO,GAAG,IAAI,GAAG,mBAAmB,CAAC;QAC3C,IAAI,IAAI,CAAC,cAAc,EAAE,GAAG,CAAC,IAAI,OAAO,GAAG,OAAO,EAAE,CAAC;YACpD,MAAM,SAAS,GAAG,OAAO,GAAG,OAAO,CAAC;YACpC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CACpB,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC,EACxC,IAAI,CAAC,GAAG,CAAC,2BAA2B,EAAE,SAAS,CAAC,CAChD,CAAC;YACF,IAAI,CAAC,QAAQ,CAAC,GAAG,CAChB,OAAO,EACP,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,EAAE,IAAI,CAAC,CACnD,CAAC;YACF,OAAO;QACR,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC1B,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAEzB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC;QACtD,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC9B,2CAA2C;IADb,CAE9B;IAED;;;;;;;OAOG;IACK,QAAQ,CAAC,SAA2B,EAAQ;QACnD,IAAI,SAAS,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC;YACvB,eAAe,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QAChC,CAAC;aAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;YACtC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACnC,CAAC;IAAA,CACD;IAEO,OAAO,CAAC,OAAe,EAAQ;QACtC,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC3C,IAAI,OAAO,EAAE,CAAC;YACb,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,CAAC;QACD,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAAA,CAC7B;IAEO,uBAAuB,GAAS;QACvC,MAAM,QAAQ,GAAG,GAAG,EAAE,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QAC/C,IAAI,CAAC,qBAAqB,GAAG,QAAQ,CAAC;QACtC,OAAO,CAAC,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,eAAe,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;QACjE,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;QACjC,OAAO,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAAA,CAClC;IAEO,gBAAgB,GAAS;QAChC,2EAA2E;QAC3E,0EAA0E;QAC1E,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;YACjD,IAAI,CAAC;gBACJ,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,IAAI,SAAS,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC;oBACvD,OAAO,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;gBACzC,CAAC;qBAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;oBACtC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBACnC,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,gBAAgB;YACjB,CAAC;QACF,CAAC;QAED,0CAA0C;QAC1C,UAAU,CAAC,GAAG,EAAE,CAAC;YAChB,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;gBACjD,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;YAC1B,CAAC;QAAA,CACD,EAAE,wBAAwB,CAAC,CAAC,KAAK,EAAE,CAAC;IAAA,CACrC;IAEO,cAAc,GAAS;QAC9B,MAAM,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC9C,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;YAAE,OAAO;QAErC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,MAAM,MAAM,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,WAAW;QAE/C,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,WAAW,CAAC,EAAE,CAAC;YAC9C,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,CAAC;YAC3C,IAAI,CAAC;gBACJ,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;gBAClC,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE;oBAAE,SAAS;gBAEnC,IAAI,GAAG,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,EAAE,CAAC;oBAClC,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC;oBACpD,IAAI,CAAC,aAAa,EAAE,CAAC;wBACpB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACtB,CAAC;gBACF,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,uCAAuC;YACxC,CAAC;QACF,CAAC;IAAA,CACD;IAEO,aAAa,CAAC,GAAW,EAAW;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACjC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;QAEvC,IAAI,CAAC;YACJ,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC;YAChE,IAAI,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC;gBAAE,OAAO,KAAK,CAAC;YACpC,OAAO,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,0BAA0B;YAChD,OAAO,IAAI,CAAC;QACb,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,KAAK,CAAC;QACd,CAAC;IAAA,CACD;IAEO,IAAI,CAAC,GAAW,EAAQ;QAC/B,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC;YACtC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YACnC,IAAI,CAAC;gBACJ,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;gBAClC,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;oBACzB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBACtB,CAAC;qBAAM,CAAC;oBACP,UAAU,CAAC,SAAS,CAAC,CAAC;gBACvB,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS;YACV,CAAC;QACF,CAAC;QACD,IAAI,CAAC;YACJ,SAAS,CAAC,GAAG,CAAC,CAAC;QAChB,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;IAAA,CACD;CACD","sourcesContent":["import type { ChildProcess } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, readdirSync, readFileSync, rmdirSync, statSync, unlinkSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { getDispatchRoot } from \"../config.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\n\nconst TIMEOUTS_MS: Record<string, number> = {\n\texplore: 5 * 60 * 1000,\n\tedit: 10 * 60 * 1000,\n\ttest: 10 * 60 * 1000,\n\treview: 8 * 60 * 1000,\n\tdoc: 5 * 60 * 1000,\n};\n\nconst HEARTBEAT_MISS_THRESHOLD_MS = 60000;\nconst HEARTBEAT_CHECK_INTERVAL_MS = 5000;\nconst PARENT_SHUTDOWN_GRACE_MS = 5000;\n\n/**\n * Concurrency tolerance. Each *additional* concurrently-monitored subagent adds\n * this fraction to the heartbeat-miss and hard-timeout budgets.\n *\n * When several subagents run at once (plus background MCP tools), they saturate\n * the CPU and starve the parent's event loop: its `setInterval` heartbeat check\n * fires late, and it cannot read a child's `{\"ping\":true}` line in time even\n * though the child is healthy and still working. Scaling the budgets by load\n * stops that contention from false-positive SIGKILLing healthy subagents — the\n * failure the demo hit when running many agents + MCP tools in the background.\n */\nconst LOAD_TOLERANCE_PER_PROCESS = 0.5;\n\n/**\n * Hard ceiling on the load tolerance multiplier, so a genuinely stuck subagent\n * is still eventually reaped no matter how busy the pool is.\n */\nconst MAX_LOAD_MULTIPLIER = 4;\n\nexport interface LifeguardProcess {\n\tpid: number;\n\ttask_id: string;\n\tagent_type: string;\n\tprocess: ChildProcess;\n}\n\n/**\n * Monitors running subagent processes for heartbeats, hard timeouts,\n * and parent-exit cleanup. Emits \"stalled\" and \"timeout\" events when\n * processes are terminated.\n */\nexport class SubagentLifeguard extends EventEmitter {\n\tprivate processes = new Map<string, LifeguardProcess>();\n\tprivate lastHeartbeat = new Map<string, number>();\n\tprivate timeouts = new Map<string, NodeJS.Timeout>();\n\t/** When each task started, used to compute the load-scaled hard timeout. */\n\tprivate startedAt = new Map<string, number>();\n\t/** Per-agent base hard timeout (before load scaling), captured at monitor(). */\n\tprivate baseTimeoutMs = new Map<string, number>();\n\tprivate checkInterval: NodeJS.Timeout | null = null;\n\t/** Wall-clock time the heartbeat check last ran, to measure event-loop lag. */\n\tprivate lastCheckAt = Date.now();\n\t/**\n\t * Count of external in-process tasks (e.g. background MCP tools) running in the\n\t * parent alongside the monitored subagents. These don't show up in `processes`\n\t * but still saturate the parent's CPU/event loop, so they're folded into the\n\t * load multiplier. Updated by the pool via setExternalLoad().\n\t */\n\tprivate externalLoad = 0;\n\t/**\n\t * Tasks already reaped (stalled/timeout kill sent) but whose `exit` has not\n\t * fired yet. The heartbeat check runs every 5s, so without this a stalled task\n\t * would re-emit \"stalled\" (and re-kill) on every tick until the process exits.\n\t */\n\tprivate reaping = new Set<string>();\n\tprivate disposed = false;\n\tprivate readonly cwd: string;\n\tprivate parentShutdownHandler?: () => void;\n\n\tconstructor(cwd: string) {\n\t\tsuper();\n\t\tthis.cwd = cwd;\n\t\tthis.setupParentExitHandlers();\n\t\tthis.sweepOldAgents();\n\t\tthis.lastCheckAt = Date.now();\n\t\tthis.checkInterval = setInterval(() => this.checkHeartbeats(), HEARTBEAT_CHECK_INTERVAL_MS);\n\t}\n\n\t/**\n\t * Set the count of external in-process tasks (background MCP tools) sharing the\n\t * parent's CPU/event loop. Folded into loadMultiplier() so concurrent MCP work\n\t * widens the heartbeat/timeout budgets just like extra monitored subagents do.\n\t * Negative values are clamped to 0.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.externalLoad = Math.max(0, Math.floor(count));\n\t}\n\n\t/**\n\t * Tolerance multiplier for the current load. 1 concurrent task → 1x; each extra\n\t * concurrent task adds LOAD_TOLERANCE_PER_PROCESS, capped at MAX_LOAD_MULTIPLIER.\n\t * Concurrency counts both monitored subagents and external load (background MCP\n\t * tools), since both contend for the same parent event loop.\n\t */\n\tprivate loadMultiplier(): number {\n\t\tconst concurrent = this.processes.size + this.externalLoad;\n\t\tconst mult = 1 + Math.max(0, concurrent - 1) * LOAD_TOLERANCE_PER_PROCESS;\n\t\treturn Math.min(mult, MAX_LOAD_MULTIPLIER);\n\t}\n\n\t/**\n\t * Begin monitoring a child process. The process must emit a\n\t * `{\"ping\":true}` JSON line on stdout every 30 seconds.\n\t */\n\tmonitor(task_id: string, agent_type: string, proc: ChildProcess): void {\n\t\tif (this.disposed) return;\n\n\t\tconst pid = proc.pid ?? 0;\n\t\tthis.processes.set(task_id, { pid, task_id, agent_type, process: proc });\n\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\n\t\tconst timeoutMs = TIMEOUTS_MS[agent_type] ?? TIMEOUTS_MS.explore;\n\t\tthis.startedAt.set(task_id, Date.now());\n\t\tthis.baseTimeoutMs.set(task_id, timeoutMs);\n\t\t// Arm the hard timeout scaled by current load. When it fires, handleTimeout\n\t\t// re-checks load and re-arms (up to MAX_LOAD_MULTIPLIER) if the pool is still\n\t\t// busy, so a slow-but-progressing subagent isn't killed for CPU contention.\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tthis.handleTimeout(task_id);\n\t\t\t},\n\t\t\tMath.round(timeoutMs * this.loadMultiplier()),\n\t\t);\n\t\tthis.timeouts.set(task_id, timeout);\n\n\t\tproc.once(\"exit\", () => {\n\t\t\tthis.untrack(task_id);\n\t\t});\n\t}\n\n\t/** Record a heartbeat for a monitored task. */\n\trecordHeartbeat(task_id: string): void {\n\t\tif (this.processes.has(task_id)) {\n\t\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\t\t}\n\t}\n\n\t/** Get the last recorded heartbeat timestamp, or null. */\n\tlastHeartbeatAt(task_id: string): number | null {\n\t\treturn this.lastHeartbeat.get(task_id) ?? null;\n\t}\n\n\t/** True if the task is currently being monitored. */\n\tisMonitoring(task_id: string): boolean {\n\t\treturn this.processes.has(task_id);\n\t}\n\n\t/** Kill all monitored processes and clean up. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tif (this.checkInterval) {\n\t\t\tclearInterval(this.checkInterval);\n\t\t\tthis.checkInterval = null;\n\t\t}\n\n\t\tfor (const timeout of this.timeouts.values()) {\n\t\t\tclearTimeout(timeout);\n\t\t}\n\t\tthis.timeouts.clear();\n\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\tthis.killTree(monitored);\n\t\t}\n\t\tthis.processes.clear();\n\t\tthis.lastHeartbeat.clear();\n\t\tthis.startedAt.clear();\n\t\tthis.baseTimeoutMs.clear();\n\t\tthis.reaping.clear();\n\t\tthis.removeAllListeners();\n\n\t\tif (this.parentShutdownHandler) {\n\t\t\tprocess.removeListener(\"SIGINT\", this.parentShutdownHandler);\n\t\t\tprocess.removeListener(\"SIGTERM\", this.parentShutdownHandler);\n\t\t}\n\t}\n\n\tprivate checkHeartbeats(): void {\n\t\tconst now = Date.now();\n\t\t// Event-loop lag: how much later than scheduled this check actually ran.\n\t\t// Heavy CPU load (many concurrent subagents + background MCP tools) starves\n\t\t// the loop, delaying both this check *and* our reading of child heartbeats.\n\t\t// Forgive that gap so the parent's own starvation isn't charged against the\n\t\t// children as a missed heartbeat.\n\t\tconst loopLag = Math.max(0, now - this.lastCheckAt - HEARTBEAT_CHECK_INTERVAL_MS);\n\t\tthis.lastCheckAt = now;\n\n\t\tconst threshold = HEARTBEAT_MISS_THRESHOLD_MS * this.loadMultiplier() + loopLag;\n\t\tfor (const [task_id] of this.processes) {\n\t\t\tif (this.reaping.has(task_id)) continue;\n\t\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\t\tif (last === undefined) continue;\n\t\t\tif (now - last > threshold) {\n\t\t\t\tthis.handleStalled(task_id);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate handleStalled(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\t\tthis.reaping.add(task_id);\n\n\t\t// Record why we reaped this child so a recurrence is diagnosable rather than\n\t\t// just \"stalled\": how long since the last heartbeat, and the load factors\n\t\t// (concurrent monitored subagents, threshold) that fed the decision.\n\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\tconst silentMs = last === undefined ? -1 : Date.now() - last;\n\t\tconsole.error(\n\t\t\t`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +\n\t\t\t\t`silent_ms=${silentMs} concurrent=${this.processes.size} ` +\n\t\t\t\t`load_mult=${this.loadMultiplier().toFixed(2)} base_threshold_ms=${HEARTBEAT_MISS_THRESHOLD_MS}`,\n\t\t);\n\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"stalled\", { task_id, pid: monitored.pid });\n\t\t// Process exit handler will call untrack()\n\t}\n\n\tprivate handleTimeout(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\n\t\t// Under load the wall-clock timer can fire while the subagent is still doing\n\t\t// real work — its turns are just slow because the CPU is shared. Re-arm rather\n\t\t// than kill, up to a hard ceiling (base * MAX_LOAD_MULTIPLIER) so a genuinely\n\t\t// stuck agent still terminates.\n\t\tconst started = this.startedAt.get(task_id) ?? Date.now();\n\t\tconst base = this.baseTimeoutMs.get(task_id) ?? TIMEOUTS_MS[monitored.agent_type] ?? TIMEOUTS_MS.explore;\n\t\tconst elapsed = Date.now() - started;\n\t\tconst ceiling = base * MAX_LOAD_MULTIPLIER;\n\t\tif (this.loadMultiplier() > 1 && elapsed < ceiling) {\n\t\t\tconst remaining = ceiling - elapsed;\n\t\t\tconst next = Math.min(\n\t\t\t\tMath.round(base * this.loadMultiplier()),\n\t\t\t\tMath.max(HEARTBEAT_CHECK_INTERVAL_MS, remaining),\n\t\t\t);\n\t\t\tthis.timeouts.set(\n\t\t\t\ttask_id,\n\t\t\t\tsetTimeout(() => this.handleTimeout(task_id), next),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tthis.reaping.add(task_id);\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"timeout\", { task_id, pid: monitored.pid });\n\t\tthis.timeouts.delete(task_id);\n\t\t// Process exit handler will call untrack()\n\t}\n\n\t/**\n\t * Kill a monitored subagent and everything it spawned. The pool spawns\n\t * children detached on POSIX (each leads its own process group), so the\n\t * group/tree kill reaches nested grandchildren (the subagent's bash commands,\n\t * its own subagents) that a single-PID SIGKILL would orphan. A pid of 0 means\n\t * the spawn never produced a process — nothing to kill (and `kill(-0)` would\n\t * signal our own process group).\n\t */\n\tprivate killTree(monitored: LifeguardProcess): void {\n\t\tif (monitored.pid > 0) {\n\t\t\tkillProcessTree(monitored.pid);\n\t\t} else if (!monitored.process.killed) {\n\t\t\tmonitored.process.kill(\"SIGKILL\");\n\t\t}\n\t}\n\n\tprivate untrack(task_id: string): void {\n\t\tconst timeout = this.timeouts.get(task_id);\n\t\tif (timeout) {\n\t\t\tclearTimeout(timeout);\n\t\t\tthis.timeouts.delete(task_id);\n\t\t}\n\t\tthis.processes.delete(task_id);\n\t\tthis.lastHeartbeat.delete(task_id);\n\t\tthis.startedAt.delete(task_id);\n\t\tthis.baseTimeoutMs.delete(task_id);\n\t\tthis.reaping.delete(task_id);\n\t}\n\n\tprivate setupParentExitHandlers(): void {\n\t\tconst shutdown = () => this.gracefulShutdown();\n\t\tthis.parentShutdownHandler = shutdown;\n\t\tprocess.setMaxListeners(Math.max(process.getMaxListeners(), 20));\n\t\tprocess.once(\"SIGINT\", shutdown);\n\t\tprocess.once(\"SIGTERM\", shutdown);\n\t}\n\n\tprivate gracefulShutdown(): void {\n\t\t// SIGTERM each child's process group (POSIX; the pool spawns them detached\n\t\t// as group leaders) so their own descendants get the graceful signal too.\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\ttry {\n\t\t\t\tif (process.platform !== \"win32\" && monitored.pid > 0) {\n\t\t\t\t\tprocess.kill(-monitored.pid, \"SIGTERM\");\n\t\t\t\t} else if (!monitored.process.killed) {\n\t\t\t\t\tmonitored.process.kill(\"SIGTERM\");\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Already gone.\n\t\t\t}\n\t\t}\n\n\t\t// Kill whole trees after the grace period\n\t\tsetTimeout(() => {\n\t\t\tfor (const monitored of this.processes.values()) {\n\t\t\t\tthis.killTree(monitored);\n\t\t\t}\n\t\t}, PARENT_SHUTDOWN_GRACE_MS).unref();\n\t}\n\n\tprivate sweepOldAgents(): void {\n\t\tconst dispatchDir = getDispatchRoot(this.cwd);\n\t\tif (!existsSync(dispatchDir)) return;\n\n\t\tconst now = Date.now();\n\t\tconst cutoff = 24 * 60 * 60 * 1000; // 24 hours\n\n\t\tfor (const entry of readdirSync(dispatchDir)) {\n\t\t\tconst entryPath = join(dispatchDir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (!stats.isDirectory()) continue;\n\n\t\t\t\tif (now - stats.mtimeMs > cutoff) {\n\t\t\t\t\tconst hasRunningPid = this.hasRunningPid(entryPath);\n\t\t\t\t\tif (!hasRunningPid) {\n\t\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore errors for individual entries\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate hasRunningPid(dir: string): boolean {\n\t\tconst pidFile = join(dir, \"pid\");\n\t\tif (!existsSync(pidFile)) return false;\n\n\t\ttry {\n\t\t\tconst pid = Number.parseInt(readFileSync(pidFile, \"utf-8\"), 10);\n\t\t\tif (Number.isNaN(pid)) return false;\n\t\t\tprocess.kill(pid, 0); // Check if process exists\n\t\t\treturn true;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate rmrf(dir: string): void {\n\t\tfor (const entry of readdirSync(dir)) {\n\t\t\tconst entryPath = join(dir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (stats.isDirectory()) {\n\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t} else {\n\t\t\t\t\tunlinkSync(entryPath);\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore\n\t\t\t}\n\t\t}\n\t\ttry {\n\t\t\trmdirSync(dir);\n\t\t} catch {\n\t\t\t// Ignore\n\t\t}\n\t}\n}\n"]}
