stageflow 0.40.0 → 0.44.0

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 (519) hide show
  1. package/README.md +15 -4
  2. package/dist/agent/activity.d.ts +18 -0
  3. package/dist/agent/claudeAdapter.js +7 -0
  4. package/dist/agent/claudeTools.d.ts +27 -18
  5. package/dist/agent/claudeTools.js +54 -15
  6. package/dist/agent/compactionEvents.d.ts +34 -0
  7. package/dist/agent/compactionEvents.js +37 -0
  8. package/dist/agent/openRouterRouting.d.ts +7 -0
  9. package/dist/agent/openRouterRouting.js +23 -0
  10. package/dist/agent/piAdapter.d.ts +6 -0
  11. package/dist/agent/piAdapter.js +128 -27
  12. package/dist/agent/piSettings.d.ts +7 -0
  13. package/dist/agent/piSettings.js +9 -0
  14. package/dist/agent/sandboxPiTools.d.ts +3 -1
  15. package/dist/agent/sandboxPiTools.js +8 -2
  16. package/dist/agent/stageTurnDriver.d.ts +11 -2
  17. package/dist/agent/stageTurnDriver.js +45 -4
  18. package/dist/agent/toolOutputBudget.d.ts +15 -1
  19. package/dist/agent/toolOutputBudget.js +18 -0
  20. package/dist/agent/toolOutputBudgetExtension.d.ts +2 -0
  21. package/dist/agent/toolOutputBudgetExtension.js +40 -3
  22. package/dist/agents/authorChat.d.ts +5 -0
  23. package/dist/agents/authorChat.js +120 -12
  24. package/dist/agents/authorSessionStore.d.ts +38 -0
  25. package/dist/agents/authorSessionStore.js +167 -0
  26. package/dist/agents/chatTurn.d.ts +2 -0
  27. package/dist/agents/chatTurn.js +17 -1
  28. package/dist/agents/errors.d.ts +1 -1
  29. package/dist/agents/errors.js +1 -0
  30. package/dist/agents/evalGate.d.ts +61 -0
  31. package/dist/agents/evalGate.js +103 -0
  32. package/dist/agents/evalSession.d.ts +26 -0
  33. package/dist/agents/evalSession.js +29 -0
  34. package/dist/agents/hostContext.d.ts +4 -1
  35. package/dist/agents/hostContext.js +4 -1
  36. package/dist/agents/runtime/piRuntime.d.ts +6 -2
  37. package/dist/agents/runtime/piRuntime.js +50 -16
  38. package/dist/agents/runtime/types.d.ts +3 -1
  39. package/dist/agents/service.d.ts +27 -2
  40. package/dist/agents/service.js +91 -15
  41. package/dist/agents/sessions.d.ts +8 -0
  42. package/dist/agents/sessions.js +53 -11
  43. package/dist/agents/store.d.ts +14 -3
  44. package/dist/agents/store.js +45 -7
  45. package/dist/agents/toolRisk.d.ts +21 -0
  46. package/dist/agents/toolRisk.js +50 -0
  47. package/dist/cli/doctorCommand.js +26 -0
  48. package/dist/cli/evalCommand.d.ts +20 -0
  49. package/dist/cli/evalCommand.js +720 -0
  50. package/dist/cli/exportAllCommand.js +1 -1
  51. package/dist/cli/hostClient.d.ts +60 -0
  52. package/dist/cli/hostClient.js +71 -0
  53. package/dist/cli/validateOutput.d.ts +2 -2
  54. package/dist/cli/validateOutput.js +2 -2
  55. package/dist/cli.js +10 -0
  56. package/dist/compaction/directives.d.ts +18 -0
  57. package/dist/compaction/directives.js +119 -0
  58. package/dist/compaction/extension.d.ts +20 -0
  59. package/dist/compaction/extension.js +78 -0
  60. package/dist/compaction/mask.d.ts +18 -0
  61. package/dist/compaction/mask.js +155 -0
  62. package/dist/compaction/pin.d.ts +21 -0
  63. package/dist/compaction/pin.js +270 -0
  64. package/dist/compaction/scrub.d.ts +7 -0
  65. package/dist/compaction/scrub.js +26 -0
  66. package/dist/compaction/sessionWiring.d.ts +22 -0
  67. package/dist/compaction/sessionWiring.js +26 -0
  68. package/dist/compaction/summarize.d.ts +79 -0
  69. package/dist/compaction/summarize.js +171 -0
  70. package/dist/compaction/toolPolicy.d.ts +3 -0
  71. package/dist/compaction/toolPolicy.js +33 -0
  72. package/dist/compaction/types.d.ts +45 -0
  73. package/dist/compaction/types.js +7 -0
  74. package/dist/compaction/workingState.d.ts +56 -0
  75. package/dist/compaction/workingState.js +198 -0
  76. package/dist/config/atomicWrite.d.ts +15 -0
  77. package/dist/config/atomicWrite.js +43 -0
  78. package/dist/config/browseCatalog.d.ts +2 -0
  79. package/dist/config/browseCatalog.js +4 -2
  80. package/dist/config/createPipeline.js +2 -0
  81. package/dist/config/createTrigger.js +7 -1
  82. package/dist/config/csv.d.ts +9 -0
  83. package/dist/config/csv.js +81 -0
  84. package/dist/config/datasetAgentCheck.d.ts +18 -0
  85. package/dist/config/datasetAgentCheck.js +109 -0
  86. package/dist/config/datasetCandidates.d.ts +22 -0
  87. package/dist/config/datasetCandidates.js +86 -0
  88. package/dist/config/datasetFromSessions.d.ts +34 -0
  89. package/dist/config/datasetFromSessions.js +105 -0
  90. package/dist/config/datasetHash.d.ts +1 -0
  91. package/dist/config/datasetHash.js +4 -0
  92. package/dist/config/datasetImport.d.ts +21 -0
  93. package/dist/config/datasetImport.js +120 -0
  94. package/dist/config/datasetStageCheck.d.ts +9 -0
  95. package/dist/config/datasetStageCheck.js +17 -0
  96. package/dist/config/datasetTypes.d.ts +27 -0
  97. package/dist/config/datasetTypes.js +1 -0
  98. package/dist/config/ensureCatalogScanRoot.d.ts +4 -1
  99. package/dist/config/ensureCatalogScanRoot.js +8 -6
  100. package/dist/config/evalCatalog.d.ts +22 -0
  101. package/dist/config/evalCatalog.js +95 -0
  102. package/dist/config/evalWriteRoots.d.ts +17 -0
  103. package/dist/config/evalWriteRoots.js +109 -0
  104. package/dist/config/githubTriggerAction.d.ts +8 -0
  105. package/dist/config/githubTriggerAction.js +29 -0
  106. package/dist/config/hostConfig.js +3 -0
  107. package/dist/config/installEvalSample.d.ts +34 -0
  108. package/dist/config/installEvalSample.js +182 -0
  109. package/dist/config/loadDataset.d.ts +16 -0
  110. package/dist/config/loadDataset.js +318 -0
  111. package/dist/config/loadEval.d.ts +32 -0
  112. package/dist/config/loadEval.js +1103 -0
  113. package/dist/config/loadOutcome.d.ts +5 -0
  114. package/dist/config/loadStageflowManifest.js +64 -0
  115. package/dist/config/pipelineNeeds.d.ts +2 -0
  116. package/dist/config/pipelineNeeds.js +16 -3
  117. package/dist/config/pipelineRoute.js +5 -4
