@monoes/monomindcli 1.11.12 → 1.11.14

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 (434) hide show
  1. package/.claude/commands/mastermind/idea.md +1 -1
  2. package/.claude/commands/mastermind/master.md +1 -1
  3. package/.claude/skills/mastermind/_protocol.md +4 -4
  4. package/.claude/skills/mastermind/architect.md +4 -7
  5. package/.claude/skills/mastermind/autodev.md +2 -4
  6. package/.claude/skills/mastermind/build.md +3 -3
  7. package/.claude/skills/mastermind/content.md +3 -3
  8. package/.claude/skills/mastermind/createorg.md +2 -2
  9. package/.claude/skills/mastermind/finance.md +3 -3
  10. package/.claude/skills/mastermind/idea.md +0 -8
  11. package/.claude/skills/mastermind/marketing.md +3 -3
  12. package/.claude/skills/mastermind/monitor.md +2 -2
  13. package/.claude/skills/mastermind/ops.md +3 -3
  14. package/.claude/skills/mastermind/release.md +3 -3
  15. package/.claude/skills/mastermind/research.md +3 -3
  16. package/.claude/skills/mastermind/review.md +3 -3
  17. package/.claude/skills/mastermind/sales.md +3 -3
  18. package/README.md +286 -129
  19. package/dist/src/agents/halt-signal.d.ts +25 -0
  20. package/dist/src/agents/halt-signal.d.ts.map +1 -0
  21. package/dist/src/agents/halt-signal.js +76 -0
  22. package/dist/src/agents/halt-signal.js.map +1 -0
  23. package/dist/src/agents/index.d.ts +18 -0
  24. package/dist/src/agents/index.d.ts.map +1 -0
  25. package/dist/src/agents/index.js +13 -0
  26. package/dist/src/agents/index.js.map +1 -0
  27. package/dist/src/agents/managed-agent.d.ts +41 -0
  28. package/dist/src/agents/managed-agent.d.ts.map +1 -0
  29. package/dist/src/agents/managed-agent.js +69 -0
  30. package/dist/src/agents/managed-agent.js.map +1 -0
  31. package/dist/src/agents/prompt-experiment.d.ts +23 -0
  32. package/dist/src/agents/prompt-experiment.d.ts.map +1 -0
  33. package/dist/src/agents/prompt-experiment.js +49 -0
  34. package/dist/src/agents/prompt-experiment.js.map +1 -0
  35. package/dist/src/agents/prompt-version-manager.d.ts +22 -0
  36. package/dist/src/agents/prompt-version-manager.d.ts.map +1 -0
  37. package/dist/src/agents/prompt-version-manager.js +80 -0
  38. package/dist/src/agents/prompt-version-manager.js.map +1 -0
  39. package/dist/src/agents/registry-query.d.ts +71 -0
  40. package/dist/src/agents/registry-query.d.ts.map +1 -0
  41. package/dist/src/agents/registry-query.js +125 -0
  42. package/dist/src/agents/registry-query.js.map +1 -0
  43. package/dist/src/agents/score-decay.d.ts +19 -0
  44. package/dist/src/agents/score-decay.d.ts.map +1 -0
  45. package/dist/src/agents/score-decay.js +22 -0
  46. package/dist/src/agents/score-decay.js.map +1 -0
  47. package/dist/src/agents/shared-instructions-loader.d.ts +13 -0
  48. package/dist/src/agents/shared-instructions-loader.d.ts.map +1 -0
  49. package/dist/src/agents/shared-instructions-loader.js +40 -0
  50. package/dist/src/agents/shared-instructions-loader.js.map +1 -0
  51. package/dist/src/agents/specialization-scorer.d.ts +54 -0
  52. package/dist/src/agents/specialization-scorer.d.ts.map +1 -0
  53. package/dist/src/agents/specialization-scorer.js +212 -0
  54. package/dist/src/agents/specialization-scorer.js.map +1 -0
  55. package/dist/src/agents/termination-watcher.d.ts +30 -0
  56. package/dist/src/agents/termination-watcher.d.ts.map +1 -0
  57. package/dist/src/agents/termination-watcher.js +84 -0
  58. package/dist/src/agents/termination-watcher.js.map +1 -0
  59. package/dist/src/agents/trigger-index.d.ts +20 -0
  60. package/dist/src/agents/trigger-index.d.ts.map +1 -0
  61. package/dist/src/agents/trigger-index.js +38 -0
  62. package/dist/src/agents/trigger-index.js.map +1 -0
  63. package/dist/src/agents/trigger-scanner.d.ts +64 -0
  64. package/dist/src/agents/trigger-scanner.d.ts.map +1 -0
  65. package/dist/src/agents/trigger-scanner.js +308 -0
  66. package/dist/src/agents/trigger-scanner.js.map +1 -0
  67. package/dist/src/agents/version-diff.d.ts +18 -0
  68. package/dist/src/agents/version-diff.d.ts.map +1 -0
  69. package/dist/src/agents/version-diff.js +64 -0
  70. package/dist/src/agents/version-diff.js.map +1 -0
  71. package/dist/src/agents/version-store.d.ts +60 -0
  72. package/dist/src/agents/version-store.d.ts.map +1 -0
  73. package/dist/src/agents/version-store.js +235 -0
  74. package/dist/src/agents/version-store.js.map +1 -0
  75. package/dist/src/benchmarks/pretrain/index.d.ts +45 -0
  76. package/dist/src/benchmarks/pretrain/index.d.ts.map +1 -0
  77. package/dist/src/benchmarks/pretrain/index.js +404 -0
  78. package/dist/src/benchmarks/pretrain/index.js.map +1 -0
  79. package/dist/src/commands/agent-wasm.d.ts +14 -0
  80. package/dist/src/commands/agent-wasm.d.ts.map +1 -0
  81. package/dist/src/commands/agent-wasm.js +333 -0
  82. package/dist/src/commands/agent-wasm.js.map +1 -0
  83. package/dist/src/commands/doctor.d.ts.map +1 -1
  84. package/dist/src/commands/doctor.js +118 -13
  85. package/dist/src/commands/doctor.js.map +1 -1
  86. package/dist/src/commands/embeddings.d.ts.map +1 -0
  87. package/dist/src/commands/embeddings.js.map +1 -0
  88. package/dist/src/commands/ui.js +68 -0
  89. package/dist/src/consensus/index.d.ts +7 -0
  90. package/dist/src/consensus/index.d.ts.map +1 -0
  91. package/dist/src/consensus/index.js +6 -0
  92. package/dist/src/consensus/index.js.map +1 -0
  93. package/dist/src/context/context-provider.d.ts +44 -0
  94. package/dist/src/context/context-provider.d.ts.map +1 -0
  95. package/dist/src/context/context-provider.js +25 -0
  96. package/dist/src/context/context-provider.js.map +1 -0
  97. package/dist/src/context/git-state-provider.d.ts +12 -0
  98. package/dist/src/context/git-state-provider.d.ts.map +1 -0
  99. package/dist/src/context/git-state-provider.js +34 -0
  100. package/dist/src/context/git-state-provider.js.map +1 -0
  101. package/dist/src/context/index.d.ts +12 -0
