@rosthq/cli 0.7.109 → 0.7.111

package/dist/generator/groups.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"groups.d.ts","sourceRoot":"","sources":["../../src/generator/groups.ts"],"names":[],"mappings":"AAuCA,OAAO,EAAgB,KAAK,SAAS,EAAE,MAAM,YAAY,CAAC;AAM1D,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAKhF,CAAC;AAitCF,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAmPjD,CAAC"}
+{"version":3,"file":"groups.d.ts","sourceRoot":"","sources":["../../src/generator/groups.ts"],"names":[],"mappings":"AAuCA,OAAO,EAAgB,KAAK,SAAS,EAAE,MAAM,YAAY,CAAC;AAM1D,eAAO,MAAM,kBAAkB,EAAE,MAAM,CAAC,MAAM,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,MAAM,CAAA;CAAE,CAKhF,CAAC;AAotCF,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,SAAS,CAmPjD,CAAC"}

package/dist/index.js

@@ -41285,6 +41285,8 @@ var signalSummarySchema = external_exports.object({
   target: external_exports.number(),
   cadence: external_exports.enum(["daily", "weekly", "monthly", "quarterly", "annual"]),
   source: external_exports.enum(["agent", "human", "integration"]),
+  // DER-2066: read-time truth of how the measurable is currently fed (DER-1904 effectiveSource) — an active binding reads "auto" even while declared source stays "human". Additive; does not replace source.
+  effective_source: external_exports.enum(["auto", "agent", "human"]),
   latest_reading_id: uuidSchema9.nullable(),
   latest_value: external_exports.number().nullable(),
   latest_period_start: external_exports.string().nullable(),
@@ -43433,6 +43435,10 @@ var fleetOverviewRowSchema = external_exports.object({
   seat_id: uuidSchema14,
   seat_name: external_exports.string(),
   seat_path: external_exports.string(),
+  // DER-2068 (N-2): the seat's immediate parent seat name (null for the root
+  // seat, which has none) — the web fleet row renders "Reports to
+  // {parent_seat_name}" instead of the internal ltree path as visible text.
+  parent_seat_name: external_exports.string().nullable(),
   lane: agentLaneSchema2.nullable(),
   agent_status: external_exports.enum(["draft", "dry_run", "live", "paused", "decommissioned", "vacant"]),
   live: external_exports.boolean(),
@@ -43452,7 +43458,12 @@ var fleetOverviewRowSchema = external_exports.object({
   // runner-lane host-resource meter; null for cloud/mcp_session lanes.
   host_resource: fleetHostResourceSchema.nullable(),
   work_progress: fleetWorkProgressSchema,
-  work_progress_conflict: external_exports.boolean()
+  work_progress_conflict: external_exports.boolean(),
+  // DER-2066: true for Forge system seats (forge.developer_team.*) so a consumer can
+  // group/collapse them the way the web /agents fleet page does (DER-1919). Kept on the
+  // SHARED row schema (not a list_fleet-only extension) so agent.list_fleet and
+  // agent.fleet_digest stay in lockstep — the DER-1895 reference-equality guard. Additive.
+  is_system: external_exports.boolean()
 }).strict();
 var agentListFleetOutputSchema = external_exports.object({
   agents: external_exports.array(fleetOverviewRowSchema)
@@ -46362,19 +46373,82 @@ var softwareVercelDeployPreviewOutputSchema = external_exports.object({
   status: softwareVercelDeploymentStatusSchema
 });
 
