oh-my-knowledge 0.29.0 → 0.31.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (449) hide show
  1. package/README.md +204 -25
  2. package/README.zh.md +204 -25
  3. package/dist/src/analysis/report-diagnostics.js +2 -0
  4. package/dist/src/analysis/report-diagnostics.js.map +1 -1
  5. package/dist/src/authoring/evolver.d.ts +32 -2
  6. package/dist/src/authoring/evolver.d.ts.map +1 -1
  7. package/dist/src/authoring/evolver.js +416 -52
  8. package/dist/src/authoring/evolver.js.map +1 -1
  9. package/dist/src/authoring/generator.d.ts +24 -23
  10. package/dist/src/authoring/generator.d.ts.map +1 -1
  11. package/dist/src/authoring/generator.js +637 -40
  12. package/dist/src/authoring/generator.js.map +1 -1
  13. package/dist/src/authoring/sample-fixer.d.ts +51 -0
  14. package/dist/src/authoring/sample-fixer.d.ts.map +1 -0
  15. package/dist/src/authoring/sample-fixer.js +213 -0
  16. package/dist/src/authoring/sample-fixer.js.map +1 -0
  17. package/dist/src/cli/commands/doctor.d.ts +25 -1
  18. package/dist/src/cli/commands/doctor.d.ts.map +1 -1
  19. package/dist/src/cli/commands/doctor.js +233 -124
  20. package/dist/src/cli/commands/doctor.js.map +1 -1
  21. package/dist/src/cli/commands/eval/gold/compare.d.ts +17 -0
  22. package/dist/src/cli/commands/eval/gold/compare.d.ts.map +1 -0
  23. package/dist/src/cli/commands/eval/gold/compare.js +91 -0
  24. package/dist/src/cli/commands/eval/gold/compare.js.map +1 -0
  25. package/dist/src/cli/commands/eval/gold/index.d.ts +9 -0
  26. package/dist/src/cli/commands/eval/gold/index.d.ts.map +1 -0
  27. package/dist/src/cli/commands/eval/gold/index.js +34 -0
  28. package/dist/src/cli/commands/eval/gold/index.js.map +1 -0
  29. package/dist/src/cli/commands/eval/gold/init.d.ts +11 -0
  30. package/dist/src/cli/commands/eval/gold/init.d.ts.map +1 -0
  31. package/dist/src/cli/commands/eval/gold/init.js +51 -0
  32. package/dist/src/cli/commands/eval/gold/init.js.map +1 -0
  33. package/dist/src/cli/commands/eval/gold/validate.d.ts +12 -0
  34. package/dist/src/cli/commands/eval/gold/validate.d.ts.map +1 -0
  35. package/dist/src/cli/commands/eval/gold/validate.js +46 -0
  36. package/dist/src/cli/commands/eval/gold/validate.js.map +1 -0
  37. package/dist/src/cli/commands/eval/index.d.ts +54 -0
  38. package/dist/src/cli/commands/eval/index.d.ts.map +1 -0
  39. package/dist/src/cli/commands/{eval-runner.js → eval/index.js} +216 -50
  40. package/dist/src/cli/commands/eval/index.js.map +1 -0
  41. package/dist/src/cli/commands/evolve.d.ts +36 -1
  42. package/dist/src/cli/commands/evolve.d.ts.map +1 -1
  43. package/dist/src/cli/commands/evolve.js +202 -41
  44. package/dist/src/cli/commands/evolve.js.map +1 -1
  45. package/dist/src/cli/commands/init.d.ts +15 -1
  46. package/dist/src/cli/commands/init.d.ts.map +1 -1
  47. package/dist/src/cli/commands/init.js +70 -28
  48. package/dist/src/cli/commands/init.js.map +1 -1
  49. package/dist/src/cli/commands/observe/inbox.d.ts +23 -0
  50. package/dist/src/cli/commands/observe/inbox.d.ts.map +1 -0
  51. package/dist/src/cli/commands/observe/inbox.js +258 -0
  52. package/dist/src/cli/commands/observe/inbox.js.map +1 -0
  53. package/dist/src/cli/commands/observe/index.d.ts +22 -0
  54. package/dist/src/cli/commands/observe/index.d.ts.map +1 -0
  55. package/dist/src/cli/commands/observe/index.js +118 -0
  56. package/dist/src/cli/commands/observe/index.js.map +1 -0
  57. package/dist/src/cli/commands/observe/ingest.d.ts +13 -0
  58. package/dist/src/cli/commands/observe/ingest.d.ts.map +1 -0
  59. package/dist/src/cli/commands/observe/ingest.js +69 -0
  60. package/dist/src/cli/commands/observe/ingest.js.map +1 -0
  61. package/dist/src/cli/commands/observe/show.d.ts +13 -0
  62. package/dist/src/cli/commands/observe/show.d.ts.map +1 -0
  63. package/dist/src/cli/commands/observe/show.js +49 -0
  64. package/dist/src/cli/commands/observe/show.js.map +1 -0
  65. package/dist/src/cli/commands/sample.d.ts +37 -1
  66. package/dist/src/cli/commands/sample.d.ts.map +1 -1
  67. package/dist/src/cli/commands/sample.js +402 -35
  68. package/dist/src/cli/commands/sample.js.map +1 -1
  69. package/dist/src/cli/commands/skill-extract.d.ts +20 -0
  70. package/dist/src/cli/commands/skill-extract.d.ts.map +1 -0
  71. package/dist/src/cli/commands/skill-extract.js +84 -0
  72. package/dist/src/cli/commands/skill-extract.js.map +1 -0
  73. package/dist/src/cli/commands/studio.d.ts +22 -1
  74. package/dist/src/cli/commands/studio.d.ts.map +1 -1
  75. package/dist/src/cli/commands/studio.js +111 -37
  76. package/dist/src/cli/commands/studio.js.map +1 -1
  77. package/dist/src/cli/index.js +15 -47
  78. package/dist/src/cli/index.js.map +1 -1
  79. package/dist/src/cli/lib/cli-exit.d.ts +16 -0
  80. package/dist/src/cli/lib/cli-exit.d.ts.map +1 -0
  81. package/dist/src/cli/lib/cli-exit.js +20 -0
  82. package/dist/src/cli/lib/cli-exit.js.map +1 -0
  83. package/dist/src/cli/lib/cmd-flags.d.ts +236 -0
  84. package/dist/src/cli/lib/cmd-flags.d.ts.map +1 -0
  85. package/dist/src/cli/lib/cmd-flags.js +46 -0
  86. package/dist/src/cli/lib/cmd-flags.js.map +1 -0
  87. package/dist/src/cli/{i18n-dict.d.ts → lib/i18n-dict.d.ts} +1 -1
  88. package/dist/src/cli/lib/i18n-dict.d.ts.map +1 -0
  89. package/dist/src/cli/{i18n-dict.js → lib/i18n-dict.js} +93 -260
  90. package/dist/src/cli/lib/i18n-dict.js.map +1 -0
  91. package/dist/src/cli/{i18n.d.ts → lib/i18n.d.ts} +11 -6
  92. package/dist/src/cli/lib/i18n.d.ts.map +1 -0
  93. package/dist/src/cli/{i18n.js → lib/i18n.js} +12 -7
  94. package/dist/src/cli/lib/i18n.js.map +1 -0
  95. package/dist/src/cli/{parse-run-config.d.ts → lib/parse-run-config.d.ts} +19 -8
  96. package/dist/src/cli/lib/parse-run-config.d.ts.map +1 -0
  97. package/dist/src/cli/{parse-run-config.js → lib/parse-run-config.js} +60 -58
  98. package/dist/src/cli/lib/parse-run-config.js.map +1 -0
  99. package/dist/src/cli/lib/progress.d.ts.map +1 -0
  100. package/dist/src/cli/lib/progress.js.map +1 -0
  101. package/dist/src/cli/{run-tally.d.ts → lib/run-tally.d.ts} +1 -1
  102. package/dist/src/cli/lib/run-tally.d.ts.map +1 -0
  103. package/dist/src/cli/lib/run-tally.js.map +1 -0
  104. package/dist/src/cli/{commands/_shared.d.ts → lib/shared.d.ts} +2 -2
  105. package/dist/src/cli/lib/shared.d.ts.map +1 -0
  106. package/dist/src/cli/{commands/_shared.js → lib/shared.js} +3 -3
  107. package/dist/src/cli/lib/shared.js.map +1 -0
  108. package/dist/src/cli/lib/update-check.d.ts.map +1 -0
  109. package/dist/src/cli/lib/update-check.js.map +1 -0
  110. package/dist/src/cli/oclif/base-command.d.ts +8 -0
  111. package/dist/src/cli/oclif/base-command.d.ts.map +1 -0
  112. package/dist/src/cli/oclif/base-command.js +27 -0
  113. package/dist/src/cli/oclif/base-command.js.map +1 -0
  114. package/dist/src/cli/oclif/help.d.ts +16 -0
  115. package/dist/src/cli/oclif/help.d.ts.map +1 -0
  116. package/dist/src/cli/oclif/help.js +33 -0
  117. package/dist/src/cli/oclif/help.js.map +1 -0
  118. package/dist/src/cli/oclif/i18n.d.ts +16 -0
  119. package/dist/src/cli/oclif/i18n.d.ts.map +1 -0
  120. package/dist/src/cli/oclif/i18n.js +62 -0
  121. package/dist/src/cli/oclif/i18n.js.map +1 -0
  122. package/dist/src/cli/oclif/parsers.d.ts +10 -0
  123. package/dist/src/cli/oclif/parsers.d.ts.map +1 -0
  124. package/dist/src/cli/oclif/parsers.js +102 -0
  125. package/dist/src/cli/oclif/parsers.js.map +1 -0
  126. package/dist/src/cli/oclif/projection.d.ts +23 -0
  127. package/dist/src/cli/oclif/projection.d.ts.map +1 -0
  128. package/dist/src/cli/oclif/projection.js +52 -0
  129. package/dist/src/cli/oclif/projection.js.map +1 -0
  130. package/dist/src/cli/oclif/run.d.ts +2 -0
  131. package/dist/src/cli/oclif/run.d.ts.map +1 -0
  132. package/dist/src/cli/oclif/run.js +11 -0
  133. package/dist/src/cli/oclif/run.js.map +1 -0
  134. package/dist/src/diagnosis/observe-mapper.d.ts +68 -0
  135. package/dist/src/diagnosis/observe-mapper.d.ts.map +1 -0
  136. package/dist/src/diagnosis/observe-mapper.js +276 -0
  137. package/dist/src/diagnosis/observe-mapper.js.map +1 -0
  138. package/dist/src/diagnosis/observe-producer.d.ts +9 -0
  139. package/dist/src/diagnosis/observe-producer.d.ts.map +1 -0
  140. package/dist/src/diagnosis/observe-producer.js +197 -0
