@indigoai-us/hq-cli 5.294.0 → 5.296.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (72) hide show
  1. package/CHANGELOG.md +33 -0
  2. package/dist/command-catalog.generated.d.ts +70 -9
  3. package/dist/command-catalog.generated.js +88 -12
  4. package/dist/commands/auth.d.ts +1 -0
  5. package/dist/commands/auth.js +76 -1
  6. package/dist/commands/core-job-native.d.ts +7 -0
  7. package/dist/commands/core-job-native.js +20 -0
  8. package/dist/commands/core-job-probe-native.d.ts +45 -0
  9. package/dist/commands/core-job-probe-native.js +564 -0
  10. package/dist/commands/core-job-remote-launcher.d.ts +7 -0
  11. package/dist/commands/core-job-remote-launcher.js +51 -0
  12. package/dist/commands/core-job-run-native.d.ts +47 -0
  13. package/dist/commands/core-job-run-native.js +1215 -0
  14. package/dist/commands/core.js +1 -1
  15. package/dist/commands/doctor.js +0 -74
  16. package/dist/commands/lanes-message.js +24 -5
  17. package/dist/commands/lanes-resume.js +1 -3
  18. package/dist/commands/lanes-shared.d.ts +1 -2
  19. package/dist/commands/lanes-stories.d.ts +1 -0
  20. package/dist/commands/lanes-stories.js +40 -1
  21. package/dist/commands/lanes.d.ts +1 -9
  22. package/dist/commands/lanes.js +197 -174
  23. package/dist/commands/run.js +5 -0
  24. package/dist/fast-scaffold.d.ts +0 -5
  25. package/dist/fast-scaffold.js +0 -10
  26. package/dist/index.js +1 -10
  27. package/dist/lib/core-utils/sentry-report.d.ts +0 -19
  28. package/dist/lib/core-utils/sentry-report.js +8 -61
  29. package/dist/lib/doctor/checks/claude-wiring.js +9 -0
  30. package/dist/lib/lanes/flag-keys.d.ts +2 -0
  31. package/dist/lib/lanes/flag-keys.js +2 -0
  32. package/dist/lib/lanes/manager-status-gate.d.ts +0 -17
  33. package/dist/lib/lanes/manager-status-gate.js +2 -43
  34. package/dist/lib/lanes/mesh-events.js +3 -0
  35. package/dist/lib/lanes/outpost-dispatch.d.ts +2 -0
  36. package/dist/lib/lanes/outpost-dispatch.js +7 -1
  37. package/dist/lib/lanes/outpost-hire-cache.d.ts +0 -1
  38. package/dist/lib/lanes/outpost-hire-cache.js +0 -1
  39. package/dist/lib/lanes/outpost-snapshot-memo.d.ts +6 -0
  40. package/dist/lib/lanes/outpost-snapshot-memo.js +14 -0
  41. package/dist/lib/lanes/reconcile.d.ts +2 -3
  42. package/dist/lib/lanes/reconcile.js +67 -57
  43. package/dist/lib/lanes/spawn.d.ts +0 -2
  44. package/dist/lib/lanes/spawn.js +57 -5
  45. package/dist/lib/lanes/store.js +54 -0
  46. package/dist/lib/lanes/stories.d.ts +7 -1
  47. package/dist/lib/lanes/stories.js +16 -0
  48. package/dist/lib/lanes/types.d.ts +30 -1
  49. package/dist/lib/lanes/types.js +2 -2
  50. package/dist/lib/lanes/watcher.d.ts +4 -3
  51. package/dist/lib/lanes/watcher.js +122 -56
  52. package/dist/lib/mesh/live/daemon/run.js +18 -1
  53. package/dist/run/env-graph-guard.d.ts +2 -2
  54. package/dist/run/env-graph-guard.js +5 -2
  55. package/dist/run/hq-plugin.js +6 -5
  56. package/dist/utils/cognito-session.d.ts +6 -0
  57. package/dist/utils/cognito-session.js +9 -0
  58. package/dist/utils/self-update.js +31 -0
  59. package/dist/utils/version-gate.js +22 -3
  60. package/package.json +2 -2
  61. package/dist/fast-lanes-list-route.d.ts +0 -8
  62. package/dist/fast-lanes-list-route.js +0 -32
  63. package/dist/fast-lanes-list.d.ts +0 -3
  64. package/dist/fast-lanes-list.js +0 -224
  65. package/dist/lib/doctor/client-health-sentry-flag.d.ts +0 -23
  66. package/dist/lib/doctor/client-health-sentry-flag.js +0 -99
  67. package/dist/lib/doctor/client-health-sentry.d.ts +0 -71
  68. package/dist/lib/doctor/client-health-sentry.js +0 -286
  69. package/dist/lib/lanes/outpost-snapshot-cache.d.ts +0 -26
  70. package/dist/lib/lanes/outpost-snapshot-cache.js +0 -215
  71. package/dist/lib/lanes/outpost-snapshot-refresh.d.ts +0 -21
  72. package/dist/lib/lanes/outpost-snapshot-refresh.js +0 -329
@@ -31,8 +31,7 @@ import { applyDefaultQuotaHold, readQuotaSignal, recordAccountQuota } from "./qu
31
31
  import { answeredQuestionEntries, questionRecordRefusalCode, raiseEnvelopeQuestions, resolveQuestionCompanyWithFallback, } from "./questions.js";
32
32
  import { SILENCE_MS, appendWatcherEvent, classifyLaneExit, isPidAlive, missingExpectedEnvelopeFile, readLatestEnvelopeFile, readLatestEnvelopeRecord, watcherEventsPath, watcherOutLogPath, writeWatcherDecision, } from "./watcher.js";
33
33
  import { createLaneMesh, emitLaneDecisionEvents, emitLaneTerminalStateEvent, laneMeshCompany, } from "./mesh-events.js";
34
- import { activeLaneStories, laneStories, storyStatusForEnvelope } from "./stories.js";
35
- import { resolveManagerStatusGateForLane } from "./manager-status-gate.js";
34
+ import { activeLaneStories, laneStories, terminalEnvelopeStoryStatus } from "./stories.js";
36
35
  import { decideEnvelopeRepair, envelopeQuestionIds, routeProviderFailure } from "./watcher.js";
37
36
