@roackb2/heddle 0.0.37 → 1.0.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (578) hide show
  1. package/README.md +11 -0
  2. package/dist/examples/conversation-engine.d.ts +2 -0
  3. package/dist/examples/conversation-engine.d.ts.map +1 -0
  4. package/dist/examples/conversation-engine.js +127 -0
  5. package/dist/examples/conversation-engine.js.map +1 -0
  6. package/dist/examples/repo-investigator.js +1 -2
  7. package/dist/examples/repo-investigator.js.map +1 -1
  8. package/dist/src/cli/ask.d.ts.map +1 -1
  9. package/dist/src/cli/ask.js +22 -3
  10. package/dist/src/cli/ask.js.map +1 -1
  11. package/dist/src/cli/chat/App.js +43 -27
  12. package/dist/src/cli/chat/App.js.map +1 -1
  13. package/dist/src/cli/chat/adapters/conversation-activity-adapter.d.ts +5 -0
  14. package/dist/src/cli/chat/adapters/conversation-activity-adapter.d.ts.map +1 -0
  15. package/dist/src/cli/chat/adapters/conversation-activity-adapter.js +43 -0
  16. package/dist/src/cli/chat/adapters/conversation-activity-adapter.js.map +1 -0
  17. package/dist/src/cli/chat/adapters/slash-command-context.d.ts +4 -0
  18. package/dist/src/cli/chat/adapters/slash-command-context.d.ts.map +1 -0
  19. package/dist/src/cli/chat/adapters/slash-command-context.js +48 -0
  20. package/dist/src/cli/chat/adapters/slash-command-context.js.map +1 -0
  21. package/dist/src/cli/chat/commands/debug-snapshot-command.d.ts +7 -0
  22. package/dist/src/cli/chat/commands/debug-snapshot-command.d.ts.map +1 -0
  23. package/dist/src/cli/chat/commands/debug-snapshot-command.js +25 -0
  24. package/dist/src/cli/chat/commands/debug-snapshot-command.js.map +1 -0
  25. package/dist/src/cli/chat/components/ApprovalComposer.js.map +1 -1
  26. package/dist/src/cli/chat/components/ConversationPanel.d.ts +42 -1
  27. package/dist/src/cli/chat/components/ConversationPanel.d.ts.map +1 -1
  28. package/dist/src/cli/chat/components/ConversationPanel.js +114 -67
  29. package/dist/src/cli/chat/components/ConversationPanel.js.map +1 -1
  30. package/dist/src/cli/chat/components/ModelPickerPanel.d.ts +2 -1
  31. package/dist/src/cli/chat/components/ModelPickerPanel.d.ts.map +1 -1
  32. package/dist/src/cli/chat/components/ModelPickerPanel.js +8 -4
  33. package/dist/src/cli/chat/components/ModelPickerPanel.js.map +1 -1
  34. package/dist/src/cli/chat/debug/tui-debug-snapshot.d.ts +2 -1
  35. package/dist/src/cli/chat/debug/tui-debug-snapshot.d.ts.map +1 -1
  36. package/dist/src/cli/chat/debug/tui-debug-snapshot.js +14 -1
  37. package/dist/src/cli/chat/debug/tui-debug-snapshot.js.map +1 -1
  38. package/dist/src/cli/chat/hooks/tui-agent-turn-result.d.ts.map +1 -1
  39. package/dist/src/cli/chat/hooks/tui-agent-turn-result.js +3 -2
  40. package/dist/src/cli/chat/hooks/tui-agent-turn-result.js.map +1 -1
  41. package/dist/src/cli/chat/hooks/tui-compaction-status.d.ts +1 -1
  42. package/dist/src/cli/chat/hooks/tui-compaction-status.d.ts.map +1 -1
  43. package/dist/src/cli/chat/hooks/tui-compaction-status.js.map +1 -1
  44. package/dist/src/cli/chat/hooks/tui-direct-shell-result.d.ts.map +1 -1
  45. package/dist/src/cli/chat/hooks/tui-direct-shell-result.js +2 -1
  46. package/dist/src/cli/chat/hooks/tui-direct-shell-result.js.map +1 -1
  47. package/dist/src/cli/chat/hooks/tui-direct-shell.d.ts.map +1 -1
  48. package/dist/src/cli/chat/hooks/tui-direct-shell.js +21 -7
  49. package/dist/src/cli/chat/hooks/tui-direct-shell.js.map +1 -1
  50. package/dist/src/cli/chat/hooks/tui-ordinary-turn.d.ts +2 -2
  51. package/dist/src/cli/chat/hooks/tui-ordinary-turn.d.ts.map +1 -1
  52. package/dist/src/cli/chat/hooks/tui-ordinary-turn.js +6 -5
  53. package/dist/src/cli/chat/hooks/tui-ordinary-turn.js.map +1 -1
  54. package/dist/src/cli/chat/hooks/tui-run-loop-events.d.ts.map +1 -1
  55. package/dist/src/cli/chat/hooks/tui-run-loop-events.js +13 -9
  56. package/dist/src/cli/chat/hooks/tui-run-loop-events.js.map +1 -1
  57. package/dist/src/cli/chat/hooks/tui-tool-approval.d.ts +8 -3
  58. package/dist/src/cli/chat/hooks/tui-tool-approval.d.ts.map +1 -1
  59. package/dist/src/cli/chat/hooks/tui-tool-approval.js +33 -25
  60. package/dist/src/cli/chat/hooks/tui-tool-approval.js.map +1 -1
  61. package/dist/src/cli/chat/hooks/useAgentRun.d.ts +2 -2
  62. package/dist/src/cli/chat/hooks/useAgentRun.d.ts.map +1 -1
  63. package/dist/src/cli/chat/hooks/useAgentRun.js +2 -2
  64. package/dist/src/cli/chat/hooks/useAgentRun.js.map +1 -1
  65. package/dist/src/cli/chat/hooks/useApprovalFlow.d.ts +2 -2
  66. package/dist/src/cli/chat/hooks/useApprovalFlow.d.ts.map +1 -1
  67. package/dist/src/cli/chat/hooks/useApprovalFlow.js +3 -3
  68. package/dist/src/cli/chat/hooks/useApprovalFlow.js.map +1 -1
  69. package/dist/src/cli/chat/hooks/useChatPickers.d.ts +6 -3
  70. package/dist/src/cli/chat/hooks/useChatPickers.d.ts.map +1 -1
  71. package/dist/src/cli/chat/hooks/useChatPickers.js +17 -4
  72. package/dist/src/cli/chat/hooks/useChatPickers.js.map +1 -1
  73. package/dist/src/cli/chat/hooks/useChatSessions.d.ts +1 -1
  74. package/dist/src/cli/chat/hooks/useChatSessions.d.ts.map +1 -1
  75. package/dist/src/cli/chat/hooks/useChatSessions.js +25 -22
  76. package/dist/src/cli/chat/hooks/useChatSessions.js.map +1 -1
  77. package/dist/src/cli/chat/hooks/useChatStatusSummary.d.ts +1 -1
  78. package/dist/src/cli/chat/hooks/useChatStatusSummary.d.ts.map +1 -1
  79. package/dist/src/cli/chat/hooks/usePromptSubmission.d.ts +2 -1
  80. package/dist/src/cli/chat/hooks/usePromptSubmission.d.ts.map +1 -1
  81. package/dist/src/cli/chat/hooks/usePromptSubmission.js +4 -1
  82. package/dist/src/cli/chat/hooks/usePromptSubmission.js.map +1 -1
  83. package/dist/src/cli/chat/state/approval-rules.d.ts +2 -20
  84. package/dist/src/cli/chat/state/approval-rules.d.ts.map +1 -1
  85. package/dist/src/cli/chat/state/approval-rules.js +1 -256
  86. package/dist/src/cli/chat/state/approval-rules.js.map +1 -1
  87. package/dist/src/cli/chat/state/compaction.d.ts +1 -1
  88. package/dist/src/cli/chat/state/compaction.d.ts.map +1 -1
  89. package/dist/src/cli/chat/state/compaction.js +1 -1
  90. package/dist/src/cli/chat/state/compaction.js.map +1 -1
  91. package/dist/src/cli/chat/state/local-commands.d.ts +5 -2
  92. package/dist/src/cli/chat/state/local-commands.d.ts.map +1 -1
  93. package/dist/src/cli/chat/state/local-commands.js +35 -410
  94. package/dist/src/cli/chat/state/local-commands.js.map +1 -1
  95. package/dist/src/cli/chat/state/storage.d.ts +1 -1
  96. package/dist/src/cli/chat/state/storage.d.ts.map +1 -1
  97. package/dist/src/cli/chat/state/storage.js +1 -1
  98. package/dist/src/cli/chat/state/storage.js.map +1 -1
  99. package/dist/src/cli/chat/state/types.d.ts +2 -1
  100. package/dist/src/cli/chat/state/types.d.ts.map +1 -1
  101. package/dist/src/cli/chat/submit.d.ts.map +1 -1
  102. package/dist/src/cli/chat/submit.js +13 -1
  103. package/dist/src/cli/chat/submit.js.map +1 -1
  104. package/dist/src/cli/chat/utils/file-mentions.js +1 -1
  105. package/dist/src/cli/chat/utils/file-mentions.js.map +1 -1
  106. package/dist/src/cli/chat/utils/format.d.ts +4 -12
  107. package/dist/src/cli/chat/utils/format.d.ts.map +1 -1
  108. package/dist/src/cli/chat/utils/format.js +12 -167
  109. package/dist/src/cli/chat/utils/format.js.map +1 -1
  110. package/dist/src/cli/chat/utils/runtime.d.ts +1 -1
  111. package/dist/src/cli/chat/utils/runtime.d.ts.map +1 -1
  112. package/dist/src/cli/chat/utils/runtime.js +1 -1
  113. package/dist/src/cli/chat/utils/runtime.js.map +1 -1
  114. package/dist/src/cli/eval/index.d.ts +24 -0
  115. package/dist/src/cli/eval/index.d.ts.map +1 -0
  116. package/dist/src/cli/eval/index.js +232 -0
  117. package/dist/src/cli/eval/index.js.map +1 -0
  118. package/dist/src/cli/main.js +25 -3
  119. package/dist/src/cli/main.js.map +1 -1
  120. package/dist/src/cli/remote/control-plane-client.d.ts +5 -1
  121. package/dist/src/cli/remote/control-plane-client.d.ts.map +1 -1
  122. package/dist/src/cli/session.js +1 -1
  123. package/dist/src/cli/session.js.map +1 -1
  124. package/dist/src/core/agent/mutation-tracking.d.ts +0 -7
  125. package/dist/src/core/agent/mutation-tracking.d.ts.map +1 -1
  126. package/dist/src/core/agent/mutation-tracking.js +5 -63
  127. package/dist/src/core/agent/mutation-tracking.js.map +1 -1
  128. package/dist/src/core/agent/post-mutation.d.ts +2 -2
  129. package/dist/src/core/agent/post-mutation.d.ts.map +1 -1
  130. package/dist/src/core/agent/post-mutation.js +5 -20
  131. package/dist/src/core/agent/post-mutation.js.map +1 -1