  141. package/dist/src/diagnosis/observe-producer.js.map +1 -0
  142. package/dist/src/diagnosis/studio-projection.d.ts +14 -0
  143. package/dist/src/diagnosis/studio-projection.d.ts.map +1 -0
  144. package/dist/src/diagnosis/studio-projection.js +84 -0
  145. package/dist/src/diagnosis/studio-projection.js.map +1 -0
  146. package/dist/src/diagnosis/types.d.ts +79 -0
  147. package/dist/src/diagnosis/types.d.ts.map +1 -0
  148. package/dist/src/diagnosis/types.js +20 -0
  149. package/dist/src/diagnosis/types.js.map +1 -0
  150. package/dist/src/doctor/fixer.d.ts +12 -0
  151. package/dist/src/doctor/fixer.d.ts.map +1 -0
  152. package/dist/src/doctor/fixer.js +326 -0
  153. package/dist/src/doctor/fixer.js.map +1 -0
  154. package/dist/src/doctor/health/builtin-dimensions.d.ts.map +1 -1
  155. package/dist/src/doctor/health/builtin-dimensions.js +20 -23
  156. package/dist/src/doctor/health/builtin-dimensions.js.map +1 -1
  157. package/dist/src/doctor/health/composer.d.ts.map +1 -1
  158. package/dist/src/doctor/health/composer.js +14 -7
  159. package/dist/src/doctor/health/composer.js.map +1 -1
  160. package/dist/src/doctor/health/dimension-spec.d.ts +1 -0
  161. package/dist/src/doctor/health/dimension-spec.d.ts.map +1 -1
  162. package/dist/src/doctor/health/parser.d.ts +0 -1
  163. package/dist/src/doctor/health/parser.d.ts.map +1 -1
  164. package/dist/src/doctor/health/parser.js +38 -0
  165. package/dist/src/doctor/health/parser.js.map +1 -1
  166. package/dist/src/doctor/health/prompt-builder.d.ts +2 -21
  167. package/dist/src/doctor/health/prompt-builder.d.ts.map +1 -1
  168. package/dist/src/doctor/health/prompt-builder.js +1 -161
  169. package/dist/src/doctor/health/prompt-builder.js.map +1 -1
  170. package/dist/src/doctor/html-renderer.d.ts +1 -1
  171. package/dist/src/doctor/html-renderer.d.ts.map +1 -1
  172. package/dist/src/doctor/html-renderer.js +78 -68
  173. package/dist/src/doctor/html-renderer.js.map +1 -1
  174. package/dist/src/doctor/index.d.ts.map +1 -1
  175. package/dist/src/doctor/index.js +1 -0
  176. package/dist/src/doctor/index.js.map +1 -1
  177. package/dist/src/doctor/renderer.d.ts +1 -1
  178. package/dist/src/doctor/renderer.d.ts.map +1 -1
  179. package/dist/src/doctor/renderer.js +2 -2
  180. package/dist/src/doctor/renderer.js.map +1 -1
  181. package/dist/src/doctor/rules.d.ts.map +1 -1
  182. package/dist/src/doctor/rules.js +33 -1
  183. package/dist/src/doctor/rules.js.map +1 -1
  184. package/dist/src/eval-core/cache.d.ts +16 -2
  185. package/dist/src/eval-core/cache.d.ts.map +1 -1
  186. package/dist/src/eval-core/cache.js +86 -15
  187. package/dist/src/eval-core/cache.js.map +1 -1
  188. package/dist/src/eval-core/evaluation-execution.d.ts +11 -1
  189. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
  190. package/dist/src/eval-core/evaluation-execution.js +78 -6
  191. package/dist/src/eval-core/evaluation-execution.js.map +1 -1
  192. package/dist/src/eval-core/evaluation-job.d.ts +2 -1
  193. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
  194. package/dist/src/eval-core/evaluation-job.js +2 -1
  195. package/dist/src/eval-core/evaluation-job.js.map +1 -1
  196. package/dist/src/eval-core/evaluation-reporting.d.ts +7 -0
  197. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
  198. package/dist/src/eval-core/evaluation-reporting.js +19 -2
  199. package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
  200. package/dist/src/eval-core/execution-strategy.d.ts +1 -1
  201. package/dist/src/eval-core/execution-strategy.d.ts.map +1 -1
  202. package/dist/src/eval-core/execution-strategy.js +12 -1
  203. package/dist/src/eval-core/execution-strategy.js.map +1 -1
  204. package/dist/src/eval-core/mock-hook.cjs +215 -0
  205. package/dist/src/eval-core/mocks-runtime.d.ts +100 -0
  206. package/dist/src/eval-core/mocks-runtime.d.ts.map +1 -0
  207. package/dist/src/eval-core/mocks-runtime.js +559 -0
  208. package/dist/src/eval-core/mocks-runtime.js.map +1 -0
  209. package/dist/src/eval-core/schema.d.ts.map +1 -1
  210. package/dist/src/eval-core/schema.js +18 -5
  211. package/dist/src/eval-core/schema.js.map +1 -1
  212. package/dist/src/eval-core/task-planner.d.ts +9 -1
  213. package/dist/src/eval-core/task-planner.d.ts.map +1 -1
  214. package/dist/src/eval-core/task-planner.js +40 -1
  215. package/dist/src/eval-core/task-planner.js.map +1 -1
  216. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +5 -1
  217. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -1
  218. package/dist/src/eval-workflows/batch-evaluation-workflow.js +2 -1
  219. package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -1
  220. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +23 -2
  221. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
  222. package/dist/src/eval-workflows/evaluation-pipeline.js +40 -13
  223. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
  224. package/dist/src/eval-workflows/evaluation-preparation.d.ts +5 -0
  225. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
  226. package/dist/src/eval-workflows/evaluation-preparation.js +3 -1
  227. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
  228. package/dist/src/eval-workflows/run-evaluation.d.ts +12 -2
  229. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
  230. package/dist/src/eval-workflows/run-evaluation.js +18 -6
  231. package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
  232. package/dist/src/executors/claude-cli.d.ts +1 -1
  233. package/dist/src/executors/claude-cli.d.ts.map +1 -1
  234. package/dist/src/executors/claude-cli.js +59 -6
  235. package/dist/src/executors/claude-cli.js.map +1 -1
  236. package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -1
  237. package/dist/src/executors/claude-sdk-trace.js +10 -4
  238. package/dist/src/executors/claude-sdk-trace.js.map +1 -1
  239. package/dist/src/executors/claude-sdk.d.ts +1 -1
  240. package/dist/src/executors/claude-sdk.d.ts.map +1 -1
  241. package/dist/src/executors/claude-sdk.js +25 -2
  242. package/dist/src/executors/claude-sdk.js.map +1 -1
  243. package/dist/src/executors/shared.d.ts +1 -1
  244. package/dist/src/executors/shared.d.ts.map +1 -1
  245. package/dist/src/executors/shared.js +6 -1
  246. package/dist/src/executors/shared.js.map +1 -1
  247. package/dist/src/grading/assertions.d.ts +5 -0
  248. package/dist/src/grading/assertions.d.ts.map +1 -1
  249. package/dist/src/grading/assertions.js +27 -1
  250. package/dist/src/grading/assertions.js.map +1 -1
  251. package/dist/src/grading/diagnostic.d.ts +40 -0
  252. package/dist/src/grading/diagnostic.d.ts.map +1 -0
  253. package/dist/src/grading/diagnostic.js +450 -0
  254. package/dist/src/grading/diagnostic.js.map +1 -0
  255. package/dist/src/grading/index.d.ts +5 -0
  256. package/dist/src/grading/index.d.ts.map +1 -1
  257. package/dist/src/grading/judge.d.ts.map +1 -1
  258. package/dist/src/grading/judge.js +54 -2
  259. package/dist/src/grading/judge.js.map +1 -1
  260. package/dist/src/grading/layered-scores.d.ts.map +1 -1
  261. package/dist/src/grading/layered-scores.js +1 -0
  262. package/dist/src/grading/layered-scores.js.map +1 -1
  263. package/dist/src/inputs/eval-config.js +19 -0
  264. package/dist/src/inputs/eval-config.js.map +1 -1
  265. package/dist/src/inputs/load-samples.d.ts +22 -2
  266. package/dist/src/inputs/load-samples.d.ts.map +1 -1
  267. package/dist/src/inputs/load-samples.js +193 -22
  268. package/dist/src/inputs/load-samples.js.map +1 -1
  269. package/dist/src/inputs/skill-loader.d.ts.map +1 -1
  270. package/dist/src/inputs/skill-loader.js +11 -3
  271. package/dist/src/inputs/skill-loader.js.map +1 -1
  272. package/dist/src/observability/experience.d.ts +441 -0
  273. package/dist/src/observability/experience.d.ts.map +1 -0
  274. package/dist/src/observability/experience.js +2400 -0
  275. package/dist/src/observability/experience.js.map +1 -0
  276. package/dist/src/observability/inbox-view-model.d.ts +14 -1
  277. package/dist/src/observability/inbox-view-model.d.ts.map +1 -1
  278. package/dist/src/observability/inbox-view-model.js +85 -3
  279. package/dist/src/observability/inbox-view-model.js.map +1 -1
  280. package/dist/src/observability/inbox.d.ts +29 -2
  281. package/dist/src/observability/inbox.d.ts.map +1 -1