  118. package/dist/config/resolveAgentTarget.d.ts +32 -0
  119. package/dist/config/resolveAgentTarget.js +118 -0
  120. package/dist/config/resolveEvalTarget.d.ts +21 -0
  121. package/dist/config/resolveEvalTarget.js +89 -0
  122. package/dist/config/scanCatalogPaths.d.ts +1 -1
  123. package/dist/config/scanCatalogPaths.js +17 -5
  124. package/dist/config/validateCatalog.d.ts +2 -2
  125. package/dist/config/validateCatalog.js +21 -1
  126. package/dist/config/validateEval.d.ts +36 -0
  127. package/dist/config/validateEval.js +446 -0
  128. package/dist/config/writeDataset.d.ts +54 -0
  129. package/dist/config/writeDataset.js +186 -0
  130. package/dist/config/writeEval.d.ts +25 -0
  131. package/dist/config/writeEval.js +398 -0
  132. package/dist/diagnostics/checks.js +1 -1
  133. package/dist/envelope/assertEmittedArtifacts.d.ts +12 -0
  134. package/dist/envelope/assertEmittedArtifacts.js +43 -0
  135. package/dist/envelope/preEmitChecks.d.ts +3 -0
  136. package/dist/evals/agentDiff.d.ts +13 -0
  137. package/dist/evals/agentDiff.js +11 -0
  138. package/dist/evals/canonical.d.ts +3 -0
  139. package/dist/evals/canonical.js +24 -0
  140. package/dist/evals/caseTask.d.ts +3 -0
  141. package/dist/evals/caseTask.js +9 -0
  142. package/dist/evals/compare.d.ts +94 -0
  143. package/dist/evals/compare.js +201 -0
  144. package/dist/evals/constants.d.ts +62 -0
  145. package/dist/evals/constants.js +97 -0
  146. package/dist/evals/engine/FakeEngine.d.ts +33 -0
  147. package/dist/evals/engine/FakeEngine.js +177 -0
  148. package/dist/evals/engine/index.d.ts +9 -0
  149. package/dist/evals/engine/index.js +19 -0
  150. package/dist/evals/engine/promptfoo/PromptfooEngine.d.ts +52 -0
  151. package/dist/evals/engine/promptfoo/PromptfooEngine.js +312 -0
  152. package/dist/evals/engine/promptfoo/errors.d.ts +5 -0
  153. package/dist/evals/engine/promptfoo/errors.js +8 -0
  154. package/dist/evals/engine/promptfoo/generate.d.ts +30 -0
  155. package/dist/evals/engine/promptfoo/generate.js +345 -0
  156. package/dist/evals/engine/promptfoo/guard.d.ts +10 -0
  157. package/dist/evals/engine/promptfoo/guard.js +116 -0
  158. package/dist/evals/engine/promptfoo/install.d.ts +17 -0
  159. package/dist/evals/engine/promptfoo/install.js +73 -0
  160. package/dist/evals/engine/promptfoo/judgeProviders.d.ts +2 -0
  161. package/dist/evals/engine/promptfoo/judgeProviders.js +9 -0
  162. package/dist/evals/engine/promptfoo/judgeServices.d.ts +6 -0
  163. package/dist/evals/engine/promptfoo/judgeServices.js +41 -0
  164. package/dist/evals/engine/promptfoo/locate.d.ts +21 -0
  165. package/dist/evals/engine/promptfoo/locate.js +68 -0
  166. package/dist/evals/engine/promptfoo/nodeVersion.d.ts +7 -0
  167. package/dist/evals/engine/promptfoo/nodeVersion.js +14 -0
  168. package/dist/evals/engine/promptfoo/normalize.d.ts +66 -0
  169. package/dist/evals/engine/promptfoo/normalize.js +261 -0
  170. package/dist/evals/engine/promptfoo/pinned.d.ts +5 -0
  171. package/dist/evals/engine/promptfoo/pinned.js +5 -0
  172. package/dist/evals/engine/promptfoo/rubricPrompt.d.ts +14 -0
  173. package/dist/evals/engine/promptfoo/rubricPrompt.js +105 -0
  174. package/dist/evals/engine/promptfoo/runTokens.d.ts +18 -0
  175. package/dist/evals/engine/promptfoo/runTokens.js +31 -0
  176. package/dist/evals/engine/promptfoo/status.d.ts +13 -0
  177. package/dist/evals/engine/promptfoo/status.js +16 -0
  178. package/dist/evals/estimate.d.ts +56 -0
  179. package/dist/evals/estimate.js +66 -0
  180. package/dist/evals/graders/codeGrader.d.ts +6 -0
  181. package/dist/evals/graders/codeGrader.js +85 -0
  182. package/dist/evals/graders/conversationCheckGrader.d.ts +6 -0
  183. package/dist/evals/graders/conversationCheckGrader.js +62 -0
  184. package/dist/evals/graders/fieldGrader.d.ts +11 -0
  185. package/dist/evals/graders/fieldGrader.js +117 -0
  186. package/dist/evals/graders/goal.d.ts +3 -0
  187. package/dist/evals/graders/goal.js +14 -0
  188. package/dist/evals/graders/index.d.ts +18 -0
  189. package/dist/evals/graders/index.js +34 -0
  190. package/dist/evals/graders/notApplicable.d.ts +5 -0
  191. package/dist/evals/graders/notApplicable.js +9 -0
  192. package/dist/evals/graders/path.d.ts +26 -0
  193. package/dist/evals/graders/path.js +75 -0
  194. package/dist/evals/graders/referenceGrader.d.ts +5 -0
  195. package/dist/evals/graders/referenceGrader.js +59 -0
  196. package/dist/evals/graders/schemaGrader.d.ts +5 -0
  197. package/dist/evals/graders/schemaGrader.js +29 -0
  198. package/dist/evals/graders/subject.d.ts +6 -0
  199. package/dist/evals/graders/subject.js +33 -0
  200. package/dist/evals/graders/toolCallsGrader.d.ts +6 -0
  201. package/dist/evals/graders/toolCallsGrader.js +100 -0
  202. package/dist/evals/graders/transcript.d.ts +12 -0
  203. package/dist/evals/graders/transcript.js +57 -0
  204. package/dist/evals/graders/types.d.ts +20 -0
  205. package/dist/evals/graders/types.js +1 -0
  206. package/dist/evals/judgeProviders.d.ts +16 -0
  207. package/dist/evals/judgeProviders.js +27 -0
  208. package/dist/evals/lineDiff.d.ts +5 -0
  209. package/dist/evals/lineDiff.js +34 -0
  210. package/dist/evals/listStatus.d.ts +31 -0
  211. package/dist/evals/listStatus.js +48 -0
  212. package/dist/evals/manager.d.ts +187 -0
  213. package/dist/evals/manager.js +795 -0
  214. package/dist/evals/plan.d.ts +65 -0
  215. package/dist/evals/plan.js +621 -0
  216. package/dist/evals/planInternalAgent.d.ts +19 -0
  217. package/dist/evals/planInternalAgent.js +205 -0
  218. package/dist/evals/providerKeys.d.ts +1 -0
  219. package/dist/evals/providerKeys.js +21 -0
  220. package/dist/evals/read/agentTargets.d.ts +37 -0
  221. package/dist/evals/read/agentTargets.js +101 -0
  222. package/dist/evals/read/api.d.ts +208 -0
  223. package/dist/evals/read/api.js +363 -0
  224. package/dist/evals/read/compare.d.ts +4 -0
  225. package/dist/evals/read/compare.js +39 -0
