@cohortapp/agent-sdk 2.17.0 → 2.18.5

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 (531) hide show
  1. package/.claude/settings.json +18 -0
  2. package/.env.example +18 -5
  3. package/README.md +1 -0
  4. package/bin/maestro.mjs +62 -0
  5. package/docs/guides/billing-console-keys.md +60 -0
  6. package/docs/guides/front-door-session.md +54 -9
  7. package/docs/guides/mac-mini.md +20 -25
  8. package/docs/guides/setup-wizard.md +1 -1
  9. package/docs/runbooks/fleet-rollout.md +156 -0
  10. package/docs/runbooks/mac-mini-bootstrap.md +12 -14
  11. package/lib/action-executor.js +19 -3
  12. package/lib/budget-guard.mjs +279 -3
  13. package/lib/channels/base-adapter.mjs +3 -1
  14. package/lib/channels/contract.mjs +2 -1
  15. package/lib/channels/inbox-item.mjs +8 -0
  16. package/lib/claude-bin.mjs +5 -6
  17. package/lib/cli/doctor-checks.mjs +141 -10
  18. package/lib/cli/global-setup-extras.mjs +5 -1
  19. package/lib/cli/inbox.mjs +100 -15
  20. package/lib/cli/seat-auth.mjs +463 -0
  21. package/lib/cli/session.mjs +80 -12
  22. package/lib/collective/capture-slots.mjs +234 -0
  23. package/lib/collective/capture.mjs +8 -6
  24. package/lib/collective/config.mjs +2 -0
  25. package/lib/collective/global-config.mjs +63 -1
  26. package/lib/collective/loop-guard.mjs +155 -0
  27. package/lib/collective/presence.mjs +142 -5
  28. package/lib/comms/send-gate.mjs +559 -1
  29. package/lib/diagnostics/alerts.mjs +49 -0
  30. package/lib/diagnostics/cadence-output-freshness.mjs +288 -0
  31. package/lib/engine/agents/definitions.mjs +343 -0
  32. package/lib/engine/agents/persist.mjs +275 -0
  33. package/lib/engine/agents/runtime.mjs +748 -0
  34. package/lib/engine/agents/usage.mjs +95 -0
  35. package/lib/engine/auth-status.mjs +139 -0
  36. package/lib/engine/budget.mjs +194 -0
  37. package/lib/engine/cli.mjs +1204 -0
  38. package/lib/engine/commands/index.mjs +269 -0
  39. package/lib/engine/context/budget.mjs +219 -0
  40. package/lib/engine/context/cache.mjs +125 -0
  41. package/lib/engine/context/child-env.mjs +215 -0
  42. package/lib/engine/context/compaction.mjs +342 -0
  43. package/lib/engine/context/images.mjs +90 -0
  44. package/lib/engine/context/instructions.mjs +327 -0
  45. package/lib/engine/context/lazy-instructions.mjs +169 -0
  46. package/lib/engine/context/manager.mjs +182 -0
  47. package/lib/engine/context/real-path.mjs +91 -0
  48. package/lib/engine/context/secret-values.mjs +163 -0
  49. package/lib/engine/context/settings.mjs +274 -0
  50. package/lib/engine/context/stream-input.mjs +159 -0
  51. package/lib/engine/guard.mjs +152 -0
  52. package/lib/engine/hooks.mjs +713 -0
  53. package/lib/engine/loop.mjs +560 -0
  54. package/lib/engine/mcp/client.mjs +254 -0
  55. package/lib/engine/mcp/config.mjs +301 -0
  56. package/lib/engine/mcp/http.mjs +201 -0
  57. package/lib/engine/mcp/index.mjs +146 -0
  58. package/lib/engine/mcp/jsonrpc.mjs +147 -0
  59. package/lib/engine/mcp/naming.mjs +66 -0
  60. package/lib/engine/mcp/resources.mjs +89 -0
  61. package/lib/engine/mcp/results.mjs +133 -0
  62. package/lib/engine/mcp/stdio.mjs +137 -0
  63. package/lib/engine/mcp/supervisor.mjs +116 -0
  64. package/lib/engine/messages.mjs +104 -0
  65. package/lib/engine/output/json.mjs +164 -0
  66. package/lib/engine/output/stream-json.mjs +266 -0
  67. package/lib/engine/permissions.mjs +845 -0
  68. package/lib/engine/process-identity.mjs +164 -0
  69. package/lib/engine/process-tree.mjs +551 -0
  70. package/lib/engine/prompt.mjs +60 -0
  71. package/lib/engine/session/store.mjs +299 -0
  72. package/lib/engine/session-runtime/args.mjs +97 -0
  73. package/lib/engine/session-runtime/host.mjs +143 -0
  74. package/lib/engine/session-runtime/inbox.mjs +122 -0
  75. package/lib/engine/session-runtime/notifications.mjs +129 -0
  76. package/lib/engine/session-runtime/registry.mjs +328 -0
  77. package/lib/engine/session-runtime/runner.mjs +344 -0
  78. package/lib/engine/session-runtime/socket.mjs +212 -0
  79. package/lib/engine/session-runtime/wakeup.mjs +115 -0
  80. package/lib/engine/skills/index.mjs +321 -0
  81. package/lib/engine/tools/bash-background.mjs +533 -0
  82. package/lib/engine/tools/bash.mjs +216 -0
  83. package/lib/engine/tools/edit.mjs +97 -0
  84. package/lib/engine/tools/glob.mjs +81 -0
  85. package/lib/engine/tools/grep.mjs +224 -0
  86. package/lib/engine/tools/index.mjs +84 -0
  87. package/lib/engine/tools/list-agents.mjs +32 -0
  88. package/lib/engine/tools/ls.mjs +127 -0
  89. package/lib/engine/tools/monitor.mjs +82 -0
  90. package/lib/engine/tools/notebook-edit.mjs +218 -0
  91. package/lib/engine/tools/read.mjs +103 -0
  92. package/lib/engine/tools/schedule-wakeup.mjs +45 -0
  93. package/lib/engine/tools/schema.mjs +144 -0
  94. package/lib/engine/tools/send-message.mjs +77 -0
  95. package/lib/engine/tools/session.mjs +70 -0
  96. package/lib/engine/tools/todo.mjs +144 -0
  97. package/lib/engine/tools/toolsearch.mjs +217 -0
  98. package/lib/engine/tools/walk.mjs +193 -0
  99. package/lib/engine/tools/web-switch.mjs +31 -0
  100. package/lib/engine/tools/webfetch-html.mjs +387 -0
  101. package/lib/engine/tools/webfetch-net.mjs +340 -0
  102. package/lib/engine/tools/webfetch.mjs +198 -0
  103. package/lib/engine/tools/websearch.mjs +91 -0
  104. package/lib/engine/tools/workflow.mjs +95 -0
  105. package/lib/engine/tools/write.mjs +76 -0
  106. package/lib/engine/tui/line-editor.mjs +137 -0
  107. package/lib/engine/tui/render.mjs +86 -0
  108. package/lib/engine/tui/tui.mjs +274 -0
  109. package/lib/engine/wire/anthropic-messages.mjs +263 -0
  110. package/lib/engine/wire/effort.mjs +36 -0
  111. package/lib/engine/wire/errors.mjs +496 -0
  112. package/lib/engine/wire/http.mjs +441 -0
  113. package/lib/engine/wire/index.mjs +76 -0
  114. package/lib/engine/wire/openai-chat.mjs +332 -0
  115. package/lib/engine/wire/prompt-cache.mjs +79 -0
  116. package/lib/engine/wire/search.mjs +140 -0
  117. package/lib/engine/wire/sse.mjs +114 -0
  118. package/lib/engine/wire/stall.mjs +349 -0
  119. package/lib/engine/wire/token-provider.mjs +175 -0
  120. package/lib/engine/wire/usage.mjs +192 -0
  121. package/lib/engine/workflow/host.mjs +524 -0