-// ../../packages/protocol/src/software-factory-events.ts
+// ../../packages/protocol/src/software-factory-verification.ts
 var uuid13 = external_exports.string().uuid();
+var DELIVERY_SURFACE_STATUS_VALUES = [
+  "merged",
+  "deployed",
+  "verification_running",
+  "verified_live",
+  "failed",
+  "inconclusive",
+  "waived"
+];
+var deliverySurfaceStatusSchema = external_exports.enum(DELIVERY_SURFACE_STATUS_VALUES);
+var DELIVERY_SURFACE_KIND_VALUES = ["page", "api", "route_health"];
+var deliverySurfaceKindSchema = external_exports.enum(DELIVERY_SURFACE_KIND_VALUES);
+var DELIVERY_VERIFICATION_STATUS_VALUES = [
+  "pending",
+  "awaiting_deployment",
+  "running",
+  "verified",
+  "failed",
+  "inconclusive",
+  "waived"
+];
+var deliveryVerificationStatusSchema = external_exports.enum(DELIVERY_VERIFICATION_STATUS_VALUES);
+var affectedSurfaceSchema = external_exports.object({
+  kind: deliverySurfaceKindSchema,
+  path: external_exports.string().min(1).max(512)
+}).strict();
+var deliverySurfaceEvidenceSchema = external_exports.object({
+  statusCode: external_exports.number().int().nullable(),
+  durationMs: external_exports.number().nonnegative().nullable(),
+  summary: external_exports.string().max(500)
+}).strict().superRefine((value, ctx) => {
+  if (hasSecretShapedValue(value.summary)) {
+    ctx.addIssue({
+      code: external_exports.ZodIssueCode.custom,
+      path: ["summary"],
+      message: "evidence summary must not contain raw secret material"
+    });
+  }
+});
+var deliverySurfaceReportSchema = external_exports.object({
+  path: external_exports.string().min(1).max(512),
+  kind: deliverySurfaceKindSchema,
+  status: deliverySurfaceStatusSchema,
+  sha: external_exports.string().nullable(),
+  evidence: external_exports.string().max(500).nullable(),
+  checkedAt: external_exports.string().nullable()
+}).strict();
+var deliveryVerificationReportSchema = external_exports.object({
+  buildRequestId: uuid13,
+  status: deliveryVerificationStatusSchema,
+  expectedSha: external_exports.string().min(1),
+  observedSha: external_exports.string().nullable(),
+  testTenantRef: external_exports.string().min(1),
+  startedAt: external_exports.string().nullable(),
+  verifiedAt: external_exports.string().nullable(),
+  kickbackWorkOrderId: uuid13.nullable(),
+  waiverDecisionId: uuid13.nullable(),
+  waiverReason: external_exports.string().nullable(),
+  surfaces: external_exports.array(deliverySurfaceReportSchema)
+}).strict();
+
+// ../../packages/protocol/src/software-factory-events.ts
+var uuid14 = external_exports.string().uuid();
 var softwareFactoryPhaseRunCompletedEventSchema = external_exports.object({
-  tenantId: uuid13,
-  phaseRunId: uuid13,
-  buildRequestId: uuid13.nullish(),
+  tenantId: uuid14,
+  phaseRunId: uuid14,
+  buildRequestId: uuid14.nullish(),
   status: external_exports.enum(["passed", "failed"]),
   idempotencyKey: external_exports.string().min(1).optional()
 }).strict();
 var softwareFactoryBudgetExhaustedEventSchema = external_exports.object({
-  tenantId: uuid13,
-  phaseRunId: uuid13,
-  buildRequestId: uuid13,
+  tenantId: uuid14,
+  phaseRunId: uuid14,
+  buildRequestId: uuid14,
   reason: external_exports.string().min(1),
   idempotencyKey: external_exports.string().min(1).optional()
 }).strict();
@@ -47765,7 +47839,7 @@ The Compass is drafted, then activated by a human through supersession.
     order: 15,
     title: "AICOS chat guide",
     summary: "How the AI Chief of Staff chat works in the authenticated app shell and what it can safely do today.",
-    version: "2026-07-15.1",
+    version: "2026-07-20.1",
     public: true,
     audiences: ["human", "in_app_agent"],
     stages: ["company_setup", "operating_rhythm"],
@@ -47783,7 +47857,10 @@ The Compass is drafted, then activated by a human through supersession.
       "agent.configure_tools",
       "agent.run_dry_run",
       "agent.go_live",
-      "charter.sign_manifest"
+      "charter.sign_manifest",
+      "tool_grants.agent.update",
+      "agent.update_schedule",
+      "skill.assign_to_seat"
     ],
     legal: {
       publicRisk: "low",
@@ -47879,6 +47956,8 @@ Durable agent lifecycle changes still use the existing command surface:
 - \`agent.go_live\` only happens after the required approval gate.
 - \`confirmation.approve\` is the human approval command for pending confirmations.
 