  226. package/dist/evals/read/datasets.d.ts +105 -0
  227. package/dist/evals/read/datasets.js +318 -0
  228. package/dist/evals/read/list.d.ts +66 -0
  229. package/dist/evals/read/list.js +182 -0
  230. package/dist/evals/read/results.d.ts +46 -0
  231. package/dist/evals/read/results.js +107 -0
  232. package/dist/evals/read/runs.d.ts +154 -0
  233. package/dist/evals/read/runs.js +295 -0
  234. package/dist/evals/read/suggestions.d.ts +43 -0
  235. package/dist/evals/read/suggestions.js +73 -0
  236. package/dist/evals/runner/FakeAgentBackend.d.ts +40 -0
  237. package/dist/evals/runner/FakeAgentBackend.js +82 -0
  238. package/dist/evals/runner/agentCaseRunner.d.ts +24 -0
  239. package/dist/evals/runner/agentCaseRunner.js +388 -0
  240. package/dist/evals/runner/caseRunner.d.ts +8 -0
  241. package/dist/evals/runner/caseRunner.js +28 -0
  242. package/dist/evals/runner/conversationSubject.d.ts +18 -0
  243. package/dist/evals/runner/conversationSubject.js +123 -0
  244. package/dist/evals/runner/internalAgentCaseRunner.d.ts +6 -0
  245. package/dist/evals/runner/internalAgentCaseRunner.js +250 -0
  246. package/dist/evals/runner/simulatorModel.d.ts +51 -0
  247. package/dist/evals/runner/simulatorModel.js +58 -0
  248. package/dist/evals/runner/simulatorPrompt.d.ts +29 -0
  249. package/dist/evals/runner/simulatorPrompt.js +59 -0
  250. package/dist/evals/runner/stageCaseRunner.d.ts +30 -0
  251. package/dist/evals/runner/stageCaseRunner.js +279 -0
  252. package/dist/evals/runner/userDriver.d.ts +71 -0
  253. package/dist/evals/runner/userDriver.js +104 -0
  254. package/dist/evals/scoring.d.ts +34 -0
  255. package/dist/evals/scoring.js +67 -0
  256. package/dist/evals/snapshot.d.ts +191 -0
  257. package/dist/evals/snapshot.js +70 -0
  258. package/dist/evals/stageDiff.d.ts +8 -0
  259. package/dist/evals/stageDiff.js +48 -0
  260. package/dist/evals/stats.d.ts +95 -0
  261. package/dist/evals/stats.js +220 -0
  262. package/dist/evals/statsAdapter.d.ts +4 -0
  263. package/dist/evals/statsAdapter.js +30 -0
  264. package/dist/evals/statsTypes.d.ts +51 -0
  265. package/dist/evals/statsTypes.js +18 -0
  266. package/dist/evals/storeError.d.ts +5 -0
  267. package/dist/evals/storeError.js +8 -0
  268. package/dist/evals/storeTypes.d.ts +207 -0
  269. package/dist/evals/storeTypes.js +1 -0
  270. package/dist/evals/tTable.d.ts +3 -0
  271. package/dist/evals/tTable.js +15 -0
  272. package/dist/evals/triage/constants.d.ts +26 -0
  273. package/dist/evals/triage/constants.js +33 -0
  274. package/dist/evals/triage/credentials.d.ts +13 -0
  275. package/dist/evals/triage/credentials.js +20 -0
  276. package/dist/evals/triage/jevClassifier.d.ts +14 -0
  277. package/dist/evals/triage/jevClassifier.js +124 -0
  278. package/dist/evals/triage/judgeClassifier.d.ts +10 -0
  279. package/dist/evals/triage/judgeClassifier.js +76 -0
  280. package/dist/evals/triage/prompts.d.ts +16 -0
  281. package/dist/evals/triage/prompts.js +64 -0
  282. package/dist/evals/triage/service.d.ts +79 -0
  283. package/dist/evals/triage/service.js +217 -0
  284. package/dist/evals/triage/stateBuilder.d.ts +19 -0
  285. package/dist/evals/triage/stateBuilder.js +125 -0
  286. package/dist/evals/triage/types.d.ts +60 -0
  287. package/dist/evals/triage/types.js +8 -0
  288. package/dist/evals/types.d.ts +357 -0
  289. package/dist/evals/types.js +4 -0
  290. package/dist/guide/cardContract.d.ts +133 -0
  291. package/dist/guide/cardContract.js +156 -0
  292. package/dist/guide/chat.d.ts +45 -0
  293. package/dist/guide/chat.js +378 -0
  294. package/dist/guide/knowledge/index.d.ts +14 -0
  295. package/dist/guide/knowledge/index.js +109 -0
  296. package/dist/guide/knowledge/topics/agents.d.ts +2 -0
  297. package/dist/guide/knowledge/topics/agents.js +58 -0
  298. package/dist/guide/knowledge/topics/bestPractices.d.ts +2 -0
  299. package/dist/guide/knowledge/topics/bestPractices.js +67 -0
  300. package/dist/guide/knowledge/topics/connections.d.ts +2 -0
  301. package/dist/guide/knowledge/topics/connections.js +66 -0
  302. package/dist/guide/knowledge/topics/consoleMap.d.ts +2 -0
  303. package/dist/guide/knowledge/topics/consoleMap.js +55 -0
  304. package/dist/guide/knowledge/topics/evals.d.ts +2 -0
  305. package/dist/guide/knowledge/topics/evals.js +72 -0
  306. package/dist/guide/knowledge/topics/limitations.d.ts +2 -0
  307. package/dist/guide/knowledge/topics/limitations.js +84 -0
  308. package/dist/guide/knowledge/topics/mcpSkillsBrowser.d.ts +2 -0
  309. package/dist/guide/knowledge/topics/mcpSkillsBrowser.js +57 -0
  310. package/dist/guide/knowledge/topics/memories.d.ts +2 -0
  311. package/dist/guide/knowledge/topics/memories.js +63 -0
  312. package/dist/guide/knowledge/topics/modelsAndProviders.d.ts +2 -0
  313. package/dist/guide/knowledge/topics/modelsAndProviders.js +56 -0
  314. package/dist/guide/knowledge/topics/overview.d.ts +2 -0
  315. package/dist/guide/knowledge/topics/overview.js +56 -0
  316. package/dist/guide/knowledge/topics/pipelines.d.ts +2 -0
  317. package/dist/guide/knowledge/topics/pipelines.js +93 -0
  318. package/dist/guide/knowledge/topics/stages.d.ts +2 -0
  319. package/dist/guide/knowledge/topics/stages.js +54 -0
  320. package/dist/guide/knowledge/topics/tasks.d.ts +2 -0
  321. package/dist/guide/knowledge/topics/tasks.js +52 -0
  322. package/dist/guide/knowledge/topics/triggers.d.ts +2 -0
  323. package/dist/guide/knowledge/topics/triggers.js +68 -0
  324. package/dist/guide/knowledge/topics/troubleshooting.d.ts +2 -0
  325. package/dist/guide/knowledge/topics/troubleshooting.js +85 -0
  326. package/dist/guide/knowledgeTools.d.ts +2 -0
  327. package/dist/guide/knowledgeTools.js +56 -0
  328. package/dist/guide/showUi.d.ts +6 -0
  329. package/dist/guide/showUi.js +546 -0
  330. package/dist/guide/threadStore.d.ts +48 -0
  331. package/dist/guide/threadStore.js +189 -0
  332. package/dist/guide/writeRoot.d.ts +11 -0
