claude-test-bench 1.0.0

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 (314) hide show
  1. package/LICENSE +201 -0
  2. package/README.md +474 -0
  3. package/dist/bin/ctb.d.ts +3 -0
  4. package/dist/bin/ctb.d.ts.map +1 -0
  5. package/dist/bin/ctb.js +107 -0
  6. package/dist/bin/ctb.js.map +1 -0
  7. package/dist/server/index.d.ts +13 -0
  8. package/dist/server/index.d.ts.map +1 -0
  9. package/dist/server/index.js +72 -0
  10. package/dist/server/index.js.map +1 -0
  11. package/dist/server/interfaces/evaluator.d.ts +15 -0
  12. package/dist/server/interfaces/evaluator.d.ts.map +1 -0
  13. package/dist/server/interfaces/evaluator.js +2 -0
  14. package/dist/server/interfaces/evaluator.js.map +1 -0
  15. package/dist/server/interfaces/logger.d.ts +9 -0
  16. package/dist/server/interfaces/logger.d.ts.map +1 -0
  17. package/dist/server/interfaces/logger.js +2 -0
  18. package/dist/server/interfaces/logger.js.map +1 -0
  19. package/dist/server/interfaces/runner.d.ts +9 -0
  20. package/dist/server/interfaces/runner.d.ts.map +1 -0
  21. package/dist/server/interfaces/runner.js +2 -0
  22. package/dist/server/interfaces/runner.js.map +1 -0
  23. package/dist/server/interfaces/storage.d.ts +36 -0
  24. package/dist/server/interfaces/storage.d.ts.map +1 -0
  25. package/dist/server/interfaces/storage.js +2 -0
  26. package/dist/server/interfaces/storage.js.map +1 -0
  27. package/dist/server/interfaces/workspace.d.ts +9 -0
  28. package/dist/server/interfaces/workspace.d.ts.map +1 -0
  29. package/dist/server/interfaces/workspace.js +2 -0
  30. package/dist/server/interfaces/workspace.js.map +1 -0
  31. package/dist/server/routes/eval-queue.d.ts +23 -0
  32. package/dist/server/routes/eval-queue.d.ts.map +1 -0
  33. package/dist/server/routes/eval-queue.js +45 -0
  34. package/dist/server/routes/eval-queue.js.map +1 -0
  35. package/dist/server/routes/evaluations.d.ts +8 -0
  36. package/dist/server/routes/evaluations.d.ts.map +1 -0
  37. package/dist/server/routes/evaluations.js +217 -0
  38. package/dist/server/routes/evaluations.js.map +1 -0
  39. package/dist/server/routes/providers.d.ts +5 -0
  40. package/dist/server/routes/providers.d.ts.map +1 -0
  41. package/dist/server/routes/providers.js +194 -0
  42. package/dist/server/routes/providers.js.map +1 -0
  43. package/dist/server/routes/run-queue.d.ts +17 -0
  44. package/dist/server/routes/run-queue.d.ts.map +1 -0
  45. package/dist/server/routes/run-queue.js +34 -0
  46. package/dist/server/routes/run-queue.js.map +1 -0
  47. package/dist/server/routes/run-sse.d.ts +18 -0
  48. package/dist/server/routes/run-sse.d.ts.map +1 -0
  49. package/dist/server/routes/run-sse.js +57 -0
  50. package/dist/server/routes/run-sse.js.map +1 -0
  51. package/dist/server/routes/runs.d.ts +9 -0
  52. package/dist/server/routes/runs.d.ts.map +1 -0
  53. package/dist/server/routes/runs.js +379 -0
  54. package/dist/server/routes/runs.js.map +1 -0
  55. package/dist/server/routes/scenarios.d.ts +5 -0
  56. package/dist/server/routes/scenarios.d.ts.map +1 -0
  57. package/dist/server/routes/scenarios.js +209 -0
  58. package/dist/server/routes/scenarios.js.map +1 -0
  59. package/dist/server/routes/setups.d.ts +5 -0
  60. package/dist/server/routes/setups.d.ts.map +1 -0
  61. package/dist/server/routes/setups.js +194 -0
  62. package/dist/server/routes/setups.js.map +1 -0
  63. package/dist/server/services/agent-mapper.d.ts +12 -0
  64. package/dist/server/services/agent-mapper.d.ts.map +1 -0
  65. package/dist/server/services/agent-mapper.js +75 -0
  66. package/dist/server/services/agent-mapper.js.map +1 -0
  67. package/dist/server/services/env-builder.d.ts +10 -0
  68. package/dist/server/services/env-builder.d.ts.map +1 -0
  69. package/dist/server/services/env-builder.js +50 -0
  70. package/dist/server/services/env-builder.js.map +1 -0
  71. package/dist/server/services/eval-helpers.d.ts +22 -0
  72. package/dist/server/services/eval-helpers.d.ts.map +1 -0
  73. package/dist/server/services/eval-helpers.js +75 -0
  74. package/dist/server/services/eval-helpers.js.map +1 -0
  75. package/dist/server/services/eval-parsers-debate-impl.d.ts +11 -0
  76. package/dist/server/services/eval-parsers-debate-impl.d.ts.map +1 -0
  77. package/dist/server/services/eval-parsers-debate-impl.js +133 -0
  78. package/dist/server/services/eval-parsers-debate-impl.js.map +1 -0
  79. package/dist/server/services/eval-parsers.d.ts +24 -0
  80. package/dist/server/services/eval-parsers.d.ts.map +1 -0
  81. package/dist/server/services/eval-parsers.js +153 -0
  82. package/dist/server/services/eval-parsers.js.map +1 -0
  83. package/dist/server/services/eval-prompts.d.ts +9 -0
  84. package/dist/server/services/eval-prompts.d.ts.map +1 -0
  85. package/dist/server/services/eval-prompts.js +170 -0
  86. package/dist/server/services/eval-prompts.js.map +1 -0
  87. package/dist/server/services/evaluator.d.ts +10 -0
  88. package/dist/server/services/evaluator.d.ts.map +1 -0
  89. package/dist/server/services/evaluator.js +156 -0
  90. package/dist/server/services/evaluator.js.map +1 -0
  91. package/dist/server/services/fs-adapter.d.ts +20 -0
  92. package/dist/server/services/fs-adapter.d.ts.map +1 -0
  93. package/dist/server/services/fs-adapter.js +13 -0
  94. package/dist/server/services/fs-adapter.js.map +1 -0
  95. package/dist/server/services/instruction-parser.d.ts +26 -0
  96. package/dist/server/services/instruction-parser.d.ts.map +1 -0
  97. package/dist/server/services/instruction-parser.js +121 -0
  98. package/dist/server/services/instruction-parser.js.map +1 -0
  99. package/dist/server/services/log-rotator.d.ts +20 -0
  100. package/dist/server/services/log-rotator.d.ts.map +1 -0
  101. package/dist/server/services/log-rotator.js +60 -0
  102. package/dist/server/services/log-rotator.js.map +1 -0
  103. package/dist/server/services/logger.d.ts +15 -0
  104. package/dist/server/services/logger.d.ts.map +1 -0
  105. package/dist/server/services/logger.js +69 -0
  106. package/dist/server/services/logger.js.map +1 -0
  107. package/dist/server/services/runner.d.ts +12 -0
  108. package/dist/server/services/runner.d.ts.map +1 -0
  109. package/dist/server/services/runner.js +161 -0
  110. package/dist/server/services/runner.js.map +1 -0
  111. package/dist/server/services/seeder.d.ts +5 -0
  112. package/dist/server/services/seeder.d.ts.map +1 -0
  113. package/dist/server/services/seeder.js +79 -0
  114. package/dist/server/services/seeder.js.map +1 -0
  115. package/dist/server/services/storage-test-helpers.d.ts +21 -0
  116. package/dist/server/services/storage-test-helpers.d.ts.map +1 -0
  117. package/dist/server/services/storage-test-helpers.js +158 -0
  118. package/dist/server/services/storage-test-helpers.js.map +1 -0
  119. package/dist/server/services/storage.d.ts +35 -0
  120. package/dist/server/services/storage.d.ts.map +1 -0
  121. package/dist/server/services/storage.js +219 -0
  122. package/dist/server/services/storage.js.map +1 -0
  123. package/dist/server/services/transcript-formatter.d.ts +18 -0
  124. package/dist/server/services/transcript-formatter.d.ts.map +1 -0
  125. package/dist/server/services/transcript-formatter.js +156 -0
  126. package/dist/server/services/transcript-formatter.js.map +1 -0
  127. package/dist/server/services/workspace.d.ts +11 -0
  128. package/dist/server/services/workspace.d.ts.map +1 -0
  129. package/dist/server/services/workspace.js +113 -0
  130. package/dist/server/services/workspace.js.map +1 -0
  131. package/dist/server/types/evaluation.d.ts +108 -0
  132. package/dist/server/types/evaluation.d.ts.map +1 -0
  133. package/dist/server/types/evaluation.js +5 -0
  134. package/dist/server/types/evaluation.js.map +1 -0
  135. package/dist/server/types/index.d.ts +5 -0
  136. package/dist/server/types/index.d.ts.map +1 -0