  132. package/dist/src/core/agent/progress-reminders.d.ts +1 -4
  133. package/dist/src/core/agent/progress-reminders.d.ts.map +1 -1
  134. package/dist/src/core/agent/progress-reminders.js +4 -56
  135. package/dist/src/core/agent/progress-reminders.js.map +1 -1
  136. package/dist/src/core/agent/run-agent.d.ts +2 -0
  137. package/dist/src/core/agent/run-agent.d.ts.map +1 -1
  138. package/dist/src/core/agent/run-agent.js +6 -103
  139. package/dist/src/core/agent/run-agent.js.map +1 -1
  140. package/dist/src/core/agent/tool-dispatch.d.ts +3 -0
  141. package/dist/src/core/agent/tool-dispatch.d.ts.map +1 -1
  142. package/dist/src/core/agent/tool-dispatch.js +24 -50
  143. package/dist/src/core/agent/tool-dispatch.js.map +1 -1
  144. package/dist/src/core/approvals/default-policies.d.ts +10 -0
  145. package/dist/src/core/approvals/default-policies.d.ts.map +1 -0
  146. package/dist/src/core/approvals/default-policies.js +37 -0
  147. package/dist/src/core/approvals/default-policies.js.map +1 -0
  148. package/dist/src/core/approvals/policy-chain.d.ts +9 -0
  149. package/dist/src/core/approvals/policy-chain.d.ts.map +1 -0
  150. package/dist/src/core/approvals/policy-chain.js +36 -0
  151. package/dist/src/core/approvals/policy-chain.js.map +1 -0
  152. package/dist/src/core/approvals/remembered-rules.d.ts +21 -0
  153. package/dist/src/core/approvals/remembered-rules.d.ts.map +1 -0
  154. package/dist/src/core/approvals/remembered-rules.js +257 -0
  155. package/dist/src/core/approvals/remembered-rules.js.map +1 -0
  156. package/dist/src/core/{chat/tool-approval-host.d.ts → approvals/surface.d.ts} +4 -6
  157. package/dist/src/core/approvals/surface.d.ts.map +1 -0
  158. package/dist/src/core/{chat/tool-approval-host.js → approvals/surface.js} +9 -1
  159. package/dist/src/core/approvals/surface.js.map +1 -0
  160. package/dist/src/core/approvals/types.d.ts +23 -0
  161. package/dist/src/core/approvals/types.d.ts.map +1 -0
  162. package/dist/src/core/approvals/types.js +2 -0
  163. package/dist/src/core/approvals/types.js.map +1 -0
  164. package/dist/src/core/chat/engine/config.d.ts +22 -0
  165. package/dist/src/core/chat/engine/config.d.ts.map +1 -0
  166. package/dist/src/core/chat/engine/config.js +27 -0
  167. package/dist/src/core/chat/engine/config.js.map +1 -0
  168. package/dist/src/core/chat/engine/conversation-engine.d.ts +3 -0
  169. package/dist/src/core/chat/engine/conversation-engine.d.ts.map +1 -0
  170. package/dist/src/core/chat/engine/conversation-engine.js +15 -0
  171. package/dist/src/core/chat/engine/conversation-engine.js.map +1 -0
  172. package/dist/src/core/chat/{compaction.d.ts → engine/history/compaction.d.ts} +3 -3
  173. package/dist/src/core/chat/engine/history/compaction.d.ts.map +1 -0
  174. package/dist/src/core/chat/{compaction.js → engine/history/compaction.js} +5 -5
  175. package/dist/src/core/chat/engine/history/compaction.js.map +1 -0
  176. package/dist/src/core/chat/engine/index.d.ts +5 -0
  177. package/dist/src/core/chat/engine/index.d.ts.map +1 -0
  178. package/dist/src/core/chat/engine/index.js +3 -0
  179. package/dist/src/core/chat/engine/index.js.map +1 -0
  180. package/dist/src/core/chat/{archive.d.ts → engine/sessions/archive.d.ts} +2 -2
  181. package/dist/src/core/chat/engine/sessions/archive.d.ts.map +1 -0
  182. package/dist/src/core/chat/engine/sessions/archive.js.map +1 -0
  183. package/dist/src/core/chat/engine/sessions/conversation-lines.d.ts +4 -0
  184. package/dist/src/core/chat/engine/sessions/conversation-lines.d.ts.map +1 -0
  185. package/dist/src/core/chat/{conversation-lines.js → engine/sessions/conversation-lines.js} +1 -1
  186. package/dist/src/core/chat/engine/sessions/conversation-lines.js.map +1 -0
  187. package/dist/src/core/chat/{session-lease.d.ts → engine/sessions/lease.d.ts} +2 -2
  188. package/dist/src/core/chat/engine/sessions/lease.d.ts.map +1 -0
  189. package/dist/src/core/chat/{session-lease.js → engine/sessions/lease.js} +1 -1
  190. package/dist/src/core/chat/engine/sessions/lease.js.map +1 -0
  191. package/dist/src/core/chat/engine/sessions/service.d.ts +6 -0
  192. package/dist/src/core/chat/engine/sessions/service.d.ts.map +1 -0
  193. package/dist/src/core/chat/engine/sessions/service.js +50 -0
  194. package/dist/src/core/chat/engine/sessions/service.js.map +1 -0
  195. package/dist/src/core/chat/{storage.d.ts → engine/sessions/storage.d.ts} +1 -1
  196. package/dist/src/core/chat/engine/sessions/storage.d.ts.map +1 -0
  197. package/dist/src/core/chat/{storage.js → engine/sessions/storage.js} +1 -1
  198. package/dist/src/core/chat/engine/sessions/storage.js.map +1 -0
  199. package/dist/src/core/chat/{session-title.d.ts → engine/sessions/title.d.ts} +2 -2
  200. package/dist/src/core/chat/engine/sessions/title.d.ts.map +1 -0
  201. package/dist/src/core/chat/{session-title.js → engine/sessions/title.js} +1 -1
  202. package/dist/src/core/chat/engine/sessions/title.js.map +1 -0
  203. package/dist/src/core/chat/engine/turns/context.d.ts +25 -0
  204. package/dist/src/core/chat/engine/turns/context.d.ts.map +1 -0
  205. package/dist/src/core/chat/engine/turns/context.js +41 -0
  206. package/dist/src/core/chat/engine/turns/context.js.map +1 -0
  207. package/dist/src/core/chat/engine/turns/host-bridge.d.ts +37 -0
  208. package/dist/src/core/chat/engine/turns/host-bridge.d.ts.map +1 -0
  209. package/dist/src/core/chat/engine/turns/host-bridge.js +20 -0
  210. package/dist/src/core/chat/engine/turns/host-bridge.js.map +1 -0
  211. package/dist/src/core/chat/engine/turns/host.d.ts +3 -0
  212. package/dist/src/core/chat/engine/turns/host.d.ts.map +1 -0
  213. package/dist/src/core/chat/engine/turns/host.js +42 -0
  214. package/dist/src/core/chat/engine/turns/host.js.map +1 -0
  215. package/dist/src/core/chat/engine/turns/memory-maintenance.d.ts +33 -0
  216. package/dist/src/core/chat/engine/turns/memory-maintenance.d.ts.map +1 -0
  217. package/dist/src/core/chat/engine/turns/memory-maintenance.js +120 -0
  218. package/dist/src/core/chat/engine/turns/memory-maintenance.js.map +1 -0
  219. package/dist/src/core/chat/engine/turns/persistence.d.ts +25 -0
  220. package/dist/src/core/chat/engine/turns/persistence.d.ts.map +1 -0
  221. package/dist/src/core/chat/engine/turns/persistence.js +45 -0
  222. package/dist/src/core/chat/engine/turns/persistence.js.map +1 -0
  223. package/dist/src/core/chat/{session-turn-preflight.d.ts → engine/turns/preflight.d.ts} +25 -6
  224. package/dist/src/core/chat/engine/turns/preflight.d.ts.map +1 -0
  225. package/dist/src/core/chat/engine/turns/preflight.js +96 -0
  226. package/dist/src/core/chat/engine/turns/preflight.js.map +1 -0
  227. package/dist/src/core/chat/{session-turn-result.d.ts → engine/turns/result.d.ts} +7 -5
  228. package/dist/src/core/chat/engine/turns/result.d.ts.map +1 -0
  229. package/dist/src/core/chat/{session-turn-result.js → engine/turns/result.js} +11 -9
  230. package/dist/src/core/chat/engine/turns/result.js.map +1 -0
  231. package/dist/src/core/chat/engine/turns/run-conversation-turn.d.ts +37 -0
  232. package/dist/src/core/chat/engine/turns/run-conversation-turn.d.ts.map +1 -0
  233. package/dist/src/core/chat/engine/turns/run-conversation-turn.js +125 -0
  234. package/dist/src/core/chat/engine/turns/run-conversation-turn.js.map +1 -0
  235. package/dist/src/core/chat/engine/turns/runtime.d.ts +26 -0
  236. package/dist/src/core/chat/engine/turns/runtime.d.ts.map +1 -0
  237. package/dist/src/core/chat/engine/turns/runtime.js +55 -0
  238. package/dist/src/core/chat/engine/turns/runtime.js.map +1 -0
  239. package/dist/src/core/chat/engine/turns/service.d.ts +6 -0
  240. package/dist/src/core/chat/engine/turns/service.d.ts.map +1 -0
  241. package/dist/src/core/chat/engine/turns/service.js +51 -0
  242. package/dist/src/core/chat/engine/turns/service.js.map +1 -0
  243. package/dist/src/core/chat/engine/turns/trace.d.ts +3 -0
  244. package/dist/src/core/chat/engine/turns/trace.d.ts.map +1 -0
  245. package/dist/src/core/chat/engine/turns/trace.js.map +1 -0
  246. package/dist/src/core/chat/engine/types.d.ts +111 -0
  247. package/dist/src/core/chat/engine/types.d.ts.map +1 -0
  248. package/dist/src/core/chat/engine/types.js +2 -0
  249. package/dist/src/core/chat/engine/types.js.map +1 -0
  250. package/dist/src/core/chat/types.d.ts +3 -20
  251. package/dist/src/core/chat/types.d.ts.map +1 -1
  252. package/dist/src/core/commands/slash/autocomplete.d.ts +6 -0
  253. package/dist/src/core/commands/slash/autocomplete.d.ts.map +1 -0