  282. package/dist/src/observability/inbox.js +62 -10
  283. package/dist/src/observability/inbox.js.map +1 -1
  284. package/dist/src/observability/problem-patterns.d.ts +52 -0
  285. package/dist/src/observability/problem-patterns.d.ts.map +1 -0
  286. package/dist/src/observability/problem-patterns.js +205 -0
  287. package/dist/src/observability/problem-patterns.js.map +1 -0
  288. package/dist/src/observability/resolved-review.d.ts +25 -0
  289. package/dist/src/observability/resolved-review.d.ts.map +1 -0
  290. package/dist/src/observability/resolved-review.js +231 -0
  291. package/dist/src/observability/resolved-review.js.map +1 -0
  292. package/dist/src/observability/review-state.d.ts +61 -0
  293. package/dist/src/observability/review-state.d.ts.map +1 -0
  294. package/dist/src/observability/review-state.js +169 -0
  295. package/dist/src/observability/review-state.js.map +1 -0
  296. package/dist/src/observability/skill-chain-advisories.d.ts +25 -0
  297. package/dist/src/observability/skill-chain-advisories.d.ts.map +1 -0
  298. package/dist/src/observability/skill-chain-advisories.js +69 -0
  299. package/dist/src/observability/skill-chain-advisories.js.map +1 -0
  300. package/dist/src/observability/skill-chain.d.ts +62 -0
  301. package/dist/src/observability/skill-chain.d.ts.map +1 -0
  302. package/dist/src/observability/skill-chain.js +270 -0
  303. package/dist/src/observability/skill-chain.js.map +1 -0
  304. package/dist/src/observability/soft-standards.d.ts +130 -0
  305. package/dist/src/observability/soft-standards.d.ts.map +1 -0
  306. package/dist/src/observability/soft-standards.js +426 -0
  307. package/dist/src/observability/soft-standards.js.map +1 -0
  308. package/dist/src/observability/text-signals.d.ts +14 -0
  309. package/dist/src/observability/text-signals.d.ts.map +1 -0
  310. package/dist/src/observability/text-signals.js +106 -0
  311. package/dist/src/observability/text-signals.js.map +1 -0
  312. package/dist/src/observability/trace-adapter.d.ts +1 -1
  313. package/dist/src/observability/trace-adapter.d.ts.map +1 -1
  314. package/dist/src/observability/trace-adapter.js +1 -1
  315. package/dist/src/observability/trace-adapter.js.map +1 -1
  316. package/dist/src/observability/trace-attribution.d.ts +29 -1
  317. package/dist/src/observability/trace-attribution.d.ts.map +1 -1
  318. package/dist/src/observability/trace-attribution.js +106 -10
  319. package/dist/src/observability/trace-attribution.js.map +1 -1
  320. package/dist/src/observability/trace-segmenter.d.ts +11 -1
  321. package/dist/src/observability/trace-segmenter.d.ts.map +1 -1
  322. package/dist/src/observability/trace-segmenter.js +121 -30
  323. package/dist/src/observability/trace-segmenter.js.map +1 -1
  324. package/dist/src/observability/trace-source.d.ts +28 -1
  325. package/dist/src/observability/trace-source.d.ts.map +1 -1
  326. package/dist/src/observability/trace-source.js +366 -6
  327. package/dist/src/observability/trace-source.js.map +1 -1
  328. package/dist/src/renderer/html-renderer.d.ts.map +1 -1
  329. package/dist/src/renderer/html-renderer.js +229 -171
  330. package/dist/src/renderer/html-renderer.js.map +1 -1
  331. package/dist/src/renderer/layout.d.ts.map +1 -1
  332. package/dist/src/renderer/layout.js +108 -61
  333. package/dist/src/renderer/layout.js.map +1 -1
  334. package/dist/src/renderer/observation-inbox-renderer.d.ts.map +1 -1
  335. package/dist/src/renderer/observation-inbox-renderer.js +9551 -85
  336. package/dist/src/renderer/observation-inbox-renderer.js.map +1 -1
  337. package/dist/src/renderer/skill-detail-renderer.d.ts +15 -0
  338. package/dist/src/renderer/skill-detail-renderer.d.ts.map +1 -0
  339. package/dist/src/renderer/skill-detail-renderer.js +1244 -0
  340. package/dist/src/renderer/skill-detail-renderer.js.map +1 -0
  341. package/dist/src/renderer/skill-list-renderer.d.ts +4 -0
  342. package/dist/src/renderer/skill-list-renderer.d.ts.map +1 -0
  343. package/dist/src/renderer/skill-list-renderer.js +334 -0
  344. package/dist/src/renderer/skill-list-renderer.js.map +1 -0
  345. package/dist/src/renderer/summary.d.ts.map +1 -1
  346. package/dist/src/renderer/summary.js +78 -50
  347. package/dist/src/renderer/summary.js.map +1 -1
  348. package/dist/src/renderer/table.d.ts.map +1 -1
  349. package/dist/src/renderer/table.js +213 -167
  350. package/dist/src/renderer/table.js.map +1 -1
  351. package/dist/src/renderer/test-view.d.ts +66 -0
  352. package/dist/src/renderer/test-view.d.ts.map +1 -0
  353. package/dist/src/renderer/test-view.js +905 -0
  354. package/dist/src/renderer/test-view.js.map +1 -0
  355. package/dist/src/server/report-server.d.ts +4 -1
  356. package/dist/src/server/report-server.d.ts.map +1 -1
  357. package/dist/src/server/report-server.js +242 -4
  358. package/dist/src/server/report-server.js.map +1 -1
  359. package/dist/src/server/report-store.d.ts.map +1 -1
  360. package/dist/src/server/report-store.js +33 -1
  361. package/dist/src/server/report-store.js.map +1 -1
  362. package/dist/src/server/skill-index.d.ts +81 -0
  363. package/dist/src/server/skill-index.d.ts.map +1 -0
  364. package/dist/src/server/skill-index.js +377 -0
  365. package/dist/src/server/skill-index.js.map +1 -0
  366. package/dist/src/server/skill-insights.d.ts +101 -0
  367. package/dist/src/server/skill-insights.d.ts.map +1 -0
  368. package/dist/src/server/skill-insights.js +788 -0
  369. package/dist/src/server/skill-insights.js.map +1 -0
  370. package/dist/src/shared/hard-rules.d.ts +42 -0
  371. package/dist/src/shared/hard-rules.d.ts.map +1 -0
  372. package/dist/src/shared/hard-rules.js +242 -0
  373. package/dist/src/shared/hard-rules.js.map +1 -0
  374. package/dist/src/shared/llm-prompts/index.d.ts +14 -0
  375. package/dist/src/shared/llm-prompts/index.d.ts.map +1 -0
  376. package/dist/src/shared/llm-prompts/index.js +35 -0
  377. package/dist/src/shared/llm-prompts/index.js.map +1 -0
  378. package/dist/src/shared/llm-prompts/skill-health.d.ts +22 -0
  379. package/dist/src/shared/llm-prompts/skill-health.d.ts.map +1 -0
  380. package/dist/src/shared/llm-prompts/skill-health.js +170 -0
  381. package/dist/src/shared/llm-prompts/skill-health.js.map +1 -0
  382. package/dist/src/shared/time.d.ts +2 -0
  383. package/dist/src/shared/time.d.ts.map +1 -0
  384. package/dist/src/shared/time.js +10 -0
  385. package/dist/src/shared/time.js.map +1 -0
  386. package/dist/src/types/doctor.d.ts +2 -0
  387. package/dist/src/types/doctor.d.ts.map +1 -1
  388. package/dist/src/types/eval.d.ts +88 -0
  389. package/dist/src/types/eval.d.ts.map +1 -1
  390. package/dist/src/types/executor.d.ts +43 -0
  391. package/dist/src/types/executor.d.ts.map +1 -1
  392. package/dist/src/types/judge.d.ts +76 -0
  393. package/dist/src/types/judge.d.ts.map +1 -1
  394. package/dist/src/types/judge.js +3 -1
  395. package/dist/src/types/judge.js.map +1 -1
  396. package/dist/src/types/report.d.ts +69 -3
  397. package/dist/src/types/report.d.ts.map +1 -1
  398. package/dist/src/util/safe-slice.d.ts +23 -0
  399. package/dist/src/util/safe-slice.d.ts.map +1 -0
  400. package/dist/src/util/safe-slice.js +33 -0
  401. package/dist/src/util/safe-slice.js.map +1 -0
  402. package/docs/prompts/llm-enhanced-review.prompt.md +111 -0
  403. package/package.json +33 -7
  404. package/dist/src/cli/cli-exit.d.ts +0 -15
  405. package/dist/src/cli/cli-exit.d.ts.map +0 -1
  406. package/dist/src/cli/cli-exit.js +0 -19
  407. package/dist/src/cli/cli-exit.js.map +0 -1
  408. package/dist/src/cli/commands/_shared.d.ts.map +0 -1
  409. package/dist/src/cli/commands/_shared.js.map +0 -1
  410. package/dist/src/cli/commands/eval-gold.d.ts +0 -2
  411. package/dist/src/cli/commands/eval-gold.d.ts.map +0 -1
  412. package/dist/src/cli/commands/eval-gold.js +0 -137
  413. package/dist/src/cli/commands/eval-gold.js.map +0 -1
  414. package/dist/src/cli/commands/eval-runner.d.ts +0 -2
  415. package/dist/src/cli/commands/eval-runner.d.ts.map +0 -1
  416. package/dist/src/cli/commands/eval-runner.js.map +0 -1
  417. package/dist/src/cli/commands/eval.d.ts +0 -2
  418. package/dist/src/cli/commands/eval.d.ts.map +0 -1
  419. package/dist/src/cli/commands/eval.js +0 -11
  420. package/dist/src/cli/commands/eval.js.map +0 -1