  import { PROVIDER_REJECTED_REASON, providerRejectionDetails, readProviderRejection, } from "./provider-rejection.js";
38
37
  /** Spec §9: unanswered at the current hop for 60 minutes → one hop. */
@@ -48,43 +47,12 @@ export const PICKUP_VALIDATION_MAX_AGE_MS = 15 * 60 * 1000;
48
47
  export const PICKUP_VALIDATION_TIMEOUT_REASON = "pickup_validation_timeout";
49
48
  export const RECONCILE_UNIT_NAME = "hq-lanes-reconcile.service";
50
49
  export const RECONCILE_TIMER_NAME = "hq-lanes-reconcile.timer";
51
- function spawnDepsForHost(deps, lane) {
50
+ function spawnDepsForHost(deps) {
52
51
  return {
53
52
  ...(deps.spawnLaneDeps ?? {}),
53
+ env: deps.spawnLaneDeps?.env ?? deps.env ?? process.env,
54
54
  hostname: deps.host.hostname,
55
55
  admissionHost: deps.host,
56
- ...(lane !== undefined && deps.managerStatusGateByLane?.has(lane.lane_id)
57
- ? { managerStatusGateEnabled: deps.managerStatusGateByLane.get(lane.lane_id) === true }
58
- : {}),
59
- };
60
- }
61
- async function prepareManagerStatusGateStarts(root, lanes, deps, summary) {
62
- const enabledByLane = new Map();
63
- let gateFailureRecorded = false;
64
- if (!deps.dryRun) {
65
- await Promise.all(lanes.filter((lane) => laneRunsHere(lane, deps.host)).map(async (lane) => {
66
- const resolution = await resolveManagerStatusGateForLane(root, lane, {
67
- env: deps.env,
68
- readFlag: deps.readRegistryFlag,
69
- onError: (err) => {
70
- if (gateFailureRecorded)
71
- return;
72
- gateFailureRecorded = true;
73
- const code = err instanceof LanesError ? err.error : "manager_status_gate_unavailable";
74
- const message = err instanceof Error ? err.message : String(err);
75
- summary.skipped.push({
76
- kind: "lane",
77
- id: lane.lane_id,
78
- why: `manager status gate disabled for this reconcile pass: ${code}: ${message}`,
79
- });
80
- },
81
- });
82
- enabledByLane.set(lane.lane_id, resolution.enabled);
83
- }));
84
- }
85
- return {
86
- ...deps,
87
- managerStatusGateByLane: enabledByLane,
88
56
  };
89
57
  }
90
58
  function logSkip(skip) {
@@ -854,7 +822,7 @@ export async function runPassBody(hqRoot, deps) {
854
822
  if (deps.afterLoad)
855
823
  await deps.afterLoad();
856
824
  await runRunawayToolGuardPass(hqRoot, lanes, config, deps, now, summary);
857
- const startDeps = await prepareManagerStatusGateStarts(hqRoot, lanes, deps, summary);
825
+ const startDeps = deps;
858
826
  markOverdueLanes(hqRoot, lanes, summary, deps, now);
859
827
  await heartbeatAndDeadHosts(hqRoot, lanes, summary, startDeps, now);
860
828
  for (const question of questions) {
@@ -1229,7 +1197,7 @@ async function heartbeatAndDeadHosts(hqRoot, lanes, summary, deps, now) {
1229
1197
  classified.account) {
1230
1198
  try {
1231
1199
  const recovered = await retryLaneOnAlternateAccount(hqRoot, classified, accountRetryReason, "reconcile", {
1232
- ...spawnDepsForHost(deps, classified),
1200
+ ...spawnDepsForHost(deps),
1233
1201
  });
1234
1202
  summary.dead_host.push({
1235
1203
  id: lane.lane_id,
@@ -1303,24 +1271,22 @@ async function heartbeatAndDeadHosts(hqRoot, lanes, summary, deps, now) {
1303
1271
  }
1304
1272
  if (terminalEnvelope && latest) {
1305
1273
  const envelope = terminalEnvelope;
1306
- next.story_status = envelope.story_status;
1307
- next.story_complete = envelope.story_complete;
1308
- next.story_complete_reason = envelope.story_complete_reason;
1309
- const storyStatus = envelope.decision === "blocked"
1310
- ? "blocked"
1311
- : envelope.decision === "done"
1312
- ? "review"
1313
- : storyStatusForEnvelope(envelope.story_status);
1314
- next.stories = laneStories(current).map((story) => story.project_id === current.project_id && story.story_id === current.story_id
1315
- ? {
1316
- ...story,
1317
- status: storyStatus,
1318
- updated_at: nowIso,
1319
- ...(envelope.decision === "blocked"
1320
- ? { blocked_reason: envelope.blocked_reason ?? "" }
1321
- : { blocked_reason: undefined }),
1322
- }
1323
- : story);
1274
+ const storyStatus = terminalEnvelopeStoryStatus(current, envelope.decision, envelope.story_status);
1275
+ if (storyStatus !== undefined) {
1276
+ next.story_status = envelope.story_status;
1277
+ next.story_complete = envelope.story_complete;
1278
+ next.story_complete_reason = envelope.story_complete_reason;
1279
+ next.stories = laneStories(current).map((story) => story.project_id === current.project_id && story.story_id === current.story_id
1280
+ ? {
1281
+ ...story,
1282
+ status: storyStatus,
1283
+ updated_at: nowIso,
1284
+ ...(envelope.decision === "blocked"
1285
+ ? { blocked_reason: envelope.blocked_reason ?? "" }
1286
+ : { blocked_reason: undefined }),
1287
+ }
1288
+ : story);
1289
+ }
1324
1290
  if (approvalGate?.violation) {
1325
1291
  next.last_envelope = {
1326
1292
  ref: latest.ref,
@@ -1512,7 +1478,7 @@ async function admitQueued(hqRoot, lanes, summary, deps, now) {
1512
1478
  nowMs,
1513
1479
  });
1514
1480
  const spawnFn = deps.spawn ??