  333. package/dist/guide/writeRoot.js +70 -0
  334. package/dist/hosted/translate.js +2 -0
  335. package/dist/mcp/agentTools.js +12 -0
  336. package/dist/mcp/catalogTools.js +18 -20
  337. package/dist/mcp/catalogWriteTools.d.ts +8 -0
  338. package/dist/mcp/catalogWriteTools.js +225 -0
  339. package/dist/mcp/connectionTools.d.ts +3 -0
  340. package/dist/mcp/connectionTools.js +85 -0
  341. package/dist/mcp/controlTools.js +34 -1
  342. package/dist/mcp/deps.d.ts +16 -0
  343. package/dist/mcp/evalTools.d.ts +3 -0
  344. package/dist/mcp/evalTools.js +943 -0
  345. package/dist/mcp/memoryHostTools.d.ts +3 -0
  346. package/dist/mcp/memoryHostTools.js +72 -0
  347. package/dist/mcp/pipelineAuthorTools.js +14 -1
  348. package/dist/mcp/server.js +6 -0
  349. package/dist/mcp/tools.js +8 -0
  350. package/dist/mcp/triggerTools.js +77 -31
  351. package/dist/models/response.d.ts +7 -2
  352. package/dist/models/response.js +6 -1
  353. package/dist/operatorAgent/agentAuthorContext.d.ts +1 -1
  354. package/dist/operatorAgent/agentAuthorContext.js +1 -1
  355. package/dist/operatorAgent/evalGate.d.ts +22 -0
  356. package/dist/operatorAgent/evalGate.js +77 -0
  357. package/dist/operatorAgent/fakeHost.js +10 -2
  358. package/dist/operatorAgent/host.js +18 -2
  359. package/dist/operatorAgent/index.d.ts +1 -0
  360. package/dist/operatorAgent/index.js +1 -0
  361. package/dist/operatorAgent/internalAgents.d.ts +11 -0
  362. package/dist/operatorAgent/internalAgents.js +13 -0
  363. package/dist/operatorAgent/piModel.d.ts +5 -1
  364. package/dist/operatorAgent/piModel.js +32 -14
  365. package/dist/operatorAgent/profiles/agentAuthor.d.ts +3 -1
  366. package/dist/operatorAgent/profiles/agentAuthor.js +201 -32
  367. package/dist/operatorAgent/profiles/agentAuthorPlaybook.d.ts +1 -1
  368. package/dist/operatorAgent/profiles/agentAuthorPlaybook.js +48 -15
  369. package/dist/operatorAgent/profiles/authorReadable.d.ts +2 -0
  370. package/dist/operatorAgent/profiles/authorReadable.js +26 -0
  371. package/dist/operatorAgent/profiles/guideConfirm.d.ts +21 -0
  372. package/dist/operatorAgent/profiles/guideConfirm.js +100 -0
  373. package/dist/operatorAgent/profiles/guideTools.d.ts +7 -0
  374. package/dist/operatorAgent/profiles/guideTools.js +123 -0
  375. package/dist/operatorAgent/profiles/stageflowGuide.d.ts +7 -0
  376. package/dist/operatorAgent/profiles/stageflowGuide.js +132 -0
  377. package/dist/operatorAgent/profiles/workshopAuthor.js +23 -0
  378. package/dist/operatorAgent/profiles/workshopAuthorPlaybook.d.ts +1 -1
  379. package/dist/operatorAgent/profiles/workshopAuthorPlaybook.js +9 -1
  380. package/dist/operatorAgent/profiles/workshopAuthorTools.d.ts +2 -0
  381. package/dist/operatorAgent/profiles/workshopAuthorTools.js +197 -9
  382. package/dist/operatorAgent/types.d.ts +23 -0
  383. package/dist/package-meta.d.ts +1 -1
  384. package/dist/package-meta.js +1 -1
  385. package/dist/runstore/port.d.ts +43 -0
  386. package/dist/runstore/retention.d.ts +2 -1
  387. package/dist/runstore/retention.js +3 -7
  388. package/dist/runstore/runStatusClock.d.ts +17 -0
  389. package/dist/runstore/runStatusClock.js +19 -0
  390. package/dist/runstore/sqlite/SqliteRunStore.d.ts +29 -0
  391. package/dist/runstore/sqlite/SqliteRunStore.js +144 -50
  392. package/dist/runstore/sqlite/evalStore.d.ts +20 -0
  393. package/dist/runstore/sqlite/evalStore.js +541 -0
  394. package/dist/runstore/sqlite/migrations/015-evals.d.ts +12 -0
  395. package/dist/runstore/sqlite/migrations/015-evals.js +157 -0
  396. package/dist/runstore/sqlite/migrations/016-eval-subject-ref.d.ts +12 -0
  397. package/dist/runstore/sqlite/migrations/016-eval-subject-ref.js +20 -0
  398. package/dist/runstore/sqlite/migrations/017-eval-result-suggestions.d.ts +11 -0
  399. package/dist/runstore/sqlite/migrations/017-eval-result-suggestions.js +36 -0
  400. package/dist/runstore/sqlite/migrations/index.js +6 -0
  401. package/dist/runstore/sqlite/suggestionStore.d.ts +14 -0
  402. package/dist/runstore/sqlite/suggestionStore.js +132 -0
  403. package/dist/runstore/workspaceLayout.js +6 -1
  404. package/dist/runtime/artifactDiskInspect.d.ts +14 -0
  405. package/dist/runtime/artifactDiskInspect.js +52 -0
  406. package/dist/runtime/artifactPath.d.ts +24 -0
  407. package/dist/runtime/artifactPath.js +84 -0
  408. package/dist/runtime/completionCheckRunner.js +15 -43
  409. package/dist/runtime/githubPollSource.d.ts +22 -4
  410. package/dist/runtime/githubPollSource.js +93 -8
  411. package/dist/runtime/pipelineRunner.d.ts +2 -1
  412. package/dist/runtime/pipelineRunner.js +2 -0
  413. package/dist/runtime/priorArtifactGrants.d.ts +8 -0
  414. package/dist/runtime/priorArtifactGrants.js +105 -0
  415. package/dist/runtime/runActionEligibility.d.ts +46 -0
  416. package/dist/runtime/runActionEligibility.js +123 -0
  417. package/dist/runtime/runManager.d.ts +6 -2
  418. package/dist/runtime/runManager.js +60 -16
  419. package/dist/runtime/runRetentionSweep.js +2 -2
  420. package/dist/runtime/runRetryCoordinator.d.ts +2 -14
  421. package/dist/runtime/runRetryCoordinator.js +3 -67
  422. package/dist/runtime/sandboxContainer.d.ts +7 -0
  423. package/dist/runtime/sandboxContainer.js +48 -6
  424. package/dist/runtime/sandboxDesk.d.ts +2 -1
  425. package/dist/runtime/sandboxDesk.js +11 -12
  426. package/dist/runtime/stageEnvironment.js +1 -0
  427. package/dist/runtime/stageRoots.d.ts +12 -0
  428. package/dist/runtime/stageSandbox.d.ts +2 -1
  429. package/dist/runtime/stageSandbox.js +18 -2
  430. package/dist/runtime/stageWorker.js +20 -0
  431. package/dist/server/agentAuthorRoutes.js +59 -1
  432. package/dist/server/bootstrap.d.ts +13 -0
  433. package/dist/server/bootstrap.js +69 -0
  434. package/dist/server/controlToken.js +3 -0
  435. package/dist/server/evalInternalRoutes.d.ts +14 -0
  436. package/dist/server/evalInternalRoutes.js +87 -0