+{"version":3,"file":"lifeguard.js","sourceRoot":"","sources":["../../src/core/lifeguard.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC3C,OAAO,EAAE,UAAU,EAAE,WAAW,EAAE,YAAY,EAAE,SAAS,EAAE,QAAQ,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AACjG,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,eAAe,EAAE,MAAM,cAAc,CAAC;AAC/C,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AACpD,OAAO,EAAE,QAAQ,EAAE,MAAM,gBAAgB,CAAC;AAE1C,MAAM,WAAW,GAA2B;IAC3C,OAAO,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;IACtB,IAAI,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;IACpB,IAAI,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;IACpB,MAAM,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;IACrB,GAAG,EAAE,CAAC,GAAG,EAAE,GAAG,IAAI;CAClB,CAAC;AAEF,MAAM,2BAA2B,GAAG,KAAK,CAAC;AAC1C,MAAM,2BAA2B,GAAG,IAAI,CAAC;AACzC,MAAM,wBAAwB,GAAG,IAAI,CAAC;AAEtC;;;;;;;;;;GAUG;AACH,MAAM,0BAA0B,GAAG,GAAG,CAAC;AAEvC;;;GAGG;AACH,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAS9B;;;;GAIG;AACH,MAAM,OAAO,iBAAkB,SAAQ,YAAY;IAC1C,SAAS,GAAG,IAAI,GAAG,EAA4B,CAAC;IAChD,aAAa,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,QAAQ,GAAG,IAAI,GAAG,EAA0B,CAAC;IACrD,4EAA4E;IACpE,SAAS,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC9C,gFAAgF;IACxE,aAAa,GAAG,IAAI,GAAG,EAAkB,CAAC;IAC1C,aAAa,GAA0B,IAAI,CAAC;IACpD,+EAA+E;IACvE,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IACjC;;;;;OAKG;IACK,YAAY,GAAG,CAAC,CAAC;IACzB;;;;OAIG;IACK,OAAO,GAAG,IAAI,GAAG,EAAU,CAAC;IAC5B,QAAQ,GAAG,KAAK,CAAC;IACR,GAAG,CAAS;IACrB,qBAAqB,CAAc;IAE3C,YAAY,GAAW,EAAE;QACxB,KAAK,EAAE,CAAC;QACR,IAAI,CAAC,GAAG,GAAG,GAAG,CAAC;QACf,IAAI,CAAC,uBAAuB,EAAE,CAAC;QAC/B,IAAI,CAAC,cAAc,EAAE,CAAC;QACtB,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC9B,IAAI,CAAC,aAAa,GAAG,WAAW,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,eAAe,EAAE,EAAE,2BAA2B,CAAC,CAAC;IAAA,CAC5F;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAa,EAAQ;QACpC,IAAI,CAAC,YAAY,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC,CAAC;IAAA,CACnD;IAED;;;;;OAKG;IACK,cAAc,GAAW;QAChC,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,GAAG,IAAI,CAAC,YAAY,CAAC;QAC3D,MAAM,IAAI,GAAG,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,UAAU,GAAG,CAAC,CAAC,GAAG,0BAA0B,CAAC;QAC1E,OAAO,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,mBAAmB,CAAC,CAAC;IAAA,CAC3C;IAED;;;OAGG;IACH,OAAO,CAAC,OAAe,EAAE,UAAkB,EAAE,IAAkB,EAAQ;QACtE,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAE1B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,CAAC;QAC1B,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,EAAE,GAAG,EAAE,OAAO,EAAE,UAAU,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;QACzE,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAE5C,MAAM,SAAS,GAAG,WAAW,CAAC,UAAU,CAAC,IAAI,WAAW,CAAC,OAAO,CAAC;QACjE,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QACxC,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;QAC3C,4EAA4E;QAC5E,8EAA8E;QAC9E,4EAA4E;QAC5E,MAAM,OAAO,GAAG,UAAU,CACzB,GAAG,EAAE,CAAC;YACL,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;QAAA,CAC5B,EACD,IAAI,CAAC,KAAK,CAAC,SAAS,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC,CAC7C,CAAC;QACF,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAEpC,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE,CAAC;YACvB,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAAA,CACtB,CAAC,CAAC;IAAA,CACH;IAED,+CAA+C;IAC/C,eAAe,CAAC,OAAe,EAAQ;QACtC,IAAI,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;YACjC,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC;QAC7C,CAAC;IAAA,CACD;IAED,0DAA0D;IAC1D,eAAe,CAAC,OAAe,EAAiB;QAC/C,OAAO,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC;IAAA,CAC/C;IAED,qDAAqD;IACrD,YAAY,CAAC,OAAe,EAAW;QACtC,OAAO,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IAAA,CACnC;IAED,iDAAiD;IACjD,OAAO,GAAS;QACf,IAAI,IAAI,CAAC,QAAQ;YAAE,OAAO;QAC1B,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC;QAErB,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACxB,aAAa,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;YAClC,IAAI,CAAC,aAAa,GAAG,IAAI,CAAC;QAC3B,CAAC;QAED,KAAK,MAAM,OAAO,IAAI,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,EAAE,CAAC;YAC9C,YAAY,CAAC,OAAO,CAAC,CAAC;QACvB,CAAC;QACD,IAAI,CAAC,QAAQ,CAAC,KAAK,EAAE,CAAC;QAEtB,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;YACjD,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAC1B,CAAC;QACD,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;QACvB,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC;QAC3B,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;QACvB,IAAI,CAAC,aAAa,CAAC,KAAK,EAAE,CAAC;QAC3B,IAAI,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QACrB,IAAI,CAAC,kBAAkB,EAAE,CAAC;QAE1B,IAAI,IAAI,CAAC,qBAAqB,EAAE,CAAC;YAChC,OAAO,CAAC,cAAc,CAAC,QAAQ,EAAE,IAAI,CAAC,qBAAqB,CAAC,CAAC;YAC7D,OAAO,CAAC,cAAc,CAAC,SAAS,EAAE,IAAI,CAAC,qBAAqB,CAAC,CAAC;QAC/D,CAAC;IAAA,CACD;IAEO,eAAe,GAAS;QAC/B,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,yEAAyE;QACzE,4EAA4E;QAC5E,4EAA4E;QAC5E,4EAA4E;QAC5E,kCAAkC;QAClC,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,GAAG,IAAI,CAAC,WAAW,GAAG,2BAA2B,CAAC,CAAC;QAClF,IAAI,CAAC,WAAW,GAAG,GAAG,CAAC;QAEvB,MAAM,SAAS,GAAG,2BAA2B,GAAG,IAAI,CAAC,cAAc,EAAE,GAAG,OAAO,CAAC;QAChF,KAAK,MAAM,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YACxC,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;gBAAE,SAAS;YACxC,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;YAC7C,IAAI,IAAI,KAAK,SAAS;gBAAE,SAAS;YACjC,IAAI,GAAG,GAAG,IAAI,GAAG,SAAS,EAAE,CAAC;gBAC5B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC;YAC7B,CAAC;QACF,CAAC;IAAA,CACD;IAEO,aAAa,CAAC,OAAe,EAAQ;QAC5C,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO;QACpD,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAE1B,6EAA6E;QAC7E,0EAA0E;QAC1E,qEAAqE;QACrE,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC7C,MAAM,QAAQ,GAAG,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;QAC7D,QAAQ,CACP,+BAA+B,OAAO,UAAU,SAAS,CAAC,UAAU,GAAG;YACtE,aAAa,QAAQ,eAAe,IAAI,CAAC,SAAS,CAAC,IAAI,GAAG;YAC1D,aAAa,IAAI,CAAC,cAAc,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,sBAAsB,2BAA2B,EAAE,CACjG,CAAC;QAEF,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAEzB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC;QACtD,2CAA2C;IADW,CAEtD;IAEO,aAAa,CAAC,OAAe,EAAQ;QAC5C,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC9C,IAAI,CAAC,SAAS,IAAI,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;YAAE,OAAO;QAEpD,6EAA6E;QAC7E,iFAA+E;QAC/E,8EAA8E;QAC9E,gCAAgC;QAChC,MAAM,OAAO,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC;QAC1D,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,WAAW,CAAC,SAAS,CAAC,UAAU,CAAC,IAAI,WAAW,CAAC,OAAO,CAAC;QACzG,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,GAAG,OAAO,CAAC;QACrC,MAAM,OAAO,GAAG,IAAI,GAAG,mBAAmB,CAAC;QAC3C,IAAI,IAAI,CAAC,cAAc,EAAE,GAAG,CAAC,IAAI,OAAO,GAAG,OAAO,EAAE,CAAC;YACpD,MAAM,SAAS,GAAG,OAAO,GAAG,OAAO,CAAC;YACpC,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CACpB,IAAI,CAAC,KAAK,CAAC,IAAI,GAAG,IAAI,CAAC,cAAc,EAAE,CAAC,EACxC,IAAI,CAAC,GAAG,CAAC,2BAA2B,EAAE,SAAS,CAAC,CAChD,CAAC;YACF,IAAI,CAAC,QAAQ,CAAC,GAAG,CAChB,OAAO,EACP,UAAU,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC,EAAE,IAAI,CAAC,CACnD,CAAC;YACF,OAAO;QACR,CAAC;QAED,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC1B,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;QAEzB,IAAI,CAAC,IAAI,CAAC,SAAS,EAAE,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC;QACtD,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC9B,2CAA2C;IADb,CAE9B;IAED;;;;;;;OAOG;IACK,QAAQ,CAAC,SAA2B,EAAQ;QACnD,IAAI,SAAS,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC;YACvB,eAAe,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;QAChC,CAAC;aAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;YACtC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACnC,CAAC;IAAA,CACD;IAEO,OAAO,CAAC,OAAe,EAAQ;QACtC,MAAM,OAAO,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;QAC3C,IAAI,OAAO,EAAE,CAAC;YACb,YAAY,CAAC,OAAO,CAAC,CAAC;YACtB,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,CAAC;QACD,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QAC/B,IAAI,CAAC,aAAa,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACnC,IAAI,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAAA,CAC7B;IAEO,uBAAuB,GAAS;QACvC,MAAM,QAAQ,GAAG,GAAG,EAAE,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QAC/C,IAAI,CAAC,qBAAqB,GAAG,QAAQ,CAAC;QACtC,OAAO,CAAC,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,eAAe,EAAE,EAAE,EAAE,CAAC,CAAC,CAAC;QACjE,OAAO,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;QACjC,OAAO,CAAC,IAAI,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAAA,CAClC;IAEO,gBAAgB,GAAS;QAChC,2EAA2E;QAC3E,0EAA0E;QAC1E,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;YACjD,IAAI,CAAC;gBACJ,IAAI,OAAO,CAAC,QAAQ,KAAK,OAAO,IAAI,SAAS,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC;oBACvD,OAAO,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,GAAG,EAAE,SAAS,CAAC,CAAC;gBACzC,CAAC;qBAAM,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,MAAM,EAAE,CAAC;oBACtC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBACnC,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,gBAAgB;YACjB,CAAC;QACF,CAAC;QAED,0CAA0C;QAC1C,UAAU,CAAC,GAAG,EAAE,CAAC;YAChB,KAAK,MAAM,SAAS,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC;gBACjD,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;YAC1B,CAAC;QAAA,CACD,EAAE,wBAAwB,CAAC,CAAC,KAAK,EAAE,CAAC;IAAA,CACrC;IAEO,cAAc,GAAS;QAC9B,MAAM,WAAW,GAAG,eAAe,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QAC9C,IAAI,CAAC,UAAU,CAAC,WAAW,CAAC;YAAE,OAAO;QAErC,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,MAAM,MAAM,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,CAAC,CAAC,WAAW;QAE/C,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,WAAW,CAAC,EAAE,CAAC;YAC9C,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,EAAE,KAAK,CAAC,CAAC;YAC3C,IAAI,CAAC;gBACJ,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;gBAClC,IAAI,CAAC,KAAK,CAAC,WAAW,EAAE;oBAAE,SAAS;gBAEnC,IAAI,GAAG,GAAG,KAAK,CAAC,OAAO,GAAG,MAAM,EAAE,CAAC;oBAClC,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,CAAC,SAAS,CAAC,CAAC;oBACpD,IAAI,CAAC,aAAa,EAAE,CAAC;wBACpB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;oBACtB,CAAC;gBACF,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,uCAAuC;YACxC,CAAC;QACF,CAAC;IAAA,CACD;IAEO,aAAa,CAAC,GAAW,EAAW;QAC3C,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;QACjC,IAAI,CAAC,UAAU,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;QAEvC,IAAI,CAAC;YACJ,MAAM,GAAG,GAAG,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC;YAChE,IAAI,MAAM,CAAC,KAAK,CAAC,GAAG,CAAC;gBAAE,OAAO,KAAK,CAAC;YACpC,OAAO,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,0BAA0B;YAChD,OAAO,IAAI,CAAC;QACb,CAAC;QAAC,MAAM,CAAC;YACR,OAAO,KAAK,CAAC;QACd,CAAC;IAAA,CACD;IAEO,IAAI,CAAC,GAAW,EAAQ;QAC/B,KAAK,MAAM,KAAK,IAAI,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC;YACtC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC;YACnC,IAAI,CAAC;gBACJ,MAAM,KAAK,GAAG,QAAQ,CAAC,SAAS,CAAC,CAAC;gBAClC,IAAI,KAAK,CAAC,WAAW,EAAE,EAAE,CAAC;oBACzB,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;gBACtB,CAAC;qBAAM,CAAC;oBACP,UAAU,CAAC,SAAS,CAAC,CAAC;gBACvB,CAAC;YACF,CAAC;YAAC,MAAM,CAAC;gBACR,SAAS;YACV,CAAC;QACF,CAAC;QACD,IAAI,CAAC;YACJ,SAAS,CAAC,GAAG,CAAC,CAAC;QAChB,CAAC;QAAC,MAAM,CAAC;YACR,SAAS;QACV,CAAC;IAAA,CACD;CACD","sourcesContent":["import type { ChildProcess } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, readdirSync, readFileSync, rmdirSync, statSync, unlinkSync } from \"node:fs\";\nimport { join } from \"node:path\";\nimport { getDispatchRoot } from \"../config.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\nimport { agentLog } from \"./agent-log.js\";\n\nconst TIMEOUTS_MS: Record<string, number> = {\n\texplore: 5 * 60 * 1000,\n\tedit: 10 * 60 * 1000,\n\ttest: 10 * 60 * 1000,\n\treview: 8 * 60 * 1000,\n\tdoc: 5 * 60 * 1000,\n};\n\nconst HEARTBEAT_MISS_THRESHOLD_MS = 60000;\nconst HEARTBEAT_CHECK_INTERVAL_MS = 5000;\nconst PARENT_SHUTDOWN_GRACE_MS = 5000;\n\n/**\n * Concurrency tolerance. Each *additional* concurrently-monitored subagent adds\n * this fraction to the heartbeat-miss and hard-timeout budgets.\n *\n * When several subagents run at once (plus background MCP tools), they saturate\n * the CPU and starve the parent's event loop: its `setInterval` heartbeat check\n * fires late, and it cannot read a child's `{\"ping\":true}` line in time even\n * though the child is healthy and still working. Scaling the budgets by load\n * stops that contention from false-positive SIGKILLing healthy subagents — the\n * failure the demo hit when running many agents + MCP tools in the background.\n */\nconst LOAD_TOLERANCE_PER_PROCESS = 0.5;\n\n/**\n * Hard ceiling on the load tolerance multiplier, so a genuinely stuck subagent\n * is still eventually reaped no matter how busy the pool is.\n */\nconst MAX_LOAD_MULTIPLIER = 4;\n\nexport interface LifeguardProcess {\n\tpid: number;\n\ttask_id: string;\n\tagent_type: string;\n\tprocess: ChildProcess;\n}\n\n/**\n * Monitors running subagent processes for heartbeats, hard timeouts,\n * and parent-exit cleanup. Emits \"stalled\" and \"timeout\" events when\n * processes are terminated.\n */\nexport class SubagentLifeguard extends EventEmitter {\n\tprivate processes = new Map<string, LifeguardProcess>();\n\tprivate lastHeartbeat = new Map<string, number>();\n\tprivate timeouts = new Map<string, NodeJS.Timeout>();\n\t/** When each task started, used to compute the load-scaled hard timeout. */\n\tprivate startedAt = new Map<string, number>();\n\t/** Per-agent base hard timeout (before load scaling), captured at monitor(). */\n\tprivate baseTimeoutMs = new Map<string, number>();\n\tprivate checkInterval: NodeJS.Timeout | null = null;\n\t/** Wall-clock time the heartbeat check last ran, to measure event-loop lag. */\n\tprivate lastCheckAt = Date.now();\n\t/**\n\t * Count of external in-process tasks (e.g. background MCP tools) running in the\n\t * parent alongside the monitored subagents. These don't show up in `processes`\n\t * but still saturate the parent's CPU/event loop, so they're folded into the\n\t * load multiplier. Updated by the pool via setExternalLoad().\n\t */\n\tprivate externalLoad = 0;\n\t/**\n\t * Tasks already reaped (stalled/timeout kill sent) but whose `exit` has not\n\t * fired yet. The heartbeat check runs every 5s, so without this a stalled task\n\t * would re-emit \"stalled\" (and re-kill) on every tick until the process exits.\n\t */\n\tprivate reaping = new Set<string>();\n\tprivate disposed = false;\n\tprivate readonly cwd: string;\n\tprivate parentShutdownHandler?: () => void;\n\n\tconstructor(cwd: string) {\n\t\tsuper();\n\t\tthis.cwd = cwd;\n\t\tthis.setupParentExitHandlers();\n\t\tthis.sweepOldAgents();\n\t\tthis.lastCheckAt = Date.now();\n\t\tthis.checkInterval = setInterval(() => this.checkHeartbeats(), HEARTBEAT_CHECK_INTERVAL_MS);\n\t}\n\n\t/**\n\t * Set the count of external in-process tasks (background MCP tools) sharing the\n\t * parent's CPU/event loop. Folded into loadMultiplier() so concurrent MCP work\n\t * widens the heartbeat/timeout budgets just like extra monitored subagents do.\n\t * Negative values are clamped to 0.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.externalLoad = Math.max(0, Math.floor(count));\n\t}\n\n\t/**\n\t * Tolerance multiplier for the current load. 1 concurrent task → 1x; each extra\n\t * concurrent task adds LOAD_TOLERANCE_PER_PROCESS, capped at MAX_LOAD_MULTIPLIER.\n\t * Concurrency counts both monitored subagents and external load (background MCP\n\t * tools), since both contend for the same parent event loop.\n\t */\n\tprivate loadMultiplier(): number {\n\t\tconst concurrent = this.processes.size + this.externalLoad;\n\t\tconst mult = 1 + Math.max(0, concurrent - 1) * LOAD_TOLERANCE_PER_PROCESS;\n\t\treturn Math.min(mult, MAX_LOAD_MULTIPLIER);\n\t}\n\n\t/**\n\t * Begin monitoring a child process. The process must emit a\n\t * `{\"ping\":true}` JSON line on stdout every 30 seconds.\n\t */\n\tmonitor(task_id: string, agent_type: string, proc: ChildProcess): void {\n\t\tif (this.disposed) return;\n\n\t\tconst pid = proc.pid ?? 0;\n\t\tthis.processes.set(task_id, { pid, task_id, agent_type, process: proc });\n\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\n\t\tconst timeoutMs = TIMEOUTS_MS[agent_type] ?? TIMEOUTS_MS.explore;\n\t\tthis.startedAt.set(task_id, Date.now());\n\t\tthis.baseTimeoutMs.set(task_id, timeoutMs);\n\t\t// Arm the hard timeout scaled by current load. When it fires, handleTimeout\n\t\t// re-checks load and re-arms (up to MAX_LOAD_MULTIPLIER) if the pool is still\n\t\t// busy, so a slow-but-progressing subagent isn't killed for CPU contention.\n\t\tconst timeout = setTimeout(\n\t\t\t() => {\n\t\t\t\tthis.handleTimeout(task_id);\n\t\t\t},\n\t\t\tMath.round(timeoutMs * this.loadMultiplier()),\n\t\t);\n\t\tthis.timeouts.set(task_id, timeout);\n\n\t\tproc.once(\"exit\", () => {\n\t\t\tthis.untrack(task_id);\n\t\t});\n\t}\n\n\t/** Record a heartbeat for a monitored task. */\n\trecordHeartbeat(task_id: string): void {\n\t\tif (this.processes.has(task_id)) {\n\t\t\tthis.lastHeartbeat.set(task_id, Date.now());\n\t\t}\n\t}\n\n\t/** Get the last recorded heartbeat timestamp, or null. */\n\tlastHeartbeatAt(task_id: string): number | null {\n\t\treturn this.lastHeartbeat.get(task_id) ?? null;\n\t}\n\n\t/** True if the task is currently being monitored. */\n\tisMonitoring(task_id: string): boolean {\n\t\treturn this.processes.has(task_id);\n\t}\n\n\t/** Kill all monitored processes and clean up. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tif (this.checkInterval) {\n\t\t\tclearInterval(this.checkInterval);\n\t\t\tthis.checkInterval = null;\n\t\t}\n\n\t\tfor (const timeout of this.timeouts.values()) {\n\t\t\tclearTimeout(timeout);\n\t\t}\n\t\tthis.timeouts.clear();\n\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\tthis.killTree(monitored);\n\t\t}\n\t\tthis.processes.clear();\n\t\tthis.lastHeartbeat.clear();\n\t\tthis.startedAt.clear();\n\t\tthis.baseTimeoutMs.clear();\n\t\tthis.reaping.clear();\n\t\tthis.removeAllListeners();\n\n\t\tif (this.parentShutdownHandler) {\n\t\t\tprocess.removeListener(\"SIGINT\", this.parentShutdownHandler);\n\t\t\tprocess.removeListener(\"SIGTERM\", this.parentShutdownHandler);\n\t\t}\n\t}\n\n\tprivate checkHeartbeats(): void {\n\t\tconst now = Date.now();\n\t\t// Event-loop lag: how much later than scheduled this check actually ran.\n\t\t// Heavy CPU load (many concurrent subagents + background MCP tools) starves\n\t\t// the loop, delaying both this check *and* our reading of child heartbeats.\n\t\t// Forgive that gap so the parent's own starvation isn't charged against the\n\t\t// children as a missed heartbeat.\n\t\tconst loopLag = Math.max(0, now - this.lastCheckAt - HEARTBEAT_CHECK_INTERVAL_MS);\n\t\tthis.lastCheckAt = now;\n\n\t\tconst threshold = HEARTBEAT_MISS_THRESHOLD_MS * this.loadMultiplier() + loopLag;\n\t\tfor (const [task_id] of this.processes) {\n\t\t\tif (this.reaping.has(task_id)) continue;\n\t\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\t\tif (last === undefined) continue;\n\t\t\tif (now - last > threshold) {\n\t\t\t\tthis.handleStalled(task_id);\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate handleStalled(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\t\tthis.reaping.add(task_id);\n\n\t\t// Record why we reaped this child so a recurrence is diagnosable rather than\n\t\t// just \"stalled\": how long since the last heartbeat, and the load factors\n\t\t// (concurrent monitored subagents, threshold) that fed the decision.