oh-my-knowledge 0.28.0 → 0.30.0

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 (281) hide show
  1. package/README.md +49 -3
  2. package/README.zh.md +49 -3
  3. package/dist/src/analysis/failure-clusterer.js +1 -1
  4. package/dist/src/analysis/failure-clusterer.js.map +1 -1
  5. package/dist/src/analysis/gap-analyzer.d.ts +8 -1
  6. package/dist/src/analysis/gap-analyzer.d.ts.map +1 -1
  7. package/dist/src/analysis/gap-analyzer.js +65 -33
  8. package/dist/src/analysis/gap-analyzer.js.map +1 -1
  9. package/dist/src/analysis/report-diagnostics.js +2 -0
  10. package/dist/src/analysis/report-diagnostics.js.map +1 -1
  11. package/dist/src/analysis/sample-diagnostics.js +14 -14
  12. package/dist/src/analysis/sample-diagnostics.js.map +1 -1
  13. package/dist/src/authoring/evolver.d.ts +7 -1
  14. package/dist/src/authoring/evolver.d.ts.map +1 -1
  15. package/dist/src/authoring/evolver.js +8 -5
  16. package/dist/src/authoring/evolver.js.map +1 -1
  17. package/dist/src/authoring/generator.d.ts +21 -23
  18. package/dist/src/authoring/generator.d.ts.map +1 -1
  19. package/dist/src/authoring/generator.js +621 -40
  20. package/dist/src/authoring/generator.js.map +1 -1
  21. package/dist/src/authoring/sample-fixer.d.ts +45 -0
  22. package/dist/src/authoring/sample-fixer.d.ts.map +1 -0
  23. package/dist/src/authoring/sample-fixer.js +186 -0
  24. package/dist/src/authoring/sample-fixer.js.map +1 -0
  25. package/dist/src/cli/commands/evolve.d.ts.map +1 -1
  26. package/dist/src/cli/commands/evolve.js +17 -0
  27. package/dist/src/cli/commands/evolve.js.map +1 -1
  28. package/dist/src/cli/commands/observe.d.ts.map +1 -1
  29. package/dist/src/cli/commands/observe.js +170 -0
  30. package/dist/src/cli/commands/observe.js.map +1 -1
  31. package/dist/src/cli/commands/registry.d.ts.map +1 -1
  32. package/dist/src/cli/commands/registry.js +9 -1
  33. package/dist/src/cli/commands/registry.js.map +1 -1
  34. package/dist/src/cli/commands/sample.d.ts +13 -0
  35. package/dist/src/cli/commands/sample.d.ts.map +1 -1
  36. package/dist/src/cli/commands/sample.js +307 -13
  37. package/dist/src/cli/commands/sample.js.map +1 -1
  38. package/dist/src/cli/commands/studio.d.ts.map +1 -1
  39. package/dist/src/cli/commands/studio.js +8 -0
  40. package/dist/src/cli/commands/studio.js.map +1 -1
  41. package/dist/src/cli/i18n-dict.d.ts +1 -1
  42. package/dist/src/cli/i18n-dict.d.ts.map +1 -1
  43. package/dist/src/cli/i18n-dict.js +189 -17
  44. package/dist/src/cli/i18n-dict.js.map +1 -1
  45. package/dist/src/cli/parse-run-config.d.ts +14 -2
  46. package/dist/src/cli/parse-run-config.d.ts.map +1 -1
  47. package/dist/src/cli/parse-run-config.js +62 -10
  48. package/dist/src/cli/parse-run-config.js.map +1 -1
  49. package/dist/src/doctor/health/builtin-dimensions.d.ts.map +1 -1
  50. package/dist/src/doctor/health/builtin-dimensions.js +20 -23
  51. package/dist/src/doctor/health/builtin-dimensions.js.map +1 -1
  52. package/dist/src/doctor/health/composer.d.ts.map +1 -1
  53. package/dist/src/doctor/health/composer.js +7 -3
  54. package/dist/src/doctor/health/composer.js.map +1 -1
  55. package/dist/src/doctor/health/dimension-spec.d.ts +1 -0
  56. package/dist/src/doctor/health/dimension-spec.d.ts.map +1 -1
  57. package/dist/src/doctor/health/parser.d.ts.map +1 -1
  58. package/dist/src/doctor/health/parser.js +1 -0
  59. package/dist/src/doctor/health/parser.js.map +1 -1
  60. package/dist/src/doctor/health/prompt-builder.js +25 -17
  61. package/dist/src/doctor/health/prompt-builder.js.map +1 -1
  62. package/dist/src/doctor/html-renderer.d.ts.map +1 -1
  63. package/dist/src/doctor/html-renderer.js +76 -66
  64. package/dist/src/doctor/html-renderer.js.map +1 -1
  65. package/dist/src/doctor/rules.d.ts.map +1 -1
  66. package/dist/src/doctor/rules.js +32 -0
  67. package/dist/src/doctor/rules.js.map +1 -1
  68. package/dist/src/eval-core/cache.d.ts +16 -2
  69. package/dist/src/eval-core/cache.d.ts.map +1 -1
  70. package/dist/src/eval-core/cache.js +86 -15
  71. package/dist/src/eval-core/cache.js.map +1 -1
  72. package/dist/src/eval-core/evaluation-execution.d.ts +11 -1
  73. package/dist/src/eval-core/evaluation-execution.d.ts.map +1 -1
  74. package/dist/src/eval-core/evaluation-execution.js +78 -6
  75. package/dist/src/eval-core/evaluation-execution.js.map +1 -1
  76. package/dist/src/eval-core/evaluation-job.d.ts +2 -1
  77. package/dist/src/eval-core/evaluation-job.d.ts.map +1 -1
  78. package/dist/src/eval-core/evaluation-job.js +2 -1
  79. package/dist/src/eval-core/evaluation-job.js.map +1 -1
  80. package/dist/src/eval-core/evaluation-reporting.d.ts +7 -0
  81. package/dist/src/eval-core/evaluation-reporting.d.ts.map +1 -1
  82. package/dist/src/eval-core/evaluation-reporting.js +19 -2
  83. package/dist/src/eval-core/evaluation-reporting.js.map +1 -1
  84. package/dist/src/eval-core/execution-strategy.d.ts +1 -1
  85. package/dist/src/eval-core/execution-strategy.d.ts.map +1 -1
  86. package/dist/src/eval-core/execution-strategy.js +12 -1
  87. package/dist/src/eval-core/execution-strategy.js.map +1 -1
  88. package/dist/src/eval-core/mock-hook.cjs +203 -0
  89. package/dist/src/eval-core/mocks-runtime.d.ts +96 -0
  90. package/dist/src/eval-core/mocks-runtime.d.ts.map +1 -0
  91. package/dist/src/eval-core/mocks-runtime.js +323 -0
  92. package/dist/src/eval-core/mocks-runtime.js.map +1 -0
  93. package/dist/src/eval-core/schema.d.ts.map +1 -1
  94. package/dist/src/eval-core/schema.js +18 -5
  95. package/dist/src/eval-core/schema.js.map +1 -1
  96. package/dist/src/eval-core/task-planner.d.ts +9 -1
  97. package/dist/src/eval-core/task-planner.d.ts.map +1 -1
  98. package/dist/src/eval-core/task-planner.js +40 -1
  99. package/dist/src/eval-core/task-planner.js.map +1 -1
  100. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts +5 -1
  101. package/dist/src/eval-workflows/batch-evaluation-workflow.d.ts.map +1 -1
  102. package/dist/src/eval-workflows/batch-evaluation-workflow.js +2 -1
  103. package/dist/src/eval-workflows/batch-evaluation-workflow.js.map +1 -1
  104. package/dist/src/eval-workflows/evaluation-pipeline.d.ts +22 -1
  105. package/dist/src/eval-workflows/evaluation-pipeline.d.ts.map +1 -1
  106. package/dist/src/eval-workflows/evaluation-pipeline.js +39 -12
  107. package/dist/src/eval-workflows/evaluation-pipeline.js.map +1 -1
  108. package/dist/src/eval-workflows/evaluation-preparation.d.ts +5 -0
  109. package/dist/src/eval-workflows/evaluation-preparation.d.ts.map +1 -1
  110. package/dist/src/eval-workflows/evaluation-preparation.js +3 -1
  111. package/dist/src/eval-workflows/evaluation-preparation.js.map +1 -1
  112. package/dist/src/eval-workflows/run-evaluation.d.ts +12 -2
  113. package/dist/src/eval-workflows/run-evaluation.d.ts.map +1 -1
  114. package/dist/src/eval-workflows/run-evaluation.js +17 -5
  115. package/dist/src/eval-workflows/run-evaluation.js.map +1 -1
  116. package/dist/src/executors/claude-cli.d.ts +1 -1
  117. package/dist/src/executors/claude-cli.d.ts.map +1 -1
  118. package/dist/src/executors/claude-cli.js +57 -6
  119. package/dist/src/executors/claude-cli.js.map +1 -1
  120. package/dist/src/executors/claude-sdk-trace.d.ts.map +1 -1
  121. package/dist/src/executors/claude-sdk-trace.js +10 -4
  122. package/dist/src/executors/claude-sdk-trace.js.map +1 -1
  123. package/dist/src/executors/claude-sdk.d.ts +1 -1
  124. package/dist/src/executors/claude-sdk.d.ts.map +1 -1
  125. package/dist/src/executors/claude-sdk.js +25 -2
  126. package/dist/src/executors/claude-sdk.js.map +1 -1
  127. package/dist/src/executors/shared.d.ts +1 -1
  128. package/dist/src/executors/shared.d.ts.map +1 -1
  129. package/dist/src/executors/shared.js +6 -1
  130. package/dist/src/executors/shared.js.map +1 -1
  131. package/dist/src/grading/assertions.d.ts +5 -0
  132. package/dist/src/grading/assertions.d.ts.map +1 -1
  133. package/dist/src/grading/assertions.js +27 -1
  134. package/dist/src/grading/assertions.js.map +1 -1
  135. package/dist/src/grading/debias-validate.js +4 -4
  136. package/dist/src/grading/debias-validate.js.map +1 -1
  137. package/dist/src/grading/diagnostic.d.ts +40 -0
  138. package/dist/src/grading/diagnostic.d.ts.map +1 -0
  139. package/dist/src/grading/diagnostic.js +450 -0
  140. package/dist/src/grading/diagnostic.js.map +1 -0
  141. package/dist/src/grading/gold-cli.js +2 -2
  142. package/dist/src/grading/gold-cli.js.map +1 -1
  143. package/dist/src/grading/index.d.ts +5 -0
  144. package/dist/src/grading/index.d.ts.map +1 -1
  145. package/dist/src/grading/judge.d.ts.map +1 -1
  146. package/dist/src/grading/judge.js +54 -2
  147. package/dist/src/grading/judge.js.map +1 -1
  148. package/dist/src/grading/layered-scores.d.ts.map +1 -1
  149. package/dist/src/grading/layered-scores.js +1 -0
  150. package/dist/src/grading/layered-scores.js.map +1 -1
  151. package/dist/src/inputs/eval-config.js +19 -0
  152. package/dist/src/inputs/eval-config.js.map +1 -1
  153. package/dist/src/inputs/load-samples.d.ts +22 -2
  154. package/dist/src/inputs/load-samples.d.ts.map +1 -1
  155. package/dist/src/inputs/load-samples.js +193 -22
  156. package/dist/src/inputs/load-samples.js.map +1 -1
  157. package/dist/src/inputs/skill-loader.d.ts.map +1 -1
  158. package/dist/src/inputs/skill-loader.js +11 -3
  159. package/dist/src/inputs/skill-loader.js.map +1 -1
  160. package/dist/src/observability/experience.d.ts +260 -0
  161. package/dist/src/observability/experience.d.ts.map +1 -0
  162. package/dist/src/observability/experience.js +1233 -0
  163. package/dist/src/observability/experience.js.map +1 -0
  164. package/dist/src/observability/inbox-view-model.d.ts +27 -0
  165. package/dist/src/observability/inbox-view-model.d.ts.map +1 -0
