@namzu/sdk 2.0.0 → 3.0.0

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 (1503) hide show
  1. package/CHANGELOG.md +2265 -0
  2. package/README.md +49 -53
  3. package/dist/__tests__/default-path-conformance.test.d.ts +2 -0
  4. package/dist/__tests__/default-path-conformance.test.d.ts.map +1 -0
  5. package/dist/__tests__/default-path-conformance.test.js +262 -0
  6. package/dist/__tests__/default-path-conformance.test.js.map +1 -0
  7. package/dist/__tests__/shipped-defaults.test.d.ts +2 -0
  8. package/dist/__tests__/shipped-defaults.test.d.ts.map +1 -0
  9. package/dist/__tests__/shipped-defaults.test.js +97 -0
  10. package/dist/__tests__/shipped-defaults.test.js.map +1 -0
  11. package/dist/advisory/__tests__/budget-enforcement.test.d.ts +2 -0
  12. package/dist/advisory/__tests__/budget-enforcement.test.d.ts.map +1 -0
  13. package/dist/advisory/__tests__/budget-enforcement.test.js +159 -0
  14. package/dist/advisory/__tests__/budget-enforcement.test.js.map +1 -0
  15. package/dist/advisory/__tests__/structured-response.test.d.ts +2 -0
  16. package/dist/advisory/__tests__/structured-response.test.d.ts.map +1 -0
  17. package/dist/advisory/__tests__/structured-response.test.js +99 -0
  18. package/dist/advisory/__tests__/structured-response.test.js.map +1 -0
  19. package/dist/advisory/budget.d.ts +15 -0
  20. package/dist/advisory/budget.d.ts.map +1 -0
  21. package/dist/advisory/budget.js +21 -0
  22. package/dist/advisory/budget.js.map +1 -0
  23. package/dist/advisory/context.d.ts +2 -0
  24. package/dist/advisory/context.d.ts.map +1 -1
  25. package/dist/advisory/context.js +16 -0
  26. package/dist/advisory/context.js.map +1 -1
  27. package/dist/advisory/executor.d.ts +14 -11
  28. package/dist/advisory/executor.d.ts.map +1 -1
  29. package/dist/advisory/executor.js +36 -24
  30. package/dist/advisory/executor.js.map +1 -1
  31. package/dist/advisory/executor.test.d.ts +8 -6
  32. package/dist/advisory/executor.test.d.ts.map +1 -1
  33. package/dist/advisory/executor.test.js +12 -8
  34. package/dist/advisory/executor.test.js.map +1 -1
  35. package/dist/advisory/index.d.ts +2 -0
  36. package/dist/advisory/index.d.ts.map +1 -1
  37. package/dist/advisory/index.js +2 -0
  38. package/dist/advisory/index.js.map +1 -1
  39. package/dist/advisory/parse.d.ts +23 -0
  40. package/dist/advisory/parse.d.ts.map +1 -0
  41. package/dist/advisory/parse.js +56 -0
  42. package/dist/advisory/parse.js.map +1 -0
  43. package/dist/agents/AbstractAgent.d.ts +48 -0
  44. package/dist/agents/AbstractAgent.d.ts.map +1 -1
  45. package/dist/agents/AbstractAgent.js +77 -0
  46. package/dist/agents/AbstractAgent.js.map +1 -1
  47. package/dist/agents/PipelineAgent.d.ts +10 -0
  48. package/dist/agents/PipelineAgent.d.ts.map +1 -1
  49. package/dist/agents/PipelineAgent.js +12 -0
  50. package/dist/agents/PipelineAgent.js.map +1 -1
  51. package/dist/agents/ReactiveAgent.d.ts +10 -0
  52. package/dist/agents/ReactiveAgent.d.ts.map +1 -1
  53. package/dist/agents/ReactiveAgent.js +40 -0
  54. package/dist/agents/ReactiveAgent.js.map +1 -1
  55. package/dist/agents/RouterAgent.d.ts +10 -0
  56. package/dist/agents/RouterAgent.d.ts.map +1 -1
  57. package/dist/agents/RouterAgent.js +23 -2
  58. package/dist/agents/RouterAgent.js.map +1 -1
  59. package/dist/agents/SupervisorAgent.d.ts +11 -1
  60. package/dist/agents/SupervisorAgent.d.ts.map +1 -1
  61. package/dist/agents/SupervisorAgent.js +23 -1
  62. package/dist/agents/SupervisorAgent.js.map +1 -1
  63. package/dist/agents/__tests__/idempotency-key.test.d.ts +2 -0
  64. package/dist/agents/__tests__/idempotency-key.test.d.ts.map +1 -0
  65. package/dist/agents/__tests__/idempotency-key.test.js +149 -0
  66. package/dist/agents/__tests__/idempotency-key.test.js.map +1 -0
  67. package/dist/agents/__tests__/invocation-lock.test.d.ts +2 -0
  68. package/dist/agents/__tests__/invocation-lock.test.d.ts.map +1 -0
  69. package/dist/agents/__tests__/invocation-lock.test.js +126 -0
  70. package/dist/agents/__tests__/invocation-lock.test.js.map +1 -0
  71. package/dist/agents/__tests__/reachability.test.d.ts +2 -0
  72. package/dist/agents/__tests__/reachability.test.d.ts.map +1 -0
  73. package/dist/agents/__tests__/reachability.test.js +173 -0
  74. package/dist/agents/__tests__/reachability.test.js.map +1 -0
  75. package/dist/agents/__tests__/supervisor-ledger.test.js +1 -1
  76. package/dist/agents/__tests__/supervisor-ledger.test.js.map +1 -1
  77. package/dist/agents/index.d.ts +1 -1
  78. package/dist/agents/index.d.ts.map +1 -1
  79. package/dist/agents/lock.d.ts +0 -1
  80. package/dist/agents/lock.d.ts.map +1 -1
  81. package/dist/agents/lock.js.map +1 -1
  82. package/dist/bridge/a2a/late-events.test.d.ts +2 -0
  83. package/dist/bridge/a2a/late-events.test.d.ts.map +1 -0
  84. package/dist/bridge/a2a/late-events.test.js +70 -0
  85. package/dist/bridge/a2a/late-events.test.js.map +1 -0
  86. package/dist/bridge/a2a/mapper.d.ts.map +1 -1
  87. package/dist/bridge/a2a/mapper.js +55 -3
  88. package/dist/bridge/a2a/mapper.js.map +1 -1
  89. package/dist/bridge/a2a/mapper.test.js +1 -1
  90. package/dist/bridge/a2a/mapper.test.js.map +1 -1
  91. package/dist/bridge/a2a/message.d.ts.map +1 -1
  92. package/dist/bridge/a2a/message.js +5 -1
  93. package/dist/bridge/a2a/message.js.map +1 -1
  94. package/dist/bridge/sse/late-events.test.d.ts +2 -0
  95. package/dist/bridge/sse/late-events.test.d.ts.map +1 -0
  96. package/dist/bridge/sse/late-events.test.js +163 -0
  97. package/dist/bridge/sse/late-events.test.js.map +1 -0
  98. package/dist/bridge/sse/mapper.d.ts.map +1 -1
  99. package/dist/bridge/sse/mapper.js +100 -0
  100. package/dist/bridge/sse/mapper.js.map +1 -1
  101. package/dist/compaction/__tests__/cleared-result-recovery.test.d.ts +2 -0
  102. package/dist/compaction/__tests__/cleared-result-recovery.test.d.ts.map +1 -0
  103. package/dist/compaction/__tests__/cleared-result-recovery.test.js +101 -0
  104. package/dist/compaction/__tests__/cleared-result-recovery.test.js.map +1 -0
  105. package/dist/compaction/__tests__/context-window.test.d.ts +2 -0
  106. package/dist/compaction/__tests__/context-window.test.d.ts.map +1 -0
  107. package/dist/compaction/__tests__/context-window.test.js +62 -0
  108. package/dist/compaction/__tests__/context-window.test.js.map +1 -0
  109. package/dist/compaction/__tests__/dangling.test.js +16 -5
  110. package/dist/compaction/__tests__/dangling.test.js.map +1 -1
  111. package/dist/compaction/__tests__/lossiness.test.d.ts +2 -0
  112. package/dist/compaction/__tests__/lossiness.test.d.ts.map +1 -0
  113. package/dist/compaction/__tests__/lossiness.test.js +163 -0
  114. package/dist/compaction/__tests__/lossiness.test.js.map +1 -0
  115. package/dist/compaction/__tests__/state-across-resume.test.d.ts +2 -0
  116. package/dist/compaction/__tests__/state-across-resume.test.d.ts.map +1 -0
  117. package/dist/compaction/__tests__/state-across-resume.test.js +88 -0
  118. package/dist/compaction/__tests__/state-across-resume.test.js.map +1 -0
  119. package/dist/compaction/__tests__/tool-result-editing.test.d.ts +2 -0
  120. package/dist/compaction/__tests__/tool-result-editing.test.d.ts.map +1 -0
  121. package/dist/compaction/__tests__/tool-result-editing.test.js +184 -0
  122. package/dist/compaction/__tests__/tool-result-editing.test.js.map +1 -0
  123. package/dist/compaction/__tests__/trim-opens-on-user.test.d.ts +2 -0
  124. package/dist/compaction/__tests__/trim-opens-on-user.test.d.ts.map +1 -0
  125. package/dist/compaction/__tests__/trim-opens-on-user.test.js +109 -0
  126. package/dist/compaction/__tests__/trim-opens-on-user.test.js.map +1 -0
  127. package/dist/compaction/__tests__/verifier-model.test.d.ts +2 -0
  128. package/dist/compaction/__tests__/verifier-model.test.d.ts.map +1 -0
  129. package/dist/compaction/__tests__/verifier-model.test.js +58 -0
  130. package/dist/compaction/__tests__/verifier-model.test.js.map +1 -0
  131. package/dist/compaction/context-window.d.ts +44 -0
  132. package/dist/compaction/context-window.d.ts.map +1 -0
  133. package/dist/compaction/context-window.js +97 -0
  134. package/dist/compaction/context-window.js.map +1 -0
  135. package/dist/compaction/dangling.js +48 -1
  136. package/dist/compaction/dangling.js.map +1 -1
  137. package/dist/compaction/extractor.d.ts.map +1 -1
  138. package/dist/compaction/extractor.js +55 -1
  139. package/dist/compaction/extractor.js.map +1 -1
  140. package/dist/compaction/index.d.ts +5 -0
  141. package/dist/compaction/index.d.ts.map +1 -1
  142. package/dist/compaction/index.js +3 -0
  143. package/dist/compaction/index.js.map +1 -1
  144. package/dist/compaction/manager.d.ts +25 -0
  145. package/dist/compaction/manager.d.ts.map +1 -1
  146. package/dist/compaction/manager.js +42 -8
  147. package/dist/compaction/manager.js.map +1 -1
  148. package/dist/compaction/managers/structured.d.ts.map +1 -1
  149. package/dist/compaction/managers/structured.js +7 -3
  150. package/dist/compaction/managers/structured.js.map +1 -1
  151. package/dist/compaction/retention.d.ts +15 -0
  152. package/dist/compaction/retention.d.ts.map +1 -0
  153. package/dist/compaction/retention.js +89 -0
  154. package/dist/compaction/retention.js.map +1 -0
  155. package/dist/compaction/serializer.d.ts.map +1 -1
  156. package/dist/compaction/serializer.js +28 -11
  157. package/dist/compaction/serializer.js.map +1 -1
  158. package/dist/compaction/tool-result-editing.d.ts +54 -0
  159. package/dist/compaction/tool-result-editing.d.ts.map +1 -0
  160. package/dist/compaction/tool-result-editing.js +164 -0
  161. package/dist/compaction/tool-result-editing.js.map +1 -0
  162. package/dist/compaction/types.d.ts +10 -0
  163. package/dist/compaction/types.d.ts.map +1 -1
  164. package/dist/compaction/verifier.d.ts +20 -1
  165. package/dist/compaction/verifier.d.ts.map +1 -1
  166. package/dist/compaction/verifier.js +11 -2
  167. package/dist/compaction/verifier.js.map +1 -1
  168. package/dist/compaction/wire.d.ts +54 -0
  169. package/dist/compaction/wire.d.ts.map +1 -0
  170. package/dist/compaction/wire.js +45 -0
  171. package/dist/compaction/wire.js.map +1 -0
  172. package/dist/config/runtime.d.ts +161 -45
  173. package/dist/config/runtime.d.ts.map +1 -1
  174. package/dist/config/runtime.js +61 -19
  175. package/dist/config/runtime.js.map +1 -1
  176. package/dist/connector/builtins/http.d.ts.map +1 -1
  177. package/dist/connector/builtins/http.js +14 -1
  178. package/dist/connector/builtins/http.js.map +1 -1
  179. package/dist/connector/builtins/http.test.js +12 -0
  180. package/dist/connector/builtins/http.test.js.map +1 -1
  181. package/dist/connector/mcp/__tests__/discovery-policy.test.d.ts +2 -0
  182. package/dist/connector/mcp/__tests__/discovery-policy.test.d.ts.map +1 -0
  183. package/dist/connector/mcp/__tests__/discovery-policy.test.js +109 -0
  184. package/dist/connector/mcp/__tests__/discovery-policy.test.js.map +1 -0
  185. package/dist/connector/mcp/__tests__/list-pagination.test.d.ts +2 -0
  186. package/dist/connector/mcp/__tests__/list-pagination.test.d.ts.map +1 -0
  187. package/dist/connector/mcp/__tests__/list-pagination.test.js +81 -0
  188. package/dist/connector/mcp/__tests__/list-pagination.test.js.map +1 -0
  189. package/dist/connector/mcp/__tests__/policy.test.d.ts +2 -0
  190. package/dist/connector/mcp/__tests__/policy.test.d.ts.map +1 -0
  191. package/dist/connector/mcp/__tests__/policy.test.js +86 -0
  192. package/dist/connector/mcp/__tests__/policy.test.js.map +1 -0
  193. package/dist/connector/mcp/__tests__/protocol-negotiation.test.d.ts +2 -0
  194. package/dist/connector/mcp/__tests__/protocol-negotiation.test.d.ts.map +1 -0
  195. package/dist/connector/mcp/__tests__/protocol-negotiation.test.js +81 -0
  196. package/dist/connector/mcp/__tests__/protocol-negotiation.test.js.map +1 -0
  197. package/dist/connector/mcp/__tests__/schema-fidelity.test.d.ts +2 -0
  198. package/dist/connector/mcp/__tests__/schema-fidelity.test.d.ts.map +1 -0
  199. package/dist/connector/mcp/__tests__/schema-fidelity.test.js +206 -0
  200. package/dist/connector/mcp/__tests__/schema-fidelity.test.js.map +1 -0
  201. package/dist/connector/mcp/__tests__/structured-result.test.d.ts +2 -0
  202. package/dist/connector/mcp/__tests__/structured-result.test.d.ts.map +1 -0
  203. package/dist/connector/mcp/__tests__/structured-result.test.js +112 -0
  204. package/dist/connector/mcp/__tests__/structured-result.test.js.map +1 -0
  205. package/dist/connector/mcp/adapter.d.ts +14 -0
  206. package/dist/connector/mcp/adapter.d.ts.map +1 -1
  207. package/dist/connector/mcp/adapter.js +292 -18
  208. package/dist/connector/mcp/adapter.js.map +1 -1
  209. package/dist/connector/mcp/client.d.ts +38 -0
  210. package/dist/connector/mcp/client.d.ts.map +1 -1
  211. package/dist/connector/mcp/client.js +138 -12
  212. package/dist/connector/mcp/client.js.map +1 -1
  213. package/dist/connector/mcp/client.test.d.ts +2 -0
  214. package/dist/connector/mcp/client.test.d.ts.map +1 -0
  215. package/dist/connector/mcp/client.test.js +105 -0
  216. package/dist/connector/mcp/client.test.js.map +1 -0
  217. package/dist/connector/mcp/discovery.d.ts +32 -1
  218. package/dist/connector/mcp/discovery.d.ts.map +1 -1
  219. package/dist/connector/mcp/discovery.js +41 -4
  220. package/dist/connector/mcp/discovery.js.map +1 -1
  221. package/dist/connector/mcp/http-sse.d.ts.map +1 -1
  222. package/dist/connector/mcp/http-sse.js +15 -1
  223. package/dist/connector/mcp/http-sse.js.map +1 -1
  224. package/dist/connector/mcp/index.d.ts +3 -0
  225. package/dist/connector/mcp/index.d.ts.map +1 -1
  226. package/dist/connector/mcp/index.js +1 -0
  227. package/dist/connector/mcp/index.js.map +1 -1
  228. package/dist/connector/mcp/policy.d.ts +57 -0
  229. package/dist/connector/mcp/policy.d.ts.map +1 -0
  230. package/dist/connector/mcp/policy.js +84 -0
  231. package/dist/connector/mcp/policy.js.map +1 -0
  232. package/dist/connector/mcp/schema-refs.d.ts +26 -0
  233. package/dist/connector/mcp/schema-refs.d.ts.map +1 -0
  234. package/dist/connector/mcp/schema-refs.js +111 -0
  235. package/dist/connector/mcp/schema-refs.js.map +1 -0
  236. package/dist/constants/agent/index.d.ts.map +1 -1
  237. package/dist/constants/agent/index.js +3 -0
  238. package/dist/constants/agent/index.js.map +1 -1
  239. package/dist/constants/compaction/index.js +1 -1
  240. package/dist/constants/compaction/index.js.map +1 -1
  241. package/dist/constants/continuation.d.ts +4 -4
  242. package/dist/constants/continuation.js +4 -4
  243. package/dist/constants/mcp/index.d.ts +31 -0
  244. package/dist/constants/mcp/index.d.ts.map +1 -1
  245. package/dist/constants/mcp/index.js +31 -0
  246. package/dist/constants/mcp/index.js.map +1 -1
  247. package/dist/constants/plugin/index.d.ts +23 -2
  248. package/dist/constants/plugin/index.d.ts.map +1 -1
  249. package/dist/constants/plugin/index.js +23 -2
  250. package/dist/constants/plugin/index.js.map +1 -1
  251. package/dist/constants/sandbox/index.d.ts +2 -3
  252. package/dist/constants/sandbox/index.d.ts.map +1 -1
  253. package/dist/constants/sandbox/index.js +6 -5
  254. package/dist/constants/sandbox/index.js.map +1 -1
  255. package/dist/constants/telemetry/index.d.ts +7 -0
  256. package/dist/constants/telemetry/index.d.ts.map +1 -1
  257. package/dist/constants/telemetry/index.js +7 -0
  258. package/dist/constants/telemetry/index.js.map +1 -1
  259. package/dist/constants/tools/index.d.ts +16 -0
  260. package/dist/constants/tools/index.d.ts.map +1 -1
  261. package/dist/constants/tools/index.js +16 -0
  262. package/dist/constants/tools/index.js.map +1 -1
  263. package/dist/contracts/a2a.d.ts +94 -94
  264. package/dist/contracts/api.d.ts +40 -2
  265. package/dist/contracts/api.d.ts.map +1 -1
  266. package/dist/contracts/run-status.d.ts +17 -0
  267. package/dist/contracts/run-status.d.ts.map +1 -0
  268. package/dist/contracts/run-status.js +41 -0
  269. package/dist/contracts/run-status.js.map +1 -0
  270. package/dist/contracts/schemas.d.ts +6 -6
  271. package/dist/eval/__tests__/case-deadline.test.d.ts +2 -0
  272. package/dist/eval/__tests__/case-deadline.test.d.ts.map +1 -0
  273. package/dist/eval/__tests__/case-deadline.test.js +128 -0
  274. package/dist/eval/__tests__/case-deadline.test.js.map +1 -0
  275. package/dist/eval/__tests__/end-to-end.test.d.ts +2 -0
  276. package/dist/eval/__tests__/end-to-end.test.d.ts.map +1 -0
  277. package/dist/eval/__tests__/end-to-end.test.js +223 -0
  278. package/dist/eval/__tests__/end-to-end.test.js.map +1 -0
  279. package/dist/eval/__tests__/harness-honesty.test.d.ts +2 -0
  280. package/dist/eval/__tests__/harness-honesty.test.d.ts.map +1 -0
  281. package/dist/eval/__tests__/harness-honesty.test.js +85 -0
  282. package/dist/eval/__tests__/harness-honesty.test.js.map +1 -0
  283. package/dist/eval/__tests__/harness.test.d.ts +2 -0
  284. package/dist/eval/__tests__/harness.test.d.ts.map +1 -0
  285. package/dist/eval/__tests__/harness.test.js +207 -0
  286. package/dist/eval/__tests__/harness.test.js.map +1 -0
  287. package/dist/eval/__tests__/judge.test.d.ts +2 -0
  288. package/dist/eval/__tests__/judge.test.d.ts.map +1 -0
  289. package/dist/eval/__tests__/judge.test.js +255 -0
  290. package/dist/eval/__tests__/judge.test.js.map +1 -0