  421. package/dist/src/cli/commands/observe.d.ts +0 -2
  422. package/dist/src/cli/commands/observe.d.ts.map +0 -1
  423. package/dist/src/cli/commands/observe.js +0 -246
  424. package/dist/src/cli/commands/observe.js.map +0 -1
  425. package/dist/src/cli/commands/registry.d.ts +0 -11
  426. package/dist/src/cli/commands/registry.d.ts.map +0 -1
  427. package/dist/src/cli/commands/registry.js +0 -32
  428. package/dist/src/cli/commands/registry.js.map +0 -1
  429. package/dist/src/cli/i18n-dict.d.ts.map +0 -1
  430. package/dist/src/cli/i18n-dict.js.map +0 -1
  431. package/dist/src/cli/i18n.d.ts.map +0 -1
  432. package/dist/src/cli/i18n.js.map +0 -1
  433. package/dist/src/cli/parse-run-config.d.ts.map +0 -1
  434. package/dist/src/cli/parse-run-config.js.map +0 -1
  435. package/dist/src/cli/parse-strict.d.ts +0 -7
  436. package/dist/src/cli/parse-strict.d.ts.map +0 -1
  437. package/dist/src/cli/parse-strict.js +0 -26
  438. package/dist/src/cli/parse-strict.js.map +0 -1
  439. package/dist/src/cli/progress.d.ts.map +0 -1
  440. package/dist/src/cli/progress.js.map +0 -1
  441. package/dist/src/cli/run-tally.d.ts.map +0 -1
  442. package/dist/src/cli/run-tally.js.map +0 -1
  443. package/dist/src/cli/update-check.d.ts.map +0 -1
  444. package/dist/src/cli/update-check.js.map +0 -1
  445. /package/dist/src/cli/{progress.d.ts → lib/progress.d.ts} +0 -0
  446. /package/dist/src/cli/{progress.js → lib/progress.js} +0 -0
  447. /package/dist/src/cli/{run-tally.js → lib/run-tally.js} +0 -0
  448. /package/dist/src/cli/{update-check.d.ts → lib/update-check.d.ts} +0 -0
  449. /package/dist/src/cli/{update-check.js → lib/update-check.js} +0 -0
@@ -0,0 +1,559 @@
1
+ /**
2
+ * Mock runtime — Sample.mocks 落地的核心逻辑。
3
+ *
4
+ * 两条路径:
5
+ * 1) `buildSdkHookCallback(mocks, baseDir)`
6
+ * → 给 claude-sdk executor 用,返回 in-process PreToolUse HookCallback。
7
+ * 纯内存,零 IO,零 spawn 开销。
8
+ *
9
+ * 2) `materializeForCliConfigDir(mocks, baseDir)`
10
+ * → 给 claude-cli executor 用,mktemp 一个临时 CLAUDE_CONFIG_DIR,
11
+ * 在里面写 settings.json + on-disk hook 脚本 + mocks.json,返回 cleanup。
12
+ * claude 子进程退出后(成功/失败/超时/SIGINT),cleanup 必删整个临时目录。
13
+ *
14
+ * 共用核心:`matchMock(mocks, toolName, toolInput, callCounter)` 命中规则评估。
15
+ */
16
+ import { mkdtempSync, rmSync, writeFileSync, mkdirSync, readFileSync, existsSync } from 'node:fs';
17
+ import { homedir, tmpdir } from 'node:os';
18
+ import { dirname, isAbsolute, join, resolve } from 'node:path';
19
+ import { fileURLToPath } from 'node:url';
20
+ // ─── Match logic ────────────────────────────────────────────────────────────
21
+ function expandHome(p) {
22
+ if (p.startsWith('~/'))
23
+ return join(homedir(), p.slice(2));
24
+ if (p === '~')
25
+ return homedir();
26
+ return p;
27
+ }
28
+ /** file_path 后缀匹配:actual 完全等于 suffix,或 actual 在 path-separator 后以 suffix 结尾。
29
+ * 避免 'state.json' 误匹配 'bad-state.json'。两端都先 expandHome。
30
+ * 例:suffix='tasks/foo/state.json' 命中 'tasks/foo/state.json' / '/abs/x/tasks/foo/state.json' /
31
+ * 'C:\\proj\\tasks\\foo\\state.json' (Windows backslash 也接受)。 */
32
+ function matchesFilePathSuffix(actualRaw, suffixRaw) {
33
+ const actual = expandHome(actualRaw);
34
+ const suffix = expandHome(suffixRaw);
35
+ if (actual === suffix)
36
+ return true;
37
+ if (!actual.endsWith(suffix))
38
+ return false;
39
+ const boundaryIdx = actual.length - suffix.length - 1;
40
+ const sep = actual.charAt(boundaryIdx);
41
+ return sep === '/' || sep === '\\';
42
+ }
43
+ /** glob 匹配(只支持 `*` 通配,不支持 `**` / `?` / `[...]`,够用且无依赖)。
44
+ * 使用 dotAll 标志(`s`),让 `*` 也能跨换行匹配 — LLM 经常用反斜杠续行写多行 bash,
45
+ * 不带 `s` 时单行 glob 会全部漏掉(integration-tool-deploy eval 上首次发现这个坑)。 */
46
+ function globMatch(pattern, value) {
47
+ const escaped = pattern.replace(/[.+?^${}()|[\]\\]/g, '\\$&');
48
+ const re = new RegExp('^' + escaped.replace(/\*/g, '.*') + '$', 's');
49
+ return re.test(value);
50
+ }
51
+ /** deep equal 子集匹配:expected 的每个 key/value 在 actual 中存在且相等(actual 可有更多字段)。 */
52
+ /** 数组 deep-equal:长度相等 + 每个元素递归比较(支持嵌套数组 / object)。 */
53
+ function arraysDeepEqual(a, b) {
54
+ if (a.length !== b.length)
55
+ return false;
56
+ for (let i = 0; i < a.length; i++) {
57
+ if (!deepEqual(a[i], b[i]))
58
+ return false;
59
+ }
60
+ return true;
61
+ }
62
+ function deepEqual(a, b) {
63
+ if (a === b)
64
+ return true;
65
+ if (Array.isArray(a) && Array.isArray(b))
66
+ return arraysDeepEqual(a, b);
67
+ if (typeof a === 'object' && a !== null && typeof b === 'object' && b !== null
68
+ && !Array.isArray(a) && !Array.isArray(b)) {
69
+ const ao = a;
70
+ const bo = b;
71
+ const ak = Object.keys(ao);
72
+ const bk = Object.keys(bo);
73
+ if (ak.length !== bk.length)
74
+ return false;
75
+ for (const k of ak) {
76
+ if (!deepEqual(ao[k], bo[k]))
77
+ return false;
78
+ }
79
+ return true;
80
+ }
81
+ return false;
82
+ }
83
+ function matchesInputSubset(expected, actual) {
84
+ if (typeof actual !== 'object' || actual === null)
85
+ return false;
86
+ const a = actual;
87
+ for (const [k, v] of Object.entries(expected)) {
88
+ if (Array.isArray(v)) {
89
+ // 数组 deep-equal,不要靠 === 引用比较 — 不同 sample 实例的同样数组永远不会等
90
+ if (!Array.isArray(a[k]) || !arraysDeepEqual(v, a[k]))
91
+ return false;
92
+ }
93
+ else if (typeof v === 'object' && v !== null) {
94
+ if (!matchesInputSubset(v, a[k]))
95
+ return false;
96
+ }
97
+ else if (a[k] !== v) {
98
+ return false;
99
+ }
100
+ }
101
+ return true;
102
+ }
103
+ function anyStringContains(obj, needle) {
104
+ if (typeof obj === 'string')
105
+ return obj.toLowerCase().includes(needle);
106
+ if (Array.isArray(obj))
107
+ return obj.some((item) => anyStringContains(item, needle));
108
+ if (typeof obj === 'object' && obj !== null)
109
+ return Object.values(obj).some((v) => anyStringContains(v, needle));
110
+ return false;
111
+ }
112
+ /** 单条 mock 是否命中给定 tool 调用。 */
113
+ export function isMockHit(mock, toolName, toolInput) {
114
+ if (mock.tool !== '*' && mock.tool !== toolName)
115
+ return false;
116
+ const m = mock.match;
117
+ if (!m)
118
+ return true;
119
+ const ti = (toolInput || {});
120
+ if (m.file_path !== undefined) {
121
+ if (typeof ti.file_path !== 'string')
122
+ return false;
123
+ if (expandHome(ti.file_path) !== expandHome(m.file_path))
124
+ return false;
125
+ }
126
+ if (m.file_path_endswith !== undefined) {
127
+ if (typeof ti.file_path !== 'string')
128
+ return false;
129
+ if (!matchesFilePathSuffix(ti.file_path, m.file_path_endswith))
130
+ return false;
131
+ }
132
+ if (m.url !== undefined) {
133
+ if (ti.url !== m.url)
134
+ return false;
135
+ }
136
+ if (m.url_glob !== undefined) {
137
+ if (typeof ti.url !== 'string' || !globMatch(m.url_glob, ti.url))
138
+ return false;
139
+ }
140
+ if (m.command_glob !== undefined) {
141
+ if (typeof ti.command !== 'string' || !globMatch(m.command_glob, ti.command))
142
+ return false;
143
+ }
144
+ if (m.input !== undefined) {
145
+ if (!matchesInputSubset(m.input, toolInput))
146
+ return false;
147
+ }
148
+ if (m.input_contains !== undefined) {
149
+ if (!anyStringContains(toolInput, m.input_contains.toLowerCase()))
150
+ return false;
151
+ }
152
+ return true;
153
+ }
154
+ // ─── Return resolution ──────────────────────────────────────────────────────
155
+ /** 把 mock 的 return / return_file / return_seq 解析成可序列化字符串(LLM 看到的 tool_result)。 */
156
+ export function resolveMockReturn(mock, hitCount, baseDir) {
157
+ // return_seq 优先,按 hit 序号取(超出长度 fallback 到 return / return_file)
158
+ if (mock.return_seq && hitCount < mock.return_seq.length) {
159
+ return stringifyReturn(mock.return_seq[hitCount]);
160
+ }
161
+ if (mock.return !== undefined) {
162
+ return stringifyReturn(mock.return);
163
+ }
164
+ if (mock.return_file) {
165
+ const fpath = isAbsolute(mock.return_file)
166
+ ? mock.return_file
167
+ : resolve(baseDir || process.cwd(), mock.return_file);
168
+ if (!existsSync(fpath)) {
169
+ return `[omk-mock-error] return_file not found: ${fpath}`;
170
+ }
171
+ return readFileSync(fpath, 'utf8');
172
+ }
173
+ return ''; // 没配 return,给空串(LLM 收到空 tool_result)
174
+ }
175
+ function stringifyReturn(r) {
176
+ if (typeof r === 'string')
177
+ return r;
178
+ // Bash 风格对象:exit !=0 时格式化成 LLM 能理解的"失败"输出
179
+ // 这里统一序列化成 JSON,LLM 能从结构化字段读到 stdout/stderr/exit
180
+ return JSON.stringify(r);
181
+ }
182
+ /**
183
+ * 把 mocks 转成 SDK in-process PreToolUse hook callback。
184