  122. package/lib/engine/workflow/journal.mjs +188 -0
  123. package/lib/engine/workflow/json-schema.mjs +171 -0
  124. package/lib/engine/workflow/meta.mjs +329 -0
  125. package/lib/engine/workflow/notifications.mjs +52 -0
  126. package/lib/engine/workflow/runtime.mjs +447 -0
  127. package/lib/engine/workflow/sandbox.mjs +534 -0
  128. package/lib/engine/workflow/worker.mjs +141 -0
  129. package/lib/engine/workflow/worktree.mjs +74 -0
  130. package/lib/execution/disposition.mjs +1 -1
  131. package/lib/execution/intake.mjs +10 -0
  132. package/lib/execution/surface-policy.mjs +15 -0
  133. package/lib/learning/curator.mjs +8 -6
  134. package/lib/learning/reflect.mjs +8 -6
  135. package/lib/model-router/catalog/cohort.yaml +137 -0
  136. package/lib/model-router/catalog.mjs +118 -1
  137. package/lib/model-router/failover.mjs +67 -16
  138. package/lib/model-router/llm-task.mjs +39 -3
  139. package/lib/model-router/resolve.mjs +89 -3
  140. package/lib/model-router/spawn.mjs +46 -47
  141. package/lib/model-router/taxonomy.mjs +126 -4
  142. package/lib/org/cost-sync.mjs +141 -11
  143. package/lib/org/inbound/broadcast.mjs +289 -0
  144. package/lib/org/inbound/collective.mjs +375 -0
  145. package/lib/org/inbound/directedness.mjs +96 -8
  146. package/lib/org/inbound/facts.mjs +78 -2
  147. package/lib/org/inbound/project.mjs +22 -0
  148. package/lib/org/inbound/surfaces.mjs +14 -0
  149. package/lib/org/llm-token.mjs +879 -0
  150. package/lib/org/mesh.mjs +61 -0
  151. package/lib/org/messaging.mjs +3 -1
  152. package/lib/org/protocol.checksum +1 -1
  153. package/lib/org/protocol.mjs +15 -0
  154. package/lib/org/quota.mjs +520 -0
  155. package/lib/org/tool-surface.mjs +104 -16
  156. package/lib/org/ui-parity.mjs +16 -1
  157. package/lib/org/work-ledger.mjs +37 -6
  158. package/lib/rate-guard.mjs +114 -1
  159. package/lib/resource-governor.mjs +41 -6
  160. package/lib/runtime/adapter.mjs +833 -0
  161. package/lib/runtime/child-env.mjs +191 -0
  162. package/lib/runtime/legacy-shell-guard.mjs +97 -0
  163. package/lib/runtime/seat-engine.mjs +162 -0
  164. package/lib/session/ask-ledger.mjs +271 -0
  165. package/lib/session/current-work.mjs +676 -0
  166. package/lib/session/feed-core.mjs +40 -3
  167. package/lib/session/launch-args.mjs +56 -4
  168. package/lib/session/status-summary.mjs +26 -9
  169. package/lib/session/upgrade-notice.mjs +42 -0
  170. package/lib/setup/claude-probe.mjs +117 -13
  171. package/lib/setup/enrich.mjs +13 -10
  172. package/lib/setup/sections/model.mjs +39 -13
  173. package/lib/telemetry/collect.mjs +229 -11
  174. package/lib/upgrade/ignored-drift.mjs +105 -0
  175. package/lib/voice/post-call-brief.mjs +30 -17
  176. package/package.json +13 -3
  177. package/plugins/maestro-skills/skills/board-work.md +5 -0
  178. package/plugins/maestro-skills/skills/inbound-triage.md +56 -15
  179. package/plugins/maestro-skills/skills/main-session.md +18 -7
  180. package/scaffold/config/collective.yaml +7 -0
  181. package/scripts/ci/check-durable-write-seam.mjs +3 -1
  182. package/scripts/ci/check-tarball-fidelity.mjs +126 -2
  183. package/scripts/ci/run-tests.mjs +47 -19
  184. package/scripts/cohort-llm/api-key-helper.mjs +92 -0
  185. package/scripts/collective/hook-runner.mjs +142 -19
  186. package/scripts/continuous-monitor.sh +13 -0
  187. package/scripts/cost/track-claude-usage.mjs +15 -0
  188. package/scripts/daemon/agent-daemon.mjs +408 -20
  189. package/scripts/daemon/assurance.mjs +48 -12
  190. package/scripts/daemon/cadence-consumer.mjs +218 -68
  191. package/scripts/daemon/cadence-handlers.mjs +73 -4
  192. package/scripts/daemon/classifier.mjs +75 -26
  193. package/scripts/daemon/context-compiler.mjs +51 -37
  194. package/scripts/daemon/deliver.mjs +30 -1
  195. package/scripts/daemon/dispatcher.mjs +595 -149
  196. package/scripts/daemon/health.mjs +14 -1
  197. package/scripts/daemon/maestro-daemon.mjs +11 -0
  198. package/scripts/daemon/prompt-builder.mjs +24 -0
  199. package/scripts/daemon/responder.mjs +246 -79
  200. package/scripts/daemon/sdk-version.mjs +98 -16
  201. package/scripts/eval/probe-gateway.mjs +635 -0
  202. package/scripts/eval/replay/extract.mjs +270 -0
  203. package/scripts/eval/replay/grade.mjs +260 -0
  204. package/scripts/eval/replay/lib/config.mjs +50 -0
  205. package/scripts/eval/replay/lib/effects.mjs +65 -0
  206. package/scripts/eval/replay/lib/fixture.mjs +188 -0
  207. package/scripts/eval/replay/lib/judge.mjs +72 -0
  208. package/scripts/eval/replay/lib/redact.mjs +136 -0
  209. package/scripts/eval/replay/lib/sandbox.mjs +170 -0
  210. package/scripts/eval/replay/lib/schema-check.mjs +63 -0
  211. package/scripts/eval/replay/lib/transcript.mjs +76 -0
  212. package/scripts/eval/replay/mcp-replay-stub.mjs +101 -0
  213. package/scripts/eval/replay/report.mjs +185 -0
  214. package/scripts/eval/replay/run.mjs +404 -0
  215. package/scripts/fleet/rollout.mjs +1151 -0
  216. package/scripts/hooks/pre-send-audit.sh +36 -245
  217. package/scripts/hooks/pre-write-yaml-validate.mjs +275 -0
  218. package/scripts/hooks/validate-state-yaml.sh +190 -0
  219. package/scripts/huddle/huddle-llm.mjs +361 -0
  220. package/scripts/huddle/huddle-server.mjs +46 -121
  221. package/scripts/local-triggers/autoupdate.sh +465 -81
  222. package/scripts/local-triggers/run-trigger.sh +13 -0
  223. package/scripts/maintenance/pin-integrity.mjs +364 -0
  224. package/scripts/poll-slack-events.sh +41 -9
  225. package/scripts/poller/slack-socket-mode.mjs +28 -3
  226. package/scripts/session/supervisor.mjs +80 -13
  227. package/scripts/spawn-session.sh +13 -0
  228. package/bin/maestro.test.mjs +0 -1574
  229. package/lib/action-executor.test.mjs +0 -871
  230. package/lib/archetype.test.mjs +0 -132
  231. package/lib/assurance/plan-note.test.mjs +0 -234
  232. package/lib/assurance/room-budget.test.mjs +0 -486
  233. package/lib/assurance/tier.test.mjs +0 -174
  234. package/lib/autonomy.test.mjs +0 -66
  235. package/lib/backlog.test.mjs +0 -302
  236. package/lib/backup/policy.test.mjs +0 -305
  237. package/lib/budget-escalate.test.mjs +0 -232
  238. package/lib/budget-guard.envelope.test.mjs +0 -476
  239. package/lib/budget-guard.test.mjs +0 -427