  291. package/dist/eval/__tests__/scorer-gates.test.d.ts +2 -0
  292. package/dist/eval/__tests__/scorer-gates.test.d.ts.map +1 -0
  293. package/dist/eval/__tests__/scorer-gates.test.js +136 -0
  294. package/dist/eval/__tests__/scorer-gates.test.js.map +1 -0
  295. package/dist/eval/experiment.d.ts +54 -0
  296. package/dist/eval/experiment.d.ts.map +1 -0
  297. package/dist/eval/experiment.js +263 -0
  298. package/dist/eval/experiment.js.map +1 -0
  299. package/dist/eval/from-run.d.ts +30 -0
  300. package/dist/eval/from-run.d.ts.map +1 -0
  301. package/dist/eval/from-run.js +46 -0
  302. package/dist/eval/from-run.js.map +1 -0
  303. package/dist/eval/index.d.ts +8 -0
  304. package/dist/eval/index.d.ts.map +1 -0
  305. package/dist/eval/index.js +5 -0
  306. package/dist/eval/index.js.map +1 -0
  307. package/dist/eval/judge.d.ts +63 -0
  308. package/dist/eval/judge.d.ts.map +1 -0
  309. package/dist/eval/judge.js +136 -0
  310. package/dist/eval/judge.js.map +1 -0
  311. package/dist/eval/scorers.d.ts +40 -0
  312. package/dist/eval/scorers.d.ts.map +1 -0
  313. package/dist/eval/scorers.js +151 -0
  314. package/dist/eval/scorers.js.map +1 -0
  315. package/dist/eval/types.d.ts +150 -0
  316. package/dist/eval/types.d.ts.map +1 -0
  317. package/dist/eval/types.js +2 -0
  318. package/dist/eval/types.js.map +1 -0
  319. package/dist/execution/local.d.ts.map +1 -1
  320. package/dist/execution/local.js +6 -1
  321. package/dist/execution/local.js.map +1 -1
  322. package/dist/execution/local.test.d.ts +2 -0
  323. package/dist/execution/local.test.d.ts.map +1 -0
  324. package/dist/execution/local.test.js +37 -0
  325. package/dist/execution/local.test.js.map +1 -0
  326. package/dist/gateway/__tests__/list-survives-eviction.test.d.ts +2 -0
  327. package/dist/gateway/__tests__/list-survives-eviction.test.d.ts.map +1 -0
  328. package/dist/gateway/__tests__/list-survives-eviction.test.js +102 -0
  329. package/dist/gateway/__tests__/list-survives-eviction.test.js.map +1 -0
  330. package/dist/gateway/local.d.ts +42 -2
  331. package/dist/gateway/local.d.ts.map +1 -1
  332. package/dist/gateway/local.js +110 -3
  333. package/dist/gateway/local.js.map +1 -1
  334. package/dist/manager/agent/__tests__/dispose-cancels.test.d.ts +2 -0
  335. package/dist/manager/agent/__tests__/dispose-cancels.test.d.ts.map +1 -0
  336. package/dist/manager/agent/__tests__/dispose-cancels.test.js +88 -0
  337. package/dist/manager/agent/__tests__/dispose-cancels.test.js.map +1 -0
  338. package/dist/manager/agent/__tests__/lifecycle.test.js +230 -0
  339. package/dist/manager/agent/__tests__/lifecycle.test.js.map +1 -1
  340. package/dist/manager/agent/lifecycle.d.ts +35 -0
  341. package/dist/manager/agent/lifecycle.d.ts.map +1 -1
  342. package/dist/manager/agent/lifecycle.js +117 -7
  343. package/dist/manager/agent/lifecycle.js.map +1 -1
  344. package/dist/manager/run/persistence.d.ts +90 -1
  345. package/dist/manager/run/persistence.d.ts.map +1 -1
  346. package/dist/manager/run/persistence.js +120 -0
  347. package/dist/manager/run/persistence.js.map +1 -1
  348. package/dist/plugin/__tests__/enable-contributions.test.js +3 -3
  349. package/dist/plugin/__tests__/enable-contributions.test.js.map +1 -1
  350. package/dist/plugin/__tests__/hook-ordering.test.d.ts +2 -0
  351. package/dist/plugin/__tests__/hook-ordering.test.d.ts.map +1 -0
  352. package/dist/plugin/__tests__/hook-ordering.test.js +227 -0
  353. package/dist/plugin/__tests__/hook-ordering.test.js.map +1 -0
  354. package/dist/plugin/__tests__/install-time-refusal.test.d.ts +2 -0
  355. package/dist/plugin/__tests__/install-time-refusal.test.d.ts.map +1 -0
  356. package/dist/plugin/__tests__/install-time-refusal.test.js +49 -0
  357. package/dist/plugin/__tests__/install-time-refusal.test.js.map +1 -0
  358. package/dist/plugin/__tests__/lifecycle.test.js +95 -68
  359. package/dist/plugin/__tests__/lifecycle.test.js.map +1 -1
  360. package/dist/plugin/lifecycle.d.ts +15 -1
  361. package/dist/plugin/lifecycle.d.ts.map +1 -1
  362. package/dist/plugin/lifecycle.js +67 -20
  363. package/dist/plugin/lifecycle.js.map +1 -1
  364. package/dist/plugin/loader.d.ts +15 -0
  365. package/dist/plugin/loader.d.ts.map +1 -1
  366. package/dist/plugin/loader.js +33 -1
  367. package/dist/plugin/loader.js.map +1 -1
  368. package/dist/probe/registry.d.ts.map +1 -1
  369. package/dist/probe/registry.js +19 -1
  370. package/dist/probe/registry.js.map +1 -1
  371. package/dist/probe/registry.test.d.ts +9 -3
  372. package/dist/probe/registry.test.d.ts.map +1 -1
  373. package/dist/probe/registry.test.js +44 -6
  374. package/dist/probe/registry.test.js.map +1 -1
  375. package/dist/provider/__tests__/mock-script.test.d.ts +2 -0
  376. package/dist/provider/__tests__/mock-script.test.d.ts.map +1 -0
  377. package/dist/provider/__tests__/mock-script.test.js +151 -0
  378. package/dist/provider/__tests__/mock-script.test.js.map +1 -0
  379. package/dist/provider/__tests__/retry-visibility.test.d.ts +2 -0
  380. package/dist/provider/__tests__/retry-visibility.test.d.ts.map +1 -0
  381. package/dist/provider/__tests__/retry-visibility.test.js +137 -0
  382. package/dist/provider/__tests__/retry-visibility.test.js.map +1 -0
  383. package/dist/provider/__tests__/retry.test.d.ts +2 -0
  384. package/dist/provider/__tests__/retry.test.d.ts.map +1 -0
  385. package/dist/provider/__tests__/retry.test.js +259 -0
  386. package/dist/provider/__tests__/retry.test.js.map +1 -0
  387. package/dist/provider/capabilities.d.ts +1 -0
  388. package/dist/provider/capabilities.d.ts.map +1 -1
  389. package/dist/provider/capabilities.js +2 -0
  390. package/dist/provider/capabilities.js.map +1 -1
  391. package/dist/provider/collect.d.ts +1 -1
  392. package/dist/provider/collect.js +1 -1
  393. package/dist/provider/collect.test.d.ts +1 -1
  394. package/dist/provider/collect.test.js +1 -1
  395. package/dist/provider/index.d.ts +3 -0
  396. package/dist/provider/index.d.ts.map +1 -1
  397. package/dist/provider/index.js +2 -0
  398. package/dist/provider/index.js.map +1 -1
  399. package/dist/provider/mock-register.d.ts.map +1 -1
  400. package/dist/provider/mock-register.js +5 -2
  401. package/dist/provider/mock-register.js.map +1 -1
  402. package/dist/provider/mock.d.ts +35 -5
  403. package/dist/provider/mock.d.ts.map +1 -1
  404. package/dist/provider/mock.js +104 -32
  405. package/dist/provider/mock.js.map +1 -1
  406. package/dist/provider/retry.d.ts +42 -0
  407. package/dist/provider/retry.d.ts.map +1 -0
  408. package/dist/provider/retry.js +161 -0
  409. package/dist/provider/retry.js.map +1 -0
  410. package/dist/public-runtime.d.ts +26 -7
  411. package/dist/public-runtime.d.ts.map +1 -1
  412. package/dist/public-runtime.js +50 -7
  413. package/dist/public-runtime.js.map +1 -1
  414. package/dist/public-types.d.ts +5 -1
  415. package/dist/public-types.d.ts.map +1 -1
  416. package/dist/rag/__tests__/retrieval-math.test.d.ts +2 -0
  417. package/dist/rag/__tests__/retrieval-math.test.d.ts.map +1 -0
  418. package/dist/rag/__tests__/retrieval-math.test.js +46 -0
  419. package/dist/rag/__tests__/retrieval-math.test.js.map +1 -0
  420. package/dist/rag/chunking.d.ts +12 -0
  421. package/dist/rag/chunking.d.ts.map +1 -1
  422. package/dist/rag/chunking.js +32 -1
  423. package/dist/rag/chunking.js.map +1 -1
  424. package/dist/rag/embedding.d.ts +4 -4
  425. package/dist/rag/embedding.d.ts.map +1 -1
  426. package/dist/rag/embedding.js +23 -3
  427. package/dist/rag/embedding.js.map +1 -1
  428. package/dist/rag/embedding.test.d.ts +1 -1
  429. package/dist/rag/embedding.test.js +59 -10
  430. package/dist/rag/embedding.test.js.map +1 -1
  431. package/dist/rag/index.d.ts +1 -1
  432. package/dist/rag/index.d.ts.map +1 -1
  433. package/dist/rag/index.js +1 -1
  434. package/dist/rag/index.js.map +1 -1
  435. package/dist/rag/retriever.d.ts +0 -1
  436. package/dist/rag/retriever.d.ts.map +1 -1
  437. package/dist/rag/retriever.js +75 -22
  438. package/dist/rag/retriever.js.map +1 -1
  439. package/dist/rag/vector-store.test.d.ts +1 -1
  440. package/dist/rag/vector-store.test.js +2 -2
  441. package/dist/rag/vector-store.test.js.map +1 -1
  442. package/dist/registry/ManagedRegistry.test.d.ts +1 -1
  443. package/dist/registry/ManagedRegistry.test.js +1 -1
  444. package/dist/registry/Registry.test.d.ts +1 -1
  445. package/dist/registry/Registry.test.js +1 -1
  446. package/dist/registry/connector/definitions.test.d.ts +1 -1
  447. package/dist/registry/connector/definitions.test.js +1 -1
  448. package/dist/registry/tool/__tests__/schema.test.d.ts +2 -0
  449. package/dist/registry/tool/__tests__/schema.test.d.ts.map +1 -0
  450. package/dist/registry/tool/__tests__/schema.test.js +176 -0
  451. package/dist/registry/tool/__tests__/schema.test.js.map +1 -0
  452. package/dist/registry/tool/__tests__/tool-name-validation.test.d.ts +2 -0
  453. package/dist/registry/tool/__tests__/tool-name-validation.test.d.ts.map +1 -0
  454. package/dist/registry/tool/__tests__/tool-name-validation.test.js +70 -0
  455. package/dist/registry/tool/__tests__/tool-name-validation.test.js.map +1 -0
  456. package/dist/registry/tool/execute.d.ts +37 -0
  457. package/dist/registry/tool/execute.d.ts.map +1 -1
  458. package/dist/registry/tool/execute.js +91 -21
  459. package/dist/registry/tool/execute.js.map +1 -1
  460. package/dist/registry/tool/schema.d.ts +26 -0
  461. package/dist/registry/tool/schema.d.ts.map +1 -0
  462. package/dist/registry/tool/schema.js +77 -0
  463. package/dist/registry/tool/schema.js.map +1 -0
  464. package/dist/registry/toolset/catalog.d.ts.map +1 -1
  465. package/dist/registry/toolset/catalog.js +0 -1
  466. package/dist/registry/toolset/catalog.js.map +1 -1
  467. package/dist/registry/toolset/catalog.test.js +0 -1
  468. package/dist/registry/toolset/catalog.test.js.map +1 -1
  469. package/dist/run/reporter.d.ts.map +1 -1
  470. package/dist/run/reporter.js +76 -0
  471. package/dist/run/reporter.js.map +1 -1
  472. package/dist/runtime/bidi/__tests__/session.test.d.ts +2 -0
  473. package/dist/runtime/bidi/__tests__/session.test.d.ts.map +1 -0
  474. package/dist/runtime/bidi/__tests__/session.test.js +201 -0
  475. package/dist/runtime/bidi/__tests__/session.test.js.map +1 -0
  476. package/dist/runtime/bidi/index.d.ts +5 -0
  477. package/dist/runtime/bidi/index.d.ts.map +1 -0
  478. package/dist/runtime/bidi/index.js +3 -0
  479. package/dist/runtime/bidi/index.js.map +1 -0
  480. package/dist/runtime/bidi/mock.d.ts +34 -0
  481. package/dist/runtime/bidi/mock.d.ts.map +1 -0
  482. package/dist/runtime/bidi/mock.js +60 -0
  483. package/dist/runtime/bidi/mock.js.map +1 -0
  484. package/dist/runtime/bidi/session.d.ts +44 -0
  485. package/dist/runtime/bidi/session.d.ts.map +1 -0
  486. package/dist/runtime/bidi/session.js +154 -0
  487. package/dist/runtime/bidi/session.js.map +1 -0
  488. package/dist/runtime/decision/parser.d.ts.map +1 -1
  489. package/dist/runtime/decision/parser.js +8 -1
  490. package/dist/runtime/decision/parser.js.map +1 -1
  491. package/dist/runtime/decision/parser.test.d.ts +2 -0
  492. package/dist/runtime/decision/parser.test.d.ts.map +1 -0
  493. package/dist/runtime/decision/parser.test.js +37 -0
  494. package/dist/runtime/decision/parser.test.js.map +1 -0
  495. package/dist/runtime/query/__tests__/answer-review.test.d.ts +2 -0
  496. package/dist/runtime/query/__tests__/answer-review.test.d.ts.map +1 -0
  497. package/dist/runtime/query/__tests__/answer-review.test.js +141 -0
  498. package/dist/runtime/query/__tests__/answer-review.test.js.map +1 -0
  499. package/dist/runtime/query/__tests__/capability-negotiation.test.js +24 -111
  500. package/dist/runtime/query/__tests__/capability-negotiation.test.js.map +1 -1
  501. package/dist/runtime/query/__tests__/checkpoint-store.test.js +12 -44
  502. package/dist/runtime/query/__tests__/checkpoint-store.test.js.map +1 -1
  503. package/dist/runtime/query/__tests__/checkpoint.test.js +2 -10
  504. package/dist/runtime/query/__tests__/checkpoint.test.js.map +1 -1
  505. package/dist/runtime/query/__tests__/citations.test.d.ts +2 -0
  506. package/dist/runtime/query/__tests__/citations.test.d.ts.map +1 -0
  507. package/dist/runtime/query/__tests__/citations.test.js +85 -0
  508. package/dist/runtime/query/__tests__/citations.test.js.map +1 -0
  509. package/dist/runtime/query/__tests__/content-budget.test.d.ts +2 -0
  510. package/dist/runtime/query/__tests__/content-budget.test.d.ts.map +1 -0
  511. package/dist/runtime/query/__tests__/content-budget.test.js +69 -0
  512. package/dist/runtime/query/__tests__/content-budget.test.js.map +1 -0
  513. package/dist/runtime/query/__tests__/context.test.js +3 -2
  514. package/dist/runtime/query/__tests__/context.test.js.map +1 -1
  515. package/dist/runtime/query/__tests__/deferred-tools.test.js +21 -130
  516. package/dist/runtime/query/__tests__/deferred-tools.test.js.map +1 -1
  517. package/dist/runtime/query/__tests__/delegation-resume.test.d.ts +2 -0
  518. package/dist/runtime/query/__tests__/delegation-resume.test.d.ts.map +1 -0
  519. package/dist/runtime/query/__tests__/delegation-resume.test.js +111 -0
  520. package/dist/runtime/query/__tests__/delegation-resume.test.js.map +1 -0
  521. package/dist/runtime/query/__tests__/document-attachments.test.d.ts +2 -0
  522. package/dist/runtime/query/__tests__/document-attachments.test.d.ts.map +1 -0
  523. package/dist/runtime/query/__tests__/document-attachments.test.js +92 -0
  524. package/dist/runtime/query/__tests__/document-attachments.test.js.map +1 -0
  525. package/dist/runtime/query/__tests__/durable-park-and-trace.test.d.ts +2 -0
  526. package/dist/runtime/query/__tests__/durable-park-and-trace.test.d.ts.map +1 -0
  527. package/dist/runtime/query/__tests__/durable-park-and-trace.test.js +194 -0
  528. package/dist/runtime/query/__tests__/durable-park-and-trace.test.js.map +1 -0
  529. package/dist/runtime/query/__tests__/durable-question-park.test.d.ts +2 -0
  530. package/dist/runtime/query/__tests__/durable-question-park.test.d.ts.map +1 -0
  531. package/dist/runtime/query/__tests__/durable-question-park.test.js +253 -0
  532. package/dist/runtime/query/__tests__/durable-question-park.test.js.map +1 -0
  533. package/dist/runtime/query/__tests__/durable-run-state.test.d.ts +2 -0
  534. package/dist/runtime/query/__tests__/durable-run-state.test.d.ts.map +1 -0
  535. package/dist/runtime/query/__tests__/durable-run-state.test.js +158 -0
  536. package/dist/runtime/query/__tests__/durable-run-state.test.js.map +1 -0
  537. package/dist/runtime/query/__tests__/executor-plugin-hooks.test.js +24 -20
  538. package/dist/runtime/query/__tests__/executor-plugin-hooks.test.js.map +1 -1
  539. package/dist/runtime/query/__tests__/finish-reason-vs-tool-calls.test.d.ts +2 -0
  540. package/dist/runtime/query/__tests__/finish-reason-vs-tool-calls.test.d.ts.map +1 -0
  541. package/dist/runtime/query/__tests__/finish-reason-vs-tool-calls.test.js +187 -0
  542. package/dist/runtime/query/__tests__/finish-reason-vs-tool-calls.test.js.map +1 -0
  543. package/dist/runtime/query/__tests__/grant-cannot-overrule-the-gate.test.d.ts +2 -0
  544. package/dist/runtime/query/__tests__/grant-cannot-overrule-the-gate.test.d.ts.map +1 -0
  545. package/dist/runtime/query/__tests__/grant-cannot-overrule-the-gate.test.js +95 -0
  546. package/dist/runtime/query/__tests__/grant-cannot-overrule-the-gate.test.js.map +1 -0
  547. package/dist/runtime/query/__tests__/guardrails-e2e.test.d.ts +2 -0
  548. package/dist/runtime/query/__tests__/guardrails-e2e.test.d.ts.map +1 -0
  549. package/dist/runtime/query/__tests__/guardrails-e2e.test.js +132 -0
  550. package/dist/runtime/query/__tests__/guardrails-e2e.test.js.map +1 -0
  551. package/dist/runtime/query/__tests__/guardrails.test.d.ts +2 -0
  552. package/dist/runtime/query/__tests__/guardrails.test.d.ts.map +1 -0
  553. package/dist/runtime/query/__tests__/guardrails.test.js +163 -0
  554. package/dist/runtime/query/__tests__/guardrails.test.js.map +1 -0
  555. package/dist/runtime/query/__tests__/hook-model-visibility.test.d.ts +2 -0
  556. package/dist/runtime/query/__tests__/hook-model-visibility.test.d.ts.map +1 -0
  557. package/dist/runtime/query/__tests__/hook-model-visibility.test.js +119 -0
  558. package/dist/runtime/query/__tests__/hook-model-visibility.test.js.map +1 -0
  559. package/dist/runtime/query/__tests__/long-document-flow.test.js +9 -2
  560. package/dist/runtime/query/__tests__/long-document-flow.test.js.map +1 -1
  561. package/dist/runtime/query/__tests__/memory-promotion.test.d.ts +2 -0
  562. package/dist/runtime/query/__tests__/memory-promotion.test.d.ts.map +1 -0
  563. package/dist/runtime/query/__tests__/memory-promotion.test.js +119 -0
  564. package/dist/runtime/query/__tests__/memory-promotion.test.js.map +1 -0
  565. package/dist/runtime/query/__tests__/orphan-task-turn-end.test.js +5 -27
  566. package/dist/runtime/query/__tests__/orphan-task-turn-end.test.js.map +1 -1
  567. package/dist/runtime/query/__tests__/overflow-relief.test.d.ts +2 -0
  568. package/dist/runtime/query/__tests__/overflow-relief.test.d.ts.map +1 -0
  569. package/dist/runtime/query/__tests__/overflow-relief.test.js +119 -0
  570. package/dist/runtime/query/__tests__/overflow-relief.test.js.map +1 -0