  102. package/dist/src/context/index.d.ts.map +1 -0
  103. package/dist/src/context/index.js +12 -0
  104. package/dist/src/context/index.js.map +1 -0
  105. package/dist/src/context/project-conventions-provider.d.ts +15 -0
  106. package/dist/src/context/project-conventions-provider.d.ts.map +1 -0
  107. package/dist/src/context/project-conventions-provider.js +19 -0
  108. package/dist/src/context/project-conventions-provider.js.map +1 -0
  109. package/dist/src/context/prompt-assembler.d.ts +26 -0
  110. package/dist/src/context/prompt-assembler.d.ts.map +1 -0
  111. package/dist/src/context/prompt-assembler.js +93 -0
  112. package/dist/src/context/prompt-assembler.js.map +1 -0
  113. package/dist/src/context/task-history-provider.d.ts +24 -0
  114. package/dist/src/context/task-history-provider.d.ts.map +1 -0
  115. package/dist/src/context/task-history-provider.js +32 -0
  116. package/dist/src/context/task-history-provider.js.map +1 -0
  117. package/dist/src/context/user-preferences-provider.d.ts +14 -0
  118. package/dist/src/context/user-preferences-provider.d.ts.map +1 -0
  119. package/dist/src/context/user-preferences-provider.js +27 -0
  120. package/dist/src/context/user-preferences-provider.js.map +1 -0
  121. package/dist/src/dlq/dlq-reader.d.ts +31 -0
  122. package/dist/src/dlq/dlq-reader.d.ts.map +1 -0
  123. package/dist/src/dlq/dlq-reader.js +81 -0
  124. package/dist/src/dlq/dlq-reader.js.map +1 -0
  125. package/dist/src/dlq/dlq-writer.d.ts +24 -0
  126. package/dist/src/dlq/dlq-writer.d.ts.map +1 -0
  127. package/dist/src/dlq/dlq-writer.js +65 -0
  128. package/dist/src/dlq/dlq-writer.js.map +1 -0
  129. package/dist/src/dlq/index.d.ts +10 -0
  130. package/dist/src/dlq/index.d.ts.map +1 -0
  131. package/dist/src/dlq/index.js +7 -0
  132. package/dist/src/dlq/index.js.map +1 -0
  133. package/dist/src/eval/dataset-manager.d.ts +33 -0
  134. package/dist/src/eval/dataset-manager.d.ts.map +1 -0
  135. package/dist/src/eval/dataset-manager.js +107 -0
  136. package/dist/src/eval/dataset-manager.js.map +1 -0
  137. package/dist/src/eval/dataset-runner.d.ts +23 -0
  138. package/dist/src/eval/dataset-runner.d.ts.map +1 -0
  139. package/dist/src/eval/dataset-runner.js +59 -0
  140. package/dist/src/eval/dataset-runner.js.map +1 -0
  141. package/dist/src/eval/index.d.ts +10 -0
  142. package/dist/src/eval/index.d.ts.map +1 -0
  143. package/dist/src/eval/index.js +7 -0
  144. package/dist/src/eval/index.js.map +1 -0
  145. package/dist/src/eval/trace-collector.d.ts +40 -0
  146. package/dist/src/eval/trace-collector.d.ts.map +1 -0
  147. package/dist/src/eval/trace-collector.js +102 -0
  148. package/dist/src/eval/trace-collector.js.map +1 -0
  149. package/dist/src/graph/enrich.mjs +362 -0
  150. package/dist/src/infrastructure/in-memory-repositories.d.ts +68 -0
  151. package/dist/src/infrastructure/in-memory-repositories.d.ts.map +1 -0
  152. package/dist/src/infrastructure/in-memory-repositories.js +264 -0
  153. package/dist/src/infrastructure/in-memory-repositories.js.map +1 -0
  154. package/dist/src/init/statusline-generator.js +3 -3
  155. package/dist/src/interactive/interrupt.d.ts +22 -0
  156. package/dist/src/interactive/interrupt.d.ts.map +1 -0
  157. package/dist/src/interactive/interrupt.js +71 -0
  158. package/dist/src/interactive/interrupt.js.map +1 -0
  159. package/dist/src/mcp/deprecation-injector.d.ts +25 -0
  160. package/dist/src/mcp/deprecation-injector.d.ts.map +1 -0
  161. package/dist/src/mcp/deprecation-injector.js +48 -0
  162. package/dist/src/mcp/deprecation-injector.js.map +1 -0
  163. package/dist/src/mcp/tool-registry.d.ts +61 -0
  164. package/dist/src/mcp/tool-registry.d.ts.map +1 -0
  165. package/dist/src/mcp/tool-registry.js +246 -0
  166. package/dist/src/mcp/tool-registry.js.map +1 -0
  167. package/dist/src/mcp-tools/wasm-agent-tools.d.ts +9 -0
  168. package/dist/src/mcp-tools/wasm-agent-tools.d.ts.map +1 -0
  169. package/dist/src/mcp-tools/wasm-agent-tools.js +230 -0
  170. package/dist/src/mcp-tools/wasm-agent-tools.js.map +1 -0
  171. package/dist/src/model/complexity-scorer.d.ts +21 -0
  172. package/dist/src/model/complexity-scorer.d.ts.map +1 -0
  173. package/dist/src/model/complexity-scorer.js +106 -0
  174. package/dist/src/model/complexity-scorer.js.map +1 -0
  175. package/dist/src/model/index.d.ts +4 -0
  176. package/dist/src/model/index.d.ts.map +1 -0
  177. package/dist/src/model/index.js +4 -0
  178. package/dist/src/model/index.js.map +1 -0
  179. package/dist/src/model/model-settings.d.ts +22 -0
  180. package/dist/src/model/model-settings.d.ts.map +1 -0
  181. package/dist/src/model/model-settings.js +33 -0
  182. package/dist/src/model/model-settings.js.map +1 -0
  183. package/dist/src/model/model-tier-resolver.d.ts +24 -0
  184. package/dist/src/model/model-tier-resolver.d.ts.map +1 -0
  185. package/dist/src/model/model-tier-resolver.js +65 -0
  186. package/dist/src/model/model-tier-resolver.js.map +1 -0
  187. package/dist/src/monovector/capabilities.d.ts +34 -0
  188. package/dist/src/monovector/capabilities.d.ts.map +1 -0
  189. package/dist/src/monovector/capabilities.js +37 -0
  190. package/dist/src/monovector/capabilities.js.map +1 -0
  191. package/dist/src/observability/replay-reader.d.ts +1 -1
  192. package/dist/src/observability/replay-reader.d.ts.map +1 -1
  193. package/dist/src/orchestration/index.d.ts +7 -0
  194. package/dist/src/orchestration/index.d.ts.map +1 -0
  195. package/dist/src/orchestration/index.js +6 -0
  196. package/dist/src/orchestration/index.js.map +1 -0
  197. package/dist/src/orchestration/mode-dispatcher.d.ts +11 -0
  198. package/dist/src/orchestration/mode-dispatcher.d.ts.map +1 -0
  199. package/dist/src/orchestration/mode-dispatcher.js +31 -0
  200. package/dist/src/orchestration/mode-dispatcher.js.map +1 -0
  201. package/dist/src/orchestration/routing-modes.d.ts +68 -0
  202. package/dist/src/orchestration/routing-modes.d.ts.map +1 -0