  254. package/dist/src/core/commands/slash/autocomplete.js +56 -0
  255. package/dist/src/core/commands/slash/autocomplete.js.map +1 -0
  256. package/dist/src/core/commands/slash/modules/auth/auth-commands.d.ts +5 -0
  257. package/dist/src/core/commands/slash/modules/auth/auth-commands.d.ts.map +1 -0
  258. package/dist/src/core/commands/slash/modules/auth/auth-commands.js +67 -0
  259. package/dist/src/core/commands/slash/modules/auth/auth-commands.js.map +1 -0
  260. package/dist/src/core/commands/slash/modules/compaction/compaction-commands.d.ts +5 -0
  261. package/dist/src/core/commands/slash/modules/compaction/compaction-commands.d.ts.map +1 -0
  262. package/dist/src/core/commands/slash/modules/compaction/compaction-commands.js +20 -0
  263. package/dist/src/core/commands/slash/modules/compaction/compaction-commands.js.map +1 -0
  264. package/dist/src/core/commands/slash/modules/context.d.ts +48 -0
  265. package/dist/src/core/commands/slash/modules/context.d.ts.map +1 -0
  266. package/dist/src/core/commands/slash/modules/context.js +2 -0
  267. package/dist/src/core/commands/slash/modules/context.js.map +1 -0
  268. package/dist/src/core/commands/slash/modules/core-command-modules.d.ts +5 -0
  269. package/dist/src/core/commands/slash/modules/core-command-modules.d.ts.map +1 -0
  270. package/dist/src/core/commands/slash/modules/core-command-modules.js +17 -0
  271. package/dist/src/core/commands/slash/modules/core-command-modules.js.map +1 -0
  272. package/dist/src/core/commands/slash/modules/drift/drift-commands.d.ts +6 -0
  273. package/dist/src/core/commands/slash/modules/drift/drift-commands.d.ts.map +1 -0
  274. package/dist/src/core/commands/slash/modules/drift/drift-commands.js +54 -0
  275. package/dist/src/core/commands/slash/modules/drift/drift-commands.js.map +1 -0
  276. package/dist/src/core/commands/slash/modules/heartbeat/heartbeat-commands.d.ts +12 -0
  277. package/dist/src/core/commands/slash/modules/heartbeat/heartbeat-commands.d.ts.map +1 -0
  278. package/dist/src/core/commands/slash/modules/heartbeat/heartbeat-commands.js +199 -0
  279. package/dist/src/core/commands/slash/modules/heartbeat/heartbeat-commands.js.map +1 -0
  280. package/dist/src/core/commands/slash/modules/model/model-commands.d.ts +7 -0
  281. package/dist/src/core/commands/slash/modules/model/model-commands.d.ts.map +1 -0
  282. package/dist/src/core/commands/slash/modules/model/model-commands.js +77 -0
  283. package/dist/src/core/commands/slash/modules/model/model-commands.js.map +1 -0
  284. package/dist/src/core/commands/slash/modules/results.d.ts +6 -0
  285. package/dist/src/core/commands/slash/modules/results.d.ts.map +1 -0
  286. package/dist/src/core/commands/slash/modules/results.js +15 -0
  287. package/dist/src/core/commands/slash/modules/results.js.map +1 -0
  288. package/dist/src/core/commands/slash/modules/session/session-commands.d.ts +11 -0
  289. package/dist/src/core/commands/slash/modules/session/session-commands.d.ts.map +1 -0
  290. package/dist/src/core/commands/slash/modules/session/session-commands.js +146 -0
  291. package/dist/src/core/commands/slash/modules/session/session-commands.js.map +1 -0
  292. package/dist/src/core/commands/slash/parser.d.ts +7 -0
  293. package/dist/src/core/commands/slash/parser.d.ts.map +1 -0
  294. package/dist/src/core/commands/slash/parser.js +44 -0
  295. package/dist/src/core/commands/slash/parser.js.map +1 -0
  296. package/dist/src/core/commands/slash/registry.d.ts +9 -0
  297. package/dist/src/core/commands/slash/registry.d.ts.map +1 -0
  298. package/dist/src/core/commands/slash/registry.js +50 -0
  299. package/dist/src/core/commands/slash/registry.js.map +1 -0
  300. package/dist/src/core/commands/slash/result-types.d.ts +20 -0
  301. package/dist/src/core/commands/slash/result-types.d.ts.map +1 -0
  302. package/dist/src/core/commands/slash/result-types.js +2 -0
  303. package/dist/src/core/commands/slash/result-types.js.map +1 -0
  304. package/dist/src/core/commands/slash/types.d.ts +28 -0
  305. package/dist/src/core/commands/slash/types.d.ts.map +1 -0
  306. package/dist/src/core/commands/slash/types.js +2 -0
  307. package/dist/src/core/commands/slash/types.js.map +1 -0
  308. package/dist/src/core/config.d.ts +1 -1
  309. package/dist/src/core/config.d.ts.map +1 -1
  310. package/dist/src/core/config.js +1 -1
  311. package/dist/src/core/config.js.map +1 -1
  312. package/dist/src/core/eval/agent-runner.d.ts +24 -0
  313. package/dist/src/core/eval/agent-runner.d.ts.map +1 -0
  314. package/dist/src/core/eval/agent-runner.js +151 -0
  315. package/dist/src/core/eval/agent-runner.js.map +1 -0
  316. package/dist/src/core/eval/case-loader.d.ts +7 -0
  317. package/dist/src/core/eval/case-loader.d.ts.map +1 -0
  318. package/dist/src/core/eval/case-loader.js +34 -0
  319. package/dist/src/core/eval/case-loader.js.map +1 -0
  320. package/dist/src/core/eval/check-runner.d.ts +8 -0
  321. package/dist/src/core/eval/check-runner.d.ts.map +1 -0
  322. package/dist/src/core/eval/check-runner.js +33 -0
  323. package/dist/src/core/eval/check-runner.js.map +1 -0
  324. package/dist/src/core/eval/cleanup.d.ts +20 -0
  325. package/dist/src/core/eval/cleanup.d.ts.map +1 -0
  326. package/dist/src/core/eval/cleanup.js +42 -0
  327. package/dist/src/core/eval/cleanup.js.map +1 -0
  328. package/dist/src/core/eval/git-artifacts.d.ts +26 -0
  329. package/dist/src/core/eval/git-artifacts.d.ts.map +1 -0
  330. package/dist/src/core/eval/git-artifacts.js +211 -0
  331. package/dist/src/core/eval/git-artifacts.js.map +1 -0
  332. package/dist/src/core/eval/process.d.ts +22 -0
  333. package/dist/src/core/eval/process.d.ts.map +1 -0
  334. package/dist/src/core/eval/process.js +65 -0
  335. package/dist/src/core/eval/process.js.map +1 -0
  336. package/dist/src/core/eval/progress.d.ts +28 -0
  337. package/dist/src/core/eval/progress.d.ts.map +1 -0
  338. package/dist/src/core/eval/progress.js +94 -0
  339. package/dist/src/core/eval/progress.js.map +1 -0
  340. package/dist/src/core/eval/report-writer.d.ts +7 -0
  341. package/dist/src/core/eval/report-writer.d.ts.map +1 -0
  342. package/dist/src/core/eval/report-writer.js +159 -0
  343. package/dist/src/core/eval/report-writer.js.map +1 -0
  344. package/dist/src/core/eval/schema.d.ts +206 -0
  345. package/dist/src/core/eval/schema.d.ts.map +1 -0
  346. package/dist/src/core/eval/schema.js +104 -0
  347. package/dist/src/core/eval/schema.js.map +1 -0
  348. package/dist/src/core/eval/trace-analyzer.d.ts +6 -0
  349. package/dist/src/core/eval/trace-analyzer.d.ts.map +1 -0
  350. package/dist/src/core/eval/trace-analyzer.js +106 -0
  351. package/dist/src/core/eval/trace-analyzer.js.map +1 -0
  352. package/dist/src/core/eval/workspace-fixture.d.ts +14 -0
  353. package/dist/src/core/eval/workspace-fixture.d.ts.map +1 -0
  354. package/dist/src/core/eval/workspace-fixture.js +235 -0
  355. package/dist/src/core/eval/workspace-fixture.js.map +1 -0
  356. package/dist/src/core/llm/model-policy.d.ts +26 -0
  357. package/dist/src/core/llm/model-policy.d.ts.map +1 -1
  358. package/dist/src/core/llm/model-policy.js +47 -0
  359. package/dist/src/core/llm/model-policy.js.map +1 -1
  360. package/dist/src/core/memory/maintainer-tools.js +1 -1
  361. package/dist/src/core/memory/maintainer-tools.js.map +1 -1
  362. package/dist/src/core/memory/visibility.js +1 -1
  363. package/dist/src/core/memory/visibility.js.map +1 -1
  364. package/dist/src/core/observability/conversation-activity.d.ts +114 -0
  365. package/dist/src/core/observability/conversation-activity.d.ts.map +1 -0
  366. package/dist/src/core/observability/conversation-activity.js +249 -0
  367. package/dist/src/core/observability/conversation-activity.js.map +1 -0
  368. package/dist/src/core/observability/semantic-conventions.d.ts +33 -0
  369. package/dist/src/core/observability/semantic-conventions.d.ts.map +1 -0
  370. package/dist/src/core/observability/semantic-conventions.js +33 -0
  371. package/dist/src/core/observability/semantic-conventions.js.map +1 -0
  372. package/dist/src/core/observability/trace-summarizers.d.ts +23 -0
  373. package/dist/src/core/observability/trace-summarizers.d.ts.map +1 -0
  374. package/dist/src/core/observability/trace-summarizers.js +60 -0
  375. package/dist/src/core/observability/trace-summarizers.js.map +1 -0
  376. package/dist/src/core/prompts/system-prompt.d.ts +1 -1
  377. package/dist/src/core/prompts/system-prompt.d.ts.map +1 -1
  378. package/dist/src/core/prompts/system-prompt.js +19 -100