  571. package/dist/runtime/query/__tests__/partial-batch-resume.test.d.ts +2 -0
  572. package/dist/runtime/query/__tests__/partial-batch-resume.test.d.ts.map +1 -0
  573. package/dist/runtime/query/__tests__/partial-batch-resume.test.js +205 -0
  574. package/dist/runtime/query/__tests__/partial-batch-resume.test.js.map +1 -0
  575. package/dist/runtime/query/__tests__/plugin-hooks.test.js +0 -1
  576. package/dist/runtime/query/__tests__/plugin-hooks.test.js.map +1 -1
  577. package/dist/runtime/query/__tests__/prepare-step-chain.test.d.ts +2 -0
  578. package/dist/runtime/query/__tests__/prepare-step-chain.test.d.ts.map +1 -0
  579. package/dist/runtime/query/__tests__/prepare-step-chain.test.js +104 -0
  580. package/dist/runtime/query/__tests__/prepare-step-chain.test.js.map +1 -0
  581. package/dist/runtime/query/__tests__/prepare-step.test.d.ts +2 -0
  582. package/dist/runtime/query/__tests__/prepare-step.test.d.ts.map +1 -0
  583. package/dist/runtime/query/__tests__/prepare-step.test.js +194 -0
  584. package/dist/runtime/query/__tests__/prepare-step.test.js.map +1 -0
  585. package/dist/runtime/query/__tests__/probe-veto-is-error.test.d.ts +2 -0
  586. package/dist/runtime/query/__tests__/probe-veto-is-error.test.d.ts.map +1 -0
  587. package/dist/runtime/query/__tests__/probe-veto-is-error.test.js +111 -0
  588. package/dist/runtime/query/__tests__/probe-veto-is-error.test.js.map +1 -0
  589. package/dist/runtime/query/__tests__/prompt.test.js +1 -1
  590. package/dist/runtime/query/__tests__/prompt.test.js.map +1 -1
  591. package/dist/runtime/query/__tests__/provider-fault-pauses.test.d.ts +2 -0
  592. package/dist/runtime/query/__tests__/provider-fault-pauses.test.d.ts.map +1 -0
  593. package/dist/runtime/query/__tests__/provider-fault-pauses.test.js +86 -0
  594. package/dist/runtime/query/__tests__/provider-fault-pauses.test.js.map +1 -0
  595. package/dist/runtime/query/__tests__/reasoning-stream.test.d.ts +2 -0
  596. package/dist/runtime/query/__tests__/reasoning-stream.test.d.ts.map +1 -0
  597. package/dist/runtime/query/__tests__/reasoning-stream.test.js +146 -0
  598. package/dist/runtime/query/__tests__/reasoning-stream.test.js.map +1 -0
  599. package/dist/runtime/query/__tests__/recoverable-settle.test.d.ts +2 -0
  600. package/dist/runtime/query/__tests__/recoverable-settle.test.d.ts.map +1 -0
  601. package/dist/runtime/query/__tests__/recoverable-settle.test.js +119 -0
  602. package/dist/runtime/query/__tests__/recoverable-settle.test.js.map +1 -0
  603. package/dist/runtime/query/__tests__/resume-budget.test.d.ts +2 -0
  604. package/dist/runtime/query/__tests__/resume-budget.test.d.ts.map +1 -0
  605. package/dist/runtime/query/__tests__/resume-budget.test.js +130 -0
  606. package/dist/runtime/query/__tests__/resume-budget.test.js.map +1 -0
  607. package/dist/runtime/query/__tests__/resume-pending.test.d.ts +2 -0
  608. package/dist/runtime/query/__tests__/resume-pending.test.d.ts.map +1 -0
  609. package/dist/runtime/query/__tests__/resume-pending.test.js +226 -0
  610. package/dist/runtime/query/__tests__/resume-pending.test.js.map +1 -0
  611. package/dist/runtime/query/__tests__/rich-content-budget-wiring.test.d.ts +2 -0
  612. package/dist/runtime/query/__tests__/rich-content-budget-wiring.test.d.ts.map +1 -0
  613. package/dist/runtime/query/__tests__/rich-content-budget-wiring.test.js +89 -0
  614. package/dist/runtime/query/__tests__/rich-content-budget-wiring.test.js.map +1 -0
  615. package/dist/runtime/query/__tests__/stop-condition.test.d.ts +2 -0
  616. package/dist/runtime/query/__tests__/stop-condition.test.d.ts.map +1 -0
  617. package/dist/runtime/query/__tests__/stop-condition.test.js +205 -0
  618. package/dist/runtime/query/__tests__/stop-condition.test.js.map +1 -0
  619. package/dist/runtime/query/__tests__/structured-output.test.d.ts +2 -0
  620. package/dist/runtime/query/__tests__/structured-output.test.d.ts.map +1 -0
  621. package/dist/runtime/query/__tests__/structured-output.test.js +219 -0
  622. package/dist/runtime/query/__tests__/structured-output.test.js.map +1 -0
  623. package/dist/runtime/query/__tests__/structured-run-failure.test.d.ts +2 -0
  624. package/dist/runtime/query/__tests__/structured-run-failure.test.d.ts.map +1 -0
  625. package/dist/runtime/query/__tests__/structured-run-failure.test.js +117 -0
  626. package/dist/runtime/query/__tests__/structured-run-failure.test.js.map +1 -0
  627. package/dist/runtime/query/__tests__/terminal-tool.test.d.ts +2 -0
  628. package/dist/runtime/query/__tests__/terminal-tool.test.d.ts.map +1 -0
  629. package/dist/runtime/query/__tests__/terminal-tool.test.js +119 -0
  630. package/dist/runtime/query/__tests__/terminal-tool.test.js.map +1 -0
  631. package/dist/runtime/query/__tests__/tool-grants.test.d.ts +2 -0
  632. package/dist/runtime/query/__tests__/tool-grants.test.d.ts.map +1 -0
  633. package/dist/runtime/query/__tests__/tool-grants.test.js +85 -0
  634. package/dist/runtime/query/__tests__/tool-grants.test.js.map +1 -0
  635. package/dist/runtime/query/__tests__/tool-output-budget.test.d.ts +2 -0
  636. package/dist/runtime/query/__tests__/tool-output-budget.test.d.ts.map +1 -0
  637. package/dist/runtime/query/__tests__/tool-output-budget.test.js +0 -0
  638. package/dist/runtime/query/__tests__/tool-output-budget.test.js.map +1 -0
  639. package/dist/runtime/query/__tests__/tool-pause.test.d.ts +2 -0
  640. package/dist/runtime/query/__tests__/tool-pause.test.d.ts.map +1 -0
  641. package/dist/runtime/query/__tests__/tool-pause.test.js +165 -0
  642. package/dist/runtime/query/__tests__/tool-pause.test.js.map +1 -0
  643. package/dist/runtime/query/__tests__/tool-progress.test.d.ts +2 -0
  644. package/dist/runtime/query/__tests__/tool-progress.test.d.ts.map +1 -0
  645. package/dist/runtime/query/__tests__/tool-progress.test.js +122 -0
  646. package/dist/runtime/query/__tests__/tool-progress.test.js.map +1 -0
  647. package/dist/runtime/query/__tests__/tool-repair-retry.test.d.ts +2 -0
  648. package/dist/runtime/query/__tests__/tool-repair-retry.test.d.ts.map +1 -0
  649. package/dist/runtime/query/__tests__/tool-repair-retry.test.js +196 -0
  650. package/dist/runtime/query/__tests__/tool-repair-retry.test.js.map +1 -0
  651. package/dist/runtime/query/__tests__/tool-result-wire.test.d.ts +2 -0
  652. package/dist/runtime/query/__tests__/tool-result-wire.test.d.ts.map +1 -0
  653. package/dist/runtime/query/__tests__/tool-result-wire.test.js +106 -0
  654. package/dist/runtime/query/__tests__/tool-result-wire.test.js.map +1 -0
  655. package/dist/runtime/query/__tests__/tool-timeout.test.d.ts +2 -0
  656. package/dist/runtime/query/__tests__/tool-timeout.test.d.ts.map +1 -0
  657. package/dist/runtime/query/__tests__/tool-timeout.test.js +184 -0
  658. package/dist/runtime/query/__tests__/tool-timeout.test.js.map +1 -0
  659. package/dist/runtime/query/__tests__/trace-parenting.test.d.ts +2 -0
  660. package/dist/runtime/query/__tests__/trace-parenting.test.d.ts.map +1 -0
  661. package/dist/runtime/query/__tests__/trace-parenting.test.js +111 -0
  662. package/dist/runtime/query/__tests__/trace-parenting.test.js.map +1 -0
  663. package/dist/runtime/query/__tests__/wiring-seams.test.d.ts +2 -0
  664. package/dist/runtime/query/__tests__/wiring-seams.test.d.ts.map +1 -0
  665. package/dist/runtime/query/__tests__/wiring-seams.test.js +237 -0
  666. package/dist/runtime/query/__tests__/wiring-seams.test.js.map +1 -0
  667. package/dist/runtime/query/checkpoint.d.ts +124 -6
  668. package/dist/runtime/query/checkpoint.d.ts.map +1 -1
  669. package/dist/runtime/query/checkpoint.js +207 -6
  670. package/dist/runtime/query/checkpoint.js.map +1 -1
  671. package/dist/runtime/query/events.js +2 -2
  672. package/dist/runtime/query/events.js.map +1 -1
  673. package/dist/runtime/query/executor.d.ts +224 -7
  674. package/dist/runtime/query/executor.d.ts.map +1 -1
  675. package/dist/runtime/query/executor.js +662 -35
  676. package/dist/runtime/query/executor.js.map +1 -1
  677. package/dist/runtime/query/guard.d.ts +18 -0
  678. package/dist/runtime/query/guard.d.ts.map +1 -1
  679. package/dist/runtime/query/guard.js +16 -1
  680. package/dist/runtime/query/guard.js.map +1 -1
  681. package/dist/runtime/query/guardrail-presets.d.ts +36 -0
  682. package/dist/runtime/query/guardrail-presets.d.ts.map +1 -0
  683. package/dist/runtime/query/guardrail-presets.js +96 -0
  684. package/dist/runtime/query/guardrail-presets.js.map +1 -0
  685. package/dist/runtime/query/guardrails.d.ts +35 -0
  686. package/dist/runtime/query/guardrails.d.ts.map +1 -0
  687. package/dist/runtime/query/guardrails.js +102 -0
  688. package/dist/runtime/query/guardrails.js.map +1 -0
  689. package/dist/runtime/query/index.d.ts +213 -2
  690. package/dist/runtime/query/index.d.ts.map +1 -1
  691. package/dist/runtime/query/index.js +458 -10
  692. package/dist/runtime/query/index.js.map +1 -1
  693. package/dist/runtime/query/iteration/__tests__/cancel-settles-turn.test.d.ts +2 -0
  694. package/dist/runtime/query/iteration/__tests__/cancel-settles-turn.test.d.ts.map +1 -0
  695. package/dist/runtime/query/iteration/__tests__/cancel-settles-turn.test.js +131 -0
  696. package/dist/runtime/query/iteration/__tests__/cancel-settles-turn.test.js.map +1 -0
  697. package/dist/runtime/query/iteration/index.d.ts +76 -1
  698. package/dist/runtime/query/iteration/index.d.ts.map +1 -1
  699. package/dist/runtime/query/iteration/index.js +577 -34
  700. package/dist/runtime/query/iteration/index.js.map +1 -1
  701. package/dist/runtime/query/iteration/phases/advisory.d.ts.map +1 -1
  702. package/dist/runtime/query/iteration/phases/advisory.js +14 -2
  703. package/dist/runtime/query/iteration/phases/advisory.js.map +1 -1
  704. package/dist/runtime/query/iteration/phases/advisory.test.d.ts +1 -1
  705. package/dist/runtime/query/iteration/phases/advisory.test.js +4 -1
  706. package/dist/runtime/query/iteration/phases/advisory.test.js.map +1 -1
  707. package/dist/runtime/query/iteration/phases/checkpoint.js +2 -2
  708. package/dist/runtime/query/iteration/phases/checkpoint.js.map +1 -1
  709. package/dist/runtime/query/iteration/phases/compaction-safe-cut.test.js +1 -1
  710. package/dist/runtime/query/iteration/phases/compaction-safe-cut.test.js.map +1 -1
  711. package/dist/runtime/query/iteration/phases/compaction.d.ts +22 -1
  712. package/dist/runtime/query/iteration/phases/compaction.d.ts.map +1 -1
  713. package/dist/runtime/query/iteration/phases/compaction.js +297 -64
  714. package/dist/runtime/query/iteration/phases/compaction.js.map +1 -1
  715. package/dist/runtime/query/iteration/phases/context-measurement.test.d.ts +21 -0
  716. package/dist/runtime/query/iteration/phases/context-measurement.test.d.ts.map +1 -0
  717. package/dist/runtime/query/iteration/phases/context-measurement.test.js +161 -0
  718. package/dist/runtime/query/iteration/phases/context-measurement.test.js.map +1 -0
  719. package/dist/runtime/query/iteration/phases/context.d.ts +62 -2
  720. package/dist/runtime/query/iteration/phases/context.d.ts.map +1 -1
  721. package/dist/runtime/query/iteration/phases/context.js +87 -0
  722. package/dist/runtime/query/iteration/phases/context.js.map +1 -1
  723. package/dist/runtime/query/iteration/phases/hitl-answer-question.test.js +3 -1
  724. package/dist/runtime/query/iteration/phases/hitl-answer-question.test.js.map +1 -1
  725. package/dist/runtime/query/iteration/phases/overflow-escalation.test.d.ts +15 -0
  726. package/dist/runtime/query/iteration/phases/overflow-escalation.test.d.ts.map +1 -0
  727. package/dist/runtime/query/iteration/phases/overflow-escalation.test.js +130 -0
  728. package/dist/runtime/query/iteration/phases/overflow-escalation.test.js.map +1 -0
  729. package/dist/runtime/query/iteration/phases/plan.d.ts.map +1 -1
  730. package/dist/runtime/query/iteration/phases/plan.js +8 -2
  731. package/dist/runtime/query/iteration/phases/plan.js.map +1 -1
  732. package/dist/runtime/query/iteration/phases/retained-messages.test.d.ts +2 -0
  733. package/dist/runtime/query/iteration/phases/retained-messages.test.d.ts.map +1 -0
  734. package/dist/runtime/query/iteration/phases/retained-messages.test.js +180 -0
  735. package/dist/runtime/query/iteration/phases/retained-messages.test.js.map +1 -0
  736. package/dist/runtime/query/iteration/phases/tool-review-denials.test.d.ts +2 -0
  737. package/dist/runtime/query/iteration/phases/tool-review-denials.test.d.ts.map +1 -0
  738. package/dist/runtime/query/iteration/phases/tool-review-denials.test.js +224 -0
  739. package/dist/runtime/query/iteration/phases/tool-review-denials.test.js.map +1 -0
  740. package/dist/runtime/query/iteration/phases/tool-review.d.ts +29 -1
  741. package/dist/runtime/query/iteration/phases/tool-review.d.ts.map +1 -1
  742. package/dist/runtime/query/iteration/phases/tool-review.js +114 -44
  743. package/dist/runtime/query/iteration/phases/tool-review.js.map +1 -1
  744. package/dist/runtime/query/iteration/phases/working-memory-compaction.test.js +20 -9
  745. package/dist/runtime/query/iteration/phases/working-memory-compaction.test.js.map +1 -1
  746. package/dist/runtime/query/iteration/stream-turn.d.ts +2 -26
  747. package/dist/runtime/query/iteration/stream-turn.d.ts.map +1 -1
  748. package/dist/runtime/query/iteration/stream-turn.js +254 -16
  749. package/dist/runtime/query/iteration/stream-turn.js.map +1 -1
  750. package/dist/runtime/query/plugin-hooks.d.ts.map +1 -1
  751. package/dist/runtime/query/plugin-hooks.js +6 -2
  752. package/dist/runtime/query/plugin-hooks.js.map +1 -1
  753. package/dist/runtime/query/prompt.d.ts.map +1 -1
  754. package/dist/runtime/query/prompt.js +25 -1
  755. package/dist/runtime/query/prompt.js.map +1 -1
  756. package/dist/runtime/query/question-park.d.ts +85 -0
  757. package/dist/runtime/query/question-park.d.ts.map +1 -0
  758. package/dist/runtime/query/question-park.js +71 -0
  759. package/dist/runtime/query/question-park.js.map +1 -0
  760. package/dist/runtime/query/result.d.ts +9 -0
  761. package/dist/runtime/query/result.d.ts.map +1 -1
  762. package/dist/runtime/query/result.js +49 -0
  763. package/dist/runtime/query/result.js.map +1 -1
  764. package/dist/runtime/query/resume-pending.d.ts +106 -0
  765. package/dist/runtime/query/resume-pending.d.ts.map +1 -0
  766. package/dist/runtime/query/resume-pending.js +312 -0
  767. package/dist/runtime/query/resume-pending.js.map +1 -0
  768. package/dist/runtime/query/run-state.d.ts +52 -0
  769. package/dist/runtime/query/run-state.d.ts.map +1 -0
  770. package/dist/runtime/query/run-state.js +87 -0
  771. package/dist/runtime/query/run-state.js.map +1 -0
  772. package/dist/runtime/query/tool-grants.d.ts +44 -0
  773. package/dist/runtime/query/tool-grants.d.ts.map +1 -0
  774. package/dist/runtime/query/tool-grants.js +63 -0
  775. package/dist/runtime/query/tool-grants.js.map +1 -0
  776. package/dist/runtime/query/tool-output-budget.d.ts +83 -0
  777. package/dist/runtime/query/tool-output-budget.d.ts.map +1 -0
  778. package/dist/runtime/query/tool-output-budget.js +147 -0
  779. package/dist/runtime/query/tool-output-budget.js.map +1 -0
  780. package/dist/runtime/query/tool-pause.d.ts +41 -0
  781. package/dist/runtime/query/tool-pause.d.ts.map +1 -0
  782. package/dist/runtime/query/tool-pause.js +94 -0
  783. package/dist/runtime/query/tool-pause.js.map +1 -0
  784. package/dist/runtime/query/tooling.d.ts +10 -1
  785. package/dist/runtime/query/tooling.d.ts.map +1 -1
  786. package/dist/runtime/query/tooling.js +13 -0
  787. package/dist/runtime/query/tooling.js.map +1 -1
  788. package/dist/sandbox/__tests__/isolation.test.d.ts +2 -0
  789. package/dist/sandbox/__tests__/isolation.test.d.ts.map +1 -0
  790. package/dist/sandbox/__tests__/isolation.test.js +139 -0
  791. package/dist/sandbox/__tests__/isolation.test.js.map +1 -0
  792. package/dist/sandbox/__tests__/output-truncation.test.d.ts +2 -0
  793. package/dist/sandbox/__tests__/output-truncation.test.d.ts.map +1 -0
  794. package/dist/sandbox/__tests__/output-truncation.test.js +88 -0
  795. package/dist/sandbox/__tests__/output-truncation.test.js.map +1 -0
  796. package/dist/sandbox/__tests__/resource-limits.test.d.ts +2 -0
  797. package/dist/sandbox/__tests__/resource-limits.test.d.ts.map +1 -0
  798. package/dist/sandbox/__tests__/resource-limits.test.js +86 -0
  799. package/dist/sandbox/__tests__/resource-limits.test.js.map +1 -0
  800. package/dist/sandbox/factory.js +1 -1
  801. package/dist/sandbox/factory.js.map +1 -1
  802. package/dist/sandbox/index.d.ts +2 -0
  803. package/dist/sandbox/index.d.ts.map +1 -1
  804. package/dist/sandbox/index.js +1 -0
  805. package/dist/sandbox/index.js.map +1 -1
  806. package/dist/sandbox/isolation.d.ts +15 -0
  807. package/dist/sandbox/isolation.d.ts.map +1 -0
  808. package/dist/sandbox/isolation.js +53 -0
  809. package/dist/sandbox/isolation.js.map +1 -0
  810. package/dist/sandbox/provider/local.d.ts +51 -2
  811. package/dist/sandbox/provider/local.d.ts.map +1 -1
  812. package/dist/sandbox/provider/local.js +139 -69
  813. package/dist/sandbox/provider/local.js.map +1 -1
  814. package/dist/session/__tests__/integration/spawn-rollback.test.d.ts +1 -1
  815. package/dist/session/__tests__/integration/spawn-rollback.test.js +1 -1