  203. package/dist/src/orchestration/routing-modes.js +180 -0
  204. package/dist/src/orchestration/routing-modes.js.map +1 -0
  205. package/dist/src/plugins/tests/demo-plugin-store.d.ts +7 -0
  206. package/dist/src/plugins/tests/demo-plugin-store.d.ts.map +1 -0
  207. package/dist/src/plugins/tests/demo-plugin-store.js +126 -0
  208. package/dist/src/plugins/tests/demo-plugin-store.js.map +1 -0
  209. package/dist/src/plugins/tests/standalone-test.d.ts +12 -0
  210. package/dist/src/plugins/tests/standalone-test.d.ts.map +1 -0
  211. package/dist/src/plugins/tests/standalone-test.js +188 -0
  212. package/dist/src/plugins/tests/standalone-test.js.map +1 -0
  213. package/dist/src/plugins/tests/test-plugin-store.d.ts +7 -0
  214. package/dist/src/plugins/tests/test-plugin-store.d.ts.map +1 -0
  215. package/dist/src/plugins/tests/test-plugin-store.js +206 -0
  216. package/dist/src/plugins/tests/test-plugin-store.js.map +1 -0
  217. package/dist/src/runtime/headless.d.ts +60 -0
  218. package/dist/src/runtime/headless.d.ts.map +1 -0
  219. package/dist/src/runtime/headless.js +284 -0
  220. package/dist/src/runtime/headless.js.map +1 -0
  221. package/dist/src/services/agentic-flow-bridge.d.ts +50 -0
  222. package/dist/src/services/agentic-flow-bridge.d.ts.map +1 -0
  223. package/dist/src/services/agentic-flow-bridge.js +95 -0
  224. package/dist/src/services/agentic-flow-bridge.js.map +1 -0
  225. package/dist/src/services/container-worker-pool.d.ts +197 -0
  226. package/dist/src/services/container-worker-pool.d.ts.map +1 -0
  227. package/dist/src/services/container-worker-pool.js +623 -0
  228. package/dist/src/services/container-worker-pool.js.map +1 -0
  229. package/dist/src/services/index.d.ts +13 -0
  230. package/dist/src/services/index.d.ts.map +1 -0
  231. package/dist/src/services/index.js +11 -0
  232. package/dist/src/services/index.js.map +1 -0
  233. package/dist/src/services/worker-queue.d.ts +201 -0
  234. package/dist/src/services/worker-queue.d.ts.map +1 -0
  235. package/dist/src/services/worker-queue.js +594 -0
  236. package/dist/src/services/worker-queue.js.map +1 -0
  237. package/dist/src/swarm/communication-graph.d.ts +25 -0
  238. package/dist/src/swarm/communication-graph.d.ts.map +1 -0
  239. package/dist/src/swarm/communication-graph.js +77 -0
  240. package/dist/src/swarm/communication-graph.js.map +1 -0
  241. package/dist/src/swarm/flow-enforcer.d.ts +31 -0
  242. package/dist/src/swarm/flow-enforcer.d.ts.map +1 -0
  243. package/dist/src/swarm/flow-enforcer.js +61 -0
  244. package/dist/src/swarm/flow-enforcer.js.map +1 -0
  245. package/dist/src/swarm/flow-visualizer.d.ts +19 -0
  246. package/dist/src/swarm/flow-visualizer.d.ts.map +1 -0
  247. package/dist/src/swarm/flow-visualizer.js +68 -0
  248. package/dist/src/swarm/flow-visualizer.js.map +1 -0
  249. package/dist/src/transfer/deploy-seraphine.d.ts +13 -0
  250. package/dist/src/transfer/deploy-seraphine.d.ts.map +1 -0
  251. package/dist/src/transfer/deploy-seraphine.js +205 -0
  252. package/dist/src/transfer/deploy-seraphine.js.map +1 -0
  253. package/dist/src/transfer/store/tests/standalone-test.d.ts +12 -0
  254. package/dist/src/transfer/store/tests/standalone-test.d.ts.map +1 -0
  255. package/dist/src/transfer/store/tests/standalone-test.js +190 -0
  256. package/dist/src/transfer/store/tests/standalone-test.js.map +1 -0
  257. package/dist/src/transfer/test-seraphine.d.ts +6 -0
  258. package/dist/src/transfer/test-seraphine.d.ts.map +1 -0
  259. package/dist/src/transfer/test-seraphine.js +105 -0
  260. package/dist/src/transfer/test-seraphine.js.map +1 -0
  261. package/dist/src/transfer/tests/test-store.d.ts +7 -0
  262. package/dist/src/transfer/tests/test-store.d.ts.map +1 -0
  263. package/dist/src/transfer/tests/test-store.js +214 -0
  264. package/dist/src/transfer/tests/test-store.js.map +1 -0
  265. package/dist/src/ui/collector.mjs +755 -0
  266. package/dist/src/ui/dashboard.html +10639 -0
  267. package/dist/src/ui/data/agent-avatars.html +763 -0
  268. package/dist/src/ui/data/agent-avatars.json +966 -0
  269. package/dist/src/ui/data/avatars/account-strategist.svg +58 -0
  270. package/dist/src/ui/data/avatars/accounts-payable.svg +54 -0
  271. package/dist/src/ui/data/avatars/adaptive-coordinator.svg +55 -0
  272. package/dist/src/ui/data/avatars/adaptive-coordinator2.svg +54 -0
  273. package/dist/src/ui/data/avatars/ai-citation.svg +57 -0
  274. package/dist/src/ui/data/avatars/ai-engineer.svg +61 -0
  275. package/dist/src/ui/data/avatars/analytics-reporter.svg +53 -0
  276. package/dist/src/ui/data/avatars/api-tester.svg +53 -0
  277. package/dist/src/ui/data/avatars/architecture.svg +54 -0
  278. package/dist/src/ui/data/avatars/automation-governance.svg +55 -0
  279. package/dist/src/ui/data/avatars/backend-dev.svg +53 -0
  280. package/dist/src/ui/data/avatars/benchmarker.svg +54 -0
  281. package/dist/src/ui/data/avatars/blockchain-auditor.svg +53 -0
  282. package/dist/src/ui/data/avatars/byzantine-coord.svg +57 -0
  283. package/dist/src/ui/data/avatars/case-analyst.svg +57 -0
  284. package/dist/src/ui/data/avatars/cicd-engineer.svg +55 -0
  285. package/dist/src/ui/data/avatars/cloud-architect.svg +54 -0
  286. package/dist/src/ui/data/avatars/code-review-swarm.svg +57 -0
  287. package/dist/src/ui/data/avatars/coder-v119.svg +57 -0
  288. package/dist/src/ui/data/avatars/coder.svg +58 -0
  289. package/dist/src/ui/data/avatars/collective-coord.svg +54 -0
  290. package/dist/src/ui/data/avatars/compliance-auditor.svg +58 -0
  291. package/dist/src/ui/data/avatars/consensus-coordinator.svg +54 -0
  292. package/dist/src/ui/data/avatars/content-creator.svg +54 -0
  293. package/dist/src/ui/data/avatars/crdt-synchronizer.svg +53 -0
  294. package/dist/src/ui/data/avatars/cro-specialist.svg +58 -0
  295. package/dist/src/ui/data/avatars/data-consolidator.svg +54 -0