  379. package/dist/src/core/prompts/system-prompt.js.map +1 -1
  380. package/dist/src/core/runtime/agent-loop.d.ts +2 -0
  381. package/dist/src/core/runtime/agent-loop.d.ts.map +1 -1
  382. package/dist/src/core/runtime/agent-loop.js +1 -0
  383. package/dist/src/core/runtime/agent-loop.js.map +1 -1
  384. package/dist/src/core/runtime/default-tools.d.ts.map +1 -1
  385. package/dist/src/core/runtime/default-tools.js +33 -54
  386. package/dist/src/core/runtime/default-tools.js.map +1 -1
  387. package/dist/src/core/runtime/heartbeat.d.ts +2 -0
  388. package/dist/src/core/runtime/heartbeat.d.ts.map +1 -1
  389. package/dist/src/core/runtime/heartbeat.js +1 -0
  390. package/dist/src/core/runtime/heartbeat.js.map +1 -1
  391. package/dist/src/core/tools/toolkit.d.ts +20 -0
  392. package/dist/src/core/tools/toolkit.d.ts.map +1 -0
  393. package/dist/src/core/tools/toolkit.js +21 -0
  394. package/dist/src/core/tools/toolkit.js.map +1 -0
  395. package/dist/src/core/tools/{delete-file.d.ts → toolkits/coding-files/delete-file.d.ts} +1 -1
  396. package/dist/src/core/tools/toolkits/coding-files/delete-file.d.ts.map +1 -0
  397. package/dist/src/core/tools/toolkits/coding-files/delete-file.js.map +1 -0
  398. package/dist/src/core/tools/{edit-file.d.ts → toolkits/coding-files/edit-file.d.ts} +1 -1
  399. package/dist/src/core/tools/toolkits/coding-files/edit-file.d.ts.map +1 -0
  400. package/dist/src/core/tools/toolkits/coding-files/edit-file.js.map +1 -0
  401. package/dist/src/core/tools/{file-edit-core.d.ts → toolkits/coding-files/file-edit-core.d.ts} +1 -1
  402. package/dist/src/core/tools/toolkits/coding-files/file-edit-core.d.ts.map +1 -0
  403. package/dist/src/core/tools/toolkits/coding-files/file-edit-core.js.map +1 -0
  404. package/dist/src/core/tools/toolkits/coding-files/index.d.ts +8 -0
  405. package/dist/src/core/tools/toolkits/coding-files/index.d.ts.map +1 -0
  406. package/dist/src/core/tools/toolkits/coding-files/index.js +8 -0
  407. package/dist/src/core/tools/toolkits/coding-files/index.js.map +1 -0
  408. package/dist/src/core/tools/{list-files.d.ts → toolkits/coding-files/list-files.d.ts} +1 -1
  409. package/dist/src/core/tools/toolkits/coding-files/list-files.d.ts.map +1 -0
  410. package/dist/src/core/tools/toolkits/coding-files/list-files.js.map +1 -0
  411. package/dist/src/core/tools/{move-file.d.ts → toolkits/coding-files/move-file.d.ts} +1 -1
  412. package/dist/src/core/tools/toolkits/coding-files/move-file.d.ts.map +1 -0
  413. package/dist/src/core/tools/toolkits/coding-files/move-file.js.map +1 -0
  414. package/dist/src/core/tools/{read-file.d.ts → toolkits/coding-files/read-file.d.ts} +1 -1
  415. package/dist/src/core/tools/toolkits/coding-files/read-file.d.ts.map +1 -0
  416. package/dist/src/core/tools/toolkits/coding-files/read-file.js.map +1 -0
  417. package/dist/src/core/tools/{search-files.d.ts → toolkits/coding-files/search-files.d.ts} +1 -1
  418. package/dist/src/core/tools/toolkits/coding-files/search-files.d.ts.map +1 -0
  419. package/dist/src/core/tools/toolkits/coding-files/search-files.js.map +1 -0
  420. package/dist/src/core/tools/toolkits/coding-files/toolkit.d.ts +3 -0
  421. package/dist/src/core/tools/toolkits/coding-files/toolkit.d.ts.map +1 -0
  422. package/dist/src/core/tools/toolkits/coding-files/toolkit.js +23 -0
  423. package/dist/src/core/tools/toolkits/coding-files/toolkit.js.map +1 -0
  424. package/dist/src/core/tools/toolkits/external-context/toolkit.d.ts +3 -0
  425. package/dist/src/core/tools/toolkits/external-context/toolkit.d.ts.map +1 -0
  426. package/dist/src/core/tools/toolkits/external-context/toolkit.js +23 -0
  427. package/dist/src/core/tools/toolkits/external-context/toolkit.js.map +1 -0
  428. package/dist/src/core/tools/{view-image.d.ts → toolkits/external-context/view-image.d.ts} +3 -3
  429. package/dist/src/core/tools/toolkits/external-context/view-image.d.ts.map +1 -0
  430. package/dist/src/core/tools/{view-image.js → toolkits/external-context/view-image.js} +5 -5
  431. package/dist/src/core/tools/toolkits/external-context/view-image.js.map +1 -0
  432. package/dist/src/core/tools/{web-search.d.ts → toolkits/external-context/web-search.d.ts} +3 -3
  433. package/dist/src/core/tools/toolkits/external-context/web-search.d.ts.map +1 -0
  434. package/dist/src/core/tools/{web-search.js → toolkits/external-context/web-search.js} +5 -5
  435. package/dist/src/core/tools/toolkits/external-context/web-search.js.map +1 -0
  436. package/dist/src/core/tools/toolkits/internal/toolkit.d.ts +3 -0
  437. package/dist/src/core/tools/toolkits/internal/toolkit.d.ts.map +1 -0
  438. package/dist/src/core/tools/toolkits/internal/toolkit.js +8 -0
  439. package/dist/src/core/tools/toolkits/internal/toolkit.js.map +1 -0
  440. package/dist/src/core/tools/{update-plan.d.ts → toolkits/internal/update-plan.d.ts} +1 -1
  441. package/dist/src/core/tools/toolkits/internal/update-plan.d.ts.map +1 -0
  442. package/dist/src/core/tools/toolkits/internal/update-plan.js.map +1 -0
  443. package/dist/src/core/tools/{memory-checkpoint.d.ts → toolkits/knowledge/memory-checkpoint.d.ts} +1 -1
  444. package/dist/src/core/tools/toolkits/knowledge/memory-checkpoint.d.ts.map +1 -0
  445. package/dist/src/core/tools/toolkits/knowledge/memory-checkpoint.js.map +1 -0
  446. package/dist/src/core/tools/{memory-notes.d.ts → toolkits/knowledge/memory-notes.d.ts} +1 -1
  447. package/dist/src/core/tools/toolkits/knowledge/memory-notes.d.ts.map +1 -0
  448. package/dist/src/core/tools/{memory-notes.js → toolkits/knowledge/memory-notes.js} +1 -1
  449. package/dist/src/core/tools/toolkits/knowledge/memory-notes.js.map +1 -0
  450. package/dist/src/core/tools/{record-knowledge.d.ts → toolkits/knowledge/record-knowledge.d.ts} +1 -1
  451. package/dist/src/core/tools/toolkits/knowledge/record-knowledge.d.ts.map +1 -0
  452. package/dist/src/core/tools/toolkits/knowledge/record-knowledge.js.map +1 -0
  453. package/dist/src/core/tools/toolkits/knowledge/toolkit.d.ts +3 -0
  454. package/dist/src/core/tools/toolkits/knowledge/toolkit.d.ts.map +1 -0
  455. package/dist/src/core/tools/toolkits/knowledge/toolkit.js +29 -0
  456. package/dist/src/core/tools/toolkits/knowledge/toolkit.js.map +1 -0
  457. package/dist/src/core/tools/toolkits/shell-process/index.d.ts +4 -0
  458. package/dist/src/core/tools/toolkits/shell-process/index.d.ts.map +1 -0
  459. package/dist/src/core/tools/toolkits/shell-process/index.js +3 -0
  460. package/dist/src/core/tools/toolkits/shell-process/index.js.map +1 -0
  461. package/dist/src/core/tools/toolkits/shell-process/run-shell.d.ts +18 -0
  462. package/dist/src/core/tools/toolkits/shell-process/run-shell.d.ts.map +1 -0
  463. package/dist/src/core/tools/toolkits/shell-process/run-shell.js +210 -0
  464. package/dist/src/core/tools/toolkits/shell-process/run-shell.js.map +1 -0
  465. package/dist/src/core/tools/{run-shell.d.ts → toolkits/shell-process/shell-policy.d.ts} +4 -17
  466. package/dist/src/core/tools/toolkits/shell-process/shell-policy.d.ts.map +1 -0
  467. package/dist/src/core/tools/{run-shell.js → toolkits/shell-process/shell-policy.js} +41 -238
  468. package/dist/src/core/tools/toolkits/shell-process/shell-policy.js.map +1 -0
  469. package/dist/src/core/tools/toolkits/shell-process/toolkit.d.ts +3 -0
  470. package/dist/src/core/tools/toolkits/shell-process/toolkit.d.ts.map +1 -0
  471. package/dist/src/core/tools/toolkits/shell-process/toolkit.js +11 -0
  472. package/dist/src/core/tools/toolkits/shell-process/toolkit.js.map +1 -0
  473. package/dist/src/index.d.ts +37 -25
  474. package/dist/src/index.d.ts.map +1 -1
  475. package/dist/src/index.js +24 -15
  476. package/dist/src/index.js.map +1 -1
  477. package/dist/src/server/features/control-plane/router.d.ts +5 -1
  478. package/dist/src/server/features/control-plane/router.d.ts.map +1 -1
  479. package/dist/src/server/features/control-plane/router.js +17 -2
  480. package/dist/src/server/features/control-plane/router.js.map +1 -1
  481. package/dist/src/server/features/control-plane/services/chat-session-events.d.ts +1 -1
  482. package/dist/src/server/features/control-plane/services/chat-session-events.d.ts.map +1 -1
  483. package/dist/src/server/features/control-plane/services/chat-sessions.d.ts +5 -2
  484. package/dist/src/server/features/control-plane/services/chat-sessions.d.ts.map +1 -1
  485. package/dist/src/server/features/control-plane/services/chat-sessions.js +67 -29