  137. package/dist/server/types/index.js +5 -0
  138. package/dist/server/types/index.js.map +1 -0
  139. package/dist/server/types/provider.d.ts +99 -0
  140. package/dist/server/types/provider.d.ts.map +1 -0
  141. package/dist/server/types/provider.js +5 -0
  142. package/dist/server/types/provider.js.map +1 -0
  143. package/dist/server/types/run.d.ts +31 -0
  144. package/dist/server/types/run.d.ts.map +1 -0
  145. package/dist/server/types/run.js +5 -0
  146. package/dist/server/types/run.js.map +1 -0
  147. package/dist/server/types/scenario.d.ts +32 -0
  148. package/dist/server/types/scenario.d.ts.map +1 -0
  149. package/dist/server/types/scenario.js +5 -0
  150. package/dist/server/types/scenario.js.map +1 -0
  151. package/dist/server/types/setup.d.ts +99 -0
  152. package/dist/server/types/setup.d.ts.map +1 -0
  153. package/dist/server/types/setup.js +5 -0
  154. package/dist/server/types/setup.js.map +1 -0
  155. package/dist/src/server/index.d.ts +13 -0
  156. package/dist/src/server/index.d.ts.map +1 -0
  157. package/dist/src/server/index.js +72 -0
  158. package/dist/src/server/index.js.map +1 -0
  159. package/dist/src/server/interfaces/evaluator.d.ts +15 -0
  160. package/dist/src/server/interfaces/evaluator.d.ts.map +1 -0
  161. package/dist/src/server/interfaces/evaluator.js +2 -0
  162. package/dist/src/server/interfaces/evaluator.js.map +1 -0
  163. package/dist/src/server/interfaces/logger.d.ts +9 -0
  164. package/dist/src/server/interfaces/logger.d.ts.map +1 -0
  165. package/dist/src/server/interfaces/logger.js +2 -0
  166. package/dist/src/server/interfaces/logger.js.map +1 -0
  167. package/dist/src/server/interfaces/runner.d.ts +9 -0
  168. package/dist/src/server/interfaces/runner.d.ts.map +1 -0
  169. package/dist/src/server/interfaces/runner.js +2 -0
  170. package/dist/src/server/interfaces/runner.js.map +1 -0
  171. package/dist/src/server/interfaces/storage.d.ts +36 -0
  172. package/dist/src/server/interfaces/storage.d.ts.map +1 -0
  173. package/dist/src/server/interfaces/storage.js +2 -0
  174. package/dist/src/server/interfaces/storage.js.map +1 -0
  175. package/dist/src/server/interfaces/workspace.d.ts +9 -0
  176. package/dist/src/server/interfaces/workspace.d.ts.map +1 -0
  177. package/dist/src/server/interfaces/workspace.js +2 -0
  178. package/dist/src/server/interfaces/workspace.js.map +1 -0
  179. package/dist/src/server/routes/eval-queue.d.ts +23 -0
  180. package/dist/src/server/routes/eval-queue.d.ts.map +1 -0
  181. package/dist/src/server/routes/eval-queue.js +45 -0
  182. package/dist/src/server/routes/eval-queue.js.map +1 -0
  183. package/dist/src/server/routes/evaluations.d.ts +8 -0
  184. package/dist/src/server/routes/evaluations.d.ts.map +1 -0
  185. package/dist/src/server/routes/evaluations.js +217 -0
  186. package/dist/src/server/routes/evaluations.js.map +1 -0
  187. package/dist/src/server/routes/providers.d.ts +5 -0
  188. package/dist/src/server/routes/providers.d.ts.map +1 -0
  189. package/dist/src/server/routes/providers.js +194 -0
  190. package/dist/src/server/routes/providers.js.map +1 -0
  191. package/dist/src/server/routes/run-queue.d.ts +17 -0
  192. package/dist/src/server/routes/run-queue.d.ts.map +1 -0
  193. package/dist/src/server/routes/run-queue.js +34 -0
  194. package/dist/src/server/routes/run-queue.js.map +1 -0
  195. package/dist/src/server/routes/run-sse.d.ts +18 -0
  196. package/dist/src/server/routes/run-sse.d.ts.map +1 -0
  197. package/dist/src/server/routes/run-sse.js +57 -0
  198. package/dist/src/server/routes/run-sse.js.map +1 -0
  199. package/dist/src/server/routes/runs.d.ts +9 -0
  200. package/dist/src/server/routes/runs.d.ts.map +1 -0
  201. package/dist/src/server/routes/runs.js +379 -0
  202. package/dist/src/server/routes/runs.js.map +1 -0
  203. package/dist/src/server/routes/scenarios.d.ts +5 -0
  204. package/dist/src/server/routes/scenarios.d.ts.map +1 -0
  205. package/dist/src/server/routes/scenarios.js +209 -0
  206. package/dist/src/server/routes/scenarios.js.map +1 -0
  207. package/dist/src/server/routes/setups.d.ts +5 -0
  208. package/dist/src/server/routes/setups.d.ts.map +1 -0
  209. package/dist/src/server/routes/setups.js +194 -0
  210. package/dist/src/server/routes/setups.js.map +1 -0
  211. package/dist/src/server/services/agent-mapper.d.ts +12 -0
  212. package/dist/src/server/services/agent-mapper.d.ts.map +1 -0
  213. package/dist/src/server/services/agent-mapper.js +75 -0
  214. package/dist/src/server/services/agent-mapper.js.map +1 -0
  215. package/dist/src/server/services/env-builder.d.ts +10 -0
  216. package/dist/src/server/services/env-builder.d.ts.map +1 -0
  217. package/dist/src/server/services/env-builder.js +50 -0
  218. package/dist/src/server/services/env-builder.js.map +1 -0
  219. package/dist/src/server/services/eval-helpers.d.ts +22 -0
  220. package/dist/src/server/services/eval-helpers.d.ts.map +1 -0
  221. package/dist/src/server/services/eval-helpers.js +75 -0
  222. package/dist/src/server/services/eval-helpers.js.map +1 -0
  223. package/dist/src/server/services/eval-parsers-debate-impl.d.ts +11 -0
  224. package/dist/src/server/services/eval-parsers-debate-impl.d.ts.map +1 -0
  225. package/dist/src/server/services/eval-parsers-debate-impl.js +133 -0
  226. package/dist/src/server/services/eval-parsers-debate-impl.js.map +1 -0
  227. package/dist/src/server/services/eval-parsers.d.ts +24 -0
  228. package/dist/src/server/services/eval-parsers.d.ts.map +1 -0
  229. package/dist/src/server/services/eval-parsers.js +153 -0
  230. package/dist/src/server/services/eval-parsers.js.map +1 -0
  231. package/dist/src/server/services/eval-prompts.d.ts +9 -0
  232. package/dist/src/server/services/eval-prompts.d.ts.map +1 -0
  233. package/dist/src/server/services/eval-prompts.js +170 -0
  234. package/dist/src/server/services/eval-prompts.js.map +1 -0
  235. package/dist/src/server/services/evaluator.d.ts +10 -0
  236. package/dist/src/server/services/evaluator.d.ts.map +1 -0
  237. package/dist/src/server/services/evaluator.js +156 -0
  238. package/dist/src/server/services/evaluator.js.map +1 -0
  239. package/dist/src/server/services/fs-adapter.d.ts +20 -0
  240. package/dist/src/server/services/fs-adapter.d.ts.map +1 -0
  241. package/dist/src/server/services/fs-adapter.js +13 -0
  242. package/dist/src/server/services/fs-adapter.js.map +1 -0
  243. package/dist/src/server/services/instruction-parser.d.ts +26 -0
  244. package/dist/src/server/services/instruction-parser.d.ts.map +1 -0
  245. package/dist/src/server/services/instruction-parser.js +121 -0
  246. package/dist/src/server/services/instruction-parser.js.map +1 -0
  247. package/dist/src/server/services/log-rotator.d.ts +20 -0
  248. package/dist/src/server/services/log-rotator.d.ts.map +1 -0
  249. package/dist/src/server/services/log-rotator.js +60 -0
  250. package/dist/src/server/services/log-rotator.js.map +1 -0
  251. package/dist/src/server/services/logger.d.ts +15 -0
  252. package/dist/src/server/services/logger.d.ts.map +1 -0
  253. package/dist/src/server/services/logger.js +69 -0
  254. package/dist/src/server/services/logger.js.map +1 -0
  255. package/dist/src/server/services/runner.d.ts +12 -0
  256. package/dist/src/server/services/runner.d.ts.map +1 -0
  257. package/dist/src/server/services/runner.js +161 -0
  258. package/dist/src/server/services/runner.js.map +1 -0
  259. package/dist/src/server/services/seeder.d.ts +5 -0
  260. package/dist/src/server/services/seeder.d.ts.map +1 -0
  261. package/dist/src/server/services/seeder.js +79 -0
  262. package/dist/src/server/services/seeder.js.map +1 -0
  263. package/dist/src/server/services/storage.d.ts +35 -0
  264. package/dist/src/server/services/storage.d.ts.map +1 -0
  265. package/dist/src/server/services/storage.js +219 -0