  296. package/dist/src/ui/data/avatars/data-engineer.svg +53 -0
  297. package/dist/src/ui/data/avatars/database-optimizer.svg +61 -0
  298. package/dist/src/ui/data/avatars/deal-strategist.svg +54 -0
  299. package/dist/src/ui/data/avatars/defender.svg +53 -0
  300. package/dist/src/ui/data/avatars/devops-automator.svg +56 -0
  301. package/dist/src/ui/data/avatars/discovery-coach.svg +54 -0
  302. package/dist/src/ui/data/avatars/email-marketing.svg +57 -0
  303. package/dist/src/ui/data/avatars/embedded-firmware.svg +61 -0
  304. package/dist/src/ui/data/avatars/evidence-collector.svg +57 -0
  305. package/dist/src/ui/data/avatars/experiment-tracker.svg +53 -0
  306. package/dist/src/ui/data/avatars/feedback-synthesizer.svg +54 -0
  307. package/dist/src/ui/data/avatars/finance-tracker.svg +54 -0
  308. package/dist/src/ui/data/avatars/frontend-developer.svg +54 -0
  309. package/dist/src/ui/data/avatars/game-audio-engineer.svg +59 -0
  310. package/dist/src/ui/data/avatars/game-designer.svg +54 -0
  311. package/dist/src/ui/data/avatars/gossip-coordinator.svg +54 -0
  312. package/dist/src/ui/data/avatars/hierarchical-coord.svg +54 -0
  313. package/dist/src/ui/data/avatars/incident-commander.svg +57 -0
  314. package/dist/src/ui/data/avatars/infrastructure.svg +54 -0
  315. package/dist/src/ui/data/avatars/input-validator.svg +53 -0
  316. package/dist/src/ui/data/avatars/ios-developer.svg +54 -0
  317. package/dist/src/ui/data/avatars/issue-tracker.svg +53 -0
  318. package/dist/src/ui/data/avatars/judge.svg +55 -0
  319. package/dist/src/ui/data/avatars/launch-strategist.svg +54 -0
  320. package/dist/src/ui/data/avatars/legal-compliance.svg +53 -0
  321. package/dist/src/ui/data/avatars/level-designer.svg +53 -0
  322. package/dist/src/ui/data/avatars/load-balancer.svg +57 -0
  323. package/dist/src/ui/data/avatars/mcp-builder.svg +53 -0
  324. package/dist/src/ui/data/avatars/memory-coordinator.svg +55 -0
  325. package/dist/src/ui/data/avatars/mesh-coordinator.svg +55 -0
  326. package/dist/src/ui/data/avatars/ml-developer.svg +58 -0
  327. package/dist/src/ui/data/avatars/mobile-app-builder.svg +53 -0
  328. package/dist/src/ui/data/avatars/mobile-dev.svg +54 -0
  329. package/dist/src/ui/data/avatars/model-qa.svg +58 -0
  330. package/dist/src/ui/data/avatars/narrative-designer.svg +58 -0
  331. package/dist/src/ui/data/avatars/outbound-strategist.svg +55 -0
  332. package/dist/src/ui/data/avatars/path-validator.svg +54 -0
  333. package/dist/src/ui/data/avatars/payment-agent.svg +53 -0
  334. package/dist/src/ui/data/avatars/perf-analyzer.svg +58 -0
  335. package/dist/src/ui/data/avatars/pipeline-analyst.svg +54 -0
  336. package/dist/src/ui/data/avatars/planner.svg +55 -0
  337. package/dist/src/ui/data/avatars/pr-manager.svg +54 -0
  338. package/dist/src/ui/data/avatars/pricing-strategist.svg +54 -0
  339. package/dist/src/ui/data/avatars/product-manager.svg +54 -0
  340. package/dist/src/ui/data/avatars/production-validator.svg +54 -0
  341. package/dist/src/ui/data/avatars/project-shepherd.svg +54 -0
  342. package/dist/src/ui/data/avatars/proposal-strategist.svg +54 -0
  343. package/dist/src/ui/data/avatars/prosecutor.svg +57 -0
  344. package/dist/src/ui/data/avatars/pseudocode.svg +53 -0
  345. package/dist/src/ui/data/avatars/queen-coordinator.svg +55 -0
  346. package/dist/src/ui/data/avatars/quorum-manager.svg +53 -0
  347. package/dist/src/ui/data/avatars/raft-manager.svg +53 -0
  348. package/dist/src/ui/data/avatars/reality-checker.svg +58 -0
  349. package/dist/src/ui/data/avatars/recruitment.svg +58 -0
  350. package/dist/src/ui/data/avatars/refinement.svg +53 -0
  351. package/dist/src/ui/data/avatars/release-manager.svg +54 -0
  352. package/dist/src/ui/data/avatars/repo-architect.svg +54 -0
  353. package/dist/src/ui/data/avatars/researcher.svg +58 -0
  354. package/dist/src/ui/data/avatars/resource-allocator.svg +53 -0
  355. package/dist/src/ui/data/avatars/reviewer.svg +53 -0
  356. package/dist/src/ui/data/avatars/safe-executor.svg +53 -0
  357. package/dist/src/ui/data/avatars/sales-coach.svg +53 -0
  358. package/dist/src/ui/data/avatars/sales-engineer.svg +58 -0
  359. package/dist/src/ui/data/avatars/scout-explorer.svg +58 -0
  360. package/dist/src/ui/data/avatars/security-architect.svg +54 -0
  361. package/dist/src/ui/data/avatars/security-auditor.svg +55 -0
  362. package/dist/src/ui/data/avatars/senior-developer.svg +58 -0
  363. package/dist/src/ui/data/avatars/senior-pm.svg +58 -0
  364. package/dist/src/ui/data/avatars/seo-specialist.svg +57 -0
  365. package/dist/src/ui/data/avatars/social-media.svg +54 -0
  366. package/dist/src/ui/data/avatars/solidity-engineer.svg +58 -0
  367. package/dist/src/ui/data/avatars/sparc-coder.svg +58 -0
  368. package/dist/src/ui/data/avatars/sparc-coord.svg +56 -0
  369. package/dist/src/ui/data/avatars/specification.svg +57 -0
  370. package/dist/src/ui/data/avatars/sprint-prioritizer.svg +53 -0
  371. package/dist/src/ui/data/avatars/sre.svg +54 -0
  372. package/dist/src/ui/data/avatars/studio-operations.svg +53 -0
  373. package/dist/src/ui/data/avatars/studio-producer.svg +55 -0
  374. package/dist/src/ui/data/avatars/support-responder.svg +56 -0
  375. package/dist/src/ui/data/avatars/system-architect.svg +54 -0
  376. package/dist/src/ui/data/avatars/task-orchestrator.svg +56 -0
  377. package/dist/src/ui/data/avatars/technical-artist.svg +53 -0
  378. package/dist/src/ui/data/avatars/technical-writer.svg +59 -0
  379. package/dist/src/ui/data/avatars/tester.svg +53 -0
  380. package/dist/src/ui/data/avatars/threat-detection.svg +61 -0
  381. package/dist/src/ui/data/avatars/trend-researcher.svg +54 -0
  382. package/dist/src/ui/data/avatars/trial-director.svg +55 -0
  383. package/dist/src/ui/data/avatars/unity-architect.svg +54 -0
  384. package/dist/src/ui/data/avatars/visionos-engineer.svg +57 -0
  385. package/dist/src/ui/data/avatars/worker-specialist.svg +55 -0