1515
- ((root, lane) => spawnLaneWorker(root, lane, spawnDepsForHost(deps, lane)));
1481
+ ((root, lane) => spawnLaneWorker(root, lane, spawnDepsForHost(deps)));
1516
1482
  const running = lanes.filter((lane) => (lane.state === "running" ||
1517
1483
  hasFreshAdmissionReservation(lane, now.getTime()) ||
1518
1484
  hasActivePendingResume(lane, now.getTime())) &&
@@ -1699,6 +1665,21 @@ function recordKeepAliveEvent(hqRoot, laneId, summary, event) {
1699
1665
  });
1700
1666
  }
1701
1667
  }
1668
+ function keepAliveAccountSelectionEvidence(selection) {
1669
+ if (selection === undefined)
1670
+ return undefined;
1671
+ return {
1672
+ account: selection.account,
1673
+ selected_at: selection.selected_at,
1674
+ reason: selection.reason,
1675
+ considered: selection.considered.map((row) => ({
1676
+ account: row.account,
1677
+ live_lanes: row.live_lanes,
1678
+ eligible: row.eligible,
1679
+ reason: row.ineligible_reason ?? null,
1680
+ })),
1681
+ };
1682
+ }
1702
1683
  function keepAliveSkipGuidance(laneId, reason) {
1703
1684
  switch (reason) {
1704
1685
  case "missing_opt_in_time":
@@ -1892,10 +1873,12 @@ async function resumeBlockedKeepAlive(hqRoot, lane, questions, summary, deps, no
1892
1873
  return;
1893
1874
  }
1894
1875
  let current = latest;
1876
+ let accountSelection;
1895
1877
  if (current.provider === "claude" && deps.prepareKeepAliveClaudeResume !== undefined) {
1896
1878
  try {
1897
1879
  const prepared = await deps.prepareKeepAliveClaudeResume(current);
1898
1880
  current = prepared.lane;
1881
+ accountSelection = prepared.accountSelection;
1899
1882
  }
1900
1883
  catch (error) {
1901
1884
  markDirty(summary, { kind: "lane", id: lane.lane_id, why: `blocked keep-alive account preparation failed: ${error instanceof Error ? error.message : String(error)}` });
@@ -1917,8 +1900,16 @@ async function resumeBlockedKeepAlive(hqRoot, lane, questions, summary, deps, no
1917
1900
  const queued = casLaneIf(hqRoot, latest.lane_id, (latestRecord) => latestRecord.version === expectedVersion && latestRecord.state === "blocked" && latestRecord.last_envelope?.decision === "blocked" && latestRecord.last_envelope.at === terminalAt, (current) => ({
1918
1901
  ...result.lane,
1919
1902
  version: current.version,
1903
+ pending_resume: result.lane.pending_resume === undefined
1904
+ ? undefined
1905
+ : { ...result.lane.pending_resume, keep_alive: true },
1920
1906
  keep_alive_blocked_resume_streak: priorResumes + 1,
1921
1907
  keep_alive_resume_count: (current.keep_alive_resume_count ?? 0) + 1,
1908
+ keep_alive_history: appendCappedHistory(current.keep_alive_history ?? [], {
1909
+ at: now.toISOString(),
1910
+ resumed_from: "blocked",
1911
+ account: current.account ?? null,
1912
+ }),
1922
1913
  keep_alive_last_resume_at: now.toISOString(),
1923
1914
  status_gate_history: appendCappedHistory(current.status_gate_history ?? [], {
1924
1915
  event: "keep_alive_blocked_resumed",
@@ -1935,6 +1926,9 @@ async function resumeBlockedKeepAlive(hqRoot, lane, questions, summary, deps, no
1935
1926
  at: now.toISOString(),
1936
1927
  attempt: (lane.keep_alive_resume_count ?? 0) + 1,
1937
1928
  resumed_from: "blocked",
1929
+ ...(accountSelection !== undefined
1930
+ ? { account_selection: keepAliveAccountSelectionEvidence(accountSelection) }
1931
+ : {}),
1938
1932
  });
1939
1933
  }
1940
1934
  }
@@ -2390,6 +2384,7 @@ async function resumeKeepAlive(hqRoot, lanes, questions, summary, deps, now, fla
2390
2384
  writeAnsweredQuestions(hqRoot, current, questions);
2391
2385
  }
2392
2386
  let accountSelectionFailure;
2387
+ let accountSelection;
2393
2388
  if (current.provider === "claude" && deps.prepareKeepAliveClaudeResume !== undefined) {
2394
2389
  const optInBeforePreparation = current.keep_alive_opted_in_at;
2395
2390
  let preparedLane;
@@ -2397,6 +2392,7 @@ async function resumeKeepAlive(hqRoot, lanes, questions, summary, deps, now, fla
2397
2392
  const prepared = await deps.prepareKeepAliveClaudeResume(current);
2398
2393
  preparedLane = prepared.lane;
2399
2394
  accountSelectionFailure = prepared.selectionFailure;
2395
+ accountSelection = prepared.accountSelection;
2400
2396
  }
2401
2397
  catch (err) {
2402
2398
  // Account balancing is best-effort for keep-alive. A failure to read
@@ -2514,12 +2510,18 @@ async function resumeKeepAlive(hqRoot, lanes, questions, summary, deps, now, fla
2514
2510
  pending_resume: {
2515
2511
  session_id: currentSessionId,
2516
2512
  trigger: carriesAnsweredQuestions ? "answered_questions" : "resume",
2513
+ keep_alive: true,
2517
2514
  resumed_from: resumedFrom,
2518
2515
  text: resumeText,
2519
2516
  compact: true,
2520
2517
  at: nowIso,
2521
2518
  },
2522
2519
  keep_alive_resume_count: attempt,
2520
+ keep_alive_history: appendCappedHistory(latestLane.keep_alive_history ?? [], {
2521
+ at: nowIso,
2522
+ resumed_from: resumedFrom,
2523
+ account: latestLane.account ?? null,
2524
+ }),
2523
2525
  keep_alive_last_resume_at: nowIso,
2524
2526
  keep_alive_next_eligible_at: nextEligibleAt,
2525
2527
  keep_alive_exhausted_at: undefined,