  266. package/dist/src/server/services/storage.js.map +1 -0
  267. package/dist/src/server/services/transcript-formatter.d.ts +18 -0
  268. package/dist/src/server/services/transcript-formatter.d.ts.map +1 -0
  269. package/dist/src/server/services/transcript-formatter.js +156 -0
  270. package/dist/src/server/services/transcript-formatter.js.map +1 -0
  271. package/dist/src/server/services/workspace.d.ts +11 -0
  272. package/dist/src/server/services/workspace.d.ts.map +1 -0
  273. package/dist/src/server/services/workspace.js +113 -0
  274. package/dist/src/server/services/workspace.js.map +1 -0
  275. package/dist/src/server/types/evaluation.d.ts +108 -0
  276. package/dist/src/server/types/evaluation.d.ts.map +1 -0
  277. package/dist/src/server/types/evaluation.js +5 -0
  278. package/dist/src/server/types/evaluation.js.map +1 -0
  279. package/dist/src/server/types/index.d.ts +5 -0
  280. package/dist/src/server/types/index.d.ts.map +1 -0
  281. package/dist/src/server/types/index.js +5 -0
  282. package/dist/src/server/types/index.js.map +1 -0
  283. package/dist/src/server/types/provider.d.ts +99 -0
  284. package/dist/src/server/types/provider.d.ts.map +1 -0
  285. package/dist/src/server/types/provider.js +5 -0
  286. package/dist/src/server/types/provider.js.map +1 -0
  287. package/dist/src/server/types/run.d.ts +31 -0
  288. package/dist/src/server/types/run.d.ts.map +1 -0
  289. package/dist/src/server/types/run.js +5 -0
  290. package/dist/src/server/types/run.js.map +1 -0
  291. package/dist/src/server/types/scenario.d.ts +32 -0
  292. package/dist/src/server/types/scenario.d.ts.map +1 -0
  293. package/dist/src/server/types/scenario.js +5 -0
  294. package/dist/src/server/types/scenario.js.map +1 -0
  295. package/dist/src/server/types/setup.d.ts +99 -0
  296. package/dist/src/server/types/setup.d.ts.map +1 -0
  297. package/dist/src/server/types/setup.js +5 -0
  298. package/dist/src/server/types/setup.js.map +1 -0
  299. package/dist/web/assets/index-C4dw8OpW.css +1 -0
  300. package/dist/web/assets/index-wve8IczO.js +76 -0
  301. package/dist/web/index.html +15 -0
  302. package/docs/schemas/provider-api.example.json +16 -0
  303. package/docs/schemas/provider-oauth.example.json +15 -0
  304. package/docs/schemas/provider.example.json +16 -0
  305. package/docs/schemas/scenario-baseline.example.json +35 -0
  306. package/docs/schemas/scenario-carwash-baseline.example.json +33 -0
  307. package/docs/schemas/scenario-carwash-with-claude-md.example.json +40 -0
  308. package/docs/schemas/scenario-golden-rules-baseline.example.json +51 -0
  309. package/docs/schemas/scenario-golden-rules-with-claude-md.example.json +61 -0
  310. package/docs/schemas/scenario-negative-analysis-baseline.example.json +34 -0
  311. package/docs/schemas/scenario-negative-analysis-with-claude-md.example.json +41 -0
  312. package/docs/schemas/scenario-with-claude-md.example.json +41 -0
  313. package/docs/schemas/scenario.example.json +33 -0
  314. package/package.json +92 -0
@@ -0,0 +1,21 @@
1
+ import type { FsAdapter } from './fs-adapter.js';
2
+ import type { Provider, Scenario, Run, Evaluation } from '../types/index.js';
3
+ export declare const BASE_PROVIDER: {
4
+ kind: "api";
5
+ baseUrl: string;
6
+ apiKey: string;
7
+ model: string;
8
+ };
9
+ export declare function makeProvider(overrides?: Partial<Provider>): Provider;
10
+ export declare function makeScenario(overrides?: Partial<Scenario>): Scenario;
11
+ export declare function makeRun(overrides?: Partial<Run>): Run;
12
+ export declare function makeEvaluation(overrides?: Partial<Evaluation>): Evaluation;
13
+ export interface MockFsState {
14
+ files: Map<string, string>;
15
+ dirs: Set<string>;
16
+ }
17
+ export declare function createMockFs(): {
18
+ fs: FsAdapter;
19
+ state: MockFsState;
20
+ };
21
+ //# sourceMappingURL=storage-test-helpers.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"storage-test-helpers.d.ts","sourceRoot":"","sources":["../../../src/server/services/storage-test-helpers.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AACjD,OAAO,KAAK,EACV,QAAQ,EACR,QAAQ,EACR,GAAG,EACH,UAAU,EACX,MAAM,mBAAmB,CAAC;AAI3B,eAAO,MAAM,aAAa;;;;;CAKzB,CAAC;AAEF,wBAAgB,YAAY,CAAC,SAAS,GAAE,OAAO,CAAC,QAAQ,CAAM,GAAG,QAAQ,CAWxE;AAED,wBAAgB,YAAY,CAAC,SAAS,GAAE,OAAO,CAAC,QAAQ,CAAM,GAAG,QAAQ,CAqBxE;AAED,wBAAgB,OAAO,CAAC,SAAS,GAAE,OAAO,CAAC,GAAG,CAAM,GAAG,GAAG,CAiBzD;AAED,wBAAgB,cAAc,CAAC,SAAS,GAAE,OAAO,CAAC,UAAU,CAAM,GAAG,UAAU,CA+B9E;AAID,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAC3B,IAAI,EAAE,GAAG,CAAC,MAAM,CAAC,CAAC;CACnB;AAED,wBAAgB,YAAY,IAAI;IAAE,EAAE,EAAE,SAAS,CAAC;IAAC,KAAK,EAAE,WAAW,CAAA;CAAE,CAyEpE"}
@@ -0,0 +1,158 @@
1
+ import { vi } from 'vitest';
2
+ // ─── Factory helpers ─────────────────────────────────────────────────
3
+ export const BASE_PROVIDER = {
4
+ kind: 'api',
5
+ baseUrl: 'https://api.anthropic.com',
6
+ apiKey: 'sk-test',
7
+ model: 'claude-sonnet-4-6',
8
+ };
9
+ export function makeProvider(overrides = {}) {
10
+ return {
11
+ id: 'provider-1',
12
+ name: 'Test Provider',
13
+ description: 'A test provider',
14
+ provider: BASE_PROVIDER,
15
+ timeoutSeconds: 300,
16
+ createdAt: '2026-01-01T00:00:00.000Z',
17
+ updatedAt: '2026-01-01T00:00:00.000Z',
18
+ ...overrides,
19
+ };
20
+ }
21
+ export function makeScenario(overrides = {}) {
22
+ return {
23
+ id: 'scenario-1',
24
+ name: 'Test Scenario',
25
+ category: 'planning',
26
+ claudeMdFiles: [],
27
+ rules: [],
28
+ skills: [],
29
+ subagents: [],
30
+ mcpServers: [],
31
+ permissionMode: 'default',
32
+ prompt: 'Do something',
33
+ workspaceFiles: [],
34
+ expectedAnswer: 'done',
35
+ criticalRequirements: [],
36
+ gradingGuidelines: '',
37
+ scoringDimensions: [],
38
+ createdAt: '2026-01-01T00:00:00.000Z',
39
+ updatedAt: '2026-01-01T00:00:00.000Z',
40
+ ...overrides,
41
+ };
42
+ }
43
+ export function makeRun(overrides = {}) {
44
+ return {
45
+ id: 'run-1',
46
+ providerId: 'provider-1',
47
+ scenarioId: 'scenario-1',
48
+ status: 'completed',
49
+ providerSnapshot: makeProvider(),
50
+ scenarioSnapshot: makeScenario(),
51
+ messages: [],
52
+ resultText: 'done',
53
+ totalCostUsd: 0.01,
54
+ durationMs: 1000,
55
+ numTurns: 3,
56
+ createdAt: '2026-01-01T00:00:00.000Z',
57
+ updatedAt: '2026-01-01T00:00:01.000Z',
58
+ ...overrides,
59
+ };
60
+ }
61
+ export function makeEvaluation(overrides = {}) {
62
+ return {
63
+ id: 'eval-1',
64
+ runId: 'run-1',
65
+ status: 'completed',
66
+ evaluators: [],
67
+ rounds: [],
68
+ answerComparison: { matches: true, explanation: 'Good', similarity: 0.9 },
69
+ criticalResults: [],
70
+ setupCompliance: {
71
+ instructionCompliance: {
72
+ followed: [],
73
+ violated: [],
74
+ notApplicable: [],
75
+ overallCompliance: 1,
76
+ },
77
+ skillUsage: [],
78
+ subagentUsage: [],
79
+ },
80
+ synthesis: {
81
+ dimensionScores: {},
82
+ weightedTotal: 8,
83
+ confidence: 0.9,
84
+ dissenting: [],
85
+ },
86
+ ledger: [],
87
+ totalCostUsd: 0.05,
88
+ createdAt: '2026-01-01T00:00:00.000Z',
89
+ updatedAt: '2026-01-01T00:00:02.000Z',
90
+ ...overrides,
91
+ };
92
+ }
93
+ export function createMockFs() {
94
+ const state = {
95
+ files: new Map(),
96
+ dirs: new Set(),
97
+ };
98
+ const mockFs = {
99
+ mkdir: vi.fn((_dirPath) => {
100
+ state.dirs.add(_dirPath);
101
+ return Promise.resolve();
102
+ }),
103
+ writeFile: vi.fn((_filePath, data) => {
104
+ state.files.set(_filePath, data);
105
+ return Promise.resolve();
106
+ }),
107
+ readFile: vi.fn((filePath) => {
108
+ const data = state.files.get(filePath);
109
+ if (data === undefined) {
110
+ return Promise.reject(new Error(`ENOENT: no such file: ${filePath}`));
111
+ }
112
+ return Promise.resolve(data);
113
+ }),
114
+ readdir: vi.fn((dirPath) => {
115
+ const prefix = dirPath.endsWith('/') ? dirPath : dirPath + '/';
116
+ const entries = [];