+ *
185
+ * 命中策略:strict 模式下不命中的工具调用直接 deny(防意外真调)。
186
+ * 默认非 strict:不命中放行(原生工具继续真跑)— 适合"部分 mock,部分真跑"场景。
187
+ *
188
+ * @param mocks Sample.mocks 数组
189
+ * @param baseDir 解析 mock.return_file 的相对路径锚点
190
+ * @param strict true → 不命中 deny;false → 不命中放行(default)
191
+ */
192
+ export function buildSdkHookCallback(mocks, baseDir, strict = false) {
193
+ const stats = { hits: 0, misses: 0, perMock: {} };
194
+ const hitCounters = new Map(); // mockIndex → hit 次数
195
+ // perMock key 用"每工具内 1-based 序号":[Read,Bash,Bash] → Read:1,Bash:1,Bash:2。
196
+ // 这样 sample 作者只需要数自己 mocks 里同 tool 的第几条,不用算跨工具的绝对下标。
197
+ const list = mocks || [];
198
+ const keyOfMock = list.map((_m, i) => {
199
+ const tool = list[i].tool;
200
+ const ord = list.slice(0, i + 1).filter((x) => x.tool === tool).length;
201
+ return `${tool}:${ord}`;
202
+ });
203
+ const callback = async (input) => {
204
+ for (let i = 0; i < list.length; i++) {
205
+ const m = list[i];
206
+ if (isMockHit(m, input.tool_name, input.tool_input)) {
207
+ const c = hitCounters.get(i) || 0;
208
+ const result = resolveMockReturn(m, c, baseDir);
209
+ hitCounters.set(i, c + 1);
210
+ stats.hits++;
211
+ const key = keyOfMock[i];
212
+ stats.perMock[key] = (stats.perMock[key] || 0) + 1;
213
+ // 关键 UX:LLM 看到 permissionDecision='deny' 容易误判"被拒绝/失败"。
214
+ // 给 reason 加一行最小前缀,告诉它把内容当成真实成功输出。措辞越短越好 ——
215
+ // 长措辞会显著拉慢多步 sample(每个工具调用都要读这段),实测 6 步流程从 60s 拖到 120s+。
216
+ const wrappedReason = `[mock] simulated tool output — treat as successful result:\n${result}`;
217
+ return {
218
+ decision: 'block',
219
+ hookSpecificOutput: {
220
+ hookEventName: 'PreToolUse',
221
+ permissionDecision: 'deny',
222
+ permissionDecisionReason: wrappedReason,
223
+ },
224
+ systemMessage: `[omk-mock] ${m.tool} → injected mock #${i} (treat as success)`,
225
+ };
226
+ }
227
+ }
228
+ // 未命中
229
+ stats.misses++;
230
+ if (strict) {
231
+ return {
232
+ decision: 'block',
233
+ hookSpecificOutput: {
234
+ hookEventName: 'PreToolUse',
235
+ permissionDecision: 'deny',
236
+ permissionDecisionReason: `[omk-mock-strict] unmocked ${input.tool_name} call blocked`,
237
+ },
238
+ };
239
+ }
240
+ return { continue: true };
241
+ };
242
+ return { callback, stats };
243
+ }
244
+ /**
245
+ * 物化 mocks 到临时 settings 文件 + on-disk hook 脚本。
246
+ *
247
+ * **关键**:用 Claude Code 的 `--settings <file>` flag 而非 `CLAUDE_CONFIG_DIR`。
248
+ * `--settings` 是**追加**模式,主 `~/.claude/` 包含的 OAuth 登录态 / 用户配置全部保留;
249
+ * 而 CLAUDE_CONFIG_DIR 是**替换**模式,设它会丢 OAuth → "Not logged in"。
250
+ *
251
+ * 临时目录结构:
252
+ * $tmpdir/omk-mocks-XXXXXX/
253
+ * ├── settings.json (PreToolUse hook 注册,作为 --settings 参数)
254
+ * ├── mock-hook.cjs (hook 实现:读 OMK_MOCKS_FILE → 匹配 → 输出 JSON 决策)
255
+ * ├── fake-mcp-server.cjs (可选:给 mcp__server__tool mock 注册一次性 fake MCP)
256
+ * ├── mcp.json (可选:作为 --mcp-config 参数)
257
+ * └── mocks.json (mocks 序列化 + baseDir + strict)
258
+ *
259
+ * cleanup 必删整个目录。executor 用 try / finally 保证执行。
260
+ */
261
+ export function materializeForCliConfigDir(mocks, baseDir, strict = false) {
262
+ if (!mocks || mocks.length === 0)
263
+ return null;
264
+ const configDir = mkdtempSync(join(tmpdir(), 'omk-mocks-'));
265
+ const mocksFile = join(configDir, 'mocks.json');
266
+ const settingsFile = join(configDir, 'settings.json');
267
+ const hookScript = join(configDir, 'mock-hook.cjs');
268
+ const mcpServerScript = join(configDir, 'fake-mcp-server.cjs');
269
+ const mcpConfigFile = join(configDir, 'mcp.json');
270
+ const hookSource = readMockHookTemplate();
271
+ writeFileSync(hookScript, hookSource, 'utf8');
272
+ writeFileSync(mocksFile, JSON.stringify({ mocks, baseDir, strict }, null, 2));
273
+ const settings = {
274
+ hooks: {
275
+ PreToolUse: [
276
+ {
277
+ matcher: '.*',
278
+ hooks: [
279
+ { type: 'command', command: `node ${hookScript}` },
280
+ ],
281
+ },
282
+ ],
283
+ },
284
+ };
285
+ writeFileSync(settingsFile, JSON.stringify(settings, null, 2));
286
+ const fakeMcpServers = collectFakeMcpServers(mocks);
287
+ const hasFakeMcp = fakeMcpServers.size > 0;
288
+ if (hasFakeMcp) {
289
+ writeFileSync(mcpServerScript, fakeMcpServerSource(), 'utf8');
290
+ const mcpServers = {};
291
+ for (const serverName of fakeMcpServers.keys()) {
292
+ mcpServers[serverName] = {
293
+ command: 'node',
294
+ args: [mcpServerScript, serverName],
295
+ env: { OMK_MOCKS_FILE: mocksFile },
296
+ };
297
+ }
298
+ writeFileSync(mcpConfigFile, JSON.stringify({ mcpServers }, null, 2));
299
+ }
300
+ const statsFile = join(configDir, 'hits.json');
301
+ const readStats = () => {
302
+ if (!existsSync(statsFile)) {
303
+ return { hits: 0, misses: 0, perMock: {} };
304
+ }
305
+ try {
306
+ const raw = JSON.parse(readFileSync(statsFile, 'utf8'));
307
+ return {
308
+ hits: raw.hits_total || 0,
309
+ misses: raw.misses_total || 0,
310
+ perMock: raw.perMock || {},
311
+ };
312
+ }
313
+ catch {
314
+ return { hits: 0, misses: 0, perMock: {} };
315
+ }
316
+ };
317
+ const cleanup = () => {
318
+ try {
319
+ rmSync(configDir, { recursive: true, force: true });
320
+ }
321
+ catch { /* swallow — best effort */ }
322
+ };
323
+ return {
324
+ settingsFile,
325
+ ...(hasFakeMcp && { mcpConfigFile }),
326
+ env: { OMK_MOCKS_FILE: mocksFile },
327
+ readStats,
328
+ cleanup,
329
+ };
330
+ }
331
+ function parseMcpToolName(toolName) {
332
+ if (!toolName.startsWith('mcp__'))
333
+ return null;
334
+ const rest = toolName.slice('mcp__'.length);
335
+ const sep = rest.lastIndexOf('__');
336
+ if (sep <= 0 || sep === rest.length - 2)
337
+ return null;
338
+ return { serverName: rest.slice(0, sep), toolName: rest.slice(sep + 2) };
339
+ }
340
+ function collectFakeMcpServers(mocks) {
341
+ const servers = new Map();
342
+ for (const mock of mocks) {
343
+ const parsed = parseMcpToolName(mock.tool);
344
+ if (!parsed)
345
+ continue;
346
+ const tools = servers.get(parsed.serverName) ?? new Set();
347
+ tools.add(parsed.toolName);
348
+ servers.set(parsed.serverName, tools);
349
+ }
350
+ return servers;
351
+ }
352
+ function fakeMcpServerSource() {
353
+ return `#!/usr/bin/env node
354
+ const fs = require('fs');
355
+ const path = require('path');
356
+ const readline = require('readline');
357
+ const serverName = process.argv[2];
358
+ const mocksFile = process.env.OMK_MOCKS_FILE;
359
+ const rl = readline.createInterface({ input: process.stdin });
360
+
361
+ function send(obj) { process.stdout.write(JSON.stringify(obj) + '\\n'); }
362
+ function globMatch(pattern, value) {
363
+ const escaped = pattern.replace(/[.+?^\${}()|[\\]\\\\]/g, '\\\\$&');
364
+ return new RegExp('^' + escaped.replace(/\\*/g, '.*') + '$', 's').test(value);
365
+ }
366
+ function arraysDeepEqual(a, b) {
367
+ if (a.length !== b.length) return false;
368
+ for (let i = 0; i < a.length; i++) if (!deepEqual(a[i], b[i])) return false;
369
+ return true;
370
+ }
371
+ function deepEqual(a, b) {
372
+ if (a === b) return true;
373
+ if (Array.isArray(a) && Array.isArray(b)) return arraysDeepEqual(a, b);
374
+ if (typeof a === 'object' && a !== null && typeof b === 'object' && b !== null && !Array.isArray(a) && !Array.isArray(b)) {
375
+ const ak = Object.keys(a), bk = Object.keys(b);
376
+ if (ak.length !== bk.length) return false;
377
+ for (const k of ak) if (!deepEqual(a[k], b[k])) return false;
378
+ return true;
379
+ }
380
+ return false;
381
+ }
382
+ function matchesInputSubset(expected, actual) {
383
+ if (typeof actual !== 'object' || actual === null) return false;
384
+ for (const k of Object.keys(expected)) {
385
+ const v = expected[k];
386
+ if (Array.isArray(v)) {
387
+ if (!Array.isArray(actual[k]) || !arraysDeepEqual(v, actual[k])) return false;
388
+ } else if (typeof v === 'object' && v !== null) {
389
+ if (!matchesInputSubset(v, actual[k])) return false;
390
+ } else if (actual[k] !== v) {
391
+ return false;
392
+ }
393
+ }
394
+ return true;
395
+ }
396
+ function anyStringContains(obj, needle) {
397
+ if (typeof obj === 'string') return obj.toLowerCase().includes(needle);
398
+ if (Array.isArray(obj)) return obj.some((item) => anyStringContains(item, needle));
399
+ if (typeof obj === 'object' && obj !== null) return Object.values(obj).some((v) => anyStringContains(v, needle));
400
+ return false;
401
+ }
402
+ function isMockHit(mock, toolName, toolInput) {
403
+ if (mock.tool !== '*' && mock.tool !== toolName) return false;