+When a user asks AICOS to change an agent's capabilities or configuration \u2014 grant a tool to a seat, set or clear an agent's schedule, or assign a skill \u2014 AICOS acts on the request instead of only explaining it. It stages the change as a draft-and-confirm proposal through the governed command and links the pending confirmation on the Approvals queue for a human to approve. Each of these proposals (\`tool_grants.agent.update\`, \`agent.update_schedule\`, \`skill.assign_to_seat\`) is human-required and always-human, so nothing durable changes until a person confirms it and AICOS never applies a capability change on its own. When AICOS does not hold the command for a request, it names the exact gap and points to the seat page where a human can make the change, rather than returning a generic how-to answer.
+
 If the user asks for report-only or no-action help, AICOS stays in plan-only mode and does not create or advance a lifecycle change. Seat-scoped staging only uses the server-authorized pinned seat from the session; client-supplied tenant or seat context is ignored as authority. Runner and MCP lanes do not bypass the same server command guards, and they do not grant hidden browser tool execution. Raw secrets, vault material, and secret-shaped metadata stay out of transcript storage.
 
 ## Seat-scoped AICOS
@@ -48488,7 +48567,7 @@ Decisions should be recorded as human decisions. Handoffs should attach to seats
     order: 45,
     title: "How agents work",
     summary: "How {{brand}} agents operate inside seats, use Charters, report work, and escalate beyond authority.",
-    version: "2026-07-18.1",
+    version: "2026-07-20.1",
     public: true,
     audiences: ["human", "cli", "mcp", "in_app_agent"],
     stages: ["staffing", "operating_rhythm"],
@@ -48588,7 +48667,7 @@ Beyond each seat's own manifest, the company can set one ceiling on how much an
 
 The Agents page shows fleet health first: live state, work evidence, failed runs, open held actions, cost, and the next operator action. Work evidence distinguishes fresh real runs, stale runs, no completed run yet, queued or running work orders, and recent failures. A failure link opens the failed run itself; queued or stale rows route the operator to the seat's agent operations instead of pretending there is a hidden run to inspect. The page also shows configuration completeness for each staffed agent seat: Steward chain, lane or runtime substrate, vaulted credentials, granted tools, model config, sandbox dry run, and go-live state, each with a remediation link when incomplete. A seat's Trust Card drills into the same facts: run history, one-run detail, tool-call outcomes, product-visible errors, transcript references, loaded Skill versions, and held confirmations or escalations. The same seat's Activity tab also shows a curated business-activity ledger (counter tiles plus a run-grouped \`work.record\` feed) and a sessions browser with the full persisted transcript and artifacts for the selected run \u2014 a layer above the Trust Card's forensic trace, not a replacement for it. The fleet page's own **Work** tab mirrors that ledger across every seat in one filterable, seat-labelled stream. Open confirmations stay visible until they are decided or expire, even when they are older than the default activity window. Approving or rejecting a held confirmation from the seat page is scoped to that seat; rejection can include a short reason. For a live agent, the seat page's agent operations panel can queue one immediate governed run through the same \`agent.run_now\` command path used by CLI and MCP; it does not expand authority or change the saved schedule. Seat pages render legacy Charter manifest scope text with customer-safe labels when old setup data used internal placeholder wording.
 
-Review the first dry runs, fleet overview, tool-call audit rows, escalations, deliverables, and Signal impact. A seat's completed-work area combines explicit deliverables, work-log evidence, and successful run summaries into one operator-facing trail with safe links labelled as internal, Linear, GitHub, or external. The fleet view at \`/agents\` shows every staffed agent seat at a glance; the agent-native equivalents are \`{{cli}} command agent.list_fleet --json '{}'\` / \`rost_list_agent_fleet\` for the compact overview, and \`{{cli}} agent fleet-digest --json\` / \`agent.fleet_digest\` / \`rost_get_dogfood_fleet_health_digest\` for the daily dogfood evidence bundle. The digest returns live/idle state, 24h/7d turns, 7d/30d spend, recent failed runs, unresolved product errors, failed notifications, and a next action per seat.
+Review the first dry runs, fleet overview, tool-call audit rows, escalations, deliverables, and Signal impact. A seat's completed-work area combines explicit deliverables, work-log evidence, and successful run summaries into one operator-facing trail with safe links labelled as internal, Linear, GitHub, or external. The fleet view at \`/agents\` shows every staffed agent seat at a glance; the agent-native equivalents are \`{{cli}} command agent.list_fleet --json '{}'\` / \`rost_list_agent_fleet\` for the compact overview (each seat marked with \`is_system\` for Forge system seats under \`forge.developer_team.*\`, so a caller can group or collapse them the way the web fleet page does), and \`{{cli}} agent fleet-digest --json\` / \`agent.fleet_digest\` / \`rost_get_dogfood_fleet_health_digest\` for the daily dogfood evidence bundle. The digest returns live/idle state, 24h/7d turns, 7d/30d spend, recent failed runs, unresolved product errors, failed notifications, and a next action per seat.
 