  166. package/dist/src/observability/inbox-view-model.js +135 -0
  167. package/dist/src/observability/inbox-view-model.js.map +1 -0
  168. package/dist/src/observability/inbox.d.ts +125 -0
  169. package/dist/src/observability/inbox.d.ts.map +1 -0
  170. package/dist/src/observability/inbox.js +741 -0
  171. package/dist/src/observability/inbox.js.map +1 -0
  172. package/dist/src/observability/problem-patterns.d.ts +51 -0
  173. package/dist/src/observability/problem-patterns.d.ts.map +1 -0
  174. package/dist/src/observability/problem-patterns.js +205 -0
  175. package/dist/src/observability/problem-patterns.js.map +1 -0
  176. package/dist/src/observability/review-state.d.ts +54 -0
  177. package/dist/src/observability/review-state.d.ts.map +1 -0
  178. package/dist/src/observability/review-state.js +131 -0
  179. package/dist/src/observability/review-state.js.map +1 -0
  180. package/dist/src/observability/skill-chain-advisories.d.ts +25 -0
  181. package/dist/src/observability/skill-chain-advisories.d.ts.map +1 -0
  182. package/dist/src/observability/skill-chain-advisories.js +69 -0
  183. package/dist/src/observability/skill-chain-advisories.js.map +1 -0
  184. package/dist/src/observability/skill-chain.d.ts +61 -0
  185. package/dist/src/observability/skill-chain.d.ts.map +1 -0
  186. package/dist/src/observability/skill-chain.js +233 -0
  187. package/dist/src/observability/skill-chain.js.map +1 -0
  188. package/dist/src/observability/text-signals.d.ts +7 -0
  189. package/dist/src/observability/text-signals.d.ts.map +1 -0
  190. package/dist/src/observability/text-signals.js +22 -0
  191. package/dist/src/observability/text-signals.js.map +1 -0
  192. package/dist/src/observability/trace-adapter.d.ts +12 -64
  193. package/dist/src/observability/trace-adapter.d.ts.map +1 -1
  194. package/dist/src/observability/trace-adapter.js +9 -355
  195. package/dist/src/observability/trace-adapter.js.map +1 -1
  196. package/dist/src/observability/trace-attribution.d.ts +56 -0
  197. package/dist/src/observability/trace-attribution.d.ts.map +1 -0
  198. package/dist/src/observability/trace-attribution.js +200 -0
  199. package/dist/src/observability/trace-attribution.js.map +1 -0
  200. package/dist/src/observability/trace-segmenter.d.ts +52 -0
  201. package/dist/src/observability/trace-segmenter.d.ts.map +1 -0
  202. package/dist/src/observability/trace-segmenter.js +320 -0
  203. package/dist/src/observability/trace-segmenter.js.map +1 -0
  204. package/dist/src/observability/trace-source.d.ts +99 -0
  205. package/dist/src/observability/trace-source.d.ts.map +1 -0
  206. package/dist/src/observability/trace-source.js +492 -0
  207. package/dist/src/observability/trace-source.js.map +1 -0
  208. package/dist/src/renderer/html-renderer.d.ts.map +1 -1
  209. package/dist/src/renderer/html-renderer.js +324 -218
  210. package/dist/src/renderer/html-renderer.js.map +1 -1
  211. package/dist/src/renderer/layout.d.ts.map +1 -1
  212. package/dist/src/renderer/layout.js +155 -71
  213. package/dist/src/renderer/layout.js.map +1 -1
  214. package/dist/src/renderer/observation-inbox-renderer.d.ts +4 -0
  215. package/dist/src/renderer/observation-inbox-renderer.d.ts.map +1 -0
  216. package/dist/src/renderer/observation-inbox-renderer.js +5376 -0
  217. package/dist/src/renderer/observation-inbox-renderer.js.map +1 -0
  218. package/dist/src/renderer/skill-detail-renderer.d.ts +15 -0
  219. package/dist/src/renderer/skill-detail-renderer.d.ts.map +1 -0
  220. package/dist/src/renderer/skill-detail-renderer.js +1234 -0
  221. package/dist/src/renderer/skill-detail-renderer.js.map +1 -0
  222. package/dist/src/renderer/skill-list-renderer.d.ts +4 -0
  223. package/dist/src/renderer/skill-list-renderer.d.ts.map +1 -0
  224. package/dist/src/renderer/skill-list-renderer.js +327 -0
  225. package/dist/src/renderer/skill-list-renderer.js.map +1 -0
  226. package/dist/src/renderer/summary.d.ts +17 -0
  227. package/dist/src/renderer/summary.d.ts.map +1 -1
  228. package/dist/src/renderer/summary.js +346 -74
  229. package/dist/src/renderer/summary.js.map +1 -1
  230. package/dist/src/renderer/table.d.ts.map +1 -1
  231. package/dist/src/renderer/table.js +213 -167
  232. package/dist/src/renderer/table.js.map +1 -1
  233. package/dist/src/renderer/test-view.d.ts +66 -0
  234. package/dist/src/renderer/test-view.d.ts.map +1 -0
  235. package/dist/src/renderer/test-view.js +905 -0
  236. package/dist/src/renderer/test-view.js.map +1 -0
  237. package/dist/src/renderer/trends.d.ts.map +1 -1
  238. package/dist/src/renderer/trends.js +3 -1
  239. package/dist/src/renderer/trends.js.map +1 -1
  240. package/dist/src/server/report-server.d.ts +6 -1
  241. package/dist/src/server/report-server.d.ts.map +1 -1
  242. package/dist/src/server/report-server.js +182 -6
  243. package/dist/src/server/report-server.js.map +1 -1
  244. package/dist/src/server/report-store.d.ts.map +1 -1
  245. package/dist/src/server/report-store.js +33 -1
  246. package/dist/src/server/report-store.js.map +1 -1
  247. package/dist/src/server/skill-index.d.ts +77 -0
  248. package/dist/src/server/skill-index.d.ts.map +1 -0
  249. package/dist/src/server/skill-index.js +331 -0
  250. package/dist/src/server/skill-index.js.map +1 -0
  251. package/dist/src/server/skill-insights.d.ts +92 -0
  252. package/dist/src/server/skill-insights.d.ts.map +1 -0
  253. package/dist/src/server/skill-insights.js +676 -0
  254. package/dist/src/server/skill-insights.js.map +1 -0
  255. package/dist/src/shared/hard-rules.d.ts +40 -0
  256. package/dist/src/shared/hard-rules.d.ts.map +1 -0
  257. package/dist/src/shared/hard-rules.js +200 -0
  258. package/dist/src/shared/hard-rules.js.map +1 -0
  259. package/dist/src/shared/time.d.ts +2 -0
  260. package/dist/src/shared/time.d.ts.map +1 -0
  261. package/dist/src/shared/time.js +10 -0
  262. package/dist/src/shared/time.js.map +1 -0
  263. package/dist/src/shared/tool-search.d.ts +9 -0
  264. package/dist/src/shared/tool-search.d.ts.map +1 -0
  265. package/dist/src/shared/tool-search.js +104 -0
  266. package/dist/src/shared/tool-search.js.map +1 -0
  267. package/dist/src/types/eval.d.ts +83 -0
  268. package/dist/src/types/eval.d.ts.map +1 -1
  269. package/dist/src/types/executor.d.ts +47 -0
  270. package/dist/src/types/executor.d.ts.map +1 -1
  271. package/dist/src/types/judge.d.ts +76 -0
  272. package/dist/src/types/judge.d.ts.map +1 -1
  273. package/dist/src/types/judge.js +3 -1
  274. package/dist/src/types/judge.js.map +1 -1
  275. package/dist/src/types/report.d.ts +63 -3
  276. package/dist/src/types/report.d.ts.map +1 -1
  277. package/dist/src/util/safe-slice.d.ts +23 -0
  278. package/dist/src/util/safe-slice.d.ts.map +1 -0
  279. package/dist/src/util/safe-slice.js +33 -0
  280. package/dist/src/util/safe-slice.js.map +1 -0
  281. package/package.json +5 -4
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-job.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,uBAAuB,EAAE,aAAa,EAAE,iBAAiB,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAMzI,wBAAgB,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,QAAQ,EACR,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GACP,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,WAAW,EAAE,CAAC;IAC3B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;CACjD,GAAG,iBAAiB,CAyBpB;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG;IAAE,GAAG,EAAE,aAAa,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAS1H;AAED,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMtG;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMlG;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAoB,GACrB,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAQhB;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG,aAAa,CAQ7G;AAED,wBAAgB,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GACX,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAYhB;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,GAAG,uBAAuB,CA4B9E;AAED,wBAAgB,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAA8C,EAC9C,UAAqB,GACtB,EAAE;IACD,GAAG,EAAE,aAAa,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,uBAAuB,CAAC;IACxC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,aAAa,CAShB"}
1
+ {"version":3,"file":"evaluation-job.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,QAAQ,EAAE,uBAAuB,EAAE,aAAa,EAAE,iBAAiB,EAAE,aAAa,EAAE,WAAW,EAAE,MAAM,mBAAmB,CAAC;AAMzI,wBAAgB,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,QAAQ,EACR,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,MAAM,GACP,EAAE;IACD,WAAW,EAAE,MAAM,CAAC;IACpB,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,OAAO,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,MAAM,EAAE,OAAO,CAAC;IAChB,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,WAAW,EAAE,CAAC;IAC3B,SAAS,CAAC,EAAE,OAAO,CAAC;IACpB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;IACvB,MAAM,CAAC,EAAE,OAAO,mBAAmB,EAAE,UAAU,CAAC;IAChD,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;CACtD,GAAG,iBAAiB,CA0BpB;AAED,wBAAgB,mBAAmB,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG;IAAE,GAAG,EAAE,aAAa,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAS1H;AAED,wBAAgB,qBAAqB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMtG;AAED,wBAAgB,iBAAiB,CAAC,GAAG,EAAE,aAAa,EAAE,UAAU,GAAE,MAAiB,GAAG,aAAa,CAMlG;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAoB,GACrB,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAQhB;AAED,wBAAgB,cAAc,CAAC,GAAG,EAAE,aAAa,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,GAAE,MAAiB,GAAG,aAAa,CAQ7G;AAED,wBAAgB,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GACX,EAAE;IACD,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,iBAAiB,CAAC;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;CACpB,GAAG,aAAa,CAYhB;AAED,wBAAgB,uBAAuB,CAAC,KAAK,EAAE,MAAM,GAAG,uBAAuB,CA4B9E;AAED,wBAAgB,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAA8C,EAC9C,UAAqB,GACtB,EAAE;IACD,GAAG,EAAE,aAAa,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,aAAa,CAAC,EAAE,uBAAuB,CAAC;IACxC,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB,GAAG,aAAa,CAShB"}