  486. package/dist/src/server/features/control-plane/services/chat-sessions.js.map +1 -1
  487. package/dist/src/server/router.d.ts +5 -1
  488. package/dist/src/server/router.d.ts.map +1 -1
  489. package/dist/src/web/assets/{MonacoDiffViewer-DM8Cy5Xf.js → MonacoDiffViewer-tvXEFlgh.js} +1 -1
  490. package/dist/src/web/assets/index-CYd4sslC.css +2 -0
  491. package/dist/src/web/assets/index-WQmlPSO9.js +56 -0
  492. package/dist/src/web/index.html +2 -2
  493. package/package.json +8 -2
  494. package/dist/src/core/chat/archive.d.ts.map +0 -1
  495. package/dist/src/core/chat/archive.js.map +0 -1
  496. package/dist/src/core/chat/compaction.d.ts.map +0 -1
  497. package/dist/src/core/chat/compaction.js.map +0 -1
  498. package/dist/src/core/chat/conversation-lines.d.ts +0 -4
  499. package/dist/src/core/chat/conversation-lines.d.ts.map +0 -1
  500. package/dist/src/core/chat/conversation-lines.js.map +0 -1
  501. package/dist/src/core/chat/ordinary-turn.d.ts +0 -34
  502. package/dist/src/core/chat/ordinary-turn.d.ts.map +0 -1
  503. package/dist/src/core/chat/ordinary-turn.js +0 -274
  504. package/dist/src/core/chat/ordinary-turn.js.map +0 -1
  505. package/dist/src/core/chat/session-lease.d.ts.map +0 -1
  506. package/dist/src/core/chat/session-lease.js.map +0 -1
  507. package/dist/src/core/chat/session-submit.d.ts +0 -35
  508. package/dist/src/core/chat/session-submit.d.ts.map +0 -1
  509. package/dist/src/core/chat/session-submit.js +0 -30
  510. package/dist/src/core/chat/session-submit.js.map +0 -1
  511. package/dist/src/core/chat/session-title.d.ts.map +0 -1
  512. package/dist/src/core/chat/session-title.js.map +0 -1
  513. package/dist/src/core/chat/session-turn-preflight.d.ts.map +0 -1
  514. package/dist/src/core/chat/session-turn-preflight.js +0 -43
  515. package/dist/src/core/chat/session-turn-preflight.js.map +0 -1
  516. package/dist/src/core/chat/session-turn-result.d.ts.map +0 -1
  517. package/dist/src/core/chat/session-turn-result.js.map +0 -1
  518. package/dist/src/core/chat/storage.d.ts.map +0 -1
  519. package/dist/src/core/chat/storage.js.map +0 -1
  520. package/dist/src/core/chat/tool-approval-host.d.ts.map +0 -1
  521. package/dist/src/core/chat/tool-approval-host.js.map +0 -1
  522. package/dist/src/core/chat/trace-summary.d.ts +0 -4
  523. package/dist/src/core/chat/trace-summary.d.ts.map +0 -1
  524. package/dist/src/core/chat/trace-summary.js +0 -76
  525. package/dist/src/core/chat/trace-summary.js.map +0 -1
  526. package/dist/src/core/chat/trace.d.ts +0 -3
  527. package/dist/src/core/chat/trace.d.ts.map +0 -1
  528. package/dist/src/core/chat/trace.js.map +0 -1
  529. package/dist/src/core/chat/turn-host.d.ts +0 -25
  530. package/dist/src/core/chat/turn-host.d.ts.map +0 -1
  531. package/dist/src/core/chat/turn-host.js +0 -2
  532. package/dist/src/core/chat/turn-host.js.map +0 -1
  533. package/dist/src/core/tools/delete-file.d.ts.map +0 -1
  534. package/dist/src/core/tools/delete-file.js.map +0 -1
  535. package/dist/src/core/tools/edit-file.d.ts.map +0 -1
  536. package/dist/src/core/tools/edit-file.js.map +0 -1
  537. package/dist/src/core/tools/file-edit-core.d.ts.map +0 -1
  538. package/dist/src/core/tools/file-edit-core.js.map +0 -1
  539. package/dist/src/core/tools/list-files.d.ts.map +0 -1
  540. package/dist/src/core/tools/list-files.js.map +0 -1
  541. package/dist/src/core/tools/memory-checkpoint.d.ts.map +0 -1
  542. package/dist/src/core/tools/memory-checkpoint.js.map +0 -1
  543. package/dist/src/core/tools/memory-notes.d.ts.map +0 -1
  544. package/dist/src/core/tools/memory-notes.js.map +0 -1
  545. package/dist/src/core/tools/move-file.d.ts.map +0 -1
  546. package/dist/src/core/tools/move-file.js.map +0 -1
  547. package/dist/src/core/tools/read-file.d.ts.map +0 -1
  548. package/dist/src/core/tools/read-file.js.map +0 -1
  549. package/dist/src/core/tools/record-knowledge.d.ts.map +0 -1
  550. package/dist/src/core/tools/record-knowledge.js.map +0 -1
  551. package/dist/src/core/tools/report-state.d.ts +0 -3
  552. package/dist/src/core/tools/report-state.d.ts.map +0 -1
  553. package/dist/src/core/tools/report-state.js +0 -63
  554. package/dist/src/core/tools/report-state.js.map +0 -1
  555. package/dist/src/core/tools/run-shell.d.ts.map +0 -1
  556. package/dist/src/core/tools/run-shell.js.map +0 -1
  557. package/dist/src/core/tools/search-files.d.ts.map +0 -1
  558. package/dist/src/core/tools/search-files.js.map +0 -1
  559. package/dist/src/core/tools/update-plan.d.ts.map +0 -1
  560. package/dist/src/core/tools/update-plan.js.map +0 -1
  561. package/dist/src/core/tools/view-image.d.ts.map +0 -1
  562. package/dist/src/core/tools/view-image.js.map +0 -1
  563. package/dist/src/core/tools/web-search.d.ts.map +0 -1
  564. package/dist/src/core/tools/web-search.js.map +0 -1
  565. package/dist/src/web/assets/index-BEeN-RT5.css +0 -2
  566. package/dist/src/web/assets/index-BKDg9H_-.js +0 -56
  567. /package/dist/src/core/chat/{archive.js → engine/sessions/archive.js} +0 -0
  568. /package/dist/src/core/chat/{trace.js → engine/turns/trace.js} +0 -0
  569. /package/dist/src/core/tools/{delete-file.js → toolkits/coding-files/delete-file.js} +0 -0
  570. /package/dist/src/core/tools/{edit-file.js → toolkits/coding-files/edit-file.js} +0 -0
  571. /package/dist/src/core/tools/{file-edit-core.js → toolkits/coding-files/file-edit-core.js} +0 -0
  572. /package/dist/src/core/tools/{list-files.js → toolkits/coding-files/list-files.js} +0 -0
  573. /package/dist/src/core/tools/{move-file.js → toolkits/coding-files/move-file.js} +0 -0
  574. /package/dist/src/core/tools/{read-file.js → toolkits/coding-files/read-file.js} +0 -0
  575. /package/dist/src/core/tools/{search-files.js → toolkits/coding-files/search-files.js} +0 -0
  576. /package/dist/src/core/tools/{update-plan.js → toolkits/internal/update-plan.js} +0 -0
  577. /package/dist/src/core/tools/{memory-checkpoint.js → toolkits/knowledge/memory-checkpoint.js} +0 -0
  578. /package/dist/src/core/tools/{record-knowledge.js → toolkits/knowledge/record-knowledge.js} +0 -0
@@ -0,0 +1 @@
1
+ {"version":3,"file":"report-writer.js","sourceRoot":"","sources":["../../../../src/core/eval/report-writer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,SAAS,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AACnD,OAAO,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAG3C,MAAM,UAAU,oBAAoB,CAAC,MAAuB;IAC1D,SAAS,CAAC,MAAM,CAAC,UAAU,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAClD,MAAM,QAAQ,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,aAAa,CAAC,CAAC;IACxD,MAAM,YAAY,GAAG,IAAI,CAAC,MAAM,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC;IAC1D,aAAa,CAAC,QAAQ,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,MAAM,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACxE,aAAa,CAAC,YAAY,EAAE,uBAAuB,CAAC,MAAM,CAAC,EAAE,MAAM,CAAC,CAAC;IACrE,OAAO,EAAE,QAAQ,EAAE,YAAY,EAAE,CAAC;AACpC,CAAC;AAED,MAAM,UAAU,uBAAuB,CAAC,MAAuB;IAC7D,MAAM,MAAM,GAAG,MAAM,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAC;IACpF,MAAM,KAAK,GAAG;QACZ,4BAA4B;QAC5B,EAAE;QACF,WAAW,MAAM,CAAC,MAAM,EAAE;QAC1B,YAAY,MAAM,CAAC,SAAS,EAAE;QAC9B,aAAa,MAAM,CAAC,UAAU,EAAE;QAChC,YAAY,MAAM,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,SAAS;QACpD,EAAE;QACF,gGAAgG;QAChG,uDAAuD;QACvD,GAAG,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,gBAAgB,CAAC;QACvC,EAAE;KACH,CAAC;IAEF,KAAK,MAAM,MAAM,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;QACpC,KAAK,CAAC,IAAI,CAAC,GAAG,eAAe,CAAC,MAAM,EAAE,MAAM,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC,CAAC;IAChE,CAAC;IAED,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,IAAI,CAAC;AAC3C,CAAC;AAED,SAAS,gBAAgB,CAAC,MAAqB;IAC7C,MAAM,YAAY,GAAG,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IAC1E,OAAO;QACL,MAAM,CAAC,MAAM;QACb,MAAM,CAAC,MAAM;QACb,MAAM,CAAC,KAAK,IAAI,SAAS;QACzB,GAAG,YAAY,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE;QACzC,MAAM,CAAC,OAAO,CAAC,OAAO,IAAI,QAAQ,MAAM,CAAC,KAAK,CAAC,QAAQ,IAAI,SAAS,EAAE;QACtE,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,cAAc,CAAC;QACrC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC;QAChC,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,iCAAiC,CAAC;KACzD,CAAC,GAAG,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;AACtE,CAAC;AAED,SAAS,eAAe,CAAC,MAAqB,EAAE,UAAkB;IAChE,MAAM,UAAU,GAAG,MAAM,CAAC,SAAS,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,UAAU,CAAC,IAAI,EAAE,UAAU,CAAC,CAAC,CAAC;IAC3F,MAAM,KAAK,GAAG;QACZ,MAAM,MAAM,CAAC,MAAM,EAAE;QACrB,EAAE;QACF,mBAAmB;QACnB,eAAe;QACf,cAAc,UAAU,CAAC,MAAM,CAAC,MAAM,CAAC,IAAI;QAC3