  816. package/dist/session/handoff/broadcast.js +3 -3
  817. package/dist/session/handoff/broadcast.js.map +1 -1
  818. package/dist/session/migration/__tests__/filesystem.test.js +5 -1
  819. package/dist/session/migration/__tests__/filesystem.test.js.map +1 -1
  820. package/dist/session/migration/filesystem.d.ts.map +1 -1
  821. package/dist/session/migration/filesystem.js +7 -13
  822. package/dist/session/migration/filesystem.js.map +1 -1
  823. package/dist/session/migration/marker.d.ts.map +1 -1
  824. package/dist/session/migration/marker.js +5 -10
  825. package/dist/session/migration/marker.js.map +1 -1
  826. package/dist/session/retention/disk-backend.d.ts.map +1 -1
  827. package/dist/session/retention/disk-backend.js +5 -12
  828. package/dist/session/retention/disk-backend.js.map +1 -1
  829. package/dist/session/workspace/__tests__/git-worktree.test.js +6 -2
  830. package/dist/session/workspace/__tests__/git-worktree.test.js.map +1 -1
  831. package/dist/session/workspace/__tests__/shared-run.test.js +15 -0
  832. package/dist/session/workspace/__tests__/shared-run.test.js.map +1 -1
  833. package/dist/session/workspace/shared-run.js +8 -1
  834. package/dist/session/workspace/shared-run.js.map +1 -1
  835. package/dist/skills/__tests__/frontmatter.test.d.ts +2 -0
  836. package/dist/skills/__tests__/frontmatter.test.d.ts.map +1 -0
  837. package/dist/skills/__tests__/frontmatter.test.js +106 -0
  838. package/dist/skills/__tests__/frontmatter.test.js.map +1 -0
  839. package/dist/skills/loader.d.ts.map +1 -1
  840. package/dist/skills/loader.js +42 -3
  841. package/dist/skills/loader.js.map +1 -1
  842. package/dist/store/__tests__/checkpoint-integrity.test.d.ts +2 -0
  843. package/dist/store/__tests__/checkpoint-integrity.test.d.ts.map +1 -0
  844. package/dist/store/__tests__/checkpoint-integrity.test.js +112 -0
  845. package/dist/store/__tests__/checkpoint-integrity.test.js.map +1 -0
  846. package/dist/store/__tests__/checkpoint-versioning.test.d.ts +2 -0
  847. package/dist/store/__tests__/checkpoint-versioning.test.d.ts.map +1 -0
  848. package/dist/store/__tests__/checkpoint-versioning.test.js +151 -0
  849. package/dist/store/__tests__/checkpoint-versioning.test.js.map +1 -0
  850. package/dist/store/__tests__/schema-adoption.test.d.ts +2 -0
  851. package/dist/store/__tests__/schema-adoption.test.d.ts.map +1 -0
  852. package/dist/store/__tests__/schema-adoption.test.js +90 -0
  853. package/dist/store/__tests__/schema-adoption.test.js.map +1 -0
  854. package/dist/store/__tests__/schema.test.d.ts +2 -0
  855. package/dist/store/__tests__/schema.test.d.ts.map +1 -0
  856. package/dist/store/__tests__/schema.test.js +93 -0
  857. package/dist/store/__tests__/schema.test.js.map +1 -0
  858. package/dist/store/memory/disk.d.ts.map +1 -1
  859. package/dist/store/memory/disk.js +16 -16
  860. package/dist/store/memory/disk.js.map +1 -1
  861. package/dist/store/run/disk.d.ts +23 -0
  862. package/dist/store/run/disk.d.ts.map +1 -1
  863. package/dist/store/run/disk.js +148 -23
  864. package/dist/store/run/disk.js.map +1 -1
  865. package/dist/store/schema.d.ts +73 -0
  866. package/dist/store/schema.d.ts.map +1 -0
  867. package/dist/store/schema.js +103 -0
  868. package/dist/store/schema.js.map +1 -0
  869. package/dist/store/session/disk.d.ts.map +1 -1
  870. package/dist/store/session/disk.js +25 -13
  871. package/dist/store/session/disk.js.map +1 -1
  872. package/dist/store/task/__tests__/lookup-across-runs.test.d.ts +2 -0
  873. package/dist/store/task/__tests__/lookup-across-runs.test.d.ts.map +1 -0
  874. package/dist/store/task/__tests__/lookup-across-runs.test.js +60 -0
  875. package/dist/store/task/__tests__/lookup-across-runs.test.js.map +1 -0
  876. package/dist/store/task/disk.d.ts +18 -0
  877. package/dist/store/task/disk.d.ts.map +1 -1
  878. package/dist/store/task/disk.js +61 -16
  879. package/dist/store/task/disk.js.map +1 -1
  880. package/dist/store/thread/disk.d.ts.map +1 -1
  881. package/dist/store/thread/disk.js +20 -11
  882. package/dist/store/thread/disk.js.map +1 -1
  883. package/dist/streaming/coalesce.d.ts +1 -1
  884. package/dist/streaming/coalesce.js +1 -1
  885. package/dist/telemetry/__tests__/latency-instruments.test.d.ts +2 -0
  886. package/dist/telemetry/__tests__/latency-instruments.test.d.ts.map +1 -0
  887. package/dist/telemetry/__tests__/latency-instruments.test.js +82 -0
  888. package/dist/telemetry/__tests__/latency-instruments.test.js.map +1 -0
  889. package/dist/telemetry/__tests__/metrics.test.d.ts +2 -0
  890. package/dist/telemetry/__tests__/metrics.test.d.ts.map +1 -0
  891. package/dist/telemetry/__tests__/metrics.test.js +112 -0
  892. package/dist/telemetry/__tests__/metrics.test.js.map +1 -0
  893. package/dist/telemetry/__tests__/span-hierarchy.test.d.ts +2 -0
  894. package/dist/telemetry/__tests__/span-hierarchy.test.d.ts.map +1 -0
  895. package/dist/telemetry/__tests__/span-hierarchy.test.js +84 -0
  896. package/dist/telemetry/__tests__/span-hierarchy.test.js.map +1 -0
  897. package/dist/telemetry/attributes.d.ts +44 -0
  898. package/dist/telemetry/attributes.d.ts.map +1 -1
  899. package/dist/telemetry/attributes.js +80 -0
  900. package/dist/telemetry/attributes.js.map +1 -1
  901. package/dist/telemetry/metrics.d.ts +41 -0
  902. package/dist/telemetry/metrics.d.ts.map +1 -0
  903. package/dist/telemetry/metrics.js +160 -0
  904. package/dist/telemetry/metrics.js.map +1 -0
  905. package/dist/test-setup.d.ts +3 -3
  906. package/dist/test-setup.js +3 -3
  907. package/dist/test-support/paths.d.ts +14 -0
  908. package/dist/test-support/paths.d.ts.map +1 -0
  909. package/dist/test-support/paths.js +16 -0
  910. package/dist/test-support/paths.js.map +1 -0
  911. package/dist/tools/__tests__/path-containment.test.d.ts +2 -0
  912. package/dist/tools/__tests__/path-containment.test.d.ts.map +1 -0
  913. package/dist/tools/__tests__/path-containment.test.js +133 -0
  914. package/dist/tools/__tests__/path-containment.test.js.map +1 -0
  915. package/dist/tools/__tests__/posix-path.test.d.ts +2 -0
  916. package/dist/tools/__tests__/posix-path.test.d.ts.map +1 -0
  917. package/dist/tools/__tests__/posix-path.test.js +67 -0
  918. package/dist/tools/__tests__/posix-path.test.js.map +1 -0
  919. package/dist/tools/__tests__/sandboxed-search.test.d.ts +2 -0
  920. package/dist/tools/__tests__/sandboxed-search.test.d.ts.map +1 -0
  921. package/dist/tools/__tests__/sandboxed-search.test.js +114 -0
  922. package/dist/tools/__tests__/sandboxed-search.test.js.map +1 -0
  923. package/dist/tools/builtins/__tests__/computer-use.test.js +12 -2
  924. package/dist/tools/builtins/__tests__/computer-use.test.js.map +1 -1
  925. package/dist/tools/builtins/__tests__/edit.test.js +10 -216
  926. package/dist/tools/builtins/__tests__/edit.test.js.map +1 -1
  927. package/dist/tools/builtins/__tests__/payload-budget.test.js +16 -30
  928. package/dist/tools/builtins/__tests__/payload-budget.test.js.map +1 -1
  929. package/dist/tools/builtins/__tests__/read-file.test.js +23 -1
  930. package/dist/tools/builtins/__tests__/read-file.test.js.map +1 -1
  931. package/dist/tools/builtins/__tests__/write-file.test.js +13 -37
  932. package/dist/tools/builtins/__tests__/write-file.test.js.map +1 -1
  933. package/dist/tools/builtins/bash.d.ts.map +1 -1
  934. package/dist/tools/builtins/bash.js +33 -4
  935. package/dist/tools/builtins/bash.js.map +1 -1
  936. package/dist/tools/builtins/computer-use.d.ts +22 -22
  937. package/dist/tools/builtins/computer-use.d.ts.map +1 -1
  938. package/dist/tools/builtins/computer-use.js +7 -1
  939. package/dist/tools/builtins/computer-use.js.map +1 -1
  940. package/dist/tools/builtins/edit.d.ts +5 -2
  941. package/dist/tools/builtins/edit.d.ts.map +1 -1
  942. package/dist/tools/builtins/edit.js +125 -105
  943. package/dist/tools/builtins/edit.js.map +1 -1
  944. package/dist/tools/builtins/glob.d.ts +1 -1
  945. package/dist/tools/builtins/glob.d.ts.map +1 -1
  946. package/dist/tools/builtins/glob.js +63 -6
  947. package/dist/tools/builtins/glob.js.map +1 -1
  948. package/dist/tools/builtins/grep.d.ts.map +1 -1
  949. package/dist/tools/builtins/grep.js +61 -8
  950. package/dist/tools/builtins/grep.js.map +1 -1
  951. package/dist/tools/builtins/index.js +6 -6
  952. package/dist/tools/builtins/index.js.map +1 -1
  953. package/dist/tools/builtins/ls.d.ts.map +1 -1
  954. package/dist/tools/builtins/ls.js +4 -2
  955. package/dist/tools/builtins/ls.js.map +1 -1
  956. package/dist/tools/builtins/read-file.d.ts +1 -1
  957. package/dist/tools/builtins/read-file.d.ts.map +1 -1
  958. package/dist/tools/builtins/read-file.js +38 -4
  959. package/dist/tools/builtins/read-file.js.map +1 -1
  960. package/dist/tools/builtins/write-file.d.ts +2 -1
  961. package/dist/tools/builtins/write-file.d.ts.map +1 -1
  962. package/dist/tools/builtins/write-file.js +40 -66
  963. package/dist/tools/builtins/write-file.js.map +1 -1
  964. package/dist/tools/coordinator/__tests__/agent.test.d.ts +1 -1
  965. package/dist/tools/coordinator/__tests__/agent.test.js +2 -2
  966. package/dist/tools/coordinator/__tests__/agent.test.js.map +1 -1
  967. package/dist/tools/coordinator/agent.d.ts +20 -9
  968. package/dist/tools/coordinator/agent.d.ts.map +1 -1
  969. package/dist/tools/coordinator/agent.js +2 -1
  970. package/dist/tools/coordinator/agent.js.map +1 -1
  971. package/dist/tools/coordinator/index.d.ts +20 -0
  972. package/dist/tools/coordinator/index.d.ts.map +1 -1
  973. package/dist/tools/coordinator/index.js +80 -32
  974. package/dist/tools/coordinator/index.js.map +1 -1
  975. package/dist/tools/defineTool.d.ts +16 -0
  976. package/dist/tools/defineTool.d.ts.map +1 -1
  977. package/dist/tools/defineTool.js +4 -0
  978. package/dist/tools/defineTool.js.map +1 -1
  979. package/dist/tools/glob-match.d.ts +18 -0
  980. package/dist/tools/glob-match.d.ts.map +1 -0
  981. package/dist/tools/glob-match.js +50 -0
  982. package/dist/tools/glob-match.js.map +1 -0
  983. package/dist/tools/paths.d.ts +20 -0
  984. package/dist/tools/paths.d.ts.map +1 -0
  985. package/dist/tools/paths.js +41 -0
  986. package/dist/tools/paths.js.map +1 -0
  987. package/dist/tools/posix-path.d.ts +24 -0
  988. package/dist/tools/posix-path.d.ts.map +1 -0
  989. package/dist/tools/posix-path.js +67 -0
  990. package/dist/tools/posix-path.js.map +1 -0
  991. package/dist/types/advisory/config.d.ts +21 -3
  992. package/dist/types/advisory/config.d.ts.map +1 -1
  993. package/dist/types/advisory/result.d.ts +12 -10
  994. package/dist/types/advisory/result.d.ts.map +1 -1
  995. package/dist/types/agent/base.d.ts +20 -2
  996. package/dist/types/agent/base.d.ts.map +1 -1
  997. package/dist/types/agent/factory.d.ts +15 -9
  998. package/dist/types/agent/factory.d.ts.map +1 -1
  999. package/dist/types/agent/gateway.d.ts +18 -0
  1000. package/dist/types/agent/gateway.d.ts.map +1 -1
  1001. package/dist/types/agent/manager.d.ts +19 -0
  1002. package/dist/types/agent/manager.d.ts.map +1 -1
  1003. package/dist/types/agent/reactive.d.ts +48 -0
  1004. package/dist/types/agent/reactive.d.ts.map +1 -1
  1005. package/dist/types/agent/router.d.ts +11 -0
  1006. package/dist/types/agent/router.d.ts.map +1 -1
  1007. package/dist/types/agent/supervisor.d.ts +2 -2
  1008. package/dist/types/agent/task.d.ts +24 -0
  1009. package/dist/types/agent/task.d.ts.map +1 -1
  1010. package/dist/types/bidi/index.d.ts +140 -0
  1011. package/dist/types/bidi/index.d.ts.map +1 -0
  1012. package/dist/types/bidi/index.js +2 -0
  1013. package/dist/types/bidi/index.js.map +1 -0
  1014. package/dist/types/common/__tests__/numeric-correctness.test.d.ts +2 -0
  1015. package/dist/types/common/__tests__/numeric-correctness.test.d.ts.map +1 -0
  1016. package/dist/types/common/__tests__/numeric-correctness.test.js +108 -0
  1017. package/dist/types/common/__tests__/numeric-correctness.test.js.map +1 -0
  1018. package/dist/types/common/index.d.ts +1 -1
  1019. package/dist/types/common/index.d.ts.map +1 -1
  1020. package/dist/types/common/index.js +15 -4
  1021. package/dist/types/common/index.js.map +1 -1
  1022. package/dist/types/connector/mcp.d.ts +29 -0
  1023. package/dist/types/connector/mcp.d.ts.map +1 -1
  1024. package/dist/types/errors/__tests__/catalog.test.d.ts +2 -0
  1025. package/dist/types/errors/__tests__/catalog.test.d.ts.map +1 -0
  1026. package/dist/types/errors/__tests__/catalog.test.js +106 -0
  1027. package/dist/types/errors/__tests__/catalog.test.js.map +1 -0
  1028. package/dist/types/errors/__tests__/errors.test.d.ts +2 -0
  1029. package/dist/types/errors/__tests__/errors.test.d.ts.map +1 -0
  1030. package/dist/types/errors/__tests__/errors.test.js +104 -0
  1031. package/dist/types/errors/__tests__/errors.test.js.map +1 -0
  1032. package/dist/types/errors/catalog.d.ts +83 -0
  1033. package/dist/types/errors/catalog.d.ts.map +1 -0
  1034. package/dist/types/errors/catalog.js +148 -0
  1035. package/dist/types/errors/catalog.js.map +1 -0
  1036. package/dist/types/errors/index.d.ts +73 -0
  1037. package/dist/types/errors/index.d.ts.map +1 -0
  1038. package/dist/types/errors/index.js +70 -0
  1039. package/dist/types/errors/index.js.map +1 -0
  1040. package/dist/types/guardrail/index.d.ts +57 -0
  1041. package/dist/types/guardrail/index.d.ts.map +1 -0
  1042. package/dist/types/guardrail/index.js +2 -0
  1043. package/dist/types/guardrail/index.js.map +1 -0
  1044. package/dist/types/hitl/index.d.ts +97 -14
  1045. package/dist/types/hitl/index.d.ts.map +1 -1
  1046. package/dist/types/hitl/index.js.map +1 -1
  1047. package/dist/types/ids/index.d.ts +2 -2
  1048. package/dist/types/message/content.d.ts +22 -0
  1049. package/dist/types/message/content.d.ts.map +1 -0
  1050. package/dist/types/message/content.js +49 -0
  1051. package/dist/types/message/content.js.map +1 -0
  1052. package/dist/types/message/index.d.ts +193 -8
  1053. package/dist/types/message/index.d.ts.map +1 -1
  1054. package/dist/types/message/index.js +19 -4
  1055. package/dist/types/message/index.js.map +1 -1
  1056. package/dist/types/plugin/index.d.ts +78 -3
  1057. package/dist/types/plugin/index.d.ts.map +1 -1
  1058. package/dist/types/plugin/index.js +0 -1
  1059. package/dist/types/plugin/index.js.map +1 -1
  1060. package/dist/types/provider/__tests__/cause-chain.test.d.ts +2 -0
  1061. package/dist/types/provider/__tests__/cause-chain.test.d.ts.map +1 -0
  1062. package/dist/types/provider/__tests__/cause-chain.test.js +108 -0
  1063. package/dist/types/provider/__tests__/cause-chain.test.js.map +1 -0
  1064. package/dist/types/provider/__tests__/declared-retryable.test.d.ts +2 -0
  1065. package/dist/types/provider/__tests__/declared-retryable.test.d.ts.map +1 -0
  1066. package/dist/types/provider/__tests__/declared-retryable.test.js +70 -0
  1067. package/dist/types/provider/__tests__/declared-retryable.test.js.map +1 -0
  1068. package/dist/types/provider/__tests__/overflow-classification.test.d.ts +2 -0
  1069. package/dist/types/provider/__tests__/overflow-classification.test.d.ts.map +1 -0
  1070. package/dist/types/provider/__tests__/overflow-classification.test.js +80 -0
  1071. package/dist/types/provider/__tests__/overflow-classification.test.js.map +1 -0
  1072. package/dist/types/provider/chat.d.ts +27 -1
  1073. package/dist/types/provider/chat.d.ts.map +1 -1
  1074. package/dist/types/provider/config.d.ts +83 -2
  1075. package/dist/types/provider/config.d.ts.map +1 -1
  1076. package/dist/types/provider/errors.d.ts +102 -0
  1077. package/dist/types/provider/errors.d.ts.map +1 -0
  1078. package/dist/types/provider/errors.js +393 -0
  1079. package/dist/types/provider/errors.js.map +1 -0
  1080. package/dist/types/provider/index.d.ts +3 -2
  1081. package/dist/types/provider/index.d.ts.map +1 -1
  1082. package/dist/types/provider/stream.d.ts +59 -3
  1083. package/dist/types/provider/stream.d.ts.map +1 -1
  1084. package/dist/types/rag/embedding.d.ts +11 -2
  1085. package/dist/types/rag/embedding.d.ts.map +1 -1
  1086. package/dist/types/run/__tests__/step.test.d.ts +2 -0
  1087. package/dist/types/run/__tests__/step.test.d.ts.map +1 -0
  1088. package/dist/types/run/__tests__/step.test.js +107 -0
  1089. package/dist/types/run/__tests__/step.test.js.map +1 -0
  1090. package/dist/types/run/answer-review.d.ts +46 -0
  1091. package/dist/types/run/answer-review.d.ts.map +1 -0
  1092. package/dist/types/run/answer-review.js +2 -0
  1093. package/dist/types/run/answer-review.js.map +1 -0
  1094. package/dist/types/run/config.d.ts +33 -0
  1095. package/dist/types/run/config.d.ts.map +1 -1
  1096. package/dist/types/run/derive-status.d.ts +29 -0
  1097. package/dist/types/run/derive-status.d.ts.map +1 -0
  1098. package/dist/types/run/derive-status.js +49 -0
  1099. package/dist/types/run/derive-status.js.map +1 -0
  1100. package/dist/types/run/entity.d.ts +16 -0
  1101. package/dist/types/run/entity.d.ts.map +1 -1
  1102. package/dist/types/run/events.d.ts +228 -4
  1103. package/dist/types/run/events.d.ts.map +1 -1
  1104. package/dist/types/run/events.js +7 -0
  1105. package/dist/types/run/events.js.map +1 -1
  1106. package/dist/types/run/index.d.ts +5 -0
  1107. package/dist/types/run/index.d.ts.map +1 -1
  1108. package/dist/types/run/index.js +3 -0
  1109. package/dist/types/run/index.js.map +1 -1