  240. package/lib/cadence-bus-requeue.test.mjs +0 -83
  241. package/lib/cadence-bus-schedule.test.mjs +0 -194
  242. package/lib/cadence-bus.test.mjs +0 -720
  243. package/lib/cadences.test.mjs +0 -230
  244. package/lib/capability/inventory.test.mjs +0 -232
  245. package/lib/capability.test.mjs +0 -78
  246. package/lib/channels/base-adapter.test.mjs +0 -590
  247. package/lib/channels/channels.test.mjs +0 -371
  248. package/lib/channels/contract.test.mjs +0 -162
  249. package/lib/channels/inbox-item.test.mjs +0 -368
  250. package/lib/channels/orgmail/adapter.test.mjs +0 -448
  251. package/lib/channels/pairing.test.mjs +0 -270
  252. package/lib/channels/repeat-suppressor.test.mjs +0 -134
  253. package/lib/channels/slack-adapter.test.mjs +0 -212
  254. package/lib/channels/telegram-adapter.test.mjs +0 -306
  255. package/lib/channels/voice/adapter.test.mjs +0 -278
  256. package/lib/channels/whatsapp/adapter-baileys.test.mjs +0 -359
  257. package/lib/channels/whatsapp/baileys-typing.test.mjs +0 -154
  258. package/lib/charter.test.mjs +0 -89
  259. package/lib/claude-bin.test.mjs +0 -131
  260. package/lib/cli/board.test.mjs +0 -227
  261. package/lib/cli/design.test.mjs +0 -270
  262. package/lib/cli/doctor-checks.test.mjs +0 -336
  263. package/lib/cli/global-setup-extras.test.mjs +0 -462
  264. package/lib/cli/inbox.test.mjs +0 -230
  265. package/lib/cli/session-ack.test.mjs +0 -63
  266. package/lib/cli/session.test.mjs +0 -613
  267. package/lib/collective/capture.test.mjs +0 -121
  268. package/lib/collective/cards.test.mjs +0 -114
  269. package/lib/collective/config.test.mjs +0 -123
  270. package/lib/collective/global-config.test.mjs +0 -220
  271. package/lib/collective/global-skills.test.mjs +0 -126
  272. package/lib/collective/presence.test.mjs +0 -95
  273. package/lib/collective/recall.test.mjs +0 -116
  274. package/lib/collective/vendor-skills.test.mjs +0 -306
  275. package/lib/comms/send-gate.test.mjs +0 -770
  276. package/lib/comms.test.mjs +0 -41
  277. package/lib/context/budget.test.mjs +0 -252
  278. package/lib/context/history-scope.test.mjs +0 -79
  279. package/lib/cost/ledger-row.test.mjs +0 -183
  280. package/lib/design/design-md.test.mjs +0 -318
  281. package/lib/design/fixtures/DESIGN.golden.md +0 -238
  282. package/lib/design/fixtures/PRODUCT.golden.md +0 -67
  283. package/lib/design/fixtures/foundation.json +0 -133
  284. package/lib/design/refresh-gate.test.mjs +0 -144
  285. package/lib/design/write.test.mjs +0 -241
  286. package/lib/diagnostics/alerts.test.mjs +0 -318
  287. package/lib/diagnostics/backup-freshness.test.mjs +0 -185
  288. package/lib/diagnostics/counters.test.mjs +0 -206
  289. package/lib/diagnostics/events.test.mjs +0 -290
  290. package/lib/diagnostics/otel.test.mjs +0 -196
  291. package/lib/diagnostics/trace.test.mjs +0 -251
  292. package/lib/env-compat.test.mjs +0 -104
  293. package/lib/execution/disposition.test.mjs +0 -553
  294. package/lib/execution/drive.test.mjs +0 -270
  295. package/lib/execution/effects.test.mjs +0 -344
  296. package/lib/execution/intake.test.mjs +0 -389
  297. package/lib/execution/journal.test.mjs +0 -261
  298. package/lib/execution/match.test.mjs +0 -235
  299. package/lib/execution/pipeline.test.mjs +0 -392
  300. package/lib/execution/route.test.mjs +0 -186
  301. package/lib/execution/surface-policy.test.mjs +0 -162
  302. package/lib/fs-atomic.test.mjs +0 -72
  303. package/lib/fs-ownership.test.mjs +0 -158
  304. package/lib/goals/admission.test.mjs +0 -164
  305. package/lib/goals/classify.test.mjs +0 -167
  306. package/lib/goals/collaborate.test.mjs +0 -336
  307. package/lib/goals/gaps.test.mjs +0 -284
  308. package/lib/goals/loop.test.mjs +0 -845
  309. package/lib/hooks/bus.test.mjs +0 -387
  310. package/lib/identity/persona.test.mjs +0 -142
  311. package/lib/kpi-sensors.test.mjs +0 -278
  312. package/lib/kpi.test.mjs +0 -244
  313. package/lib/learning/config.test.mjs +0 -75
  314. package/lib/learning/counters.test.mjs +0 -69
  315. package/lib/learning/curator-consolidate.test.mjs +0 -238
  316. package/lib/learning/curator.test.mjs +0 -106
  317. package/lib/learning/reflect.test.mjs +0 -0
  318. package/lib/learning/session-index.test.mjs +0 -125
  319. package/lib/learning/skill-writer.test.mjs +0 -210
  320. package/lib/mandate/audit.test.mjs +0 -195
  321. package/lib/mandate/contract.test.mjs +0 -185
  322. package/lib/mandate/derive.test.mjs +0 -274
  323. package/lib/mandate/model.test.mjs +0 -164
  324. package/lib/mandate/refresh.test.mjs +0 -389
  325. package/lib/mcp/server.test.mjs +0 -426
  326. package/lib/model-router/auth-profiles.test.mjs +0 -580
  327. package/lib/model-router/catalog.test.mjs +0 -385
  328. package/lib/model-router/economics.test.mjs +0 -438
  329. package/lib/model-router/failover.test.mjs +0 -439
  330. package/lib/model-router/health.test.mjs +0 -338
  331. package/lib/model-router/integration-coverage.test.mjs +0 -831
  332. package/lib/model-router/integration.test.mjs +0 -564
  333. package/lib/model-router/ledger.test.mjs +0 -415
  334. package/lib/model-router/llm-task.test.mjs +0 -392
  335. package/lib/model-router/org-credentials.test.mjs +0 -265
  336. package/lib/model-router/pricing-refresh.test.mjs +0 -286
  337. package/lib/model-router/reconcile.test.mjs +0 -316
  338. package/lib/model-router/repair.test.mjs +0 -180
  339. package/lib/model-router/spawn.test.mjs +0 -446
  340. package/lib/model-router/taxonomy.test.mjs +0 -410
  341. package/lib/model-router.test.mjs +0 -1207
  342. package/lib/org/activity.test.mjs +0 -134
  343. package/lib/org/approvals.test.mjs +0 -216
  344. package/lib/org/awareness.test.mjs +0 -159
  345. package/lib/org/board-mine-cache.test.mjs +0 -53
  346. package/lib/org/board.test.mjs +0 -187
  347. package/lib/org/bootstrap-context.test.mjs +0 -153
  348. package/lib/org/client.test.mjs +0 -1206
  349. package/lib/org/cohort-client.test.mjs +0 -126
  350. package/lib/org/cost-sync.test.mjs +0 -153
  351. package/lib/org/doctor.test.mjs +0 -346
  352. package/lib/org/engagement-ledger.test.mjs +0 -112
  353. package/lib/org/engagement.test.mjs +0 -739
  354. package/lib/org/handoff.test.mjs +0 -269
  355. package/lib/org/inbound/directedness.test.mjs +0 -668