  437. package/dist/server/evalRouteContext.d.ts +46 -0
  438. package/dist/server/evalRouteContext.js +43 -0
  439. package/dist/server/evalRoutes.d.ts +5 -0
  440. package/dist/server/evalRoutes.js +1456 -0
  441. package/dist/server/evalTriageRoutes.d.ts +2 -0
  442. package/dist/server/evalTriageRoutes.js +95 -0
  443. package/dist/server/guideRoutes.d.ts +5 -0
  444. package/dist/server/guideRoutes.js +166 -0
  445. package/dist/server/http.d.ts +15 -0
  446. package/dist/server/http.js +67 -3
  447. package/dist/server/internalAgentRoutes.d.ts +6 -0
  448. package/dist/server/internalAgentRoutes.js +12 -0
  449. package/dist/server/mcpHost.js +5 -0
  450. package/dist/toolkit/meta.d.ts +1 -1
  451. package/dist/toolkit/meta.js +53 -0
  452. package/dist/toolkit/profiles.d.ts +5 -0
  453. package/dist/toolkit/profiles.js +13 -1
  454. package/dist/tools/askOperator.js +2 -1
  455. package/dist/tools/emitStageEnvelope.d.ts +17 -16
  456. package/dist/tools/emitStageEnvelope.js +46 -32
  457. package/dist/tools/stageToolNames.d.ts +4 -0
  458. package/dist/tools/stageToolNames.js +4 -0
  459. package/dist/tools/workingState.d.ts +40 -0
  460. package/dist/tools/workingState.js +108 -0
  461. package/dist/tools/writeStageArtifact.d.ts +0 -9
  462. package/dist/tools/writeStageArtifact.js +2 -1
  463. package/dist/types/stageflowManifest.d.ts +6 -0
  464. package/dist/ui/assets/{MarkdownText-D7tYaIFO.js → MarkdownText-S5C_vemc.js} +2 -2
  465. package/dist/ui/assets/adapters-C0l4ruiR.js +17 -0
  466. package/dist/ui/assets/chat-CjnsmvqR.js +2 -0
  467. package/dist/ui/assets/chat-DFDd4UjS.css +2 -0
  468. package/dist/ui/assets/{main-BFo79krC.css → main-Del1GeLf.css} +1 -1
  469. package/dist/ui/assets/main-_yXVqcW9.js +277 -0
  470. package/dist/ui/chat.html +3 -4
  471. package/dist/ui/index.html +3 -4
  472. package/dist/workshop/chatTurn.js +1 -1
  473. package/dist/workshop/workshopPin.d.ts +3 -3
  474. package/dist/workshop/workshopPin.js +16 -15
  475. package/examples/README.md +2 -0
  476. package/examples/compaction-stress/README.md +30 -0
  477. package/examples/compaction-stress/check-run.mjs +53 -0
  478. package/examples/compaction-stress/collect-evidence.yaml +60 -0
  479. package/examples/compaction-stress/compaction-stress.pipeline.yaml +141 -0
  480. package/examples/compaction-stress/compaction-stress.task.yaml +15 -0
  481. package/examples/compaction-stress/generate-evidence.mjs +110 -0
  482. package/examples/compaction-stress/root-cause.yaml +26 -0
  483. package/examples/compaction-stress/stageflow.yaml +9 -0
  484. package/examples/compaction-stress/write-postmortem.yaml +40 -0
  485. package/examples/document-policy-review/README.md +35 -0
  486. package/examples/document-policy-review/agents/compliance.agent.yaml +42 -0
  487. package/examples/document-policy-review/documents/acme-vendor-sow.md +15 -0
  488. package/examples/document-policy-review/documents/support-runbook.md +13 -0
  489. package/examples/document-policy-review/policies/acceptable-use.md +13 -0
  490. package/examples/document-policy-review/policies/data-handling.md +17 -0
  491. package/examples/document-policy-review/policies/vendor-security.md +17 -0
  492. package/examples/evals/README.md +141 -0
  493. package/examples/evals/critique-sample.dataset.jsonl +8 -0
  494. package/examples/evals/critique-sample.eval.yaml +34 -0
  495. package/examples/evals/critique-sample.task.yaml +4 -0
  496. package/examples/evals/eval-sample.pipeline.yaml +35 -0
  497. package/examples/evals/guide-behavior.dataset.jsonl +15 -0
  498. package/examples/evals/guide-behavior.eval.yaml +35 -0
  499. package/examples/evals/guide-depth.dataset.jsonl +35 -0
  500. package/examples/evals/guide-depth.eval.yaml +169 -0
  501. package/examples/evals/support-sample.agent.yaml +20 -0
  502. package/examples/evals/support-sample.dataset.jsonl +6 -0
  503. package/examples/evals/support-sample.eval.yaml +38 -0
  504. package/examples/memory/evals/guide-behavior.dataset.jsonl +15 -0
  505. package/examples/memory/evals/guide-behavior.eval.yaml +35 -0
  506. package/examples/memory/evals/guide-depth.dataset.jsonl +35 -0
  507. package/examples/memory/evals/guide-depth.eval.yaml +169 -0
  508. package/examples/stageflow.yaml +6 -0
  509. package/package.json +1 -1
  510. package/skills/stageflow/SKILL.md +5 -2
  511. package/skills/stageflow/references/control-surface.md +4 -2
  512. package/skills/stageflow/references/evals.md +68 -0
  513. package/skills/stageflow-author/SKILL.md +4 -0
  514. package/skills/stageflow-run/SKILL.md +4 -0
  515. package/dist/ui/assets/AssistantRuntimeProvider-CsiPFhCT.js +0 -9
  516. package/dist/ui/assets/ReadonlyStore-BfXqHykm.js +0 -6
  517. package/dist/ui/assets/chat-BbLukq9W.js +0 -5
  518. package/dist/ui/assets/chat-zIZU_bP1.css +0 -2
  519. package/dist/ui/assets/main-CijYMeM0.js +0 -259
@@ -0,0 +1,59 @@
1
+ import { isDeepStrictEqual } from "node:util";
2
+ import { resolveValue } from "./path.js";
3
+ import { show, subjectRoot, truncate } from "./subject.js";
4
+ const norm = (s) => s.replace(/\s+/g, " ").trim().toLowerCase();
5
+ function toText(value) {
6
+ if (typeof value === "string")
7
+ return value;
8
+ return JSON.stringify(value) ?? "";
9
+ }
10
+ export function gradeReference(grader, subject, testCase, _context) {
11
+ const base = { graderId: grader.id, graderType: "reference" };
12
+ const missing = (reason) => ({
13
+ ...base,
14
+ status: "error",
15
+ errorCode: "EVAL_CASE_MISSING_EXPECTED",
16
+ reason,
17
+ });
18
+ const expectedRef = resolveValue(testCase, grader.expected);
19
+ if (!expectedRef.found || expectedRef.value === undefined || expectedRef.value === null) {
20
+ return missing(`case has no value at ${grader.expected}`);
21
+ }
22
+ const actual = resolveValue(subjectRoot(subject), grader.path);
23
+ if (grader.mode === "equals") {
24
+ const passed = actual.found && isDeepStrictEqual(actual.value, expectedRef.value);