  1110. package/dist/types/run/memory-promotion.d.ts +90 -0
  1111. package/dist/types/run/memory-promotion.d.ts.map +1 -0
  1112. package/dist/types/run/memory-promotion.js +38 -0
  1113. package/dist/types/run/memory-promotion.js.map +1 -0
  1114. package/dist/types/run/prepare-step.d.ts +100 -0
  1115. package/dist/types/run/prepare-step.d.ts.map +1 -0
  1116. package/dist/types/run/prepare-step.js +2 -0
  1117. package/dist/types/run/prepare-step.js.map +1 -0
  1118. package/dist/types/run/state.d.ts +85 -0
  1119. package/dist/types/run/state.d.ts.map +1 -0
  1120. package/dist/types/run/state.js +33 -0
  1121. package/dist/types/run/state.js.map +1 -0
  1122. package/dist/types/run/step.d.ts +77 -0
  1123. package/dist/types/run/step.d.ts.map +1 -0
  1124. package/dist/types/run/step.js +26 -0
  1125. package/dist/types/run/step.js.map +1 -0
  1126. package/dist/types/run/stop-reason.d.ts +31 -3
  1127. package/dist/types/run/stop-reason.d.ts.map +1 -1
  1128. package/dist/types/sandbox/index.d.ts +82 -10
  1129. package/dist/types/sandbox/index.d.ts.map +1 -1
  1130. package/dist/types/sandbox/index.js +13 -1
  1131. package/dist/types/sandbox/index.js.map +1 -1
  1132. package/dist/types/structured-output/index.d.ts +6 -5
  1133. package/dist/types/structured-output/index.d.ts.map +1 -1
  1134. package/dist/types/thread/entity.d.ts +2 -3
  1135. package/dist/types/thread/entity.d.ts.map +1 -1
  1136. package/dist/types/tool/index.d.ts +193 -1
  1137. package/dist/types/tool/index.d.ts.map +1 -1
  1138. package/dist/types/tool/index.js +1 -1
  1139. package/dist/types/tool/index.js.map +1 -1
  1140. package/dist/types/tool/repair.d.ts +49 -0
  1141. package/dist/types/tool/repair.d.ts.map +1 -0
  1142. package/dist/types/tool/repair.js +2 -0
  1143. package/dist/types/tool/repair.js.map +1 -0
  1144. package/dist/types/toolset/index.d.ts +2 -5
  1145. package/dist/types/toolset/index.d.ts.map +1 -1
  1146. package/dist/utils/__tests__/atomic-write.test.d.ts +2 -0
  1147. package/dist/utils/__tests__/atomic-write.test.d.ts.map +1 -0
  1148. package/dist/utils/__tests__/atomic-write.test.js +91 -0
  1149. package/dist/utils/__tests__/atomic-write.test.js.map +1 -0
  1150. package/dist/utils/__tests__/shell-compress.test.d.ts +2 -0
  1151. package/dist/utils/__tests__/shell-compress.test.d.ts.map +1 -0
  1152. package/dist/utils/__tests__/shell-compress.test.js +41 -0
  1153. package/dist/utils/__tests__/shell-compress.test.js.map +1 -0
  1154. package/dist/utils/atomic-write.d.ts +10 -0
  1155. package/dist/utils/atomic-write.d.ts.map +1 -0
  1156. package/dist/utils/atomic-write.js +83 -0
  1157. package/dist/utils/atomic-write.js.map +1 -0
  1158. package/dist/utils/shell-compress.d.ts.map +1 -1
  1159. package/dist/utils/shell-compress.js +12 -1
  1160. package/dist/utils/shell-compress.js.map +1 -1
  1161. package/dist/verification/presets.d.ts +6 -3
  1162. package/dist/verification/presets.d.ts.map +1 -1
  1163. package/dist/verification/presets.js +6 -3
  1164. package/dist/verification/presets.js.map +1 -1
  1165. package/package.json +3 -3
  1166. package/src/__tests__/default-path-conformance.test.ts +306 -0
  1167. package/src/__tests__/shipped-defaults.test.ts +111 -0
  1168. package/src/advisory/__tests__/budget-enforcement.test.ts +208 -0
  1169. package/src/advisory/__tests__/structured-response.test.ts +128 -0
  1170. package/src/advisory/budget.ts +27 -0
  1171. package/src/advisory/context.ts +19 -0
  1172. package/src/advisory/executor.test.ts +12 -8
  1173. package/src/advisory/executor.ts +37 -25
  1174. package/src/advisory/index.ts +2 -0
  1175. package/src/advisory/parse.ts +63 -0
  1176. package/src/agents/AbstractAgent.ts +82 -0
  1177. package/src/agents/PipelineAgent.ts +19 -0
  1178. package/src/agents/ReactiveAgent.ts +47 -0
  1179. package/src/agents/RouterAgent.ts +32 -2
  1180. package/src/agents/SupervisorAgent.ts +31 -1
  1181. package/src/agents/__tests__/idempotency-key.test.ts +177 -0
  1182. package/src/agents/__tests__/invocation-lock.test.ts +154 -0
  1183. package/src/agents/__tests__/reachability.test.ts +224 -0
  1184. package/src/agents/__tests__/supervisor-ledger.test.ts +1 -1
  1185. package/src/agents/index.ts +1 -1
  1186. package/src/agents/lock.ts +0 -2
  1187. package/src/bridge/a2a/late-events.test.ts +84 -0
  1188. package/src/bridge/a2a/mapper.test.ts +1 -1
  1189. package/src/bridge/a2a/mapper.ts +70 -5
  1190. package/src/bridge/a2a/message.ts +5 -1
  1191. package/src/bridge/sse/late-events.test.ts +182 -0
  1192. package/src/bridge/sse/mapper.ts +109 -0
  1193. package/src/compaction/__tests__/cleared-result-recovery.test.ts +118 -0
  1194. package/src/compaction/__tests__/context-window.test.ts +77 -0
  1195. package/src/compaction/__tests__/dangling.test.ts +16 -6
  1196. package/src/compaction/__tests__/lossiness.test.ts +187 -0
  1197. package/src/compaction/__tests__/state-across-resume.test.ts +99 -0
  1198. package/src/compaction/__tests__/tool-result-editing.test.ts +213 -0
  1199. package/src/compaction/__tests__/trim-opens-on-user.test.ts +118 -0
  1200. package/src/compaction/__tests__/verifier-model.test.ts +83 -0
  1201. package/src/compaction/context-window.ts +107 -0
  1202. package/src/compaction/dangling.ts +48 -1
  1203. package/src/compaction/extractor.ts +56 -1
  1204. package/src/compaction/index.ts +16 -0
  1205. package/src/compaction/manager.ts +53 -8
  1206. package/src/compaction/managers/structured.ts +7 -6
  1207. package/src/compaction/retention.ts +89 -0
  1208. package/src/compaction/serializer.ts +42 -11
  1209. package/src/compaction/tool-result-editing.ts +229 -0
  1210. package/src/compaction/types.ts +10 -0
  1211. package/src/compaction/verifier.ts +24 -1
  1212. package/src/compaction/wire.ts +91 -0
  1213. package/src/config/runtime.ts +63 -19
  1214. package/src/connector/builtins/http.test.ts +13 -0
  1215. package/src/connector/builtins/http.ts +15 -1
  1216. package/src/connector/mcp/__tests__/discovery-policy.test.ts +133 -0
  1217. package/src/connector/mcp/__tests__/list-pagination.test.ts +98 -0
  1218. package/src/connector/mcp/__tests__/policy.test.ts +100 -0
  1219. package/src/connector/mcp/__tests__/protocol-negotiation.test.ts +105 -0
  1220. package/src/connector/mcp/__tests__/schema-fidelity.test.ts +250 -0
  1221. package/src/connector/mcp/__tests__/structured-result.test.ts +143 -0
  1222. package/src/connector/mcp/adapter.ts +306 -19
  1223. package/src/connector/mcp/client.test.ts +159 -0
  1224. package/src/connector/mcp/client.ts +160 -13
  1225. package/src/connector/mcp/discovery.ts +74 -5
  1226. package/src/connector/mcp/http-sse.ts +16 -1
  1227. package/src/connector/mcp/index.ts +4 -0
  1228. package/src/connector/mcp/policy.ts +134 -0
  1229. package/src/connector/mcp/schema-refs.ts +126 -0
  1230. package/src/constants/agent/index.ts +3 -0
  1231. package/src/constants/compaction/index.ts +1 -1
  1232. package/src/constants/continuation.ts +4 -4
  1233. package/src/constants/mcp/index.ts +34 -0
  1234. package/src/constants/plugin/index.ts +24 -2
  1235. package/src/constants/sandbox/index.ts +6 -5
  1236. package/src/constants/telemetry/index.ts +8 -0
  1237. package/src/constants/tools/index.ts +20 -0
  1238. package/src/contracts/api.ts +40 -0
  1239. package/src/contracts/run-status.ts +43 -0
  1240. package/src/eval/__tests__/case-deadline.test.ts +147 -0
  1241. package/src/eval/__tests__/end-to-end.test.ts +259 -0
  1242. package/src/eval/__tests__/harness-honesty.test.ts +99 -0
  1243. package/src/eval/__tests__/harness.test.ts +249 -0
  1244. package/src/eval/__tests__/judge.test.ts +306 -0
  1245. package/src/eval/__tests__/scorer-gates.test.ts +156 -0
  1246. package/src/eval/experiment.ts +335 -0
  1247. package/src/eval/from-run.ts +48 -0
  1248. package/src/eval/index.ts +21 -0
  1249. package/src/eval/judge.ts +240 -0
  1250. package/src/eval/scorers.ts +169 -0
  1251. package/src/eval/types.ts +157 -0
  1252. package/src/execution/local.test.ts +46 -0
  1253. package/src/execution/local.ts +6 -1
  1254. package/src/gateway/__tests__/list-survives-eviction.test.ts +124 -0
  1255. package/src/gateway/local.ts +117 -3
  1256. package/src/manager/agent/__tests__/dispose-cancels.test.ts +104 -0
  1257. package/src/manager/agent/__tests__/lifecycle.test.ts +297 -0
  1258. package/src/manager/agent/lifecycle.ts +127 -7
  1259. package/src/manager/run/persistence.ts +132 -1
  1260. package/src/plugin/__tests__/enable-contributions.test.ts +3 -3
  1261. package/src/plugin/__tests__/hook-ordering.test.ts +252 -0
  1262. package/src/plugin/__tests__/install-time-refusal.test.ts +68 -0
  1263. package/src/plugin/__tests__/lifecycle.test.ts +95 -73
  1264. package/src/plugin/lifecycle.ts +76 -22
  1265. package/src/plugin/loader.ts +37 -1
  1266. package/src/probe/registry.test.ts +49 -6
  1267. package/src/probe/registry.ts +21 -1
  1268. package/src/provider/__tests__/mock-script.test.ts +179 -0
  1269. package/src/provider/__tests__/registry.test.ts +1 -1
  1270. package/src/provider/__tests__/retry-visibility.test.ts +160 -0
  1271. package/src/provider/__tests__/retry.test.ts +296 -0
  1272. package/src/provider/capabilities.ts +4 -0
  1273. package/src/provider/collect.test.ts +1 -1
  1274. package/src/provider/collect.ts +1 -1
  1275. package/src/provider/index.ts +8 -0
  1276. package/src/provider/mock-register.ts +5 -2
  1277. package/src/provider/mock.ts +117 -34
  1278. package/src/provider/retry.ts +199 -0
  1279. package/src/public-runtime.ts +111 -7
  1280. package/src/public-types.ts +5 -1
  1281. package/src/rag/__tests__/retrieval-math.test.ts +54 -0
  1282. package/src/rag/chunking.ts +32 -1
  1283. package/src/rag/embedding.test.ts +65 -10
  1284. package/src/rag/embedding.ts +26 -5
  1285. package/src/rag/index.ts +1 -1
  1286. package/src/rag/retriever.ts +92 -21
  1287. package/src/rag/vector-store.test.ts +2 -2
  1288. package/src/registry/ManagedRegistry.test.ts +1 -1
  1289. package/src/registry/Registry.test.ts +1 -1
  1290. package/src/registry/connector/definitions.test.ts +1 -1
  1291. package/src/registry/tool/__tests__/schema.test.ts +208 -0
  1292. package/src/registry/tool/__tests__/tool-name-validation.test.ts +88 -0
  1293. package/src/registry/tool/execute.ts +104 -22
  1294. package/src/registry/tool/schema.ts +80 -0
  1295. package/src/registry/toolset/catalog.test.ts +0 -1
  1296. package/src/registry/toolset/catalog.ts +0 -1
  1297. package/src/run/reporter.ts +84 -0
  1298. package/src/runtime/bidi/__tests__/session.test.ts +232 -0
  1299. package/src/runtime/bidi/index.ts +4 -0
  1300. package/src/runtime/bidi/mock.ts +90 -0
  1301. package/src/runtime/bidi/session.ts +205 -0
  1302. package/src/runtime/decision/parser.test.ts +46 -0
  1303. package/src/runtime/decision/parser.ts +8 -1
  1304. package/src/runtime/query/__tests__/answer-review.test.ts +172 -0
  1305. package/src/runtime/query/__tests__/capability-negotiation.test.ts +25 -129
  1306. package/src/runtime/query/__tests__/checkpoint-store.test.ts +13 -50
  1307. package/src/runtime/query/__tests__/checkpoint.test.ts +2 -10
  1308. package/src/runtime/query/__tests__/citations.test.ts +111 -0
  1309. package/src/runtime/query/__tests__/content-budget.test.ts +83 -0
  1310. package/src/runtime/query/__tests__/context.test.ts +3 -2
  1311. package/src/runtime/query/__tests__/deferred-tools.test.ts +23 -148
  1312. package/src/runtime/query/__tests__/delegation-resume.test.ts +159 -0
  1313. package/src/runtime/query/__tests__/document-attachments.test.ts +125 -0
  1314. package/src/runtime/query/__tests__/durable-park-and-trace.test.ts +245 -0
  1315. package/src/runtime/query/__tests__/durable-question-park.test.ts +305 -0
  1316. package/src/runtime/query/__tests__/durable-run-state.test.ts +194 -0
  1317. package/src/runtime/query/__tests__/executor-plugin-hooks.test.ts +24 -20
  1318. package/src/runtime/query/__tests__/finish-reason-vs-tool-calls.test.ts +206 -0
  1319. package/src/runtime/query/__tests__/grant-cannot-overrule-the-gate.test.ts +115 -0
  1320. package/src/runtime/query/__tests__/guardrails-e2e.test.ts +164 -0
  1321. package/src/runtime/query/__tests__/guardrails.test.ts +236 -0
  1322. package/src/runtime/query/__tests__/hook-model-visibility.test.ts +143 -0
  1323. package/src/runtime/query/__tests__/long-document-flow.test.ts +9 -2
  1324. package/src/runtime/query/__tests__/memory-promotion.test.ts +148 -0
  1325. package/src/runtime/query/__tests__/orphan-task-turn-end.test.ts +5 -31
  1326. package/src/runtime/query/__tests__/overflow-relief.test.ts +139 -0
  1327. package/src/runtime/query/__tests__/partial-batch-resume.test.ts +264 -0
  1328. package/src/runtime/query/__tests__/plugin-hooks.test.ts +0 -1
  1329. package/src/runtime/query/__tests__/prepare-step-chain.test.ts +127 -0
  1330. package/src/runtime/query/__tests__/prepare-step.test.ts +233 -0
  1331. package/src/runtime/query/__tests__/probe-veto-is-error.test.ts +145 -0
  1332. package/src/runtime/query/__tests__/prompt.test.ts +1 -1
  1333. package/src/runtime/query/__tests__/provider-fault-pauses.test.ts +104 -0
  1334. package/src/runtime/query/__tests__/reasoning-stream.test.ts +177 -0
  1335. package/src/runtime/query/__tests__/recoverable-settle.test.ts +145 -0
  1336. package/src/runtime/query/__tests__/resume-budget.test.ts +158 -0
  1337. package/src/runtime/query/__tests__/resume-pending.test.ts +273 -0
  1338. package/src/runtime/query/__tests__/rich-content-budget-wiring.test.ts +109 -0
  1339. package/src/runtime/query/__tests__/stop-condition.test.ts +265 -0
  1340. package/src/runtime/query/__tests__/structured-output.test.ts +268 -0
  1341. package/src/runtime/query/__tests__/structured-run-failure.test.ts +144 -0
  1342. package/src/runtime/query/__tests__/terminal-tool.test.ts +148 -0
  1343. package/src/runtime/query/__tests__/tool-grants.test.ts +102 -0
  1344. package/src/runtime/query/__tests__/tool-output-budget.test.ts +0 -0
  1345. package/src/runtime/query/__tests__/tool-pause.test.ts +209 -0
  1346. package/src/runtime/query/__tests__/tool-progress.test.ts +154 -0
  1347. package/src/runtime/query/__tests__/tool-repair-retry.test.ts +255 -0
  1348. package/src/runtime/query/__tests__/tool-result-wire.test.ts +135 -0
  1349. package/src/runtime/query/__tests__/tool-timeout.test.ts +230 -0
  1350. package/src/runtime/query/__tests__/trace-parenting.test.ts +128 -0
  1351. package/src/runtime/query/__tests__/wiring-seams.test.ts +288 -0
  1352. package/src/runtime/query/checkpoint.ts +242 -10
  1353. package/src/runtime/query/events.ts +2 -2
  1354. package/src/runtime/query/executor.ts +847 -39
  1355. package/src/runtime/query/guard.ts +26 -1
  1356. package/src/runtime/query/guardrail-presets.ts +124 -0
  1357. package/src/runtime/query/guardrails.ts +144 -0
  1358. package/src/runtime/query/index.ts +759 -12
  1359. package/src/runtime/query/iteration/__tests__/cancel-settles-turn.test.ts +177 -0
  1360. package/src/runtime/query/iteration/index.ts +656 -40
  1361. package/src/runtime/query/iteration/phases/advisory.test.ts +4 -1
  1362. package/src/runtime/query/iteration/phases/advisory.ts +14 -2
  1363. package/src/runtime/query/iteration/phases/checkpoint.ts +2 -2
  1364. package/src/runtime/query/iteration/phases/compaction-safe-cut.test.ts +3 -1
  1365. package/src/runtime/query/iteration/phases/compaction.ts +329 -64
  1366. package/src/runtime/query/iteration/phases/context-measurement.test.ts +196 -0
  1367. package/src/runtime/query/iteration/phases/context.ts +155 -2
  1368. package/src/runtime/query/iteration/phases/hitl-answer-question.test.ts +3 -1
  1369. package/src/runtime/query/iteration/phases/overflow-escalation.test.ts +167 -0
  1370. package/src/runtime/query/iteration/phases/plan.ts +10 -2
  1371. package/src/runtime/query/iteration/phases/retained-messages.test.ts +226 -0
  1372. package/src/runtime/query/iteration/phases/tool-review-denials.test.ts +270 -0
  1373. package/src/runtime/query/iteration/phases/tool-review.ts +140 -51
  1374. package/src/runtime/query/iteration/phases/working-memory-compaction.test.ts +21 -9
  1375. package/src/runtime/query/iteration/stream-turn.ts +285 -15
  1376. package/src/runtime/query/plugin-hooks.ts +6 -2
  1377. package/src/runtime/query/prompt.ts +26 -1
  1378. package/src/runtime/query/question-park.ts +120 -0
  1379. package/src/runtime/query/result.ts +64 -0
  1380. package/src/runtime/query/resume-pending.ts +394 -0
  1381. package/src/runtime/query/run-state.ts +124 -0
  1382. package/src/runtime/query/tool-grants.ts +83 -0
  1383. package/src/runtime/query/tool-output-budget.ts +183 -0
  1384. package/src/runtime/query/tool-pause.ts +119 -0
  1385. package/src/runtime/query/tooling.ts +23 -1
  1386. package/src/sandbox/__tests__/isolation.test.ts +164 -0
  1387. package/src/sandbox/__tests__/output-truncation.test.ts +106 -0
  1388. package/src/sandbox/__tests__/resource-limits.test.ts +106 -0
  1389. package/src/sandbox/factory.ts +1 -1
  1390. package/src/sandbox/index.ts +2 -0
  1391. package/src/sandbox/isolation.ts +72 -0
  1392. package/src/sandbox/provider/local.ts +173 -74
  1393. package/src/session/__tests__/integration/spawn-rollback.test.ts +1 -1
  1394. package/src/session/handoff/broadcast.ts +3 -3
  1395. package/src/session/migration/__tests__/filesystem.test.ts +26 -19