\n\t\tconst last = this.lastHeartbeat.get(task_id);\n\t\tconst silentMs = last === undefined ? -1 : Date.now() - last;\n\t\tagentLog(\n\t\t\t`[LIFEGUARD] stalled task_id=${task_id} agent=${monitored.agent_type} ` +\n\t\t\t\t`silent_ms=${silentMs} concurrent=${this.processes.size} ` +\n\t\t\t\t`load_mult=${this.loadMultiplier().toFixed(2)} base_threshold_ms=${HEARTBEAT_MISS_THRESHOLD_MS}`,\n\t\t);\n\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"stalled\", { task_id, pid: monitored.pid });\n\t\t// Process exit handler will call untrack()\n\t}\n\n\tprivate handleTimeout(task_id: string): void {\n\t\tconst monitored = this.processes.get(task_id);\n\t\tif (!monitored || this.reaping.has(task_id)) return;\n\n\t\t// Under load the wall-clock timer can fire while the subagent is still doing\n\t\t// real work — its turns are just slow because the CPU is shared. Re-arm rather\n\t\t// than kill, up to a hard ceiling (base * MAX_LOAD_MULTIPLIER) so a genuinely\n\t\t// stuck agent still terminates.\n\t\tconst started = this.startedAt.get(task_id) ?? Date.now();\n\t\tconst base = this.baseTimeoutMs.get(task_id) ?? TIMEOUTS_MS[monitored.agent_type] ?? TIMEOUTS_MS.explore;\n\t\tconst elapsed = Date.now() - started;\n\t\tconst ceiling = base * MAX_LOAD_MULTIPLIER;\n\t\tif (this.loadMultiplier() > 1 && elapsed < ceiling) {\n\t\t\tconst remaining = ceiling - elapsed;\n\t\t\tconst next = Math.min(\n\t\t\t\tMath.round(base * this.loadMultiplier()),\n\t\t\t\tMath.max(HEARTBEAT_CHECK_INTERVAL_MS, remaining),\n\t\t\t);\n\t\t\tthis.timeouts.set(\n\t\t\t\ttask_id,\n\t\t\t\tsetTimeout(() => this.handleTimeout(task_id), next),\n\t\t\t);\n\t\t\treturn;\n\t\t}\n\n\t\tthis.reaping.add(task_id);\n\t\tthis.killTree(monitored);\n\n\t\tthis.emit(\"timeout\", { task_id, pid: monitored.pid });\n\t\tthis.timeouts.delete(task_id);\n\t\t// Process exit handler will call untrack()\n\t}\n\n\t/**\n\t * Kill a monitored subagent and everything it spawned. The pool spawns\n\t * children detached on POSIX (each leads its own process group), so the\n\t * group/tree kill reaches nested grandchildren (the subagent's bash commands,\n\t * its own subagents) that a single-PID SIGKILL would orphan. A pid of 0 means\n\t * the spawn never produced a process — nothing to kill (and `kill(-0)` would\n\t * signal our own process group).\n\t */\n\tprivate killTree(monitored: LifeguardProcess): void {\n\t\tif (monitored.pid > 0) {\n\t\t\tkillProcessTree(monitored.pid);\n\t\t} else if (!monitored.process.killed) {\n\t\t\tmonitored.process.kill(\"SIGKILL\");\n\t\t}\n\t}\n\n\tprivate untrack(task_id: string): void {\n\t\tconst timeout = this.timeouts.get(task_id);\n\t\tif (timeout) {\n\t\t\tclearTimeout(timeout);\n\t\t\tthis.timeouts.delete(task_id);\n\t\t}\n\t\tthis.processes.delete(task_id);\n\t\tthis.lastHeartbeat.delete(task_id);\n\t\tthis.startedAt.delete(task_id);\n\t\tthis.baseTimeoutMs.delete(task_id);\n\t\tthis.reaping.delete(task_id);\n\t}\n\n\tprivate setupParentExitHandlers(): void {\n\t\tconst shutdown = () => this.gracefulShutdown();\n\t\tthis.parentShutdownHandler = shutdown;\n\t\tprocess.setMaxListeners(Math.max(process.getMaxListeners(), 20));\n\t\tprocess.once(\"SIGINT\", shutdown);\n\t\tprocess.once(\"SIGTERM\", shutdown);\n\t}\n\n\tprivate gracefulShutdown(): void {\n\t\t// SIGTERM each child's process group (POSIX; the pool spawns them detached\n\t\t// as group leaders) so their own descendants get the graceful signal too.\n\t\tfor (const monitored of this.processes.values()) {\n\t\t\ttry {\n\t\t\t\tif (process.platform !== \"win32\" && monitored.pid > 0) {\n\t\t\t\t\tprocess.kill(-monitored.pid, \"SIGTERM\");\n\t\t\t\t} else if (!monitored.process.killed) {\n\t\t\t\t\tmonitored.process.kill(\"SIGTERM\");\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Already gone.\n\t\t\t}\n\t\t}\n\n\t\t// Kill whole trees after the grace period\n\t\tsetTimeout(() => {\n\t\t\tfor (const monitored of this.processes.values()) {\n\t\t\t\tthis.killTree(monitored);\n\t\t\t}\n\t\t}, PARENT_SHUTDOWN_GRACE_MS).unref();\n\t}\n\n\tprivate sweepOldAgents(): void {\n\t\tconst dispatchDir = getDispatchRoot(this.cwd);\n\t\tif (!existsSync(dispatchDir)) return;\n\n\t\tconst now = Date.now();\n\t\tconst cutoff = 24 * 60 * 60 * 1000; // 24 hours\n\n\t\tfor (const entry of readdirSync(dispatchDir)) {\n\t\t\tconst entryPath = join(dispatchDir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (!stats.isDirectory()) continue;\n\n\t\t\t\tif (now - stats.mtimeMs > cutoff) {\n\t\t\t\t\tconst hasRunningPid = this.hasRunningPid(entryPath);\n\t\t\t\t\tif (!hasRunningPid) {\n\t\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t\t}\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore errors for individual entries\n\t\t\t}\n\t\t}\n\t}\n\n\tprivate hasRunningPid(dir: string): boolean {\n\t\tconst pidFile = join(dir, \"pid\");\n\t\tif (!existsSync(pidFile)) return false;\n\n\t\ttry {\n\t\t\tconst pid = Number.parseInt(readFileSync(pidFile, \"utf-8\"), 10);\n\t\t\tif (Number.isNaN(pid)) return false;\n\t\t\tprocess.kill(pid, 0); // Check if process exists\n\t\t\treturn true;\n\t\t} catch {\n\t\t\treturn false;\n\t\t}\n\t}\n\n\tprivate rmrf(dir: string): void {\n\t\tfor (const entry of readdirSync(dir)) {\n\t\t\tconst entryPath = join(dir, entry);\n\t\t\ttry {\n\t\t\t\tconst stats = statSync(entryPath);\n\t\t\t\tif (stats.isDirectory()) {\n\t\t\t\t\tthis.rmrf(entryPath);\n\t\t\t\t} else {\n\t\t\t\t\tunlinkSync(entryPath);\n\t\t\t\t}\n\t\t\t} catch {\n\t\t\t\t// Ignore\n\t\t\t}\n\t\t}\n\t\ttry {\n\t\t\trmdirSync(dir);\n\t\t} catch {\n\t\t\t// Ignore\n\t\t}\n\t}\n}\n"]}

package/dist/core/subagent-pool.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"subagent-pool.d.ts","sourceRoot":"","sources":["../../src/core/subagent-pool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAG3C,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,yBAAyB,CAAC;AAY1D,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AAmBtD,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;OAIG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8FAA8F;IAC9F,yBAAyB,CAAC,EAAE,OAAO,CAAC;CACpC;AAED,MAAM,WAAW,YAAY;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,UAAU,CAAC,OAAO,KAAK,CAAC,CAAC;CAClC;AAED,MAAM,WAAW,cAAc;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,yEAAyE;IACzE,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,0DAA0D;IAC1D,MAAM,CAAC,EAAE,UAAU,GAAG,SAAS,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,WAAW,CAAC;IACjF,8EAA8E;IAC9E,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACtC,2FAA2F;IAC3F,0BAA0B,CAAC,EAAE,OAAO,CAAC;CACrC;AAED,MAAM,WAAW,UAAU;IAC1B,qFAAqF;IACrF,cAAc,EAAE,OAAO,CAAC;IACxB,2CAA2C;IAC3C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,sCAAsC;IACtC,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,+CAA+C;IAC/C,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,eAAe;IAC/B;gFAC4E;IAC5E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,wEAAwE;IACxE,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iCAAiC;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,kEAAkE;IAClE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;sFACkF;IAClF,MAAM,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,mBAAmB;IACnC,0FAA0F;IAC1F,UAAU,EAAE,MAAM,CAAC;IACnB,+EAA+E;IAC/E,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,yDAAyD;IACzD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,0EAA0E;IAC1E,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,sDAAsD;IACtD,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC;IACxB,oDAAoD;IACpD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,8CAA8C;IAC9C,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB;;;;OAIG;IACH,eAAe,CAAC,EAAE,SAAS,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;CACxC;AAED;;;;GAIG;AACH,eAAO,MAAM,0BAA0B,KAAK,CAAC;AAE7C;;;;;;;GAOG;AACH,eAAO,MAAM,yBAAyB,qBAA2B,CAAC;AAuBlE,mFAAmF;AACnF,MAAM,MAAM,kBAAkB,GAC3B;IAAE,IAAI,EAAE,WAAW,CAAA;CAAE,GACrB;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,GACpD;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,CAAC;AAEtB;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,kBAAkB,CAcrE;AAED;;;;;;;;;;;;;;;GAeG;AACH,qBAAa,YAAa,SAAQ,YAAY;IAC7C,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAS;IACxC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;IACpC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAW;IACtC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAS;IAC7B,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAoB;IACxC,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAS;IAC5C,+EAA+E;IAC/E,OAAO,CAAC,UAAU,CAAW;IAE7B,OAAO,CAAC,KAAK,CAAmC;IAChD,OAAO,CAAC,KAAK,CAA0B;IACvC,OAAO,CAAC,SAAS,CAAqC;IACtD,OAAO,CAAC,OAAO,CAAkG;IACjH,OAAO,CAAC,OAAO,CAAkC;IACjD,OAAO,CAAC,QAAQ,CAAwB;IACxC,OAAO,CAAC,SAAS,CAAoB;IACrC,OAAO,CAAC,QAAQ,CAAS;IACzB,kFAAkF;IAClF,OAAO,CAAC,QAAQ,CAAC,CAAgB;IACjC,gGAAgG;IAChG,OAAO,CAAC,WAAW,CAA0D;IAC7E,qEAAqE;IACrE,OAAO,CAAC,UAAU,CAA8E;IAChG,8CAA8C;IAC9C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAW;IACrC,kFAAkF;IAClF,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAwB;IAExD,YAAY,OAAO,EAAE,mBAAmB,EAqBvC;IAED,qEAAqE;IACrE,gBAAgB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,IAAI,CAEtC;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAEnC;IAED,0DAA0D;IAC1D,OAAO,CAAC,WAAW;IAOnB,gDAAgD;IAChD,OAAO,CAAC,UAAU;IAMlB,qDAAqD;IACrD,KAAK,CAAC,IAAI,EAAE,gBAAgB,GAAG,IAAI,CAoBlC;IAED,gCAAgC;IAChC,UAAU,CACT,OAAO,EAAE,MAAM,GACb,SAAS,GAAG,QAAQ,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,WAAW,GAAG,SAAS,CAc5F;IAED;;;;;;OAMG;IACH,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CA0B/B;IAED,yDAAyD;IACzD,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,cAAc,CAAC,CAcjD;IAED,6CAA6C;IAC7C,aAAa,IAAI,MAAM,CAEtB;IAED,4CAA4C;IAC5C,YAAY,IAAI,MAAM,CAErB;IAED;;;;;;;;OAQG;IACG,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,GAAE,eAAoB,GAAG,OAAO,CAAC,UAAU,CAAC,CAiB/E;IAED;;;OAGG;IACH,gBAAgB,CACf,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,eAAoB,GAC3B;QAAE,cAAc,EAAE,OAAO,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CASrF;IAED;;;OAGG;IACH,OAAO,CAAC,aAAa;IA4CrB;;;;OAIG;IACH,OAAO,CAAC,OAAO,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAEnD;IAED,8DAA8D;IAC9D,cAAc,CAAC,OAAO,EAAE,MAAM,EAAE,GAAG,GAAE,MAAiB,GAAG,MAAM,CAE9D;IAED;;;;OAIG;IACG,MAAM,CACX,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,MAAM,EACd,OAAO,GAAE,IAAI,CAAC,eAAe,EAAE,YAAY,GAAG,aAAa,CAAM,GAC/D,OAAO,CAAC,UAAU,CAAC,CAUrB;IAED,gFAAgF;IAChF,OAAO,CAAC,qBAAqB;IAW7B,OAAO,CAAC,gBAAgB;IA0BxB,OAAO,CAAC,eAAe;IAsBvB;;;OAGG;IACH,OAAO,CAAC,kBAAkB;IAQ1B,+EAA+E;IAC/E,OAAO,IAAI,IAAI,CA8Bd;IAED,2DAA2D;IAC3D,OAAO,CAAC,IAAI;IAOZ,sCAAsC;IACtC,OAAO,CAAC,SAAS;IAsGjB;;;;;;;;;OASG;IACH,OAAO,CAAC,aAAa;IAkBrB,kEAAkE;IAClE,OAAO,CAAC,SAAS;IA4SjB,kFAAkF;IAClF,OAAO,CAAC,6BAA6B;IAiBrC,oFAAoF;IACpF,OAAO,CAAC,6BAA6B;IAQrC,yEAAyE;IACzE,OAAO,CAAC,qBAAqB;IAa7B;;;;OAIG;IACH,OAAO,CAAC,mBAAmB;IAiB3B,OAAO,CAAC,iBAAiB;IAWzB,OAAO,CAAC,aAAa;CAgBrB","sourcesContent":["import { spawn } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport type { Api, Model } from \"@kolisachint/hoocode-ai\";\nimport { getDispatchTaskDir } from \"../config.js\";\nimport { attachJsonlLineReader } from \"../modes/rpc/jsonl.js\";\nimport { writeFileAtomicSync } from \"../utils/atomic-file.js\";\nimport { waitForChildProcess } from \"../utils/child-process.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\nimport { MODEL_INHERIT } from \"./agent-frontmatter.js\";\nimport { type AgentRegistry, loadAgentRegistry } from \"./agent-registry.js\";\nimport { DispatchEvaluator } from \"./dispatch-evaluator.js\";\nimport { SubagentLifeguard } from \"./lifeguard.js\";\nimport { resolveModelReference } from \"./model-categories.js\";\nimport { OutputVerifier } from \"./output-verifier.js\";\nimport type { Settings } from \"./settings-manager.js\";\nimport {\n\tcurrentSubagentDepth,\n\tDEFER_MCP_SCHEMAS_ENV,\n\tresolveMaxSubagentDepth,\n\tSUBAGENT_DEPTH_ENV,\n\tSUBAGENT_SKIP_MCP_ENV,\n\ttoolAllowlistNeedsMcp,\n} from \"./subagent-depth.js\";\nimport { SUBAGENT_PROGRESS_EVENTS } from \"./subagent-events.js\";\nimport { TokenBudget } from \"./token-budget.js\";\n\n/**\n * Provider/model failure signatures where inheriting the parent model can\n * recover. Compiled once at module load instead of on every error check.\n */\nconst INHERITED_MODEL_FALLBACK_ERROR =\n\t/usage[_\\s-]?limit|subscription|quota|rate.?limit|too many requests|429|insufficient|out of credit|credit balance|billing|payment required|402|model[^\\n]*(not found|unavailable|not available|not supported|does not exist|invalid|unsupported)|no api key|no auth configured|authentication|unauthorized|forbidden|permission/i;\n\nexport interface SubagentPoolTask {\n\ttask_id: string;\n\tagent_type: string;\n\ttask: string;\n\tcontext?: string;\n\ttoken_budget?: number;\n\tcwd?: string;\n\tmodel?: string;\n\tprovider?: string;\n\t/**\n\t * Explicit session file for the child to persist/continue. When omitted the\n\t * child uses its own dispatch dir (`<dispatch>/<task_id>/session.jsonl`).\n\t * Resume reuses the original task's session file to continue the transcript.\n\t */\n\tsessionFile?: string;\n\t/** Internal: retry using the caller's model when a built-in agent's preferred model fails. */\n\tuseInheritedModelFallback?: boolean;\n}\n\nexport interface SubagentSlot {\n\tpid: number;\n\tagent_type: string;\n\ttask_id: string;\n\tspawned_at: number;\n\ttoken_budget: number;\n\tprocess: ReturnType<typeof spawn>;\n}\n\nexport interface SubagentResult {\n\ttask_id: string;\n\tok: boolean;\n\tstdout: string;\n\tstderr: string;\n\texit_code: number | null;\n\terror?: string;\n\t/** True when the task exceeded its token budget and was hard-stopped. */\n\tbudget_exceeded?: boolean;\n\t/** Terminal status derived from how the task finished. */\n\tstatus?: \"complete\" | \"partial\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\";\n\t/** Parsed result.json content when available (e.g. on partial completion). */\n\tresult_data?: Record<string, unknown>;\n\t/** True when this run used the inherited-model fallback (preferred model failed first). */\n\tusedInheritedModelFallback?: boolean;\n}\n\nexport interface TaskResult {\n\t/** True when the evaluator decided the task is simple enough for inline handling. */\n\thandled_inline: boolean;\n\t/** Present when the task was delegated. */\n\ttask_id?: string;\n\tagent_type?: string;\n\treason?: string;\n\t/** Subagent result when delegated. */\n\tresult?: SubagentResult;\n\t/** Duration in milliseconds when delegated. */\n\tduration?: number;\n}\n\nexport interface DispatchOptions {\n\t/** Skip evaluation and force this agent type (user/explicit override).\n\t *  Accepts any registry-defined agent name, not just the built-in modes. */\n\tforceAgent?: string;\n\t/** Context distilled from the calling agent, passed to the subagent. */\n\tcontext?: string;\n\t/** Model id for the subagent (defaults to the child's configured default). */\n\tmodel?: string;\n\t/** Provider for the subagent. */\n\tprovider?: string;\n\t/** Explicit session file to persist/continue (used by resume). */\n\tsessionFile?: string;\n\t/** Caller-supplied task id. Defaults to a generated `dispatch-…` id. Lets a\n\t *  caller register liveness/inbox state under the id before dispatch resolves. */\n\ttaskId?: string;\n}\n\nexport interface SubagentPoolOptions {\n\t/** Path to the hoocode executable (or the runtime, e.g. node, when prefixArgs is set). */\n\texecutable: string;\n\t/** Args inserted before task args (e.g. the CLI entry script for node/tsx). */\n\tprefixArgs?: string[];\n\t/** Maximum concurrent child processes. Defaults to 5. */\n\tmaxConcurrency?: number;\n\t/** Working directory for spawned processes. Defaults to process.cwd(). */\n\tcwd?: string;\n\t/** Environment variables. Defaults to process.env. */\n\tenv?: NodeJS.ProcessEnv;\n\t/** Default token budget per task. Defaults to 0. */\n\tdefaultTokenBudget?: number;\n\t/**\n\t * Non-default skill paths to forward to every spawned subagent via --skill.\n\t * Subagents auto-discover skills from standard locations; only paths that\n\t * won't be found by default discovery need to be forwarded here.\n\t */\n\tskillPaths?: string[];\n\t/** Settings for model category resolution. */\n\tsettings?: Settings;\n\t/**\n\t * Available/configured models used to derive default model-category mappings\n\t * when a tier is not explicitly set in `settings.modelCategories`. Snapshotted\n\t * at pool creation, mirroring how `settings` is captured.\n\t */\n\tavailableModels?: readonly Model<Api>[];\n}\n\n/**\n * Default hard cap on assistant turns for a spawned subagent when its definition\n * does not set `maxTurns`. The token budget is advisory (it warns but never\n * kills), so this turn cap is the guaranteed hard stop for every subagent.\n */\nexport const DEFAULT_SUBAGENT_MAX_TURNS = 50;\n\n/**\n * AgentSession event `type`s forwarded from a subagent's json event stream as\n * `task_progress` events. Deliberately coarse; the child now also filters its\n * stdout to this set plus `message_end` at the source (see\n * SUBAGENT_STDOUT_EVENT_TYPES), so the per-delta firehose no longer crosses the\n * pipe. Re-exported from the shared module so the child emitter and this\n * consumer stay in lockstep.\n */\nexport const FORWARDED_SUBAGENT_EVENTS = SUBAGENT_PROGRESS_EVENTS;\n\n/**\n * Cap on the captured stdout/stderr text kept per task (tail-truncated). The\n * capture exists for diagnostics (failure reasons, output.json), so keeping the\n * most recent tail is enough; without a cap a chatty child could grow the\n * parent's memory without bound across a long swarm.\n */\nconst MAX_CAPTURED_STREAM_CHARS = 256 * 1024;\n\n/**\n * Cap on a single un-terminated JSONL line buffered from a child's stdout. The\n * forwarded events are small; anything approaching this is a runaway writer,\n * and the reader drops the line rather than buffering toward OOM.\n */\nconst MAX_SUBAGENT_EVENT_LINE_CHARS = 8 * 1024 * 1024;\n\n/** Append a chunk to a capped capture buffer, keeping the most recent tail. */\nfunction appendTail(current: string, chunk: string, cap: number = MAX_CAPTURED_STREAM_CHARS): string {\n\tconst next = current + chunk;\n\treturn next.length > cap ? next.slice(next.length - cap) : next;\n}\n\n/** The action the pool should take for one JSONL line from a subagent's stdout. */\nexport type SubagentStdoutLine =\n\t| { kind: \"heartbeat\" }\n\t| { kind: \"progress\"; event: Record<string, unknown> }\n\t| { kind: \"ignore\" };\n\n/**\n * Classify one JSONL line from a subagent's stdout into the action to take.\n * Pure (no side effects) so the ping/forward/drop policy is unit-testable without\n * spawning a child. Line framing — UTF-8-safe reassembly of chunks split mid-line\n * — is handled upstream by attachJsonlLineReader; this only sees complete lines.