C,aAAa,UAAU,CAAC,MAAM,CAAC,KAAK,IAAI,SAAS,CAAC,IAAI;QACtD,iBAAiB,UAAU,CAAC,MAAM,CAAC,MAAM,CAAC,QAAQ,IAAI,SAAS,CAAC,CAAC,IAAI;QACrE,kBAAkB,UAAU,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,IAAI,SAAS,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI;QACvH,eAAe,UAAU,CAAC,aAAa,CAAC,MAAM,CAAC,CAAC,IAAI;QACpD,mBAAmB,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC,MAAM;QACjF,gBAAgB,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,EAAE,UAAU,CAAC,CAAC,MAAM;QAC1E,cAAc,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,WAAW,EAAE,UAAU,CAAC,CAAC,MAAM;QACpF,mBAAmB,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,eAAe,EAAE,UAAU,CAAC,CAAC,MAAM;QAC7F,4BAA4B,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,gBAAgB,EAAE,UAAU,CAAC,CAAC,MAAM;QACvG,oBAAoB,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,aAAa,EAAE,UAAU,CAAC,CAAC,MAAM;QAC5F,gBAAgB,MAAM,CAAC,SAAS,CAAC,YAAY,CAAC,CAAC,CAAC,KAAK,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,IAAI;QACvI,uBAAuB,MAAM,CAAC,SAAS,CAAC,kBAAkB,CAAC,CAAC,CAAC,KAAK,UAAU,CAAC,UAAU,CAAC,MAAM,CAAC,SAAS,CAAC,kBAAkB,EAAE,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,IAAI;QAC1J,mBAAmB,UAAU,CAAC,MAAM,CAAC,CAAC,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,IAAI;QACpH,EAAE;QACF,sBAAsB;QACtB,EAAE;KACH,CAAC;IAEF,KAAK,CAAC,IAAI,CAAC,GAAG,mBAAmB,CAAC,MAAM,CAAC,EAAE,EAAE,EAAE,mBAAmB,EAAE,EAAE,CAAC,CAAC;IACxE,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC/C,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvB,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CAAC,2BAA2B,EAAE,6BAA6B,CAAC,CAAC;QACvE,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,SAAS,CAAC,YAAY,EAAE,CAAC;YACjD,KAAK,CAAC,IAAI,CAAC,KAAK,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,MAAM,IAAI,CAAC,SAAS,IAAI,EAAE,MAAM,IAAI,CAAC,SAAS,IAAI,EAAE,IAAI,CAAC,CAAC;QAC9H,CAAC;IACH,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,qBAAqB,EAAE,EAAE,CAAC,CAAC;IAC1C,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC/B,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvB,CAAC;SAAM,CAAC;QACN,KAAK,MAAM,KAAK,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC;YAClC,KAAK,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,IAAI,KAAK,CAAC,IAAI,OAAO,KAAK,CAAC,OAAO,OAAO,KAAK,CAAC,QAAQ,IAAI,SAAS,KAAK,KAAK,CAAC,UAAU,KAAK,CAAC,CAAC;QAChJ,CAAC;IACH,CAAC;IAED,KAAK,CAAC,IAAI,CACR,EAAE,EACF,aAAa,EACb,EAAE,EACF,oBAAoB,EACpB,gBAAgB,EAChB,uBAAuB,MAAM,CAAC,OAAO,CAAC,cAAc,IAAI,EACxD,kBAAkB,MAAM,CAAC,OAAO,CAAC,SAAS,IAAI,EAC9C,iBAAiB,MAAM,CAAC,OAAO,CAAC,SAAS,IAAI,EAC7C,mCAAmC,MAAM,CAAC,OAAO,CAAC,iCAAiC,IAAI,EACvF,2BAA2B,MAAM,CAAC,OAAO,CAAC,kBAAkB,IAAI,EAChE,0BAA0B,MAAM,CAAC,OAAO,CAAC,iBAAiB,IAAI,EAC9D,mBAAmB,MAAM,CAAC,OAAO,CAAC,UAAU,IAAI,CACjD,CAAC;IAEF,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,iCAAiC,EAAE,EAAE,CAAC,CAAC;IACtD,IAAI,MAAM,CAAC,OAAO,CAAC,0BAA0B,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3D,KAAK,CAAC,IAAI,CAAC,sCAAsC,CAAC,CAAC;IACrD,CAAC;SAAM,CAAC;QACN,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,OAAO,CAAC,0BAA0B,EAAE,CAAC;YAChE,KAAK,CAAC,IAAI,CAAC,OAAO,UAAU,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QAC7C,CAAC;IACH,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,YAAY,EAAE,EAAE,CAAC,CAAC;IACjC,IAAI,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,MAAM,KAAK,CAAC,IAAI,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC7F,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;IACvB,CAAC;SAAM,CAAC;QACN,KAAK,MAAM,OAAO,IAAI,MAAM,CAAC,MAAM,CAAC,gBAAgB,EAAE,CAAC;YACrD,KAAK,CAAC,IAAI,CAAC,SAAS,OAAO,EAAE,CAAC,CAAC;QACjC,CAAC;QACD,KAAK,MAAM,QAAQ,IAAI,MAAM,CAAC,MAAM,CAAC,cAAc,EAAE,CAAC;YACpD,KAAK,CAAC,IAAI,CAAC,SAAS,QAAQ,EAAE,CAAC,CAAC;QAClC,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,OAAO,CAAC,OAAO,EAAE,CAAC;QAC3B,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,mBAAmB,EAAE,EAAE,EAAE,MAAM,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAClE,CAAC;IAED,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,mBAAmB,CAAC,MAAqB;IAChD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,IAAI,MAAM,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC;QAC5B,KAAK,CAAC,IAAI,CAAC,cAAc,MAAM,CAAC,MAAM,CAAC,SAAS,EAAE,CAAC,CAAC;IACtD,CAAC;IACD,IAAI,MAAM,CAAC,MAAM,CAAC,MAAM,EAAE,CAAC;QACzB,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,MAAM,CAAC,MAAM,CAAC,MAAM,CAAC,CAAC;IACvC,CAAC;IACD,KAAK,CAAC,IAAI,CAAC,EAAE,EAAE,0BAA0B,EAAE,eAAe,CAAC,CAAC;IAC5D,KAAK,CAAC,IAAI,CAAC,yBAAyB,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC;IACxF,KAAK,CAAC,IAAI,CAAC,qBAAqB,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC;IAChF,KAAK,CAAC,IAAI,CAAC,oBAAoB,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAC7E,KAAK,CAAC,IAAI,CAAC,uBAAuB,cAAc,CAAC,MAAM,CAAC,MAAM,CAAC,cAAc,CAAC,IAAI,CAAC,CAAC;IACpF,OAAO,KAAK,CAAC;AACf,CAAC;AAED,SAAS,cAAc,CAAC,KAAe;IACrC,OAAO,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,KAAK,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC;AAC3F,CAAC;AAED,SAAS,aAAa,CAAC,MAAqB;IAC1C,IAAI,MAAM,CAAC,OAAO,CAAC,IAAI,KAAK,cAAc,EAAE,CAAC;QAC3C,OAAO;YACL,cAAc;YACd,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,MAAM,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,SAAS;YAC5D,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,UAAU,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,EAAE,CAAC,CAAC,CAAC,SAAS;YACzF,MAAM,CAAC,OAAO,CAAC,cAAc,IAAI,MAAM,CAAC,OAAO,CAAC,cAAc,KAAK,MAAM,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC;gBAC7F,YAAY,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,cAAc,CAAC,EAAE;gBACvD,CAAC,CAAC,SAAS;SACZ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC/B,CAAC;IACD,OAAO,MAAM,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC,CAAC,oBAAoB,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,cAAc,CAAC,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC;AAClH,CAAC;AAED,SAAS,QAAQ,CAAC,KAAa;IAC7B,OAAO,KAAK,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;AAC5B,CAAC;AAED,SAAS,UAAU,CAAC,KAAa;IAC/B,OAAO,KAAK,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,UAAU,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;AAC5D,CAAC;AAED,SAAS,UAAU,CAAC,IAAY,EAAE,QAAgB;IAChD,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;QACtB,OAAO,GAAG,CAAC;IACb,CAAC;IAED,MAAM,YAAY,GAAG,QAAQ,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;IAC9C,OAAO,YAAY,IAAI,CAAC,YAAY,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAC;AAC9E,CAAC"}
@@ -0,0 +1,206 @@
1
+ import { z } from 'zod';
2
+ export declare const evalCheckSchema: z.ZodObject<{
3
+ name: z.ZodOptional<z.ZodString>;
4
+ command: z.ZodString;
5
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
6
+ }, z.core.$strip>;
7
+ export declare const evalSetupSchema: z.ZodObject<{
8
+ copyFiles: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
9
+ files: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
10
+ commands: z.ZodOptional<z.ZodArray<z.ZodObject<{
11
+ name: z.ZodOptional<z.ZodString>;
12
+ command: z.ZodString;
13
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
14
+ }, z.core.$strip>>>;
15
+ commitMessage: z.ZodOptional<z.ZodString>;
16
+ }, z.core.$strip>;
17
+ export declare const evalInlineFixtureSchema: z.ZodObject<{
18
+ type: z.ZodLiteral<"inline">;
19
+ }, z.core.$strip>;
20
+ export declare const evalGitWorktreeFixtureSchema: z.ZodObject<{
21
+ type: z.ZodLiteral<"git-worktree">;
22
+ repo: z.ZodDefault<z.ZodString>;
23
+ ref: z.ZodString;
24
+ }, z.core.$strip>;
25
+ export declare const evalFixtureSchema: z.ZodDiscriminatedUnion<[z.ZodObject<{
26
+ type: z.ZodLiteral<"inline">;
27
+ }, z.core.$strip>, z.ZodObject<{
28
+ type: z.ZodLiteral<"git-worktree">;
29
+ repo: z.ZodDefault<z.ZodString>;
30
+ ref: z.ZodString;
31
+ }, z.core.$strip>], "type">;
32
+ export declare const evalMilestoneReviewSchema: z.ZodObject<{
33
+ milestone: z.ZodOptional<z.ZodString>;
34
+ intent: z.ZodOptional<z.ZodString>;
35
+ requiredOutcomes: z.ZodDefault<z.ZodArray<z.ZodString>>;
36
+ allowedScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
37
+ outOfScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
38
+ humanQuestions: z.ZodDefault<z.ZodArray<z.ZodString>>;
39
+ }, z.core.$strip>;
40
+ export declare const agentEvalCaseSchema: z.ZodObject<{
41
+ id: z.ZodString;
42
+ kind: z.ZodLiteral<"coding">;
43
+ description: z.ZodOptional<z.ZodString>;
44
+ prompt: z.ZodString;
45
+ model: z.ZodOptional<z.ZodString>;
46
+ maxSteps: z.ZodOptional<z.ZodNumber>;
47
+ setup: z.ZodDefault<z.ZodObject<{
48
+ copyFiles: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
49