  1396. package/src/session/migration/filesystem.ts +7 -12
  1397. package/src/session/migration/marker.ts +5 -9
  1398. package/src/session/retention/disk-backend.ts +5 -11
  1399. package/src/session/workspace/__tests__/git-worktree.test.ts +6 -2
  1400. package/src/session/workspace/__tests__/shared-run.test.ts +18 -0
  1401. package/src/session/workspace/shared-run.ts +7 -1
  1402. package/src/skills/__tests__/frontmatter.test.ts +135 -0
  1403. package/src/skills/loader.ts +47 -3
  1404. package/src/store/__tests__/checkpoint-integrity.test.ts +134 -0
  1405. package/src/store/__tests__/checkpoint-versioning.test.ts +194 -0
  1406. package/src/store/__tests__/schema-adoption.test.ts +118 -0
  1407. package/src/store/__tests__/schema.test.ts +119 -0
  1408. package/src/store/memory/disk.ts +17 -16
  1409. package/src/store/run/disk.ts +169 -22
  1410. package/src/store/schema.ts +135 -0
  1411. package/src/store/session/disk.ts +30 -21
  1412. package/src/store/task/__tests__/lookup-across-runs.test.ts +77 -0
  1413. package/src/store/task/disk.ts +59 -15
  1414. package/src/store/thread/disk.ts +21 -10
  1415. package/src/streaming/coalesce.ts +1 -1
  1416. package/src/telemetry/__tests__/latency-instruments.test.ts +120 -0
  1417. package/src/telemetry/__tests__/metrics.test.ts +160 -0
  1418. package/src/telemetry/__tests__/span-hierarchy.test.ts +94 -0
  1419. package/src/telemetry/attributes.ts +109 -0
  1420. package/src/telemetry/metrics.ts +190 -0
  1421. package/src/test-setup.ts +3 -3
  1422. package/src/test-support/paths.ts +17 -0
  1423. package/src/tools/__tests__/path-containment.test.ts +167 -0
  1424. package/src/tools/__tests__/posix-path.test.ts +83 -0
  1425. package/src/tools/__tests__/sandboxed-search.test.ts +140 -0
  1426. package/src/tools/builtins/__tests__/computer-use.test.ts +15 -2
  1427. package/src/tools/builtins/__tests__/edit.test.ts +10 -270
  1428. package/src/tools/builtins/__tests__/payload-budget.test.ts +20 -40
  1429. package/src/tools/builtins/__tests__/read-file.test.ts +29 -1
  1430. package/src/tools/builtins/__tests__/write-file.test.ts +13 -52
  1431. package/src/tools/builtins/bash.ts +34 -4
  1432. package/src/tools/builtins/computer-use.ts +7 -1
  1433. package/src/tools/builtins/edit.ts +162 -116
  1434. package/src/tools/builtins/glob.ts +71 -6
  1435. package/src/tools/builtins/grep.ts +82 -8
  1436. package/src/tools/builtins/index.ts +6 -6
  1437. package/src/tools/builtins/ls.ts +4 -2
  1438. package/src/tools/builtins/read-file.ts +40 -4
  1439. package/src/tools/builtins/write-file.ts +44 -68
  1440. package/src/tools/coordinator/__tests__/agent.test.ts +2 -2
  1441. package/src/tools/coordinator/agent.ts +23 -10
  1442. package/src/tools/coordinator/index.ts +111 -31
  1443. package/src/tools/defineTool.ts +20 -0
  1444. package/src/tools/glob-match.ts +50 -0
  1445. package/src/tools/paths.ts +44 -0
  1446. package/src/tools/posix-path.ts +73 -0
  1447. package/src/types/advisory/config.ts +21 -3
  1448. package/src/types/advisory/result.ts +12 -7
  1449. package/src/types/agent/base.ts +22 -2
  1450. package/src/types/agent/factory.ts +15 -9
  1451. package/src/types/agent/gateway.ts +20 -0
  1452. package/src/types/agent/manager.ts +19 -0
  1453. package/src/types/agent/reactive.ts +50 -0
  1454. package/src/types/agent/router.ts +11 -0
  1455. package/src/types/agent/supervisor.ts +2 -2
  1456. package/src/types/agent/task.ts +26 -0
  1457. package/src/types/bidi/index.ts +119 -0
  1458. package/src/types/common/__tests__/numeric-correctness.test.ts +123 -0
  1459. package/src/types/common/index.ts +16 -4
  1460. package/src/types/computer-use/index.ts +1 -1
  1461. package/src/types/connector/mcp.ts +29 -0
  1462. package/src/types/errors/__tests__/catalog.test.ts +130 -0
  1463. package/src/types/errors/__tests__/errors.test.ts +118 -0
  1464. package/src/types/errors/catalog.ts +211 -0
  1465. package/src/types/errors/index.ts +121 -0
  1466. package/src/types/guardrail/index.ts +62 -0
  1467. package/src/types/hitl/index.ts +102 -20
  1468. package/src/types/ids/index.ts +2 -2
  1469. package/src/types/message/content.ts +50 -0
  1470. package/src/types/message/index.ts +210 -9
  1471. package/src/types/plugin/index.ts +82 -2
  1472. package/src/types/provider/__tests__/cause-chain.test.ts +132 -0
  1473. package/src/types/provider/__tests__/declared-retryable.test.ts +85 -0
  1474. package/src/types/provider/__tests__/overflow-classification.test.ts +113 -0
  1475. package/src/types/provider/chat.ts +29 -1
  1476. package/src/types/provider/config.ts +83 -2
  1477. package/src/types/provider/errors.ts +472 -0
  1478. package/src/types/provider/index.ts +5 -0
  1479. package/src/types/provider/stream.ts +63 -3
  1480. package/src/types/rag/embedding.ts +11 -2
  1481. package/src/types/run/__tests__/step.test.ts +123 -0
  1482. package/src/types/run/answer-review.ts +51 -0
  1483. package/src/types/run/config.ts +35 -0
  1484. package/src/types/run/derive-status.ts +62 -0
  1485. package/src/types/run/entity.ts +18 -0
  1486. package/src/types/run/events.ts +230 -3
  1487. package/src/types/run/index.ts +5 -0
  1488. package/src/types/run/memory-promotion.ts +112 -0
  1489. package/src/types/run/prepare-step.ts +109 -0
  1490. package/src/types/run/state.ts +111 -0
  1491. package/src/types/run/step.ts +95 -0
  1492. package/src/types/run/stop-reason.ts +29 -1
  1493. package/src/types/sandbox/index.ts +93 -12
  1494. package/src/types/structured-output/index.ts +6 -5
  1495. package/src/types/thread/entity.ts +2 -3
  1496. package/src/types/tool/index.ts +204 -1
  1497. package/src/types/tool/repair.ts +54 -0
  1498. package/src/types/toolset/index.ts +2 -6
  1499. package/src/utils/__tests__/atomic-write.test.ts +108 -0
  1500. package/src/utils/__tests__/shell-compress.test.ts +49 -0
  1501. package/src/utils/atomic-write.ts +85 -0
  1502. package/src/utils/shell-compress.ts +12 -1
  1503. package/src/verification/presets.ts +6 -3
@@ -0,0 +1,136 @@
1
+ import { describe, expect, it } from 'vitest';
2
+ import { formatReport, runExperiment } from '../experiment.js';
3
+ /**
4
+ * A case's verdict was one unweighted mean over every scorer against one
5
+ * suite-wide threshold, and the two halves of that fought each other.
6
+ *
7
+ * At the default threshold of 1 the harness never reports a false pass —
8
+ * but a trajectory F1 and a graded judge can essentially never reach 1, so
9
+ * every real suite lowers it. And every step down buys tolerance for the
10
+ * fuzzy scorers by buying exactly the same tolerance for the deterministic
11
+ * ones: at 0.75, a trajectory score of 0 alongside three perfect scores
12
+ * averages to 0.75 and reports PASSED. The regression the harness exists
13
+ * to catch comes back green.
14
+ *
15
+ * The workarounds were lossy too — folding every hard check into one
16
+ * custom scorer destroys per-dimension reporting, and two experiments at
17
+ * two thresholds re-runs the dataset and splits one case's verdict across
18
+ * two reports.
19
+ */
20
+ const run = () => ({
21
+ output: 'answer',
22
+ steps: [],
23
+ toolCalls: [],
24
+ totalTokens: 0,
25
+ totalCostUsd: 0,
26
+ durationMs: 0,
27
+ });
28
+ const scorer = (name, score, extra = {}) => ({
29
+ name,
30
+ score: () => ({ score, reason: `${name} scored ${score}` }),
31
+ ...extra,
32
+ });
33
+ const experiment = (scorers, passThreshold = 0.75) => runExperiment({
34
+ name: 'gates',
35
+ cases: [{ name: 'one', input: null }],
36
+ scorers,
37
+ passThreshold,
38
+ run: async () => run(),
39
+ });
40
+ describe('a gate scorer', () => {
41
+ it('fails the case even when the mean clears the threshold', async () => {
42
+ const report = await experiment([
43
+ scorer('trajectory', 0, { severity: 'gate' }),
44
+ scorer('completion', 1),
45
+ scorer('contains', 1),
46
+ scorer('judge', 1),
47
+ ]);
48
+ // Mean is exactly 0.75 against a 0.75 threshold — reported passed
49
+ // before, with the one deterministic check at zero.
50
+ expect(report.cases[0]?.mean).toBeCloseTo(0.75);
51
+ expect(report.cases[0]?.status).toBe('failed');
52
+ });
53
+ it('names itself, so the reader does not have to guess', async () => {
54
+ const report = await experiment([
55
+ scorer('trajectory', 0, { severity: 'gate' }),
56
+ scorer('judge', 1),
57
+ ]);
58
+ expect(report.cases[0]?.failedGates).toEqual(['trajectory']);
59
+ });
60
+ it('honours its own threshold when it sets one', async () => {
61
+ // "Good enough" is not one number across dimensions: a trajectory
62
+ // match at 0.8 may be fine while a completion check at 0.8 is
63
+ // meaningless.
64
+ const lenient = await experiment([
65
+ scorer('trajectory', 0.8, { severity: 'gate', threshold: 0.7 }),
66
+ ]);
67
+ expect(lenient.cases[0]?.status).toBe('passed');
68
+ const strict = await experiment([
69
+ scorer('trajectory', 0.8, { severity: 'gate', threshold: 0.9 }),
70
+ ]);
71
+ expect(strict.cases[0]?.status).toBe('failed');
72
+ });
73
+ it('falls back to the suite threshold when it sets none', async () => {
74
+ const report = await experiment([scorer('trajectory', 0.5, { severity: 'gate' })], 0.75);
75
+ expect(report.cases[0]?.failedGates).toEqual(['trajectory']);
76
+ });
77
+ it('does not fail the case when it could not judge at all', async () => {
78
+ // Unavailable is not zero. A gate that threw did not judge the run
79
+ // badly, it failed to judge it — the inconclusive path, not a
80
+ // failure, and the two demand opposite responses.
81
+ const report = await runExperiment({
82
+ name: 'gates',
83
+ cases: [{ name: 'one', input: null }],
84
+ scorers: [
85
+ {
86
+ name: 'judge-gate',
87
+ severity: 'gate',
88
+ score: () => {
89
+ throw new Error('provider unreachable');
90
+ },
91
+ },
92
+ ],
93
+ run: async () => run(),
94
+ });
95
+ expect(report.cases[0]?.status).toBe('inconclusive');
96
+ expect(report.cases[0]?.failedGates).toBeUndefined();
97
+ });
98
+ it('lets a passing gate through', async () => {
99
+ const report = await experiment([scorer('trajectory', 1, { severity: 'gate' })]);
100
+ expect(report.cases[0]?.status).toBe('passed');
101
+ });
102
+ });
103
+ describe('a soft scorer', () => {
104
+ it('only moves the mean', async () => {
105
+ // The fuzzy dimensions, where a number below 1 is normal and a
106
+ // threshold is a judgement call.
107
+ const report = await experiment([scorer('judge', 0.5), scorer('other', 1)]);
108
+ expect(report.cases[0]?.status).toBe('passed');
109
+ expect(report.cases[0]?.failedGates).toBeUndefined();
110
+ });
111
+ it('is the default, so a suite that sets nothing is unchanged', async () => {
112
+ const report = await experiment([
113
+ scorer('a', 0),
114
+ scorer('b', 1),
115
+ scorer('c', 1),
116
+ scorer('d', 1),
117
+ ]);
118
+ expect(report.cases[0]?.status).toBe('passed');
119
+ });
120
+ it('can still fail the case through the mean', async () => {
121
+ const report = await experiment([scorer('a', 0), scorer('b', 0.5)]);
122
+ expect(report.cases[0]?.status).toBe('failed');
123
+ expect(report.cases[0]?.failedGates).toBeUndefined();
124
+ });
125
+ });
126
+ describe('the report', () => {
127
+ it('says which gate missed, above the individual scores', async () => {
128
+ const report = await experiment([
129
+ scorer('trajectory', 0, { severity: 'gate' }),
130
+ scorer('judge', 1),
131
+ ]);
132
+ const text = formatReport(report);
133
+ expect(text).toContain('gate missed: trajectory');
134
+ });
135
+ });
136
+ //# sourceMappingURL=scorer-gates.test.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"scorer-gates.test.js","sourceRoot":"","sources":["../../../src/eval/__tests__/scorer-gates.test.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,EAAE,EAAE,EAAE,MAAM,QAAQ,CAAA;AAE7C,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAA;AAG9D;;;;;;;;;;;;;;;;GAgBG;AAEH,MAAM,GAAG,GAAG,GAAY,EAAE,CAAC,CAAC;IAC3B,MAAM,EAAE,QAAQ;IAChB,KAAK,EAAE,EAAE;IACT,SAAS,EAAE,EAAE;IACb,WAAW,EAAE,CAAC;IACd,YAAY,EAAE,CAAC;IACf,UAAU,EAAE,CAAC;CACb,CAAC,CAAA;AAEF,MAAM,MAAM,GAAG,CAAC,IAAY,EAAE,KAAa,EAAE,QAAyB,EAAE,EAAU,EAAE,CAAC,CAAC;IACrF,IAAI;IACJ,KAAK,EAAE,GAAU,EAAE,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,IAAI,WAAW,KAAK,EAAE,EAAE,CAAC;IAClE,GAAG,KAAK;CACR,CAAC,CAAA;AAEF,MAAM,UAAU,GAAG,CAAC,OAAiB,EAAE,aAAa,GAAG,IAAI,EAAE,EAAE,CAC9D,aAAa,CAAC;IACb,IAAI,EAAE,OAAO;IACb,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;IACrC,OAAO;IACP,aAAa;IACb,GAAG,EAAE,KAAK,IAAI,EAAE,CAAC,GAAG,EAAE;CACtB,CAAC,CAAA;AAEH,QAAQ,CAAC,eAAe,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,wDAAwD,EAAE,KAAK,IAAI,EAAE;QACvE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;YAC/B,MAAM,CAAC,YAAY,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;YAC7C,MAAM,CAAC,YAAY,EAAE,CAAC,CAAC;YACvB,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC;YACrB,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;SAClB,CAAC,CAAA;QAEF,kEAAkE;QAClE,oDAAoD;QACpD,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,CAAA;QAC/C,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,oDAAoD,EAAE,KAAK,IAAI,EAAE;QACnE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;YAC/B,MAAM,CAAC,YAAY,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;YAC7C,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;SAClB,CAAC,CAAA;QACF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAA;IAC7D,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,4CAA4C,EAAE,KAAK,IAAI,EAAE;QAC3D,kEAAkE;QAClE,8DAA8D;QAC9D,eAAe;QACf,MAAM,OAAO,GAAG,MAAM,UAAU,CAAC;YAChC,MAAM,CAAC,YAAY,EAAE,GAAG,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC;SAC/D,CAAC,CAAA;QACF,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;QAE/C,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;YAC/B,MAAM,CAAC,YAAY,EAAE,GAAG,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,SAAS,EAAE,GAAG,EAAE,CAAC;SAC/D,CAAC,CAAA;QACF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,qDAAqD,EAAE,KAAK,IAAI,EAAE;QACpE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,CAAC,MAAM,CAAC,YAAY,EAAE,GAAG,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC,EAAE,IAAI,CAAC,CAAA;QACxF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC,OAAO,CAAC,CAAC,YAAY,CAAC,CAAC,CAAA;IAC7D,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,uDAAuD,EAAE,KAAK,IAAI,EAAE;QACtE,mEAAmE;QACnE,8DAA8D;QAC9D,kDAAkD;QAClD,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC;YAClC,IAAI,EAAE,OAAO;YACb,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC;YACrC,OAAO,EAAE;gBACR;oBACC,IAAI,EAAE,YAAY;oBAClB,QAAQ,EAAE,MAAM;oBAChB,KAAK,EAAE,GAAG,EAAE;wBACX,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAA;oBACxC,CAAC;iBACD;aACD;YACD,GAAG,EAAE,KAAK,IAAI,EAAE,CAAC,GAAG,EAAE;SACtB,CAAC,CAAA;QAEF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,cAAc,CAAC,CAAA;QACpD,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC,aAAa,EAAE,CAAA;IACrD,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,6BAA6B,EAAE,KAAK,IAAI,EAAE;QAC5C,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,CAAC,MAAM,CAAC,YAAY,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC,CAAC,CAAA;QAChF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA;AAEF,QAAQ,CAAC,eAAe,EAAE,GAAG,EAAE;IAC9B,EAAE,CAAC,qBAAqB,EAAE,KAAK,IAAI,EAAE;QACpC,+DAA+D;QAC/D,iCAAiC;QACjC,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,CAAC,MAAM,CAAC,OAAO,EAAE,GAAG,CAAC,EAAE,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC,CAAC,CAAA;QAC3E,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;QAC9C,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC,aAAa,EAAE,CAAA;IACrD,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,2DAA2D,EAAE,KAAK,IAAI,EAAE;QAC1E,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;YAC/B,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;YACd,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;YACd,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;YACd,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC;SACd,CAAC,CAAA;QACF,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;IAC/C,CAAC,CAAC,CAAA;IAEF,EAAE,CAAC,0CAA0C,EAAE,KAAK,IAAI,EAAE;QACzD,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC,CAAC,MAAM,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAA;QACnE,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAA;QAC9C,MAAM,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,WAAW,CAAC,CAAC,aAAa,EAAE,CAAA;IACrD,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA;AAEF,QAAQ,CAAC,YAAY,EAAE,GAAG,EAAE;IAC3B,EAAE,CAAC,qDAAqD,EAAE,KAAK,IAAI,EAAE;QACpE,MAAM,MAAM,GAAG,MAAM,UAAU,CAAC;YAC/B,MAAM,CAAC,YAAY,EAAE,CAAC,EAAE,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;YAC7C,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC;SAClB,CAAC,CAAA;QAEF,MAAM,IAAI,GAAG,YAAY,CAAC,MAAM,CAAC,CAAA;QACjC,MAAM,CAAC,IAAI,CAAC,CAAC,SAAS,CAAC,yBAAyB,CAAC,CAAA;IAClD,CAAC,CAAC,CAAA;AACH,CAAC,CAAC,CAAA"}
@@ -0,0 +1,54 @@
1
+ import type { CaseResult, EvalCase, EvalRun, ExperimentReport, Scorer } from './types.js';
2
+ export interface ExperimentConfig<TInput = unknown> {
3
+ name: string;
4
+ cases: readonly EvalCase<TInput>[];
5
+ /** Applied to every case unless the case overrides them. */
6
+ scorers: readonly Scorer[];
7
+ /**
8
+ * Execute one case. Returning an `EvalRun` rather than driving `query()`
9
+ * here keeps the harness independent of how you construct a run —
10
+ * scripted mock, real provider, or a whole agent behind a facade.