\n */\nexport function classifySubagentLine(line: string): SubagentStdoutLine {\n\tconst trimmed = line.trim();\n\tif (!trimmed.startsWith(\"{\")) return { kind: \"ignore\" };\n\tlet parsed: Record<string, unknown>;\n\ttry {\n\t\tparsed = JSON.parse(trimmed) as Record<string, unknown>;\n\t} catch {\n\t\treturn { kind: \"ignore\" };\n\t}\n\tif (parsed.ping === true) return { kind: \"heartbeat\" };\n\tif (typeof parsed.type === \"string\" && FORWARDED_SUBAGENT_EVENTS.has(parsed.type)) {\n\t\treturn { kind: \"progress\", event: parsed };\n\t}\n\treturn { kind: \"ignore\" };\n}\n\n/**\n * Pool for running hoocode subagents as child processes with bounded concurrency,\n * FIFO queuing with priority support, and automatic slot refill.\n *\n * Events:\n * - \"task_done\"    – task completed successfully and output was verified\n * - \"task_failed\"  – task failed (spawn error, bad exit code, verification failure)\n * - \"task_stalled\" – heartbeat missed past the load-scaled threshold (60s base,\n *                    widened under concurrency/event-loop lag), process SIGKILLed\n * - \"task_timeout\" – hard timeout exceeded, process was SIGKILLed\n * - \"task_cancelled\" – user-initiated cancel (see cancel()); process tree killed\n * - \"budget_warning\" – token usage crossed 80% threshold (advisory)\n * - \"budget_exceeded\" – token usage crossed 100% threshold (advisory; never kills)\n * - \"task_progress\" – coarse lifecycle event (turn_end, tool start/end) parsed\n *                    from the child's json event stream, for live UI updates\n */\nexport class SubagentPool extends EventEmitter {\n\tprivate readonly maxConcurrency: number;\n\tprivate readonly executable: string;\n\tprivate readonly prefixArgs: string[];\n\tprivate readonly cwd: string;\n\tprivate readonly env: NodeJS.ProcessEnv;\n\tprivate readonly defaultTokenBudget: number;\n\t/** Non-default skill paths forwarded to every spawned subagent via --skill. */\n\tprivate skillPaths: string[];\n\n\tprivate slots = new Map<string, SubagentSlot>();\n\tprivate queue: SubagentPoolTask[] = [];\n\tprivate completed = new Map<string, SubagentResult>();\n\tprivate waiters = new Map<string, { resolve: (result: SubagentResult) => void; reject: (err: Error) => void }>();\n\tprivate budgets = new Map<string, TokenBudget>();\n\tprivate verifier = new OutputVerifier();\n\tprivate lifeguard: SubagentLifeguard;\n\tprivate disposed = false;\n\t/** Lazily-loaded agent registry (frontmatter definitions) for this pool's cwd. */\n\tprivate registry?: AgentRegistry;\n\t/** Tracks why a task was killed (stalled / timeout / user cancel) before exit handler fires. */\n\tprivate killReasons = new Map<string, \"stalled\" | \"timeout\" | \"cancelled\">();\n\t/** Persistent terminal status map, survives wait_for consumption. */\n\tprivate taskStatus = new Map<string, \"done\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\">();\n\t/** Settings for model category resolution. */\n\tprivate readonly settings?: Settings;\n\t/** Available models used to derive default model-category mappings (snapshot). */\n\tprivate readonly availableModels: readonly Model<Api>[];\n\n\tconstructor(options: SubagentPoolOptions) {\n\t\tsuper();\n\t\tthis.maxConcurrency = options.maxConcurrency ?? 5;\n\t\tthis.executable = options.executable;\n\t\tthis.prefixArgs = options.prefixArgs ?? [];\n\t\tthis.cwd = options.cwd ?? process.cwd();\n\t\tthis.env = options.env ?? process.env;\n\t\tthis.defaultTokenBudget = options.defaultTokenBudget ?? 0;\n\t\tthis.skillPaths = options.skillPaths ? [...options.skillPaths] : [];\n\t\tthis.settings = options.settings;\n\t\tthis.availableModels = options.availableModels ?? [];\n\t\tthis.verifier = new OutputVerifier(this.cwd);\n\t\tthis.lifeguard = new SubagentLifeguard(this.cwd);\n\t\tthis.lifeguard.on(\"stalled\", (data: { task_id: string; pid: number }) => {\n\t\t\tthis.killReasons.set(data.task_id, \"stalled\");\n\t\t\tthis.emit(\"task_stalled\", data);\n\t\t});\n\t\tthis.lifeguard.on(\"timeout\", (data: { task_id: string; pid: number }) => {\n\t\t\tthis.killReasons.set(data.task_id, \"timeout\");\n\t\t\tthis.emit(\"task_timeout\", data);\n\t\t});\n\t}\n\n\t/** Update the non-default skill paths forwarded to new subagents. */\n\tupdateSkillPaths(paths: string[]): void {\n\t\tthis.skillPaths = [...paths];\n\t}\n\n\t/**\n\t * Report external in-process load (e.g. the number of background MCP tools\n\t * currently executing in the parent) to the lifeguard. This widens its\n\t * heartbeat/timeout tolerance so monitored subagents aren't false-positive\n\t * reaped when the parent's event loop is busy with concurrent background work.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.lifeguard.setExternalLoad(count);\n\t}\n\n\t/** Lazily load the agent registry for this pool's cwd. */\n\tprivate getRegistry(): AgentRegistry {\n\t\tif (!this.registry) {\n\t\t\tthis.registry = loadAgentRegistry({ cwd: this.cwd });\n\t\t}\n\t\treturn this.registry;\n\t}\n\n\t/** Priority value: higher numbers run first. */\n\tprivate priorityOf(agent_type: string): number {\n\t\t// Read-only investigation (explore/plan) often unblocks downstream work, so\n\t\t// it runs ahead of other agents.\n\t\treturn agent_type === \"explore\" || agent_type === \"plan\" ? 2 : 1;\n\t}\n\n\t/** Queue a task. It will run when a slot is free. */\n\tspawn(task: SubagentPoolTask): void {\n\t\tif (this.disposed) {\n\t\t\tthrow new Error(\"SubagentPool has been disposed\");\n\t\t}\n\t\tif (\n\t\t\tthis.slots.has(task.task_id) ||\n\t\t\tthis.queue.some((t) => t.task_id === task.task_id) ||\n\t\t\tthis.completed.has(task.task_id)\n\t\t) {\n\t\t\tthrow new Error(`Duplicate task_id: ${task.task_id}`);\n\t\t}\n\n\t\tconst p = this.priorityOf(task.agent_type);\n\t\tconst idx = this.queue.findIndex((t) => this.priorityOf(t.agent_type) < p);\n\t\tif (idx === -1) {\n\t\t\tthis.queue.push(task);\n\t\t} else {\n\t\t\tthis.queue.splice(idx, 0, task);\n\t\t}\n\t\tthis.pull();\n\t}\n\n\t/** Current status of a task. */\n\tget_status(\n\t\ttask_id: string,\n\t): \"running\" | \"queued\" | \"done\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\" | \"unknown\" {\n\t\tif (this.slots.has(task_id)) return \"running\";\n\t\tif (this.queue.some((t) => t.task_id === task_id)) return \"queued\";\n\t\tconst persisted = this.taskStatus.get(task_id);\n\t\tif (persisted) return persisted;\n\t\tconst result = this.completed.get(task_id);\n\t\tif (result) {\n\t\t\tif (result.status === \"stalled\") return \"stalled\";\n\t\t\tif (result.status === \"timeout\") return \"timeout\";\n\t\t\tif (result.status === \"cancelled\") return \"cancelled\";\n\t\t\tif (result.ok) return \"done\";\n\t\t\treturn \"failed\";\n\t\t}\n\t\treturn \"unknown\";\n\t}\n\n\t/**\n\t * Cancel a task on the user's behalf (Esc/abort mid-turn). A queued task is\n\t * removed and settles immediately; a running task's whole process tree is\n\t * killed and settles with status \"cancelled\" when its exit is observed\n\t * (unless it already wrote a valid result.json — completed work is honored).\n\t * Returns false for unknown/settled task ids.\n\t */\n\tcancel(task_id: string): boolean {\n\t\tconst queued = this.queue.findIndex((t) => t.task_id === task_id);\n\t\tif (queued !== -1) {\n\t\t\tthis.queue.splice(queued, 1);\n\t\t\tconst result: SubagentResult = {\n\t\t\t\ttask_id,\n\t\t\t\tok: false,\n\t\t\t\tstdout: \"\",\n\t\t\t\tstderr: \"\",\n\t\t\t\texit_code: null,\n\t\t\t\terror: \"cancelled before start\",\n\t\t\t\tstatus: \"cancelled\",\n\t\t\t};\n\t\t\tthis.emit(\"task_cancelled\", { task_id });\n\t\t\tthis.resolveWaiter(task_id, result);\n\t\t\treturn true;\n\t\t}\n\t\tconst slot = this.slots.get(task_id);\n\t\tif (!slot) return false;\n\t\tthis.killReasons.set(task_id, \"cancelled\");\n\t\tif (slot.pid > 0) {\n\t\t\tkillProcessTree(slot.pid);\n\t\t} else if (!slot.process.killed) {\n\t\t\tslot.process.kill(\"SIGKILL\");\n\t\t}\n\t\treturn true;\n\t}\n\n\t/** Wait for a task to complete and return its result. */\n\twait_for(task_id: string): Promise<SubagentResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\n\t\tconst existing = this.completed.get(task_id);\n\t\tif (existing) {\n\t\t\tthis.completed.delete(task_id);\n\t\t\treturn Promise.resolve(existing);\n\t\t}\n\n\t\treturn new Promise((resolve, reject) => {\n\t\t\tthis.waiters.set(task_id, { resolve, reject });\n\t\t});\n\t}\n\n\t/** Number of currently running subagents. */\n\trunning_count(): number {\n\t\treturn this.slots.size;\n\t}\n\n\t/** Number of tasks waiting in the queue. */\n\tqueued_count(): number {\n\t\treturn this.queue.length;\n\t}\n\n\t/**\n\t * Dispatch a task through the evaluator.\n\t *\n\t * - If `options.forceAgent` is provided, skip evaluation and spawn directly.\n\t * - Otherwise evaluate the task. If it should be handled inline, return\n\t *   `{ handled_inline: true }` immediately.\n\t * - If delegating, spawn the subagent, wait for completion, write\n\t *   `output.json`, and return the result.\n\t */\n\tasync dispatch(task: string, options: DispatchOptions = {}): Promise<TaskResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\t\tconst begin = this.beginDispatch(task, options);\n\t\tif (begin.handled_inline) {\n\t\t\treturn { handled_inline: true, reason: begin.reason };\n\t\t}\n\t\tconst result = await this.wait_for(begin.task_id);\n\t\treturn {\n\t\t\thandled_inline: false,\n\t\t\ttask_id: begin.task_id,\n\t\t\tagent_type: begin.agent_type,\n\t\t\treason: begin.reason,\n\t\t\tresult,\n\t\t\tduration: Date.now() - begin.startTime,\n\t\t};\n\t}\n\n\t/**\n\t * Fire-and-forget dispatch for background agents. Spawns the subagent and\n\t * returns its handle immediately; the caller polls get_status()/collect().\n\t */\n\tdispatchDetached(\n\t\ttask: string,\n\t\toptions: DispatchOptions = {},\n\t): { handled_inline: boolean; task_id?: string; agent_type?: string; reason?: string } {\n\t\tif (this.disposed) {\n\t\t\tthrow new Error(\"SubagentPool has been disposed\");\n\t\t}\n\t\tconst begin = this.beginDispatch(task, options);\n\t\tif (begin.handled_inline) {\n\t\t\treturn { handled_inline: true, reason: begin.reason };\n\t\t}\n\t\treturn { handled_inline: false, task_id: begin.task_id, agent_type: begin.agent_type, reason: begin.reason };\n\t}\n\n\t/**\n\t * Evaluate, log, and spawn a task without waiting. Shared by dispatch()\n\t * (blocking) and dispatchDetached() (background).\n\t */\n\tprivate beginDispatch(\n\t\ttask: string,\n\t\toptions: DispatchOptions,\n\t):\n\t\t| { handled_inline: true; reason?: string }\n\t\t| { handled_inline: false; task_id: string; agent_type: string; reason?: string; startTime: number } {\n\t\tconst { forceAgent, context, model, provider, sessionFile } = options;\n\t\tconst evaluator = new DispatchEvaluator();\n\t\tconst analysis = evaluator.evaluate(task);\n\n\t\tif (!forceAgent && !analysis.should_delegate) {\n\t\t\treturn { handled_inline: true, reason: analysis.reason };\n\t\t}\n\n\t\tconst agent_type = forceAgent ?? \"general-purpose\";\n\t\tconst task_id = options.taskId ?? `dispatch-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;\n\t\tconst reason = forceAgent ? \"user_override\" : analysis.reason;\n\t\tconst complexity = analysis.estimated_complexity;\n\t\t// Depth of the child about to be spawned (this process's depth + 1). Surfaced\n\t\t// so a delegation tree's nesting is visible in logs without extra tooling.\n\t\tconst childDepth = currentSubagentDepth(this.env) + 1;\n\n\t\t// Pre-dispatch logging. Use stderr: stdout is reserved for the JSON event\n\t\t// stream / TUI render and must not be polluted.\n\t\tconsole.error(\n\t\t\t`[DISPATCH] agent=${agent_type} depth=${childDepth} reason=${reason} complexity=${complexity} task_id=${task_id}`,\n\t\t);\n\t\tthis.writeDispatchLog(task_id, agent_type, reason, complexity, task, childDepth);\n\n\t\tconst poolTask: SubagentPoolTask = {\n\t\t\ttask_id,\n\t\t\tagent_type,\n\t\t\ttask,\n\t\t\tcontext,\n\t\t\tmodel,\n\t\t\tprovider,\n\t\t\tsessionFile,\n\t\t\tcwd: this.cwd,\n\t\t};\n\t\tconst startTime = Date.now();\n\t\tthis.spawn(poolTask);\n\t\treturn { handled_inline: false, task_id, agent_type, reason, startTime };\n\t}\n\n\t/**\n\t * Non-destructively read a completed task's result (for background polling).\n\t * Returns undefined while the task is still running/queued, or if its result\n\t * was already consumed via wait_for().\n\t */\n\tcollect(task_id: string): SubagentResult | undefined {\n\t\treturn this.completed.get(task_id);\n\t}\n\n\t/** Absolute path of the persisted session file for a task. */\n\tgetSessionFile(task_id: string, cwd: string = this.cwd): string {\n\t\treturn join(getDispatchTaskDir(cwd, task_id), \"session.jsonl\");\n\t}\n\n\t/**\n\t * Resume a previously dispatched subagent, continuing its persisted session\n\t * with a follow-up prompt. Recovers the original agent type from its dispatch\n\t * log. Rejects if no resumable session exists for the task.\n\t */\n\tasync resume(\n\t\ttask_id: string,\n\t\tprompt: string,\n\t\toptions: Omit<DispatchOptions, \"forceAgent\" | \"sessionFile\"> = {},\n\t): Promise<TaskResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\t\tconst sessionFile = this.getSessionFile(task_id);\n\t\tif (!existsSync(sessionFile)) {\n\t\t\treturn Promise.reject(new Error(`No resumable session for task \"${task_id}\" (expected ${sessionFile}).`));\n\t\t}\n\t\tconst agent_type = this.readDispatchAgentType(task_id) ?? \"general-purpose\";\n\t\treturn this.dispatch(prompt, { ...options, forceAgent: agent_type, sessionFile });\n\t}\n\n\t/** Recover the agent type a task was dispatched with, from its dispatch log. */\n\tprivate readDispatchAgentType(task_id: string): string | undefined {\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"dispatch-log.json\");\n\t\tif (!existsSync(path)) return undefined;\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(readFileSync(path, \"utf-8\")) as { agent_type?: string };\n\t\t\treturn typeof parsed.agent_type === \"string\" ? parsed.agent_type : undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate writeDispatchLog(\n\t\ttask_id: string,\n\t\tagent_type: string,\n\t\treason: string,\n\t\tcomplexity: string,\n\t\ttask: string,\n\t\tdepth: number,\n\t): void {\n\t\tconst log = {\n\t\t\ttimestamp: new Date().toISOString(),\n\t\t\ttask_id,\n\t\t\tagent_type,\n\t\t\tdepth,\n\t\t\treason,\n\t\t\tcomplexity,\n\t\t\ttask,\n\t\t};\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"dispatch-log.json\");\n\t\ttry {\n\t\t\tmkdirSync(dirname(path), { recursive: true });\n\t\t\twriteFileSync(path, JSON.stringify(log, null, 2));\n\t\t} catch {\n\t\t\t// Best-effort persistence\n\t\t}\n\t}\n\n\tprivate writeOutputJson(task_id: string, result: SubagentResult): void {\n\t\tconst output = {\n\t\t\ttask_id: result.task_id,\n\t\t\tok: result.ok,\n\t\t\texit_code: result.exit_code,\n\t\t\tstatus: result.status,\n\t\t\tstdout: result.stdout,\n\t\t\tstderr: result.stderr,\n\t\t\terror: result.error,\n\t\t\tbudget_exceeded: result.budget_exceeded,\n\t\t\tresult_data: result.result_data,\n\t\t};\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"output.json\");\n\t\ttry {\n\t\t\t// Atomic like result.json: background pollers may read output.json while\n\t\t\t// it is being (re)written.\n\t\t\twriteFileAtomicSync(path, JSON.stringify(output, null, 2));\n\t\t} catch {\n\t\t\t// Best-effort persistence\n\t\t}\n\t}\n\n\t/**\n\t * Remove a task's dispatch dir after a clean, verified success. Best-effort:\n\t * a cleanup failure must never fail an otherwise successful task.\n\t */\n\tprivate cleanupDispatchDir(task_id: string, cwd: string): void {\n\t\ttry {\n\t\t\trmSync(getDispatchTaskDir(cwd, task_id), { recursive: true, force: true });\n\t\t} catch {\n\t\t\t// Best-effort cleanup\n\t\t}\n\t}\n\n\t/** Kill all running processes, clear the queue, and reject pending waiters. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tfor (const slot of this.slots.values()) {\n\t\t\t// Kill the whole process group/tree, not just the direct child: a\n\t\t\t// subagent's own children (bash commands, nested subagents) must not\n\t\t\t// outlive the pool.\n\t\t\tif (slot.pid > 0) {\n\t\t\t\tkillProcessTree(slot.pid);\n\t\t\t} else if (!slot.process.killed) {\n\t\t\t\tslot.process.kill(\"SIGTERM\");\n\t\t\t}\n\t\t}\n\t\tthis.slots.clear();\n\t\tthis.queue = [];\n\n\t\tfor (const [task_id, waiter] of this.waiters) {\n\t\t\twaiter.reject(new Error(\"SubagentPool disposed\"));\n\t\t\tthis.waiters.delete(task_id);\n\t\t}\n\t\tthis.completed.clear();\n\t\tfor (const budget of this.budgets.values()) {\n\t\t\tbudget.removeAllListeners();\n\t\t}\n\t\tthis.budgets.clear();\n\t\tthis.killReasons.clear();\n\t\tthis.taskStatus.clear();\n\t\tthis.lifeguard.dispose();\n\t\tthis.removeAllListeners();\n\t}\n\n\t/** Pull tasks from the queue while slots are available. */\n\tprivate pull(): void {\n\t\twhile (this.slots.size < this.maxConcurrency && this.queue.length > 0) {\n\t\t\tconst task = this.queue.shift()!;\n\t\t\tthis.startTask(task, false);\n\t\t}\n\t}\n\n\t/** Build CLI arguments for a task. */\n\tprivate buildArgs(task: SubagentPoolTask): string[] {\n\t\t// Persist the child's session so a finished/interrupted subagent can be\n\t\t// resumed later (see resume()). SessionManager.open() creates the file on\n\t\t// first run and continues it on subsequent runs.\n\t\tconst sessionFile = task.sessionFile ?? this.getSessionFile(task.task_id, task.cwd ?? this.cwd);\n\t\tconst args: string[] = [\n\t\t\t...this.prefixArgs,\n\t\t\t\"--mode\",\n\t\t\t\"json\",\n\t\t\t\"--session\",\n\t\t\tsessionFile,\n\t\t\t\"--task-id\",\n\t\t\ttask.task_id,\n\t\t];\n\n\t\t// Prefer the data-driven agent definition from the registry; fall back to the\n\t\t// built-in mode prompt/allowlist for legacy modes not present in the registry.\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\n\t\tif (task.agent_type) {\n\t\t\tconst systemPrompt = def?.prompt;\n\t\t\tif (systemPrompt) {\n\t\t\t\targs.push(\"--system-prompt\", systemPrompt);\n\t\t\t}\n\n\t\t\t// A `delegate: true` agent may itself dispatch via the Task tool, but only\n\t\t\t// while the child it becomes can still nest (childDepth < cap) — so the\n\t\t\t// deepest permitted level cannot delegate further. Gating here keeps the\n\t\t\t// authorization explicit and bounded by the same cap as the depth guard.\n\t\t\tconst childDepth = currentSubagentDepth(this.env) + 1;\n\t\t\tconst canChildDelegate = def?.delegate === true && childDepth < resolveMaxSubagentDepth(undefined, this.env);\n\n\t\t\t// Tool allowlist comes from the agent definition's frontmatter `tools`\n\t\t\t// field (read-only built-ins declare their own sandbox). When omitted, no\n\t\t\t// --tools is passed and the subagent inherits all parent tools (so the Task\n\t\t\t// tool already survives). A delegating agent with an explicit allowlist must\n\t\t\t// have Task/TaskOutput added, or the child would filter them out.