@@ -1,7 +1,7 @@
1
1
  function nowIso() {
2
2
  return new Date().toISOString();
3
3
  }
4
- export function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model, executor, noJudge, concurrency, timeoutMs, noCache, dryRun, blind, project, owner, tags, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, }) {
4
+ export function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model, executor, noJudge, concurrency, timeoutMs, noCache, dryRun, blind, project, owner, tags, repeat, batch, judgeRepeat, judgeModels, bootstrap, bootstrapSamples, lengthDebias, budget, effort, }) {
5
5
  return {
6
6
  samplesPath,
7
7
  skillDir,
@@ -25,6 +25,7 @@ export function buildEvaluationRequest({ samplesPath, skillDir, artifacts, model
25
25
  bootstrapSamples,
26
26
  lengthDebias,
27
27
  budget,
28
+ ...(effort ? { effort } : {}),
28
29
  };
29
30
  }
30
31
  export function createEvaluationRun(runId, startedAt = nowIso()) {
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-job.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAEA,SAAS,MAAM;IACb,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,QAAQ,EACR,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,GAwBP;IACC,OAAO;QACL,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,QAAQ;QACR,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM;QACN,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;KACP,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC7E,OAAO;QACL,SAAS;QACT,GAAG,EAAE;YACH,KAAK;YACL,SAAS;YACT,MAAM,EAAE,SAAS;SAClB;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACrF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,WAAW;KACpB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACjF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,QAAQ;KACjB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAS,GAAG,MAAM,EAAE,GAKrB;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,QAAQ;QAChB,SAAS;QACT,SAAS,EAAE,SAAS;QACpB,OAAO;KACR,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,GAAkB,EAAE,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC5F,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,SAAS;QACjB,KAAK;QACL,SAAS;QACT,SAAS,EAAE,SAAS;KACrB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GASX;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,WAAW;QACnB,SAAS;QACT,SAAS;QACT,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,OAAO;QACP,KAAK;QACL,cAAc,EAAE,QAAQ;KACzB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAa;IACnD,MAAM,UAAU,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;IACvC,IACE,UAAU,CAAC,QAAQ,CAAC,wBAAwB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,sBAAsB,CAAC;WAC3C,UAAU,CAAC,QAAQ,CAAC,iBAAiB,CAAC;WACtC,UAAU,CAAC,QAAQ,CAAC,cAAc,CAAC;WACnC,UAAU,CAAC,QAAQ,CAAC,aAAa,CAAC,EACrC,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC;WACzB,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,gCAAgC,CAAC,EACxD,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,WAAW,CAAC;WAC7B,UAAU,CAAC,QAAQ,CAAC,uBAAuB,CAAC;WAC5C,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,KAAK,CAAC;WAC1B,UAAU,CAAC,QAAQ,CAAC,SAAS,CAAC,EACjC,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAAa,GAAG,uBAAuB,CAAC,KAAK,CAAC,EAC9C,UAAU,GAAG,MAAM,EAAE,GAMtB;IACC,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,QAAQ;QAChB,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,KAAK;QACL,aAAa;KACd,CAAC;AACJ,CAAC"}
1
+ {"version":3,"file":"evaluation-job.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-job.ts"],"names":[],"mappings":"AAEA,SAAS,MAAM;IACb,OAAO,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,sBAAsB,CAAC,EACrC,WAAW,EACX,QAAQ,EACR,SAAS,EACT,KAAK,EACL,QAAQ,EACR,OAAO,EACP,WAAW,EACX,SAAS,EACT,OAAO,EACP,MAAM,EACN,KAAK,EACL,OAAO,EACP,KAAK,EACL,IAAI,EACJ,MAAM,EACN,KAAK,EACL,WAAW,EACX,WAAW,EACX,SAAS,EACT,gBAAgB,EAChB,YAAY,EACZ,MAAM,EACN,MAAM,GAyBP;IACC,OAAO;QACL,WAAW;QACX,QAAQ;QACR,SAAS;QACT,KAAK;QACL,QAAQ;QACR,OAAO;QACP,WAAW;QACX,SAAS;QACT,OAAO;QACP,MAAM;QACN,KAAK;QACL,OAAO;QACP,KAAK;QACL,IAAI;QACJ,MAAM;QACN,KAAK;QACL,WAAW;QACX,WAAW;QACX,SAAS;QACT,gBAAgB;QAChB,YAAY;QACZ,MAAM;QACN,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAC9B,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC7E,OAAO;QACL,SAAS;QACT,GAAG,EAAE;YACH,KAAK;YACL,SAAS;YACT,MAAM,EAAE,SAAS;SAClB;KACF,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACrF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,WAAW;KACpB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,GAAkB,EAAE,aAAqB,MAAM,EAAE;IACjF,OAAO;QACL,GAAG,GAAG;QACN,UAAU;QACV,MAAM,EAAE,QAAQ;KACjB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,OAAO,EACP,SAAS,GAAG,MAAM,EAAE,GAKrB;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,QAAQ;QAChB,SAAS;QACT,SAAS,EAAE,SAAS;QACpB,OAAO;KACR,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,GAAkB,EAAE,KAAa,EAAE,YAAoB,MAAM,EAAE;IAC5F,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,SAAS;QACjB,KAAK;QACL,SAAS;QACT,SAAS,EAAE,SAAS;KACrB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,EACjC,KAAK,EACL,KAAK,EACL,QAAQ,EACR,OAAO,EACP,SAAS,EACT,SAAS,EACT,UAAU,GASX;IACC,OAAO;QACL,KAAK;QACL,MAAM,EAAE,WAAW;QACnB,SAAS;QACT,SAAS;QACT,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,OAAO;QACP,KAAK;QACL,cAAc,EAAE,QAAQ;KACzB,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,KAAa;IACnD,MAAM,UAAU,GAAG,KAAK,CAAC,WAAW,EAAE,CAAC;IACvC,IACE,UAAU,CAAC,QAAQ,CAAC,wBAAwB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,sBAAsB,CAAC;WAC3C,UAAU,CAAC,QAAQ,CAAC,iBAAiB,CAAC;WACtC,UAAU,CAAC,QAAQ,CAAC,cAAc,CAAC;WACnC,UAAU,CAAC,QAAQ,CAAC,aAAa,CAAC,EACrC,CAAC;QACD,OAAO,MAAM,CAAC;IAChB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC;WACzB,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,gCAAgC,CAAC,EACxD,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IACD,IACE,UAAU,CAAC,QAAQ,CAAC,WAAW,CAAC;WAC7B,UAAU,CAAC,QAAQ,CAAC,uBAAuB,CAAC;WAC5C,UAAU,CAAC,QAAQ,CAAC,qBAAqB,CAAC;WAC1C,UAAU,CAAC,QAAQ,CAAC,KAAK,CAAC;WAC1B,UAAU,CAAC,QAAQ,CAAC,SAAS,CAAC,EACjC,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;IACD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,eAAe,CAAC,EAC9B,GAAG,EACH,KAAK,EACL,aAAa,GAAG,uBAAuB,CAAC,KAAK,CAAC,EAC9C,UAAU,GAAG,MAAM,EAAE,GAMtB;IACC,OAAO;QACL,GAAG,GAAG;QACN,MAAM,EAAE,QAAQ;QAChB,UAAU;QACV,SAAS,EAAE,UAAU;QACrB,KAAK;QACL,aAAa;KACd,CAAC;AACJ,CAAC"}
@@ -1,5 +1,12 @@
1
1
  import type { Artifact, Report, Sample, Task, VariantResult, GitInfo, EvaluationJob, EvaluationRequest, EvaluationRun } from '../types/index.js';
2
2
  export declare const DEFAULT_OUTPUT_DIR: string;
3
+ export declare function hashString(str: string): string;
4
+ /**
5
+ * Stable content hash of a sample. Hashes the prompt + assertions + dimensions/rubric
6
+ * (the parts that determine what's being measured). Two samples with the same hash
7
+ * across runs measure the same thing; mismatched hashes mean the sample changed.
8
+ */
9
+ export declare function hashSample(sample: Sample): string;
3
10
  export declare function getCliVersion(): string;
4
11
  export declare function getGitInfo(): GitInfo | null;
5
12
  interface AggregateReportOptions {
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAGb,OAAO,EACP,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,mBAAmB,CAAC;AAgB3B,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AAmCzF,wBAAgB,aAAa,IAAI,MAAM,CAEtC;AAED,wBAAgB,UAAU,IAAI,OAAO,GAAG,IAAI,CAS3C;AAqDD,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CA0HjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAwC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
1
+ {"version":3,"file":"evaluation-reporting.d.ts","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EACV,QAAQ,EACR,MAAM,EACN,MAAM,EACN,IAAI,EACJ,aAAa,EAGb,OAAO,EACP,aAAa,EACb,iBAAiB,EACjB,aAAa,EACd,MAAM,mBAAmB,CAAC;AAgB3B,eAAO,MAAM,kBAAkB,EAAE,MAAuD,CAAC;AAEzF,wBAAgB,UAAU,CAAC,GAAG,EAAE,MAAM,GAAG,MAAM,CAE9C;AAeD;;;;GAIG;AACH,wBAAgB,UAAU,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CASjD;AAED,wBAAgB,aAAa,IAAI,MAAM,CAEtC;AAED,wBAAgB,UAAU,IAAI,OAAO,GAAG,IAAI,CAS3C;AAqDD,UAAU,sBAAsB;IAC9B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,OAAO,CAAC;IACjB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,KAAK,EAAE,IAAI,EAAE,CAAC;IACd,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,aAAa,CAAC,CAAC,CAAC;IACvD,YAAY,EAAE,MAAM,CAAC;IACrB,SAAS,EAAE,QAAQ,EAAE,CAAC;IACtB,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,GAAG,CAAC,EAAE,aAAa,CAAC;IACpB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAED,wBAAgB,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACb,EAAE,sBAAsB,GAAG,MAAM,CA2IjC;AAED,wBAAgB,cAAc,CAAC,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,CAwC1F;AAED,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,iBAAiB,EAAE,SAAS,EAAE,MAAM,GAAG,IAAI,GAAG,MAAM,GAAG,IAAI,CAMhG;AAED,wBAAgB,aAAa,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,MAAM,CASxD"}
@@ -22,7 +22,7 @@ function findPackageJson(startDir) {
22
22
  }
23
23
  const PKG = JSON.parse(readFileSync(findPackageJson(__dirname), 'utf-8'));
24
24
  export const DEFAULT_OUTPUT_DIR = join(homedir(), '.oh-my-knowledge', 'reports');
25
- function hashString(str) {
25
+ export function hashString(str) {
26
26
  return createHash('sha256').update(str).digest('hex').slice(0, 12);
27
27
  }
28
28
  /**
@@ -44,7 +44,7 @@ function canonicalStringify(value) {
44
44
  * (the parts that determine what's being measured). Two samples with the same hash
45
45
  * across runs measure the same thing; mismatched hashes mean the sample changed.