11
+ */
12
+ run: (input: TInput, evalCase: EvalCase<TInput>, signal: AbortSignal) => Promise<EvalRun>;
13
+ /**
14
+ * Deadline for a single case. Unset means no deadline.
15
+ *
16
+ * `executeCase` used to be a bare await, so a `run` closure that never
17
+ * settled blocked its worker and `runExperiment` never returned — no
18
+ * report, no partial results, nothing to read. The documented path
19
+ * inherits deadlines from the runtime it drives, so the residual is a
20
+ * closure that does not go through `query()` and a mid-iteration
21
+ * provider stall the between-iterations guard cannot see. Both are
22
+ * reachable, and neither is the suite's fault to absorb silently.
23
+ *
24
+ * A timed-out case is REPORTED and the suite continues, exactly like a
25
+ * case that threw: forty cases should not be lost to one that hung.
26
+ */
27
+ timeoutMs?: number;
28
+ /** Mean score a case must reach to count as passed. Default 1. */
29
+ passThreshold?: number;
30
+ /** Cases to run at once. Default 1 — deterministic ordering by default. */
31
+ concurrency?: number;
32
+ onCaseFinish?: (result: CaseResult) => void;
33
+ }
34
+ /**
35
+ * Run a dataset and score it.
36
+ *
37
+ * There was no evaluation harness of any kind: no dataset, no scorer, no
38
+ * judge, no trajectory assertion. So every behavior change in this SDK —
39
+ * a tool description, the `search_tools` top-k, the compaction threshold —
40
+ * shipped with no regression signal behind it, and the only way to notice
41
+ * a degradation was for a user to hit it.
42
+ *
43
+ * The scorer contract requires a `reason` on every score, so a report says
44
+ * what got worse rather than only that something did.
45
+ */
46
+ export declare function runExperiment<TInput>(config: ExperimentConfig<TInput>): Promise<ExperimentReport>;
47
+ /**
48
+ * Render a report as text.
49
+ *
50
+ * Failures print their scorer reasons, because a CI log that says
51
+ * "0.62" is a log that sends someone back to reproduce it by hand.
52
+ */
53
+ export declare function formatReport(report: ExperimentReport): string;
54
+ //# sourceMappingURL=experiment.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"experiment.d.ts","sourceRoot":"","sources":["../../src/eval/experiment.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACX,UAAU,EAEV,QAAQ,EACR,OAAO,EACP,gBAAgB,EAEhB,MAAM,EACN,MAAM,YAAY,CAAA;AAEnB,MAAM,WAAW,gBAAgB,CAAC,MAAM,GAAG,OAAO;IACjD,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,SAAS,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAA;IAClC,4DAA4D;IAC5D,OAAO,EAAE,SAAS,MAAM,EAAE,CAAA;IAC1B;;;;OAIG;IACH,GAAG,EAAE,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,OAAO,CAAC,CAAA;IACzF;;;;;;;;;;;;;OAaG;IACH,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,kEAAkE;IAClE,aAAa,CAAC,EAAE,MAAM,CAAA;IACtB,2EAA2E;IAC3E,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,YAAY,CAAC,EAAE,CAAC,MAAM,EAAE,UAAU,KAAK,IAAI,CAAA;CAC3C;AAED;;;;;;;;;;;GAWG;AACH,wBAAsB,aAAa,CAAC,MAAM,EACzC,MAAM,EAAE,gBAAgB,CAAC,MAAM,CAAC,GAC9B,OAAO,CAAC,gBAAgB,CAAC,CAsG3B;AAmHD;;;;;GAKG;AACH,wBAAgB,YAAY,CAAC,MAAM,EAAE,gBAAgB,GAAG,MAAM,CAsD7D"}
@@ -0,0 +1,263 @@
1
+ /**
2
+ * Run a dataset and score it.
3
+ *
4
+ * There was no evaluation harness of any kind: no dataset, no scorer, no
5
+ * judge, no trajectory assertion. So every behavior change in this SDK —
6
+ * a tool description, the `search_tools` top-k, the compaction threshold —
7
+ * shipped with no regression signal behind it, and the only way to notice
8
+ * a degradation was for a user to hit it.
9
+ *
10
+ * The scorer contract requires a `reason` on every score, so a report says
11
+ * what got worse rather than only that something did.
12
+ */
13
+ export async function runExperiment(config) {
14
+ const startedAt = Date.now();
15
+ const threshold = config.passThreshold ?? 1;
16
+ const concurrency = Math.max(1, config.concurrency ?? 1);
17
+ const results = new Array(config.cases.length);
18
+ let cursor = 0;
19
+ const worker = async () => {
20
+ for (;;) {
21
+ const index = cursor++;
22
+ const evalCase = config.cases[index];
23
+ if (!evalCase)
24
+ return;
25
+ const run = await executeCase(config, evalCase);
26
+ const scorers = evalCase.scorers ?? config.scorers;
27
+ const scores = {};
28
+ // Scores are keyed by name, so two scorers sharing one collapse:
29
+ // the mean's denominator becomes the count of distinct NAMES and
30
+ // the surviving score is whichever ran last. Two
31
+ // `containsScorer(...)` instances are both called 'contains', so
32
+ // this is easy to hit by accident and silently halves the
33
+ // evidence. Ambiguous results are worse than a loud failure.
34
+ const seen = new Set();
35
+ for (const scorer of scorers) {
36
+ if (seen.has(scorer.name)) {
37
+ throw new Error(`Duplicate scorer name "${scorer.name}" for case "${evalCase.name}". Scores are keyed by name, so the second would overwrite the first and the case mean would be computed over the wrong denominator. Give each scorer a distinct name.`);
38
+ }
39
+ seen.add(scorer.name);
40
+ scores[scorer.name] = await safeScore(scorer, run, evalCase);
41
+ }
42
+ // Only scores that were actually produced count — an unavailable
43
+ // scorer leaves the denominator alone rather than dragging the
44
+ // mean toward zero with a measurement that never happened.
45
+ const values = Object.values(scores)
46
+ .filter((s) => s.unavailable !== true)
47
+ .map((s) => s.score);
48
+ const mean = values.length === 0 ? 0 : values.reduce((a, b) => a + b, 0) / values.length;
49
+ // A gate miss fails the case whatever the mean says. Averaging a
50
+ // hard check together with a fuzzy one lets three good scores
51
+ // carry a zero: trajectory 0 + completion 1 + contains 1 +
52
+ // judge 1 averages to 0.75 and reports passed at a threshold of
53
+ // 0.75 — the exact regression the harness exists to catch,
54
+ // reported green. An UNAVAILABLE gate does not fail the case;
55
+ // it did not judge the run at all, which is the inconclusive
56
+ // path, not a failure.
57
+ const failedGates = scorers
58
+ .filter((s) => s.severity === 'gate')
59
+ .filter((s) => {
60
+ const score = scores[s.name];
61
+ if (!score || score.unavailable === true)
62
+ return false;
63
+ return score.score < (s.threshold ?? threshold);
64
+ })
65
+ .map((s) => s.name);
66
+ const status = Object.keys(scores).length > 0 && values.length === 0
67
+ ? 'inconclusive'
68
+ : failedGates.length > 0
69
+ ? 'failed'
70
+ : mean >= threshold
71
+ ? 'passed'
72
+ : 'failed';
73
+ const result = {
74
+ case: evalCase.name,
75
+ run,
76
+ scores,
77
+ mean,
78
+ status,
79
+ passed: status === 'passed',
80
+ // Named, not just counted: "failed" with a mean of 0.75 sends
81
+ // somebody to read four scores and guess which one mattered.
82
+ ...(failedGates.length > 0 ? { failedGates } : {}),
83
+ };
84
+ results[index] = result;
85
+ config.onCaseFinish?.(result);
86
+ }
87
+ };
88
+ await Promise.all(Array.from({ length: Math.min(concurrency, config.cases.length) }, worker));
89
+ const settled = results.filter(Boolean);
90
+ // An inconclusive case has no mean to contribute; averaging its zero in
91
+ // would report a suite-wide regression caused by the harness.
92
+ const scored = settled.filter((r) => r.status !== 'inconclusive');
93
+ const mean = scored.length === 0 ? 0 : scored.reduce((sum, r) => sum + r.mean, 0) / scored.length;
94
+ return {
95
+ name: config.name,
96
+ cases: settled,
97
+ mean,
98
+ passed: settled.filter((r) => r.status === 'passed').length,
99
+ failed: settled.filter((r) => r.status === 'failed').length,
100
+ inconclusive: settled.filter((r) => r.status === 'inconclusive').length,
101
+ byScorer: meanByScorer(settled),
102
+ durationMs: Date.now() - startedAt,
103
+ };
104
+ }
105
+ /**
106
+ * A case that throws is a RESULT, not a crash. An eval suite whose first
107
+ * broken case aborts the run tells you nothing about the other forty.
108
+ */
109
+ async function executeCase(config, evalCase) {
110
+ const startedAt = Date.now();
111
+ const controller = new AbortController();
112
+ const timeoutMs = config.timeoutMs;
113
+ // The signal is handed to `run` so a closure that drives `query()` can
114
+ // pass it through and actually stop working. A closure that ignores it
115
+ // is merely detached rather than stopped — the same bargain every other
116
+ // deadline in the SDK makes — but the SUITE is unblocked either way,
117
+ // which is the part that was missing.
118
+ const timer = timeoutMs !== undefined && timeoutMs > 0
119
+ ? setTimeout(() => controller.abort(new Error(`case timed out after ${timeoutMs}ms`)), timeoutMs)
120
+ : undefined;
121
+ const deadline = timer === undefined
122
+ ? undefined
123
+ : new Promise((_resolve, reject) => {
124
+ controller.signal.addEventListener('abort', () => reject(controller.signal.reason), {
125
+ once: true,
126
+ });
127
+ });
128
+ try {
129
+ const work = config.run(evalCase.input, evalCase, controller.signal);
130
+ // A rejection of the loser must never surface as an unhandled
131
+ // rejection after the race has already been decided.
132
+ if (deadline)
133
+ work.catch(() => { });
134
+ return await (deadline ? Promise.race([work, deadline]) : work);
135
+ }
136
+ catch (err) {
137
+ return {
138
+ output: null,
139
+ steps: [],
140
+ toolCalls: [],
141
+ totalTokens: 0,
142
+ totalCostUsd: 0,
143
+ // Real elapsed time, not zero: a case that burned its whole
144
+ // deadline is the most interesting number in the report, and
145
+ // zero would hide it.
146
+ durationMs: Date.now() - startedAt,
147
+ error: err instanceof Error ? err.message : String(err),
148
+ };
149
+ }
150
+ finally {
151
+ // Only the timer needs clearing. Aborting here would fire a spurious
152
+ // signal at a `run` that has already settled — the deadline branch
153
+ // has aborted already, and every other path is done.
154
+ clearTimeout(timer);
155
+ }
156
+ }
157
+ /** A throwing scorer scores zero with the throw as its reason. */
158
+ async function safeScore(scorer, run, evalCase) {
159
+ // A run that THREW scores zero, whatever the scorer would have said.
160
+ // `executeCase` catches the failure and returns an empty run, and an
161
+ // empty run walks straight into every scorer's happy path:
162
+ // `stepBudgetScorer` sees 0 steps against its allowance and returns 1,
163
+ // `trajectoryScorer` sees no tools expected and none called and returns
164
+ // 1. So a suite whose runs were all dying reported green. The failure is
165
+ // recorded on `run.error` and nothing consulted it.
166
+ if (run.error !== undefined) {
167
+ return { score: 0, reason: `run failed: ${run.error}`, details: { error: run.error } };
168
+ }
169
+ try {
170
+ return await scorer.score(run, evalCase);
171
+ }
172
+ catch (err) {
173
+ // UNAVAILABLE, not zero. A scorer that threw did not judge the run
174
+ // badly — it failed to judge it at all, and the two call for
175
+ // opposite responses. Scoring the throw zero was survivable while
176
+ // every scorer was a pure function; a scorer that reaches a provider
177
+ // makes it actively misleading, because one rate limit would look
178
+ // exactly like a behavioural regression.
179
+ return {
180
+ score: 0,
181
+ unavailable: true,
182
+ reason: `scorer "${scorer.name}" could not judge this run: ${err instanceof Error ? err.message : String(err)}`,
183
+ };
184
+ }
185
+ }
186
+ function meanByScorer(results) {
187
+ const sums = new Map();
188
+ for (const result of results) {
189
+ for (const [name, score] of Object.entries(result.scores)) {
190
+ if (score.unavailable === true)
191
+ continue;
192
+ const entry = sums.get(name) ?? { total: 0, count: 0 };
193
+ entry.total += score.score;
194
+ entry.count++;
195
+ sums.set(name, entry);
196
+ }
197
+ }
198
+ const out = {};
199
+ // A scorer that was never available anywhere is omitted rather than
200
+ // reported as 0 — a dimension with no measurements is not a dimension
201
+ // that scored badly.
202
+ for (const [name, { total, count }] of sums) {
203
+ if (count > 0)
204
+ out[name] = total / count;
205
+ }
206
+ return out;
207
+ }
208
+ /**
209
+ * Render a report as text.
210
+ *
211
+ * Failures print their scorer reasons, because a CI log that says
212
+ * "0.62" is a log that sends someone back to reproduce it by hand.
213
+ */
214
+ export function formatReport(report) {
215
+ const lines = [
216
+ `${report.name}: ${report.passed}/${report.cases.length} passed (mean ${report.mean.toFixed(2)}) in ${report.durationMs}ms`,
217
+ '',
218
+ ];
219
+ // Surfaced above the failures, because an inconclusive case means the
220
+ // harness is broken and every number below it is measured over less
221
+ // evidence than it looks like.
222
+ if (report.inconclusive > 0) {
223
+ lines.push(` ${report.inconclusive} case${report.inconclusive === 1 ? '' : 's'} could not be judged — the numbers below cover the rest`, '');
224
+ }
225
+ for (const [name, mean] of Object.entries(report.byScorer)) {
226
+ lines.push(` ${name}: ${mean.toFixed(2)}`);
227
+ }
228
+ const failures = report.cases.filter((c) => c.status === 'failed');
229
+ if (failures.length > 0) {
230
+ lines.push('', 'Failures:');
231
+ for (const failure of failures) {
232
+ lines.push(` ✗ ${failure.case} (${failure.mean.toFixed(2)})`);
233
+ if (failure.failedGates && failure.failedGates.length > 0) {
234
+ // First line under the case, because it says WHY this failed.
235
+ // A mean of 0.75 next to four scores leaves the reader
236
+ // guessing which one mattered.