 For an operational audit, start with the \`/health\` dashboard or \`{{cli}} system health --json\` / \`system.health\` / \`rost_system_health\` before manually chaining fleet, run, Signal, Friction, task, integration, runner, and Sync commands. The dashboard, CLI, and MCP tool share the same command-backed system-health model, with server-derived tenant, member, requested-seat, or seat-token scope. Owners see tenant health; members see occupied-seat subtrees; seat-scoped tokens see their own seat plus authorized downstream seats, and optional \`seat_id\` narrowing can only reduce that server-derived scope. Seat pages link to \`/health?seat_id=<seat>\` for scoped follow-up. Every health finding includes a remediation action with a human label/link plus command id, schema reference, CLI command, help text, and human-gated marker for agents. Open Friction without an action task links to the Friction detail and gives agents the \`task.create\` then \`friction.link_task\` next step; the final issue resolution still requires the human \`friction.resolve\` decision.
 
@@ -48753,7 +48832,7 @@ External connectors are being rolled out provider by provider, conservatively (r
     order: 48,
     title: "CLI and MCP installation guide",
     summary: "Install the public CLI, register remote token-backed MCP clients, and find the full command and tool catalog.",
-    version: "2026-07-19.2",
+    version: "2026-07-20.1",
     public: true,
     audiences: ["human", "cli", "mcp", "in_app_agent"],
     stages: ["company_setup", "staffing"],
@@ -49365,7 +49444,7 @@ Several rows here are seat-operating commands (\`task.create\`, the \`signal.*\`
 | \`rost_get_current_compass\` | \`compass.get_current\` | Read the active and draft Compass versions and source documents. | Tenant | Call with \`{}\`. |
 | \`rost_list_compass_gaps\` | \`compass.list_gaps\` | List unanswered and answered Compass context gaps. | Tenant | Call with \`{}\` before answering gaps. |
 | \`rost_get_agent_status\` | \`agent.status\` | Read agent lane, live state, steward chain, dry-run result, Runner availability. | Seat or tenant-admin | Call with \`{"seat_id":"<seat-id>"}\`. |
-| \`rost_list_agent_fleet\` | \`agent.list_fleet\` | Read every staffed agent seat at once: lane, live state, last real turn, 24h/7d real turns, measurable status, escalations, and 7-day spend, plus the liveness/health/work-progress axes the web fleet page renders (liveness state, health status, work evidence, host-resource pressure). | Tenant | Call with \`{}\`; counts use the same seat-run association as \`agent.list_runs\`, with sandbox dry runs excluded from real turns. |
+| \`rost_list_agent_fleet\` | \`agent.list_fleet\` | Read every staffed agent seat at once: lane, live state, last real turn, 24h/7d real turns, measurable status, escalations, and 7-day spend, plus the liveness/health/work-progress axes the web fleet page renders (liveness state, health status, work evidence, host-resource pressure), plus an \`is_system\` marker for Forge system seats. | Tenant | Call with \`{}\`; counts use the same seat-run association as \`agent.list_runs\`, with sandbox dry runs excluded from real turns. |
 | \`rost_get_dogfood_fleet_health_digest\` | \`agent.fleet_digest\` | Capture the daily dogfood fleet-health digest: live/idle state, 24h/7d turns, spend, recent failed runs, unresolved errors, failed notifications, and next actions. | Tenant | Call with \`{}\`; optional \`window_hours\` bounds recent failed-run evidence, while \`error_limit\` caps linked evidence per seat. |