  356. package/lib/org/inbound/facts.test.mjs +0 -471
  357. package/lib/org/inbound/hydrate.test.mjs +0 -908
  358. package/lib/org/inbound/index.test.mjs +0 -429
  359. package/lib/org/inbound/project.test.mjs +0 -287
  360. package/lib/org/integration-tools.test.mjs +0 -160
  361. package/lib/org/keys.test.mjs +0 -92
  362. package/lib/org/knowledge.test.mjs +0 -326
  363. package/lib/org/leases.test.mjs +0 -235
  364. package/lib/org/mesh-directives.test.mjs +0 -110
  365. package/lib/org/mesh-integration.test.mjs +0 -127
  366. package/lib/org/mesh.test.mjs +0 -400
  367. package/lib/org/messaging.test.mjs +0 -471
  368. package/lib/org/param-contract.test.mjs +0 -477
  369. package/lib/org/policy.test.mjs +0 -237
  370. package/lib/org/protocol.checksum.test.mjs +0 -90
  371. package/lib/org/protocol.test.mjs +0 -323
  372. package/lib/org/push.test.mjs +0 -792
  373. package/lib/org/registry.test.mjs +0 -100
  374. package/lib/org/resource-tools.test.mjs +0 -361
  375. package/lib/org/tool-access.test.mjs +0 -144
  376. package/lib/org/tool-surface-integration.test.mjs +0 -120
  377. package/lib/org/tool-surface.test.mjs +0 -1268
  378. package/lib/org/typing.test.mjs +0 -291
  379. package/lib/org/ui-parity.test.mjs +0 -560
  380. package/lib/org/verify.test.mjs +0 -194
  381. package/lib/org/work-ledger.test.mjs +0 -273
  382. package/lib/plan/adoption-e2e.test.mjs +0 -366
  383. package/lib/plan/budget-enforcement.test.mjs +0 -400
  384. package/lib/plan/compile.test.mjs +0 -382
  385. package/lib/plan/emit.test.mjs +0 -269
  386. package/lib/plan/explain.test.mjs +0 -188
  387. package/lib/prompts/parallelism.test.mjs +0 -177
  388. package/lib/rag/rag.test.mjs +0 -505
  389. package/lib/rate-guard.test.mjs +0 -272
  390. package/lib/reactive-gate.test.mjs +0 -57
  391. package/lib/render.test.mjs +0 -68
  392. package/lib/resource-governor.test.mjs +0 -488
  393. package/lib/scheduling/dynamic-jobs.test.mjs +0 -344
  394. package/lib/scheduling/jitter.test.mjs +0 -140
  395. package/lib/secrets/broker.test.mjs +0 -280
  396. package/lib/secrets/providers.test.mjs +0 -274
  397. package/lib/security/audit-engine.test.mjs +0 -424
  398. package/lib/security/coerce-args.test.mjs +0 -281
  399. package/lib/security/dangerous-tools.test.mjs +0 -68
  400. package/lib/security/external-content.test.mjs +0 -84
  401. package/lib/security/redact.test.mjs +0 -441
  402. package/lib/security/secret-equal.test.mjs +0 -55
  403. package/lib/session/config.test.mjs +0 -92
  404. package/lib/session/feed-core.test.mjs +0 -198
  405. package/lib/session/first-run.test.mjs +0 -121
  406. package/lib/session/frontdoor.test.mjs +0 -205
  407. package/lib/session/handoffs.test.mjs +0 -183
  408. package/lib/session/identity.test.mjs +0 -180
  409. package/lib/session/inbox-claims.test.mjs +0 -286
  410. package/lib/session/launch-args.test.mjs +0 -157
  411. package/lib/session/liveness.test.mjs +0 -100
  412. package/lib/session/status-summary.test.mjs +0 -118
  413. package/lib/session-permissions.test.mjs +0 -120
  414. package/lib/setup/claude-probe.test.mjs +0 -187
  415. package/lib/setup/completeness.test.mjs +0 -110
  416. package/lib/setup/context-pack.test.mjs +0 -89
  417. package/lib/setup/enrich.test.mjs +0 -115
  418. package/lib/setup/enroll-from-cohort.test.mjs +0 -300
  419. package/lib/setup/integration.test.mjs +0 -162
  420. package/lib/setup/io.test.mjs +0 -77
  421. package/lib/setup/runner.test.mjs +0 -132
  422. package/lib/setup/sections/identity.test.mjs +0 -234
  423. package/lib/setup/sections/inventory.test.mjs +0 -198
  424. package/lib/setup/sections/learning.test.mjs +0 -81
  425. package/lib/setup/sections/mandate.test.mjs +0 -388
  426. package/lib/setup/sections/messaging.test.mjs +0 -127
  427. package/lib/setup/sections/model.test.mjs +0 -240
  428. package/lib/setup/sections/org.test.mjs +0 -346
  429. package/lib/setup/sections/orgmail.test.mjs +0 -118
  430. package/lib/setup/sections/recovery.test.mjs +0 -98
  431. package/lib/setup/sections/subagents.test.mjs +0 -429
  432. package/lib/setup/sections/verify.test.mjs +0 -175
  433. package/lib/setup/sot.test.mjs +0 -81
  434. package/lib/setup/state.test.mjs +0 -115
  435. package/lib/singleton.test.mjs +0 -151
  436. package/lib/subagents/cli.test.mjs +0 -389
  437. package/lib/subagents/client.test.mjs +0 -309
  438. package/lib/subagents/gap.test.mjs +0 -234
  439. package/lib/subagents/lock.test.mjs +0 -248
  440. package/lib/subagents/manifest.test.mjs +0 -175
  441. package/lib/subagents/refs.test.mjs +0 -204
  442. package/lib/subagents/resolve.test.mjs +0 -422
  443. package/lib/subagents/schema.test.mjs +0 -328
  444. package/lib/telemetry/alerts.test.mjs +0 -109
  445. package/lib/telemetry/collect.test.mjs +0 -1274
  446. package/lib/tool-definitions-integration.test.mjs +0 -83
  447. package/lib/tool-definitions.test.mjs +0 -437
  448. package/lib/upgrade/global-refresh.test.mjs +0 -65
  449. package/lib/upgrade/launchd-reconcile.test.mjs +0 -272
  450. package/lib/upgrade/post-steps.test.mjs +0 -200
  451. package/lib/upgrade/verify.test.mjs +0 -164
  452. package/lib/util/fetch-timeout.test.mjs +0 -202
  453. package/lib/util/reconnect.test.mjs +0 -369
  454. package/lib/util/unhandled.test.mjs +0 -216
  455. package/lib/voice/outbound.test.mjs +0 -69
  456. package/lib/voice/session-rotation.test.mjs +0 -114
  457. package/lib/voice/stt.test.mjs +0 -226
  458. package/lib/voice/voice.test.mjs +0 -990
  459. package/scripts/cadence/enqueue-cadence-tick.test.mjs +0 -187
  460. package/scripts/ci/check-docs-accuracy.test.mjs +0 -409
  461. package/scripts/ci/check-durable-write-seam.test.mjs +0 -90
  462. package/scripts/ci/check-no-build-artifacts.test.mjs +0 -71
  463. package/scripts/ci/check-no-residual-identity.test.mjs +0 -202
  464. package/scripts/ci/check-skill-packs.test.mjs +0 -495
  465. package/scripts/ci/check-subagent-frontmatter.test.mjs +0 -124
  466. package/scripts/ci/check.test.mjs +0 -194
  467. package/scripts/ci/conformance-org-api.test.mjs +0 -425
  468. package/scripts/cloud-relay/voice/relay-identity.test.mjs +0 -96
  469. package/scripts/collective/hook-runner.test.mjs +0 -173