117
+ for (const key of state.files.keys()) {
118
+ if (key.startsWith(prefix)) {
119
+ const rest = key.slice(prefix.length);
120
+ if (!rest.includes('/')) {
121
+ entries.push(rest);
122
+ }
123
+ }
124
+ }
125
+ return Promise.resolve(entries);
126
+ }),
127
+ unlink: vi.fn((filePath) => {
128
+ if (!state.files.has(filePath)) {
129
+ return Promise.reject(new Error(`ENOENT: no such file: ${filePath}`));
130
+ }
131
+ state.files.delete(filePath);
132
+ return Promise.resolve();
133
+ }),
134
+ rename: vi.fn((oldPath, newPath) => {
135
+ const data = state.files.get(oldPath);
136
+ if (data === undefined) {
137
+ return Promise.reject(new Error(`ENOENT: no such file: ${oldPath}`));
138
+ }
139
+ state.files.set(newPath, data);
140
+ state.files.delete(oldPath);
141
+ return Promise.resolve();
142
+ }),
143
+ stat: vi.fn((filePath) => {
144
+ if (!state.files.has(filePath)) {
145
+ return Promise.reject(new Error(`ENOENT: no such file: ${filePath}`));
146
+ }
147
+ return Promise.resolve({ isFile: () => true });
148
+ }),
149
+ access: vi.fn((filePath) => {
150
+ if (!state.files.has(filePath)) {
151
+ return Promise.reject(new Error(`ENOENT: no such file: ${filePath}`));
152
+ }
153
+ return Promise.resolve();
154
+ }),
155
+ };
156
+ return { fs: mockFs, state };
157
+ }
158
+ //# sourceMappingURL=storage-test-helpers.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"storage-test-helpers.js","sourceRoot":"","sources":["../../../src/server/services/storage-test-helpers.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAC;AAS5B,wEAAwE;AAExE,MAAM,CAAC,MAAM,aAAa,GAAG;IAC3B,IAAI,EAAE,KAAc;IACpB,OAAO,EAAE,2BAA2B;IACpC,MAAM,EAAE,SAAS;IACjB,KAAK,EAAE,mBAAmB;CAC3B,CAAC;AAEF,MAAM,UAAU,YAAY,CAAC,YAA+B,EAAE;IAC5D,OAAO;QACL,EAAE,EAAE,YAAY;QAChB,IAAI,EAAE,eAAe;QACrB,WAAW,EAAE,iBAAiB;QAC9B,QAAQ,EAAE,aAAa;QACvB,cAAc,EAAE,GAAG;QACnB,SAAS,EAAE,0BAA0B;QACrC,SAAS,EAAE,0BAA0B;QACrC,GAAG,SAAS;KACb,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,YAA+B,EAAE;IAC5D,OAAO;QACL,EAAE,EAAE,YAAY;QAChB,IAAI,EAAE,eAAe;QACrB,QAAQ,EAAE,UAAU;QACpB,aAAa,EAAE,EAAE;QACjB,KAAK,EAAE,EAAE;QACT,MAAM,EAAE,EAAE;QACV,SAAS,EAAE,EAAE;QACb,UAAU,EAAE,EAAE;QACd,cAAc,EAAE,SAAS;QACzB,MAAM,EAAE,cAAc;QACtB,cAAc,EAAE,EAAE;QAClB,cAAc,EAAE,MAAM;QACtB,oBAAoB,EAAE,EAAE;QACxB,iBAAiB,EAAE,EAAE;QACrB,iBAAiB,EAAE,EAAE;QACrB,SAAS,EAAE,0BAA0B;QACrC,SAAS,EAAE,0BAA0B;QACrC,GAAG,SAAS;KACb,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,OAAO,CAAC,YAA0B,EAAE;IAClD,OAAO;QACL,EAAE,EAAE,OAAO;QACX,UAAU,EAAE,YAAY;QACxB,UAAU,EAAE,YAAY;QACxB,MAAM,EAAE,WAAW;QACnB,gBAAgB,EAAE,YAAY,EAAE;QAChC,gBAAgB,EAAE,YAAY,EAAE;QAChC,QAAQ,EAAE,EAAE;QACZ,UAAU,EAAE,MAAM;QAClB,YAAY,EAAE,IAAI;QAClB,UAAU,EAAE,IAAI;QAChB,QAAQ,EAAE,CAAC;QACX,SAAS,EAAE,0BAA0B;QACrC,SAAS,EAAE,0BAA0B;QACrC,GAAG,SAAS;KACb,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,YAAiC,EAAE;IAChE,OAAO;QACL,EAAE,EAAE,QAAQ;QACZ,KAAK,EAAE,OAAO;QACd,MAAM,EAAE,WAAW;QACnB,UAAU,EAAE,EAAE;QACd,MAAM,EAAE,EAAE;QACV,gBAAgB,EAAE,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,MAAM,EAAE,UAAU,EAAE,GAAG,EAAE;QACzE,eAAe,EAAE,EAAE;QACnB,eAAe,EAAE;YACf,qBAAqB,EAAE;gBACrB,QAAQ,EAAE,EAAE;gBACZ,QAAQ,EAAE,EAAE;gBACZ,aAAa,EAAE,EAAE;gBACjB,iBAAiB,EAAE,CAAC;aACrB;YACD,UAAU,EAAE,EAAE;YACd,aAAa,EAAE,EAAE;SAClB;QACD,SAAS,EAAE;YACT,eAAe,EAAE,EAAE;YACnB,aAAa,EAAE,CAAC;YAChB,UAAU,EAAE,GAAG;YACf,UAAU,EAAE,EAAE;SACf;QACD,MAAM,EAAE,EAAE;QACV,YAAY,EAAE,IAAI;QAClB,SAAS,EAAE,0BAA0B;QACrC,SAAS,EAAE,0BAA0B;QACrC,GAAG,SAAS;KACb,CAAC;AACJ,CAAC;AASD,MAAM,UAAU,YAAY;IAC1B,MAAM,KAAK,GAAgB;QACzB,KAAK,EAAE,IAAI,GAAG,EAAE;QAChB,IAAI,EAAE,IAAI,GAAG,EAAE;KAChB,CAAC;IAEF,MAAM,MAAM,GAAc;QACxB,KAAK,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,QAAgB,EAAE,EAAE;YAChC,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;YACzB,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,CAAC,CAAC;QAEF,SAAS,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,SAAiB,EAAE,IAAY,EAAE,EAAE;YACnD,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,EAAE,IAAI,CAAC,CAAC;YACjC,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,CAAC,CAAC;QAEF,QAAQ,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,QAAgB,EAAE,EAAE;YACnC,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;YACvC,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;gBACvB,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,QAAQ,EAAE,CAAC,CAAC,CAAC;YACxE,CAAC;YACD,OAAO,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC/B,CAAC,CAAC;QAEF,OAAO,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,OAAe,EAAE,EAAE;YACjC,MAAM,MAAM,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,GAAG,GAAG,CAAC;YAC/D,MAAM,OAAO,GAAa,EAAE,CAAC;YAC7B,KAAK,MAAM,GAAG,IAAI,KAAK,CAAC,KAAK,CAAC,IAAI,EAAE,EAAE,CAAC;gBACrC,IAAI,GAAG,CAAC,UAAU,CAAC,MAAM,CAAC,EAAE,CAAC;oBAC3B,MAAM,IAAI,GAAG,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;oBACtC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;wBACxB,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBACrB,CAAC;gBACH,CAAC;YACH,CAAC;YACD,OAAO,OAAO,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;QAClC,CAAC,CAAC;QAEF,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,QAAgB,EAAE,EAAE;YACjC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC/B,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,QAAQ,EAAE,CAAC,CAAC,CAAC;YACxE,CAAC;YACD,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC7B,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,CAAC,CAAC;QAEF,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,OAAe,EAAE,OAAe,EAAE,EAAE;YACjD,MAAM,IAAI,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;YACtC,IAAI,IAAI,KAAK,SAAS,EAAE,CAAC;gBACvB,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,OAAO,EAAE,CAAC,CAAC,CAAC;YACvE,CAAC;YACD,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC;YAC/B,KAAK,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAC5B,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,CAAC,CAAC;QAEF,IAAI,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,QAAgB,EAAE,EAAE;YAC/B,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC/B,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,QAAQ,EAAE,CAAC,CAAC,CAAC;YACxE,CAAC;YACD,OAAO,OAAO,CAAC,OAAO,CAAC,EAAE,MAAM,EAAE,GAAG,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC;QACjD,CAAC,CAAC;QAEF,MAAM,EAAE,EAAE,CAAC,EAAE,CAAC,CAAC,QAAgB,EAAE,EAAE;YACjC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,EAAE,CAAC;gBAC/B,OAAO,OAAO,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,yBAAyB,QAAQ,EAAE,CAAC,CAAC,CAAC;YACxE,CAAC;YACD,OAAO,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,CAAC,CAAC;KACH,CAAC;IAEF,OAAO,EAAE,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;AAC/B,CAAC"}
@@ -0,0 +1,35 @@
1
+ import type { IStorage, ProviderFilter, ScenarioFilter, RunFilter, EvaluationFilter } from '../interfaces/storage.js';
2
+ import type { Provider, Scenario, Run, Evaluation } from '../types/index.js';
3
+ import type { FsAdapter } from './fs-adapter.js';
4
+ export declare class JsonFileStorage implements IStorage {
5
+ private readonly basePath;
6
+ private readonly fs;
7
+ private initialized;
8
+ private readonly entities;
9
+ constructor(basePath?: string, fsAdapter?: FsAdapter);
10
+ private ensureInit;
11
+ private entityDir;
12
+ private entityPath;
13
+ private getEntity;
14
+ private listEntities;
15
+ private saveEntity;
16
+ private deleteEntity;
17
+ getProvider(id: string): Promise<Provider | undefined>;
18
+ listProviders(filter?: ProviderFilter): Promise<readonly Provider[]>;
19
+ saveProvider(provider: Provider): Promise<void>;
20
+ deleteProvider(id: string): Promise<boolean>;
21
+ getScenario(id: string): Promise<Scenario | undefined>;
22
+ listScenarios(filter?: ScenarioFilter): Promise<readonly Scenario[]>;
23
+ saveScenario(scenario: Scenario): Promise<void>;
24
+ deleteScenario(id: string): Promise<boolean>;
25
+ private normalizeRun;
26