 | \`rost_system_health\` | \`system.health\` | Read the scoped system-health snapshot across agent runs, unresolved errors, Signals, Cascade setup, work loop, integrations, runners, notifications, and Sync Brief readiness. | Tenant, member, or seat token | Call with \`{}\`; optional \`seat_id\` narrows inside the caller's server-derived scope. |
 | \`rost_get_run_heartbeats\` | \`agent.run_heartbeats\` | Read the per-seat run-heartbeat roll-up (quiet-but-healthy vs stalled) for long-running agent programs. | Tenant | Call with \`{}\`; optional \`seat_ids\` narrows to specific seats. |
@@ -49437,7 +49516,7 @@ Several rows here are seat-operating commands (\`task.create\`, the \`signal.*\`
 | \`rost_update_model_gateway_policy\` | \`tenant.model_gateway_policy.update\` | Unlock or relock managed models that lack a zero-retention (ZDR) tier. | Tenant-admin | Owner-only, human-gated; call with \`{"allow_non_zdr_models":true}\` (or \`false\` to relock). Unlocking establishes an ADR-0018 \xA76 standing authorization and surfaces a per-model retention disclosure. |
 | \`rost_get_aicos_brain_settings\` | \`aicos.brain_settings.get\` | Read AICOS lane and brain labels for cloud managed/BYOK, runner Claude/Codex, and MCP client. | Tenant | Call with \`{}\`; returns enum labels only, never keys, vault refs, or runner secrets. |
 | \`rost_update_aicos_brain_settings\` | \`aicos.brain_settings.update\` | Update AICOS lane and cloud/runner brain preferences. | Tenant-admin | Owner-only and human-gated; call with \`{"lane":"runner","runner_brain":"claude-cli"}\` or \`{"cloud_brain":"byok"}\`. BYOK requires an active tenant Anthropic key. For AICOS, selecting \`codex-cli\` is currently rejected until the governed interactive runner path is verified. |
-| \`rost_list_signals\` | \`signal.list\` | List measurables with their latest reading and on/off-track state. | Seat or tenant-admin | Call with \`{}\` to find measurable ids. |
+| \`rost_list_signals\` | \`signal.list\` | List measurables with their latest reading, on/off-track state, and \`effective_source\` (read-time measured-by). | Seat or tenant-admin | Call with \`{}\` to find measurable ids. |
 | \`rost_get_signal\` | \`signal.get\` | Read a measurable with its full reading history. | Seat or tenant-admin | Call with \`{"measurable_id":"<id>"}\`. |
 | \`rost_confirm_signal_reading\` | \`signal.confirm_reading\` | Confirm an unconfirmed reading as human-verified. | Seat or tenant-admin | Humans confirm; call with \`{"reading_id":"<id>"}\`. |
 | \`rost_correct_signal_reading\` | \`signal.correct_reading\` | Overwrite a reading with a human-confirmed value. | Seat or tenant-admin | Manual correction; expect confirmation. |
@@ -50174,7 +50253,7 @@ Agents can suggest commitments and report progress. They should not create a new
     order: 61,
     title: "Signal guide",
     summary: "How to define and read measurables so the company runs on evidence instead of status theater.",
-    version: "2026-07-18.3",
+    version: "2026-07-20.2",
     public: true,
     audiences: ["human", "cli", "mcp", "in_app_agent"],
     stages: ["operating_rhythm"],
@@ -50229,11 +50308,11 @@ Avoid vanity numbers, manual-only status fields, and metrics nobody can act on.
 