\n\t\t\tconst tools = def?.tools ? [...def.tools] : undefined;\n\t\t\tif (canChildDelegate && tools) {\n\t\t\t\tfor (const t of [\"Task\", \"TaskOutput\"]) {\n\t\t\t\t\tif (!tools.includes(t)) tools.push(t);\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (tools && tools.length > 0) {\n\t\t\t\targs.push(\"--tools\", tools.join(\",\"));\n\t\t\t}\n\t\t\tif (def?.disallowedTools && def.disallowedTools.length > 0) {\n\t\t\t\targs.push(\"--disallowed-tools\", def.disallowedTools.join(\",\"));\n\t\t\t}\n\n\t\t\t// Propagate subagent enablement so the child registers the Task tool; without\n\t\t\t// this the flag-based enablement would not reach a spawned child.\n\t\t\tif (canChildDelegate) {\n\t\t\t\targs.push(\"--enable-subagents\");\n\t\t\t\t// Scoped delegation: restrict which agent types this child may spawn.\n\t\t\t\tif (def?.delegateTo && def.delegateTo.length > 0) {\n\t\t\t\t\targs.push(\"--delegate-allow\", def.delegateTo.join(\",\"));\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Model precedence: a definition's explicit model wins (unless it is the\n\t\t// `inherit` sentinel), otherwise use the caller-provided model. Built-in\n\t\t// agents can retry with the inherited model when their preferred model is\n\t\t// unavailable or quota-limited.\n\t\tconst explicitModel =\n\t\t\t!task.useInheritedModelFallback && def?.model && def.model !== MODEL_INHERIT ? def.model : undefined;\n\t\t// Resolve a model-category reference (fast/standard/capable) to its\n\t\t// configured or derived model id. An unresolvable category yields undefined,\n\t\t// so no `--model` is passed and the child keeps its default model.\n\t\tconst rawModel = explicitModel ?? task.model;\n\t\tconst modelToUse = rawModel ? resolveModelReference(rawModel, this.settings, this.availableModels) : undefined;\n\t\tif (modelToUse) {\n\t\t\targs.push(\"--model\", modelToUse);\n\t\t}\n\t\t// Only pass --provider when the model doesn't already encode the provider\n\t\t// (e.g. \"anthropic/claude-sonnet-4-5\"). When both --model and --provider are\n\t\t// present, --provider wins and filters candidates to that provider, which\n\t\t// breaks models whose IDs contain a provider prefix from a different provider.\n\t\tif (task.provider && (!modelToUse || !modelToUse.includes(\"/\"))) {\n\t\t\targs.push(\"--provider\", task.provider);\n\t\t}\n\n\t\t// Always give subagents a hard turn cap. With the token budget now advisory\n\t\t// (warn-only), this is the guaranteed hard stop for a runaway subagent.\n\t\tconst maxTurns = def?.maxTurns && def.maxTurns > 0 ? def.maxTurns : DEFAULT_SUBAGENT_MAX_TURNS;\n\t\targs.push(\"--max-turns\", String(maxTurns));\n\n\t\t// Forward non-default skill paths so the subagent has access to all parent skills.\n\t\t// Standard discovery locations (~/.hoocode/, .hoocode/, .claude/) are found automatically.\n\t\tfor (const skillPath of this.skillPaths) {\n\t\t\targs.push(\"--skill\", skillPath);\n\t\t}\n\n\t\tconst prompt = task.context?.trim()\n\t\t\t? `Context from the calling agent:\\n\\n${task.context.trim()}\\n\\nTask: ${task.task.trim()}`\n\t\t\t: `Task: ${task.task.trim()}`;\n\t\targs.push(prompt);\n\n\t\treturn args;\n\t}\n\n\t/**\n\t * Environment for a spawned child.\n\t *\n\t * Stamps the child's depth (parent depth + 1) so its own guard knows where it\n\t * sits in the tree; the tree-wide cap (HOOCODE_SUBAGENT_MAX_DEPTH) is inherited\n\t * via the spread, so at the default cap of 1 the child lands at depth 1 and\n\t * cannot spawn further subagents. Also flags the child to skip MCP server\n\t * connection when its tool allowlist is explicit and MCP-free — connecting\n\t * external servers it can never call is pure boot latency.\n\t */\n\tprivate childSpawnEnv(task: SubagentPoolTask): NodeJS.ProcessEnv {\n\t\tconst env: NodeJS.ProcessEnv = {\n\t\t\t...this.env,\n\t\t\t[SUBAGENT_DEPTH_ENV]: String(currentSubagentDepth(this.env) + 1),\n\t\t};\n\t\t// Use the agent's own frontmatter allowlist (the same source buildArgs passes\n\t\t// via --tools). A delegating child also gets Task/TaskOutput appended there,\n\t\t// but neither is an MCP tool, so the decision is unchanged by that.\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\t\tif (!toolAllowlistNeedsMcp(def?.tools)) {\n\t\t\tenv[SUBAGENT_SKIP_MCP_ENV] = \"1\";\n\t\t}\n\t\t// A child never defers MCP schemas: if it needs MCP it eager-registers its\n\t\t// allowlisted tools at dispatch so they are immediately callable (spec §2).\n\t\tdelete env[DEFER_MCP_SCHEMAS_ENV];\n\t\treturn env;\n\t}\n\n\t/** Start a task in a child process, with one retry on failure. */\n\tprivate startTask(task: SubagentPoolTask, isRetry: boolean): void {\n\t\t// Get or create a TokenBudget tracker. On retry, reuse the existing one\n\t\t// so cumulative usage persists across retries.\n\t\tlet budget = this.budgets.get(task.task_id);\n\t\tif (!budget) {\n\t\t\tbudget = new TokenBudget(task.task_id, task.agent_type, {\n\t\t\t\tlimit: task.token_budget,\n\t\t\t\tcwd: task.cwd ?? this.cwd,\n\t\t\t});\n\t\t\tbudget.on(\"budget_warning\", (data: { task_id: string; message: string; used: number; limit: number }) => {\n\t\t\t\tthis.emit(\"budget_warning\", data);\n\t\t\t});\n\t\t\t// The token budget is advisory: surface telemetry but never kill. The\n\t\t\t// guaranteed hard stop is the per-subagent turn cap (--max-turns); see\n\t\t\t// DEFAULT_SUBAGENT_MAX_TURNS.\n\t\t\tbudget.on(\"budget_exceeded\", (data: { task_id: string; used: number; limit: number }) => {\n\t\t\t\tthis.emit(\"budget_exceeded\", data);\n\t\t\t});\n\t\t\tthis.budgets.set(task.task_id, budget);\n\t\t}\n\n\t\tlet proc: ReturnType<typeof spawn>;\n\t\ttry {\n\t\t\tproc = spawn(this.executable, this.buildArgs(task), {\n\t\t\t\tcwd: task.cwd ?? this.cwd,\n\t\t\t\tenv: this.childSpawnEnv(task),\n\t\t\t\tshell: false,\n\t\t\t\tstdio: [\"ignore\", \"pipe\", \"pipe\"],\n\t\t\t\t// POSIX: the child leads its own process group so a kill can reach its\n\t\t\t\t// whole tree (see lifeguard/dispose). A single-PID SIGKILL orphans any\n\t\t\t\t// grandchildren the subagent spawned (its own bash commands, nested\n\t\t\t\t// subagents), which kept burning CPU after their parent was reaped.\n\t\t\t\tdetached: process.platform !== \"win32\",\n\t\t\t});\n\t\t} catch {\n\t\t\tif (!isRetry) {\n\t\t\t\tthis.startTask(task, true);\n\t\t\t} else {\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\terror: \"Spawn failed synchronously\",\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: false,\n\t\t\t\t\tstdout: \"\",\n\t\t\t\t\tstderr: \"\",\n\t\t\t\t\texit_code: null,\n\t\t\t\t\terror: \"Spawn failed synchronously\",\n\t\t\t\t\tstatus: \"failed\",\n\t\t\t\t});\n\t\t\t\tthis.pull();\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\n\t\tconst slot: SubagentSlot = {\n\t\t\tpid: proc.pid ?? 0,\n\t\t\tagent_type: task.agent_type,\n\t\t\ttask_id: task.task_id,\n\t\t\tspawned_at: Date.now(),\n\t\t\ttoken_budget: task.token_budget ?? this.defaultTokenBudget,\n\t\t\tprocess: proc,\n\t\t};\n\n\t\tthis.slots.set(task.task_id, slot);\n\t\tthis.lifeguard.monitor(task.task_id, task.agent_type, proc);\n\n\t\tlet stdout = \"\";\n\t\tlet stderr = \"\";\n\t\tlet detachStdoutReader: (() => void) | undefined;\n\n\t\tproc.stdout?.on(\"data\", (data: Buffer) => {\n\t\t\t// Capture (tail-capped) for diagnostics, and treat any output as a\n\t\t\t// heartbeat: a child busily streaming events is alive even when the\n\t\t\t// parent's starved event loop hasn't parsed its {\"ping\":true} line yet.\n\t\t\t// All *parsing* happens in the single JSONL reader below.\n\t\t\tstdout = appendTail(stdout, data.toString());\n\t\t\tthis.lifeguard.recordHeartbeat(task.task_id);\n\t\t});\n\n\t\t// The one parser on the child's stdout: UTF-8-safe, LF-only framing —\n\t\t// multi-byte characters and large events split across pipe chunks are\n\t\t// reassembled before parsing. Each complete line feeds the token budget\n\t\t// (usage telemetry) and the ping/forward/drop classifier; previously the\n\t\t// budget ran its own naive chunk-splitting parser on the same stream, which\n\t\t// corrupted lines split mid-character. The line buffer is capped so a\n\t\t// runaway writer cannot OOM the parent. Detached when the child exits (see\n\t\t// cleanup below); the reader also self-detaches on stream error/close.\n\t\tif (proc.stdout) {\n\t\t\tdetachStdoutReader = attachJsonlLineReader(\n\t\t\t\tproc.stdout,\n\t\t\t\t(line) => {\n\t\t\t\t\tbudget.processLine(line);\n\t\t\t\t\tconst action = classifySubagentLine(line);\n\t\t\t\t\tif (action.kind === \"heartbeat\") {\n\t\t\t\t\t\tthis.lifeguard.recordHeartbeat(task.task_id);\n\t\t\t\t\t} else if (action.kind === \"progress\") {\n\t\t\t\t\t\tthis.emit(\"task_progress\", {\n\t\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\t\tevent: action.event,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t},\n\t\t\t\t{ maxBuffer: MAX_SUBAGENT_EVENT_LINE_CHARS },\n\t\t\t);\n\t\t}\n\t\tproc.stderr?.on(\"data\", (data: Buffer) => {\n\t\t\tstderr = appendTail(stderr, data.toString());\n\t\t});\n\n\t\t// True when this attempt scheduled another run of the same task id (spawn\n\t\t// retry or inherited-model fallback). The retry reuses the task's TokenBudget\n\t\t// so usage accumulates across attempts; the .finally below must then leave\n\t\t// the budget (and its listeners) alone instead of tearing it down under the\n\t\t// retry's feet.\n\t\tlet retryScheduled = false;\n\n\t\twaitForChildProcess(proc)\n\t\t\t.then((code) => {\n\t\t\t\tthis.slots.delete(task.task_id);\n\t\t\t\tbudget.flush();\n\n\t\t\t\tconst killReason = this.killReasons.get(task.task_id);\n\t\t\t\tthis.killReasons.delete(task.task_id);\n\n\t\t\t\tconst duration = Date.now() - slot.spawned_at;\n\t\t\t\tconst tokens_used = budget.getUsed();\n\t\t\t\tconst budgetExceeded = budget.isExceeded();\n\n\t\t\t\t// A subagent's success is defined by a valid, verified result.json, not by\n\t\t\t\t// its exit code. A child that finished its work and wrote a valid result can\n\t\t\t\t// still be SIGKILLed by the lifeguard before it exits on its own (lingering\n\t\t\t\t// open handles delay a natural exit past the heartbeat threshold), which forces\n\t\t\t\t// exit_code === null. Keying completion off the verified result, not code === 0,\n\t\t\t\t// honors that genuine success instead of discarding it as a false stall.\n\t\t\t\tconst verification = this.verifier.verify(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\t// A well-formed result.json counts as clean completion unless its own\n\t\t\t\t// status field declares failure (e.g. \"failed\" from a provider quota\n\t\t\t\t// error). Without this check the pool would treat a child that wrote a\n\t\t\t\t// valid-but-failed result.json and exited non-zero as a success.\n\t\t\t\tlet cleanlyCompleted = code === 0 || verification.valid;\n\t\t\t\tif (cleanlyCompleted && verification.valid) {\n\t\t\t\t\tconst rd = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\t\tif (rd && (rd as Record<string, unknown>).status === \"failed\") {\n\t\t\t\t\t\tcleanlyCompleted = false;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// If killed (lifeguard reap or user cancel) before producing a valid\n\t\t\t\t// result, honor the kill; a child that already wrote a verified\n\t\t\t\t// result.json completed its work and settles as a success below.\n\t\t\t\tif (killReason !== undefined && !verification.valid) {\n\t\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tok: false,\n\t\t\t\t\t\tstdout,\n\t\t\t\t\t\tstderr,\n\t\t\t\t\t\texit_code: code,\n\t\t\t\t\t\tstatus: killReason,\n\t\t\t\t\t};\n\t\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\t\tthis.emit(`task_${killReason}`, {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\tduration,\n\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t});\n\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: cleanlyCompleted,\n\t\t\t\t\tstdout,\n\t\t\t\t\tstderr,\n\t\t\t\t\texit_code: code,\n\t\t\t\t\t// Advisory telemetry only: exceeding the budget never fails the task.\n\t\t\t\t\tbudget_exceeded: budgetExceeded,\n\t\t\t\t\tstatus: cleanlyCompleted ? \"complete\" : \"failed\",\n\t\t\t\t\tusedInheritedModelFallback: task.useInheritedModelFallback === true,\n\t\t\t\t};\n\n\t\t\t\tif (result.ok) {\n\t\t\t\t\tif (!verification.valid) {\n\t\t\t\t\t\tresult.ok = false;\n\t\t\t\t\t\tresult.error = verification.reason;\n\t\t\t\t\t\tresult.status = \"failed\";\n\t\t\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\t\tduration,\n\t\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t\t\terror: verification.reason,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\t// Attach the verified result.json so callers can read the summary\n\t\t\t\t\t// without parsing the raw event stream.\n\t\t\t\t\tresult.result_data = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\n\t\t\t\t\t// Clean success: discard the per-task dispatch dir entirely\n\t\t\t\t\t// (session.jsonl, result.json, dispatch-log.json, budget.json). The\n\t\t\t\t\t// in-memory result already carries result_data, so callers lose\n\t\t\t\t\t// nothing. Trade-off: resume() only works for non-successful tasks.\n\t\t\t\t\tthis.cleanupDispatchDir(task.task_id, task.cwd ?? this.cwd);\n\n\t\t\t\t\tthis.emit(\"task_done\", {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\tduration,\n\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t\tstatus: \"complete\",\n\t\t\t\t\t});\n\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\t// Failure path: keep the dispatch dir for debugging and persist output.\n\t\t\t\t// Attach the child's result.json (if any) and derive a concrete failure\n\t\t\t\t// reason so callers see the real cause (e.g. a provider usage/quota\n\t\t\t\t// error) instead of a generic \"subagent failed\".\n\t\t\t\tresult.result_data = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\tif (!result.error) {\n\t\t\t\t\tresult.error = this.deriveFailureReason(result);\n\t\t\t\t}\n\t\t\t\tif (this.shouldRetryWithInheritedModel(task, result)) {\n\t\t\t\t\tconsole.error(\n\t\t\t\t\t\t`[DISPATCH] agent=${task.agent_type} task_id=${task.task_id} preferred model failed; retrying with inherited model`,\n\t\t\t\t\t);\n\t\t\t\t\tthis.cleanupRetryArtifacts(task);\n\t\t\t\t\tretryScheduled = true;\n\t\t\t\t\tthis.queue.unshift({ ...task, useInheritedModelFallback: true });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\tduration,\n\t\t\t\t\ttokens_used,\n\t\t\t\t\terror: result.error ?? `Exited with code ${code}`,\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t})\n\t\t\t.catch((err) => {\n\t\t\t\tthis.slots.delete(task.task_id);\n\t\t\t\tbudget.flush();\n\t\t\t\tconst duration = Date.now() - slot.spawned_at;\n\t\t\t\tconst tokens_used = budget.getUsed();\n\t\t\t\tif (!isRetry) {\n\t\t\t\t\t// The retry (started synchronously here) looks the budget up from\n\t\t\t\t\t// this.budgets and reuses it; flag it so the .finally below doesn't\n\t\t\t\t\t// strip the budget's listeners or delete the entry the retry now owns.\n\t\t\t\t\tretryScheduled = true;\n\t\t\t\t\tthis.startTask(task, true);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst error = err instanceof Error ? err.message : String(err);\n\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: false,\n\t\t\t\t\tstdout,\n\t\t\t\t\tstderr,\n\t\t\t\t\texit_code: null,\n\t\t\t\t\terror,\n\t\t\t\t\tstatus: \"failed\",\n\t\t\t\t\tusedInheritedModelFallback: task.useInheritedModelFallback === true,\n\t\t\t\t};\n\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\tduration,\n\t\t\t\t\ttokens_used,\n\t\t\t\t\terror,\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t})\n\t\t\t.finally(() => {\n\t\t\t\t// Runs on success AND failure: the stdout reader used to be detached only\n\t\t\t\t// on the success path, leaking listeners (and their buffers) whenever a\n\t\t\t\t// child failed or the wait rejected.\n\t\t\t\tdetachStdoutReader?.();\n\t\t\t\tproc.stdout?.removeAllListeners(\"data\");\n\t\t\t\tproc.stderr?.removeAllListeners(\"data\");\n\t\t\t\tproc.stdout?.destroy();\n\t\t\t\tproc.stderr?.destroy();\n\t\t\t\tif (!retryScheduled) {\n\t\t\t\t\tbudget.removeAllListeners();\n\t\t\t\t\tthis.budgets.delete(task.task_id);\n\t\t\t\t}\n\t\t\t\tthis.pull();\n\t\t\t});\n\t}\n\n\t/** Whether a failed built-in subagent should be retried with `model: inherit`. */\n\tprivate shouldRetryWithInheritedModel(task: SubagentPoolTask, result: SubagentResult): boolean {\n\t\tif (task.useInheritedModelFallback) return false;\n\t\tif (task.sessionFile) return false;\n\t\t// Only the parent model is required: the provider may be unset when the\n\t\t// harness routes through a gateway. The retry inherits the parent model and\n\t\t// lets the child resolve the provider from its own default when none was threaded through.\n\t\tif (!task.model) return false;\n\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\t\t// Built-in agents always inherit; project agents may pin an explicit model in\n\t\t// frontmatter, so let them fall back too when that model is rejected.\n\t\tif (def?.source !== \"builtin\" && def?.source !== \"project\") return false;\n\t\tif (!def.model || def.model === MODEL_INHERIT) return false;\n\n\t\treturn this.isInheritedModelFallbackError(result);\n\t}\n\n\t/** Detect provider/model failures where inheriting the parent model can recover. */\n\tprivate isInheritedModelFallbackError(result: SubagentResult): boolean {\n\t\tconst text = [result.error, result.stderr, JSON.stringify(result.result_data ?? {})]\n\t\t\t.filter((part): part is string => typeof part === \"string\" && part.length > 0)\n\t\t\t.join(\"\\n\");\n\n\t\treturn INHERITED_MODEL_FALLBACK_ERROR.test(text);\n\t}\n\n\t/** Remove failed attempt artifacts before rerunning the same task id. */\n\tprivate cleanupRetryArtifacts(task: SubagentPoolTask): void {\n\t\tconst cwd = task.cwd ?? this.cwd;\n\t\tconst taskDir = getDispatchTaskDir(cwd, task.task_id);\n\t\tconst sessionFile = task.sessionFile ?? this.getSessionFile(task.task_id, cwd);\n\t\ttry {\n\t\t\trmSync(sessionFile, { force: true });\n\t\t\trmSync(join(taskDir, \"result.json\"), { force: true });\n\t\t\trmSync(join(taskDir, \"output.json\"), { force: true });\n\t\t} catch {\n\t\t\t// Best-effort cleanup; retry can still proceed with existing artifacts.\n\t\t}\n\t}\n\n\t/**\n\t * Best-effort concrete failure reason for a non-zero-exit subagent. Prefers\n\t * the child's result.json summary (which carries the provider/model error\n\t * message on failure), then the tail of stderr, then the exit code.\n\t */\n\tprivate deriveFailureReason(result: SubagentResult): string {\n\t\tconst summary = (result.result_data as { summary?: string } | undefined)?.summary?.trim();\n\t\tif (summary) {\n\t\t\treturn summary;\n\t\t}\n\t\tconst stderrTail = result.stderr\n\t\t\t.split(\"\\n\")\n\t\t\t.map((line) => line.trim())\n\t\t\t.filter((line) => line.length > 0)\n\t\t\t.slice(-5)\n\t\t\t.join(\"\\n\");\n\t\tif (stderrTail) {\n\t\t\treturn stderrTail;\n\t\t}\n\t\treturn `Exited with code ${result.exit_code}`;\n\t}\n\n\tprivate tryReadResultJson(task_id: string, cwd: string): Record<string, unknown> | undefined {\n\t\tconst path = join(getDispatchTaskDir(cwd, task_id), \"result.json\");\n\t\tif (!existsSync(path)) return undefined;\n\t\ttry {\n\t\t\tconst raw = readFileSync(path, \"utf-8\");\n\t\t\treturn JSON.parse(raw) as Record<string, unknown>;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate resolveWaiter(task_id: string, result: SubagentResult): void {\n\t\t// Persist terminal status for get_status() even after wait_for consumes the result\n\t\tif (result.status === \"stalled\") this.taskStatus.set(task_id, \"stalled\");\n\t\telse if (result.status === \"timeout\") this.taskStatus.set(task_id, \"timeout\");\n\t\telse if (result.status === \"cancelled\") this.taskStatus.set(task_id, \"cancelled\");\n\t\telse if (result.ok) this.taskStatus.set(task_id, \"done\");\n\t\telse this.taskStatus.set(task_id, \"failed\");\n\n\t\tconst waiter = this.waiters.get(task_id);\n\t\tif (waiter) {\n\t\t\twaiter.resolve(result);\n\t\t\tthis.waiters.delete(task_id);\n\t\t\treturn;\n\t\t}\n\t\tthis.completed.set(task_id, result);\n\t}\n}\n"]}