46
46
  */
47
- function hashSample(sample) {
47
+ export function hashSample(sample) {
48
48
  const stableForm = canonicalStringify({
49
49
  prompt: sample.prompt,
50
50
  rubric: sample.rubric ?? null,
@@ -181,6 +181,7 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
181
181
  variants,
182
182
  model,
183
183
  executor: executorName,
184
+ ...(request?.effort ? { effort: request.effort } : {}),
184
185
  sampleCount: samples.length,
185
186
  taskCount: tasks.length,
186
187
  totalCostUSD: Number(totalCostUSD.toFixed(6)),
@@ -215,6 +216,22 @@ export function aggregateReport({ runId, variants, model, judgeModel, noJudge, e
215
216
  sample_id,
216
217
  variants: variantData,
217
218
  })),
219
+ // 用例设计快照,供单测视角渲染。只挑渲染需要的字段,跳过 cwd / allowedTools /
220
+ // expectedTools / dimensions / environment(对单测视图无附加价值)。Sample 字段
221
+ // 全选会让 report 体积接近翻倍,选子集 size 增长 ~10-20%。
222
+ sampleSnapshots: Object.fromEntries(samples.map((s) => [s.sample_id, {
223
+ sample_id: s.sample_id,
224
+ prompt: s.prompt,
225
+ ...(s.rubric ? { rubric: s.rubric } : {}),
226
+ ...(s.context ? { context: s.context } : {}),
227
+ ...(s.assertions && s.assertions.length > 0 ? { assertions: s.assertions } : {}),
228
+ ...(s.mocks && s.mocks.length > 0 ? { mocks: s.mocks } : {}),
229
+ ...(s.capability && s.capability.length > 0 ? { capability: s.capability } : {}),
230
+ ...(s.difficulty ? { difficulty: s.difficulty } : {}),
231
+ ...(s.construct ? { construct: s.construct } : {}),
232
+ ...(s.provenance ? { provenance: s.provenance } : {}),
233
+ ...(s.tripwire ? { tripwire: true } : {}),
234
+ }])),
218
235
  };
219
236
  }
220
237
  export function applyBlindMode(report, variants, blindSeed) {
@@ -1 +1 @@
1
- {"version":3,"file":"evaluation-reporting.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACvF,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAClE,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AAepF,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,SAAS,eAAe,CAAC,QAAgB;IACvC,IAAI,GAAG,GAAG,QAAQ,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;QAC5C,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,GAAG,GAAwB,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,eAAe,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;AAEtH,MAAM,CAAC,MAAM,kBAAkB,GAAW,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,SAAS,CAAC,CAAC;AAEzF,SAAS,UAAU,CAAC,GAAW;IAC7B,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACrE,CAAC;AAED;;;;;GAKG;AACH,SAAS,kBAAkB,CAAC,KAAc;IACxC,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IAC9E,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;IACrF,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,OAAO,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,kBAAkB,CAAE,KAAiC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;AACvI,CAAC;AAED;;;;GAIG;AACH,SAAS,UAAU,CAAC,MAAc;IAChC,MAAM,UAAU,GAAG,kBAAkB,CAAC;QACpC,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;KAC9B,CAAC,CAAC;IACH,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa;IAC3B,OAAO,GAAG,CAAC,OAAO,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,UAAU;IACxB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxG,MAAM,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QACtG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,QAA0E;IAC/F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACxB,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAClG,CAAC;AAED,SAAS,qBAAqB,CAAC,QAA0E;IACvG,OAAO,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,8BAA8B,CAAC,EACtC,QAAQ,EACR,KAAK,EACL,YAAY,EACZ,KAAK,EACL,SAAS,EACT,OAAO,GAQR;IACC,MAAM,QAAQ,GAAqE,EAAE,CAAC;IACtF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;QACvF,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YAC1E,QAAQ,EAAE,aAAa,CAAC,KAAK,CAAC,QAAQ;SACvC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QAChC,MAAM,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC;QAC3D,MAAM,gBAAgB,GAAG,QAAQ,EAAE,IAAI,KAAK,UAAU;YACpD,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,QAAQ,EAAE,OAAO,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK;gBAC9C,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAC3B,CAAC,CAAC,OAAO,EAAE,QAAQ,CAAC;QACxB,QAAQ,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YACrE,QAAQ,EAAE,gBAAgB;SAC3B,CAAC,CAAC;IACL,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAoBD,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACW;IACvB,MAAM,OAAO,GAAmC,EAAE,CAAC;IACnD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACxF,OAAO,CAAC,OAAO,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC;IAED,kFAAkF;IAClF,mFAAmF;IACnF,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,KAAK,IAAI,CAAC;IACrD,MAAM,gBAAgB,GAAG,OAAO,EAAE,gBAAgB,IAAI,IAAI,CAAC;IAC3D,IAAI,eAAoD,CAAC;IACzD,IAAI,gBAAgB,EAAE,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,OAAO;iBAC5B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;gBAChC,OAAO,CAAC,OAAO,CAAC,CAAC,WAAW,GAAG,eAAe,CAAC,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC,CAAC;YAC1F,CAAC;QACH,CAAC;QAED,iFAAiF;QACjF,8DAA8D;QAC9D,IAAI,QAAQ,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YACzB,eAAe,GAAG,EAAE,CAAC;YACrB,MAAM,WAAW,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACzF,MAAM,aAAa,GAAG,cAAc;iBACjC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACzC,MAAM,aAAa,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAClC,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC7F,MAAM,eAAe,GAAG,gBAAgB;qBACrC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;qBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;gBACjC,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;oBAC7D,eAAe,CAAC,IAAI,CAAC;wBACnB,OAAO,EAAE,WAAW;wBACpB,SAAS,EAAE,aAAa;wBACxB,eAAe,EAAE,eAAe,CAAC,aAAa,EAAE,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC;qBACzF,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,WAAW,CACvC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAC3G,CAAC;IAEF,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1F,MAAM,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,OAAO,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACtG,MAAM,cAAc,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;IACvD,MAAM,gBAAgB,GAAG,8BAA8B,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;IACtH,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,qBAAqB,CAAC,gBAAgB,CAAC;WACvC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IACxE,oFAAoF;IACpF,iFAAiF;IACjF,sFAAsF;IACtF,MAAM,aAAa,GAAG,OAAO,EAAE,WAAW,IAAI,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAC9F,MAAM,eAAe,GAAoD,aAAa,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAClG,QAAQ,EAAE,EAAE,CAAC,QAAQ;QACrB,KAAK,EAAE,EAAE,CAAC,KAAK;QACf,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,6BAA6B,CAAC,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,EAAE,cAAc,CAAC,EAAE,CAAC;KACtG,CAAC,CAAC,CAAC;IACJ,2DAA2D;IAC3D,0EAA0E;IAC1E,4EAA4E;IAC5E,MAAM,cAAc,GAAG,OAAO,EAAE,YAAY,KAAK,KAAK,CAAC;IACvD,MAAM,cAAc,GAAiC,EAAE,CAAC;IACxD,IAAI,cAAc;QAAE,cAAc,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAClD,MAAM,iBAAiB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CACjE,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC;IAE7E,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK;QACT,IAAI,EAAE;YACJ,QAAQ;YACR,KAAK;YACL,QAAQ,EAAE,YAAY;YACtB,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,EAAE,CAAC;YAC1D,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,cAAc;YACd,YAAY;YACZ,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,kBAAkB,CAAC,cAAc,CAAC,EAAE,CAAC;YAC3E,eAAe;YACf,gBAAgB;YAChB,WAAW,EAAE,eAAe;YAC5B,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,GAAG,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpE,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,MAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrE,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/C,cAAc,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,kBAAkB,CAAC,QAAQ,CAAC,CAAC;YACzE,6EAA6E;YAC7E,6DAA6D;YAC7D,4CAA4C;YAC5C,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,CAAC,CAAC,CACxD;YACD,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;YACrB,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChD;QACD,OAAO;QACP,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC;YAClE,SAAS;YACT,QAAQ,EAAE,WAAW;SACtB,CAAC,CAAC;KACJ,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAkB,EAAE,SAAiB;IAClF,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAC/D,MAAM,YAAY,GAAG,GAAW,EAAE;QAChC,IAAI,IAAI,CAAC,CAAC;QACV,IAAI,GAAG,IAAI,GAAG,UAAU,GAAG,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,KAAK,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACpD,KAAK,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;QAC5D,OAAO,CAAC,CAAC,KAAK,GAAG,KAAK,KAAK,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACrD,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/C,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GAA2B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,UAAU,GAA2B,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAEpI,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAChC,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC;IAC9B,IAAI,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAC/C,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,OAAO,EAAE,OAAO,CAAC,CAAC,CACpH,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAmC,EAAE,CAAC;IACtD,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,UAAU,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,KAAK,CAAC;IAC1C,CAAC;IACD,MAAM,CAAC,OAAO,GAAG,UAAU,CAAC;IAE5B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,MAAM,WAAW,GAAkC,EAAE,CAAC;QACtD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9D,WAAW,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC;QAC1C,CAAC;QACD,MAAM,CAAC,QAAQ,GAAG,WAAW,CAAC;IAChC,CAAC;AACH,CAAC;AAMD,MAAM,UAAU,aAAa,CAAC,MAAyB,EAAE,SAAwB;IAC/E,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,GAAG,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC;IACtD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACzD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAkB;IAC9C,MAAM,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;IACrB,MAAM,GAAG,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,WAAW,GAAG,QAAQ;SACzB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC;SACzF,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO,GAAG,WAAW,IAAI,IAAI,IAAI,IAAI,EAAE,CAAC;AAC1C,CAAC"}