+ files: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
50
+ commands: z.ZodOptional<z.ZodArray<z.ZodObject<{
51
+ name: z.ZodOptional<z.ZodString>;
52
+ command: z.ZodString;
53
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
54
+ }, z.core.$strip>>>;
55
+ commitMessage: z.ZodOptional<z.ZodString>;
56
+ }, z.core.$strip>>;
57
+ fixture: z.ZodDefault<z.ZodDiscriminatedUnion<[z.ZodObject<{
58
+ type: z.ZodLiteral<"inline">;
59
+ }, z.core.$strip>, z.ZodObject<{
60
+ type: z.ZodLiteral<"git-worktree">;
61
+ repo: z.ZodDefault<z.ZodString>;
62
+ ref: z.ZodString;
63
+ }, z.core.$strip>], "type">>;
64
+ review: z.ZodDefault<z.ZodObject<{
65
+ milestone: z.ZodOptional<z.ZodString>;
66
+ intent: z.ZodOptional<z.ZodString>;
67
+ requiredOutcomes: z.ZodDefault<z.ZodArray<z.ZodString>>;
68
+ allowedScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
69
+ outOfScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
70
+ humanQuestions: z.ZodDefault<z.ZodArray<z.ZodString>>;
71
+ }, z.core.$strip>>;
72
+ checks: z.ZodDefault<z.ZodArray<z.ZodObject<{
73
+ name: z.ZodOptional<z.ZodString>;
74
+ command: z.ZodString;
75
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
76
+ }, z.core.$strip>>>;
77
+ rubric: z.ZodDefault<z.ZodArray<z.ZodString>>;
78
+ tags: z.ZodDefault<z.ZodArray<z.ZodString>>;
79
+ }, z.core.$strip>;
80
+ export declare const evalCaseSchema: z.ZodObject<{
81
+ id: z.ZodString;
82
+ kind: z.ZodLiteral<"coding">;
83
+ description: z.ZodOptional<z.ZodString>;
84
+ prompt: z.ZodString;
85
+ model: z.ZodOptional<z.ZodString>;
86
+ maxSteps: z.ZodOptional<z.ZodNumber>;
87
+ setup: z.ZodDefault<z.ZodObject<{
88
+ copyFiles: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
89
+ files: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
90
+ commands: z.ZodOptional<z.ZodArray<z.ZodObject<{
91
+ name: z.ZodOptional<z.ZodString>;
92
+ command: z.ZodString;
93
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
94
+ }, z.core.$strip>>>;
95
+ commitMessage: z.ZodOptional<z.ZodString>;
96
+ }, z.core.$strip>>;
97
+ fixture: z.ZodDefault<z.ZodDiscriminatedUnion<[z.ZodObject<{
98
+ type: z.ZodLiteral<"inline">;
99
+ }, z.core.$strip>, z.ZodObject<{
100
+ type: z.ZodLiteral<"git-worktree">;
101
+ repo: z.ZodDefault<z.ZodString>;
102
+ ref: z.ZodString;
103
+ }, z.core.$strip>], "type">>;
104
+ review: z.ZodDefault<z.ZodObject<{
105
+ milestone: z.ZodOptional<z.ZodString>;
106
+ intent: z.ZodOptional<z.ZodString>;
107
+ requiredOutcomes: z.ZodDefault<z.ZodArray<z.ZodString>>;
108
+ allowedScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
109
+ outOfScope: z.ZodDefault<z.ZodArray<z.ZodString>>;
110
+ humanQuestions: z.ZodDefault<z.ZodArray<z.ZodString>>;
111
+ }, z.core.$strip>>;
112
+ checks: z.ZodDefault<z.ZodArray<z.ZodObject<{
113
+ name: z.ZodOptional<z.ZodString>;
114
+ command: z.ZodString;
115
+ timeoutMs: z.ZodOptional<z.ZodNumber>;
116
+ }, z.core.$strip>>>;
117
+ rubric: z.ZodDefault<z.ZodArray<z.ZodString>>;
118
+ tags: z.ZodDefault<z.ZodArray<z.ZodString>>;
119
+ }, z.core.$strip>;
120
+ export type EvalCheck = z.infer<typeof evalCheckSchema>;
121
+ export type EvalSetup = z.infer<typeof evalSetupSchema>;
122
+ export type EvalFixture = z.infer<typeof evalFixtureSchema>;
123
+ export type EvalMilestoneReview = z.infer<typeof evalMilestoneReviewSchema>;
124
+ export type AgentEvalCase = z.infer<typeof agentEvalCaseSchema>;
125
+ export type EvalCase = z.infer<typeof evalCaseSchema>;
126
+ export type EvalCheckResult = {
127
+ name: string;
128
+ command: string;
129
+ exitCode: number | null;
130
+ stdout: string;
131
+ stderr: string;
132
+ durationMs: number;
133
+ passed: boolean;
134
+ timedOut: boolean;
135
+ };
136
+ export type EvalChangedFile = {
137
+ path: string;
138
+ status: string;
139
+ additions?: number;
140
+ deletions?: number;
141
+ };
142
+ export type EvalTraceMetrics = {
143
+ assistantTurns: number;
144
+ toolCalls: number;
145
+ toolResults: number;
146
+ mutations: number;
147
+ approvalsRequested: number;
148
+ approvalsResolved: number;
149
+ toolErrors: number;
150
+ verificationCommandsAfterMutation: number;
151
+ firstMutationStep?: number;
152
+ outcome?: string;
153
+ summary?: string;
154
+ toolsByName: Record<string, number>;
155
+ readOrSearchBeforeMutation: string[];
156
+ verificationCommandDetails: string[];
157
+ };
158
+ export type EvalRunResult = {
159
+ caseId: string;
160
+ target: string;
161
+ status: 'passed' | 'failed';
162
+ workspaceRoot: string;
163
+ outputDir: string;
164
+ fixture: {
165
+ type: EvalFixture['type'];
166
+ repo?: string;
167
+ ref?: string;
168
+ resolvedRef?: string;
169
+ baselineCommit?: string;
170
+ };
171
+ startedAt: string;
172
+ finishedAt: string;
173
+ durationMs: number;
174
+ agent: {
175
+ command: string[];
176
+ exitCode: number | null;
177
+ stdoutPath: string;
178
+ stderrPath: string;
179
+ timedOut: boolean;
180
+ };
181
+ artifacts: {
182
+ gitStatusPath: string;
183
+ gitDiffPath: string;
184
+ gitDiffStatPath: string;
185
+ changedFilesPath: string;
186
+ progressPath?: string;
187
+ sessionCatalogPath?: string;
188
+ traceFiles: string[];
189
+ changedFiles: EvalChangedFile[];
190
+ };
191
+ checks: EvalCheckResult[];
192
+ metrics: EvalTraceMetrics;
193
+ review: EvalMilestoneReview;
194
+ model?: string;
195
+ maxSteps?: number;
196
+ };
197
+ export type EvalSuiteReport = {
198
+ version: 1;
199
+ target: string;
200
+ repoRoot: string;
201
+ startedAt: string;
202
+ finishedAt: string;
203
+ resultsDir: string;
204
+ results: EvalRunResult[];
205
+ };
206
+ //# sourceMappingURL=schema.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../../src/core/eval/schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,eAAO,MAAM,eAAe;;;;iBASoG,CAAC;AAEjI,eAAO,MAAM,eAAe;;;;;;;;;iBAmB8E,CAAC;AAE3G,eAAO,MAAM,uBAAuB;;iBAG+C,CAAC;AAEpF,eAAO,MAAM,4BAA4B;;;;iBAQ8C,CAAC;AAExF,eAAO,MAAM,iBAAiB;;;;;;2BAGyC,CAAC;AAExE,eAAO,MAAM,yBAAyB;;;;;;;iBA2BoE,CAAC;AAE3G,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBA2CmF,CAAC;AAEpH,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;iBAAsB,CAAC;AAElD,MAAM,MAAM,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AACxD,MAAM,MAAM,SAAS,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,eAAe,CAAC,CAAC;AACxD,MAAM,MAAM,WAAW,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,iBAAiB,CAAC,CAAC;AAC5D,MAAM,MAAM,mBAAmB,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,yBAAyB,CAAC,CAAC;AAC5E,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AAEtD,MAAM,MAAM,eAAe,GAAG;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,MAAM,CAAC;IAChB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,MAAM,EAAE,OAAO,CAAC;IAChB,QAAQ,EAAE,OAAO,CAAC;CACnB,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB,CAAC;AAEF,MAAM,MAAM,gBAAgB,GAAG;IAC7B,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,iBAAiB,EAAE,MAAM,CAAC;IAC1B,UAAU,EAAE,MAAM,CAAC;IACnB,iCAAiC,EAAE,MAAM,CAAC;IAC1C,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACpC,0BAA0B,EAAE,MAAM,EAAE,CAAC;IACrC,0BAA0B,EAAE,MAAM,EAAE,CAAC;CACtC,CAAC;AAEF,MAAM,MAAM,aAAa,GAAG;IAC1B,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,QAAQ,GAAG,QAAQ,CAAC;IAC5B,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE;QACP,IAAI,EAAE,WAAW,CAAC,MAAM,CAAC,CAAC;QAC1B,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,GAAG,CAAC,EAAE,MAAM,CAAC;QACb,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,cAAc,CAAC,EAAE,MAAM,CAAC;KACzB,CAAC;IACF,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,KAAK,EAAE;QACL,OAAO,EAAE,MAAM,EAAE,CAAC;QAClB,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAC;QACxB,UAAU,EAAE,MAAM,CAAC;QACnB,UAAU,EAAE,MAAM,CAAC;QACnB,QAAQ,EAAE,OAAO,CAAC;KACnB,CAAC;IACF,SAAS,EAAE;QACT,aAAa,EAAE,MAAM,CAAC;QACtB,WAAW,EAAE,MAAM,CAAC;QACpB,eAAe,EAAE,MAAM,CAAC;QACxB,gBAAgB,EAAE,MAAM,CAAC;QACzB,YAAY,CAAC,EAAE,MAAM,CAAC;QACtB,kBAAkB,CAAC,EAAE,MAAM,CAAC;QAC5B,UAAU,EAAE,MAAM,EAAE,CAAC;QACrB,YAAY,EAAE,eAAe,EAAE,CAAC;KACjC,CAAC;IACF,MAAM,EAAE,eAAe,EAAE,CAAC;IAC1B,OAAO,EAAE,gBAAgB,CAAC;IAC1B,MAAM,EAAE,mBAAmB,CAAC;IAC5B,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB,CAAC;AAEF,MAAM,MAAM,eAAe,GAAG;IAC5B,OAAO,EAAE,CAAC,CAAC;IACX,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,EAAE,MAAM,CAAC;IACjB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,aAAa,EAAE,CAAC;CAC1B,CAAC"}
@@ -0,0 +1,104 @@
1
+ import { z } from 'zod';
2
+ export const evalCheckSchema = z.object({
3
+ name: z.string().trim().min(1)
4
+ .describe('Human-readable label for this check in reports. Defaults to the command when omitted.')