@@ -2573,6 +2575,9 @@ async function resumeKeepAlive(hqRoot, lanes, questions, summary, deps, now, fla
2573
2575
  attempt,
2574
2576
  next_eligible_at: nextEligibleAt,
2575
2577
  resumed_from: resumedFrom,
2578
+ ...(accountSelection !== undefined
2579
+ ? { account_selection: keepAliveAccountSelectionEvidence(accountSelection) }
2580
+ : {}),
2576
2581
  });
2577
2582
  }
2578
2583
  }
@@ -2604,7 +2609,7 @@ async function resumeAwaiting(hqRoot, lanes, questions, summary, deps, onlyLaneI
2604
2609
  const throttle = await decideAdmission(hqRoot, config.throttle, throttleDeps, deps.cpuBusyFraction, deps.cpuObservation);
2605
2610
  throttleDecision = throttle.decision;
2606
2611
  }
2607
- const spawnFn = deps.spawn ?? ((root, lane) => spawnLaneWorker(root, lane, spawnDepsForHost(deps, lane)));
2612
+ const spawnFn = deps.spawn ?? ((root, lane) => spawnLaneWorker(root, lane, spawnDepsForHost(deps)));
2608
2613
  const nowMs = deps.now().getTime();
2609
2614
  const fallbackAnswerAt = new Date(nowMs).toISOString();
2610
2615
  const answerAtByLane = new Map();
@@ -2676,6 +2681,10 @@ async function resumeAwaiting(hqRoot, lanes, questions, summary, deps, onlyLaneI
2676
2681
  if (!deps.dryRun)
2677
2682
  writeAnsweredQuestions(hqRoot, current, questions);
2678
2683
  if (queueReason !== undefined) {
2684
+ const answerEnv = deps.spawnLaneDeps?.env ?? deps.env ?? process.env;
2685
+ const answerCaller = typeof answerEnv.HQ_LANE_ID === "string"
2686
+ ? answerEnv.HQ_LANE_ID.trim() || "user"
2687
+ : "user";
2679
2688
  const queued = casLaneIf(hqRoot, current.lane_id, (latest) => latest.state === "awaiting_input", (latest) => ({
2680
2689
  ...latest,
2681
2690
  state: "queued",
@@ -2687,6 +2696,7 @@ async function resumeAwaiting(hqRoot, lanes, questions, summary, deps, onlyLaneI
2687
2696
  ? { session_id: latest.provider_session_id }
2688
2697
  : {}),
2689
2698
  trigger: "answered_questions",
2699
+ caller: answerCaller,
2690
2700
  at: answeredAt,
2691
2701
  },
2692
2702
  }), deps.dryRun);
@@ -88,8 +88,6 @@ export interface SpawnLaneDeps {
88
88
  runCompaction?: RunCompactionFn;
89
89
  /** Finds a session transcript under an account home. Injected in tests. */
90
90
  locateTranscript?: (home: string, sessionId: string) => string | null;
91
- /** Pre-read by the async command/reconcile boundary before this sync spawn. */
92
- managerStatusGateEnabled?: boolean;
93
91
  }
94
92
  export interface ProviderLaunch {
95
93
  command: string;
@@ -354,12 +354,13 @@ function archivedResumeMessageText(hqRoot, laneId, messageId) {
354
354
  }
355
355
  function pendingResumeForRetry(hqRoot, lane, sessionId, at) {
356
356
  const last = lane.last_resume;
357
+ const keepAlive = last?.keep_alive === true ? { keep_alive: true } : {};
357
358
  if (last === undefined || last.session_id !== sessionId) {
358
- return { session_id: sessionId, trigger: "resume", resumed_from: "interrupted", at };
359
+ return { session_id: sessionId, trigger: "resume", resumed_from: "interrupted", at, ...keepAlive };
359
360
  }
360
361
  const compact = last.compact === undefined ? {} : { compact: last.compact };
361
362
  if (last.trigger === "answered_questions") {
362
- return { session_id: sessionId, trigger: "answered_questions", at, ...compact };
363
+ return { session_id: sessionId, trigger: "answered_questions", at, ...compact, ...keepAlive };
363
364
  }
364
365
  if (last.trigger === "message") {
365
366
  if (last.message_id === undefined) {
@@ -376,6 +377,7 @@ function pendingResumeForRetry(hqRoot, lane, sessionId, at) {
376
377
  message_id: last.message_id,
377
378
  at,
378
379
  ...compact,
380
+ ...keepAlive,
379
381
  };
380
382
  }
381
383
  return {
@@ -386,6 +388,7 @@ function pendingResumeForRetry(hqRoot, lane, sessionId, at) {
386
388
  ...(last.message_id !== undefined ? { message_id: last.message_id } : {}),
387
389
  at,
388
390
  ...compact,
391
+ ...keepAlive,
389
392
  };
390
393
  }
391
394
  /** Change account only after a provider/runtime account failure, then re-admit. */
@@ -1112,6 +1115,8 @@ export function envelopeRepairPrompt(input) {
1112
1115
  `Missing expected envelope file: ${input.envelopePath}`,
1113
1116
  "The previous envelope is stale and must not be reused or copied into this run's result.",
1114
1117
  "Printing an envelope in the conversation or writing it to last-message.txt is not enough; write the JSON object directly to the expected file.",
1118
+ "If the task is unfinished, write decision: done, story_complete: false, and set story_status and story_complete_reason to describe the work that remains. The repair turn itself is never a reason for blocked.",
1119
+ "This repair turn only writes the missing envelope; do not resume the task.",
1115
1120
  "Schema:",
1116
1121
  JSON.stringify(LANE_ENVELOPE_JSON_SCHEMA, null, 2),
1117
1122
  `Write the corrected JSON object to ${input.envelopePath}, overwriting it. ` +
@@ -2000,10 +2005,30 @@ export function spawnLaneWorker(hqRoot, lane, deps = {}, launch = {}) {
2000
2005
  // Claim running before spawning. Two callers that both observe queued must
2001
2006
  // not both start processes; the CAS loser returns without spawn.