237
+ lines.push(` gate missed: ${failure.failedGates.join(', ')}`);
238
+ }
239
+ for (const [name, score] of Object.entries(failure.scores)) {
240
+ if (score.unavailable === true) {
241
+ lines.push(` ${name}: not judged — ${score.reason}`);
242
+ }
243
+ else if (score.score < 1) {
244
+ lines.push(` ${name}: ${score.reason}`);
245
+ }
246
+ }
247
+ if (failure.run.error)
248
+ lines.push(` error: ${failure.run.error}`);
249
+ }
250
+ }
251
+ const unjudged = report.cases.filter((c) => c.status === 'inconclusive');
252
+ if (unjudged.length > 0) {
253
+ lines.push('', 'Not judged:');
254
+ for (const item of unjudged) {
255
+ lines.push(` ? ${item.case}`);
256
+ for (const [name, score] of Object.entries(item.scores)) {
257
+ lines.push(` ${name}: ${score.reason}`);
258
+ }
259
+ }
260
+ }
261
+ return lines.join('\n');
262
+ }
263
+ //# sourceMappingURL=experiment.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"experiment.js","sourceRoot":"","sources":["../../src/eval/experiment.ts"],"names":[],"mappings":"AA2CA;;;;;;;;;;;GAWG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAClC,MAAgC;IAEhC,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;IAC5B,MAAM,SAAS,GAAG,MAAM,CAAC,aAAa,IAAI,CAAC,CAAA;IAC3C,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,WAAW,IAAI,CAAC,CAAC,CAAA;IAExD,MAAM,OAAO,GAAiB,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAA;IAC5D,IAAI,MAAM,GAAG,CAAC,CAAA;IAEd,MAAM,MAAM,GAAG,KAAK,IAAmB,EAAE;QACxC,SAAS,CAAC;YACT,MAAM,KAAK,GAAG,MAAM,EAAE,CAAA;YACtB,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,CAAC,CAAA;YACpC,IAAI,CAAC,QAAQ;gBAAE,OAAM;YAErB,MAAM,GAAG,GAAG,MAAM,WAAW,CAAC,MAAM,EAAE,QAAQ,CAAC,CAAA;YAC/C,MAAM,OAAO,GAAG,QAAQ,CAAC,OAAO,IAAI,MAAM,CAAC,OAAO,CAAA;YAClD,MAAM,MAAM,GAA0B,EAAE,CAAA;YAExC,iEAAiE;YACjE,iEAAiE;YACjE,iDAAiD;YACjD,iEAAiE;YACjE,0DAA0D;YAC1D,6DAA6D;YAC7D,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAA;YAC9B,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;gBAC9B,IAAI,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,CAAC;oBAC3B,MAAM,IAAI,KAAK,CACd,0BAA0B,MAAM,CAAC,IAAI,eAAe,QAAQ,CAAC,IAAI,wKAAwK,CACzO,CAAA;gBACF,CAAC;gBACD,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;gBACrB,MAAM,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,MAAM,SAAS,CAAC,MAAM,EAAE,GAAG,EAAE,QAAQ,CAAC,CAAA;YAC7D,CAAC;YAED,iEAAiE;YACjE,+DAA+D;YAC/D,2DAA2D;YAC3D,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC;iBAClC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,IAAI,CAAC;iBACrC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAA;YACrB,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAA;YAExF,iEAAiE;YACjE,8DAA8D;YAC9D,2DAA2D;YAC3D,gEAAgE;YAChE,2DAA2D;YAC3D,8DAA8D;YAC9D,6DAA6D;YAC7D,uBAAuB;YACvB,MAAM,WAAW,GAAG,OAAO;iBACzB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,MAAM,CAAC;iBACpC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE;gBACb,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;gBAC5B,IAAI,CAAC,KAAK,IAAI,KAAK,CAAC,WAAW,KAAK,IAAI;oBAAE,OAAO,KAAK,CAAA;gBACtD,OAAO,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC,SAAS,IAAI,SAAS,CAAC,CAAA;YAChD,CAAC,CAAC;iBACD,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;YAEpB,MAAM,MAAM,GACX,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;gBACpD,CAAC,CAAC,cAAc;gBAChB,CAAC,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC;oBACvB,CAAC,CAAC,QAAQ;oBACV,CAAC,CAAC,IAAI,IAAI,SAAS;wBAClB,CAAC,CAAC,QAAQ;wBACV,CAAC,CAAC,QAAQ,CAAA;YACd,MAAM,MAAM,GAAe;gBAC1B,IAAI,EAAE,QAAQ,CAAC,IAAI;gBACnB,GAAG;gBACH,MAAM;gBACN,IAAI;gBACJ,MAAM;gBACN,MAAM,EAAE,MAAM,KAAK,QAAQ;gBAC3B,8DAA8D;gBAC9D,6DAA6D;gBAC7D,GAAG,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,WAAW,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aAClD,CAAA;YACD,OAAO,CAAC,KAAK,CAAC,GAAG,MAAM,CAAA;YACvB,MAAM,CAAC,YAAY,EAAE,CAAC,MAAM,CAAC,CAAA;QAC9B,CAAC;IACF,CAAC,CAAA;IAED,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,MAAM,CAAC,CAAC,CAAA;IAE7F,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO,CAAC,CAAA;IACvC,wEAAwE;IACxE,8DAA8D;IAC9D,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,cAAc,CAAC,CAAA;IACjE,MAAM,IAAI,GAAG,MAAM,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAA;IAEjG,OAAO;QACN,IAAI,EAAE,MAAM,CAAC,IAAI;QACjB,KAAK,EAAE,OAAO;QACd,IAAI;QACJ,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM;QAC3D,MAAM,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM;QAC3D,YAAY,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,cAAc,CAAC,CAAC,MAAM;QACvE,QAAQ,EAAE,YAAY,CAAC,OAAO,CAAC;QAC/B,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;KAClC,CAAA;AACF,CAAC;AAED;;;GAGG;AACH,KAAK,UAAU,WAAW,CACzB,MAAgC,EAChC,QAA0B;IAE1B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;IAC5B,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAA;IACxC,MAAM,SAAS,GAAG,MAAM,CAAC,SAAS,CAAA;IAClC,uEAAuE;IACvE,uEAAuE;IACvE,wEAAwE;IACxE,qEAAqE;IACrE,sCAAsC;IACtC,MAAM,KAAK,GACV,SAAS,KAAK,SAAS,IAAI,SAAS,GAAG,CAAC;QACvC,CAAC,CAAC,UAAU,CACV,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,wBAAwB,SAAS,IAAI,CAAC,CAAC,EACxE,SAAS,CACT;QACF,CAAC,CAAC,SAAS,CAAA;IAEb,MAAM,QAAQ,GACb,KAAK,KAAK,SAAS;QAClB,CAAC,CAAC,SAAS;QACX,CAAC,CAAC,IAAI,OAAO,CAAQ,CAAC,QAAQ,EAAE,MAAM,EAAE,EAAE;YACxC,UAAU,CAAC,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,GAAG,EAAE,CAAC,MAAM,CAAC,UAAU,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE;gBACnF,IAAI,EAAE,IAAI;aACV,CAAC,CAAA;QACH,CAAC,CAAC,CAAA;IAEL,IAAI,CAAC;QACJ,MAAM,IAAI,GAAG,MAAM,CAAC,GAAG,CAAC,QAAQ,CAAC,KAAK,EAAE,QAAQ,EAAE,UAAU,CAAC,MAAM,CAAC,CAAA;QACpE,8DAA8D;QAC9D,qDAAqD;QACrD,IAAI,QAAQ;YAAE,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAA;QAClC,OAAO,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAA;IAChE,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACd,OAAO;YACN,MAAM,EAAE,IAAI;YACZ,KAAK,EAAE,EAAE;YACT,SAAS,EAAE,EAAE;YACb,WAAW,EAAE,CAAC;YACd,YAAY,EAAE,CAAC;YACf,4DAA4D;YAC5D,6DAA6D;YAC7D,sBAAsB;YACtB,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAClC,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;SACvD,CAAA;IACF,CAAC;YAAS,CAAC;QACV,qEAAqE;QACrE,mEAAmE;QACnE,qDAAqD;QACrD,YAAY,CAAC,KAAK,CAAC,CAAA;IACpB,CAAC;AACF,CAAC;AAED,kEAAkE;AAClE,KAAK,UAAU,SAAS,CAAC,MAAc,EAAE,GAAY,EAAE,QAAkB;IACxE,qEAAqE;IACrE,qEAAqE;IACrE,2DAA2D;IAC3D,uEAAuE;IACvE,wEAAwE;IACxE,yEAAyE;IACzE,oDAAoD;IACpD,IAAI,GAAG,CAAC,KAAK,KAAK,SAAS,EAAE,CAAC;QAC7B,OAAO,EAAE,KAAK,EAAE,CAAC,EAAE,MAAM,EAAE,eAAe,GAAG,CAAC,KAAK,EAAE,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,GAAG,CAAC,KAAK,EAAE,EAAE,CAAA;IACvF,CAAC;IAED,IAAI,CAAC;QACJ,OAAO,MAAM,MAAM,CAAC,KAAK,CAAC,GAAG,EAAE,QAAQ,CAAC,CAAA;IACzC,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACd,mEAAmE;QACnE,6DAA6D;QAC7D,kEAAkE;QAClE,qEAAqE;QACrE,kEAAkE;QAClE,yCAAyC;QACzC,OAAO;YACN,KAAK,EAAE,CAAC;YACR,WAAW,EAAE,IAAI;YACjB,MAAM,EAAE,WAAW,MAAM,CAAC,IAAI,+BAC7B,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAChD,EAAE;SACF,CAAA;IACF,CAAC;AACF,CAAC;AAED,SAAS,YAAY,CAAC,OAA8B;IACnD,MAAM,IAAI,GAAG,IAAI,GAAG,EAA4C,CAAA;IAChE,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC9B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC;YAC3D,IAAI,KAAK,CAAC,WAAW,KAAK,IAAI;gBAAE,SAAQ;YACxC,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAA;YACtD,KAAK,CAAC,KAAK,IAAI,KAAK,CAAC,KAAK,CAAA;YAC1B,KAAK,CAAC,KAAK,EAAE,CAAA;YACb,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,KAAK,CAAC,CAAA;QACtB,CAAC;IACF,CAAC;IACD,MAAM,GAAG,GAA2B,EAAE,CAAA;IACtC,oEAAoE;IACpE,sEAAsE;IACtE,qBAAqB;IACrB,KAAK,MAAM,CAAC,IAAI,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,CAAC,IAAI,IAAI,EAAE,CAAC;QAC7C,IAAI,KAAK,GAAG,CAAC;YAAE,GAAG,CAAC,IAAI,CAAC,GAAG,KAAK,GAAG,KAAK,CAAA;IACzC,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,YAAY,CAAC,MAAwB;IACpD,MAAM,KAAK,GAAa;QACvB,GAAG,MAAM,CAAC,IAAI,KAAK,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,iBAAiB,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,MAAM,CAAC,UAAU,IAAI;QAC3H,EAAE;KACF,CAAA;IAED,sEAAsE;IACtE,oEAAoE;IACpE,+BAA+B;IAC/B,IAAI,MAAM,CAAC,YAAY,GAAG,CAAC,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CACT,KAAK,MAAM,CAAC,YAAY,QAAQ,MAAM,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,yDAAyD,EAC7H,EAAE,CACF,CAAA;IACF,CAAC;IAED,KAAK,MAAM,CAAC,IAAI,EAAE,IAAI,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QAC5D,KAAK,CAAC,IAAI,CAAC,KAAK,IAAI,KAAK,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAA;IAC5C,CAAC;IAED,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAA;IAClE,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,WAAW,CAAC,CAAA;QAC3B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,OAAO,OAAO,CAAC,IAAI,KAAK,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAA;YAC9D,IAAI,OAAO,CAAC,WAAW,IAAI,OAAO,CAAC,WAAW,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC3D,8DAA8D;gBAC9D,uDAAuD;gBACvD,+BAA+B;gBAC/B,KAAK,CAAC,IAAI,CAAC,sBAAsB,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;YACnE,CAAC;YACD,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,EAAE,CAAC;gBAC5D,IAAI,KAAK,CAAC,WAAW,KAAK,IAAI,EAAE,CAAC;oBAChC,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,kBAAkB,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;gBAC1D,CAAC;qBAAM,IAAI,KAAK,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC;oBAC5B,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,KAAK,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;gBAC7C,CAAC;YACF,CAAC;YACD,IAAI,OAAO,CAAC,GAAG,CAAC,KAAK;gBAAE,KAAK,CAAC,IAAI,CAAC,gBAAgB,OAAO,CAAC,GAAG,CAAC,KAAK,EAAE,CAAC,CAAA;QACvE,CAAC;IACF,CAAC;IAED,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,cAAc,CAAC,CAAA;IACxE,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACzB,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,aAAa,CAAC,CAAA;QAC7B,KAAK,MAAM,IAAI,IAAI,QAAQ,EAAE,CAAC;YAC7B,KAAK,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,IAAI,EAAE,CAAC,CAAA;YAC9B,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,EAAE,CAAC;gBACzD,KAAK,CAAC,IAAI,CAAC,SAAS,IAAI,KAAK,KAAK,CAAC,MAAM,EAAE,CAAC,CAAA;YAC7C,CAAC;QACF,CAAC;IACF,CAAC;IAED,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACxB,CAAC"}
@@ -0,0 +1,30 @@
1
+ import type { Run } from '../types/run/entity.js';
2
+ import type { EvalRun } from './types.js';
3
+ /**
4
+ * Project a finished `Run` into the shape scorers consume.
5
+ *
6
+ * This is the whole bridge between the harness and the runtime, and it is
7
+ * three lines of mapping because `Run.steps` now exists. Before that, a
8
+ * trajectory scorer would have had to correlate raw `RunEvent`s by
9
+ * iteration number and diff cumulative counters — which is why the step
10
+ * record had to land before the harness could.
11
+ */
12
+ export declare function evalRunFromRun(run: Run, opts?: {
13
+ durationMs?: number;
14
+ }): EvalRun;
15
+ /**
16
+ * Drain a `query()` generator to its returned `Run`, then project it.
17
+ *
18
+ * The common case, so it does not need writing per suite:
19
+ *
20
+ * ```ts
21
+ * runExperiment({
22
+ * name: 'file-editing',
23
+ * cases,
24
+ * scorers: [trajectoryScorer(), completionScorer()],
25
+ * run: (input) => evalRunFromQuery(query({ provider, tools, messages: input, … })),
26
+ * })
27
+ * ```
28
+ */
29
+ export declare function evalRunFromQuery(generator: AsyncGenerator<unknown, Run>): Promise<EvalRun>;
30
+ //# sourceMappingURL=from-run.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"from-run.d.ts","sourceRoot":"","sources":["../../src/eval/from-run.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,GAAG,EAAE,MAAM,wBAAwB,CAAA;AACjD,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,YAAY,CAAA;AAEzC;;;;;;;;GAQG;AACH,wBAAgB,cAAc,CAAC,GAAG,EAAE,GAAG,EAAE,IAAI,GAAE;IAAE,UAAU,CAAC,EAAE,MAAM,CAAA;CAAO,GAAG,OAAO,CAcpF;AAED;;;;;;;;;;;;;GAaG;AACH,wBAAsB,gBAAgB,CAAC,SAAS,EAAE,cAAc,CAAC,OAAO,EAAE,GAAG,CAAC,GAAG,OAAO,CAAC,OAAO,CAAC,CAKhG"}
@@ -0,0 +1,46 @@
1
+ /**
2
+ * Project a finished `Run` into the shape scorers consume.
3
+ *
4
+ * This is the whole bridge between the harness and the runtime, and it is
5
+ * three lines of mapping because `Run.steps` now exists. Before that, a
6
+ * trajectory scorer would have had to correlate raw `RunEvent`s by
7
+ * iteration number and diff cumulative counters — which is why the step
8
+ * record had to land before the harness could.
9
+ */
10
+ export function evalRunFromRun(run, opts = {}) {
11
+ const steps = run.steps ?? [];
12
+ return {
13
+ output: run.result ?? null,
14
+ ...(run.structuredOutput !== undefined ? { structuredOutput: run.structuredOutput } : {}),
15
+ steps,
16
+ // Flattened in call order across steps — the trajectory.
17
+ toolCalls: steps.flatMap((s) => s.toolCalls.map((c) => c.function.name)),
18
+ ...(run.stopReason ? { stopReason: run.stopReason } : {}),
19
+ totalTokens: run.tokenUsage.totalTokens,
20
+ totalCostUsd: run.costInfo.totalCost,
21
+ durationMs: opts.durationMs ?? (run.endedAt ?? Date.now()) - run.startedAt,
22
+ ...(run.lastError ? { error: run.lastError } : {}),
23
+ };
24
+ }
25
+ /**
26
+ * Drain a `query()` generator to its returned `Run`, then project it.
27
+ *
28
+ * The common case, so it does not need writing per suite:
29
+ *
30
+ * ```ts
31
+ * runExperiment({
32
+ * name: 'file-editing',
33
+ * cases,
34
+ * scorers: [trajectoryScorer(), completionScorer()],
35
+ * run: (input) => evalRunFromQuery(query({ provider, tools, messages: input, … })),
36
+ * })
37
+ * ```
38
+ */
39
+ export async function evalRunFromQuery(generator) {
40
+ const startedAt = Date.now();
41
+ let next = await generator.next();
42
+ while (!next.done)
43
+ next = await generator.next();
44
+ return evalRunFromRun(next.value, { durationMs: Date.now() - startedAt });
45
+ }
46
+ //# sourceMappingURL=from-run.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"from-run.js","sourceRoot":"","sources":["../../src/eval/from-run.ts"],"names":[],"mappings":"AAGA;;;;;;;;GAQG;AACH,MAAM,UAAU,cAAc,CAAC,GAAQ,EAAE,OAAgC,EAAE;IAC1E,MAAM,KAAK,GAAG,GAAG,CAAC,KAAK,IAAI,EAAE,CAAA;IAC7B,OAAO;QACN,MAAM,EAAE,GAAG,CAAC,MAAM,IAAI,IAAI;QAC1B,GAAG,CAAC,GAAG,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,gBAAgB,EAAE,GAAG,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzF,KAAK;QACL,yDAAyD;QACzD,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QACxE,GAAG,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,EAAE,UAAU,EAAE,GAAG,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;QACzD,WAAW,EAAE,GAAG,CAAC,UAAU,CAAC,WAAW;QACvC,YAAY,EAAE,GAAG,CAAC,QAAQ,CAAC,SAAS;QACpC,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,GAAG,CAAC,OAAO,IAAI,IAAI,CAAC,GAAG,EAAE,CAAC,GAAG,GAAG,CAAC,SAAS;QAC1E,GAAG,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,KAAK,EAAE,GAAG,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KAClD,CAAA;AACF,CAAC;AAED;;;;;;;;;;;;;GAaG;AACH,MAAM,CAAC,KAAK,UAAU,gBAAgB,CAAC,SAAuC;IAC7E,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAA;IAC5B,IAAI,IAAI,GAAG,MAAM,SAAS,CAAC,IAAI,EAAE,CAAA;IACjC,OAAO,CAAC,IAAI,CAAC,IAAI;QAAE,IAAI,GAAG,MAAM,SAAS,CAAC,IAAI,EAAE,CAAA;IAChD,OAAO,cAAc,CAAC,IAAI,CAAC,KAAK,EAAE,EAAE,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS,EAAE,CAAC,CAAA;AAC1E,CAAC"}
@@ -0,0 +1,8 @@
1
+ export { evalRunFromQuery, evalRunFromRun } from './from-run.js';
2
+ export { formatReport, runExperiment } from './experiment.js';
3
+ export type { ExperimentConfig } from './experiment.js';
4
+ export { judgeScorer } from './judge.js';
5
+ export type { JudgeScorerConfig } from './judge.js';
6
+ export { completionScorer, containsScorer, customScorer, stepBudgetScorer, trajectoryScorer, } from './scorers.js';
7
+ export type { CaseResult, CaseStatus, EvalCase, EvalRun, ExperimentReport, Score, Scorer, } from './types.js';
8
+ //# sourceMappingURL=index.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,eAAe,CAAA;AAChE,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAA;AAC7D,YAAY,EAAE,gBAAgB,EAAE,MAAM,iBAAiB,CAAA;AACvD,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA;AACxC,YAAY,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAA;AACnD,OAAO,EACN,gBAAgB,EAChB,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,GAChB,MAAM,cAAc,CAAA;AACrB,YAAY,EACX,UAAU,EACV,UAAU,EACV,QAAQ,EACR,OAAO,EACP,gBAAgB,EAChB,KAAK,EACL,MAAM,GACN,MAAM,YAAY,CAAA"}
@@ -0,0 +1,5 @@
1
+ export { evalRunFromQuery, evalRunFromRun } from './from-run.js';
2
+ export { formatReport, runExperiment } from './experiment.js';
3
+ export { judgeScorer } from './judge.js';
4
+ export { completionScorer, containsScorer, customScorer, stepBudgetScorer, trajectoryScorer, } from './scorers.js';
5
+ //# sourceMappingURL=index.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/eval/index.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAE,cAAc,EAAE,MAAM,eAAe,CAAA;AAChE,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,iBAAiB,CAAA;AAE7D,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAA;AAExC,OAAO,EACN,gBAAgB,EAChB,cAAc,EACd,YAAY,EACZ,gBAAgB,EAChB,gBAAgB,GAChB,MAAM,cAAc,CAAA"}