+ getRun(id: string): Promise<Run | undefined>;
27
+ listRuns(filter?: RunFilter): Promise<readonly Run[]>;
28
+ saveRun(run: Run): Promise<void>;
29
+ deleteRun(id: string): Promise<boolean>;
30
+ getEvaluation(id: string): Promise<Evaluation | undefined>;
31
+ listEvaluations(filter?: EvaluationFilter): Promise<readonly Evaluation[]>;
32
+ saveEvaluation(evaluation: Evaluation): Promise<void>;
33
+ deleteEvaluation(id: string): Promise<boolean>;
34
+ }
35
+ //# sourceMappingURL=storage.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"storage.d.ts","sourceRoot":"","sources":["../../../src/server/services/storage.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,QAAQ,EAAE,cAAc,EAAE,cAAc,EAAE,SAAS,EAAE,gBAAgB,EAAE,MAAM,0BAA0B,CAAC;AACtH,OAAO,KAAK,EACV,QAAQ,EACR,QAAQ,EACR,GAAG,EACH,UAAU,EACX,MAAM,mBAAmB,CAAC;AAE3B,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAQjD,qBAAa,eAAgB,YAAW,QAAQ;IAC9C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAS;IAClC,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAY;IAC/B,OAAO,CAAC,WAAW,CAAS;IAE5B,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAKvB;gBAEU,QAAQ,CAAC,EAAE,MAAM,EAAE,SAAS,CAAC,EAAE,SAAS;YAgDtC,UAAU;IAwBxB,OAAO,CAAC,SAAS;IAIjB,OAAO,CAAC,UAAU;YAMJ,SAAS;YAWT,YAAY;YA6BZ,UAAU;YAeV,YAAY;IAc1B,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,GAAG,SAAS,CAAC;IAItD,aAAa,CAAC,MAAM,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,SAAS,QAAQ,EAAE,CAAC;IAKpE,YAAY,CAAC,QAAQ,EAAE,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;IAK/C,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAM5C,WAAW,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,GAAG,SAAS,CAAC;IAItD,aAAa,CAAC,MAAM,CAAC,EAAE,cAAc,GAAG,OAAO,CAAC,SAAS,QAAQ,EAAE,CAAC;IAKpE,YAAY,CAAC,QAAQ,EAAE,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC;IAK/C,cAAc,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAK5C,OAAO,CAAC,YAAY;IAed,MAAM,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,GAAG,GAAG,SAAS,CAAC;IAI5C,QAAQ,CAAC,MAAM,CAAC,EAAE,SAAS,GAAG,OAAO,CAAC,SAAS,GAAG,EAAE,CAAC;IAI3D,OAAO,CAAC,GAAG,EAAE,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC;IAChC,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;IAGvC,aAAa,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,GAAG,SAAS,CAAC;IAC1D,eAAe,CAAC,MAAM,CAAC,EAAE,gBAAgB,GAAG,OAAO,CAAC,SAAS,UAAU,EAAE,CAAC;IAC1E,cAAc,CAAC,UAAU,EAAE,UAAU,GAAG,OAAO,CAAC,IAAI,CAAC;IACrD,gBAAgB,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,OAAO,CAAC;CAC/C"}
@@ -0,0 +1,219 @@
1
+ import path from 'node:path';
2
+ import { randomUUID } from 'node:crypto';
3
+ import { defaultFs } from './fs-adapter.js';
4
+ export class JsonFileStorage {
5
+ basePath;
6
+ fs;
7
+ initialized = false;
8
+ entities;
9
+ constructor(basePath, fsAdapter) {
10
+ this.basePath = basePath ?? path.join(process.cwd(), '.claude-test-bench');
11
+ this.fs = fsAdapter ?? defaultFs;
12
+ this.entities = {
13
+ providers: {
14
+ subdir: 'providers',
15
+ sensitive: true,
16
+ matchesFilter: (s, f) => {
17
+ if (!f)
18
+ return true;
19
+ if (f.provider && s.provider.kind !== f.provider)
20
+ return false;
21
+ if (f.model && s.provider.model !== f.model)
22
+ return false;
23
+ return true;
24
+ },
25
+ },
26
+ scenarios: {
27
+ subdir: 'scenarios/custom',
28
+ sensitive: false,
29
+ matchesFilter: (s, f) => {
30
+ if (!f)
31
+ return true;
32
+ if (f.category && s.category !== f.category)
33
+ return false;
34
+ return true;
35
+ },
36
+ },
37
+ runs: {
38
+ subdir: 'runs',
39
+ sensitive: false,
40
+ matchesFilter: (r, f) => {
41
+ if (!f)
42
+ return true;
43
+ if (f.providerId && r.providerId !== f.providerId)
44
+ return false;
45
+ if (f.scenarioId && r.scenarioId !== f.scenarioId)
46
+ return false;
47
+ if (f.status && r.status !== f.status)
48
+ return false;
49
+ return true;
50
+ },
51
+ },
52
+ evaluations: {
53
+ subdir: 'evaluations',
54
+ sensitive: false,
55
+ matchesFilter: (e, f) => {
56
+ if (!f)
57
+ return true;
58
+ if (f.runId && e.runId !== f.runId)
59
+ return false;
60
+ if (f.status && e.status !== f.status)
61
+ return false;
62
+ return true;
63
+ },
64
+ },
65
+ };
66
+ }
67
+ async ensureInit() {
68
+ if (this.initialized)
69
+ return;
70
+ // Migrate legacy 'setups' directory to 'providers'
71
+ const legacyDir = path.join(this.basePath, 'setups');
72
+ const newDir = path.join(this.basePath, 'providers');
73
+ try {
74
+ await this.fs.access(legacyDir);
75
+ try {
76
+ await this.fs.access(newDir);
77
+ }
78
+ catch {
79
+ // New dir doesn't exist — rename legacy dir
80
+ await this.fs.rename(legacyDir, newDir);
81
+ }
82
+ }
83
+ catch {
84
+ // Legacy dir doesn't exist — nothing to migrate
85
+ }
86
+ for (const cfg of Object.values(this.entities)) {
87
+ await this.fs.mkdir(path.join(this.basePath, cfg.subdir), { recursive: true });
88
+ }
89
+ this.initialized = true;
90
+ }
91
+ entityDir(subdir) {
92
+ return path.join(this.basePath, subdir);
93
+ }
94
+ entityPath(subdir, id) {
95
+ return path.join(this.basePath, subdir, `${id}.json`);
96
+ }
97
+ // ─── Generic CRUD ──────────────────────────────────────────────────
98
+ async getEntity(subdir, id) {
99
+ await this.ensureInit();
100
+ const filePath = this.entityPath(subdir, id);
101
+ try {
102
+ const raw = await this.fs.readFile(filePath, 'utf-8');
103
+ return JSON.parse(raw);
104
+ }
105
+ catch {
106
+ return undefined;
107
+ }
108
+ }
109
+ async listEntities(subdir, matchesFilter, filter) {
110
+ await this.ensureInit();
111
+ const dir = this.entityDir(subdir);
112
+ let files;
113
+ try {
114
+ files = await this.fs.readdir(dir);
115
+ }
116
+ catch {
117
+ return [];
118
+ }
119
+ const results = [];
120
+ for (const file of files) {
121
+ if (!file.endsWith('.json'))
122
+ continue;
123
+ try {
124
+ const raw = await this.fs.readFile(path.join(dir, file), 'utf-8');
125
+ const entity = JSON.parse(raw);
126
+ if (matchesFilter(entity, filter)) {
127
+ results.push(entity);
128
+ }
129
+ }
130
+ catch {
131
+ // Skip corrupt files
132
+ }
133
+ }
134
+ return results;
135
+ }
136
+ async saveEntity(subdir, entity, sensitive) {
137
+ await this.ensureInit();
138
+ const dir = this.entityDir(subdir);
139
+ const tmpPath = path.join(dir, `.tmp-${randomUUID()}.json`);
140
+ const finalPath = this.entityPath(subdir, entity.id);
141
+ const data = JSON.stringify(entity, null, 2);
142
+ const opts = sensitive ? { mode: 0o600 } : undefined;
143
+ await this.fs.writeFile(tmpPath, data, opts);
144
+ await this.fs.rename(tmpPath, finalPath);
145
+ }
146
+ async deleteEntity(subdir, id) {
147
+ await this.ensureInit();
148
+ const filePath = this.entityPath(subdir, id);
149
+ try {
150
+ await this.fs.access(filePath);
151
+ await this.fs.unlink(filePath);
152
+ return true;
153
+ }
154
+ catch {
155
+ return false;
156
+ }
157
+ }
158
+ // ─── Providers ──────────────────────────────────────────────────────
159
+ getProvider(id) {
160
+ return this.getEntity(this.entities.providers.subdir, id);
161
+ }
162
+ listProviders(filter) {
163
+ const cfg = this.entities.providers;
164
+ return this.listEntities(cfg.subdir, cfg.matchesFilter, filter);
165
+ }
166
+ saveProvider(provider) {
167
+ const cfg = this.entities.providers;
168
+ return this.saveEntity(cfg.subdir, provider, cfg.sensitive);
169
+ }
170
+ deleteProvider(id) {
171
+ return this.deleteEntity(this.entities.providers.subdir, id);
172
+ }
173
+ // ─── Scenarios ─────────────────────────────────────────────────────
174
+ getScenario(id) {
175
+ return this.getEntity(this.entities.scenarios.subdir, id);
176
+ }
177
+ listScenarios(filter) {
178
+ const cfg = this.entities.scenarios;
179
+ return this.listEntities(cfg.subdir, cfg.matchesFilter, filter);
180
+ }
181
+ saveScenario(scenario) {
182
+ const cfg = this.entities.scenarios;
183
+ return this.saveEntity(cfg.subdir, scenario, cfg.sensitive);
184
+ }
185