  386. package/dist/src/ui/data/avatars/workflow-architect.svg +57 -0
  387. package/dist/src/ui/data/avatars/workflow-automation.svg +54 -0
  388. package/dist/src/ui/data/avatars/zk-steward.svg +54 -0
  389. package/dist/src/ui/data/mastermind-events.jsonl +59 -0
  390. package/dist/src/ui/data/mastermind-sessions.json +1 -0
  391. package/dist/src/ui/orgs.html +1360 -0
  392. package/dist/src/ui/server.mjs +4336 -0
  393. package/dist/src/workflow/condition-evaluator.d.ts +10 -0
  394. package/dist/src/workflow/condition-evaluator.d.ts.map +1 -0
  395. package/dist/src/workflow/condition-evaluator.js +82 -0
  396. package/dist/src/workflow/condition-evaluator.js.map +1 -0
  397. package/dist/src/workflow/context-resolver.d.ts +12 -0
  398. package/dist/src/workflow/context-resolver.d.ts.map +1 -0
  399. package/dist/src/workflow/context-resolver.js +23 -0
  400. package/dist/src/workflow/context-resolver.js.map +1 -0
  401. package/dist/src/workflow/dag-builder.d.ts +17 -0
  402. package/dist/src/workflow/dag-builder.d.ts.map +1 -0
  403. package/dist/src/workflow/dag-builder.js +129 -0
  404. package/dist/src/workflow/dag-builder.js.map +1 -0
  405. package/dist/src/workflow/dag-executor.d.ts +9 -0
  406. package/dist/src/workflow/dag-executor.d.ts.map +1 -0
  407. package/dist/src/workflow/dag-executor.js +116 -0
  408. package/dist/src/workflow/dag-executor.js.map +1 -0
  409. package/dist/src/workflow/dag-types.d.ts +41 -0
  410. package/dist/src/workflow/dag-types.d.ts.map +1 -0
  411. package/dist/src/workflow/dag-types.js +8 -0
  412. package/dist/src/workflow/dag-types.js.map +1 -0
  413. package/dist/src/workflow/dsl-parser.d.ts +12 -0
  414. package/dist/src/workflow/dsl-parser.d.ts.map +1 -0
  415. package/dist/src/workflow/dsl-parser.js +20 -0
  416. package/dist/src/workflow/dsl-parser.js.map +1 -0
  417. package/dist/src/workflow/dsl-schema.d.ts +165 -0
  418. package/dist/src/workflow/dsl-schema.d.ts.map +1 -0
  419. package/dist/src/workflow/dsl-schema.js +82 -0
  420. package/dist/src/workflow/dsl-schema.js.map +1 -0
  421. package/dist/src/workflow/index.d.ts +13 -0
  422. package/dist/src/workflow/index.d.ts.map +1 -0
  423. package/dist/src/workflow/index.js +11 -0
  424. package/dist/src/workflow/index.js.map +1 -0
  425. package/dist/src/workflow/template-engine.d.ts +11 -0
  426. package/dist/src/workflow/template-engine.d.ts.map +1 -0
  427. package/dist/src/workflow/template-engine.js +40 -0
  428. package/dist/src/workflow/template-engine.js.map +1 -0
  429. package/dist/src/workflow/workflow-executor.d.ts +29 -0
  430. package/dist/src/workflow/workflow-executor.d.ts.map +1 -0
  431. package/dist/src/workflow/workflow-executor.js +227 -0
  432. package/dist/src/workflow/workflow-executor.js.map +1 -0
  433. package/dist/tsconfig.tsbuildinfo +1 -1
  434. package/package.json +9 -9
@@ -0,0 +1,107 @@
1
+ /**
2
+ * DatasetManager - JSONL-based eval dataset management (Task 33)
3
+ */
4
+ import { randomUUID } from 'crypto';
5
+ import { appendFileSync, readFileSync, writeFileSync, renameSync, existsSync, statSync } from 'fs';
6
+ import { resolve, sep } from 'path';
7
+ import { parseJsonl } from '../utils/parse-jsonl.js';
8
+ export class DatasetManager {
9
+ datasetsPath;
10
+ entriesPath;
11
+ constructor(datasetsPath, entriesPath) {
12
+ this.datasetsPath = datasetsPath;
13
+ this.entriesPath = entriesPath;
14
+ }
15
+ /**
16
+ * Create a dataset from a set of filtered traces.
17
+ */
18
+ createFromTraces(opts) {
19
+ const now = new Date().toISOString();
20
+ const agentSlugs = opts.agentSlugs ?? [...new Set(opts.traces.map((t) => t.agentSlug))];
21
+ const dataset = {
22
+ datasetId: randomUUID(),
23
+ name: opts.name,
24
+ description: opts.description,
25
+ agentSlugs,
26
+ createdAt: now,
27
+ updatedAt: now,
28
+ entryCount: opts.traces.length,
29
+ };
30
+ appendFileSync(this.datasetsPath, JSON.stringify(dataset) + '\n', 'utf-8');
31
+ for (const trace of opts.traces) {
32
+ const entry = {
33
+ entryId: randomUUID(),
34
+ datasetId: dataset.datasetId,
35
+ traceId: trace.traceId,
36
+ addedAt: now,
37
+ };
38
+ appendFileSync(this.entriesPath, JSON.stringify(entry) + '\n', 'utf-8');
39
+ }
40
+ return dataset;
41
+ }
42
+ /**
43
+ * List all datasets.
44
+ */
45
+ listDatasets() {
46
+ if (!existsSync(this.datasetsPath))
47
+ return [];
48
+ if (statSync(this.datasetsPath).size > 50 * 1024 * 1024) {
49
+ throw new Error('Dataset file exceeds 50MB — run cleanup');
50
+ }
51
+ const content = readFileSync(this.datasetsPath, 'utf-8');
52
+ return parseJsonl(content);
53
+ }
54
+ /**
55
+ * Get entries for a specific dataset.
56
+ */
57
+ getEntries(datasetId) {
58
+ if (!existsSync(this.entriesPath))
59
+ return [];
60
+ if (statSync(this.entriesPath).size > 50 * 1024 * 1024) {
61
+ throw new Error('Entries file exceeds 50MB — run cleanup');
62
+ }
63
+ const content = readFileSync(this.entriesPath, 'utf-8');
64
+ return parseJsonl(content).filter((e) => e.datasetId === datasetId);
65
+ }
66
+ /**
67
+ * Add a single trace to an existing dataset.
68
+ */
69
+ addTraceToDataset(datasetId, traceId) {
70
+ const entry = {
71
+ entryId: randomUUID(),
72
+ datasetId,
73
+ traceId,
74
+ addedAt: new Date().toISOString(),
75
+ };
76
+ appendFileSync(this.entriesPath, JSON.stringify(entry) + '\n', 'utf-8');
77
+ // Update dataset entryCount via atomic write
78
+ const datasets = this.listDatasets();
79
+ const updated = datasets.map((d) => {
80
+ if (d.datasetId === datasetId) {
81
+ return { ...d, entryCount: d.entryCount + 1, updatedAt: new Date().toISOString() };
82
+ }
83
+ return d;
84
+ });
85
+ const tmp = `${this.datasetsPath}.${randomUUID()}.tmp`;
86
+ writeFileSync(tmp, updated.map((d) => JSON.stringify(d)).join('\n') + '\n', 'utf-8');
87
+ renameSync(tmp, this.datasetsPath);
88
+ return entry;
89
+ }
90
+ /**
91
+ * Export a dataset to a JSON file. Output path must be within `allowedRoot`.