25
+ return {
26
+ ...base,
27
+ status: "graded",
28
+ passed,
29
+ score: passed ? 1 : 0,
30
+ rawValue: passed ? 1 : 0,
31
+ ...(passed
32
+ ? {}
33
+ : {
34
+ reason: truncate(`${grader.path} is ${actual.found ? show(actual.value) : "missing"}, expected ${show(expectedRef.value)}`),
35
+ }),
36
+ };
37
+ }
38
+ const expected = expectedRef.value;
39
+ if (!Array.isArray(expected) || !expected.every((e) => typeof e === "string")) {
40
+ return missing(`${grader.expected} must be an array of strings`);
41
+ }
42
+ const text = actual.found ? norm(toText(actual.value)) : "";
43
+ const items = expected.map((t) => ({ text: t, found: text.includes(norm(t)) }));
44
+ const foundCount = items.filter((i) => i.found).length;
45
+ const score = items.length === 0 ? 1 : foundCount / items.length;
46
+ const passed = score + 1e-9 >= grader.passAt;
47
+ const firstMissing = items.find((i) => !i.found);
48
+ return {
49
+ ...base,
50
+ status: "graded",
51
+ passed,
52
+ score,
53
+ rawValue: score,
54
+ ...(firstMissing
55
+ ? { reason: truncate(`found ${foundCount} of ${items.length}; missing ${show(firstMissing.text)}`) }
56
+ : {}),
57
+ details: { items },
58
+ };
59
+ }
@@ -0,0 +1,5 @@
1
+ import type { ResolvedGrader } from "../types.js";
2
+ import type { GradeData, GraderContext, Subject } from "./types.js";
3
+ export declare function gradeSchema(grader: Extract<ResolvedGrader, {
4
+ type: "schema";
5
+ }>, subject: Subject, _testCase: unknown, context: GraderContext): GradeData;
@@ -0,0 +1,29 @@
1
+ import { payloadInstanceMismatch } from "../../envelope/payloadSchema.js";
2
+ import { notApplicable } from "./notApplicable.js";
3
+ import { isConversation, truncate } from "./subject.js";
4
+ export function gradeSchema(grader, subject, _testCase, context) {
5
+ if (isConversation(subject))
6
+ return notApplicable(grader, "conversation");
7
+ const base = { graderId: grader.id, graderType: "schema" };
8
+ const fail = (reason) => ({
9
+ ...base,
10
+ status: "graded",
11
+ passed: false,
12
+ score: 0,
13
+ rawValue: 0,
14
+ reason: truncate(reason, 500),
15
+ });
16
+ if (subject.status !== "success") {
17
+ return fail(`envelope status is ${subject.status}, expected success`);
18
+ }
19
+ let mismatch;
20
+ try {
21
+ mismatch = payloadInstanceMismatch(subject.payload ?? {}, context.outputSchema);
22
+ }
23
+ catch (error) {
24
+ return fail(error instanceof Error ? error.message : String(error));
25
+ }
26
+ if (mismatch !== undefined)
27
+ return fail(mismatch);
28
+ return { ...base, status: "graded", passed: true, score: 1, rawValue: 1 };
29
+ }
@@ -0,0 +1,6 @@
1
+ import type { ConversationSubject, EnvelopeSubject, Subject } from "./types.js";
2
+ export declare function isConversation(subject: Subject): subject is ConversationSubject;
3
+ export declare function subjectEnvelope(subject: EnvelopeSubject): Record<string, unknown>;
4
+ export declare function subjectRoot(subject: Subject): Record<string, unknown>;
5
+ export declare function truncate(text: string, max?: number): string;
6
+ export declare function show(value: unknown): string;
@@ -0,0 +1,33 @@
1
+ export function isConversation(subject) {
2
+ return subject.kind === "conversation";
3
+ }
4
+ export function subjectEnvelope(subject) {
5
+ const env = {
6
+ status: subject.status,
7
+ summary: subject.summary,
8
+ artifacts: subject.artifacts,
9
+ };
10
+ if (subject.payload !== undefined)
11
+ env.payload = subject.payload;
12
+ return env;
13
+ }
14
+ export function subjectRoot(subject) {
15
+ if (isConversation(subject)) {
16
+ const { transcript: _transport, ...root } = subject;
17
+ return root;
18
+ }
19
+ return subjectEnvelope(subject);
20
+ }
21
+ export function truncate(text, max = 200) {
22
+ return text.length > max ? `${text.slice(0, max - 1)}…` : text;
23
+ }
24
+ export function show(value) {
25
+ if (value === undefined)
26
+ return "undefined";
27
+ try {
28
+ return JSON.stringify(value) ?? String(value);
29
+ }
30
+ catch {
31
+ return String(value);
32
+ }
33
+ }
@@ -0,0 +1,6 @@
1
+ import type { ResolvedGrader } from "../types.js";
2
+ import type { GradeCase, GradeData, GraderContext, Subject } from "./types.js";
3
+ export declare function toolNameMatches(pattern: string, name: string): boolean;
4
+ export declare function gradeToolCalls(grader: Extract<ResolvedGrader, {
5
+ type: "tool_calls";
6
+ }>, subject: Subject, testCase: GradeCase, _context?: GraderContext): GradeData;
@@ -0,0 +1,100 @@
1
+ import { describeCheck, runCheck } from "./fieldGrader.js";
2
+ import { notApplicable } from "./notApplicable.js";
3
+ import { isConversation, truncate } from "./subject.js";
4
+ function isRecord(v) {
5
+ return typeof v === "object" && v !== null && !Array.isArray(v);
6
+ }
7
+ function expectedList(testCase, key) {
8
+ const expected = testCase.expected;
9
+ if (!isRecord(expected))
10
+ return undefined;
11
+ const list = expected[key];
12
+ return Array.isArray(list) && list.every((x) => typeof x === "string") ? list : undefined;
13
+ }
14
+ export function toolNameMatches(pattern, name) {
15
+ if (pattern.endsWith(".*")) {
16
+ const prefix = pattern.slice(0, -2);
17
+ return name === prefix || name.startsWith(`${prefix}.`);
18
+ }
19
+ return name === pattern;
20
+ }
21
+ function plural(n, word) {
22
+ return `${n} ${word}${n === 1 ? "" : "s"}`;
23
+ }
24
+ export function gradeToolCalls(grader, subject, testCase, _context) {
25
+ if (!isConversation(subject))
26
+ return notApplicable(grader, "envelope");
27
+ const mustCall = grader.mustCall ?? expectedList(testCase, "tools");
28
+ const mustNotCall = grader.mustNotCall ?? expectedList(testCase, "forbidden_tools");
29
+ const { mustAskFirst, maxCalls } = grader;
30
+ if (mustCall === undefined &&
31
+ mustNotCall === undefined &&
32
+ mustAskFirst === undefined &&
33
+ maxCalls === undefined &&
34
+ grader.args.length === 0) {
35
+ return {
36
+ graderId: grader.id,
37