+ deleteScenario(id) {
186
+ return this.deleteEntity(this.entities.scenarios.subdir, id);
187
+ }
188
+ // ─── Runs ──────────────────────────────────────────────────────────
189
+ normalizeRun(raw) {
190
+ // Accept both old and new field names for backward compat
191
+ const r = raw;
192
+ if ('setupId' in r && !('providerId' in r)) {
193
+ return {
194
+ ...raw,
195
+ providerId: r['setupId'],
196
+ providerSnapshot: (r['setupSnapshot'] ?? raw.providerSnapshot),
197
+ reviewerProviderIds: (r['reviewerSetupIds'] ?? raw.reviewerProviderIds),
198
+ reviewerProviderSnapshots: (r['reviewerSetupSnapshots'] ?? raw.reviewerProviderSnapshots),
199
+ };
200
+ }
201
+ return raw;
202
+ }
203
+ async getRun(id) {
204
+ const raw = await this.getEntity(this.entities.runs.subdir, id);
205
+ return raw ? this.normalizeRun(raw) : undefined;
206
+ }
207
+ async listRuns(filter) {
208
+ const results = await this.listEntities(this.entities.runs.subdir, this.entities.runs.matchesFilter, filter);
209
+ return results.map((r) => this.normalizeRun(r));
210
+ }
211
+ saveRun(run) { return this.saveEntity(this.entities.runs.subdir, run, this.entities.runs.sensitive); }
212
+ deleteRun(id) { return this.deleteEntity(this.entities.runs.subdir, id); }
213
+ // ─── Evaluations ───────────────────────────────────────────────────
214
+ getEvaluation(id) { return this.getEntity(this.entities.evaluations.subdir, id); }
215
+ listEvaluations(filter) { return this.listEntities(this.entities.evaluations.subdir, this.entities.evaluations.matchesFilter, filter); }
216
+ saveEvaluation(evaluation) { return this.saveEntity(this.entities.evaluations.subdir, evaluation, this.entities.evaluations.sensitive); }
217
+ deleteEvaluation(id) { return this.deleteEntity(this.entities.evaluations.subdir, id); }
218
+ }
219
+ //# sourceMappingURL=storage.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"storage.js","sourceRoot":"","sources":["../../../src/server/services/storage.ts"],"names":[],"mappings":"AAAA,OAAO,IAAI,MAAM,WAAW,CAAC;AAC7B,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAQzC,OAAO,EAAE,SAAS,EAAE,MAAM,iBAAiB,CAAC;AAS5C,MAAM,OAAO,eAAe;IACT,QAAQ,CAAS;IACjB,EAAE,CAAY;IACvB,WAAW,GAAG,KAAK,CAAC;IAEX,QAAQ,CAKvB;IAEF,YAAY,QAAiB,EAAE,SAAqB;QAClD,IAAI,CAAC,QAAQ,GAAG,QAAQ,IAAI,IAAI,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,EAAE,EAAE,oBAAoB,CAAC,CAAC;QAC3E,IAAI,CAAC,EAAE,GAAG,SAAS,IAAI,SAAS,CAAC;QAEjC,IAAI,CAAC,QAAQ,GAAG;YACd,SAAS,EAAE;gBACT,MAAM,EAAE,WAAW;gBACnB,SAAS,EAAE,IAAI;gBACf,aAAa,EAAE,CAAC,CAAW,EAAE,CAAkB,EAAE,EAAE;oBACjD,IAAI,CAAC,CAAC;wBAAE,OAAO,IAAI,CAAC;oBACpB,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,QAAQ,CAAC,IAAI,KAAK,CAAC,CAAC,QAAQ;wBAAE,OAAO,KAAK,CAAC;oBAC/D,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,QAAQ,CAAC,KAAK,KAAK,CAAC,CAAC,KAAK;wBAAE,OAAO,KAAK,CAAC;oBAC1D,OAAO,IAAI,CAAC;gBACd,CAAC;aACF;YACD,SAAS,EAAE;gBACT,MAAM,EAAE,kBAAkB;gBAC1B,SAAS,EAAE,KAAK;gBAChB,aAAa,EAAE,CAAC,CAAW,EAAE,CAAkB,EAAE,EAAE;oBACjD,IAAI,CAAC,CAAC;wBAAE,OAAO,IAAI,CAAC;oBACpB,IAAI,CAAC,CAAC,QAAQ,IAAI,CAAC,CAAC,QAAQ,KAAK,CAAC,CAAC,QAAQ;wBAAE,OAAO,KAAK,CAAC;oBAC1D,OAAO,IAAI,CAAC;gBACd,CAAC;aACF;YACD,IAAI,EAAE;gBACJ,MAAM,EAAE,MAAM;gBACd,SAAS,EAAE,KAAK;gBAChB,aAAa,EAAE,CAAC,CAAM,EAAE,CAAa,EAAE,EAAE;oBACvC,IAAI,CAAC,CAAC;wBAAE,OAAO,IAAI,CAAC;oBACpB,IAAI,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,UAAU;wBAAE,OAAO,KAAK,CAAC;oBAChE,IAAI,CAAC,CAAC,UAAU,IAAI,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,UAAU;wBAAE,OAAO,KAAK,CAAC;oBAChE,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;wBAAE,OAAO,KAAK,CAAC;oBACpD,OAAO,IAAI,CAAC;gBACd,CAAC;aACF;YACD,WAAW,EAAE;gBACX,MAAM,EAAE,aAAa;gBACrB,SAAS,EAAE,KAAK;gBAChB,aAAa,EAAE,CAAC,CAAa,EAAE,CAAoB,EAAE,EAAE;oBACrD,IAAI,CAAC,CAAC;wBAAE,OAAO,IAAI,CAAC;oBACpB,IAAI,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC,KAAK,KAAK,CAAC,CAAC,KAAK;wBAAE,OAAO,KAAK,CAAC;oBACjD,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM;wBAAE,OAAO,KAAK,CAAC;oBACpD,OAAO,IAAI,CAAC;gBACd,CAAC;aACF;SACF,CAAC;IACJ,CAAC;IAEO,KAAK,CAAC,UAAU;QACtB,IAAI,IAAI,CAAC,WAAW;YAAE,OAAO;QAE7B,mDAAmD;QACnD,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;QACrD,MAAM,MAAM,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC;QACrD,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;YAChC,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;YAC/B,CAAC;YAAC,MAAM,CAAC;gBACP,4CAA4C;gBAC5C,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;YAC1C,CAAC;QACH,CAAC;QAAC,MAAM,CAAC;YACP,gDAAgD;QAClD,CAAC;QAED,KAAK,MAAM,GAAG,IAAI,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,QAAQ,CAAC,EAAE,CAAC;YAC/C,MAAM,IAAI,CAAC,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,GAAG,CAAC,MAAM,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACjF,CAAC;QACD,IAAI,CAAC,WAAW,GAAG,IAAI,CAAC;IAC1B,CAAC;IAEO,SAAS,CAAC,MAAc;QAC9B,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,CAAC,CAAC;IAC1C,CAAC;IAEO,UAAU,CAAC,MAAc,EAAE,EAAU;QAC3C,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,EAAE,OAAO,CAAC,CAAC;IACxD,CAAC;IAED,sEAAsE;IAE9D,KAAK,CAAC,SAAS,CAAI,MAAc,EAAE,EAAU;QACnD,MAAM,IAAI,CAAC,UAAU,EAAE,CAAC;QACxB,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAC7C,IAAI,CAAC;YACH,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC,QAAQ,EAAE,OAAO,CAAC,CAAC;YACtD,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAM,CAAC;QAC9B,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,SAAS,CAAC;QACnB,CAAC;IACH,CAAC;IAEO,KAAK,CAAC,YAAY,CACxB,MAAc,EACd,aAAiD,EACjD,MAAU;QAEV,MAAM,IAAI,CAAC,UAAU,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QACnC,IAAI,KAAe,CAAC;QACpB,IAAI,CAAC;YACH,KAAK,GAAG,MAAM,IAAI,CAAC,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;QACrC,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,EAAE,CAAC;QACZ,CAAC;QACD,MAAM,OAAO,GAAQ,EAAE,CAAC;QACxB,KAAK,MAAM,IAAI,IAAI,KAAK,EAAE,CAAC;YACzB,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,OAAO,CAAC;gBAAE,SAAS;YACtC,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,EAAE,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,IAAI,CAAC,EAAE,OAAO,CAAC,CAAC;gBAClE,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAM,CAAC;gBACpC,IAAI,aAAa,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC;oBAClC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACvB,CAAC;YACH,CAAC;YAAC,MAAM,CAAC;gBACP,qBAAqB;YACvB,CAAC;QACH,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IAEO,KAAK,CAAC,UAAU,CACtB,MAAc,EACd,MAAS,EACT,SAAkB;QAElB,MAAM,IAAI,CAAC,UAAU,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC;QACnC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,QAAQ,UAAU,EAAE,OAAO,CAAC,CAAC;QAC5D,MAAM,SAAS,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QACrD,MAAM,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,CAAC;QAC7C,MAAM,IAAI,GAAG,SAAS,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;QACrD,MAAM,IAAI,CAAC,EAAE,CAAC,SAAS,CAAC,OAAO,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;QAC7C,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,CAAC;IAC3C,CAAC;IAEO,KAAK,CAAC,YAAY,CAAC,MAAc,EAAE,EAAU;QACnD,MAAM,IAAI,CAAC,UAAU,EAAE,CAAC;QACxB,MAAM,QAAQ,GAAG,IAAI,CAAC,UAAU,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QAC7C,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC/B,MAAM,IAAI,CAAC,EAAE,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;YAC/B,OAAO,IAAI,CAAC;QACd,CAAC;QAAC,MAAM,CAAC;YACP,OAAO,KAAK,CAAC;QACf,CAAC;IACH,CAAC;IAED,uEAAuE;IAEvE,WAAW,CAAC,EAAU;QACpB,OAAO,IAAI,CAAC,SAAS,CAAW,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACtE,CAAC;IAED,aAAa,CAAC,MAAuB;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC;QACpC,OAAO,IAAI,CAAC,YAAY,CAA2B,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;IAC5F,CAAC;IAED,YAAY,CAAC,QAAkB;QAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC;QACpC,OAAO,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,EAAE,GAAG,CAAC,SAAS,CAAC,CAAC;IAC9D,CAAC