  470. package/scripts/cost/fleet-digest.test.mjs +0 -207
  471. package/scripts/cost/track-claude-usage-pricing.test.mjs +0 -183
  472. package/scripts/cost/track-claude-usage.test.mjs +0 -148
  473. package/scripts/daemon/agent-daemon-board-mine.test.mjs +0 -96
  474. package/scripts/daemon/agent-daemon-design.test.mjs +0 -238
  475. package/scripts/daemon/agent-daemon-frontdoor.test.mjs +0 -60
  476. package/scripts/daemon/agent-daemon.test.mjs +0 -995
  477. package/scripts/daemon/assurance-e2e.test.mjs +0 -613
  478. package/scripts/daemon/assurance.test.mjs +0 -1791
  479. package/scripts/daemon/board-mirror.test.mjs +0 -165
  480. package/scripts/daemon/cadence-consumer-frontdoor.test.mjs +0 -393
  481. package/scripts/daemon/cadence-consumer-governance.test.mjs +0 -276
  482. package/scripts/daemon/cadence-consumer.test.mjs +0 -776
  483. package/scripts/daemon/cadence-handlers.test.mjs +0 -837
  484. package/scripts/daemon/classifier-identity.test.mjs +0 -137
  485. package/scripts/daemon/classifier.test.mjs +0 -266
  486. package/scripts/daemon/classify-kind.test.mjs +0 -40
  487. package/scripts/daemon/context-compiler.test.mjs +0 -406
  488. package/scripts/daemon/deliver.test.mjs +0 -564
  489. package/scripts/daemon/dispatcher-cooldown.test.mjs +0 -122
  490. package/scripts/daemon/dispatcher-governance.test.mjs +0 -1013
  491. package/scripts/daemon/dispatcher-resume.test.mjs +0 -166
  492. package/scripts/daemon/dispatcher-session-continuity.test.mjs +0 -365
  493. package/scripts/daemon/execution-ladder.test.mjs +0 -470
  494. package/scripts/daemon/goal-steward-cadence.test.mjs +0 -312
  495. package/scripts/daemon/inbox-deferral-session.test.mjs +0 -49
  496. package/scripts/daemon/inbox-deferral.test.mjs +0 -336
  497. package/scripts/daemon/inbox-wake.test.mjs +0 -199
  498. package/scripts/daemon/integration.test.mjs +0 -149
  499. package/scripts/daemon/lib/self-echo.test.mjs +0 -153
  500. package/scripts/daemon/lib/session-router.test.mjs +0 -554
  501. package/scripts/daemon/prompt-builder-preamble.test.mjs +0 -210
  502. package/scripts/daemon/prompt-builder.test.mjs +0 -556
  503. package/scripts/daemon/responder-cost.test.mjs +0 -68
  504. package/scripts/daemon/responder-history.test.mjs +0 -221
  505. package/scripts/daemon/sdk-version.test.mjs +0 -31
  506. package/scripts/daemon/session-lock.test.mjs +0 -252
  507. package/scripts/daemon/session-outcomes.test.mjs +0 -533
  508. package/scripts/daemon/typing-registry.test.mjs +0 -102
  509. package/scripts/hooks/pre-send-audit.test.mjs +0 -354
  510. package/scripts/huddle/huddle-prompt.test.mjs +0 -176
  511. package/scripts/local-triggers/autoupdate.test.mjs +0 -518
  512. package/scripts/local-triggers/generate-plists.test.mjs +0 -456
  513. package/scripts/media-generation/brand-clause.test.mjs +0 -135
  514. package/scripts/org/send-orgmail.first-contact.test.mjs +0 -102
  515. package/scripts/poller/inbox-privilege-injection.test.mjs +0 -167
  516. package/scripts/poller/inbox-scan-poller.test.mjs +0 -295
  517. package/scripts/poller/lib/cloud-relay-dedup.test.mjs +0 -133
  518. package/scripts/poller/slack-socket-mode.test.mjs +0 -805
  519. package/scripts/poller-launchd/install.test.mjs +0 -243
  520. package/scripts/restore-from-backup.test.mjs +0 -181
  521. package/scripts/session/feed.test.mjs +0 -196
  522. package/scripts/session/supervisor-sh.test.mjs +0 -218
  523. package/scripts/session/supervisor.test.mjs +0 -482
  524. package/scripts/setup/configure-macos.test.mjs +0 -306
  525. package/scripts/setup/gen-subagent-manifest.test.mjs +0 -124
  526. package/scripts/setup/generate-agent-package-json.test.mjs +0 -143
  527. package/scripts/setup/generate-capability.test.mjs +0 -134
  528. package/scripts/setup/init-agent.test.mjs +0 -370
  529. package/scripts/setup/init-skill-marketplace.test.mjs +0 -193
  530. package/scripts/vendor/sync-skill-packs.test.mjs +0 -103
  531. package/scripts/watchdog/memory-watchdog.test.mjs +0 -64
@@ -1,235 +0,0 @@
1
- /**
2
- * lib/execution/match.test.mjs — binding events to the compiled plan.
3
- *
4
- * The negatives here are the ones that keep the plan honest: a suspended
5
- * obligation must NOT bind (and must not look like a coverage gap either), an
6
- * unparseable predicate must fail closed rather than matching everything, and an
7
- * event genuinely nobody planned for must produce a proposal the schema accepts.
8
- */
9
-
10
- import test from "node:test";
11
- import assert from "node:assert/strict";
12
-
13
- import {
14
- matchObligation,
15
- proposeReact,
16
- reactObligations,
17
- kindAliases,
18
- evalPredicate,
19
- predicateView,
20
- loadReactObligations,
21
- PREDICATE_FIELDS,
22
- } from "./match.mjs";
23
- import { validateObligation } from "../plan/schema.mjs";
24
- import { fromInboxItem } from "./intake.mjs";
25
- import { decide } from "./disposition.mjs";
26
-
27
- const ME = "mem_self";
28
-
29
- function ob(over = {}) {
30
- return {
31
- key: "react.task.assigned",
32
- kind: "REACT",
33
- source: { origin: "react-standard" },
34
- trigger: { topic: "task", kind: "assigned", predicate: null },
35
- status: "active",
36
- uses: [],
37
- allowed_tools: [],
38
- action_classes: [],
39
- offline_safe: true,
40
- ...over,
41
- };
42
- }
43
-
44
- function boardEvent() {
45
- return {
46
- candidate: {
47
- seq: 90,
48
- family: "board",
49
- kind: "assigned",
50
- entityId: "task_1",
51
- actor: "mem_peer",
52
- at: "2026-08-11T09:00:00.000Z",
53
- topic: "task",
54
- surfaces: ["task_assigned"],
55
- ids: { taskId: "task_1", channelId: "chan_1" },
56
- directTo: [ME],
57
- payload: {},
58
- },
59
- verdict: { directed: true, surface: "task_assigned", reason: "assignee" },
60
- };
61
- }
62
-
63
- // ---------------------------------------------------------------------------
64
- // aliases + predicates
65
- // ---------------------------------------------------------------------------
66
-
67
- test("kind aliases bridge chain coordinates and surface names", () => {
68
- const a = kindAliases({ kind: "assigned" }, { surface: "task_assigned" });
69
- assert.ok(a.includes("assigned"));
70
- assert.ok(a.includes("task_assigned"));
71
- // The reconstructed-from-disk form must reach the same obligation.