92
+ */
93
+ exportToFile(datasetId, outputPath, allowedRoot) {
94
+ if (allowedRoot) {
95
+ const resolvedOut = resolve(outputPath);
96
+ const resolvedRoot = resolve(allowedRoot);
97
+ if (!resolvedOut.startsWith(resolvedRoot + sep) && resolvedOut !== resolvedRoot) {
98
+ throw new Error(`Export path escapes allowed root: ${resolvedOut}`);
99
+ }
100
+ }
101
+ const datasets = this.listDatasets();
102
+ const dataset = datasets.find((d) => d.datasetId === datasetId);
103
+ const entries = this.getEntries(datasetId);
104
+ writeFileSync(outputPath, JSON.stringify({ dataset, entries }, null, 2), 'utf-8');
105
+ }
106
+ }
107
+ //# sourceMappingURL=dataset-manager.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"dataset-manager.js","sourceRoot":"","sources":["../../../src/eval/dataset-manager.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,aAAa,EAAE,UAAU,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,IAAI,CAAC;AACnG,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,MAAM,MAAM,CAAC;AAEpC,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AASrD,MAAM,OAAO,cAAc;IACjB,YAAY,CAAS;IACrB,WAAW,CAAS;IAE5B,YAAY,YAAoB,EAAE,WAAmB;QACnD,IAAI,CAAC,YAAY,GAAG,YAAY,CAAC;QACjC,IAAI,CAAC,WAAW,GAAG,WAAW,CAAC;IACjC,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,IAA0B;QACzC,MAAM,GAAG,GAAG,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;QACrC,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;QACxF,MAAM,OAAO,GAAgB;YAC3B,SAAS,EAAE,UAAU,EAAE;YACvB,IAAI,EAAE,IAAI,CAAC,IAAI;YACf,WAAW,EAAE,IAAI,CAAC,WAAW;YAC7B,UAAU;YACV,SAAS,EAAE,GAAG;YACd,SAAS,EAAE,GAAG;YACd,UAAU,EAAE,IAAI,CAAC,MAAM,CAAC,MAAM;SAC/B,CAAC;QAEF,cAAc,CAAC,IAAI,CAAC,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAE3E,KAAK,MAAM,KAAK,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChC,MAAM,KAAK,GAAqB;gBAC9B,OAAO,EAAE,UAAU,EAAE;gBACrB,SAAS,EAAE,OAAO,CAAC,SAAS;gBAC5B,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,OAAO,EAAE,GAAG;aACb,CAAC;YACF,cAAc,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAC1E,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,YAAY;QACV,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,YAAY,CAAC;YAAE,OAAO,EAAE,CAAC;QAC9C,IAAI,QAAQ,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,IAAI,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,EAAE,CAAC;YACxD,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;QAC7D,CAAC;QACD,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,YAAY,EAAE,OAAO,CAAC,CAAC;QACzD,OAAO,UAAU,CAAc,OAAO,CAAC,CAAC;IAC1C,CAAC;IAED;;OAEG;IACH,UAAU,CAAC,SAAiB;QAC1B,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;YAAE,OAAO,EAAE,CAAC;QAC7C,IAAI,QAAQ,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,IAAI,GAAG,EAAE,GAAG,IAAI,GAAG,IAAI,EAAE,CAAC;YACvD,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAC;QAC7D,CAAC;QACD,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,CAAC,CAAC;QACxD,OAAO,UAAU,CAAmB,OAAO,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC;IACxF,CAAC;IAED;;OAEG;IACH,iBAAiB,CAAC,SAAiB,EAAE,OAAe;QAClD,MAAM,KAAK,GAAqB;YAC9B,OAAO,EAAE,UAAU,EAAE;YACrB,SAAS;YACT,OAAO;YACP,OAAO,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;SAClC,CAAC;QACF,cAAc,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAExE,6CAA6C;QAC7C,MAAM,QAAQ,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;YACjC,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;gBAC9B,OAAO,EAAE,GAAG,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,GAAG,CAAC,EAAE,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,EAAE,CAAC;YACrF,CAAC;YACD,OAAO,CAAC,CAAC;QACX,CAAC,CAAC,CAAC;QACH,MAAM,GAAG,GAAG,GAAG,IAAI,CAAC,YAAY,IAAI,UAAU,EAAE,MAAM,CAAC;QACvD,aAAa,CAAC,GAAG,EAAE,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QACrF,UAAU,CAAC,GAAG,EAAE,IAAI,CAAC,YAAY,CAAC,CAAC;QAEnC,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;OAEG;IACH,YAAY,CAAC,SAAiB,EAAE,UAAkB,EAAE,WAAoB;QACtE,IAAI,WAAW,EAAE,CAAC;YAChB,MAAM,WAAW,GAAG,OAAO,CAAC,UAAU,CAAC,CAAC;YACxC,MAAM,YAAY,GAAG,OAAO,CAAC,WAAW,CAAC,CAAC;YAC1C,IAAI,CAAC,WAAW,CAAC,UAAU,CAAC,YAAY,GAAG,GAAG,CAAC,IAAI,WAAW,KAAK,YAAY,EAAE,CAAC;gBAChF,MAAM,IAAI,KAAK,CAAC,qCAAqC,WAAW,EAAE,CAAC,CAAC;YACtE,CAAC;QACH,CAAC;QACD,MAAM,QAAQ,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC;QACrC,MAAM,OAAO,GAAG,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC;QAChE,MAAM,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC;QAC3C,aAAa,CAAC,UAAU,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;IACpF,CAAC;CACF"}
@@ -0,0 +1,23 @@
1
+ import type { EvalRunResult, EvalTrace } from '../../../shared/src/types/eval.js';
2
+ export interface AgentRunnerResult {
3
+ agentOutput: string;
4
+ outcome: 'success' | 'failure' | 'timeout';
5
+ qualityScore: number;
6
+ latencyMs: number;
7
+ }
8
+ export interface DatasetRunOpts {
9
+ datasetId: string;
10
+ agentVersion: string;
11
+ traces: EvalTrace[];
12
+ agentRunner: (trace: EvalTrace) => Promise<AgentRunnerResult>;
13
+ baselineResult?: EvalRunResult;
14
+ regressionThreshold?: number;
15
+ }
16
+ export declare class DatasetRunner {
17
+ /**
18
+ * Run all traces through the agent runner and compute stats.
19
+ * Optionally compare against a baseline to detect regressions.
20
+ */
21
+ run(opts: DatasetRunOpts): Promise<EvalRunResult>;
22
+ }
23
+ //# sourceMappingURL=dataset-runner.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"dataset-runner.d.ts","sourceRoot":"","sources":["../../../src/eval/dataset-runner.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,aAAa,EAAE,SAAS,EAAoB,MAAM,mCAAmC,CAAC;AAEpG,MAAM,WAAW,iBAAiB;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,SAAS,GAAG,SAAS,GAAG,SAAS,CAAC;IAC3C,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,WAAW,EAAE,CAAC,KAAK,EAAE,SAAS,KAAK,OAAO,CAAC,iBAAiB,CAAC,CAAC;IAC9D,cAAc,CAAC,EAAE,aAAa,CAAC;IAC/B,mBAAmB,CAAC,EAAE,MAAM,CAAC;CAC9B;AAED,qBAAa,aAAa;IACxB;;;OAGG;IACG,GAAG,CAAC,IAAI,EAAE,cAAc,GAAG,OAAO,CAAC,aAAa,CAAC;CA6DxD"}
@@ -0,0 +1,59 @@
1
+ /**
2
+ * DatasetRunner - Run eval datasets against agents and detect regressions (Task 33)
3
+ */
4
+ import { randomUUID } from 'crypto';
5
+ export class DatasetRunner {
6
+ /**
7
+ * Run all traces through the agent runner and compute stats.
8
+ * Optionally compare against a baseline to detect regressions.