 ## Operate Signal from CLI or MCP
 
-- Read: \`{{cli}} signal list --json\` / \`signal.list\` / \`rost_list_signals\` returns measurables with their latest reading and on/off-track state. \`signal.get\` / \`rost_get_signal\` returns one measurable's full reading history.
+- Read: \`{{cli}} signal list --json\` / \`signal.list\` / \`rost_list_signals\` returns measurables with their latest reading and on/off-track state, plus \`effective_source\` \u2014 the read-time measured-by (\`auto\` when a source binding is active, otherwise \`agent\`/\`human\`), matching the \`/signal\` UI and distinct from the declared \`source\`. \`signal.get\` / \`rost_get_signal\` returns one measurable's full reading history.
 - Add a measurable: \`measurable.create\` (scope: seat) defines a measurable a seat owns \u2014 name, unit, direction, target, cadence. The seat owns it; readings attach to it afterward.
 - Record a reading: \`{{cli}} status record --measurable-id <id> --value <n>\` (\`status.record\`, scope: seat) writes a status event with the reading. This is not gated. An agent's \`status.record\` never downgrades a human-confirmed reading: a routine agent read that lands on a period a human already confirmed leaves the confirmed value and its confirmation intact (invariants #7/#8 \u2014 agents recommend; humans decide).
 - Confirm a reading: \`{{cli}} signal confirm\` / \`signal.confirm_reading\` / \`rost_confirm_signal_reading\` marks a reading human-verified. \`signal.correct_reading\` / \`rost_correct_signal_reading\` overwrites a reading with a human-confirmed value.
-- Draft first readings from a connected source: \`signal.draft_first_readings\` pulls a measurable's bound source once (read-only, SSRF-guarded \u2014 the same pull path as \`signal.preview\`) and lands the value as a DRAFT reading for the current period, for a human to confirm. It is for a measurable that has a connected source but no confirmed reading yet: the Signal page shows a "Draft first readings" button on those rows so the row stops being a permanent blank. Nothing is published \u2014 a human clears the draft with \`signal.confirm_reading\` (invariant #7) \u2014 and it never fabricates a value: a non-numeric or blocked pull drafts nothing and says so.
+- Draft first readings from a connected source: \`signal.draft_first_readings\` pulls a measurable's bound source once (read-only, SSRF-guarded \u2014 the same pull path as \`signal.preview\`) and lands the value as a DRAFT reading for the current period, for a human to confirm. It is strictly for a measurable that has a connected source but has never had a confirmed reading \u2014 in ANY period, not just the current one. A measurable that already carries a confirmed reading (even an older one) is rejected with an "already has readings" outcome and no pull happens, so the first-reading flow can never be repurposed as an ongoing draft source; use \`signal.report\` to add a current-period datapoint or \`signal.correct_reading\` to change a confirmed value instead. The Signal page shows a "Draft first readings" button on the eligible (never-measured) rows so the row stops being a permanent blank. Nothing is published \u2014 a human clears the draft with \`signal.confirm_reading\` (invariant #7) \u2014 and it never fabricates a value: a non-numeric or blocked pull drafts nothing and says so.
 