+ graderType: "tool_calls",
38
+ status: "error",
39
+ errorCode: "EVAL_CASE_MISSING_EXPECTED",
40
+ reason: "no tool rule: set a rule on the grader or expected.tools / expected.forbidden_tools on the case",
41
+ };
42
+ }
43
+ const calls = subject.tool_calls;
44
+ const named = (pattern) => calls.filter((c) => toolNameMatches(pattern, c.name));
45
+ const checks = [];
46
+ for (const name of mustCall ?? []) {
47
+ const n = named(name).length;
48
+ checks.push({ rule: `must_call ${name}`, passed: n > 0, found: n > 0 ? plural(n, "call") : "not called" });
49
+ }
50
+ for (const name of mustNotCall ?? []) {
51
+ const n = named(name).length;
52
+ checks.push({ rule: `must_not_call ${name}`, passed: n === 0, found: n === 0 ? "not called" : plural(n, "call") });
53
+ }
54
+ for (const name of mustAskFirst ?? []) {
55
+ const matching = named(name);
56
+ const asked = matching.filter((c) => c.policy_outcome === "would_ask_first").length;
57
+ checks.push({
58
+ rule: `must_ask_first ${name}`,
59
+ passed: asked > 0,
60
+ found: asked > 0 ? plural(asked, "approval request") : matching.length > 0 ? "called without asking" : "not called",
61
+ });
62
+ }
63
+ if (maxCalls !== undefined) {
64
+ checks.push({
65
+ rule: `max_calls ${maxCalls}`,
66
+ passed: calls.length <= maxCalls,
67
+ found: plural(calls.length, "call"),
68
+ });
69
+ }
70
+ for (const arg of grader.args) {
71
+ const check = { path: arg.path, op: arg.op, ...(arg.value !== undefined ? { value: arg.value } : {}) };
72
+ const matching = named(arg.tool);
73
+ const rule = `args ${arg.tool} ${describeCheck(check)}`;
74
+ let failure;
75
+ for (const call of matching) {
76
+ const outcome = runCheck(check, call.args);
77
+ if (!outcome.passed) {
78
+ failure = `call ${call.id}: ${outcome.found}`;
79
+ break;
80
+ }
81
+ }
82
+ checks.push({
83
+ rule,
84
+ passed: failure === undefined,
85
+ found: failure ?? (matching.length === 0 ? "no call to check" : plural(matching.length, "call")),
86
+ });
87
+ }
88
+ const first = checks.find((c) => !c.passed);
89
+ const passed = first === undefined;
90
+ return {
91
+ graderId: grader.id,
92
+ graderType: "tool_calls",
93
+ status: "graded",
94
+ passed,
95
+ score: passed ? 1 : 0,
96
+ rawValue: passed ? 1 : 0,
97
+ ...(first ? { reason: truncate(`${first.rule}: ${first.found}`) } : {}),
98
+ details: { checks: checks.map((c) => ({ ...c, found: truncate(c.found) })) },
99
+ };
100
+ }
@@ -0,0 +1,12 @@
1
+ import type { ConversationOutput } from "../types.js";
2
+ export type TranscriptLimits = {
3
+ turns: number;
4
+ messageChars: number;
5
+ toolArgsChars: number;
6
+ toolResultChars: number;
7
+ };
8
+ /**
9
+ * The bounded transcript the judge reads: one line block per message or tool call, tool arguments and results
10
+ * capped, and only the first and last turns kept past the turn limit.
11
+ */
12
+ export declare function renderTranscript(subject: Pick<ConversationOutput, "messages" | "tool_calls">, limits?: TranscriptLimits): string;
@@ -0,0 +1,57 @@
1
+ import { JUDGE_TRANSCRIPT_LIMITS } from "../constants.js";
2
+ const KEEP_FIRST_TURNS = 3;
3
+ function cut(text, max) {
4
+ if (text.length <= max)
5
+ return text;
6
+ return `${text.slice(0, max)}…[truncated ${text.length - max} chars]`;
7
+ }
8
+ function neutralize(text) {
9
+ return text.replace(/<(\/?)(transcript|rubric|examples|case_input|case_expected)\b/gi, "<\\$1$2");
10
+ }
11
+ function json(value) {
12
+ try {
13
+ return JSON.stringify(value ?? {}) ?? "{}";
14
+ }
15
+ catch {
16
+ return "{}";
17
+ }
18
+ }
19
+ /**
20
+ * The bounded transcript the judge reads: one line block per message or tool call, tool arguments and results
21
+ * capped, and only the first and last turns kept past the turn limit.
22
+ */
23
+ export function renderTranscript(subject, limits = JUDGE_TRANSCRIPT_LIMITS) {
24
+ const turns = [
25
+ ...new Set([...subject.messages.map((m) => m.turn), ...subject.tool_calls.map((c) => c.turn)]),
26
+ ].sort((a, b) => a - b);
27
+ let kept = turns;
28
+ let omitted = 0;
29
+ if (turns.length > limits.turns) {
30
+ const tail = Math.max(0, limits.turns - KEEP_FIRST_TURNS);
31
+ omitted = turns.length - limits.turns;
32
+ kept = [...turns.slice(0, KEEP_FIRST_TURNS), ...turns.slice(turns.length - tail)];
33
+ }
34
+ const lines = [];
35
+ kept.forEach((turn, index) => {
36
+ if (omitted > 0 && index === KEEP_FIRST_TURNS)
37
+ lines.push(`[${omitted} turns omitted]`);
38
+ const say = (role) => {
39
+ for (const m of subject.messages) {
40
+ if (m.turn === turn && m.role === role) {
41
+ lines.push(`[turn ${turn}] ${role === "user" ? "USER" : "AGENT"}: ${neutralize(cut(m.text, limits.messageChars))}`);
42
+ }
43
+ }
44
+ };
45
+ say("user");
46
+ for (const call of subject.tool_calls) {
47
+ if (call.turn !== turn)
48
+ continue;
49
+ const outcome = call.status === "error" ? `${call.policy_outcome}, error` : call.policy_outcome;
50
+ const args = neutralize(cut(json(call.args), limits.toolArgsChars));
51
+ const result = neutralize(cut(call.result_summary, limits.toolResultChars));
52
+ lines.push(`[turn ${turn}] TOOL ${call.name} (${outcome}) args=${args} -> ${result}`);
53
+ }
54
+ say("assistant");
55
+ });
56
+ return lines.join("\n");
57
+ }
@@ -0,0 +1,20 @@
1
+ import type { ConversationOutput, NormalizedGrade, PlanCase } from "../types.js";
2
+ export type EnvelopeSubject = {
3
+ kind?: "envelope";
4
+ status: "success" | "failure";
5
+ summary: string;
6
+ payload?: Record<string, unknown>;
7
+ artifacts: string[];
8
+ };
9
+ export type ConversationSubject = ConversationOutput;
10
+ export type Subject = EnvelopeSubject | ConversationSubject;
11
+ export type GraderContext = {
12
+ evalId: string;
13
+ runId: string;
14
+ repeat: number;
15
+ passThreshold: number;
16
+ outputSchema?: unknown;
17
+ codeTimeoutMs?: number;
18
+ };
19
+ export type GradeData = NormalizedGrade;
20
+ export type GradeCase = PlanCase;
@@ -0,0 +1 @@
1
+ export {};