9
+ */
10
+ async run(opts) {
11
+ const { datasetId, agentVersion, traces, agentRunner, baselineResult, regressionThreshold = 0.1, } = opts;
12
+ const results = [];
13
+ for (const trace of traces) {
14
+ const result = await agentRunner(trace);
15
+ results.push({ trace, result });
16
+ }
17
+ const passCount = results.filter((r) => r.result.outcome === 'success').length;
18
+ const failCount = results.length - passCount;
19
+ const totalQuality = results.reduce((sum, r) => sum + r.result.qualityScore, 0);
20
+ const totalLatency = results.reduce((sum, r) => sum + r.result.latencyMs, 0);
21
+ const avgQualityScore = results.length > 0 ? totalQuality / results.length : 0;
22
+ const avgLatencyMs = results.length > 0 ? totalLatency / results.length : 0;
23
+ // Regression detection
24
+ const regressionDetails = [];
25
+ let regressionDetected = false;
26
+ if (baselineResult) {
27
+ const delta = baselineResult.avgQualityScore - avgQualityScore;
28
+ if (delta > regressionThreshold) {
29
+ regressionDetected = true;
30
+ // Report per-trace regressions for traces with quality below baseline average
31
+ for (const { trace, result } of results) {
32
+ if (result.qualityScore < baselineResult.avgQualityScore) {
33
+ regressionDetails.push({
34
+ traceId: trace.traceId,
35
+ agentSlug: trace.agentSlug,
36
+ baselineScore: baselineResult.avgQualityScore,
37
+ currentScore: result.qualityScore,
38
+ delta: baselineResult.avgQualityScore - result.qualityScore,
39
+ });
40
+ }
41
+ }
42
+ }
43
+ }
44
+ return {
45
+ runId: randomUUID(),
46
+ datasetId,
47
+ runAt: new Date().toISOString(),
48
+ agentVersion,
49
+ entriesTested: results.length,
50
+ passCount,
51
+ failCount,
52
+ avgQualityScore,
53
+ avgLatencyMs,
54
+ regressionDetected,
55
+ regressionDetails,
56
+ };
57
+ }
58
+ }
59
+ //# sourceMappingURL=dataset-runner.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"dataset-runner.js","sourceRoot":"","sources":["../../../src/eval/dataset-runner.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AAmBpC,MAAM,OAAO,aAAa;IACxB;;;OAGG;IACH,KAAK,CAAC,GAAG,CAAC,IAAoB;QAC5B,MAAM,EACJ,SAAS,EACT,YAAY,EACZ,MAAM,EACN,WAAW,EACX,cAAc,EACd,mBAAmB,GAAG,GAAG,GAC1B,GAAG,IAAI,CAAC;QAET,MAAM,OAAO,GAA2D,EAAE,CAAC;QAE3E,KAAK,MAAM,KAAK,IAAI,MAAM,EAAE,CAAC;YAC3B,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,KAAK,CAAC,CAAC;YACxC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC;QAClC,CAAC;QAED,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC,MAAM,CAAC;QAC/E,MAAM,SAAS,GAAG,OAAO,CAAC,MAAM,GAAG,SAAS,CAAC;QAC7C,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC,CAAC;QAChF,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC;QAC7E,MAAM,eAAe,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAC/E,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;QAE5E,uBAAuB;QACvB,MAAM,iBAAiB,GAAuB,EAAE,CAAC;QACjD,IAAI,kBAAkB,GAAG,KAAK,CAAC;QAE/B,IAAI,cAAc,EAAE,CAAC;YACnB,MAAM,KAAK,GAAG,cAAc,CAAC,eAAe,GAAG,eAAe,CAAC;YAC/D,IAAI,KAAK,GAAG,mBAAmB,EAAE,CAAC;gBAChC,kBAAkB,GAAG,IAAI,CAAC;gBAC1B,8EAA8E;gBAC9E,KAAK,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,OAAO,EAAE,CAAC;oBACxC,IAAI,MAAM,CAAC,YAAY,GAAG,cAAc,CAAC,eAAe,EAAE,CAAC;wBACzD,iBAAiB,CAAC,IAAI,CAAC;4BACrB,OAAO,EAAE,KAAK,CAAC,OAAO;4BACtB,SAAS,EAAE,KAAK,CAAC,SAAS;4BAC1B,aAAa,EAAE,cAAc,CAAC,eAAe;4BAC7C,YAAY,EAAE,MAAM,CAAC,YAAY;4BACjC,KAAK,EAAE,cAAc,CAAC,eAAe,GAAG,MAAM,CAAC,YAAY;yBAC5D,CAAC,CAAC;oBACL,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;QAED,OAAO;YACL,KAAK,EAAE,UAAU,EAAE;YACnB,SAAS;YACT,KAAK,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YAC/B,YAAY;YACZ,aAAa,EAAE,OAAO,CAAC,MAAM;YAC7B,SAAS;YACT,SAAS;YACT,eAAe;YACf,YAAY;YACZ,kBAAkB;YAClB,iBAAiB;SAClB,CAAC;IACJ,CAAC;CACF"}
@@ -0,0 +1,10 @@
1
+ /**
2
+ * Eval Module - Automated eval dataset from production traces (Task 33)
3
+ */
4
+ export { TraceCollector } from './trace-collector.js';
5
+ export type { RecordTraceInput } from './trace-collector.js';
6
+ export { DatasetManager } from './dataset-manager.js';
7
+ export type { CreateFromTracesOpts } from './dataset-manager.js';
8
+ export { DatasetRunner } from './dataset-runner.js';
9
+ export type { AgentRunnerResult, DatasetRunOpts } from './dataset-runner.js';
10
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../../src/eval/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,YAAY,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAE7D,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,YAAY,EAAE,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAEjE,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,YAAY,EAAE,iBAAiB,EAAE,cAAc,EAAE,MAAM,qBAAqB,CAAC"}
@@ -0,0 +1,7 @@
1
+ /**
2
+ * Eval Module - Automated eval dataset from production traces (Task 33)
3
+ */
4
+ export { TraceCollector } from './trace-collector.js';
5
+ export { DatasetManager } from './dataset-manager.js';
6
+ export { DatasetRunner } from './dataset-runner.js';
7
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../../src/eval/index.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAGtD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAGtD,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC"}
@@ -0,0 +1,40 @@
1
+ import type { EvalTrace } from '../../../shared/src/types/eval.js';
2
+ export interface RecordTraceInput {
3
+ agentSlug: string;
4
+ agentVersion: string;
5
+ taskDescription: string;
6
+ taskInput: string;
7
+ agentOutput: string;
8
+ retryCount: number;
9
+ qualityScore?: number;
10
+ outcome: 'success' | 'failure' | 'timeout';
11
+ latencyMs: number;
12
+ tokenCount?: number;
13
+ costUsd?: number;
14
+ correctedOutput?: string;
15
+ }
16
+ export declare class TraceCollector {
17
+ private filePath;
18
+ constructor(filePath: string);
19
+ /**
20
+ * Determine auto review status based on trace quality signals.
21
+ */
22
+ autoReviewStatus(input: RecordTraceInput): 'pending' | 'approved';
23
+ /**
24
+ * Auto-generate tags based on trace characteristics.
25
+ */
26
+ autoTag(input: RecordTraceInput): string[];
27
+ /**
28
+ * Record a trace, auto-generating traceId, capturedAt, reviewStatus, and tags.
29
+ */
30
+ record(input: RecordTraceInput): EvalTrace;
31
+ /**
32
+ * Read all traces from the JSONL file.
33
+ */
34
+ readAll(): EvalTrace[];
35
+ /**
36
+ * Get traces pending review, with optional limit.
37
+ */
38
+ getTracesPendingReview(limit?: number): EvalTrace[];
39
+ }
40
+ //# sourceMappingURL=trace-collector.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"trace-collector.d.ts","sourceRoot":"","sources":["../../../src/eval/trace-collector.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,mCAAmC,CAAC;AAGnE,MAAM,WAAW,gBAAgB;IAC/B,SAAS,EAAE,MAAM,CAAC;IAClB,YAAY,EAAE,MAAM,CAAC;IACrB,eAAe,EAAE,MAAM,CAAC;IACxB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,EAAE,SAAS,GAAG,SAAS,GAAG,SAAS,CAAC;IAC3C,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,eAAe,CAAC,EAAE,MAAM,CAAC;CAC1B;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAS;gBAEb,QAAQ,EAAE,MAAM;IAI5B;;OAEG;IACH,gBAAgB,CAAC,KAAK,EAAE,gBAAgB,GAAG,SAAS,GAAG,UAAU;IAOjE;;OAEG;IACH,OAAO,CAAC,KAAK,EAAE,gBAAgB,GAAG,MAAM,EAAE;IAQ1C;;OAEG;IACH,MAAM,CAAC,KAAK,EAAE,gBAAgB,GAAG,SAAS;IAsC1C;;OAEG;IACH,OAAO,IAAI,SAAS,EAAE;IAWtB;;OAEG;IACH,sBAAsB,CAAC,KAAK,CAAC,EAAE,MAAM,GAAG,SAAS,EAAE;CAKpD"}
@@ -0,0 +1,102 @@
1
+ /**
2
+ * TraceCollector - JSONL-based production trace collection (Task 33)
3
+ */
4
+ import { randomUUID } from 'crypto';
5
+ import { appendFileSync, readFileSync, existsSync, statSync } from 'fs';
6
+ import { parseJsonl } from '../utils/parse-jsonl.js';
7
+ export class TraceCollector {
8
+ filePath;
9
+ constructor(filePath) {
10
+ this.filePath = filePath;
11
+ }
12
+ /**
13
+ * Determine auto review status based on trace quality signals.
14
+ */
15
+ autoReviewStatus(input) {
16
+ if (input.retryCount > 1)
17
+ return 'pending';
18
+ if (input.qualityScore !== undefined && input.qualityScore < 0.6)
19
+ return 'pending';
20
+ if (input.outcome === 'failure')
21
+ return 'pending';
22
+ return 'approved';
23
+ }
24
+ /**
25
+ * Auto-generate tags based on trace characteristics.