+{"version":3,"file":"subagent-pool.d.ts","sourceRoot":"","sources":["../../src/core/subagent-pool.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC3C,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAG3C,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,MAAM,yBAAyB,CAAC;AAa1D,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,uBAAuB,CAAC;AAmBtD,MAAM,WAAW,gBAAgB;IAChC,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB;;;;OAIG;IACH,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,8FAA8F;IAC9F,yBAAyB,CAAC,EAAE,OAAO,CAAC;CACpC;AAED,MAAM,WAAW,YAAY;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,UAAU,CAAC,OAAO,KAAK,CAAC,CAAC;CAClC;AAED,MAAM,WAAW,cAAc;IAC9B,OAAO,EAAE,MAAM,CAAC;IAChB,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,yEAAyE;IACzE,eAAe,CAAC,EAAE,OAAO,CAAC;IAC1B,0DAA0D;IAC1D,MAAM,CAAC,EAAE,UAAU,GAAG,SAAS,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,WAAW,CAAC;IACjF,8EAA8E;IAC9E,WAAW,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACtC,2FAA2F;IAC3F,0BAA0B,CAAC,EAAE,OAAO,CAAC;CACrC;AAED,MAAM,WAAW,UAAU;IAC1B,qFAAqF;IACrF,cAAc,EAAE,OAAO,CAAC;IACxB,2CAA2C;IAC3C,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,sCAAsC;IACtC,MAAM,CAAC,EAAE,cAAc,CAAC;IACxB,+CAA+C;IAC/C,QAAQ,CAAC,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,eAAe;IAC/B;gFAC4E;IAC5E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,wEAAwE;IACxE,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,8EAA8E;IAC9E,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,iCAAiC;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,kEAAkE;IAClE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB;sFACkF;IAClF,MAAM,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,mBAAmB;IACnC,0FAA0F;IAC1F,UAAU,EAAE,MAAM,CAAC;IACnB,+EAA+E;IAC/E,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,yDAAyD;IACzD,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,0EAA0E;IAC1E,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,sDAAsD;IACtD,GAAG,CAAC,EAAE,MAAM,CAAC,UAAU,CAAC;IACxB,oDAAoD;IACpD,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B;;;;OAIG;IACH,UAAU,CAAC,EAAE,MAAM,EAAE,CAAC;IACtB,8CAA8C;IAC9C,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB;;;;OAIG;IACH,eAAe,CAAC,EAAE,SAAS,KAAK,CAAC,GAAG,CAAC,EAAE,CAAC;CACxC;AAED;;;;GAIG;AACH,eAAO,MAAM,0BAA0B,KAAK,CAAC;AAE7C;;;;;;;GAOG;AACH,eAAO,MAAM,yBAAyB,qBAA2B,CAAC;AAuBlE,mFAAmF;AACnF,MAAM,MAAM,kBAAkB,GAC3B;IAAE,IAAI,EAAE,WAAW,CAAA;CAAE,GACrB;IAAE,IAAI,EAAE,UAAU,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CAAE,GACpD;IAAE,IAAI,EAAE,QAAQ,CAAA;CAAE,CAAC;AAEtB;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,kBAAkB,CAcrE;AAED;;;;;;;;;;;;;;;GAeG;AACH,qBAAa,YAAa,SAAQ,YAAY;IAC7C,OAAO,CAAC,QAAQ,CAAC,cAAc,CAAS;IACxC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAS;IACpC,OAAO,CAAC,QAAQ,CAAC,UAAU,CAAW;IACtC,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAS;IAC7B,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAoB;IACxC,OAAO,CAAC,QAAQ,CAAC,kBAAkB,CAAS;IAC5C,+EAA+E;IAC/E,OAAO,CAAC,UAAU,CAAW;IAE7B,OAAO,CAAC,KAAK,CAAmC;IAChD,OAAO,CAAC,KAAK,CAA0B;IACvC,OAAO,CAAC,SAAS,CAAqC;IACtD,OAAO,CAAC,OAAO,CAAkG;IACjH,OAAO,CAAC,OAAO,CAAkC;IACjD,OAAO,CAAC,QAAQ,CAAwB;IACxC,OAAO,CAAC,SAAS,CAAoB;IACrC,OAAO,CAAC,QAAQ,CAAS;IACzB,kFAAkF;IAClF,OAAO,CAAC,QAAQ,CAAC,CAAgB;IACjC,gGAAgG;IAChG,OAAO,CAAC,WAAW,CAA0D;IAC7E,qEAAqE;IACrE,OAAO,CAAC,UAAU,CAA8E;IAChG,8CAA8C;IAC9C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAW;IACrC,kFAAkF;IAClF,OAAO,CAAC,QAAQ,CAAC,eAAe,CAAwB;IAExD,YAAY,OAAO,EAAE,mBAAmB,EAqBvC;IAED,qEAAqE;IACrE,gBAAgB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,IAAI,CAEtC;IAED;;;;;OAKG;IACH,eAAe,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,CAEnC;IAED,0DAA0D;IAC1D,OAAO,CAAC,WAAW;IAOnB,gDAAgD;IAChD,OAAO,CAAC,UAAU;IAMlB,qDAAqD;IACrD,KAAK,CAAC,IAAI,EAAE,gBAAgB,GAAG,IAAI,CAoBlC;IAED,gCAAgC;IAChC,UAAU,CACT,OAAO,EAAE,MAAM,GACb,SAAS,GAAG,QAAQ,GAAG,MAAM,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,GAAG,WAAW,GAAG,SAAS,CAc5F;IAED;;;;;;OAMG;IACH,MAAM,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CA0B/B;IAED,yDAAyD;IACzD,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,cAAc,CAAC,CAcjD;IAED,6CAA6C;IAC7C,aAAa,IAAI,MAAM,CAEtB;IAED,4CAA4C;IAC5C,YAAY,IAAI,MAAM,CAErB;IAED;;;;;;;;OAQG;IACG,QAAQ,CAAC,IAAI,EAAE,MAAM,EAAE,OAAO,GAAE,eAAoB,GAAG,OAAO,CAAC,UAAU,CAAC,CAiB/E;IAED;;;OAGG;IACH,gBAAgB,CACf,IAAI,EAAE,MAAM,EACZ,OAAO,GAAE,eAAoB,GAC3B;QAAE,cAAc,EAAE,OAAO,CAAC;QAAC,OAAO,CAAC,EAAE,MAAM,CAAC;QAAC,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CASrF;IAED;;;OAGG;IACH,OAAO,CAAC,aAAa;IA6CrB;;;;OAIG;IACH,OAAO,CAAC,OAAO,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAEnD;IAED,8DAA8D;IAC9D,cAAc,CAAC,OAAO,EAAE,MAAM,EAAE,GAAG,GAAE,MAAiB,GAAG,MAAM,CAE9D;IAED;;;;OAIG;IACG,MAAM,CACX,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,MAAM,EACd,OAAO,GAAE,IAAI,CAAC,eAAe,EAAE,YAAY,GAAG,aAAa,CAAM,GAC/D,OAAO,CAAC,UAAU,CAAC,CAUrB;IAED,gFAAgF;IAChF,OAAO,CAAC,qBAAqB;IAW7B,OAAO,CAAC,gBAAgB;IA0BxB,OAAO,CAAC,eAAe;IAsBvB;;;OAGG;IACH,OAAO,CAAC,kBAAkB;IAQ1B,+EAA+E;IAC/E,OAAO,IAAI,IAAI,CA8Bd;IAED,2DAA2D;IAC3D,OAAO,CAAC,IAAI;IAOZ,sCAAsC;IACtC,OAAO,CAAC,SAAS;IAsGjB;;;;;;;;;OASG;IACH,OAAO,CAAC,aAAa;IAkBrB,kEAAkE;IAClE,OAAO,CAAC,SAAS;IA4SjB,kFAAkF;IAClF,OAAO,CAAC,6BAA6B;IAiBrC,oFAAoF;IACpF,OAAO,CAAC,6BAA6B;IAQrC,yEAAyE;IACzE,OAAO,CAAC,qBAAqB;IAa7B;;;;OAIG;IACH,OAAO,CAAC,mBAAmB;IAiB3B,OAAO,CAAC,iBAAiB;IAWzB,OAAO,CAAC,aAAa;CAgBrB","sourcesContent":["import { spawn } from \"node:child_process\";\nimport { EventEmitter } from \"node:events\";\nimport { existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from \"node:fs\";\nimport { dirname, join } from \"node:path\";\nimport type { Api, Model } from \"@kolisachint/hoocode-ai\";\nimport { getDispatchTaskDir } from \"../config.js\";\nimport { attachJsonlLineReader } from \"../modes/rpc/jsonl.js\";\nimport { writeFileAtomicSync } from \"../utils/atomic-file.js\";\nimport { waitForChildProcess } from \"../utils/child-process.js\";\nimport { killProcessTree } from \"../utils/shell.js\";\nimport { MODEL_INHERIT } from \"./agent-frontmatter.js\";\nimport { agentLog } from \"./agent-log.js\";\nimport { type AgentRegistry, loadAgentRegistry } from \"./agent-registry.js\";\nimport { DispatchEvaluator } from \"./dispatch-evaluator.js\";\nimport { SubagentLifeguard } from \"./lifeguard.js\";\nimport { resolveModelReference } from \"./model-categories.js\";\nimport { OutputVerifier } from \"./output-verifier.js\";\nimport type { Settings } from \"./settings-manager.js\";\nimport {\n\tcurrentSubagentDepth,\n\tDEFER_MCP_SCHEMAS_ENV,\n\tresolveMaxSubagentDepth,\n\tSUBAGENT_DEPTH_ENV,\n\tSUBAGENT_SKIP_MCP_ENV,\n\ttoolAllowlistNeedsMcp,\n} from \"./subagent-depth.js\";\nimport { SUBAGENT_PROGRESS_EVENTS } from \"./subagent-events.js\";\nimport { TokenBudget } from \"./token-budget.js\";\n\n/**\n * Provider/model failure signatures where inheriting the parent model can\n * recover. Compiled once at module load instead of on every error check.\n */\nconst INHERITED_MODEL_FALLBACK_ERROR =\n\t/usage[_\\s-]?limit|subscription|quota|rate.?limit|too many requests|429|insufficient|out of credit|credit balance|billing|payment required|402|model[^\\n]*(not found|unavailable|not available|not supported|does not exist|invalid|unsupported)|no api key|no auth configured|authentication|unauthorized|forbidden|permission/i;\n\nexport interface SubagentPoolTask {\n\ttask_id: string;\n\tagent_type: string;\n\ttask: string;\n\tcontext?: string;\n\ttoken_budget?: number;\n\tcwd?: string;\n\tmodel?: string;\n\tprovider?: string;\n\t/**\n\t * Explicit session file for the child to persist/continue. When omitted the\n\t * child uses its own dispatch dir (`<dispatch>/<task_id>/session.jsonl`).\n\t * Resume reuses the original task's session file to continue the transcript.\n\t */\n\tsessionFile?: string;\n\t/** Internal: retry using the caller's model when a built-in agent's preferred model fails. */\n\tuseInheritedModelFallback?: boolean;\n}\n\nexport interface SubagentSlot {\n\tpid: number;\n\tagent_type: string;\n\ttask_id: string;\n\tspawned_at: number;\n\ttoken_budget: number;\n\tprocess: ReturnType<typeof spawn>;\n}\n\nexport interface SubagentResult {\n\ttask_id: string;\n\tok: boolean;\n\tstdout: string;\n\tstderr: string;\n\texit_code: number | null;\n\terror?: string;\n\t/** True when the task exceeded its token budget and was hard-stopped. */\n\tbudget_exceeded?: boolean;\n\t/** Terminal status derived from how the task finished. */\n\tstatus?: \"complete\" | \"partial\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\";\n\t/** Parsed result.json content when available (e.g. on partial completion). */\n\tresult_data?: Record<string, unknown>;\n\t/** True when this run used the inherited-model fallback (preferred model failed first). */\n\tusedInheritedModelFallback?: boolean;\n}\n\nexport interface TaskResult {\n\t/** True when the evaluator decided the task is simple enough for inline handling. */\n\thandled_inline: boolean;\n\t/** Present when the task was delegated. */\n\ttask_id?: string;\n\tagent_type?: string;\n\treason?: string;\n\t/** Subagent result when delegated. */\n\tresult?: SubagentResult;\n\t/** Duration in milliseconds when delegated. */\n\tduration?: number;\n}\n\nexport interface DispatchOptions {\n\t/** Skip evaluation and force this agent type (user/explicit override).\n\t *  Accepts any registry-defined agent name, not just the built-in modes. */\n\tforceAgent?: string;\n\t/** Context distilled from the calling agent, passed to the subagent. */\n\tcontext?: string;\n\t/** Model id for the subagent (defaults to the child's configured default). */\n\tmodel?: string;\n\t/** Provider for the subagent. */\n\tprovider?: string;\n\t/** Explicit session file to persist/continue (used by resume). */\n\tsessionFile?: string;\n\t/** Caller-supplied task id. Defaults to a generated `dispatch-…` id. Lets a\n\t *  caller register liveness/inbox state under the id before dispatch resolves. */\n\ttaskId?: string;\n}\n\nexport interface SubagentPoolOptions {\n\t/** Path to the hoocode executable (or the runtime, e.g. node, when prefixArgs is set). */\n\texecutable: string;\n\t/** Args inserted before task args (e.g. the CLI entry script for node/tsx). */\n\tprefixArgs?: string[];\n\t/** Maximum concurrent child processes. Defaults to 5. */\n\tmaxConcurrency?: number;\n\t/** Working directory for spawned processes. Defaults to process.cwd(). */\n\tcwd?: string;\n\t/** Environment variables. Defaults to process.env. */\n\tenv?: NodeJS.ProcessEnv;\n\t/** Default token budget per task. Defaults to 0. */\n\tdefaultTokenBudget?: number;\n\t/**\n\t * Non-default skill paths to forward to every spawned subagent via --skill.\n\t * Subagents auto-discover skills from standard locations; only paths that\n\t * won't be found by default discovery need to be forwarded here.\n\t */\n\tskillPaths?: string[];\n\t/** Settings for model category resolution. */\n\tsettings?: Settings;\n\t/**\n\t * Available/configured models used to derive default model-category mappings\n\t * when a tier is not explicitly set in `settings.modelCategories`. Snapshotted\n\t * at pool creation, mirroring how `settings` is captured.\n\t */\n\tavailableModels?: readonly Model<Api>[];\n}\n\n/**\n * Default hard cap on assistant turns for a spawned subagent when its definition\n * does not set `maxTurns`. The token budget is advisory (it warns but never\n * kills), so this turn cap is the guaranteed hard stop for every subagent.\n */\nexport const DEFAULT_SUBAGENT_MAX_TURNS = 50;\n\n/**\n * AgentSession event `type`s forwarded from a subagent's json event stream as\n * `task_progress` events. Deliberately coarse; the child now also filters its\n * stdout to this set plus `message_end` at the source (see\n * SUBAGENT_STDOUT_EVENT_TYPES), so the per-delta firehose no longer crosses the\n * pipe. Re-exported from the shared module so the child emitter and this\n * consumer stay in lockstep.\n */\nexport const FORWARDED_SUBAGENT_EVENTS = SUBAGENT_PROGRESS_EVENTS;\n\n/**\n * Cap on the captured stdout/stderr text kept per task (tail-truncated). The\n * capture exists for diagnostics (failure reasons, output.json), so keeping the\n * most recent tail is enough; without a cap a chatty child could grow the\n * parent's memory without bound across a long swarm.\n */\nconst MAX_CAPTURED_STREAM_CHARS = 256 * 1024;\n\n/**\n * Cap on a single un-terminated JSONL line buffered from a child's stdout. The\n * forwarded events are small; anything approaching this is a runaway writer,\n * and the reader drops the line rather than buffering toward OOM.\n */\nconst MAX_SUBAGENT_EVENT_LINE_CHARS = 8 * 1024 * 1024;\n\n/** Append a chunk to a capped capture buffer, keeping the most recent tail. */\nfunction appendTail(current: string, chunk: string, cap: number = MAX_CAPTURED_STREAM_CHARS): string {\n\tconst next = current + chunk;\n\treturn next.length > cap ? next.slice(next.length - cap) : next;\n}\n\n/** The action the pool should take for one JSONL line from a subagent's stdout. */\nexport type SubagentStdoutLine =\n\t| { kind: \"heartbeat\" }\n\t| { kind: \"progress\"; event: Record<string, unknown> }\n\t| { kind: \"ignore\" };\n\n/**\n * Classify one JSONL line from a subagent's stdout into the action to take.\n * Pure (no side effects) so the ping/forward/drop policy is unit-testable without\n * spawning a child. Line framing — UTF-8-safe reassembly of chunks split mid-line\n * — is handled upstream by attachJsonlLineReader; this only sees complete lines.\n */\nexport function classifySubagentLine(line: string): SubagentStdoutLine {\n\tconst trimmed = line.trim();\n\tif (!trimmed.startsWith(\"{\")) return { kind: \"ignore\" };\n\tlet parsed: Record<string, unknown>;\n\ttry {\n\t\tparsed = JSON.parse(trimmed) as Record<string, unknown>;\n\t} catch {\n\t\treturn { kind: \"ignore\" };\n\t}\n\tif (parsed.ping === true) return { kind: \"heartbeat\" };\n\tif (typeof parsed.type === \"string\" && FORWARDED_SUBAGENT_EVENTS.has(parsed.type)) {\n\t\treturn { kind: \"progress\", event: parsed };\n\t}\n\treturn { kind: \"ignore\" };\n}\n\n/**\n * Pool for running hoocode subagents as child processes with bounded concurrency,\n * FIFO queuing with priority support, and automatic slot refill.\n *\n * Events:\n * - \"task_done\"    – task completed successfully and output was verified\n * - \"task_failed\"  – task failed (spawn error, bad exit code, verification failure)\n * - \"task_stalled\" – heartbeat missed past the load-scaled threshold (60s base,\n *                    widened under concurrency/event-loop lag), process SIGKILLed\n * - \"task_timeout\" – hard timeout exceeded, process was SIGKILLed\n * - \"task_cancelled\" – user-initiated cancel (see cancel()); process tree killed\n * - \"budget_warning\" – token usage crossed 80% threshold (advisory)\n * - \"budget_exceeded\" – token usage crossed 100% threshold (advisory; never kills)\n * - \"task_progress\" – coarse lifecycle event (turn_end, tool start/end) parsed\n *                    from the child's json event stream, for live UI updates\n */\nexport class SubagentPool extends EventEmitter {\n\tprivate readonly maxConcurrency: number;\n\tprivate readonly executable: string;\n\tprivate readonly prefixArgs: string[];\n\tprivate readonly cwd: string;\n\tprivate readonly env: NodeJS.ProcessEnv;\n\tprivate readonly defaultTokenBudget: number;\n\t/** Non-default skill paths forwarded to every spawned subagent via --skill. */\n\tprivate skillPaths: string[];\n\n\tprivate slots = new Map<string, SubagentSlot>();\n\tprivate queue: SubagentPoolTask[] = [];\n\tprivate completed = new Map<string, SubagentResult>();\n\tprivate waiters = new Map<string, { resolve: (result: SubagentResult) => void; reject: (err: Error) => void }>();\n\tprivate budgets = new Map<string, TokenBudget>();\n\tprivate verifier = new OutputVerifier();\n\tprivate lifeguard: SubagentLifeguard;\n\tprivate disposed = false;\n\t/** Lazily-loaded agent registry (frontmatter definitions) for this pool's cwd. */\n\tprivate registry?: AgentRegistry;\n\t/** Tracks why a task was killed (stalled / timeout / user cancel) before exit handler fires. */\n\tprivate killReasons = new Map<string, \"stalled\" | \"timeout\" | \"cancelled\">();\n\t/** Persistent terminal status map, survives wait_for consumption. */\n\tprivate taskStatus = new Map<string, \"done\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\">();\n\t/** Settings for model category resolution. */\n\tprivate readonly settings?: Settings;\n\t/** Available models used to derive default model-category mappings (snapshot). */\n\tprivate readonly availableModels: readonly Model<Api>[];\n\n\tconstructor(options: SubagentPoolOptions) {\n\t\tsuper();\n\t\tthis.maxConcurrency = options.maxConcurrency ?? 5;\n\t\tthis.executable = options.executable;\n\t\tthis.prefixArgs = options.prefixArgs ?? [];\n\t\tthis.cwd = options.cwd ?? process.cwd();\n\t\tthis.env = options.env ?? process.env;\n\t\tthis.defaultTokenBudget = options.defaultTokenBudget ?? 