72
- const b = kindAliases({ kind: "task_assigned" }, { surface: "task_assigned" });
73
- assert.ok(b.includes("assigned"), "an event rebuilt from a flattened item must still bind");
74
- });
75
-
76
- test("a predicate matches on a known field, including any-of lists", () => {
77
- const view = predicateView({ actor: "mem_peer", topic: "task" }, { surface: "task_assigned", reason: "assignee" });
78
- assert.equal(evalPredicate(null, view).ok, true);
79
- assert.equal(evalPredicate({}, view).ok, true);
80
- assert.equal(evalPredicate({ surface: "task_assigned" }, view).ok, true);
81
- assert.equal(evalPredicate({ reason: ["assignee", "reviewer"] }, view).ok, true);
82
- assert.equal(evalPredicate({ reason: "reviewer" }, view).ok, false);
83
- });
84
-
85
- test("NEGATIVE: a predicate this module cannot evaluate FAILS CLOSED", () => {
86
- const view = predicateView({}, {});
87
- const unknown = evalPredicate({ mood: "urgent" }, view);
88
- assert.equal(unknown.ok, false, "an unknown field must not be ignored — that would widen the obligation");
89
- assert.ok(unknown.why.includes("mood"));
90
- assert.equal(evalPredicate("reason=mention", view).ok, false, "a string predicate is not a field map");
91
- assert.equal(evalPredicate([{ surface: "dm" }], view).ok, false);
92
- for (const f of PREDICATE_FIELDS) assert.equal(evalPredicate({ [f]: "" }, view).ok, true, `${f} should compare blank-to-blank`);
93
- });
94
-
95
- // ---------------------------------------------------------------------------
96
- // matching
97
- // ---------------------------------------------------------------------------
98
-
99
- test("an event binds the obligation whose trigger covers it", () => {
100
- const r = matchObligation({ ...boardEvent(), plan: [ob()] });
101
- assert.equal(r.obligation.key, "react.task.assigned");
102
- assert.ok(r.why.join(" ").includes("bound to react.task.assigned"));
103
- });
104
-
105
- test("a trigger may be written against the surface as well as the coarse topic", () => {
106
- const r = matchObligation({ ...boardEvent(), plan: [ob({ key: "react.surface", trigger: { topic: "task_assigned", kind: null } })] });
107
- assert.equal(r.obligation.key, "react.surface");
108
- });
109
-
110
- test("the most specific match wins, deterministically", () => {
111
- const broad = ob({ key: "react.task.any", trigger: { topic: "task", kind: null, predicate: null } });
112
- const narrow = ob({ key: "react.task.assigned", trigger: { topic: "task", kind: "assigned", predicate: { reason: "assignee" } } });
113
- const a = matchObligation({ ...boardEvent(), plan: [broad, narrow] });
114
- const b = matchObligation({ ...boardEvent(), plan: [narrow, broad] });
115
- assert.equal(a.obligation.key, "react.task.assigned");
116
- assert.equal(b.obligation.key, a.obligation.key, "input order must not change the binding");
117
- assert.ok(a.why.some((w) => w.includes("also matched, less specific")));
118
- });
119
-
120
- test("NEGATIVE: a suspended obligation does not bind, and does not read as a coverage gap", () => {
121
- const r = matchObligation({ ...boardEvent(), plan: [ob({ status: "suspended" })] });
122
- assert.equal(r.obligation, null);
123
- assert.deepEqual(r.suspended.map((x) => x.key), ["react.task.assigned"]);
124
- assert.ok(r.why.join(" ").includes('status "suspended"'), "the reason must distinguish suspended from uncovered");
125
- });
126
-
127
- test("NEGATIVE: a proposed obligation is not bindable either", () => {
128
- const r = matchObligation({ ...boardEvent(), plan: [ob({ status: "proposed" })] });
129
- assert.equal(r.obligation, null);
130
- });
131
-
132
- test("NEGATIVE: an undirected event is never matched", () => {
133
- const r = matchObligation({
134
- candidate: boardEvent().candidate,
135
- verdict: { directed: false, reason: "ambient_channel", surface: null },
136
- plan: [ob({ trigger: { topic: "task", kind: null } })],
137
- });
138
- assert.equal(r.obligation, null);
139
- assert.ok(r.why[0].includes("no directed event"));
140
- });
141
-
142
- test("NEGATIVE: a plan with no REACT obligations says so rather than failing silently", () => {
143
- const r = matchObligation({ ...boardEvent(), plan: [{ key: "schedule.x", kind: "SCHEDULE" }] });
144
- assert.equal(r.obligation, null);
145
- assert.ok(r.why.join(" ").includes("no REACT obligations"));
146
- });
147
-
148
- test("SCHEDULE and OUTCOME obligations are never considered for an inbound event", () => {
149
- const list = reactObligations({ obligations: [ob(), { key: "outcome.x", kind: "OUTCOME" }, { key: "schedule.x", kind: "SCHEDULE" }] });
150
- assert.deepEqual(list.map((x) => x.key), ["react.task.assigned"]);
151
- });
152
-
153
- // ---------------------------------------------------------------------------
154
- // binding changes the decision
155
- // ---------------------------------------------------------------------------
156
-
157
- test("binding an obligation lifts the rung cap the uncovered path imposes", () => {
158
- const ev = boardEvent();
159
- const need = { steps: 8, filesystem: true };
160
- const uncovered = decide({ ...ev, me: ME, history: null, need });
161
- const covered = decide({
162
- ...ev,
163
- me: ME,
164
- history: null,
165
- need,
166
- obligation: ob({ objective_id: "obj_1" }),
167
- });
168
- assert.ok(uncovered.rung <= 1, "an uncovered event must be capped at rung ≤1");
169
- assert.ok(covered.rung > uncovered.rung, "a planned event may route to the rung the work actually needs");
170
- assert.equal(covered.obligationKey, "react.task.assigned");
171
- assert.ok(uncovered.drift && uncovered.drift.kind === "uncovered_event");
172
- });
173
-
174
- // ---------------------------------------------------------------------------
175
- // proposals
176
- // ---------------------------------------------------------------------------
177
-
178
- test("an uncovered event proposes a schema-valid REACT obligation", () => {
179
- const ev = boardEvent();
180
- const p = proposeReact({ ...ev, tier: "T0" });
181
- assert.ok(p.obligation, p.why);
182
- assert.deepEqual(validateObligation(p.obligation), [], "the proposal must satisfy the real schema");
183
- assert.equal(p.obligation.status, "proposed", "nothing an agent proposes is active");
184
- assert.deepEqual(p.obligation.uses, [], "an unprobed proposal may not cite a capability (§5.1)");
185
- assert.equal(p.obligation.trigger.topic, "task");
186
- });
187
-
188
- test("a proposal inherits the surface's blast radius, so email proposals are external", () => {
189
- const norm = fromInboxItem(
190
- { raw_ref: "cohort:email:mail_1:5", kind: "email", service: "cohort", timestamp: "2026-08-11T09:00:00.000Z", content: "hi", priority_signals: {} },
191
- { me: ME },
192
- );
193
- const p = proposeReact({ candidate: norm.candidate, verdict: norm.verdict, tier: "T0" });
194
- assert.ok(p.obligation.action_classes.includes("external"));
195
- assert.equal(p.obligation.offline_safe, false);
196
- });
197
-
198
- test("a proposal that would be invalid is dropped loudly, never written", () => {
199
- const p = proposeReact({ candidate: {}, verdict: {} });
200
- assert.equal(p.obligation, null);
201
- assert.ok(p.errors.length > 0);
202
- });
203
-
204
- test("proposed keys are stable slugs, so the same gap proposes the same obligation twice", () => {
205
- const ev = boardEvent();
206
- const a = proposeReact({ ...ev, tier: "T0" });
207
- const b = proposeReact({ ...ev, tier: "T1" });
208
- assert.equal(a.obligation.key, b.obligation.key);
209
- assert.match(a.obligation.key, /^[a-z0-9][a-z0-9._-]*$/);
210
- });
211
-
212
- // ---------------------------------------------------------------------------
213
- // loading
214
- // ---------------------------------------------------------------------------
215
-
216
- test("a missing plan.yaml degrades with a reason instead of throwing", async () => {
217
- const r = await loadReactObligations("/nonexistent/agent/root");
218
- assert.deepEqual(r.obligations, []);
219
- assert.equal(r.degraded, true);
220
- assert.ok(r.reason, "a degradation must always carry a reason");
221
- });
222
-
223
- test("a plan is read through the one loader, and its REACT obligations come back", async () => {
224
- const plan = {
225
- enforcement: "observe_only",
226
- obligations: [ob(), { key: "schedule.a", kind: "SCHEDULE" }],
227
- };
228
- const r = await loadReactObligations("/anywhere", {
229
- readPlan: () => plan,
230
- yaml: { load: () => plan },
231
- });
232
- assert.equal(r.degraded, false);
233
- assert.equal(r.enforcement, "observe_only");
234
- assert.deepEqual(r.obligations.map((x) => x.key), ["react.task.assigned"]);
235
- });
@@ -1,392 +0,0 @@
1
- /**
2
- * pipeline.test.mjs — the one call the inbox pipeline makes, end to end.
3
- * Run: node --test lib/execution/pipeline.test.mjs
4
- *
5
- * What is actually under test here is the BATCH behaviour, because that is the
6
- * part a caller re-implementing the loop would get wrong: the same event
7
- * arriving twice inside one sweep (SSE push + poll backstop, which is the
8
- * design), a burst from one actor, and a human speaking mid-batch.