2002
2007
  const nowIso = (deps.now ?? (() => new Date()))().toISOString();
2003
- const statusGateHeartbeat = deps.managerStatusGateEnabled === true
2004
- ? managerStatusGateHeartbeat(hqRoot, existing, laneTurnId, nowIso)
2005
- : undefined;
2008
+ const resumeHistoryTrigger = pendingResume?.keep_alive === true
2009
+ ? "keep_alive"
2010
+ : pendingResume?.trigger === "message" || trigger.kind === "message"
2011
+ ? "message"
2012
+ : pendingResume?.trigger === "resume" || trigger.kind === "resume"
2013
+ ? "resume_cli"
2014
+ : "reconcile";
2015
+ const resumeHistoryCaller = pendingResume?.caller ||
2016
+ (typeof env.HQ_LANE_ID === "string" ? env.HQ_LANE_ID.trim() : "") ||
2017
+ (trigger.kind === "answered_questions" || trigger.kind === "resume" || trigger.kind === "message"
2018
+ ? "user"
2019
+ : undefined);
2020
+ let statusGateHeartbeat;
2021
+ try {
2022
+ statusGateHeartbeat = managerStatusGateHeartbeat(hqRoot, existing, laneTurnId, nowIso);
2023
+ }
2024
+ catch (err) {
2025
+ const code = err instanceof LanesError ? err.error : "manager_status_gate_unavailable";
2026
+ const message = err instanceof Error ? err.message : String(err);
2027
+ process.stderr.write(`hq lanes: manager status gate disabled for ${lane.lane_id}: ${code}: ${message}\n`);
2028
+ statusGateHeartbeat = undefined;
2029
+ }
2006
2030
  let claimed = false;
2031
+ let priorStartRecorded = false;
2007
2032
  const claimedRecord = retryUpdateLane(hqRoot, lane.lane_id, "spawn", (current) => {
2008
2033
  claimed = false;
2009
2034
  if (current.state === "running" && current.account === resolved.id) {
@@ -2015,6 +2040,12 @@ export function spawnLaneWorker(hqRoot, lane, deps = {}, launch = {}) {
2015
2040
  if (!claimableStatesFor(trigger, current).includes(current.state)) {
2016
2041
  throw new LanesError("spawn_state_mismatch", `Cannot mark ${current.lane_id} running from ${current.state}`, { lane_id: current.lane_id, state: current.state, trigger: trigger.kind });
2017
2042
  }
2043
+ // The host report is written in the claim CAS before exec proves that a
2044
+ // worker started. Use the prior spawn manifest's turn id (written only
2045
+ // after detachment is proven), with last_resume for legacy manifests.
2046
+ priorStartRecorded = previousSpawn.turn_id !== undefined ||
2047
+ current.last_resume !== undefined ||
2048
+ (current.resume_history?.length ?? 0) > 0;
2018
2049
  if (failedResumeClaim &&
2019
2050
  (failedResumeExpectedSenior === undefined ||
2020
2051
  !sameSeniorRef(current.senior, failedResumeExpectedSenior))) {
@@ -2102,6 +2133,7 @@ export function spawnLaneWorker(hqRoot, lane, deps = {}, launch = {}) {
2102
2133
  at: nowIso,
2103
2134
  account: resolved.id,
2104
2135
  trigger: pendingResume.trigger,
2136
+ ...(pendingResume.keep_alive === true ? { keep_alive: true } : {}),
2105
2137
  ...(pendingResume.message_id !== undefined ? { message_id: pendingResume.message_id } : {}),
2106
2138
  ...(pendingResume.monitor_event === true ? { monitor_event: true } : {}),
2107
2139
  ...(pendingResume.resumed_from !== undefined ? { resumed_from: pendingResume.resumed_from } : {}),
@@ -2247,6 +2279,16 @@ export function spawnLaneWorker(hqRoot, lane, deps = {}, launch = {}) {
2247
2279
  }
2248
2280
  throw err;
2249
2281
  }
2282
+ if (priorStartRecorded) {
2283
+ retryUpdateLane(hqRoot, lane.lane_id, "spawn", (current) => ({
2284
+ ...current,
2285
+ resume_history: appendCappedHistory(current.resume_history ?? [], {
2286
+ at: nowIso,
2287
+ trigger: resumeHistoryTrigger,
2288
+ ...(resumeHistoryCaller ? { caller: resumeHistoryCaller } : {}),
2289
+ }),
2290
+ }));
2291
+ }
2250
2292
  return laneLaunchResult(readLane(hqRoot, lane.lane_id));
2251
2293
  }
2252
2294
  /**
@@ -2550,6 +2592,16 @@ export function spawnEnvelopeRepair(hqRoot, lane, input, deps = {}) {
2550
2592
  killQuietly(killProcess, workerPid, "worker");
2551
2593
  throw err;
2552
2594
  }
2595
+ const startedAt = (deps.now ?? (() => new Date()))().toISOString();
2596
+ const callerLaneId = typeof env.HQ_LANE_ID === "string" ? env.HQ_LANE_ID.trim() : "";
2597
+ retryUpdateLane(hqRoot, lane.lane_id, "spawn", (current) => ({
2598
+ ...current,
2599
+ resume_history: appendCappedHistory(current.resume_history ?? [], {
2600
+ at: startedAt,
2601
+ trigger: "envelope_repair",
2602
+ ...(callerLaneId ? { caller: callerLaneId } : {}),
2603
+ }),
2604
+ }));
2553
2605
  return workerPid;
2554
2606
  }
2555
2607
  export { PROVIDER_HOME_ENV, PROVIDER_COMMAND };
@@ -751,6 +751,55 @@ export function parseLaneRecord(raw, filePath) {
751
751
  path: filePath,
752
752
  });
753
753
  }
754