1
+ {"version":3,"file":"evaluation-reporting.js","sourceRoot":"","sources":["../../../src/eval-core/evaluation-reporting.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,SAAS,EAAE,UAAU,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAClC,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,aAAa,EAAE,MAAM,UAAU,CAAC;AACzC,OAAO,EAAE,mBAAmB,EAAE,MAAM,aAAa,CAAC;AAClD,OAAO,EAAE,kBAAkB,EAAE,wBAAwB,EAAE,MAAM,yBAAyB,CAAC;AACvF,OAAO,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,EAAE,eAAe,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AAClE,OAAO,EAAE,6BAA6B,EAAE,MAAM,qCAAqC,CAAC;AAepF,MAAM,SAAS,GAAG,OAAO,CAAC,aAAa,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC;AAE1D,SAAS,eAAe,CAAC,QAAgB;IACvC,IAAI,GAAG,GAAG,QAAQ,CAAC;IACnB,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC3B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,cAAc,CAAC,CAAC;QAC5C,IAAI,UAAU,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;QAC5C,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,QAAQ,EAAE,IAAI,EAAE,cAAc,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,GAAG,GAAwB,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,eAAe,CAAC,SAAS,CAAC,EAAE,OAAO,CAAC,CAAwB,CAAC;AAEtH,MAAM,CAAC,MAAM,kBAAkB,GAAW,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,SAAS,CAAC,CAAC;AAEzF,MAAM,UAAU,UAAU,CAAC,GAAW;IACpC,OAAO,UAAU,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AACrE,CAAC;AAED;;;;;GAKG;AACH,SAAS,kBAAkB,CAAC,KAAc;IACxC,IAAI,KAAK,KAAK,IAAI,IAAI,OAAO,KAAK,KAAK,QAAQ;QAAE,OAAO,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;IAC9E,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,GAAG,GAAG,KAAK,CAAC,GAAG,CAAC,kBAAkB,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;IACrF,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,CAAC,KAAgC,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,OAAO,GAAG,GAAG,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,kBAAkB,CAAE,KAAiC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,GAAG,GAAG,CAAC;AACvI,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,UAAU,CAAC,MAAc;IACvC,MAAM,UAAU,GAAG,kBAAkB,CAAC;QACpC,MAAM,EAAE,MAAM,CAAC,MAAM;QACrB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;QAC7B,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,UAAU,EAAE,MAAM,CAAC,UAAU,IAAI,IAAI;QACrC,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,IAAI;KAC9B,CAAC,CAAC;IACH,OAAO,UAAU,CAAC,UAAU,CAAC,CAAC;AAChC,CAAC;AAED,MAAM,UAAU,aAAa;IAC3B,OAAO,GAAG,CAAC,OAAO,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,UAAU;IACxB,IAAI,CAAC;QACH,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxF,MAAM,MAAM,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,WAAW,EAAE,cAAc,EAAE,MAAM,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACxG,MAAM,KAAK,GAAG,YAAY,CAAC,KAAK,EAAE,CAAC,QAAQ,EAAE,aAAa,CAAC,EAAE,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QACtG,OAAO,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;IACpE,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,IAAI,CAAC;IACd,CAAC;AACH,CAAC;AAED,SAAS,aAAa,CAAC,QAA0E;IAC/F,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,SAAS,CAAC;IAC1C,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC;IACxB,OAAO,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,WAAW,KAAK,KAAK,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;AAClG,CAAC;AAED,SAAS,qBAAqB,CAAC,QAA0E;IACvG,OAAO,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,8BAA8B,CAAC,EACtC,QAAQ,EACR,KAAK,EACL,YAAY,EACZ,KAAK,EACL,SAAS,EACT,OAAO,GAQR;IACC,MAAM,QAAQ,GAAqE,EAAE,CAAC;IACtF,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;QACzB,IAAI,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,aAAa,GAAG,wBAAwB,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,EAAE,SAAS,EAAE,KAAK,CAAC,CAAC;QACvF,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YAC1E,QAAQ,EAAE,aAAa,CAAC,KAAK,CAAC,QAAQ;SACvC,CAAC,CAAC;IACL,CAAC;IAED,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,QAAQ,CAAC,OAAO,CAAC;YAAE,SAAS;QAChC,MAAM,QAAQ,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,OAAO,CAAC,CAAC;QAC3D,MAAM,gBAAgB,GAAG,QAAQ,EAAE,IAAI,KAAK,UAAU;YACpD,CAAC,CAAC,IAAI;YACN,CAAC,CAAC,QAAQ,EAAE,OAAO,IAAI,QAAQ,CAAC,MAAM,KAAK,KAAK;gBAC9C,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAC3B,CAAC,CAAC,OAAO,EAAE,QAAQ,CAAC;QACxB,QAAQ,CAAC,OAAO,CAAC,GAAG,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE;YACrE,QAAQ,EAAE,gBAAgB;SAC3B,CAAC,CAAC;IACL,CAAC;IAED,OAAO,QAAQ,CAAC;AAClB,CAAC;AAoBD,MAAM,UAAU,eAAe,CAAC,EAC9B,KAAK,EACL,QAAQ,EACR,KAAK,EACL,UAAU,EACV,OAAO,EACP,YAAY,EACZ,OAAO,EACP,KAAK,EACL,OAAO,EACP,YAAY,EACZ,SAAS,EACT,OAAO,EACP,GAAG,EACH,GAAG,EACH,YAAY,GACW;IACvB,MAAM,OAAO,GAAmC,EAAE,CAAC;IACnD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;QACxF,OAAO,CAAC,OAAO,CAAC,GAAG,mBAAmB,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC;IAED,kFAAkF;IAClF,mFAAmF;IACnF,MAAM,gBAAgB,GAAG,OAAO,EAAE,SAAS,KAAK,IAAI,CAAC;IACrD,MAAM,gBAAgB,GAAG,OAAO,EAAE,gBAAgB,IAAI,IAAI,CAAC;IAC3D,IAAI,eAAoD,CAAC;IACzD,IAAI,gBAAgB,EAAE,CAAC;QACrB,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,MAAM,OAAO,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC9E,MAAM,eAAe,GAAG,OAAO;iBAC5B,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;gBAChC,OAAO,CAAC,OAAO,CAAC,CAAC,WAAW,GAAG,eAAe,CAAC,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC,CAAC;YAC1F,CAAC;QACH,CAAC;QAED,iFAAiF;QACjF,8DAA8D;QAC9D,IAAI,QAAQ,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YACzB,eAAe,GAAG,EAAE,CAAC;YACrB,MAAM,WAAW,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YAChC,MAAM,cAAc,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YACzF,MAAM,aAAa,GAAG,cAAc;iBACjC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;iBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;YACjC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACzC,MAAM,aAAa,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;gBAClC,MAAM,gBAAgB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;gBAC7F,MAAM,eAAe,GAAG,gBAAgB;qBACrC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,OAAO,CAAC,CAAC,cAAc,KAAK,QAAQ,IAAI,CAAC,CAAC,cAAe,GAAG,CAAC,CAAC;qBAC5E,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAe,CAAC,CAAC;gBACjC,IAAI,aAAa,CAAC,MAAM,IAAI,CAAC,IAAI,eAAe,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;oBAC7D,eAAe,CAAC,IAAI,CAAC;wBACnB,OAAO,EAAE,WAAW;wBACpB,SAAS,EAAE,aAAa;wBACxB,eAAe,EAAE,eAAe,CAAC,aAAa,EAAE,eAAe,EAAE,IAAI,EAAE,gBAAgB,CAAC;qBACzF,CAAC,CAAC;gBACL,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IAED,MAAM,cAAc,GAAG,MAAM,CAAC,WAAW,CACvC,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAC3G,CAAC;IAEF,MAAM,YAAY,GAAG,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1F,MAAM,WAAW,GAAG,OAAO,EAAE,WAAW,IAAI,OAAO,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC;IACtG,MAAM,cAAc,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,QAAQ,EAAE,CAAC;IACvD,MAAM,gBAAgB,GAAG,8BAA8B,CAAC,EAAE,QAAQ,EAAE,KAAK,EAAE,YAAY,EAAE,KAAK,EAAE,SAAS,EAAE,OAAO,EAAE,CAAC,CAAC;IACtH,MAAM,eAAe,GAAG,aAAa,CAAC,gBAAgB,CAAC;WAClD,qBAAqB,CAAC,gBAAgB,CAAC;WACvC,6BAA6B,CAAC,YAAY,EAAE,KAAK,EAAE,cAAc,CAAC,CAAC;IACxE,oFAAoF;IACpF,iFAAiF;IACjF,sFAAsF;IACtF,MAAM,aAAa,GAAG,OAAO,EAAE,WAAW,IAAI,CAAC,EAAE,QAAQ,EAAE,YAAY,EAAE,KAAK,EAAE,UAAU,EAAE,CAAC,CAAC;IAC9F,MAAM,eAAe,GAAoD,aAAa,CAAC,GAAG,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,CAAC;QAClG,QAAQ,EAAE,EAAE,CAAC,QAAQ;QACrB,KAAK,EAAE,EAAE,CAAC,KAAK;QACf,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,6BAA6B,CAAC,EAAE,CAAC,QAAQ,EAAE,EAAE,CAAC,KAAK,EAAE,cAAc,CAAC,EAAE,CAAC;KACtG,CAAC,CAAC,CAAC;IACJ,2DAA2D;IAC3D,0EAA0E;IAC1E,4EAA4E;IAC5E,MAAM,cAAc,GAAG,OAAO,EAAE,YAAY,KAAK,KAAK,CAAC;IACvD,MAAM,cAAc,GAAiC,EAAE,CAAC;IACxD,IAAI,cAAc;QAAE,cAAc,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAClD,MAAM,iBAAiB,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,OAAO,EAAE,EAAE,CACjE,OAAO,CAAC,gBAAgB,KAAK,KAAK,IAAI,OAAO,CAAC,iBAAiB,KAAK,KAAK,CAAC,CAAC;IAE7E,OAAO;QACL,IAAI,EAAE,YAAY;QAClB,EAAE,EAAE,KAAK;QACT,IAAI,EAAE;YACJ,QAAQ;YACR,KAAK;YACL,QAAQ,EAAE,YAAY;YACtB,GAAG,CAAC,OAAO,EAAE,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACtD,WAAW,EAAE,OAAO,CAAC,MAAM;YAC3B,SAAS,EAAE,KAAK,CAAC,MAAM;YACvB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;YAC7C,GAAG,CAAC,iBAAiB,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,iBAAiB,EAAE,KAAK,EAAE,CAAC;YAC1D,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;YACnC,UAAU,EAAE,aAAa,EAAE;YAC3B,WAAW,EAAE,OAAO,CAAC,OAAO;YAC5B,cAAc;YACd,YAAY;YACZ,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,kBAAkB,CAAC,cAAc,CAAC,EAAE,CAAC;YAC3E,eAAe;YACf,gBAAgB;YAChB,WAAW,EAAE,eAAe;YAC5B,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrC,GAAG,CAAC,WAAW,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACvC,GAAG,CAAC,cAAc,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,cAAc,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACpE,GAAG,CAAC,gBAAgB,CAAC,CAAC,CAAC,EAAE,mBAAmB,EAAE,MAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YACrE,GAAG,CAAC,eAAe,CAAC,CAAC,CAAC,EAAE,eAAe,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;YAC/C,cAAc,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,QAAQ,EAAE,EAAE,CAAC,kBAAkB,CAAC,QAAQ,CAAC,CAAC;YACzE,6EAA6E;YAC7E,6DAA6D;YAC7D,4CAA4C;YAC5C,cAAc,EAAE,MAAM,CAAC,WAAW,CAChC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,aAAa,IAAI,IAAI,CAAC,CAAC,CACxD;YACD,OAAO;YACP,GAAG;YACH,GAAG;YACH,OAAO,EAAE,UAAU,EAAE;YACrB,GAAG,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,YAAY,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;SAChD;QACD,OAAO;QACP,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,WAAW,CAAC,EAAE,EAAE,CAAC,CAAC;YAClE,SAAS;YACT,QAAQ,EAAE,WAAW;SACtB,CAAC,CAAC;QACH,mDAAmD;QACnD,iEAAiE;QACjE,0CAA0C;QAC1C,eAAe,EAAE,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,SAAS,EAAE;gBACnE,SAAS,EAAE,CAAC,CAAC,SAAS;gBACtB,MAAM,EAAE,CAAC,CAAC,MAAM;gBAChB,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACzC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5C,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAC5D,GAAG,CAAC,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAChF,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBAClD,GAAG,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,CAAC,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;gBACrD,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAC1C,CAAC,CAAC,CAAC;KACL,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAc,EAAE,QAAkB,EAAE,SAAiB;IAClF,MAAM,MAAM,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,YAAY,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC;IACnE,IAAI,IAAI,GAAG,QAAQ,CAAC,UAAU,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC;IAC/D,MAAM,YAAY,GAAG,GAAW,EAAE;QAChC,IAAI,IAAI,CAAC,CAAC;QACV,IAAI,GAAG,IAAI,GAAG,UAAU,GAAG,CAAC,CAAC;QAC7B,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,KAAK,EAAE,EAAE,CAAC,GAAG,IAAI,CAAC,CAAC;QACpD,KAAK,IAAI,KAAK,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,GAAG,KAAK,KAAK,CAAC,EAAE,EAAE,GAAG,KAAK,CAAC,CAAC;QAC5D,OAAO,CAAC,CAAC,KAAK,GAAG,KAAK,KAAK,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,CAAC;IACrD,CAAC,CAAC;IACF,MAAM,QAAQ,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC7C,MAAM,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,EAAE,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/C,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC;IAC1D,CAAC;IACD,MAAM,QAAQ,GAA2B,MAAM,CAAC,WAAW,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC;IAChH,MAAM,UAAU,GAA2B,MAAM,CAAC,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,EAAE,KAAK,CAAC,CAAC,CAAC,CAAC;IAEpI,MAAM,CAAC,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC;IACzB,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,QAAQ,CAAC;IAChC,MAAM,CAAC,IAAI,CAAC,QAAQ,GAAG,MAAM,CAAC;IAC9B,IAAI,MAAM,CAAC,IAAI,CAAC,gBAAgB,EAAE,CAAC;QACjC,MAAM,CAAC,IAAI,CAAC,gBAAgB,GAAG,MAAM,CAAC,WAAW,CAC/C,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,OAAO,CAAC,EAAE,EAAE,CAAC,CAAC,UAAU,CAAC,OAAO,CAAC,IAAI,OAAO,EAAE,OAAO,CAAC,CAAC,CACpH,CAAC;IACJ,CAAC;IAED,MAAM,UAAU,GAAmC,EAAE,CAAC;IACtD,KAAK,MAAM,CAAC,OAAO,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAAC;QAC9D,UAAU,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,KAAK,CAAC;IAC1C,CAAC;IACD,MAAM,CAAC,OAAO,GAAG,UAAU,CAAC;IAE5B,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,MAAM,WAAW,GAAkC,EAAE,CAAC;QACtD,KAAK,MAAM,CAAC,OAAO,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC9D,WAAW,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC,GAAG,IAAI,CAAC;QAC1C,CAAC;QACD,MAAM,CAAC,QAAQ,GAAG,WAAW,CAAC;IAChC,CAAC;AACH,CAAC;AAMD,MAAM,UAAU,aAAa,CAAC,MAAyB,EAAE,SAAwB;IAC/E,IAAI,CAAC,SAAS;QAAE,OAAO,IAAI,CAAC;IAC5B,IAAI,CAAC,UAAU,CAAC,SAAS,CAAC;QAAE,SAAS,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtE,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,GAAG,MAAM,CAAC,EAAE,OAAO,CAAC,CAAC;IACtD,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IACzD,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,QAAkB;IAC9C,MAAM,CAAC,GAAG,IAAI,IAAI,EAAE,CAAC;IACrB,MAAM,GAAG,GAAG,CAAC,CAAS,EAAU,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;IAC9D,MAAM,IAAI,GAAG,GAAG,CAAC,CAAC,WAAW,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,EAAE,CAAC;IAC7E,MAAM,IAAI,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,UAAU,EAAE,CAAC,EAAE,CAAC;IAC1D,MAAM,WAAW,GAAG,QAAQ;SACzB,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,UAAU,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC,UAAU,CAAC,mBAAmB,EAAE,GAAG,CAAC,CAAC;SACzF,IAAI,CAAC,MAAM,CAAC,CAAC;IAChB,OAAO,GAAG,WAAW,IAAI,IAAI,IAAI,IAAI,EAAE,CAAC;AAC1C,CAAC"}
@@ -7,5 +7,5 @@ export interface ExecutionPlan {
7
7
  export declare function resolveArtifactExecutionStrategy(artifact: Artifact): ExecutionStrategyKind;
8
8
  export declare function resolveExperimentType(artifact: Artifact): ExperimentType;
9
9
  export declare function buildVariantConfig(artifact: Artifact): VariantConfig;
10
- export declare function resolveExecutionStrategy(task: Task, model: string, timeoutMs?: number, verbose?: boolean): ExecutionPlan;
10
+ export declare function resolveExecutionStrategy(task: Task, model: string, timeoutMs?: number, verbose?: boolean, effort?: 'low' | 'medium' | 'high' | 'xhigh' | 'max', samplesBaseDir?: string): ExecutionPlan;
11
11
  //# sourceMappingURL=execution-strategy.d.ts.map
@@ -1 +1 @@
1
- {"version":3,"file":"execution-strategy.d.ts","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,QAAQ,EAAE,qBAAqB,EAAE,aAAa,EAAE,cAAc,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAmB7H,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,qBAAqB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,CAAC;CACtB;AAOD,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,QAAQ,GAAG,qBAAqB,CAc1F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,QAAQ,GAAG,cAAc,CAIxE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,QAAQ,GAAG,aAAa,CA4BpE;AAOD,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,GAAG,aAAa,CA4ExH"}
1
+ {"version":3,"file":"execution-strategy.d.ts","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AAGA,OAAO,KAAK,EAAE,QAAQ,EAAE,qBAAqB,EAAE,aAAa,EAAE,cAAc,EAAE,IAAI,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAmB7H,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,qBAAqB,CAAC;IAChC,WAAW,EAAE,MAAM,CAAC;IACpB,KAAK,EAAE,aAAa,CAAC;CACtB;AAOD,wBAAgB,gCAAgC,CAAC,QAAQ,EAAE,QAAQ,GAAG,qBAAqB,CAc1F;AAED,wBAAgB,qBAAqB,CAAC,QAAQ,EAAE,QAAQ,GAAG,cAAc,CAIxE;AAED,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,QAAQ,GAAG,aAAa,CA4BpE;AAOD,wBAAgB,wBAAwB,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,OAAO,EAAE,MAAM,CAAC,EAAE,KAAK,GAAG,QAAQ,GAAG,MAAM,GAAG,OAAO,GAAG,KAAK,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,aAAa,CAuFvM"}
@@ -76,7 +76,7 @@ function extractSkillDir(artifact) {
76
76
  return null;
77
77
  return dirname(artifact.locator);
78
78
  }
79
- export function resolveExecutionStrategy(task, model, timeoutMs, verbose) {
79
+ export function resolveExecutionStrategy(task, model, timeoutMs, verbose, effort, samplesBaseDir) {
80
80
  const skillDir = extractSkillDir(task.artifact);
81
81
  // strict-baseline cwd 沙箱:baseline + allowedSkills===[] + 用户没显式
82
82
  // cwd 时,改用 isolated empty dir。否则 baseline 的 Glob/Read 会走进用户工作目录
@@ -92,10 +92,21 @@ export function resolveExecutionStrategy(task, model, timeoutMs, verbose) {
92
92
  skillDir,
93
93
  timeoutMs,
94
94
  verbose,
95
+ ...(effort && { effort }),
95
96
  // pass skill-isolation declaration to executors. undefined keeps
96
97
  // SDK default; [] = strict isolation (skills:[] + disallowedTools:['Skill']);
97
98
  // [...] = whitelist (skills:[...] only).
98
99
  ...(task.artifact.allowedSkills !== undefined && { allowedSkills: task.artifact.allowedSkills }),
100
+ // Sample.mocks 透传到 executor。executor(claude-sdk / claude-cli)
101
+ // 自决定怎么落地(in-process hook vs 临时 CLAUDE_CONFIG_DIR + on-disk hook)。
102
+ // mocksBaseDir 优先 samplesBaseDir(sample bundle 根,跟 ExecutorInput 类型注释一致),
103
+ // 让 mock.return_file 的相对路径相对 sample 文件所在目录解析;
104
+ // 没传时 fallback skillDir → effectiveCwd,兼容老调用。
105
+ ...(task._sample.mocks && task._sample.mocks.length > 0 && {
106
+ mocks: task._sample.mocks,
107
+ mocksBaseDir: samplesBaseDir || skillDir || effectiveCwd || undefined,
108
+ ...(task._sample.mocksStrict && { mocksStrict: true }),
109
+ }),
99
110
  };
100
111
  switch (task.artifact.kind) {
101
112
  case 'baseline':
@@ -1 +1 @@
1