9
- */
10
- "use strict";
11
-
12
- import { test } from "node:test";
13
- import assert from "node:assert/strict";
14
- import { mkdtempSync, rmSync, existsSync, readFileSync } from "node:fs";
15
- import { tmpdir } from "node:os";
16
- import { join } from "node:path";
17
-
18
- import { processInbound, processOne, batchHistory } from "./pipeline.mjs";
19
- import { JOURNAL_REL } from "./journal.mjs";
20
- import { INBOUND_QUEUE_REL, PROPOSED_REL } from "./effects.mjs";
21
-
22
- const ME = "mem_self";
23
- const NOW = Date.parse("2026-08-11T12:00:00Z");
24
-
25
- function tmp() {
26
- return mkdtempSync(join(tmpdir(), "exec-pipeline-"));
27
- }
28
-
29
- /** A recording effect set — no network, every call captured. */
30
- function fakeEffects() {
31
- const calls = { react: [], schedule: [], delegate: [], escalate: [], drift: [], propose: [], obligation: [], arbitrate: [] };
32
- return {
33
- calls,
34
- react: async (a) => { calls.react.push(a); return { ok: true, ref: { sent: true } }; },
35
- schedule: async (a) => { calls.schedule.push(a); return { ok: true, ref: { queued: true } }; },
36
- delegate: async (a) => { calls.delegate.push(a); return { ok: true, ref: {} }; },
37
- escalate: async (a) => { calls.escalate.push(a); return { ok: true, ref: {} }; },
38
- arbitrate: async (a) => { calls.arbitrate.push(a); return { ok: true, ref: { granted: true } }; },
39
- reportDrift: async (a) => { calls.drift.push(a); return { ok: true }; },
40
- propose: async (a) => { calls.propose.push(a); return { ok: true }; },
41
- proposeObligation: async (a) => { calls.obligation.push(a); return { ok: true }; },
42
- };
43
- }
44
-
45
- function dm(over = {}) {
46
- return {
47
- id: "cohort-msg_1",
48
- service: "cohort",
49
- channel: "dm",
50
- channel_id: "chan_dm",
51
- sender: "peer",
52
- sender_id: "mem_peer",
53
- timestamp: "2026-08-11T11:59:00.000Z",
54
- content: "quick question",
55
- thread_id: "msg_1",
56
- is_dm: true,
57
- priority_signals: { mentions_agent: true },
58
- raw_ref: "cohort:dm:msg_1:11",
59
- kind: "message",
60
- ...over,
61
- };
62
- }
63
-
64
- const REACT_DM = {
65
- key: "react.message.dm",
66
- kind: "REACT",
67
- source: { origin: "react-standard" },
68
- trigger: { topic: "message", kind: null, predicate: null },
69
- status: "active",
70
- uses: [],
71
- allowed_tools: [],
72
- action_classes: [],
73
- offline_safe: true,
74
- };
75
-
76
- function run(items, over = {}) {
77
- return processInbound({
78
- items,
79
- me: ME,
80
- nowMs: NOW,
81
- loadPlan: false,
82
- obligations: [REACT_DM],
83
- ...over,
84
- });
85
- }
86
-
87
- // ---------------------------------------------------------------------------
88
- // the happy path, per surface
89
- // ---------------------------------------------------------------------------
90
-
91
- test("a directed DM is answered, and the journal records both halves", async () => {
92
- const root = tmp();
93
- try {
94
- const fx = fakeEffects();
95
- const { results, stats } = await run([dm()], { agentRoot: root, effects: fx });
96
- assert.equal(results[0].decision.disposition, "react_now");
97
- assert.equal(results[0].origin, "inbox_item");
98
- assert.equal(results[0].obligationKey, "react.message.dm");
99
- assert.equal(fx.calls.react.length, 1);
100
- assert.equal(stats.acted, 1);
101
-
102
- const journal = readFileSync(join(root, JOURNAL_REL), "utf-8").trim().split("\n").map(JSON.parse);
103
- assert.ok(journal.some((r) => r.event === "decision"), "the reasoning is on disk before the effect runs");
104
- assert.ok(journal.some((r) => r.event === "outcome"), "and the outcome after it");
105
- } finally {
106
- rmSync(root, { recursive: true, force: true });
107
- }
108
- });
109
-
110
- test("a mixture of surfaces is dispositioned per surface in one batch", async () => {
111
- const root = tmp();
112
- try {
113
- const fx = fakeEffects();
114
- const { stats } = await run(
115
- [
116
- dm(),
117
- dm({ id: "cohort-task-2", raw_ref: "cohort:task_assigned:task_2:12", kind: "task_assigned", is_dm: false, channel_id: "chan_b" }),
118
- { id: "daemon_down", severity: "critical", detail: "no beat" },
119
- { id: "warm", severity: "info", detail: "warmed" },
120
- { id: "s1", service: "slack", channel_id: "C1", content: "unrelated chatter", timestamp: "2026-08-11T11:00:00.000Z", priority_signals: {} },
121
- ],
122
- { agentRoot: root, effects: fx, obligations: [] },
123
- );
124
- assert.equal(stats.processed, 5);
125
- assert.equal(stats.byDisposition.react_now, 1, "only the DM warrants an immediate answer");
126
- assert.equal(stats.byDisposition.schedule, 2, "the task and the critical alert become work");
127
- assert.equal(stats.byDisposition.ignore, 2, "the info alert and the ambient channel message do not");
128
- assert.equal(stats.bySurface.alert, 1);
129
- } finally {
130
- rmSync(root, { recursive: true, force: true });
131
- }
132
- });
133
-
134
- // ---------------------------------------------------------------------------
135
- // batch semantics — the reason this module exists
136
- // ---------------------------------------------------------------------------
137
-
138
- test("the same event delivered twice in one batch is answered ONCE", async () => {
139
- const root = tmp();
140
- try {
141
- const fx = fakeEffects();
142
- // The SSE push and the 300s sweep both deliver; §6.1 says they converge here.
143
- const { results, stats } = await run([dm(), dm()], { agentRoot: root, effects: fx });
144
- assert.equal(results[0].decision.disposition, "react_now");
145
- assert.equal(results[1].decision.disposition, "ignore");
146
- assert.equal(results[1].decision.reason, "duplicate");
147
- assert.equal(fx.calls.react.length, 1, "the agent must not answer the same message twice");
148
- assert.equal(stats.acted, 1);
149
- } finally {
150
- rmSync(root, { recursive: true, force: true });
151
- }
152
- });
153
-
154
- test("a burst from one actor is batched rather than answered turn-by-turn", async () => {
155
- const root = tmp();
156
- try {
157
- const fx = fakeEffects();
158
- const items = [];
159
- for (let i = 0; i < 6; i += 1) {
160
- // A burst of @mentions (NOT DMs): a directed 1:1 DM is now exempt from
161
- // actor-flood, so flood batching is exercised on a channel surface where
162
- // it still applies. `is_dm:false` + mentions_agent → surface "mention".
163
- items.push(dm({ id: `cohort-msg_${i}`, raw_ref: `cohort:mention:msg_${i}:${100 + i}`, thread_id: `t_${i}`, channel_id: `c_${i}`, is_dm: false }));
164
- }
165
- const { results } = await run(items, { agentRoot: root, effects: fx, actorFloodLimit: 3 });
166
- const dispositions = results.map((r) => r.decision.disposition);
167
- assert.equal(dispositions[0], "react_now");
168
- assert.ok(dispositions.includes("schedule"), "the flood limit must downgrade, not silence");
169
- assert.ok(
170
- results.some((r) => r.decision.reason === "actor_flood"),
171
- "and it must say that is why",
172
- );
173
- // Nothing was dropped: every event still produced a decision.
174
- assert.equal(results.length, 6);
175
- } finally {
176
- rmSync(root, { recursive: true, force: true });
177
- }
178
- });
179
-
180
- test("a ping-pong chain stops inside a single batch", async () => {
181
- const root = tmp();
182
- try {
183
- const fx = fakeEffects();
184
- const items = [];
185
- for (let i = 0; i < 6; i += 1) {
186
- // Same thread, different events, different actors so the flood guard is not
187
- // what stops it.