+ if (isPlainObject(raw.pending_resume) &&
755
+ raw.pending_resume.caller !== undefined) {
756
+ requireString(raw.pending_resume.caller, "pending_resume.caller");
757
+ }
758
+ if (raw.keep_alive_history !== undefined) {
759
+ if (!Array.isArray(raw.keep_alive_history) || raw.keep_alive_history.length > HISTORY_CAP) {
760
+ throw new LaneSchemaError(`keep_alive_history must be an array capped at ${HISTORY_CAP} entries`, {
761
+ path: filePath,
762
+ field: "keep_alive_history",
763
+ });
764
+ }
765
+ for (const [index, value] of raw.keep_alive_history.entries()) {
766
+ const field = `keep_alive_history[${index}]`;
767
+ if (!isPlainObject(value) ||
768
+ Object.keys(value).some((key) => !["at", "resumed_from", "account"].includes(key)) ||
769
+ (value.resumed_from !== "done" && value.resumed_from !== "interrupted" && value.resumed_from !== "blocked") ||
770
+ (value.account !== null && typeof value.account !== "string")) {
771
+ throw new LaneSchemaError(`${field} must be a keep-alive resume history entry`, {
772
+ path: filePath,
773
+ field,
774
+ value,
775
+ });
776
+ }
777
+ requireTimestamp(value.at, `${field}.at`, filePath);
778
+ }
779
+ }
780
+ if (raw.resume_history !== undefined) {
781
+ if (!Array.isArray(raw.resume_history) || raw.resume_history.length > HISTORY_CAP) {
782
+ throw new LaneSchemaError(`resume_history must be an array capped at ${HISTORY_CAP} entries`, {
783
+ path: filePath,
784
+ field: "resume_history",
785
+ });
786
+ }
787
+ for (const [index, value] of raw.resume_history.entries()) {
788
+ const field = `resume_history[${index}]`;
789
+ if (!isPlainObject(value) ||
790
+ Object.keys(value).some((key) => !["at", "trigger", "caller"].includes(key))) {
791
+ throw new LaneSchemaError(`${field} must be a resume history entry`, {
792
+ path: filePath,
793
+ field,
794
+ value,
795
+ });
796
+ }
797
+ requireTimestamp(value.at, `${field}.at`, filePath);
798
+ requireString(value.trigger, `${field}.trigger`);
799
+ if (value.caller !== undefined)
800
+ requireString(value.caller, `${field}.caller`);
801
+ }
802
+ }
754
803
  if (raw.stop_history !== undefined) {
755
804
  if (!Array.isArray(raw.stop_history)) {
756
805
  throw new LaneSchemaError("stop_history must be an array when present", {
@@ -1141,6 +1190,11 @@ export function mergeLaneRecords(live, sidecars) {
1141
1190
  failed_resume_history: unionHistory(all.map((record) => record.failed_resume_history ?? []), entryIdentity, base.failed_resume_history ?? []),
1142
1191
  }
1143
1192
  : {}),
1193
+ ...(all.some((record) => record.resume_history !== undefined)
1194
+ ? {
1195
+ resume_history: unionHistory(all.map((record) => record.resume_history ?? []), entryIdentity, base.resume_history ?? []),
1196
+ }
1197
+ : {}),
1144
1198
  ...(all.some((record) => record.settings_history !== undefined)
1145
1199
  ? {
1146
1200
  settings_history: unionHistory(all.map((record) => record.settings_history ?? []), entryIdentity, base.settings_history ?? []),
@@ -1,4 +1,4 @@
1
- import type { LaneRecord, LaneStory } from "./types.js";
1
+ import type { LaneRecord, LaneStory, LaneStoryStatus } from "./types.js";
2
2
  /** Board-facing states for the stories a lane currently carries. */
3
3
  export declare const ACTIVE_LANE_STORY_STATUSES: readonly ["in_progress", "blocked", "review"];
4
4
  export declare function storyKey(projectId: string, storyId: string): string;
@@ -14,4 +14,10 @@ export declare function storyWarning(lane: LaneRecord): string | undefined;
14
14
  export declare function withCurrentStory(lane: LaneRecord, projectId: string, storyId: string): LaneRecord;
15
15
  export declare function withLaneStories(lane: LaneRecord, stories: LaneStory[]): LaneRecord;
16
16
  export declare function storyStatusForEnvelope(status: string): "in_progress" | "review";
17
+ /**
18
+ * Resolve the story status requested by a terminal envelope. Already-final
19
+ * carried stories are immutable through the envelope path; senior verification
20
+ * and explicit story commands own those transitions.
21
+ */
22
+ export declare function terminalEnvelopeStoryStatus(lane: LaneRecord, decision: "done" | "blocked" | "ask", envelopeStatus?: string): LaneStoryStatus | undefined;
17
23
  //# sourceMappingURL=stories.d.ts.map
@@ -73,4 +73,20 @@ export function storyStatusForEnvelope(status) {
73
73
  return "review";
74
74
  return "in_progress";
75
75
  }
76
+ /**
77
+ * Resolve the story status requested by a terminal envelope. Already-final
78
+ * carried stories are immutable through the envelope path; senior verification
79
+ * and explicit story commands own those transitions.