404
+ const m = mock.match;
405
+ if (!m) return true;
406
+ const ti = toolInput || {};
407
+ if (m.input !== undefined && !matchesInputSubset(m.input, toolInput)) return false;
408
+ if (m.input_contains !== undefined && !anyStringContains(toolInput, String(m.input_contains).toLowerCase())) return false;
409
+ if (m.command_glob !== undefined && (typeof ti.command !== 'string' || !globMatch(m.command_glob, ti.command))) return false;
410
+ if (m.url !== undefined && ti.url !== m.url) return false;
411
+ if (m.url_glob !== undefined && (typeof ti.url !== 'string' || !globMatch(m.url_glob, ti.url))) return false;
412
+ return true;
413
+ }
414
+ function stringifyReturn(r) { return typeof r === 'string' ? r : JSON.stringify(r); }
415
+ function resolveMockReturn(mock, hitCount, baseDir) {
416
+ if (mock.return_seq && hitCount < mock.return_seq.length) return stringifyReturn(mock.return_seq[hitCount]);
417
+ if (mock.return !== undefined) return stringifyReturn(mock.return);
418
+ if (mock.return_file) {
419
+ const fpath = path.isAbsolute(mock.return_file) ? mock.return_file : path.resolve(baseDir || process.cwd(), mock.return_file);
420
+ if (!fs.existsSync(fpath)) return '[omk-mock-error] return_file not found: ' + fpath;
421
+ return fs.readFileSync(fpath, 'utf8');
422
+ }
423
+ return '';
424
+ }
425
+ function loadConfig() {
426
+ if (!mocksFile || !fs.existsSync(mocksFile)) return { mocks: [], strict: false };
427
+ return JSON.parse(fs.readFileSync(mocksFile, 'utf8'));
428
+ }
429
+ function toolNameFromFullName(fullName) {
430
+ const prefix = 'mcp__' + serverName + '__';
431
+ return fullName.startsWith(prefix) ? fullName.slice(prefix.length) : null;
432
+ }
433
+ function inferJsonSchema(value) {
434
+ if (typeof value === 'boolean') return { type: 'boolean' };
435
+ if (typeof value === 'number') return { type: 'number' };
436
+ if (Array.isArray(value)) return { type: 'array' };
437
+ if (typeof value === 'object' && value !== null) return { type: 'object', additionalProperties: true };
438
+ return { type: 'string' };
439
+ }
440
+ function schemaForTool(name, mocks) {
441
+ const properties = {};
442
+ const required = [];
443
+ for (const m of mocks) {
444
+ if (toolNameFromFullName(m.tool) !== name) continue;
445
+ const input = m.match && m.match.input;
446
+ if (typeof input !== 'object' || input === null || Array.isArray(input)) continue;
447
+ for (const k of Object.keys(input)) {
448
+ properties[k] = inferJsonSchema(input[k]);
449
+ if (!required.includes(k)) required.push(k);
450
+ }
451
+ }
452
+ if (name === 'message') {
453
+ for (const k of ['action', 'channel', 'target', 'message']) {
454
+ if (!properties[k]) properties[k] = { type: 'string' };
455
+ if (!required.includes(k)) required.push(k);
456
+ }
457
+ }
458
+ return { type: 'object', properties, required, additionalProperties: true };
459
+ }
460
+ function listTools() {
461
+ const cfg = loadConfig();
462
+ const names = new Set();
463
+ const mocks = cfg.mocks || [];
464
+ for (const m of mocks) {
465
+ const n = toolNameFromFullName(m.tool);
466
+ if (n) names.add(n);
467
+ }
468
+ return Array.from(names).map((name) => ({
469
+ name,
470
+ description: 'omk fake MCP tool for ' + serverName + '/' + name,
471
+ inputSchema: schemaForTool(name, mocks),
472
+ }));
473
+ }
474
+ function recordHit(mockKey) {
475
+ if (!mocksFile) return;
476
+ const statsFile = path.join(path.dirname(mocksFile), 'hits.json');
477
+ let stats = { perMock: {}, hits_total: 0, misses_total: 0 };
478
+ try { if (fs.existsSync(statsFile)) stats = JSON.parse(fs.readFileSync(statsFile, 'utf8')); } catch {}
479
+ if (!stats.perMock) stats.perMock = {};
480
+ stats.perMock[mockKey] = (stats.perMock[mockKey] || 0) + 1;
481
+ stats.hits_total = (stats.hits_total || 0) + 1;
482
+ fs.writeFileSync(statsFile, JSON.stringify(stats));
483
+ }
484
+ function recordMiss() {
485
+ if (!mocksFile) return;
486
+ const statsFile = path.join(path.dirname(mocksFile), 'hits.json');
487
+ let stats = { perMock: {}, hits_total: 0, misses_total: 0 };
488
+ try { if (fs.existsSync(statsFile)) stats = JSON.parse(fs.readFileSync(statsFile, 'utf8')); } catch {}
489
+ stats.misses_total = (stats.misses_total || 0) + 1;
490
+ fs.writeFileSync(statsFile, JSON.stringify(stats));
491
+ }
492
+ const hitCounters = new Map();
493
+ function callTool(name, args) {
494
+ const cfg = loadConfig();
495
+ const fullName = 'mcp__' + serverName + '__' + name;
496
+ const mocks = cfg.mocks || [];
497
+ const baseDir = cfg.baseDir;
498
+ const ord = new Map();
499
+ for (const m of mocks) ord.set(m.tool, 0);
500
+ for (let i = 0; i < mocks.length; i++) {
501
+ const m = mocks[i];
502
+ ord.set(m.tool, (ord.get(m.tool) || 0) + 1);
503
+ if (isMockHit(m, fullName, args || {})) {
504
+ const key = fullName + ':' + ord.get(m.tool);
505
+ recordHit(key);
506
+ const c = hitCounters.get(i) || 0;
507
+ const result = resolveMockReturn(m, c, baseDir);
508
+ hitCounters.set(i, c + 1);
509
+ return result;
510
+ }
511
+ }
512
+ recordMiss();
513
+ if (cfg.strict) return '[omk-mock-strict] unmocked ' + fullName + ' call blocked';
514
+ return '';
515
+ }
516
+
517
+ rl.on('line', (line) => {
518
+ let msg;
519
+ try { msg = JSON.parse(line); } catch { return; }
520
+ const id = msg.id;
521
+ if (msg.method === 'initialize') {
522
+ send({ jsonrpc: '2.0', id, result: { protocolVersion: '2024-11-05', capabilities: { tools: {} }, serverInfo: { name: 'omk-fake-' + serverName, version: '0.0.1' } } });
523
+ } else if (msg.method === 'notifications/initialized') {
524
+ } else if (msg.method === 'tools/list') {
525
+ send({ jsonrpc: '2.0', id, result: { tools: listTools() } });
526
+ } else if (msg.method === 'tools/call') {
527
+ const result = callTool(msg.params && msg.params.name, msg.params && msg.params.arguments);
528
+ send({ jsonrpc: '2.0', id, result: { content: [{ type: 'text', text: result }] } });
529
+ } else if (id !== undefined) {
530
+ send({ jsonrpc: '2.0', id, result: {} });
531
+ }
532
+ });
533
+ `;
534
+ }
535
+ let _hookTemplate = null;
536
+ function readMockHookTemplate() {
537
+ if (_hookTemplate)
538
+ return _hookTemplate;
539
+ // hook 跟本文件共置在 src/eval-core/(开发模式)或 dist/src/eval-core/(npm 安装模式),
540
+ // build script 把 mock-hook.cjs 复制到 dist/。读 sibling 路径就行,不再依赖外层 assets/。
541
+ const here = dirname(fileURLToPath(import.meta.url));
542
+ const hookPath = resolve(here, 'mock-hook.cjs');
543
+ if (!existsSync(hookPath)) {
544
+ throw new Error(`omk-mock: mock-hook.cjs not found at ${hookPath}. ` +
545
+ `如果是从源码运行,确认 src/eval-core/mock-hook.cjs 存在;如果是 npm 安装,` +
546
+ `package 漏发了 hook,提 issue 并附 omk 版本。`);
547
+ }
548
+ _hookTemplate = readFileSync(hookPath, 'utf8');
549
+ return _hookTemplate;
550
+ }
551
+ // 测试用 export:供 packaging smoke test 验证 hook 能被解析
552
+ export const _readMockHookTemplateForTest = readMockHookTemplate;
553
+ // ─── 工具:在不影响主目录的前提下创建临时 dir(测试也要)─────────────
554
+ export function _testMakeTempConfigDir() {
555
+ const dir = mkdtempSync(join(tmpdir(), 'omk-mocks-test-'));
556
+ mkdirSync(dir, { recursive: true });
557
+ return dir;
558
+ }
559