0;\n\t\tthis.skillPaths = options.skillPaths ? [...options.skillPaths] : [];\n\t\tthis.settings = options.settings;\n\t\tthis.availableModels = options.availableModels ?? [];\n\t\tthis.verifier = new OutputVerifier(this.cwd);\n\t\tthis.lifeguard = new SubagentLifeguard(this.cwd);\n\t\tthis.lifeguard.on(\"stalled\", (data: { task_id: string; pid: number }) => {\n\t\t\tthis.killReasons.set(data.task_id, \"stalled\");\n\t\t\tthis.emit(\"task_stalled\", data);\n\t\t});\n\t\tthis.lifeguard.on(\"timeout\", (data: { task_id: string; pid: number }) => {\n\t\t\tthis.killReasons.set(data.task_id, \"timeout\");\n\t\t\tthis.emit(\"task_timeout\", data);\n\t\t});\n\t}\n\n\t/** Update the non-default skill paths forwarded to new subagents. */\n\tupdateSkillPaths(paths: string[]): void {\n\t\tthis.skillPaths = [...paths];\n\t}\n\n\t/**\n\t * Report external in-process load (e.g. the number of background MCP tools\n\t * currently executing in the parent) to the lifeguard. This widens its\n\t * heartbeat/timeout tolerance so monitored subagents aren't false-positive\n\t * reaped when the parent's event loop is busy with concurrent background work.\n\t */\n\tsetExternalLoad(count: number): void {\n\t\tthis.lifeguard.setExternalLoad(count);\n\t}\n\n\t/** Lazily load the agent registry for this pool's cwd. */\n\tprivate getRegistry(): AgentRegistry {\n\t\tif (!this.registry) {\n\t\t\tthis.registry = loadAgentRegistry({ cwd: this.cwd });\n\t\t}\n\t\treturn this.registry;\n\t}\n\n\t/** Priority value: higher numbers run first. */\n\tprivate priorityOf(agent_type: string): number {\n\t\t// Read-only investigation (explore/plan) often unblocks downstream work, so\n\t\t// it runs ahead of other agents.\n\t\treturn agent_type === \"explore\" || agent_type === \"plan\" ? 2 : 1;\n\t}\n\n\t/** Queue a task. It will run when a slot is free. */\n\tspawn(task: SubagentPoolTask): void {\n\t\tif (this.disposed) {\n\t\t\tthrow new Error(\"SubagentPool has been disposed\");\n\t\t}\n\t\tif (\n\t\t\tthis.slots.has(task.task_id) ||\n\t\t\tthis.queue.some((t) => t.task_id === task.task_id) ||\n\t\t\tthis.completed.has(task.task_id)\n\t\t) {\n\t\t\tthrow new Error(`Duplicate task_id: ${task.task_id}`);\n\t\t}\n\n\t\tconst p = this.priorityOf(task.agent_type);\n\t\tconst idx = this.queue.findIndex((t) => this.priorityOf(t.agent_type) < p);\n\t\tif (idx === -1) {\n\t\t\tthis.queue.push(task);\n\t\t} else {\n\t\t\tthis.queue.splice(idx, 0, task);\n\t\t}\n\t\tthis.pull();\n\t}\n\n\t/** Current status of a task. */\n\tget_status(\n\t\ttask_id: string,\n\t): \"running\" | \"queued\" | \"done\" | \"failed\" | \"stalled\" | \"timeout\" | \"cancelled\" | \"unknown\" {\n\t\tif (this.slots.has(task_id)) return \"running\";\n\t\tif (this.queue.some((t) => t.task_id === task_id)) return \"queued\";\n\t\tconst persisted = this.taskStatus.get(task_id);\n\t\tif (persisted) return persisted;\n\t\tconst result = this.completed.get(task_id);\n\t\tif (result) {\n\t\t\tif (result.status === \"stalled\") return \"stalled\";\n\t\t\tif (result.status === \"timeout\") return \"timeout\";\n\t\t\tif (result.status === \"cancelled\") return \"cancelled\";\n\t\t\tif (result.ok) return \"done\";\n\t\t\treturn \"failed\";\n\t\t}\n\t\treturn \"unknown\";\n\t}\n\n\t/**\n\t * Cancel a task on the user's behalf (Esc/abort mid-turn). A queued task is\n\t * removed and settles immediately; a running task's whole process tree is\n\t * killed and settles with status \"cancelled\" when its exit is observed\n\t * (unless it already wrote a valid result.json — completed work is honored).\n\t * Returns false for unknown/settled task ids.\n\t */\n\tcancel(task_id: string): boolean {\n\t\tconst queued = this.queue.findIndex((t) => t.task_id === task_id);\n\t\tif (queued !== -1) {\n\t\t\tthis.queue.splice(queued, 1);\n\t\t\tconst result: SubagentResult = {\n\t\t\t\ttask_id,\n\t\t\t\tok: false,\n\t\t\t\tstdout: \"\",\n\t\t\t\tstderr: \"\",\n\t\t\t\texit_code: null,\n\t\t\t\terror: \"cancelled before start\",\n\t\t\t\tstatus: \"cancelled\",\n\t\t\t};\n\t\t\tthis.emit(\"task_cancelled\", { task_id });\n\t\t\tthis.resolveWaiter(task_id, result);\n\t\t\treturn true;\n\t\t}\n\t\tconst slot = this.slots.get(task_id);\n\t\tif (!slot) return false;\n\t\tthis.killReasons.set(task_id, \"cancelled\");\n\t\tif (slot.pid > 0) {\n\t\t\tkillProcessTree(slot.pid);\n\t\t} else if (!slot.process.killed) {\n\t\t\tslot.process.kill(\"SIGKILL\");\n\t\t}\n\t\treturn true;\n\t}\n\n\t/** Wait for a task to complete and return its result. */\n\twait_for(task_id: string): Promise<SubagentResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\n\t\tconst existing = this.completed.get(task_id);\n\t\tif (existing) {\n\t\t\tthis.completed.delete(task_id);\n\t\t\treturn Promise.resolve(existing);\n\t\t}\n\n\t\treturn new Promise((resolve, reject) => {\n\t\t\tthis.waiters.set(task_id, { resolve, reject });\n\t\t});\n\t}\n\n\t/** Number of currently running subagents. */\n\trunning_count(): number {\n\t\treturn this.slots.size;\n\t}\n\n\t/** Number of tasks waiting in the queue. */\n\tqueued_count(): number {\n\t\treturn this.queue.length;\n\t}\n\n\t/**\n\t * Dispatch a task through the evaluator.\n\t *\n\t * - If `options.forceAgent` is provided, skip evaluation and spawn directly.\n\t * - Otherwise evaluate the task. If it should be handled inline, return\n\t *   `{ handled_inline: true }` immediately.\n\t * - If delegating, spawn the subagent, wait for completion, write\n\t *   `output.json`, and return the result.\n\t */\n\tasync dispatch(task: string, options: DispatchOptions = {}): Promise<TaskResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\t\tconst begin = this.beginDispatch(task, options);\n\t\tif (begin.handled_inline) {\n\t\t\treturn { handled_inline: true, reason: begin.reason };\n\t\t}\n\t\tconst result = await this.wait_for(begin.task_id);\n\t\treturn {\n\t\t\thandled_inline: false,\n\t\t\ttask_id: begin.task_id,\n\t\t\tagent_type: begin.agent_type,\n\t\t\treason: begin.reason,\n\t\t\tresult,\n\t\t\tduration: Date.now() - begin.startTime,\n\t\t};\n\t}\n\n\t/**\n\t * Fire-and-forget dispatch for background agents. Spawns the subagent and\n\t * returns its handle immediately; the caller polls get_status()/collect().\n\t */\n\tdispatchDetached(\n\t\ttask: string,\n\t\toptions: DispatchOptions = {},\n\t): { handled_inline: boolean; task_id?: string; agent_type?: string; reason?: string } {\n\t\tif (this.disposed) {\n\t\t\tthrow new Error(\"SubagentPool has been disposed\");\n\t\t}\n\t\tconst begin = this.beginDispatch(task, options);\n\t\tif (begin.handled_inline) {\n\t\t\treturn { handled_inline: true, reason: begin.reason };\n\t\t}\n\t\treturn { handled_inline: false, task_id: begin.task_id, agent_type: begin.agent_type, reason: begin.reason };\n\t}\n\n\t/**\n\t * Evaluate, log, and spawn a task without waiting. Shared by dispatch()\n\t * (blocking) and dispatchDetached() (background).\n\t */\n\tprivate beginDispatch(\n\t\ttask: string,\n\t\toptions: DispatchOptions,\n\t):\n\t\t| { handled_inline: true; reason?: string }\n\t\t| { handled_inline: false; task_id: string; agent_type: string; reason?: string; startTime: number } {\n\t\tconst { forceAgent, context, model, provider, sessionFile } = options;\n\t\tconst evaluator = new DispatchEvaluator();\n\t\tconst analysis = evaluator.evaluate(task);\n\n\t\tif (!forceAgent && !analysis.should_delegate) {\n\t\t\treturn { handled_inline: true, reason: analysis.reason };\n\t\t}\n\n\t\tconst agent_type = forceAgent ?? \"general-purpose\";\n\t\tconst task_id = options.taskId ?? `dispatch-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;\n\t\tconst reason = forceAgent ? \"user_override\" : analysis.reason;\n\t\tconst complexity = analysis.estimated_complexity;\n\t\t// Depth of the child about to be spawned (this process's depth + 1). Surfaced\n\t\t// so a delegation tree's nesting is visible in logs without extra tooling.\n\t\tconst childDepth = currentSubagentDepth(this.env) + 1;\n\n\t\t// Pre-dispatch logging. Goes through agentLog rather than console.error:\n\t\t// stdout is reserved for the JSON event stream, and stderr is not free\n\t\t// either while the interactive TUI owns the terminal (see agent-log.ts).\n\t\tagentLog(\n\t\t\t`[DISPATCH] agent=${agent_type} depth=${childDepth} reason=${reason} complexity=${complexity} task_id=${task_id}`,\n\t\t);\n\t\tthis.writeDispatchLog(task_id, agent_type, reason, complexity, task, childDepth);\n\n\t\tconst poolTask: SubagentPoolTask = {\n\t\t\ttask_id,\n\t\t\tagent_type,\n\t\t\ttask,\n\t\t\tcontext,\n\t\t\tmodel,\n\t\t\tprovider,\n\t\t\tsessionFile,\n\t\t\tcwd: this.cwd,\n\t\t};\n\t\tconst startTime = Date.now();\n\t\tthis.spawn(poolTask);\n\t\treturn { handled_inline: false, task_id, agent_type, reason, startTime };\n\t}\n\n\t/**\n\t * Non-destructively read a completed task's result (for background polling).\n\t * Returns undefined while the task is still running/queued, or if its result\n\t * was already consumed via wait_for().\n\t */\n\tcollect(task_id: string): SubagentResult | undefined {\n\t\treturn this.completed.get(task_id);\n\t}\n\n\t/** Absolute path of the persisted session file for a task. */\n\tgetSessionFile(task_id: string, cwd: string = this.cwd): string {\n\t\treturn join(getDispatchTaskDir(cwd, task_id), \"session.jsonl\");\n\t}\n\n\t/**\n\t * Resume a previously dispatched subagent, continuing its persisted session\n\t * with a follow-up prompt. Recovers the original agent type from its dispatch\n\t * log. Rejects if no resumable session exists for the task.\n\t */\n\tasync resume(\n\t\ttask_id: string,\n\t\tprompt: string,\n\t\toptions: Omit<DispatchOptions, \"forceAgent\" | \"sessionFile\"> = {},\n\t): Promise<TaskResult> {\n\t\tif (this.disposed) {\n\t\t\treturn Promise.reject(new Error(\"SubagentPool has been disposed\"));\n\t\t}\n\t\tconst sessionFile = this.getSessionFile(task_id);\n\t\tif (!existsSync(sessionFile)) {\n\t\t\treturn Promise.reject(new Error(`No resumable session for task \"${task_id}\" (expected ${sessionFile}).`));\n\t\t}\n\t\tconst agent_type = this.readDispatchAgentType(task_id) ?? \"general-purpose\";\n\t\treturn this.dispatch(prompt, { ...options, forceAgent: agent_type, sessionFile });\n\t}\n\n\t/** Recover the agent type a task was dispatched with, from its dispatch log. */\n\tprivate readDispatchAgentType(task_id: string): string | undefined {\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"dispatch-log.json\");\n\t\tif (!existsSync(path)) return undefined;\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(readFileSync(path, \"utf-8\")) as { agent_type?: string };\n\t\t\treturn typeof parsed.agent_type === \"string\" ? parsed.agent_type : undefined;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate writeDispatchLog(\n\t\ttask_id: string,\n\t\tagent_type: string,\n\t\treason: string,\n\t\tcomplexity: string,\n\t\ttask: string,\n\t\tdepth: number,\n\t): void {\n\t\tconst log = {\n\t\t\ttimestamp: new Date().toISOString(),\n\t\t\ttask_id,\n\t\t\tagent_type,\n\t\t\tdepth,\n\t\t\treason,\n\t\t\tcomplexity,\n\t\t\ttask,\n\t\t};\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"dispatch-log.json\");\n\t\ttry {\n\t\t\tmkdirSync(dirname(path), { recursive: true });\n\t\t\twriteFileSync(path, JSON.stringify(log, null, 2));\n\t\t} catch {\n\t\t\t// Best-effort persistence\n\t\t}\n\t}\n\n\tprivate writeOutputJson(task_id: string, result: SubagentResult): void {\n\t\tconst output = {\n\t\t\ttask_id: result.task_id,\n\t\t\tok: result.ok,\n\t\t\texit_code: result.exit_code,\n\t\t\tstatus: result.status,\n\t\t\tstdout: result.stdout,\n\t\t\tstderr: result.stderr,\n\t\t\terror: result.error,\n\t\t\tbudget_exceeded: result.budget_exceeded,\n\t\t\tresult_data: result.result_data,\n\t\t};\n\t\tconst path = join(getDispatchTaskDir(this.cwd, task_id), \"output.json\");\n\t\ttry {\n\t\t\t// Atomic like result.json: background pollers may read output.json while\n\t\t\t// it is being (re)written.\n\t\t\twriteFileAtomicSync(path, JSON.stringify(output, null, 2));\n\t\t} catch {\n\t\t\t// Best-effort persistence\n\t\t}\n\t}\n\n\t/**\n\t * Remove a task's dispatch dir after a clean, verified success. Best-effort:\n\t * a cleanup failure must never fail an otherwise successful task.\n\t */\n\tprivate cleanupDispatchDir(task_id: string, cwd: string): void {\n\t\ttry {\n\t\t\trmSync(getDispatchTaskDir(cwd, task_id), { recursive: true, force: true });\n\t\t} catch {\n\t\t\t// Best-effort cleanup\n\t\t}\n\t}\n\n\t/** Kill all running processes, clear the queue, and reject pending waiters. */\n\tdispose(): void {\n\t\tif (this.disposed) return;\n\t\tthis.disposed = true;\n\n\t\tfor (const slot of this.slots.values()) {\n\t\t\t// Kill the whole process group/tree, not just the direct child: a\n\t\t\t// subagent's own children (bash commands, nested subagents) must not\n\t\t\t// outlive the pool.\n\t\t\tif (slot.pid > 0) {\n\t\t\t\tkillProcessTree(slot.pid);\n\t\t\t} else if (!slot.process.killed) {\n\t\t\t\tslot.process.kill(\"SIGTERM\");\n\t\t\t}\n\t\t}\n\t\tthis.slots.clear();\n\t\tthis.queue = [];\n\n\t\tfor (const [task_id, waiter] of this.waiters) {\n\t\t\twaiter.reject(new Error(\"SubagentPool disposed\"));\n\t\t\tthis.waiters.delete(task_id);\n\t\t}\n\t\tthis.completed.clear();\n\t\tfor (const budget of this.budgets.values()) {\n\t\t\tbudget.removeAllListeners();\n\t\t}\n\t\tthis.budgets.clear();\n\t\tthis.killReasons.clear();\n\t\tthis.taskStatus.clear();\n\t\tthis.lifeguard.dispose();\n\t\tthis.removeAllListeners();\n\t}\n\n\t/** Pull tasks from the queue while slots are available. */\n\tprivate pull(): void {\n\t\twhile (this.slots.size < this.maxConcurrency && this.queue.length > 0) {\n\t\t\tconst task = this.queue.shift()!;\n\t\t\tthis.startTask(task, false);\n\t\t}\n\t}\n\n\t/** Build CLI arguments for a task. */\n\tprivate buildArgs(task: SubagentPoolTask): string[] {\n\t\t// Persist the child's session so a finished/interrupted subagent can be\n\t\t// resumed later (see resume()). SessionManager.open() creates the file on\n\t\t// first run and continues it on subsequent runs.\n\t\tconst sessionFile = task.sessionFile ?? this.getSessionFile(task.task_id, task.cwd ?? this.cwd);\n\t\tconst args: string[] = [\n\t\t\t...this.prefixArgs,\n\t\t\t\"--mode\",\n\t\t\t\"json\",\n\t\t\t\"--session\",\n\t\t\tsessionFile,\n\t\t\t\"--task-id\",\n\t\t\ttask.task_id,\n\t\t];\n\n\t\t// Prefer the data-driven agent definition from the registry; fall back to the\n\t\t// built-in mode prompt/allowlist for legacy modes not present in the registry.\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\n\t\tif (task.agent_type) {\n\t\t\tconst systemPrompt = def?.prompt;\n\t\t\tif (systemPrompt) {\n\t\t\t\targs.push(\"--system-prompt\", systemPrompt);\n\t\t\t}\n\n\t\t\t// A `delegate: true` agent may itself dispatch via the Task tool, but only\n\t\t\t// while the child it becomes can still nest (childDepth < cap) — so the\n\t\t\t// deepest permitted level cannot delegate further. Gating here keeps the\n\t\t\t// authorization explicit and bounded by the same cap as the depth guard.\n\t\t\tconst childDepth = currentSubagentDepth(this.env) + 1;\n\t\t\tconst canChildDelegate = def?.delegate === true && childDepth < resolveMaxSubagentDepth(undefined, this.env);\n\n\t\t\t// Tool allowlist comes from the agent definition's frontmatter `tools`\n\t\t\t// field (read-only built-ins declare their own sandbox). When omitted, no\n\t\t\t// --tools is passed and the subagent inherits all parent tools (so the Task\n\t\t\t// tool already survives). A delegating agent with an explicit allowlist must\n\t\t\t// have Task/TaskOutput added, or the child would filter them out.\n\t\t\tconst tools = def?.tools ? [...def.tools] : undefined;\n\t\t\tif (canChildDelegate && tools) {\n\t\t\t\tfor (const t of [\"Task\", \"TaskOutput\"]) {\n\t\t\t\t\tif (!tools.includes(t)) tools.push(t);\n\t\t\t\t}\n\t\t\t}\n\t\t\tif (tools && tools.length > 0) {\n\t\t\t\targs.push(\"--tools\", tools.join(\",\"));\n\t\t\t}\n\t\t\tif (def?.disallowedTools && def.disallowedTools.length > 0) {\n\t\t\t\targs.push(\"--disallowed-tools\", def.disallowedTools.join(\",\"));\n\t\t\t}\n\n\t\t\t// Propagate subagent enablement so the child registers the Task tool; without\n\t\t\t// this the flag-based enablement would not reach a spawned child.\n\t\t\tif (canChildDelegate) {\n\t\t\t\targs.push(\"--enable-subagents\");\n\t\t\t\t// Scoped delegation: restrict which agent types this child may spawn.\n\t\t\t\tif (def?.delegateTo && def.delegateTo.length > 0) {\n\t\t\t\t\targs.push(\"--delegate-allow\", def.delegateTo.join(\",\"));\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Model precedence: a definition's explicit model wins (unless it is the\n\t\t// `inherit` sentinel), otherwise use the caller-provided model. Built-in\n\t\t// agents can retry with the inherited model when their preferred model is\n\t\t// unavailable or quota-limited.\n\t\tconst explicitModel =\n\t\t\t!task.useInheritedModelFallback && def?.model && def.model !== MODEL_INHERIT ? def.model : undefined;\n\t\t// Resolve a model-category reference (fast/standard/capable) to its\n\t\t// configured or derived model id. An unresolvable category yields undefined,\n\t\t// so no `--model` is passed and the child keeps its default model.\n\t\tconst rawModel = explicitModel ?? task.model;\n\t\tconst modelToUse = rawModel ? resolveModelReference(rawModel, this.settings, this.availableModels) : undefined;\n\t\tif (modelToUse) {\n\t\t\targs.push(\"--model\", modelToUse);\n\t\t}\n\t\t// Only pass --provider when the model doesn't already encode the provider\n\t\t// (e.g. \"anthropic/claude-sonnet-4-5\"). When both --model and --provider are\n\t\t// present, --provider wins and filters candidates to that provider, which\n\t\t// breaks models whose IDs contain a provider prefix from a different provider.\n\t\tif (task.provider && (!modelToUse || !modelToUse.includes(\"/\"))) {\n\t\t\targs.push(\"--provider\", task.provider);\n\t\t}\n\n\t\t// Always give subagents a hard turn cap. With the token budget now advisory\n\t\t// (warn-only), this is the guaranteed hard stop for a runaway subagent.\n\t\tconst maxTurns = def?.maxTurns && def.maxTurns > 0 ? def.maxTurns : DEFAULT_SUBAGENT_MAX_TURNS;\n\t\targs.push(\"--max-turns\", String(maxTurns));\n\n\t\t// Forward non-default skill paths so the subagent has access to all parent skills.\n\t\t// Standard discovery locations (~/.hoocode/, .hoocode/, .claude/) are found automatically.\n\t\tfor (const skillPath of this.skillPaths) {\n\t\t\targs.push(\"--skill\", skillPath);\n\t\t}\n\n\t\tconst prompt = task.context?.trim()\n\t\t\t? `Context from the calling agent:\\n\\n${task.context.trim()}\\n\\nTask: ${task.task.trim()}`\n\t\t\t: `Task: ${task.task.trim()}`;\n\t\targs.push(prompt);\n\n\t\treturn args;\n\t}\n\n\t/**\n\t * Environment for a spawned child.\n\t *\n\t * Stamps the child's depth (parent depth + 1) so its own guard knows where it\n\t * sits in the tree; the tree-wide cap (HOOCODE_SUBAGENT_MAX_DEPTH) is inherited\n\t * via the spread, so at the default cap of 1 the child lands at depth 1 and\n\t * cannot spawn further subagents. Also flags the child to skip MCP server\n\t * connection when its tool allowlist is explicit and MCP-free — connecting\n\t * external servers it can never call is pure boot latency.