188
- items.push(dm({ id: `cohort-p_${i}`, raw_ref: `cohort:dm:p_${i}:${200 + i}`, sender_id: `mem_peer_${i}` }));
189
- }
190
- const { results } = await run(items, { agentRoot: root, effects: fx });
191
- const stopped = results.filter((r) => r.decision.reason === "reply_chain_depth");
192
- assert.ok(stopped.length > 0, "an agent talking into a thread forever must stop inside the same sweep");
193
- assert.ok(results.filter((r) => r.spoke).length <= 4, "the DM ceiling is 4 turns");
194
- } finally {
195
- rmSync(root, { recursive: true, force: true });
196
- }
197
- });
198
-
199
- test("a human speaking mid-batch unblocks the agent in the SAME sweep", async () => {
200
- const root = tmp();
201
- try {
202
- const fx = fakeEffects();
203
- const items = [];
204
- for (let i = 0; i < 5; i += 1) items.push(dm({ id: `cohort-h_${i}`, raw_ref: `cohort:dm:h_${i}:${300 + i}`, sender_id: `mem_peer_${i}` }));
205
- const { results } = await run(items, {
206
- agentRoot: root,
207
- effects: fx,
208
- // the fifth item is a human re-entering the conversation
209
- isHumanTurn: (it) => it.id === "cohort-h_4",
210
- });
211
- assert.notEqual(
212
- results[4].decision.reason,
213
- "reply_chain_depth",
214
- "a human turn must reset the chain before the decision that follows it",
215
- );
216
- } finally {
217
- rmSync(root, { recursive: true, force: true });
218
- }
219
- });
220
-
221
- test("batchHistory unions the on-disk snapshot with in-batch decisions", () => {
222
- const base = {
223
- seen: (k) => k === "on-disk",
224
- chainDepth: () => 2,
225
- actorActivity: () => 1,
226
- failures: () => 3,
227
- rows: [],
228
- degraded: false,
229
- degradedReason: null,
230
- };
231
- const h = batchHistory(base);
232
- assert.equal(h.seen("on-disk"), true);
233
- assert.equal(h.seen("fresh"), false);
234
- h.note({ key: "fresh", actor: "a", thread: "t" }, true);
235
- assert.equal(h.seen("fresh"), true);
236
- assert.equal(h.chainDepth("t"), 3, "in-batch turns add to the on-disk count");
237
- assert.equal(h.actorActivity("a"), 2);
238
- assert.equal(h.failures("k"), 3, "failures come from the journal only");
239
- h.humanTurn("t");
240
- assert.equal(h.chainDepth("t"), 0, "a human turn masks the stale on-disk count, not just the in-batch one");
241
- });
242
-
243
- test("batchHistory tolerates no base at all, and says the guards are weakened", () => {
244
- const h = batchHistory(null);
245
- assert.equal(h.seen("x"), false);
246
- assert.equal(h.chainDepth("t"), 0);
247
- assert.equal(h.degraded, true);
248
- assert.ok(h.degradedReason);
249
- });
250
-
251
- // ---------------------------------------------------------------------------
252
- // coverage, drift and proposals
253
- // ---------------------------------------------------------------------------
254
-
255
- test("an uncovered directed event records drift AND proposes the obligation that would cover it", async () => {
256
- const root = tmp();
257
- try {
258
- const fx = fakeEffects();
259
- const { results, stats } = await run([dm()], { agentRoot: root, effects: fx, obligations: [] });
260
- assert.equal(results[0].obligationKey, null);
261
- assert.equal(stats.uncovered, 1);
262
- assert.equal(fx.calls.drift.length, 1);
263
- assert.equal(fx.calls.drift[0].drift.kind, "uncovered_event");
264
- assert.equal(fx.calls.obligation.length, 1, "the plan must learn, not just complain");
265
- assert.equal(fx.calls.obligation[0].obligation.status, "proposed");
266
- assert.deepEqual(stats.proposed, [fx.calls.obligation[0].obligation.key]);
267
- } finally {
268
- rmSync(root, { recursive: true, force: true });
269
- }
270
- });
271
-
272
- test("a suspended obligation is NOT reported as a coverage gap", async () => {
273
- const root = tmp();
274
- try {
275
- const fx = fakeEffects();
276
- const { results } = await run([dm()], {
277
- agentRoot: root,
278
- effects: fx,
279
- obligations: [{ ...REACT_DM, status: "suspended" }],
280
- });
281
- assert.deepEqual(results[0].suspendedObligations, ["react.message.dm"]);
282
- assert.equal(fx.calls.obligation.length, 0, "the plan already covers this — do not propose a duplicate");
283
- } finally {
284
- rmSync(root, { recursive: true, force: true });
285
- }
286
- });
287
-
288
- test("an ignored event never proposes an obligation", async () => {
289
- const root = tmp();
290
- try {
291
- const fx = fakeEffects();
292
- await run([{ id: "warm", severity: "info" }], { agentRoot: root, effects: fx, obligations: [] });
293
- assert.equal(fx.calls.obligation.length, 0, "coverage is only owed for events the agent would act on");
294
- } finally {
295
- rmSync(root, { recursive: true, force: true });
296
- }
297
- });
298
-
299
- // ---------------------------------------------------------------------------
300
- // failure isolation and degradation reporting — never silent
301
- // ---------------------------------------------------------------------------
302
-
303
- test("one poisoned item cannot take the batch down, and it is logged as an error", async () => {
304
- const root = tmp();
305
- try {
306
- const fx = fakeEffects();
307
- const logs = [];
308
- const poison = { get service() { throw new Error("boom"); } };
309
- const { results, stats } = await run([dm(), poison, dm({ id: "cohort-msg_2", raw_ref: "cohort:dm:msg_2:12", thread_id: "t2", channel_id: "c2" })], {
310
- agentRoot: root,
311
- effects: fx,
312
- log: (lvl, msg) => logs.push([lvl, msg]),
313
- });
314
- assert.equal(results.length, 3);
315
- assert.equal(stats.processed >= 2, true, "the healthy items still processed");
316
- // The poisoned item is either caught by intake's own guard or by the batch
317
- // guard — either way it produced a decision or an error, and it was LOUD.
318
- assert.ok(
319
- logs.some(([lvl]) => lvl === "error" || lvl === "warn"),
320
- "a degradation must always be logged",
321
- );
322
- } finally {
323
- rmSync(root, { recursive: true, force: true });
324
- }
325
- });
326
-
327
- test("no compiled plan is reported as a degradation, not treated as normal", async () => {
328
- const root = tmp();
329
- try {
330
- const logs = [];
331
- const { stats } = await processInbound({
332
- items: [dm()],
333
- me: ME,
334
- agentRoot: root,
335
- nowMs: NOW,
336
- effects: fakeEffects(),
337
- log: (lvl, msg) => logs.push([lvl, msg]),
338
- });
339
- assert.ok(
340
- stats.degraded.some((d) => d.startsWith("plan_unavailable")),
341
- "an agent with no plan runs the whole reactive lane at rung ≤1 — the operator must be able to see that",
342
- );
343
- assert.ok(logs.some(([lvl]) => lvl === "warn"));
344
- } finally {
345
- rmSync(root, { recursive: true, force: true });
346
- }
347
- });
348
-
349
- test("an empty batch is a no-op that still returns a well-formed summary", async () => {
350
- const { results, stats } = await processInbound({ items: [], me: ME });
351
- assert.deepEqual(results, []);
352
- assert.equal(stats.total, 0);
353
- assert.equal(stats.processed, 0);
354
- });
355
-
356
- test("processOne works standalone, with the default on-disk history", async () => {
357
- const root = tmp();
358
- try {
359
- const fx = fakeEffects();
360
- const r = await processOne({ item: dm(), me: ME, agentRoot: root, effects: fx, nowMs: NOW, obligations: [REACT_DM] });
361
- assert.equal(r.decision.disposition, "react_now");
362
- assert.ok(existsSync(join(root, JOURNAL_REL)));
363
- } finally {
364
- rmSync(root, { recursive: true, force: true });
365
- }
366
- });
367
-
368
- // ---------------------------------------------------------------------------
369
- // the real effects, wired for once, to prove the shapes line up
370
- // ---------------------------------------------------------------------------
371
-
372
- test("the default effects actually write the queue and the proposal files", async () => {
373
- const root = tmp();
374
- try {
375
- const { defaultEffects } = await import("./effects.mjs");
376
- const effects = defaultEffects({ agentRoot: root, me: ME, nowMs: NOW });
377
- await processInbound({
378
- items: [dm({ id: "cohort-task_9", raw_ref: "cohort:task_assigned:task_9:44", kind: "task_assigned", is_dm: false })],
379
- me: ME,
380
- agentRoot: root,
381
- nowMs: NOW,
382
- effects,
383
- loadPlan: false,
384
- obligations: [],
385
- });
386
- assert.ok(existsSync(join(root, INBOUND_QUEUE_REL)), "queued work must land on the queue the sweep reads");
387
- const proposed = readFileSync(join(root, PROPOSED_REL), "utf-8");
388
- assert.ok(proposed.includes('"type":"obligation"'), "the proposed obligation must be reviewable on disk");
389
- } finally {
390
- rmSync(root, { recursive: true, force: true });
391
- }
392
- });