26
+ */
27
+ autoTag(input) {
28
+ const tags = [];
29
+ if (input.retryCount > 1)
30
+ tags.push('high-retry');
31
+ if (input.outcome === 'failure')
32
+ tags.push('failure');
33
+ if (input.outcome === 'timeout')
34
+ tags.push('timeout');
35
+ return tags;
36
+ }
37
+ /**
38
+ * Record a trace, auto-generating traceId, capturedAt, reviewStatus, and tags.
39
+ */
40
+ record(input) {
41
+ const trace = {
42
+ traceId: randomUUID(),
43
+ agentSlug: input.agentSlug,
44
+ agentVersion: input.agentVersion,
45
+ taskDescription: input.taskDescription,
46
+ taskInput: input.taskInput,
47
+ agentOutput: input.agentOutput,
48
+ retryCount: input.retryCount,
49
+ qualityScore: input.qualityScore,
50
+ outcome: input.outcome,
51
+ latencyMs: input.latencyMs,
52
+ tokenCount: input.tokenCount,
53
+ costUsd: input.costUsd,
54
+ capturedAt: new Date().toISOString(),
55
+ reviewStatus: this.autoReviewStatus(input),
56
+ correctedOutput: input.correctedOutput,
57
+ tags: this.autoTag(input),
58
+ };
59
+ // Defensive serialization — agent outputs may contain circular references
60
+ // or BigInt; without this guard the writer crashes mid-trace.
61
+ let serialized;
62
+ try {
63
+ serialized = JSON.stringify(trace);
64
+ }
65
+ catch {
66
+ serialized = JSON.stringify({
67
+ traceId: trace.traceId,
68
+ agentSlug: trace.agentSlug,
69
+ capturedAt: trace.capturedAt,
70
+ reviewStatus: trace.reviewStatus,
71
+ outcome: 'serialize_failed',
72
+ });
73
+ }
74
+ appendFileSync(this.filePath, serialized + '\n', 'utf-8');
75
+ return trace;
76
+ }
77
+ /**
78
+ * Read all traces from the JSONL file.
79
+ */
80
+ readAll() {
81
+ if (!existsSync(this.filePath))
82
+ return [];
83
+ const stat = statSync(this.filePath);
84
+ if (stat.size > 256 * 1024 * 1024) {
85
+ throw new Error(`Trace file exceeds 256MB (${stat.size} bytes). Run rotation/cleanup.`);
86
+ }
87
+ const content = readFileSync(this.filePath, 'utf-8').trim();
88
+ if (!content)
89
+ return [];
90
+ return parseJsonl(content);
91
+ }
92
+ /**
93
+ * Get traces pending review, with optional limit.
94
+ */
95
+ getTracesPendingReview(limit) {
96
+ const all = this.readAll().filter((t) => t.reviewStatus === 'pending');
97
+ if (limit !== undefined)
98
+ return all.slice(0, limit);
99
+ return all;
100
+ }
101
+ }
102
+ //# sourceMappingURL=trace-collector.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"trace-collector.js","sourceRoot":"","sources":["../../../src/eval/trace-collector.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,OAAO,EAAE,UAAU,EAAE,MAAM,QAAQ,CAAC;AACpC,OAAO,EAAE,cAAc,EAAE,YAAY,EAAE,UAAU,EAAE,QAAQ,EAAE,MAAM,IAAI,CAAC;AAExE,OAAO,EAAE,UAAU,EAAE,MAAM,yBAAyB,CAAC;AAiBrD,MAAM,OAAO,cAAc;IACjB,QAAQ,CAAS;IAEzB,YAAY,QAAgB;QAC1B,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAC3B,CAAC;IAED;;OAEG;IACH,gBAAgB,CAAC,KAAuB;QACtC,IAAI,KAAK,CAAC,UAAU,GAAG,CAAC;YAAE,OAAO,SAAS,CAAC;QAC3C,IAAI,KAAK,CAAC,YAAY,KAAK,SAAS,IAAI,KAAK,CAAC,YAAY,GAAG,GAAG;YAAE,OAAO,SAAS,CAAC;QACnF,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,OAAO,SAAS,CAAC;QAClD,OAAO,UAAU,CAAC;IACpB,CAAC;IAED;;OAEG;IACH,OAAO,CAAC,KAAuB;QAC7B,MAAM,IAAI,GAAa,EAAE,CAAC;QAC1B,IAAI,KAAK,CAAC,UAAU,GAAG,CAAC;YAAE,IAAI,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;QAClD,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtD,IAAI,KAAK,CAAC,OAAO,KAAK,SAAS;YAAE,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtD,OAAO,IAAI,CAAC;IACd,CAAC;IAED;;OAEG;IACH,MAAM,CAAC,KAAuB;QAC5B,MAAM,KAAK,GAAc;YACvB,OAAO,EAAE,UAAU,EAAE;YACrB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,WAAW,EAAE,KAAK,CAAC,WAAW;YAC9B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,YAAY,EAAE,KAAK,CAAC,YAAY;YAChC,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,SAAS,EAAE,KAAK,CAAC,SAAS;YAC1B,UAAU,EAAE,KAAK,CAAC,UAAU;YAC5B,OAAO,EAAE,KAAK,CAAC,OAAO;YACtB,UAAU,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACpC,YAAY,EAAE,IAAI,CAAC,gBAAgB,CAAC,KAAK,CAAC;YAC1C,eAAe,EAAE,KAAK,CAAC,eAAe;YACtC,IAAI,EAAE,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC;SAC1B,CAAC;QAEF,0EAA0E;QAC1E,8DAA8D;QAC9D,IAAI,UAAkB,CAAC;QACvB,IAAI,CAAC;YACH,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACrC,CAAC;QAAC,MAAM,CAAC;YACP,UAAU,GAAG,IAAI,CAAC,SAAS,CAAC;gBAC1B,OAAO,EAAE,KAAK,CAAC,OAAO;gBACtB,SAAS,EAAE,KAAK,CAAC,SAAS;gBAC1B,UAAU,EAAE,KAAK,CAAC,UAAU;gBAC5B,YAAY,EAAE,KAAK,CAAC,YAAY;gBAChC,OAAO,EAAE,kBAAkB;aAC5B,CAAC,CAAC;QACL,CAAC;QACD,cAAc,CAAC,IAAI,CAAC,QAAQ,EAAE,UAAU,GAAG,IAAI,EAAE,OAAO,CAAC,CAAC;QAC1D,OAAO,KAAK,CAAC;IACf,CAAC;IAED;;OAEG;IACH,OAAO;QACL,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC;YAAE,OAAO,EAAE,CAAC;QAC1C,MAAM,IAAI,GAAG,QAAQ,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QACrC,IAAI,IAAI,CAAC,IAAI,GAAG,GAAG,GAAG,IAAI,GAAG,IAAI,EAAE,CAAC;YAClC,MAAM,IAAI,KAAK,CAAC,6BAA6B,IAAI,CAAC,IAAI,gCAAgC,CAAC,CAAC;QAC1F,CAAC;QACD,MAAM,OAAO,GAAG,YAAY,CAAC,IAAI,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC,IAAI,EAAE,CAAC;QAC5D,IAAI,CAAC,OAAO;YAAE,OAAO,EAAE,CAAC;QACxB,OAAO,UAAU,CAAY,OAAO,CAAC,CAAC;IACxC,CAAC;IAED;;OAEG;IACH,sBAAsB,CAAC,KAAc;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,SAAS,CAAC,CAAC;QACvE,IAAI,KAAK,KAAK,SAAS;YAAE,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;QACpD,OAAO,GAAG,CAAC;IACb,CAAC;CACF"}