80
+ */
81
+ export function terminalEnvelopeStoryStatus(lane, decision, envelopeStatus = "in_progress") {
82
+ const primary = laneStory(lane, lane.project_id, lane.story_id);
83
+ if (primary && (String(primary.status) === "done" || String(primary.status) === "dropped")) {
84
+ return undefined;
85
+ }
86
+ if (decision === "blocked")
87
+ return "blocked";
88
+ if (decision === "done")
89
+ return "review";
90
+ return storyStatusForEnvelope(envelopeStatus);
91
+ }
76
92
  //# sourceMappingURL=stories.js.map
@@ -407,6 +407,19 @@ export interface KeepAliveBlockedHistoryEntry {
407
407
  resume_skipped_reason?: "open_question" | "blocked_grace";
408
408
  resume_queued?: boolean;
409
409
  }
410
+ export interface KeepAliveResumeHistoryEntry {
411
+ at: string;
412
+ resumed_from: "done" | "interrupted" | "blocked";
413
+ account: string | null;
414
+ }
415
+ /** Bounded audit entry for every lane worker start after its first start. */
416
+ export interface LaneResumeHistoryEntry {
417
+ at: string;
418
+ /** Stable trigger class; path-specific starts may use their path name. */
419
+ trigger: string;
420
+ /** HQ lane id for an agent caller, or `user` for an explicit human CLI action. */
421
+ caller?: string;
422
+ }
410
423
  export interface KrFreshnessHistoryEntry {
411
424
  event: "kr_stalled";
412
425
  turn_id: string;
@@ -418,7 +431,13 @@ export interface KrFreshnessHistoryEntry {
418
431
  resume_queued?: boolean;
419
432
  resume_skipped_reason?: "open_question" | "blocked" | "miss_limit" | "no_provider_session";
420
433
  }
421
- export type LaneGateHistoryEntry = ManagerStatusGateHistoryEntry | BlockerFreshnessHistoryEntry | KrFreshnessHistoryEntry | KeepAliveBlockedHistoryEntry;
434
+ export interface EnvelopeRepairContinueHistoryEntry {
435
+ event: "envelope_repair_continued";
436
+ at: string;
437
+ envelope_ref: string;
438
+ session_id: string;
439
+ }
440
+ export type LaneGateHistoryEntry = ManagerStatusGateHistoryEntry | BlockerFreshnessHistoryEntry | KrFreshnessHistoryEntry | EnvelopeRepairContinueHistoryEntry | KeepAliveBlockedHistoryEntry;
422
441
  export interface LaneStory {
423
442
  project_id: string;
424
443
  story_id: string;
@@ -491,6 +510,8 @@ export interface LaneRecord {
491
510
  keep_alive_blocked_reason?: "not_opted_in" | "missing_opt_in_time" | "envelope_before_opt_in" | "interruption_before_opt_in" | "missing_interruption_time" | "retired" | "superseded" | "retired_list_invalid" | "terminal_state";
492
511
  /** Consecutive automatic keep-alive resumes since the last human reset. */
493
512
  keep_alive_resume_count?: number;
513
+ /** Bounded audit trail of automatic keep-alive resumes; human resets never clear it. */
514
+ keep_alive_history?: KeepAliveResumeHistoryEntry[];
494
515
  /** Blocked keep-alive resumes in the current blocked episode. */
495
516
  keep_alive_blocked_resume_streak?: number;
496
517
  /** Time the most recent automatic keep-alive resume was queued. */
@@ -646,6 +667,8 @@ export interface LaneRecord {
646
667
  provider_home_at_spawn?: string;
647
668
  /** Set at spawn when this run continued a recorded session instead of starting fresh. */
648
669
  last_resume?: LaneResumeRecord;
670
+ /** Last 50 lane worker starts after the initial start, including actor provenance. */
671
+ resume_history?: LaneResumeHistoryEntry[];
649
672
  /** Cache counters from the last exited run's result. Nothing else from the payload. */
650
673
  last_run_usage?: LaneRunUsage;
651
674
  /** Completion time paired with last_run_usage; written by the watcher. */
@@ -700,6 +723,8 @@ export interface PendingResume {
700
723
  /** Fresh answered-question launches need no previous session id. */
701
724
  session_id?: string;
702
725
  trigger: "answered_questions" | "message" | "resume";
726
+ /** True when this queued resume was initiated automatically by keep-alive. */
727
+ keep_alive?: true;
703
728
  resumed_from?: "interrupted" | "done" | "blocked" | "failed";
704
729
  from_failed?: true;
705
730
  /** Senior that authorized a failed-lane resume, rechecked at the claim CAS. */
@@ -708,6 +733,8 @@ export interface PendingResume {
708
733
  message_id?: string;
709
734
  /** The text is already framed as untrusted monitor data, not an operator instruction. */
710
735
  monitor_event?: true;
736
+ /** Caller identity captured when a deferred start was requested, when known. */
737
+ caller?: string;
711
738
  /** False when the command passed --no-compact; the deferred spawn honours it. */
712
739
  compact?: boolean;
713
740
  at: string;
@@ -727,6 +754,8 @@ export interface LaneResumeRecord {
727
754
  /** Account the resumed run was bound to. */
728
755
  account?: string;
729
756
  trigger: "answered_questions" | "message" | "resume";
757
+ /** True when this was an automatic keep-alive resume. */
758
+ keep_alive?: true;
730
759
  message_id?: string;
731
760
  monitor_event?: true;
732
761
  /**
@@ -225,7 +225,7 @@ export const LANE_FIELD_WRITERS = {
225
225
  pickup_validation_pending: ["creator", "command", "spawn", "reconcile"],
226
226
  pickup_validation_pending_since: ["creator", "command", "spawn", "reconcile"],
227
227
  priority: ["command"], worker: ["creator", "command"], worker_path: ["creator", "command"], worker_pack: ["creator", "command"], worker_profile: ["creator", "command"], worker_profile_hash: ["creator", "command", "spawn"], registry_worker: ["creator", "command", "spawn"], registry_worker_approval_history: ["command", "watcher", "reconcile"], worker_profile_repin_history: ["command"], worker_profile_ignored: ["creator", "command", "spawn"], worker_policy_delivery: ["command", "spawn"],
228