- {"version":3,"file":"execution-strategy.js","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAGlC;;;;;;;;GAQG;AACH,SAAS,cAAc;IACrB,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,cAAc,CAAC,CAAC;IAChE,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;QACrB,SAAS,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAQD,SAAS,iBAAiB,CAAC,MAAqB,EAAE,MAAc;IAC9D,IAAI,CAAC,MAAM;QAAE,OAAO,MAAM,CAAC;IAC3B,OAAO,GAAG,MAAM,OAAO,MAAM,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,gCAAgC,CAAC,QAAkB;IACjE,QAAQ,QAAQ,CAAC,IAAI,EAAE,CAAC;QACtB,KAAK,UAAU;YACb,OAAO,UAAU,CAAC;QACpB,KAAK,QAAQ;YACX,OAAO,aAAa,CAAC;QACvB,KAAK,OAAO;YACV,OAAO,eAAe,CAAC;QACzB,KAAK,UAAU;YACb,OAAO,kBAAkB,CAAC;QAC5B,KAAK,OAAO,CAAC;QACb;YACE,OAAO,eAAe,CAAC;IAC3B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,QAAkB;IACtD,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,QAAQ,CAAC,GAAG;QAAE,OAAO,UAAU,CAAC;IACrE,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,IAAI,QAAQ,CAAC,GAAG;QAAE,OAAO,sBAAsB,CAAC;IAChF,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,QAAkB;IACnD,oDAAoD;IACpD,kFAAkF;IAClF,4DAA4D;IAC5D,iEAAiE;IACjE,qEAAqE;IACrE,kCAAkC;IAClC,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CACb,aAAa,QAAQ,CAAC,IAAI,oDAAoD;cAC5E,2EAA2E,CAC9E,CAAC;IACJ,CAAC;IACD,OAAO;QACL,OAAO,EAAE,QAAQ,CAAC,IAAI;QACtB,YAAY,EAAE,QAAQ,CAAC,IAAI;QAC3B,cAAc,EAAE,QAAQ,CAAC,MAAM;QAC/B,iBAAiB,EAAE,gCAAgC,CAAC,QAAQ,CAAC;QAC7D,cAAc,EAAE,qBAAqB,CAAC,QAAQ,CAAC;QAC/C,cAAc,EAAE,QAAQ,CAAC,cAAc;QACvC,kBAAkB,EAAE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;QAC7C,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI;QACzB,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,GAAG,EAAE,QAAQ,CAAC,GAAG;QACjB,sEAAsE;QACtE,gFAAgF;QAChF,GAAG,CAAC,QAAQ,CAAC,aAAa,KAAK,SAAS,IAAI,EAAE,aAAa,EAAE,QAAQ,CAAC,aAAa,EAAE,CAAC;KACvF,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,QAAkB;IACzC,IAAI,CAAC,QAAQ,CAAC,OAAO;QAAE,OAAO,IAAI,CAAC;IACnC,OAAO,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,IAAU,EAAE,KAAa,EAAE,SAAkB,EAAE,OAAiB;IACvG,MAAM,QAAQ,GAAG,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAChD,+DAA+D;IAC/D,gEAAgE;IAChE,6DAA6D;IAC7D,uCAAuC;IACvC,MAAM,gBAAgB,GACpB,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,UAAU;WAC9B,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC;WAC1C,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,MAAM,KAAK,CAAC,CAAC;IAC9C,MAAM,YAAY,GAAG,gBAAgB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;IACjF,MAAM,SAAS,GAAG;QAChB,KAAK;QACL,GAAG,EAAE,YAAY;QACjB,QAAQ;QACR,SAAS;QACT,OAAO;QACP,iEAAiE;QACjE,8EAA8E;QAC9E,yCAAyC;QACzC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,aAAa,KAAK,SAAS,IAAI,EAAE,aAAa,EAAE,IAAI,CAAC,QAAQ,CAAC,aAAa,EAAE,CAAC;KACjG,CAAC;IAEF,QAAQ,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC3B,KAAK,UAAU;YACb,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,EAAE;gBACf,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI;oBACZ,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,QAAQ;YACX,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,EAAE;gBACf,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI;oBACZ,MAAM,EAAE,iBAAiB,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC;iBAC9D;aACF,CAAC;QACJ,KAAK,OAAO;YACV,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,UAAU;YACb,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,OAAO,CAAC;QACb;YACE,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;IACN,CAAC;AACH,CAAC"}
1
+ {"version":3,"file":"execution-strategy.js","sourceRoot":"","sources":["../../../src/eval-core/execution-strategy.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAC1C,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,SAAS,CAAC;AAChD,OAAO,EAAE,OAAO,EAAE,MAAM,SAAS,CAAC;AAGlC;;;;;;;;GAQG;AACH,SAAS,cAAc;IACrB,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,kBAAkB,EAAE,cAAc,CAAC,CAAC;IAChE,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,EAAE,CAAC;QACrB,SAAS,CAAC,GAAG,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACtC,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAQD,SAAS,iBAAiB,CAAC,MAAqB,EAAE,MAAc;IAC9D,IAAI,CAAC,MAAM;QAAE,OAAO,MAAM,CAAC;IAC3B,OAAO,GAAG,MAAM,OAAO,MAAM,EAAE,CAAC;AAClC,CAAC;AAED,MAAM,UAAU,gCAAgC,CAAC,QAAkB;IACjE,QAAQ,QAAQ,CAAC,IAAI,EAAE,CAAC;QACtB,KAAK,UAAU;YACb,OAAO,UAAU,CAAC;QACpB,KAAK,QAAQ;YACX,OAAO,aAAa,CAAC;QACvB,KAAK,OAAO;YACV,OAAO,eAAe,CAAC;QACzB,KAAK,UAAU;YACb,OAAO,kBAAkB,CAAC;QAC5B,KAAK,OAAO,CAAC;QACb;YACE,OAAO,eAAe,CAAC;IAC3B,CAAC;AACH,CAAC;AAED,MAAM,UAAU,qBAAqB,CAAC,QAAkB;IACtD,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,IAAI,CAAC,QAAQ,CAAC,GAAG;QAAE,OAAO,UAAU,CAAC;IACrE,IAAI,QAAQ,CAAC,IAAI,KAAK,UAAU,IAAI,QAAQ,CAAC,GAAG;QAAE,OAAO,sBAAsB,CAAC;IAChF,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,MAAM,UAAU,kBAAkB,CAAC,QAAkB;IACnD,oDAAoD;IACpD,kFAAkF;IAClF,4DAA4D;IAC5D,iEAAiE;IACjE,qEAAqE;IACrE,kCAAkC;IAClC,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CACb,aAAa,QAAQ,CAAC,IAAI,oDAAoD;cAC5E,2EAA2E,CAC9E,CAAC;IACJ,CAAC;IACD,OAAO;QACL,OAAO,EAAE,QAAQ,CAAC,IAAI;QACtB,YAAY,EAAE,QAAQ,CAAC,IAAI;QAC3B,cAAc,EAAE,QAAQ,CAAC,MAAM;QAC/B,iBAAiB,EAAE,gCAAgC,CAAC,QAAQ,CAAC;QAC7D,cAAc,EAAE,qBAAqB,CAAC,QAAQ,CAAC;QAC/C,cAAc,EAAE,QAAQ,CAAC,cAAc;QACvC,kBAAkB,EAAE,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC;QAC7C,GAAG,EAAE,QAAQ,CAAC,GAAG,IAAI,IAAI;QACzB,OAAO,EAAE,QAAQ,CAAC,OAAO;QACzB,GAAG,EAAE,QAAQ,CAAC,GAAG;QACjB,sEAAsE;QACtE,gFAAgF;QAChF,GAAG,CAAC,QAAQ,CAAC,aAAa,KAAK,SAAS,IAAI,EAAE,aAAa,EAAE,QAAQ,CAAC,aAAa,EAAE,CAAC;KACvF,CAAC;AACJ,CAAC;AAED,SAAS,eAAe,CAAC,QAAkB;IACzC,IAAI,CAAC,QAAQ,CAAC,OAAO;QAAE,OAAO,IAAI,CAAC;IACnC,OAAO,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,wBAAwB,CAAC,IAAU,EAAE,KAAa,EAAE,SAAkB,EAAE,OAAiB,EAAE,MAAoD,EAAE,cAAuB;IACtL,MAAM,QAAQ,GAAG,eAAe,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IAChD,+DAA+D;IAC/D,gEAAgE;IAChE,6DAA6D;IAC7D,uCAAuC;IACvC,MAAM,gBAAgB,GACpB,IAAI,CAAC,QAAQ,CAAC,IAAI,KAAK,UAAU;WAC9B,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC;WAC1C,IAAI,CAAC,QAAQ,CAAC,aAAa,CAAC,MAAM,KAAK,CAAC,CAAC;IAC9C,MAAM,YAAY,GAAG,gBAAgB,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;IACjF,MAAM,SAAS,GAAG;QAChB,KAAK;QACL,GAAG,EAAE,YAAY;QACjB,QAAQ;QACR,SAAS;QACT,OAAO;QACP,GAAG,CAAC,MAAM,IAAI,EAAE,MAAM,EAAE,CAAC;QACzB,iEAAiE;QACjE,8EAA8E;QAC9E,yCAAyC;QACzC,GAAG,CAAC,IAAI,CAAC,QAAQ,CAAC,aAAa,KAAK,SAAS,IAAI,EAAE,aAAa,EAAE,IAAI,CAAC,QAAQ,CAAC,aAAa,EAAE,CAAC;QAChG,8DAA8D;QAC9D,mEAAmE;QACnE,0EAA0E;QAC1E,8CAA8C;QAC9C,8CAA8C;QAC9C,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,IAAI,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,MAAM,GAAG,CAAC,IAAI;YACzD,KAAK,EAAE,IAAI,CAAC,OAAO,CAAC,KAAK;YACzB,YAAY,EAAE,cAAc,IAAI,QAAQ,IAAI,YAAY,IAAI,SAAS;YACrE,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,WAAW,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,CAAC;SACvD,CAAC;KACH,CAAC;IAEF,QAAQ,IAAI,CAAC,QAAQ,CAAC,IAAI,EAAE,CAAC;QAC3B,KAAK,UAAU;YACb,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,EAAE;gBACf,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI;oBACZ,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,QAAQ;YACX,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,EAAE;gBACf,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI;oBACZ,MAAM,EAAE,iBAAiB,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC;iBAC9D;aACF,CAAC;QACJ,KAAK,OAAO;YACV,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,UAAU;YACb,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;QACJ,KAAK,OAAO,CAAC;QACb;YACE,OAAO;gBACL,QAAQ,EAAE,gCAAgC,CAAC,IAAI,CAAC,QAAQ,CAAC;gBACzD,WAAW,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO,IAAI,EAAE;gBACxC,KAAK,EAAE;oBACL,GAAG,SAAS;oBACZ,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,OAAO;oBAC7B,MAAM,EAAE,IAAI,CAAC,MAAM;iBACpB;aACF,CAAC;IACN,CAAC;AACH,CAAC"}
@@ -0,0 +1,203 @@
1
+ #!/usr/bin/env node
2
+ /**
3
+ * omk-mock PreToolUse hook (CLI fallback path).
4
+ *
5
+ * 由 src/eval-core/mocks-runtime.ts:materializeForCliConfigDir 写到临时
6
+ * CLAUDE_CONFIG_DIR/mock-hook.cjs。Claude Code 子进程会以 hook 形式调它,
7
+ * stdin 收 PreToolUseHookInput,stdout 出 PreToolUseHookOutput JSON。
8
+ *
9
+ * 该脚本是无状态的 — 所有规则从 OMK_MOCKS_FILE 文件读。命中状态(状态机用)
10
+ * 通过同目录下的 hits.json 持久化(同一 spawn 内多次 hook 调用共享)。
11
+ */
12
+ const fs = require('fs');
13
+ const path = require('path');
14
+ const os = require('os');
15
+
16
+ function expandHome(p) {
17
+ if (p.startsWith('~/')) return path.join(os.homedir(), p.slice(2));
18
+ if (p === '~') return os.homedir();
19
+ return p;
20
+ }
21
+
22
+ /** 与 mocks-runtime.ts 的 matchesFilePathSuffix 对齐 — 完全相等或路径分隔符后缀。 */
23
+ function matchesFilePathSuffix(actualRaw, suffixRaw) {
24
+ const actual = expandHome(actualRaw);
25
+ const suffix = expandHome(suffixRaw);
26
+ if (actual === suffix) return true;
27
+ if (!actual.endsWith(suffix)) return false;
28
+ const sep = actual.charAt(actual.length - suffix.length - 1);
29
+ return sep === '/' || sep === '\\';
30
+ }
31
+
32
+ function globMatch(pattern, value) {
33
+ // 与 SDK 路径(mocks-runtime.ts)对齐:dotAll 让 `*` 跨换行匹配多行 bash。
34
+ const escaped = pattern.replace(/[.+?^${}()|[\]\\]/g, '\\$&');
35
+ return new RegExp('^' + escaped.replace(/\*/g, '.*') + '$', 's').test(value);
36
+ }
37
+
38
+ function arraysDeepEqual(a, b) {
39
+ if (a.length !== b.length) return false;
40
+ for (let i = 0; i < a.length; i++) if (!deepEqual(a[i], b[i])) return false;
41
+ return true;
42
+ }
43
+ function deepEqual(a, b) {
44
+ if (a === b) return true;
45
+ if (Array.isArray(a) && Array.isArray(b)) return arraysDeepEqual(a, b);
46
+ if (typeof a === 'object' && a !== null && typeof b === 'object' && b !== null
47
+ && !Array.isArray(a) && !Array.isArray(b)) {
48
+ const ak = Object.keys(a), bk = Object.keys(b);
49
+ if (ak.length !== bk.length) return false;
50
+ for (const k of ak) if (!deepEqual(a[k], b[k])) return false;
51
+ return true;
52
+ }
53
+ return false;
54
+ }
55
+ function matchesInputSubset(expected, actual) {
56
+ if (typeof actual !== 'object' || actual === null) return false;
57
+ for (const k of Object.keys(expected)) {
58
+ const v = expected[k];
59
+ if (Array.isArray(v)) {
60
+ if (!Array.isArray(actual[k]) || !arraysDeepEqual(v, actual[k])) return false;
61
+ } else if (typeof v === 'object' && v !== null) {
62
+ if (!matchesInputSubset(v, actual[k])) return false;
63
+ } else if (actual[k] !== v) {
64
+ return false;
65
+ }
66
+ }
67
+ return true;
68
+ }
69
+
70
+ function isMockHit(mock, toolName, toolInput) {
71
+ if (mock.tool !== toolName) return false;
72
+ const m = mock.match;
73
+ if (!m) return true;
74
+ const ti = toolInput || {};
75
+ if (m.file_path !== undefined) {
76
+ if (typeof ti.file_path !== 'string') return false;
77
+ if (expandHome(ti.file_path) !== expandHome(m.file_path)) return false;
78
+ }
79
+ if (m.file_path_endswith !== undefined) {
80
+ if (typeof ti.file_path !== 'string') return false;