;IAED,cAAc,CAAC,EAAU;QACvB,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,sEAAsE;IAEtE,WAAW,CAAC,EAAU;QACpB,OAAO,IAAI,CAAC,SAAS,CAAW,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IACtE,CAAC;IAED,aAAa,CAAC,MAAuB;QACnC,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC;QACpC,OAAO,IAAI,CAAC,YAAY,CAA2B,GAAG,CAAC,MAAM,EAAE,GAAG,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;IAC5F,CAAC;IAED,YAAY,CAAC,QAAkB;QAC7B,MAAM,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC;QACpC,OAAO,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,MAAM,EAAE,QAAQ,EAAE,GAAG,CAAC,SAAS,CAAC,CAAC;IAC9D,CAAC;IAED,cAAc,CAAC,EAAU;QACvB,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,SAAS,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;IAC/D,CAAC;IAED,sEAAsE;IAC9D,YAAY,CAAC,GAAQ;QAC3B,0DAA0D;QAC1D,MAAM,CAAC,GAAG,GAAyC,CAAC;QACpD,IAAI,SAAS,IAAI,CAAC,IAAI,CAAC,CAAC,YAAY,IAAI,CAAC,CAAC,EAAE,CAAC;YAC3C,OAAO;gBACL,GAAI,GAAW;gBACf,UAAU,EAAE,CAAC,CAAC,SAAS,CAAW;gBAClC,gBAAgB,EAAE,CAAC,CAAC,CAAC,eAAe,CAAC,IAAI,GAAG,CAAC,gBAAgB,CAA4B;gBACzF,mBAAmB,EAAE,CAAC,CAAC,CAAC,kBAAkB,CAAC,IAAI,GAAG,CAAC,mBAAmB,CAA+B;gBACrG,yBAAyB,EAAE,CAAC,CAAC,CAAC,wBAAwB,CAAC,IAAI,GAAG,CAAC,yBAAyB,CAAqC;aAC9H,CAAC;QACJ,CAAC;QACD,OAAO,GAAG,CAAC;IACb,CAAC;IAED,KAAK,CAAC,MAAM,CAAC,EAAU;QACrB,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,SAAS,CAAM,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC;QACrE,OAAO,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAClD,CAAC;IACD,KAAK,CAAC,QAAQ,CAAC,MAAkB;QAC/B,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,YAAY,CAAiB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC;QAC7H,OAAO,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC;IAClD,CAAC;IACD,OAAO,CAAC,GAAQ,IAAmB,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;IAC1H,SAAS,CAAC,EAAU,IAAsB,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;IAEpG,sEAAsE;IACtE,aAAa,CAAC,EAAU,IAAqC,OAAO,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;IAC3H,eAAe,CAAC,MAAyB,IAAoC,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,aAAa,EAAE,MAAM,CAAC,CAAC,CAAC,CAAC;IAC3L,cAAc,CAAC,UAAsB,IAAmB,OAAO,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,UAAU,EAAE,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC;IACpK,gBAAgB,CAAC,EAAU,IAAsB,OAAO,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;CACnH"}
@@ -0,0 +1,18 @@
1
+ import type { SDKMessageRecord } from '../types/index.js';
2
+ /** Structured summary extracted from a transcript. */
3
+ export interface TranscriptSummary {
4
+ readonly toolCallSequence: readonly string[];
5
+ readonly filesRead: readonly string[];
6
+ readonly filesModified: readonly string[];
7
+ readonly commandFailures: readonly string[];
8
+ readonly retryPatterns: readonly string[];
9
+ readonly askedClarifyingQuestions: boolean;
10
+ }
11
+ /** Full transcript output. */
12
+ export interface TranscriptResult {
13
+ readonly text: string;
14
+ readonly summary: TranscriptSummary;
15
+ }
16
+ /** Format SDK messages into a readable transcript and structured summary. */
17
+ export declare function formatTranscript(messages: readonly SDKMessageRecord[]): TranscriptResult;
18
+ //# sourceMappingURL=transcript-formatter.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"transcript-formatter.d.ts","sourceRoot":"","sources":["../../../src/server/services/transcript-formatter.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAO1D,sDAAsD;AACtD,MAAM,WAAW,iBAAiB;IAChC,QAAQ,CAAC,gBAAgB,EAAE,SAAS,MAAM,EAAE,CAAC;IAC7C,QAAQ,CAAC,SAAS,EAAE,SAAS,MAAM,EAAE,CAAC;IACtC,QAAQ,CAAC,aAAa,EAAE,SAAS,MAAM,EAAE,CAAC;IAC1C,QAAQ,CAAC,eAAe,EAAE,SAAS,MAAM,EAAE,CAAC;IAC5C,QAAQ,CAAC,aAAa,EAAE,SAAS,MAAM,EAAE,CAAC;IAC1C,QAAQ,CAAC,wBAAwB,EAAE,OAAO,CAAC;CAC5C;AAED,8BAA8B;AAC9B,MAAM,WAAW,gBAAgB;IAC/B,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,OAAO,EAAE,iBAAiB,CAAC;CACrC;AAED,6EAA6E;AAC7E,wBAAgB,gBAAgB,CAAC,QAAQ,EAAE,SAAS,gBAAgB,EAAE,GAAG,gBAAgB,CAyExF"}
@@ -0,0 +1,156 @@
1
+ // ---------------------------------------------------------------------------
2
+ // Transcript Formatter — converts SDK messages into evaluator-readable text
3
+ // ---------------------------------------------------------------------------
4
+ const MAX_TRANSCRIPT_CHARS = 100_000;
5
+ const TRUNCATED_MARKER = '\n[transcript truncated]';
6
+ const MAX_TOOL_INPUT_CHARS = 500;
7
+ const MAX_TOOL_OUTPUT_CHARS = 500;
8
+ /** Format SDK messages into a readable transcript and structured summary. */
9
+ export function formatTranscript(messages) {
10
+ const lines = [];
11
+ const toolCalls = [];
12
+ const filesRead = new Set();
13
+ const filesModified = new Set();
14
+ const commandFailures = [];
15
+ const retryPatterns = [];
16
+ let askedClarifyingQuestions = false;
17
+ let totalChars = 0;
18
+ let truncated = false;
19
+ for (const record of messages) {
20
+ if (truncated)
21
+ break;
22
+ const msg = record.message;
23
+ const msgType = msg['type'];
24
+ if (msgType === 'assistant') {
25
+ const formatted = formatAssistantMessage(msg);
26
+ if (formatted) {
27
+ const result = appendLine(lines, formatted, totalChars);
28
+ totalChars = result.totalChars;
29
+ truncated = result.truncated;
30
+ if (looksLikeQuestion(formatted)) {
31
+ askedClarifyingQuestions = true;
32
+ }
33
+ }
34
+ }
35
+ else if (msgType === 'tool_use') {
36
+ const name = msg['name'] ?? 'unknown_tool';
37
+ const input = truncateStr(JSON.stringify(msg['input'] ?? ''), MAX_TOOL_INPUT_CHARS);
38
+ toolCalls.push(name);
39
+ trackFileAccess(name, msg, filesRead, filesModified);
40
+ const line = `[Tool Call] ${name}: ${input}`;
41
+ const result = appendLine(lines, line, totalChars);
42
+ totalChars = result.totalChars;
43
+ truncated = result.truncated;
44
+ }
45
+ else if (msgType === 'tool_result') {
46
+ const output = truncateStr(extractToolResultText(msg), MAX_TOOL_OUTPUT_CHARS);
47
+ const isError = msg['is_error'] === true;
48
+ if (isError) {
49
+ commandFailures.push(output);
50
+ detectRetryPattern(toolCalls, retryPatterns);
51
+ }
52
+ const prefix = isError ? '[Tool Error]' : '[Tool Result]';
53
+ const line = `${prefix} ${output}`;
54
+ const result = appendLine(lines, line, totalChars);
55
+ totalChars = result.totalChars;
56
+ truncated = result.truncated;
57
+ }
58
+ else if (msgType === 'result') {
59
+ const resultText = msg['result'] ?? '';
60
+ if (resultText) {
61
+ const line = `[Final Result] ${resultText}`;
62
+ const result = appendLine(lines, line, totalChars);
63
+ totalChars = result.totalChars;
64
+ truncated = result.truncated;
65
+ }
66
+ }
67
+ }
68
+ if (truncated) {
69
+ lines.push(TRUNCATED_MARKER);
70
+ }
71
+ return {
72
+ text: lines.join('\n'),
73
+ summary: {
74
+ toolCallSequence: toolCalls,
75
+ filesRead: [...filesRead],
76
+ filesModified: [...filesModified],
77
+ commandFailures,
78
+ retryPatterns,
79
+ askedClarifyingQuestions,
80
+ },
81
+ };
82
+ }
83
+ // ---------------------------------------------------------------------------
84
+ // Internal helpers
85
+ // ---------------------------------------------------------------------------
86
+ function appendLine(lines, line, totalChars) {
87
+ const newTotal = totalChars + line.length + 1; // +1 for newline
88
+ if (newTotal > MAX_TRANSCRIPT_CHARS) {
89
+ return { totalChars, truncated: true };
90
+ }
91
+ lines.push(line);
92
+ return { totalChars: newTotal, truncated: false };
93
+ }
94
+ function formatAssistantMessage(msg) {
95
+ const content = msg['message'];
96
+ if (!content)
97
+ return undefined;
98
+ const contentBlocks = content['content'];
99
+ if (!Array.isArray(contentBlocks))
100
+ return undefined;
101
+ const textParts = [];
102
+ for (const block of contentBlocks) {
103
+ if (block['type'] === 'text' && typeof block['text'] === 'string') {
104
+ textParts.push(block['text']);
105
+ }
106
+ }
107
+ return textParts.length > 0 ? `[Assistant] ${textParts.join(' ')}` : undefined;