- keep_alive: ["creator", "command"], keep_alive_standing: ["command"], keep_alive_opted_in_at: ["creator", "command"], keep_alive_blocked_reason: ["command", "reconcile"], keep_alive_resume_count: ["command", "reconcile"], keep_alive_blocked_resume_streak: ["reconcile", "spawn", "watcher"], keep_alive_last_resume_at: ["command", "reconcile"], keep_alive_next_eligible_at: ["command", "reconcile"], keep_alive_exhausted_at: ["command", "reconcile"],
228
+ keep_alive: ["creator", "command"], keep_alive_standing: ["command"], keep_alive_opted_in_at: ["creator", "command"], keep_alive_blocked_reason: ["command", "reconcile"], keep_alive_resume_count: ["command", "reconcile"], keep_alive_history: ["reconcile"], keep_alive_blocked_resume_streak: ["reconcile", "spawn", "watcher"], keep_alive_last_resume_at: ["command", "reconcile"], keep_alive_next_eligible_at: ["command", "reconcile"], keep_alive_exhausted_at: ["command", "reconcile"],
229
229
  provider: ["creator"], account: ["creator", "spawn", "command", "watcher", "reconcile"], account_selection: ["creator", "command"], admission_accounts: ["creator", "command"], account_history: ["command", "spawn", "watcher", "reconcile"], failed_resume_history: ["command"], pending_resume: ["command", "spawn", "watcher", "reconcile"], placement_default: ["creator"], host: ["creator", "command"], origin_host: ["creator"], origin_host_source: ["creator"],
230
230
  version: ["creator"], created_at: ["creator"], updated_at: ["creator"], ended_at: ["reconcile"], envelope_summary: ["reconcile"], attribution_history: ["command"],
231
231
  last_envelope: ["watcher", "reconcile"], model: ["creator", "spawn", "command"], model_source: ["creator", "spawn", "command"], effort: ["creator", "spawn", "command"], effort_source: ["creator", "spawn", "command"], fast_mode: ["spawn"],
@@ -233,7 +233,7 @@ export const LANE_FIELD_WRITERS = {
233
233
  overdue: ["reconcile"], overdue_at: ["reconcile"], last_host_report_at: ["watcher", "spawn", "reconcile"], stalled: ["watcher"], stalled_at: ["watcher"], stalled_reason: ["watcher"],
234
234
  admitted_reason: ["command", "reconcile"], terminal_reason: ["command", "watcher", "spawn", "reconcile"], stop_history: ["command", "reconcile"], settings_history: ["command"], pending_settings: ["command", "spawn"], interrupted_at: ["command", "watcher", "reconcile"], terminal_details: ["command", "spawn", "watcher", "reconcile"],
235
235
  quota_reset_at: ["watcher", "reconcile"], model_quota_marker: ["watcher", "reconcile"], outpost_command_id: ["command"],
236
- provider_session_id: ["watcher", "command", "spawn"], provider_session_id_absent: ["watcher", "command"], provider_home_at_spawn: ["spawn"], last_resume: ["spawn"], last_run_usage: ["watcher"], last_run_usage_at: ["watcher"], envelope_repair: ["watcher", "reconcile"], envelope_validation_errors: ["watcher"], last_progress_event_at: ["watcher"],
236
+ provider_session_id: ["watcher", "command", "spawn"], provider_session_id_absent: ["watcher", "command"], provider_home_at_spawn: ["spawn"], last_resume: ["spawn"], resume_history: ["spawn", "watcher"], last_run_usage: ["watcher"], last_run_usage_at: ["watcher"], envelope_repair: ["watcher", "reconcile"], envelope_validation_errors: ["watcher"], last_progress_event_at: ["watcher"],
237
237
  };
238
238
  function isPlainObject(value) {
239
239
  return value !== null && typeof value === "object" && !Array.isArray(value);
@@ -41,6 +41,7 @@ import { type EnvelopeRepairSpawnInput, type SpawnLaneDeps } from "./spawn.js";
41
41
  /** Spec §5.7: no signal for 20 minutes while the process is still alive. */
42
42
  export declare const SILENCE_MS: number;
43
43
  export declare const WATCHER_POLL_MS = 5000;
44
+ export declare const ENVELOPE_REPAIR_CONTINUE_TEXT = "continue the task; the previous turn ended without an envelope.";
44
45
  /** Synced last_host_report_at cadence. Reconcile dead-host TTL is 20 minutes. */
45
46
  export declare const WATCHER_HEARTBEAT_MS: number;
46
47
  /**
@@ -49,7 +50,7 @@ export declare const WATCHER_HEARTBEAT_MS: number;
49
50
  * once the file would exceed this, the oldest complete lines are dropped.
50
51
  */
51
52
  export declare const WATCHER_OUT_MAX_BYTES: number;
52
- export declare const WATCHER_DECISION_EVENTS: readonly ["liveness_transition", "silence_deadline", "envelope_observed", "envelope_invalid", "envelope_repair", "classification", "terminal_write", "message_recovery", "message_recovery_failed", "message_undeliverable", "mesh_event", "mesh_event_failed"];
53
+ export declare const WATCHER_DECISION_EVENTS: readonly ["liveness_transition", "silence_deadline", "envelope_observed", "envelope_invalid", "envelope_repair", "envelope_repair_continue", "classification", "terminal_write", "message_recovery", "message_recovery_failed", "message_undeliverable", "mesh_event", "mesh_event_failed"];
53
54
  /** Terminal reason when the envelope failed to parse after the repair turn (spec §4). */
54
55
  export { ENVELOPE_INVALID_REASON };
55
56
  export type WatcherDecisionEvent = (typeof WATCHER_DECISION_EVENTS)[number];
@@ -232,11 +233,11 @@ export interface WatcherLoopDeps {
232
233
  /** Deliver a lane-senior escalation through the normal lane-message router. */
233
234
  deliverLaneSeniorMessage?: (seniorLaneId: string, text: string) => Promise<void>;
234
235
  /** Test seam; production reads the company-scoped hq-flags row. */
235
- readManagerStatusGateFlag?: typeof readLanesRegistryFlag;
236
- /** Test seam; production reads the company-scoped hq-flags row. */
237
236
  readBlockerFreshnessFlag?: typeof readLanesRegistryFlag;
238
237
  /** Test seam; production reads the company-scoped hq-flags row. */
239
238
  readKrFreshnessFlag?: typeof readLanesRegistryFlag;
239
+ /** Test seam; production reads the company-scoped hq-flags row. */
240
+ readEnvelopeRepairContinueFlag?: typeof readLanesRegistryFlag;
240
241
  }
241
242
  /**
242
243
  * Wrap one mesh call. A failed emit is logged as `mesh_event_failed` with