@@ -0,0 +1,16 @@
1
+ export type JudgeMapping = {
2
+ providerSpec: string;
3
+ envVar: string;
4
+ stageflowProvider: string;
5
+ };
6
+ type JudgeProviderRow = {
7
+ stageflowProvider: string;
8
+ envVar: string;
9
+ spec: (model: string) => string;
10
+ };
11
+ export declare const JUDGE_PROVIDERS: readonly JudgeProviderRow[];
12
+ export declare function supportedJudgeProviders(): string[];
13
+ export declare function mapJudge(modelId: string): JudgeMapping | null;
14
+ export declare function judgeCapableModels(modelIds: readonly string[]): string[];
15
+ export declare function unsupportedJudgeMessage(modelId: string): string;
16
+ export {};
@@ -0,0 +1,27 @@
1
+ export const JUDGE_PROVIDERS = [
2
+ { stageflowProvider: "anthropic", envVar: "ANTHROPIC_API_KEY", spec: (m) => `anthropic:messages:${m}` },
3
+ { stageflowProvider: "openai", envVar: "OPENAI_API_KEY", spec: (m) => `openai:chat:${m}` },
4
+ { stageflowProvider: "openrouter", envVar: "OPENROUTER_API_KEY", spec: (m) => `openrouter:${m}` },
5
+ ];
6
+ export function supportedJudgeProviders() {
7
+ return JUDGE_PROVIDERS.map((row) => row.stageflowProvider);
8
+ }
9
+ export function mapJudge(modelId) {
10
+ const slash = modelId.indexOf("/");
11
+ if (slash <= 0)
12
+ return null;
13
+ const provider = modelId.slice(0, slash);
14
+ const model = modelId.slice(slash + 1);
15
+ if (model.length === 0)
16
+ return null;
17
+ const row = JUDGE_PROVIDERS.find((entry) => entry.stageflowProvider === provider);
18
+ if (row === undefined)
19
+ return null;
20
+ return { providerSpec: row.spec(model), envVar: row.envVar, stageflowProvider: provider };
21
+ }
22
+ export function judgeCapableModels(modelIds) {
23
+ return modelIds.filter((id) => mapJudge(id) !== null);
24
+ }
25
+ export function unsupportedJudgeMessage(modelId) {
26
+ return `Judge model "${modelId}" cannot be used: supported judge providers are ${supportedJudgeProviders().join(", ")}.`;
27
+ }
@@ -0,0 +1,5 @@
1
+ export interface DiffLine {
2
+ op: "same" | "add" | "remove";
3
+ text: string;
4
+ }
5
+ export declare function lineDiff(a: string, b: string): DiffLine[];
@@ -0,0 +1,34 @@
1
+ function toLines(text) {
2
+ return text === "" ? [] : text.split("\n");
3
+ }
4
+ export function lineDiff(a, b) {
5
+ const x = toLines(a);
6
+ const y = toLines(b);
7
+ const lcs = Array.from({ length: x.length + 1 }, () => new Array(y.length + 1).fill(0));
8
+ for (let i = x.length - 1; i >= 0; i--) {
9
+ for (let j = y.length - 1; j >= 0; j--) {
10
+ lcs[i][j] = x[i] === y[j] ? lcs[i + 1][j + 1] + 1 : Math.max(lcs[i + 1][j], lcs[i][j + 1]);
11
+ }
12
+ }
13
+ const out = [];
14
+ let i = 0;
15
+ let j = 0;
16
+ while (i < x.length && j < y.length) {
17
+ if (x[i] === y[j]) {
18
+ out.push({ op: "same", text: x[i] });
19
+ i++;
20
+ j++;
21
+ }
22
+ else if (lcs[i + 1][j] >= lcs[i][j + 1]) {
23
+ out.push({ op: "remove", text: x[i++] });
24
+ }
25
+ else {
26
+ out.push({ op: "add", text: y[j++] });
27
+ }
28
+ }
29
+ while (i < x.length)
30
+ out.push({ op: "remove", text: x[i++] });
31
+ while (j < y.length)
32
+ out.push({ op: "add", text: y[j++] });
33
+ return out;
34
+ }
@@ -0,0 +1,31 @@
1
+ import type { EvalType } from "./statsTypes.js";
2
+ export type ListStatus = "never_run" | "saturated" | "passing" | "regressed" | "failing";
3
+ export interface ListRunGrader {
4
+ graderId: string;
5
+ hash: string;
6
+ passRate: number | null;
7
+ }
8
+ export interface ListRunSummary {
9
+ runNumber: number;
10
+ passed: boolean;
11
+ passRate: number | null;
12
+ finishedAt: string | null;
13
+ graders: ListRunGrader[];
14
+ }
15
+ export declare function listStatus(completedRunsNewestFirst: ListRunSummary[], type: EvalType): ListStatus;
16
+ export interface BannerEvalInput {
17
+ evalId: string;
18
+ status: ListStatus;
19
+ last: ListRunSummary | null;
20
+ previous: ListRunSummary | null;
21
+ }
22
+ export interface RegressionBanner {
23
+ eval_id: string;
24
+ previous_pass_rate: number | null;
25
+ last_pass_rate: number | null;
26
+ criteria_dropped: number;
27
+ last_run_number: number;
28
+ previous_run_number: number;
29
+ }
30
+ export declare function criteriaDropped(last: ListRunSummary, previous: ListRunSummary): number;
31
+ export declare function regressionBanner(evals: BannerEvalInput[]): RegressionBanner | null;
@@ -0,0 +1,48 @@
1
+ const EPS = 1e-9;
2
+ const SATURATED_PASS_RATE = 0.98;
3
+ const CRITERIA_DROP_POINTS = -5;
4
+ export function listStatus(completedRunsNewestFirst, type) {
5
+ const last = completedRunsNewestFirst[0];
6
+ if (!last)
7
+ return "never_run";
8
+ const previous = completedRunsNewestFirst[1];
9
+ if (last.passed) {
10
+ if (type === "capability" && (last.passRate ?? 0) + EPS >= SATURATED_PASS_RATE) {
11
+ return "saturated";
12
+ }
13
+ return "passing";
14
+ }
15
+ return previous?.passed ? "regressed" : "failing";
16
+ }
17
+ export function criteriaDropped(last, previous) {
18
+ let dropped = 0;
19
+ for (const g of last.graders) {
20
+ const prev = previous.graders.find((p) => p.graderId === g.graderId);
21
+ if (!prev || prev.hash !== g.hash)
22
+ continue;
23
+ if (g.passRate === null || prev.passRate === null)
24
+ continue;
25
+ if ((g.passRate - prev.passRate) * 100 <= CRITERIA_DROP_POINTS + EPS)
26
+ dropped += 1;
27
+ }
28
+ return dropped;
29
+ }
30
+ export function regressionBanner(evals) {
31
+ let best = null;
32
+ for (const e of evals) {
33
+ if (e.status !== "regressed" || !e.last || !e.previous)
34
+ continue;
35
+ if (!best || (e.last.finishedAt ?? "") > (best.last?.finishedAt ?? ""))
36
+ best = e;
37
+ }
38
+ if (!best?.last || !best.previous)
39
+ return null;
40
+ return {
41
+ eval_id: best.evalId,
42
+ previous_pass_rate: best.previous.passRate,
43
+ last_pass_rate: best.last.passRate,
44
+ criteria_dropped: criteriaDropped(best.last, best.previous),
45
+ last_run_number: best.last.runNumber,
46
+ previous_run_number: best.previous.runNumber,
47
+ };
48
+ }