108
+ }
109
+ function extractToolResultText(msg) {
110
+ const content = msg['content'];
111
+ if (typeof content === 'string')
112
+ return content;
113
+ if (Array.isArray(content)) {
114
+ const parts = [];
115
+ for (const block of content) {
116
+ if (block['type'] === 'text' && typeof block['text'] === 'string') {
117
+ parts.push(block['text']);
118
+ }
119
+ }
120
+ return parts.join(' ');
121
+ }
122
+ return '';
123
+ }
124
+ function truncateStr(s, max) {
125
+ return s.length > max ? s.slice(0, max) + '...' : s;
126
+ }
127
+ function looksLikeQuestion(text) {
128
+ return /\?\s*$/.test(text.trim());
129
+ }
130
+ function trackFileAccess(toolName, msg, filesRead, filesModified) {
131
+ const input = msg['input'];
132
+ if (!input)
133
+ return;
134
+ const filePath = (input['file_path'] ?? input['path'] ?? input['filename']);
135
+ if (!filePath)
136
+ return;
137
+ if (toolName === 'Read' || toolName === 'Glob' || toolName === 'Grep') {
138
+ filesRead.add(filePath);
139
+ }
140
+ else if (toolName === 'Edit' || toolName === 'Write' || toolName === 'NotebookEdit') {
141
+ filesModified.add(filePath);
142
+ }
143
+ }
144
+ function detectRetryPattern(toolCalls, retryPatterns) {
145
+ if (toolCalls.length < 2)
146
+ return;
147
+ const last = toolCalls[toolCalls.length - 1];
148
+ const prev = toolCalls[toolCalls.length - 2];
149
+ if (last === prev) {
150
+ const pattern = `Repeated ${last} after error`;
151
+ if (!retryPatterns.includes(pattern)) {
152
+ retryPatterns.push(pattern);
153
+ }
154
+ }
155
+ }
156
+ //# sourceMappingURL=transcript-formatter.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"transcript-formatter.js","sourceRoot":"","sources":["../../../src/server/services/transcript-formatter.ts"],"names":[],"mappings":"AAAA,8EAA8E;AAC9E,4EAA4E;AAC5E,8EAA8E;AAI9E,MAAM,oBAAoB,GAAG,OAAO,CAAC;AACrC,MAAM,gBAAgB,GAAG,0BAA0B,CAAC;AACpD,MAAM,oBAAoB,GAAG,GAAG,CAAC;AACjC,MAAM,qBAAqB,GAAG,GAAG,CAAC;AAkBlC,6EAA6E;AAC7E,MAAM,UAAU,gBAAgB,CAAC,QAAqC;IACpE,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,MAAM,SAAS,GAAa,EAAE,CAAC;IAC/B,MAAM,SAAS,GAAgB,IAAI,GAAG,EAAE,CAAC;IACzC,MAAM,aAAa,GAAgB,IAAI,GAAG,EAAE,CAAC;IAC7C,MAAM,eAAe,GAAa,EAAE,CAAC;IACrC,MAAM,aAAa,GAAa,EAAE,CAAC;IACnC,IAAI,wBAAwB,GAAG,KAAK,CAAC;IACrC,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,SAAS,GAAG,KAAK,CAAC;IAEtB,KAAK,MAAM,MAAM,IAAI,QAAQ,EAAE,CAAC;QAC9B,IAAI,SAAS;YAAE,MAAM;QACrB,MAAM,GAAG,GAAG,MAAM,CAAC,OAAO,CAAC;QAC3B,MAAM,OAAO,GAAG,GAAG,CAAC,MAAM,CAAuB,CAAC;QAElD,IAAI,OAAO,KAAK,WAAW,EAAE,CAAC;YAC5B,MAAM,SAAS,GAAG,sBAAsB,CAAC,GAAG,CAAC,CAAC;YAC9C,IAAI,SAAS,EAAE,CAAC;gBACd,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,SAAS,EAAE,UAAU,CAAC,CAAC;gBACxD,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;gBAC/B,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;gBAC7B,IAAI,iBAAiB,CAAC,SAAS,CAAC,EAAE,CAAC;oBACjC,wBAAwB,GAAG,IAAI,CAAC;gBAClC,CAAC;YACH,CAAC;QACH,CAAC;aAAM,IAAI,OAAO,KAAK,UAAU,EAAE,CAAC;YAClC,MAAM,IAAI,GAAI,GAAG,CAAC,MAAM,CAAY,IAAI,cAAc,CAAC;YACvD,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,EAAE,oBAAoB,CAAC,CAAC;YACpF,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YACrB,eAAe,CAAC,IAAI,EAAE,GAAG,EAAE,SAAS,EAAE,aAAa,CAAC,CAAC;YACrD,MAAM,IAAI,GAAG,eAAe,IAAI,KAAK,KAAK,EAAE,CAAC;YAC7C,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,IAAI,EAAE,UAAU,CAAC,CAAC;YACnD,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;YAC/B,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAC/B,CAAC;aAAM,IAAI,OAAO,KAAK,aAAa,EAAE,CAAC;YACrC,MAAM,MAAM,GAAG,WAAW,CAAC,qBAAqB,CAAC,GAAG,CAAC,EAAE,qBAAqB,CAAC,CAAC;YAC9E,MAAM,OAAO,GAAG,GAAG,CAAC,UAAU,CAAC,KAAK,IAAI,CAAC;YACzC,IAAI,OAAO,EAAE,CAAC;gBACZ,eAAe,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBAC7B,kBAAkB,CAAC,SAAS,EAAE,aAAa,CAAC,CAAC;YAC/C,CAAC;YACD,MAAM,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,eAAe,CAAC;YAC1D,MAAM,IAAI,GAAG,GAAG,MAAM,IAAI,MAAM,EAAE,CAAC;YACnC,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,IAAI,EAAE,UAAU,CAAC,CAAC;YACnD,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;YAC/B,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAC/B,CAAC;aAAM,IAAI,OAAO,KAAK,QAAQ,EAAE,CAAC;YAChC,MAAM,UAAU,GAAI,GAAG,CAAC,QAAQ,CAAY,IAAI,EAAE,CAAC;YACnD,IAAI,UAAU,EAAE,CAAC;gBACf,MAAM,IAAI,GAAG,kBAAkB,UAAU,EAAE,CAAC;gBAC5C,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,EAAE,IAAI,EAAE,UAAU,CAAC,CAAC;gBACnD,UAAU,GAAG,MAAM,CAAC,UAAU,CAAC;gBAC/B,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;YAC/B,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,SAAS,EAAE,CAAC;QACd,KAAK,CAAC,IAAI,CAAC,gBAAgB,CAAC,CAAC;IAC/B,CAAC;IAED,OAAO;QACL,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC;QACtB,OAAO,EAAE;YACP,gBAAgB,EAAE,SAAS;YAC3B,SAAS,EAAE,CAAC,GAAG,SAAS,CAAC;YACzB,aAAa,EAAE,CAAC,GAAG,aAAa,CAAC;YACjC,eAAe;YACf,aAAa;YACb,wBAAwB;SACzB;KACF,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,mBAAmB;AACnB,8EAA8E;AAE9E,SAAS,UAAU,CACjB,KAAe,EACf,IAAY,EACZ,UAAkB;IAElB,MAAM,QAAQ,GAAG,UAAU,GAAG,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,iBAAiB;IAChE,IAAI,QAAQ,GAAG,oBAAoB,EAAE,CAAC;QACpC,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC;IACzC,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACjB,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,SAAS,EAAE,KAAK,EAAE,CAAC;AACpD,CAAC;AAED,SAAS,sBAAsB,CAAC,GAAsC;IACpE,MAAM,OAAO,GAAG,GAAG,CAAC,SAAS,CAAwC,CAAC;IACtE,IAAI,CAAC,OAAO;QAAE,OAAO,SAAS,CAAC;IAE/B,MAAM,aAAa,GAAG,OAAO,CAAC,SAAS,CAAuD,CAAC;IAC/F,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC;QAAE,OAAO,SAAS,CAAC;IAEpD,MAAM,SAAS,GAAa,EAAE,CAAC;IAC/B,KAAK,MAAM,KAAK,IAAI,aAAa,EAAE,CAAC;QAClC,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,MAAM,IAAI,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,QAAQ,EAAE,CAAC;YAClE,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;QAChC,CAAC;IACH,CAAC;IACD,OAAO,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,eAAe,SAAS,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC;AACjF,CAAC;AAED,SAAS,qBAAqB,CAAC,GAAsC;IACnE,MAAM,OAAO,GAAG,GAAG,CAAC,SAAS,CAAgE,CAAC;IAC9F,IAAI,OAAO,OAAO,KAAK,QAAQ;QAAE,OAAO,OAAO,CAAC;IAChD,IAAI,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,EAAE,CAAC;QAC3B,MAAM,KAAK,GAAa,EAAE,CAAC;QAC3B,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;YAC5B,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,MAAM,IAAI,OAAO,KAAK,CAAC,MAAM,CAAC,KAAK,QAAQ,EAAE,CAAC;gBAClE,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC;YAC5B,CAAC;QACH,CAAC;QACD,OAAO,KAAK,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACzB,CAAC;IACD,OAAO,EAAE,CAAC;AACZ,CAAC;AAED,SAAS,WAAW,CAAC,CAAS,EAAE,GAAW;IACzC,OAAO,CAAC,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC;AACtD,CAAC;AAED,SAAS,iBAAiB,CAAC,IAAY;IACrC,OAAO,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;AACpC,CAAC;AAED,SAAS,eAAe,CACtB,QAAgB,EAChB,GAAsC,EACtC,SAAsB,EACtB,aAA0B;IAE1B,MAAM,KAAK,GAAG,GAAG,CAAC,OAAO,CAAwC,CAAC;IAClE,IAAI,CAAC,KAAK;QAAE,OAAO;IACnB,MAAM,QAAQ,GAAG,CAAC,KAAK,CAAC,WAAW,CAAC,IAAI,KAAK,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC,UAAU,CAAC,CAAuB,CAAC;IAClG,IAAI,CAAC,QAAQ;QAAE,OAAO;IAEtB,IAAI,QAAQ,KAAK,MAAM,IAAI,QAAQ,KAAK,MAAM,IAAI,QAAQ,KAAK,MAAM,EAAE,CAAC;QACtE,SAAS,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC1B,CAAC;SAAM,IAAI,QAAQ,KAAK,MAAM,IAAI,QAAQ,KAAK,OAAO,IAAI,QAAQ,KAAK,cAAc,EAAE,CAAC;QACtF,aAAa,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC;IAC9B,CAAC;AACH,CAAC;AAED,SAAS,kBAAkB,CAAC,SAA4B,EAAE,aAAuB;IAC/E,IAAI,SAAS,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO;IACjC,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;IAC7C,IAAI,IAAI,KAAK,IAAI,EAAE,CAAC;QAClB,MAAM,OAAO,GAAG,YAAY,IAAI,cAAc,CAAC;QAC/C,IAAI,CAAC,aAAa,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,CAAC;YACrC,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QAC9B,CAAC;IACH,CAAC;AACH,CAAC"}