\n\t */\n\tprivate childSpawnEnv(task: SubagentPoolTask): NodeJS.ProcessEnv {\n\t\tconst env: NodeJS.ProcessEnv = {\n\t\t\t...this.env,\n\t\t\t[SUBAGENT_DEPTH_ENV]: String(currentSubagentDepth(this.env) + 1),\n\t\t};\n\t\t// Use the agent's own frontmatter allowlist (the same source buildArgs passes\n\t\t// via --tools). A delegating child also gets Task/TaskOutput appended there,\n\t\t// but neither is an MCP tool, so the decision is unchanged by that.\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\t\tif (!toolAllowlistNeedsMcp(def?.tools)) {\n\t\t\tenv[SUBAGENT_SKIP_MCP_ENV] = \"1\";\n\t\t}\n\t\t// A child never defers MCP schemas: if it needs MCP it eager-registers its\n\t\t// allowlisted tools at dispatch so they are immediately callable (spec §2).\n\t\tdelete env[DEFER_MCP_SCHEMAS_ENV];\n\t\treturn env;\n\t}\n\n\t/** Start a task in a child process, with one retry on failure. */\n\tprivate startTask(task: SubagentPoolTask, isRetry: boolean): void {\n\t\t// Get or create a TokenBudget tracker. On retry, reuse the existing one\n\t\t// so cumulative usage persists across retries.\n\t\tlet budget = this.budgets.get(task.task_id);\n\t\tif (!budget) {\n\t\t\tbudget = new TokenBudget(task.task_id, task.agent_type, {\n\t\t\t\tlimit: task.token_budget,\n\t\t\t\tcwd: task.cwd ?? this.cwd,\n\t\t\t});\n\t\t\tbudget.on(\"budget_warning\", (data: { task_id: string; message: string; used: number; limit: number }) => {\n\t\t\t\tthis.emit(\"budget_warning\", data);\n\t\t\t});\n\t\t\t// The token budget is advisory: surface telemetry but never kill. The\n\t\t\t// guaranteed hard stop is the per-subagent turn cap (--max-turns); see\n\t\t\t// DEFAULT_SUBAGENT_MAX_TURNS.\n\t\t\tbudget.on(\"budget_exceeded\", (data: { task_id: string; used: number; limit: number }) => {\n\t\t\t\tthis.emit(\"budget_exceeded\", data);\n\t\t\t});\n\t\t\tthis.budgets.set(task.task_id, budget);\n\t\t}\n\n\t\tlet proc: ReturnType<typeof spawn>;\n\t\ttry {\n\t\t\tproc = spawn(this.executable, this.buildArgs(task), {\n\t\t\t\tcwd: task.cwd ?? this.cwd,\n\t\t\t\tenv: this.childSpawnEnv(task),\n\t\t\t\tshell: false,\n\t\t\t\tstdio: [\"ignore\", \"pipe\", \"pipe\"],\n\t\t\t\t// POSIX: the child leads its own process group so a kill can reach its\n\t\t\t\t// whole tree (see lifeguard/dispose). A single-PID SIGKILL orphans any\n\t\t\t\t// grandchildren the subagent spawned (its own bash commands, nested\n\t\t\t\t// subagents), which kept burning CPU after their parent was reaped.\n\t\t\t\tdetached: process.platform !== \"win32\",\n\t\t\t});\n\t\t} catch {\n\t\t\tif (!isRetry) {\n\t\t\t\tthis.startTask(task, true);\n\t\t\t} else {\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\terror: \"Spawn failed synchronously\",\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: false,\n\t\t\t\t\tstdout: \"\",\n\t\t\t\t\tstderr: \"\",\n\t\t\t\t\texit_code: null,\n\t\t\t\t\terror: \"Spawn failed synchronously\",\n\t\t\t\t\tstatus: \"failed\",\n\t\t\t\t});\n\t\t\t\tthis.pull();\n\t\t\t}\n\t\t\treturn;\n\t\t}\n\n\t\tconst slot: SubagentSlot = {\n\t\t\tpid: proc.pid ?? 0,\n\t\t\tagent_type: task.agent_type,\n\t\t\ttask_id: task.task_id,\n\t\t\tspawned_at: Date.now(),\n\t\t\ttoken_budget: task.token_budget ?? this.defaultTokenBudget,\n\t\t\tprocess: proc,\n\t\t};\n\n\t\tthis.slots.set(task.task_id, slot);\n\t\tthis.lifeguard.monitor(task.task_id, task.agent_type, proc);\n\n\t\tlet stdout = \"\";\n\t\tlet stderr = \"\";\n\t\tlet detachStdoutReader: (() => void) | undefined;\n\n\t\tproc.stdout?.on(\"data\", (data: Buffer) => {\n\t\t\t// Capture (tail-capped) for diagnostics, and treat any output as a\n\t\t\t// heartbeat: a child busily streaming events is alive even when the\n\t\t\t// parent's starved event loop hasn't parsed its {\"ping\":true} line yet.\n\t\t\t// All *parsing* happens in the single JSONL reader below.\n\t\t\tstdout = appendTail(stdout, data.toString());\n\t\t\tthis.lifeguard.recordHeartbeat(task.task_id);\n\t\t});\n\n\t\t// The one parser on the child's stdout: UTF-8-safe, LF-only framing —\n\t\t// multi-byte characters and large events split across pipe chunks are\n\t\t// reassembled before parsing. Each complete line feeds the token budget\n\t\t// (usage telemetry) and the ping/forward/drop classifier; previously the\n\t\t// budget ran its own naive chunk-splitting parser on the same stream, which\n\t\t// corrupted lines split mid-character. The line buffer is capped so a\n\t\t// runaway writer cannot OOM the parent. Detached when the child exits (see\n\t\t// cleanup below); the reader also self-detaches on stream error/close.\n\t\tif (proc.stdout) {\n\t\t\tdetachStdoutReader = attachJsonlLineReader(\n\t\t\t\tproc.stdout,\n\t\t\t\t(line) => {\n\t\t\t\t\tbudget.processLine(line);\n\t\t\t\t\tconst action = classifySubagentLine(line);\n\t\t\t\t\tif (action.kind === \"heartbeat\") {\n\t\t\t\t\t\tthis.lifeguard.recordHeartbeat(task.task_id);\n\t\t\t\t\t} else if (action.kind === \"progress\") {\n\t\t\t\t\t\tthis.emit(\"task_progress\", {\n\t\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\t\tevent: action.event,\n\t\t\t\t\t\t});\n\t\t\t\t\t}\n\t\t\t\t},\n\t\t\t\t{ maxBuffer: MAX_SUBAGENT_EVENT_LINE_CHARS },\n\t\t\t);\n\t\t}\n\t\tproc.stderr?.on(\"data\", (data: Buffer) => {\n\t\t\tstderr = appendTail(stderr, data.toString());\n\t\t});\n\n\t\t// True when this attempt scheduled another run of the same task id (spawn\n\t\t// retry or inherited-model fallback). The retry reuses the task's TokenBudget\n\t\t// so usage accumulates across attempts; the .finally below must then leave\n\t\t// the budget (and its listeners) alone instead of tearing it down under the\n\t\t// retry's feet.\n\t\tlet retryScheduled = false;\n\n\t\twaitForChildProcess(proc)\n\t\t\t.then((code) => {\n\t\t\t\tthis.slots.delete(task.task_id);\n\t\t\t\tbudget.flush();\n\n\t\t\t\tconst killReason = this.killReasons.get(task.task_id);\n\t\t\t\tthis.killReasons.delete(task.task_id);\n\n\t\t\t\tconst duration = Date.now() - slot.spawned_at;\n\t\t\t\tconst tokens_used = budget.getUsed();\n\t\t\t\tconst budgetExceeded = budget.isExceeded();\n\n\t\t\t\t// A subagent's success is defined by a valid, verified result.json, not by\n\t\t\t\t// its exit code. A child that finished its work and wrote a valid result can\n\t\t\t\t// still be SIGKILLed by the lifeguard before it exits on its own (lingering\n\t\t\t\t// open handles delay a natural exit past the heartbeat threshold), which forces\n\t\t\t\t// exit_code === null. Keying completion off the verified result, not code === 0,\n\t\t\t\t// honors that genuine success instead of discarding it as a false stall.\n\t\t\t\tconst verification = this.verifier.verify(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\t// A well-formed result.json counts as clean completion unless its own\n\t\t\t\t// status field declares failure (e.g. \"failed\" from a provider quota\n\t\t\t\t// error). Without this check the pool would treat a child that wrote a\n\t\t\t\t// valid-but-failed result.json and exited non-zero as a success.\n\t\t\t\tlet cleanlyCompleted = code === 0 || verification.valid;\n\t\t\t\tif (cleanlyCompleted && verification.valid) {\n\t\t\t\t\tconst rd = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\t\tif (rd && (rd as Record<string, unknown>).status === \"failed\") {\n\t\t\t\t\t\tcleanlyCompleted = false;\n\t\t\t\t\t}\n\t\t\t\t}\n\n\t\t\t\t// If killed (lifeguard reap or user cancel) before producing a valid\n\t\t\t\t// result, honor the kill; a child that already wrote a verified\n\t\t\t\t// result.json completed its work and settles as a success below.\n\t\t\t\tif (killReason !== undefined && !verification.valid) {\n\t\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tok: false,\n\t\t\t\t\t\tstdout,\n\t\t\t\t\t\tstderr,\n\t\t\t\t\t\texit_code: code,\n\t\t\t\t\t\tstatus: killReason,\n\t\t\t\t\t};\n\t\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\t\tthis.emit(`task_${killReason}`, {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\tduration,\n\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t});\n\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: cleanlyCompleted,\n\t\t\t\t\tstdout,\n\t\t\t\t\tstderr,\n\t\t\t\t\texit_code: code,\n\t\t\t\t\t// Advisory telemetry only: exceeding the budget never fails the task.\n\t\t\t\t\tbudget_exceeded: budgetExceeded,\n\t\t\t\t\tstatus: cleanlyCompleted ? \"complete\" : \"failed\",\n\t\t\t\t\tusedInheritedModelFallback: task.useInheritedModelFallback === true,\n\t\t\t\t};\n\n\t\t\t\tif (result.ok) {\n\t\t\t\t\tif (!verification.valid) {\n\t\t\t\t\t\tresult.ok = false;\n\t\t\t\t\t\tresult.error = verification.reason;\n\t\t\t\t\t\tresult.status = \"failed\";\n\t\t\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\t\tduration,\n\t\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t\t\terror: verification.reason,\n\t\t\t\t\t\t});\n\t\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\t\treturn;\n\t\t\t\t\t}\n\t\t\t\t\t// Attach the verified result.json so callers can read the summary\n\t\t\t\t\t// without parsing the raw event stream.\n\t\t\t\t\tresult.result_data = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\n\t\t\t\t\t// Clean success: discard the per-task dispatch dir entirely\n\t\t\t\t\t// (session.jsonl, result.json, dispatch-log.json, budget.json). The\n\t\t\t\t\t// in-memory result already carries result_data, so callers lose\n\t\t\t\t\t// nothing. Trade-off: resume() only works for non-successful tasks.\n\t\t\t\t\tthis.cleanupDispatchDir(task.task_id, task.cwd ?? this.cwd);\n\n\t\t\t\t\tthis.emit(\"task_done\", {\n\t\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\t\tduration,\n\t\t\t\t\t\ttokens_used,\n\t\t\t\t\t\tstatus: \"complete\",\n\t\t\t\t\t});\n\t\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\n\t\t\t\t// Failure path: keep the dispatch dir for debugging and persist output.\n\t\t\t\t// Attach the child's result.json (if any) and derive a concrete failure\n\t\t\t\t// reason so callers see the real cause (e.g. a provider usage/quota\n\t\t\t\t// error) instead of a generic \"subagent failed\".\n\t\t\t\tresult.result_data = this.tryReadResultJson(task.task_id, task.cwd ?? this.cwd);\n\t\t\t\tif (!result.error) {\n\t\t\t\t\tresult.error = this.deriveFailureReason(result);\n\t\t\t\t}\n\t\t\t\tif (this.shouldRetryWithInheritedModel(task, result)) {\n\t\t\t\t\tagentLog(\n\t\t\t\t\t\t`[DISPATCH] agent=${task.agent_type} task_id=${task.task_id} preferred model failed; retrying with inherited model`,\n\t\t\t\t\t);\n\t\t\t\t\tthis.cleanupRetryArtifacts(task);\n\t\t\t\t\tretryScheduled = true;\n\t\t\t\t\tthis.queue.unshift({ ...task, useInheritedModelFallback: true });\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\tduration,\n\t\t\t\t\ttokens_used,\n\t\t\t\t\terror: result.error ?? `Exited with code ${code}`,\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t})\n\t\t\t.catch((err) => {\n\t\t\t\tthis.slots.delete(task.task_id);\n\t\t\t\tbudget.flush();\n\t\t\t\tconst duration = Date.now() - slot.spawned_at;\n\t\t\t\tconst tokens_used = budget.getUsed();\n\t\t\t\tif (!isRetry) {\n\t\t\t\t\t// The retry (started synchronously here) looks the budget up from\n\t\t\t\t\t// this.budgets and reuses it; flag it so the .finally below doesn't\n\t\t\t\t\t// strip the budget's listeners or delete the entry the retry now owns.\n\t\t\t\t\tretryScheduled = true;\n\t\t\t\t\tthis.startTask(task, true);\n\t\t\t\t\treturn;\n\t\t\t\t}\n\t\t\t\tconst error = err instanceof Error ? err.message : String(err);\n\t\t\t\tconst result: SubagentResult = {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tok: false,\n\t\t\t\t\tstdout,\n\t\t\t\t\tstderr,\n\t\t\t\t\texit_code: null,\n\t\t\t\t\terror,\n\t\t\t\t\tstatus: \"failed\",\n\t\t\t\t\tusedInheritedModelFallback: task.useInheritedModelFallback === true,\n\t\t\t\t};\n\t\t\t\tthis.writeOutputJson(task.task_id, result);\n\t\t\t\tthis.emit(\"task_failed\", {\n\t\t\t\t\ttask_id: task.task_id,\n\t\t\t\t\tagent_type: task.agent_type,\n\t\t\t\t\tduration,\n\t\t\t\t\ttokens_used,\n\t\t\t\t\terror,\n\t\t\t\t});\n\t\t\t\tthis.resolveWaiter(task.task_id, result);\n\t\t\t})\n\t\t\t.finally(() => {\n\t\t\t\t// Runs on success AND failure: the stdout reader used to be detached only\n\t\t\t\t// on the success path, leaking listeners (and their buffers) whenever a\n\t\t\t\t// child failed or the wait rejected.\n\t\t\t\tdetachStdoutReader?.();\n\t\t\t\tproc.stdout?.removeAllListeners(\"data\");\n\t\t\t\tproc.stderr?.removeAllListeners(\"data\");\n\t\t\t\tproc.stdout?.destroy();\n\t\t\t\tproc.stderr?.destroy();\n\t\t\t\tif (!retryScheduled) {\n\t\t\t\t\tbudget.removeAllListeners();\n\t\t\t\t\tthis.budgets.delete(task.task_id);\n\t\t\t\t}\n\t\t\t\tthis.pull();\n\t\t\t});\n\t}\n\n\t/** Whether a failed built-in subagent should be retried with `model: inherit`. */\n\tprivate shouldRetryWithInheritedModel(task: SubagentPoolTask, result: SubagentResult): boolean {\n\t\tif (task.useInheritedModelFallback) return false;\n\t\tif (task.sessionFile) return false;\n\t\t// Only the parent model is required: the provider may be unset when the\n\t\t// harness routes through a gateway. The retry inherits the parent model and\n\t\t// lets the child resolve the provider from its own default when none was threaded through.\n\t\tif (!task.model) return false;\n\n\t\tconst def = task.agent_type ? this.getRegistry().get(task.agent_type) : undefined;\n\t\t// Built-in agents always inherit; project agents may pin an explicit model in\n\t\t// frontmatter, so let them fall back too when that model is rejected.\n\t\tif (def?.source !== \"builtin\" && def?.source !== \"project\") return false;\n\t\tif (!def.model || def.model === MODEL_INHERIT) return false;\n\n\t\treturn this.isInheritedModelFallbackError(result);\n\t}\n\n\t/** Detect provider/model failures where inheriting the parent model can recover. */\n\tprivate isInheritedModelFallbackError(result: SubagentResult): boolean {\n\t\tconst text = [result.error, result.stderr, JSON.stringify(result.result_data ?? {})]\n\t\t\t.filter((part): part is string => typeof part === \"string\" && part.length > 0)\n\t\t\t.join(\"\\n\");\n\n\t\treturn INHERITED_MODEL_FALLBACK_ERROR.test(text);\n\t}\n\n\t/** Remove failed attempt artifacts before rerunning the same task id. */\n\tprivate cleanupRetryArtifacts(task: SubagentPoolTask): void {\n\t\tconst cwd = task.cwd ?? this.cwd;\n\t\tconst taskDir = getDispatchTaskDir(cwd, task.task_id);\n\t\tconst sessionFile = task.sessionFile ?? this.getSessionFile(task.task_id, cwd);\n\t\ttry {\n\t\t\trmSync(sessionFile, { force: true });\n\t\t\trmSync(join(taskDir, \"result.json\"), { force: true });\n\t\t\trmSync(join(taskDir, \"output.json\"), { force: true });\n\t\t} catch {\n\t\t\t// Best-effort cleanup; retry can still proceed with existing artifacts.\n\t\t}\n\t}\n\n\t/**\n\t * Best-effort concrete failure reason for a non-zero-exit subagent. Prefers\n\t * the child's result.json summary (which carries the provider/model error\n\t * message on failure), then the tail of stderr, then the exit code.\n\t */\n\tprivate deriveFailureReason(result: SubagentResult): string {\n\t\tconst summary = (result.result_data as { summary?: string } | undefined)?.summary?.trim();\n\t\tif (summary) {\n\t\t\treturn summary;\n\t\t}\n\t\tconst stderrTail = result.stderr\n\t\t\t.split(\"\\n\")\n\t\t\t.map((line) => line.trim())\n\t\t\t.filter((line) => line.length > 0)\n\t\t\t.slice(-5)\n\t\t\t.join(\"\\n\");\n\t\tif (stderrTail) {\n\t\t\treturn stderrTail;\n\t\t}\n\t\treturn `Exited with code ${result.exit_code}`;\n\t}\n\n\tprivate tryReadResultJson(task_id: string, cwd: string): Record<string, unknown> | undefined {\n\t\tconst path = join(getDispatchTaskDir(cwd, task_id), \"result.json\");\n\t\tif (!existsSync(path)) return undefined;\n\t\ttry {\n\t\t\tconst raw = readFileSync(path, \"utf-8\");\n\t\t\treturn JSON.parse(raw) as Record<string, unknown>;\n\t\t} catch {\n\t\t\treturn undefined;\n\t\t}\n\t}\n\n\tprivate resolveWaiter(task_id: string, result: SubagentResult): void {\n\t\t// Persist terminal status for get_status() even after wait_for consumes the result\n\t\tif (result.status === \"stalled\") this.taskStatus.set(task_id, \"stalled\");\n\t\telse if (result.status === \"timeout\") this.taskStatus.set(task_id, \"timeout\");\n\t\telse if (result.status === \"cancelled\") this.taskStatus.set(task_id, \"cancelled\");\n\t\telse if (result.ok) this.taskStatus.set(task_id, \"done\");\n\t\telse this.taskStatus.set(task_id, \"failed\");\n\n\t\tconst waiter = this.waiters.get(task_id);\n\t\tif (waiter) {\n\t\t\twaiter.resolve(result);\n\t\t\tthis.waiters.delete(task_id);\n\t\t\treturn;\n\t\t}\n\t\tthis.completed.set(task_id, result);\n\t}\n}\n"]}

package/dist/core/subagent-pool.js

@@ -8,6 +8,7 @@ import { writeFileAtomicSync } from "../utils/atomic-file.js";
 import { waitForChildProcess } from "../utils/child-process.js";
 import { killProcessTree } from "../utils/shell.js";
 import { MODEL_INHERIT } from "./agent-frontmatter.js";
+import { agentLog } from "./agent-log.js";
 import { loadAgentRegistry } from "./agent-registry.js";
 import { DispatchEvaluator } from "./dispatch-evaluator.js";
 import { SubagentLifeguard } from "./lifeguard.js";
@@ -329,9 +330,10 @@ export class SubagentPool extends EventEmitter {
         // Depth of the child about to be spawned (this process's depth + 1). Surfaced
         // so a delegation tree's nesting is visible in logs without extra tooling.
         const childDepth = currentSubagentDepth(this.env) + 1;
-        // Pre-dispatch logging. Use stderr: stdout is reserved for the JSON event
-        // stream / TUI render and must not be polluted.
-        console.error(`[DISPATCH] agent=${agent_type} depth=${childDepth} reason=${reason} complexity=${complexity} task_id=${task_id}`);
+        // Pre-dispatch logging. Goes through agentLog rather than console.error:
+        // stdout is reserved for the JSON event stream, and stderr is not free
+        // either while the interactive TUI owns the terminal (see agent-log.ts).
+        agentLog(`[DISPATCH] agent=${agent_type} depth=${childDepth} reason=${reason} complexity=${complexity} task_id=${task_id}`);
         this.writeDispatchLog(task_id, agent_type, reason, complexity, task, childDepth);
         const poolTask = {
             task_id,
@@ -813,7 +815,7 @@ export class SubagentPool extends EventEmitter {
                 result.error = this.deriveFailureReason(result);
             }
             if (this.shouldRetryWithInheritedModel(task, result)) {
-                console.error(`[DISPATCH] agent=${task.agent_type} task_id=${task.task_id} preferred model failed; retrying with inherited model`);
+                agentLog(`[DISPATCH] agent=${task.agent_type} task_id=${task.task_id} preferred model failed; retrying with inherited model`);
                 this.cleanupRetryArtifacts(task);
                 retryScheduled = true;
                 this.queue.unshift({ ...task, useInheritedModelFallback: true });