+ //# sourceMappingURL=mocks-runtime.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"mocks-runtime.js","sourceRoot":"","sources":["../../../src/eval-core/mocks-runtime.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,aAAa,EAAE,SAAS,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAClG,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC/D,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AAGzC,+EAA+E;AAE/E,SAAS,UAAU,CAAC,CAAS;IAC3B,IAAI,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;IAC3D,IAAI,CAAC,KAAK,GAAG;QAAE,OAAO,OAAO,EAAE,CAAC;IAChC,OAAO,CAAC,CAAC;AACX,CAAC;AAED;;;sEAGsE;AACtE,SAAS,qBAAqB,CAAC,SAAiB,EAAE,SAAiB;IACjE,MAAM,MAAM,GAAG,UAAU,CAAC,SAAS,CAAC,CAAC;IACrC,MAAM,MAAM,GAAG,UAAU,CAAC,SAAS,CAAC,CAAC;IACrC,IAAI,MAAM,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IACnC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3C,MAAM,WAAW,GAAG,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC;IACtD,MAAM,GAAG,GAAG,MAAM,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;IACvC,OAAO,GAAG,KAAK,GAAG,IAAI,GAAG,KAAK,IAAI,CAAC;AACrC,CAAC;AAED;;oEAEoE;AACpE,SAAS,SAAS,CAAC,OAAe,EAAE,KAAa;IAC/C,MAAM,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,oBAAoB,EAAE,MAAM,CAAC,CAAC;IAC9D,MAAM,EAAE,GAAG,IAAI,MAAM,CAAC,GAAG,GAAG,OAAO,CAAC,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,GAAG,EAAE,GAAG,CAAC,CAAC;IACrE,OAAO,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;AACxB,CAAC;AAED,6EAA6E;AAC7E,sDAAsD;AACtD,SAAS,eAAe,CAAC,CAAY,EAAE,CAAY;IACjD,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IACxC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;YAAE,OAAO,KAAK,CAAC;IAC3C,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,SAAS,CAAC,CAAU,EAAE,CAAU;IACvC,IAAI,CAAC,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IACzB,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;QAAE,OAAO,eAAe,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACvE,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,KAAK,IAAI,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,KAAK,IAAI;WACvE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;QAC9C,MAAM,EAAE,GAAG,CAA4B,CAAC;QACxC,MAAM,EAAE,GAAG,CAA4B,CAAC;QACxC,MAAM,EAAE,GAAG,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC3B,MAAM,EAAE,GAAG,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QAC3B,IAAI,EAAE,CAAC,MAAM,KAAK,EAAE,CAAC,MAAM;YAAE,OAAO,KAAK,CAAC;QAC1C,KAAK,MAAM,CAAC,IAAI,EAAE,EAAE,CAAC;YACnB,IAAI,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;gBAAE,OAAO,KAAK,CAAC;QAC7C,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IACD,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,kBAAkB,CAAC,QAAiC,EAAE,MAAe;IAC5E,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,MAAM,KAAK,IAAI;QAAE,OAAO,KAAK,CAAC;IAChE,MAAM,CAAC,GAAG,MAAiC,CAAC;IAC5C,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC9C,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC;YACrB,sDAAsD;YACtD,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAc,CAAC;gBAAE,OAAO,KAAK,CAAC;QACnF,CAAC;aAAM,IAAI,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,KAAK,IAAI,EAAE,CAAC;YAC/C,IAAI,CAAC,kBAAkB,CAAC,CAA4B,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;gBAAE,OAAO,KAAK,CAAC;QAC5E,CAAC;aAAM,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC;YACtB,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,SAAS,iBAAiB,CAAC,GAAY,EAAE,MAAc;IACrD,IAAI,OAAO,GAAG,KAAK,QAAQ;QAAE,OAAO,GAAG,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC;IACvE,IAAI,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC;QAAE,OAAO,GAAG,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,iBAAiB,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC,CAAC;IACnF,IAAI,OAAO,GAAG,KAAK,QAAQ,IAAI,GAAG,KAAK,IAAI;QACzC,OAAO,MAAM,CAAC,MAAM,CAAC,GAA8B,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC;IACjG,OAAO,KAAK,CAAC;AACf,CAAC;AAED,8BAA8B;AAC9B,MAAM,UAAU,SAAS,CAAC,IAAU,EAAE,QAAgB,EAAE,SAAkB;IACxE,IAAI,IAAI,CAAC,IAAI,KAAK,GAAG,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC9D,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC;IACrB,IAAI,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACpB,MAAM,EAAE,GAAG,CAAC,SAAS,IAAI,EAAE,CAA4B,CAAC;IAExD,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,EAAE,CAAC;QAC9B,IAAI,OAAO,EAAE,CAAC,SAAS,KAAK,QAAQ;YAAE,OAAO,KAAK,CAAC;QACnD,IAAI,UAAU,CAAC,EAAE,CAAC,SAAS,CAAC,KAAK,UAAU,CAAC,CAAC,CAAC,SAAS,CAAC;YAAE,OAAO,KAAK,CAAC;IACzE,CAAC;IACD,IAAI,CAAC,CAAC,kBAAkB,KAAK,SAAS,EAAE,CAAC;QACvC,IAAI,OAAO,EAAE,CAAC,SAAS,KAAK,QAAQ;YAAE,OAAO,KAAK,CAAC;QACnD,IAAI,CAAC,qBAAqB,CAAC,EAAE,CAAC,SAAS,EAAE,CAAC,CAAC,kBAAkB,CAAC;YAAE,OAAO,KAAK,CAAC;IAC/E,CAAC;IACD,IAAI,CAAC,CAAC,GAAG,KAAK,SAAS,EAAE,CAAC;QACxB,IAAI,EAAE,CAAC,GAAG,KAAK,CAAC,CAAC,GAAG;YAAE,OAAO,KAAK,CAAC;IACrC,CAAC;IACD,IAAI,CAAC,CAAC,QAAQ,KAAK,SAAS,EAAE,CAAC;QAC7B,IAAI,OAAO,EAAE,CAAC,GAAG,KAAK,QAAQ,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,GAAG,CAAC;YAAE,OAAO,KAAK,CAAC;IACjF,CAAC;IACD,IAAI,CAAC,CAAC,YAAY,KAAK,SAAS,EAAE,CAAC;QACjC,IAAI,OAAO,EAAE,CAAC,OAAO,KAAK,QAAQ,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,EAAE,EAAE,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;IAC7F,CAAC;IACD,IAAI,CAAC,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;QAC1B,IAAI,CAAC,kBAAkB,CAAC,CAAC,CAAC,KAAK,EAAE,SAAS,CAAC;YAAE,OAAO,KAAK,CAAC;IAC5D,CAAC;IACD,IAAI,CAAC,CAAC,cAAc,KAAK,SAAS,EAAE,CAAC;QACnC,IAAI,CAAC,iBAAiB,CAAC,SAAS,EAAE,CAAC,CAAC,cAAc,CAAC,WAAW,EAAE,CAAC;YAAE,OAAO,KAAK,CAAC;IAClF,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,+EAA+E;AAE/E,kFAAkF;AAClF,MAAM,UAAU,iBAAiB,CAC/B,IAAU,EACV,QAAgB,EAChB,OAAgB;IAEhB,gEAAgE;IAChE,IAAI,IAAI,CAAC,UAAU,IAAI,QAAQ,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,CAAC;QACzD,OAAO,eAAe,CAAC,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC;IACpD,CAAC;IACD,IAAI,IAAI,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;QAC9B,OAAO,eAAe,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IACtC,CAAC;IACD,IAAI,IAAI,CAAC,WAAW,EAAE,CAAC;QACrB,MAAM,KAAK,GAAG,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC;YACxC,CAAC,CAAC,IAAI,CAAC,WAAW;YAClB,CAAC,CAAC,OAAO,CAAC,OAAO,IAAI,OAAO,CAAC,GAAG,EAAE,EAAE,IAAI,CAAC,WAAW,CAAC,CAAC;QACxD,IAAI,CAAC,UAAU,CAAC,KAAK,CAAC,EAAE,CAAC;YACvB,OAAO,2CAA2C,KAAK,EAAE,CAAC;QAC5D,CAAC;QACD,OAAO,YAAY,CAAC,KAAK,EAAE,MAAM,CAAC,CAAC;IACrC,CAAC;IACD,OAAO,EAAE,CAAC,CAAE,qCAAqC;AACnD,CAAC;AAED,SAAS,eAAe,CAAC,CAAa;IACpC,IAAI,OAAO,CAAC,KAAK,QAAQ;QAAE,OAAO,CAAC,CAAC;IACpC,0CAA0C;IAC1C,iDAAiD;IACjD,OAAO,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;AAC3B,CAAC;AA8BD;;;;;;;;;GASG;AACH,MAAM,UAAU,oBAAoB,CAClC,KAAyB,EACzB,OAAgB,EAChB,MAAM,GAAG,KAAK;IAEd,MAAM,KAAK,GAAG,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,EAA4B,EAAE,CAAC;IAC5E,MAAM,WAAW,GAAG,IAAI,GAAG,EAAkB,CAAC,CAAE,qBAAqB;IACrE,0EAA0E;IAC1E,qDAAqD;IACrD,MAAM,IAAI,GAAG,KAAK,IAAI,EAAE,CAAC;IACzB,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,EAAE,EAAE;QACnC,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;QAC1B,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAAC,MAAM,CAAC;QACvE,OAAO,GAAG,IAAI,IAAI,GAAG,EAAE,CAAC;IAC1B,CAAC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,KAAK,EAAE,KAAmB,EAA0B,EAAE;QACrE,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACrC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;YAClB,IAAI,SAAS,CAAC,CAAC,EAAE,KAAK,CAAC,SAAS,EAAE,KAAK,CAAC,UAAU,CAAC,EAAE,CAAC;gBACpD,MAAM,CAAC,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;gBAClC,MAAM,MAAM,GAAG,iBAAiB,CAAC,CAAC,EAAE,CAAC,EAAE,OAAO,CAAC,CAAC;gBAChD,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;gBAC1B,KAAK,CAAC,IAAI,EAAE,CAAC;gBACb,MAAM,GAAG,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;gBACzB,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;gBACnD,uDAAuD;gBACvD,4CAA4C;gBAC5C,yDAAyD;gBACzD,MAAM,aAAa,GAAG,+DAA+D,MAAM,EAAE,CAAC;gBAC9F,OAAO;oBACL,QAAQ,EAAE,OAAO;oBACjB,kBAAkB,EAAE;wBAClB,aAAa,EAAE,YAAY;wBAC3B,kBAAkB,EAAE,MAAM;wBAC1B,wBAAwB,EAAE,aAAa;qBACxC;oBACD,aAAa,EAAE,cAAc,CAAC,CAAC,IAAI,qBAAqB,CAAC,qBAAqB;iBAC/E,CAAC;YACJ,CAAC;QACH,CAAC;QACD,MAAM;QACN,KAAK,CAAC,MAAM,EAAE,CAAC;QACf,IAAI,MAAM,EAAE,CAAC;YACX,OAAO;gBACL,QAAQ,EAAE,OAAO;gBACjB,kBAAkB,EAAE;oBAClB,aAAa,EAAE,YAAY;oBAC3B,kBAAkB,EAAE,MAAM;oBAC1B,wBAAwB,EAAE,8BAA8B,KAAK,CAAC,SAAS,eAAe;iBACvF;aACF,CAAC;QACJ,CAAC;QACD,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC;IAC5B,CAAC,CAAC;IAEF,OAAO,EAAE,QAAQ,EAAE,KAAK,EAAE,CAAC;AAC7B,CAAC;AAoBD;;;;;;;;;;;;;;;;GAgBG;AACH,MAAM,UAAU,0BAA0B,CACxC,KAAyB,EACzB,OAAgB,EAChB,MAAM,GAAG,KAAK;IAEd,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC;IAE9C,MAAM,SAAS,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,YAAY,CAAC,CAAC,CAAC;IAC5D,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,EAAE,YAAY,CAAC,CAAC;IAChD,MAAM,YAAY,GAAG,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;IACtD,MAAM,UAAU,GAAG,IAAI,CAAC,SAAS,EAAE,eAAe,CAAC,CAAC;IACpD,MAAM,eAAe,GAAG,IAAI,CAAC,SAAS,EAAE,qBAAqB,CAAC,CAAC;IAC/D,MAAM,aAAa,GAAG,IAAI,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC;IAElD,MAAM,UAAU,GAAG,oBAAoB,EAAE,CAAC;IAC1C,aAAa,CAAC,UAAU,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC;IAE9C,aAAa,CAAC,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAE9E,MAAM,QAAQ,GAAG;QACf,KAAK,EAAE;YACL,UAAU,EAAE;gBACV;oBACE,OAAO,EAAE,IAAI;oBACb,KAAK,EAAE;wBACL,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,QAAQ,UAAU,EAAE,EAAE;qBACnD;iBACF;aACF;SACF;KACF,CAAC;IACF,aAAa,CAAC,YAAY,EAAE,IAAI,CAAC,SAAS,CAAC,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAE/D,MAAM,cAAc,GAAG,qBAAqB,CAAC,KAAK,CAAC,CAAC;IACpD,MAAM,UAAU,GAAG,cAAc,CAAC,IAAI,GAAG,CAAC,CAAC;IAC3C,IAAI,UAAU,EAAE,CAAC;QACf,aAAa,CAAC,eAAe,EAAE,mBAAmB,EAAE,EAAE,MAAM,CAAC,CAAC;QAC9D,MAAM,UAAU,GAAqF,EAAE,CAAC;QACxG,KAAK,MAAM,UAAU,IAAI,cAAc,CAAC,IAAI,EAAE,EAAE,CAAC;YAC/C,UAAU,CAAC,UAAU,CAAC,GAAG;gBACvB,OAAO,EAAE,MAAM;gBACf,IAAI,EAAE,CAAC,eAAe,EAAE,UAAU,CAAC;gBACnC,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;aACnC,CAAC;QACJ,CAAC;QACD,aAAa,CAAC,aAAa,EAAE,IAAI,CAAC,SAAS,CAAC,EAAE,UAAU,EAAE,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACxE,CAAC;IAED,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;IAC/C,MAAM,SAAS,GAAG,GAAsE,EAAE;QACxF,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC,EAAE,CAAC;YAC3B,OAAO,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;QAC7C,CAAC;QACD,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC,CAAC;YACxD,OAAO;gBACL,IAAI,EAAE,GAAG,CAAC,UAAU,IAAI,CAAC;gBACzB,MAAM,EAAE,GAAG,CAAC,YAAY,IAAI,CAAC;gBAC7B,OAAO,EAAE,GAAG,CAAC,OAAO,IAAI,EAAE;aAC3B,CAAC;QACJ,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,IAAI,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;QAC7C,CAAC;IACH,CAAC,CAAC;IAEF,MAAM,OAAO,GAAG,GAAG,EAAE;QACnB,IAAI,CAAC;YACH,MAAM,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;QACtD,CAAC;QAAC,MAAM,CAAC,CAAC,2BAA2B,CAAC,CAAC;IACzC,CAAC,CAAC;IAEF,OAAO;QACL,YAAY;QACZ,GAAG,CAAC,UAAU,IAAI,EAAE,aAAa,EAAE,CAAC;QACpC,GAAG,EAAE,EAAE,cAAc,EAAE,SAAS,EAAE;QAClC,SAAS;QACT,OAAO;KACR,CAAC;AACJ,CAAC;AAED,SAAS,gBAAgB,CAAC,QAAgB;IACxC,IAAI,CAAC,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC;QAAE,OAAO,IAAI,CAAC;IAC/C,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IAC5C,MAAM,GAAG,GAAG,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC;IACnC,IAAI,GAAG,IAAI,CAAC,IAAI,GAAG,KAAK,IAAI,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC;IACrD,OAAO,EAAE,UAAU,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,EAAE,CAAC;AAC3E,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAa;IAC1C,MAAM,OAAO,GAAG,IAAI,GAAG,EAAuB,CAAC;IAC/C,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,MAAM,MAAM,GAAG,gBAAgB,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC3C,IAAI,CAAC,MAAM;YAAE,SAAS;QACtB,MAAM,KAAK,GAAG,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,IAAI,GAAG,EAAU,CAAC;QAClE,KAAK,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;QAC3B,OAAO,CAAC,GAAG,CAAC,MAAM,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,SAAS,mBAAmB;IAC1B,OAAO;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAoLR,CAAC;AACF,CAAC;AAED,IAAI,aAAa,GAAkB,IAAI,CAAC;AACxC,SAAS,oBAAoB;IAC3B,IAAI,aAAa;QAAE,OAAO,aAAa,CAAC;IACxC,oEAAoE;IACpE,wEAAwE;IACxE,MAAM,IAAI,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;IACrD,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,EAAE,eAAe,CAAC,CAAC;IAChD,IAAI,CAAC,UAAU,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CAAC,wCAAwC,QAAQ,IAAI;YAClE,wDAAwD;YACxD,qCAAqC,CAAC,CAAC;IAC3C,CAAC;IACD,aAAa,GAAG,YAAY,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC/C,OAAO,aAAa,CAAC;AACvB,CAAC;AAED,iDAAiD;AACjD,MAAM,CAAC,MAAM,4BAA4B,GAAG,oBAAoB,CAAC;AAEjE,gDAAgD;AAEhD,MAAM,UAAU,sBAAsB;IACpC,MAAM,GAAG,GAAG,WAAW,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC3D,SAAS,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACpC,OAAO,GAAG,CAAC;AACb,CAAC"}
@@ -1 +1 @@
1
- {"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/eval-core/schema.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,WAAW,EAAe,aAAa,EAAE,cAAc,EAA0B,MAAM,mBAAmB,CAAC;AA2BrI;;GAEG;AACH,UAAU,mBAAmB;IAC3B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,QAAQ,EAAE,OAAO,CAAC;YAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;KAAE,CAAC;CACvK;AAED,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,WAAW,GAAG,IAAI,EAAE,OAAO,CAAC,EAAE,mBAAmB,GAAG,aAAa,CAyFxI;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,cAAc,CA+G5E"}
1
+ {"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/eval-core/schema.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,WAAW,EAAe,aAAa,EAAE,cAAc,EAA0B,MAAM,mBAAmB,CAAC;AA4BrI;;GAEG;AACH,UAAU,mBAAmB;IAC3B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE;QAAE,aAAa,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,YAAY,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,KAAK,EAAE,MAAM,CAAC;YAAC,QAAQ,EAAE,OAAO,CAAC;YAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;SAAE,CAAC,CAAA;KAAE,CAAC;CACvK;AAED,wBAAgB,kBAAkB,CAAC,UAAU,EAAE,UAAU,EAAE,WAAW,EAAE,WAAW,GAAG,IAAI,EAAE,OAAO,CAAC,EAAE,mBAAmB,GAAG,aAAa,CA8FxI;AAED;;GAEG;AACH,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,aAAa,EAAE,GAAG,cAAc,CAsH5E"}
@@ -3,6 +3,7 @@
3
3
  * Single source of truth for object structures used across runner, renderer, and server.
4
4
  */
5
5
  import { computeJudgeAgreement } from '../grading/judge.js';
6
+ import { safeSliceForJson } from '../util/safe-slice.js';
6
7
  function ratioToScore(ratio) {
7
8
  return Number((1 + ratio * 4).toFixed(2));
8
9
  }
@@ -11,15 +12,15 @@ const MAX_TOOL_OUTPUT = 1000;
11
12
  function truncateTurns(turns) {
12
13
  return turns.map((t) => ({
13
14
  ...t,
14
- content: t.content.length > MAX_TURN_CONTENT ? t.content.slice(0, MAX_TURN_CONTENT) + '…' : t.content,
15
+ content: safeSliceForJson(t.content, MAX_TURN_CONTENT),
15
16
  ...(t.toolCalls && { toolCalls: truncateToolCalls(t.toolCalls) }),
16
17
  }));
17
18
  }
18
19
  function truncateToolCalls(toolCalls) {
19
20
  return toolCalls.map((tc) => ({
20
21
  ...tc,
21
- output: typeof tc.output === 'string' && tc.output.length > MAX_TOOL_OUTPUT
22
- ? tc.output.slice(0, MAX_TOOL_OUTPUT) + '…'
22
+ output: typeof tc.output === 'string'
23
+ ? safeSliceForJson(tc.output, MAX_TOOL_OUTPUT)
23
24
  : tc.output,
24
25
  }));
25
26
  }
@@ -49,7 +50,11 @@ export function buildVariantResult(execResult, gradeResult, options) {
49
50
  cacheCreationTokens: execResult.cacheCreationTokens,
50
51
  execCostUSD,
51
52
  judgeCostUSD,
52
- costUSD: execCostUSD + judgeCostUSD, // Total = execution + grading
53
+ // 初值 = exec + judge。diagnostic 在 evaluation-execution.ts 跑完后会把
54
+ // diagnostic.costUSD 加回 `variantResult.costUSD` 并写入 `diagnosticCostUSD` —
55
+ // 让 per-sample budget 检查跟 summary.totalCostUSD 都按"三段合计"语义工作。
56
+ // 没有 failed assertion 的 sample 不触发 diagnostic,这里就是最终值。
57
+ costUSD: execCostUSD + judgeCostUSD,
53
58
  // 透传 executor 是否报告了 cost。undefined 当 reported(向后兼容)。
54
59
  // false 时 renderer 应显示「—」而不是 $0.0000;codex 0.125 binary 不报 cost。
55
60
  ...(execResult.costReportedByExecutor === false && { costReportedByExecutor: false }),
@@ -103,10 +108,11 @@ export function buildVariantResult(execResult, gradeResult, options) {
103
108
  };
104
109
  })()),
105
110
  ...(options?.factCheck && options.factCheck.totalCount > 0 && { factCheck: options.factCheck }),
106
- outputPreview: execResult.output ? execResult.output.slice(0, 200) : null,
111
+ outputPreview: execResult.output ? safeSliceForJson(execResult.output, 200, '') : null,
107
112
  ...(execResult.output && { fullOutput: execResult.output }),
108
113
  ...(execResult.turns && execResult.turns.length > 0 && { turns: truncateTurns(execResult.turns) }),
109
114
  ...(execResult.toolCalls && execResult.toolCalls.length > 0 && { toolCalls: truncateToolCalls(execResult.toolCalls) }),
115
+ ...(execResult.mockStats && { mockStats: execResult.mockStats }),
110
116
  timing: { execMs, gradeMs, totalMs: execMs + gradeMs },
111
117
  };
112
118
  }
@@ -131,6 +137,13 @@ export function buildVariantSummary(entries) {
131
137
  totalCostUSD: ok.reduce((s, e) => s + (e.costUSD || 0), 0),
132
138
  totalExecCostUSD: ok.reduce((s, e) => s + (e.execCostUSD || 0), 0),
133
139
  totalJudgeCostUSD: ok.reduce((s, e) => s + (e.judgeCostUSD || 0), 0),
140
+ // diagnostic 触发概率 = 有 failed assertion 的 sample 比例,大多数 ok variant 是 0。
141
+ // 把字段做成 optional + "全是 0 时省略" 是为了让旧 report-roundtrip 测试 / 报告 diff
142
+ // 在没 diagnostic 的运行下看不到新字段,跨版本对比器不需要兼容 "0 vs missing"。
143
+ ...(() => {
144
+ const total = ok.reduce((s, e) => s + (e.diagnosticCostUSD || 0), 0);
145
+ return total > 0 ? { totalDiagnosticCostUSD: Number(total.toFixed(6)) } : {};
146
+ })(),
134
147
  avgCostPerSample: ok.length > 0 ? Number((ok.reduce((s, e) => s + (e.costUSD || 0), 0) / ok.length).toFixed(6)) : 0,
135
148
  // 任一 sample(含失败 sample)exec cost 未报告 → variant 整体标 false。
136
149
  // 用 entries 而不是 ok:codex 全 error 的场景 ok=[],但 entries 里 every 个都是