 ## Run Signal without an agent
 
@@ -53707,7 +53786,7 @@ function renderSignalList(output) {
   }
   return signals.map((signal) => {
     const record2 = asRecord(signal);
-    return `${field(record2, "measurable_id")}  ${field(record2, "state")}  ${field(record2, "name")} (latest ${field(record2, "latest_value")} / target ${field(record2, "target")})`;
+    return `${field(record2, "measurable_id")}  ${field(record2, "state")}  ${field(record2, "name")} (latest ${field(record2, "latest_value")} / target ${field(record2, "target")}, measured by ${field(record2, "effective_source")})`;
   }).join("\n");
 }
 function renderSignalGet(output) {
@@ -57658,6 +57737,7 @@ ${usage()}
   let failed = false;
   let activeSessions = 0;
   const activeTurnExecutionIds = /* @__PURE__ */ new Set();
+  let serverHeartbeatContractVersion = null;
   do {
     const detectedCapabilities = await detectCapabilities(cliVersion, env, homeDir, config2.maxSessionsOverride);
     const capabilities = capabilitiesForConfiguredRuntime(detectedCapabilities, config2.runtime);
@@ -57666,7 +57746,8 @@ ${usage()}
       const heartbeat = await post2(fetchImpl, options.appUrl, "/api/runner/heartbeat", buildHeartbeatBody({
         capabilities,
         telemetry: detectTelemetry(capabilities, env, activeSessions, cliVersion),
-        activeTurnExecutionIds: [...activeTurnExecutionIds]
+        activeTurnExecutionIds: [...activeTurnExecutionIds],
+        serverContractVersion: serverHeartbeatContractVersion
       }), state.runner_secret);
       if (heartbeat.status !== 200) {
         options.io.stderr.write(`heartbeat ${heartbeat.status}: ${redactForLog(JSON.stringify(heartbeat.json))}
@@ -57674,6 +57755,10 @@ ${usage()}
         failed = true;
       } else {
         heartbeatSucceeded = true;
+        const learnedVersion = readHeartbeatContractVersion(heartbeat.json);
+        if (learnedVersion !== null) {
+          serverHeartbeatContractVersion = learnedVersion;
+        }
         options.io.stdout.write(`heartbeat ok #${beat} runner_id=${state.runner_id} claude=${capabilities.claude.installed} codex=${capabilities.codex.installed}
 `);
       }
@@ -57687,7 +57772,11 @@ ${usage()}
           capabilities,
           io: options.io,
           telemetry: () => detectTelemetry(capabilities, env, activeSessions, cliVersion),
-          activeTurnIds: () => [...activeTurnExecutionIds]
+          activeTurnIds: () => [...activeTurnExecutionIds],
+          serverContractVersion: () => serverHeartbeatContractVersion,
+          recordServerContractVersion: (v) => {
+            serverHeartbeatContractVersion = v;
+          }
         });
         try {
           await claimAndExecute(fetchImpl, options.appUrl, state, config2, options.io, localRuntime, activeTurnExecutionIds);
@@ -57712,13 +57801,17 @@ ${usage()}
         localRuntime,
         capabilities,
         () => detectTelemetry(capabilities, env, activeSessions, cliVersion),
-        activeTurnExecutionIds
+        activeTurnExecutionIds,
+        () => serverHeartbeatContractVersion,
+        (v) => {
+          serverHeartbeatContractVersion = v;
+        }
       );
     }
   } while (!config2.once);
   return heartbeatSucceeded && !failed ? 0 : 1;
 }
-async function waitForNextHeartbeat(fetchImpl, appUrl2, state, config2, io, runtime, capabilities, telemetry, activeTurnExecutionIds) {
+async function waitForNextHeartbeat(fetchImpl, appUrl2, state, config2, io, runtime, capabilities, telemetry, activeTurnExecutionIds, serverContractVersion, recordServerContractVersion) {
   const deadline = Date.now() + config2.heartbeatMs;
   const stopWait = runtime ? startTurnHeartbeat({
     fetchImpl,
@@ -57728,7 +57821,9 @@ async function waitForNextHeartbeat(fetchImpl, appUrl2, state, config2, io, runt
     capabilities,
     io,
     telemetry,
-    activeTurnIds: () => [...activeTurnExecutionIds]
+    activeTurnIds: () => [...activeTurnExecutionIds],
+    serverContractVersion,
+    recordServerContractVersion
   }) : null;
   try {
     while (Date.now() < deadline) {
@@ -57748,12 +57843,17 @@ async function waitForNextHeartbeat(fetchImpl, appUrl2, state, config2, io, runt
     stopWait?.();
   }
 }
+var HEARTBEAT_ACTIVE_TURN_MIN_CONTRACT_VERSION = 2;
+function readHeartbeatContractVersion(json2) {
+  const value = json2.heartbeat_contract_version;
+  return typeof value === "number" && Number.isFinite(value) ? value : null;
+}
 function buildHeartbeatBody(input) {
   const body = {
     capabilities: input.capabilities,
     telemetry: input.telemetry
   };
-  if (input.activeTurnExecutionIds.length > 0) {
+  if (input.activeTurnExecutionIds.length > 0 && (input.serverContractVersion ?? 0) >= HEARTBEAT_ACTIVE_TURN_MIN_CONTRACT_VERSION) {
     body.active_turn_execution_ids = input.activeTurnExecutionIds;
   }
   return body;
@@ -57774,11 +57874,17 @@ function startTurnHeartbeat(deps) {
         const response = await post2(deps.fetchImpl, deps.appUrl, "/api/runner/heartbeat", buildHeartbeatBody({
           capabilities: deps.capabilities,
           telemetry: deps.telemetry(),
-          activeTurnExecutionIds: deps.activeTurnIds()
+          activeTurnExecutionIds: deps.activeTurnIds(),
+          serverContractVersion: deps.serverContractVersion()
         }), deps.secret, controller.signal);
         if (!stopped && response.status !== 200) {
           deps.io.stderr.write(`turn heartbeat ${response.status}: ${redactForLog(JSON.stringify(response.json))}
 `);
+        } else if (!stopped && response.status === 200) {
+          const learnedVersion = readHeartbeatContractVersion(response.json);
+          if (learnedVersion !== null) {
+            deps.recordServerContractVersion(learnedVersion);
+          }
         }
       } catch (error51) {
         if (!stopped) {