81
+ if (!matchesFilePathSuffix(ti.file_path, m.file_path_endswith)) return false;
82
+ }
83
+ if (m.url !== undefined && ti.url !== m.url) return false;
84
+ if (m.url_glob !== undefined) {
85
+ if (typeof ti.url !== 'string' || !globMatch(m.url_glob, ti.url)) return false;
86
+ }
87
+ if (m.command_glob !== undefined) {
88
+ if (typeof ti.command !== 'string' || !globMatch(m.command_glob, ti.command)) return false;
89
+ }
90
+ if (m.input !== undefined && !matchesInputSubset(m.input, toolInput)) return false;
91
+ return true;
92
+ }
93
+
94
+ function stringifyReturn(r) {
95
+ if (typeof r === 'string') return r;
96
+ return JSON.stringify(r);
97
+ }
98
+
99
+ function resolveMockReturn(mock, hitCount, baseDir) {
100
+ if (mock.return_seq && hitCount < mock.return_seq.length) {
101
+ return stringifyReturn(mock.return_seq[hitCount]);
102
+ }
103
+ if (mock.return !== undefined) return stringifyReturn(mock.return);
104
+ if (mock.return_file) {
105
+ const fpath = path.isAbsolute(mock.return_file)
106
+ ? mock.return_file
107
+ : path.resolve(baseDir || process.cwd(), mock.return_file);
108
+ if (!fs.existsSync(fpath)) return `[omk-mock-error] return_file not found: ${fpath}`;
109
+ return fs.readFileSync(fpath, 'utf8');
110
+ }
111
+ return '';
112
+ }
113
+
114
+ // ─── main ────────────────────────────────────────────────────────────────────
115
+
116
+ let stdin = '';
117
+ process.stdin.on('data', (chunk) => { stdin += chunk; });
118
+ process.stdin.on('end', () => {
119
+ let input;
120
+ try {
121
+ input = JSON.parse(stdin);
122
+ } catch (e) {
123
+ process.stdout.write(JSON.stringify({ continue: true })); // hook 协议错误时放行
124
+ process.exit(0);
125
+ }
126
+
127
+ const mocksFile = process.env.OMK_MOCKS_FILE;
128
+ if (!mocksFile || !fs.existsSync(mocksFile)) {
129
+ // 没配 mocks 就 noop
130
+ process.stdout.write(JSON.stringify({ continue: true }));
131
+ process.exit(0);
132
+ }
133
+
134
+ const config = JSON.parse(fs.readFileSync(mocksFile, 'utf8'));
135
+ const mocks = config.mocks || [];
136
+ const baseDir = config.baseDir;
137
+ const strict = !!config.strict;
138
+
139
+ // 统计文件:perMock(每条 mock 的命中次数,用于状态机 + 报告)+ hits_total + misses_total
140
+ // 同一 spawn 内多次 hook 调用共享。
141
+ const statsFile = path.join(path.dirname(mocksFile), 'hits.json');
142
+ const stats = fs.existsSync(statsFile)
143
+ ? JSON.parse(fs.readFileSync(statsFile, 'utf8'))
144
+ : { perMock: {}, hits_total: 0, misses_total: 0 };
145
+ if (!stats.perMock) stats.perMock = {};
146
+ if (!stats.hits_total) stats.hits_total = 0;
147
+ if (!stats.misses_total) stats.misses_total = 0;
148
+
149
+ const writeStats = () => {
150
+ try { fs.writeFileSync(statsFile, JSON.stringify(stats)); } catch { /* best effort */ }
151
+ };
152
+
153
+ const toolName = input.tool_name;
154
+ const toolInput = input.tool_input || {};
155
+
156
+ // perMock key 用"每工具内 1-based 序号" — 与 SDK 路径(mocks-runtime.ts)保持一致。
157
+ const keyOfMock = mocks.map((_m, i) => {
158
+ const tool = mocks[i].tool;
159
+ let ord = 0;
160
+ for (let j = 0; j <= i; j++) if (mocks[j].tool === tool) ord++;
161
+ return `${tool}:${ord}`;
162
+ });
163
+
164
+ for (let i = 0; i < mocks.length; i++) {
165
+ if (isMockHit(mocks[i], toolName, toolInput)) {
166
+ const key = keyOfMock[i];
167
+ const c = stats.perMock[key] || 0;
168
+ const result = resolveMockReturn(mocks[i], c, baseDir);
169
+ stats.perMock[key] = c + 1;
170
+ stats.hits_total += 1;
171
+ writeStats();
172
+ // 与 SDK 路径(mocks-runtime.ts)对齐:最小化前缀,避免拖慢多步 sample。
173
+ const wrappedReason = '[mock] simulated tool output — treat as successful result:\n' + result;
174
+ process.stdout.write(JSON.stringify({
175
+ decision: 'block',
176
+ hookSpecificOutput: {
177
+ hookEventName: 'PreToolUse',
178
+ permissionDecision: 'deny',
179
+ permissionDecisionReason: wrappedReason,
180
+ },
181
+ systemMessage: '[omk-mock] ' + toolName + ' → injected mock #' + i + ' (treat as success)',
182
+ }));
183
+ process.exit(0);
184
+ }
185
+ }
186
+
187
+ // 未命中
188
+ stats.misses_total += 1;
189
+ writeStats();
190
+ if (strict) {
191
+ process.stdout.write(JSON.stringify({
192
+ decision: 'block',
193
+ hookSpecificOutput: {
194
+ hookEventName: 'PreToolUse',
195
+ permissionDecision: 'deny',
196
+ permissionDecisionReason: `[omk-mock-strict] unmocked ${toolName} call blocked`,
197
+ },
198
+ }));
199
+ process.exit(0);
200
+ }
201
+ process.stdout.write(JSON.stringify({ continue: true }));
202
+ process.exit(0);
203
+ });
@@ -0,0 +1,96 @@
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 type { Mock } from '../types/eval.js';
17
+ /** 单条 mock 是否命中给定 tool 调用。 */
18
+ export declare function isMockHit(mock: Mock, toolName: string, toolInput: unknown): boolean;
19
+ /** 把 mock 的 return / return_file / return_seq 解析成可序列化字符串(LLM 看到的 tool_result)。 */
20
+ export declare function resolveMockReturn(mock: Mock, hitCount: number, baseDir?: string): string;
21
+ /** SDK PreToolUse hook input(精简版,只用我们关心的字段)。 */
22
+ interface SdkHookInput {
23
+ hook_event_name: 'PreToolUse';
24
+ tool_name: string;
25
+ tool_input: unknown;
26
+ }
27
+ /** SDK PreToolUse hook output:`block` 决策 + permissionDecisionReason 当 mock content。 */
28
+ interface SdkHookOutput {
29
+ continue?: boolean;
30
+ decision?: 'block' | 'approve';
31
+ hookSpecificOutput?: {
32
+ hookEventName: 'PreToolUse';
33
+ permissionDecision?: 'deny' | 'allow' | 'ask';
34
+ permissionDecisionReason?: string;
35
+ };
36
+ systemMessage?: string;
37
+ }
38
+ export interface SdkHookHandle {
39
+ /** 给 SDK options.hooks.PreToolUse 用的 callback。 */
40
+ callback: (input: SdkHookInput) => Promise<SdkHookOutput>;
41
+ /** 拦截命中统计(诊断用)。 */
42
+ stats: {
43
+ hits: number;
44
+ misses: number;
45
+ perMock: Record<string, number>;
46
+ };
47
+ }
48
+ /**
49
+ * 把 mocks 转成 SDK in-process PreToolUse hook callback。
50
+ *
51
+ * 命中策略:strict 模式下不命中的工具调用直接 deny(防意外真调)。
52
+ * 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。
53
+ *
54
+ * @param mocks Sample.mocks 数组
55
+ * @param baseDir 解析 mock.return_file 的相对路径锚点
56
+ * @param strict true → 不命中 deny;false → 不命中放行(default)
57
+ */
58
+ export declare function buildSdkHookCallback(mocks: Mock[] | undefined, baseDir?: string, strict?: boolean): SdkHookHandle;
59
+ export interface CliMockHandle {
60
+ /** 临时 settings 文件路径,作为 `claude --settings <path>` 参数传入。
61
+ * Claude Code 会**追加**这个 settings 到 ~/.claude/settings.json,**不替换** —
62
+ * 这样 OAuth 登录态 / 用户主配置都不动。 */
63
+ settingsFile: string;
64
+ /** 子进程要看到的额外 env(目前只 OMK_MOCKS_FILE)。 */
65
+ env: Record<string, string>;
66
+ /** 读 hook 在临时目录写的命中统计(给 ExecResult.mockStats 用)。
67
+ * 必须在 cleanup 之前调,cleanup 后文件就没了。 */
68
+ readStats: () => {
69
+ hits: number;
70
+ misses: number;
71
+ perMock: Record<string, number>;
72
+ };
73
+ /** 必须在 spawn 完成后(无论 success/error/timeout)调用。 */
74
+ cleanup: () => void;
75
+ }
76
+ /**
77
+ * 物化 mocks 到临时 settings 文件 + on-disk hook 脚本。
78
+ *
79
+ * **关键**:用 Claude Code 的 `--settings <file>` flag 而非 `CLAUDE_CONFIG_DIR`。
80
+ * `--settings` 是**追加**模式,主 `~/.claude/` 包含的 OAuth 登录态 / 用户配置全部保留;
81
+ * 而 CLAUDE_CONFIG_DIR 是**替换**模式,设它会丢 OAuth → "Not logged in"。
82
+ *
83
+ * 临时目录结构:
84
+ * $tmpdir/omk-mocks-XXXXXX/
85
+ * ├── settings.json (PreToolUse hook 注册,作为 --settings 参数)
86
+ * ├── mock-hook.cjs (hook 实现:读 OMK_MOCKS_FILE → 匹配 → 输出 JSON 决策)
87
+ * └── mocks.json (mocks 序列化 + baseDir + strict)
88
+ *
89
+ * cleanup 必删整个目录。executor 用 try / finally 保证执行。
90
+ */
91
+ export declare function materializeForCliConfigDir(mocks: Mock[] | undefined, baseDir?: string, strict?: boolean): CliMockHandle | null;
92
+ declare function readMockHookTemplate(): string;
93
+ export declare const _readMockHookTemplateForTest: typeof readMockHookTemplate;
94
+ export declare function _testMakeTempConfigDir(): string;
95
+ export {};
96
+ //# sourceMappingURL=mocks-runtime.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"mocks-runtime.d.ts","sourceRoot":"","sources":["../../../src/eval-core/mocks-runtime.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;GAcG;AAMH,OAAO,KAAK,EAAE,IAAI,EAAc,MAAM,kBAAkB,CAAC;AA6EzD,8BAA8B;AAC9B,wBAAgB,SAAS,CAAC,IAAI,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,GAAG,OAAO,CA2BnF;AAID,kFAAkF;AAClF,wBAAgB,iBAAiB,CAC/B,IAAI,EAAE,IAAI,EACV,QAAQ,EAAE,MAAM,EAChB,OAAO,CAAC,EAAE,MAAM,GACf,MAAM,CAkBR;AAWD,gDAAgD;AAChD,UAAU,YAAY;IACpB,eAAe,EAAE,YAAY,CAAC;IAC9B,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,OAAO,CAAC;CACrB;AAED,uFAAuF;AACvF,UAAU,aAAa;IACrB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,QAAQ,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC/B,kBAAkB,CAAC,EAAE;QACnB,aAAa,EAAE,YAAY,CAAC;QAC5B,kBAAkB,CAAC,EAAE,MAAM,GAAG,OAAO,GAAG,KAAK,CAAC;QAC9C,wBAAwB,CAAC,EAAE,MAAM,CAAC;KACnC,CAAC;IACF,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,MAAM,WAAW,aAAa;IAC5B,kDAAkD;IAClD,QAAQ,EAAE,CAAC,KAAK,EAAE,YAAY,KAAK,OAAO,CAAC,aAAa,CAAC,CAAC;IAC1D,mBAAmB;IACnB,KAAK,EAAE;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,CAAC;CAC1E;AAED;;;;;;;;;GASG;AACH,wBAAgB,oBAAoB,CAClC,KAAK,EAAE,IAAI,EAAE,GAAG,SAAS,EACzB,OAAO,CAAC,EAAE,MAAM,EAChB,MAAM,UAAQ,GACb,aAAa,CAqDf;AAID,MAAM,WAAW,aAAa;IAC5B;;mCAE+B;IAC/B,YAAY,EAAE,MAAM,CAAC;IACrB,yCAAyC;IACzC,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC5B;0CACsC;IACtC,SAAS,EAAE,MAAM;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;KAAE,CAAC;IACnF,iDAAiD;IACjD,OAAO,EAAE,MAAM,IAAI,CAAC;CACrB;AAED;;;;;;;;;;;;;;GAcG;AACH,wBAAgB,0BAA0B,CACxC,KAAK,EAAE,IAAI,EAAE,GAAG,SAAS,EACzB,OAAO,CAAC,EAAE,MAAM,EAChB,MAAM,UAAQ,GACb,aAAa,GAAG,IAAI,CAwDtB;AAGD,iBAAS,oBAAoB,IAAI,MAAM,CAatC;AAGD,eAAO,MAAM,4BAA4B,6BAAuB,CAAC;AAIjE,wBAAgB,sBAAsB,IAAI,MAAM,CAI/C"}