5
+ .optional(),
6
+ command: z.string().trim().min(1)
7
+ .describe('Shell command to run inside the disposable eval workspace after the agent finishes. Exit code 0 means the check passed.'),
8
+ timeoutMs: z.number().int().positive()
9
+ .describe('Optional timeout for this check command in milliseconds.')
10
+ .optional(),
11
+ }).describe('A deterministic post-run command used to decide whether the agent-produced workspace passes objective validation.');
12
+ export const evalSetupSchema = z.object({
13
+ copyFiles: z.record(z.string().trim().min(1).describe('Repository-relative source file path to copy from the eval harness repo.'), z.string().trim().min(1).describe('Workspace-relative destination file path to create before the eval starts.'))
14
+ .describe('Fixture files to copy from the eval harness repo into the disposable workspace before committing the initial Git baseline.')
15
+ .optional(),
16
+ files: z.record(z.string().trim().min(1).describe('Workspace-relative file path to create before the eval starts.'), z.string().describe('Exact UTF-8 file contents to write for the fixture file.'))
17
+ .describe('Fixture files to write into the disposable workspace before committing the initial Git baseline.')
18
+ .optional(),
19
+ commands: z.array(evalCheckSchema)
20
+ .describe('Setup commands to run after fixture files are written and before the initial Git baseline is committed.')
21
+ .optional(),
22
+ commitMessage: z.string().trim().min(1)
23
+ .describe('Commit message for the initial fixture Git baseline. Defaults to a generated eval fixture message.')
24
+ .optional(),
25
+ }).describe('Instructions for creating the disposable repository state that the agent will work against.');
26
+ export const evalInlineFixtureSchema = z.object({
27
+ type: z.literal('inline')
28
+ .describe('Create a small disposable Git repository from the case setup files and commands.'),
29
+ }).describe('A synthetic disposable repository built from inline eval setup data.');
30
+ export const evalGitWorktreeFixtureSchema = z.object({
31
+ type: z.literal('git-worktree')
32
+ .describe('Create a disposable Git worktree from an existing repository at a pinned ref.'),
33
+ repo: z.string().trim().min(1)
34
+ .describe('Repository path to create the worktree from. Relative paths are resolved from the Heddle repo root.')
35
+ .default('.'),
36
+ ref: z.string().trim().min(1)
37
+ .describe('Pinned target ref for the worktree, such as a release tag or commit SHA. Avoid moving HEAD for comparable evals.'),
38
+ }).describe('A realistic disposable repository fixture created from a pinned Git ref.');
39
+ export const evalFixtureSchema = z.discriminatedUnion('type', [
40
+ evalInlineFixtureSchema,
41
+ evalGitWorktreeFixtureSchema,
42
+ ]).describe('How to prepare the disposable workspace the agent edits.');
43
+ export const evalMilestoneReviewSchema = z.object({
44
+ milestone: z.string().trim().min(1)
45
+ .describe('Short name for the user-intended milestone this case evaluates.')
46
+ .optional(),
47
+ intent: z.string().trim().min(1)
48
+ .describe('Human-readable statement of what the agent should accomplish beyond merely passing checks.')
49
+ .optional(),
50
+ requiredOutcomes: z.array(z.string().trim().min(1).describe('Observable outcome a human reviewer should look for in the final diff, trace, or answer.'))
51
+ .describe('Milestone outcomes expected for a high-quality completion.')
52
+ .default([]),
53
+ allowedScope: z.array(z.string().trim().min(1).describe('Files, modules, or behavior areas the agent is allowed or expected to touch.'))
54
+ .describe('Expected implementation scope for judging whether the diff stayed on task.')
55
+ .default([]),
56
+ outOfScope: z.array(z.string().trim().min(1).describe('Files, modules, or behavior areas that should not be changed for this case.'))
57
+ .describe('Boundaries a human reviewer should use to spot unrelated churn.')
58
+ .default([]),
59
+ humanQuestions: z.array(z.string().trim().min(1).describe('Question for human review after the run completes.'))
60
+ .describe('Review prompts that help judge task completion quality beyond deterministic checks.')
61
+ .default([]),
62
+ }).describe('Human-review metadata for milestone-style eval cases where pass/fail checks are not enough.');
63
+ export const agentEvalCaseSchema = z.object({
64
+ id: z.string().trim().regex(/^[a-zA-Z0-9._-]+$/, 'Use a filesystem-safe case id.')
65
+ .describe('Stable filesystem-safe case id used in result paths, filtering, and reports.'),
66
+ kind: z.literal('coding')
67
+ .describe('Eval case type. The first harness slice supports coding cases run through ask --new-session.'),
68
+ description: z.string().trim()
69
+ .describe('Optional short explanation of what behavior this case is meant to exercise.')
70
+ .optional(),
71
+ prompt: z.string().trim().min(1)
72
+ .describe('User prompt sent to Heddle in the disposable workspace. This should ask for real coding work, not just Q&A.'),
73
+ model: z.string().trim().min(1)
74
+ .describe('Optional model override for this case. The CLI-level --model takes precedence when supplied.')
75
+ .optional(),
76
+ maxSteps: z.number().int().positive()
77
+ .describe('Optional maximum agent loop steps for this case. The CLI-level --max-steps takes precedence when supplied.')
78
+ .optional(),
79
+ setup: evalSetupSchema
80
+ .describe('Disposable workspace setup for this case.')
81
+ .default({}),
82
+ fixture: evalFixtureSchema
83
+ .describe('Workspace fixture source. Defaults to an inline synthetic repository.')
84
+ .default({ type: 'inline' }),
85
+ review: evalMilestoneReviewSchema
86
+ .describe('Optional milestone-completion review guidance included in reports.')
87
+ .default({
88
+ requiredOutcomes: [],
89
+ allowedScope: [],
90
+ outOfScope: [],
91
+ humanQuestions: [],
92
+ }),
93
+ checks: z.array(evalCheckSchema)
94
+ .describe('Deterministic post-agent commands that must pass for the case to be marked passed.')
95
+ .default([]),
96
+ rubric: z.array(z.string().trim().min(1).describe('Qualitative behavior criterion for human or future LLM judging.'))
97
+ .describe('Non-deterministic quality criteria preserved in reports for manual or future judge review.')
98
+ .default([]),
99
+ tags: z.array(z.string().trim().min(1).describe('Free-form label for filtering or grouping eval cases.'))
100
+ .describe('Case labels such as bugfix, refactor, verification, multi-file, or tui.')
101
+ .default([]),
102
+ }).describe('A live coding-task eval case run in a disposable Git workspace through Heddle ask/session execution.');
103
+ export const evalCaseSchema = agentEvalCaseSchema;
104
+ //# sourceMappingURL=schema.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"schema.js","sourceRoot":"","sources":["../../../../src/core/eval/schema.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC3B,QAAQ,CAAC,uFAAuF,CAAC;SACjG,QAAQ,EAAE;IACb,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC9B,QAAQ,CAAC,yHAAyH,CAAC;IACtI,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE;SACnC,QAAQ,CAAC,0DAA0D,CAAC;SACpE,QAAQ,EAAE;CACd,CAAC,CAAC,QAAQ,CAAC,mHAAmH,CAAC,CAAC;AAEjI,MAAM,CAAC,MAAM,eAAe,GAAG,CAAC,CAAC,MAAM,CAAC;IACtC,SAAS,EAAE,CAAC,CAAC,MAAM,CACjB,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,0EAA0E,CAAC,EAC7G,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,4EAA4E,CAAC,CAChH;SACE,QAAQ,CAAC,4HAA4H,CAAC;SACtI,QAAQ,EAAE;IACb,KAAK,EAAE,CAAC,CAAC,MAAM,CACb,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,gEAAgE,CAAC,EACnG,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,0DAA0D,CAAC,CAChF;SACE,QAAQ,CAAC,kGAAkG,CAAC;SAC5G,QAAQ,EAAE;IACb,QAAQ,EAAE,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC;SAC/B,QAAQ,CAAC,yGAAyG,CAAC;SACnH,QAAQ,EAAE;IACb,aAAa,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SACpC,QAAQ,CAAC,oGAAoG,CAAC;SAC9G,QAAQ,EAAE;CACd,CAAC,CAAC,QAAQ,CAAC,6FAA6F,CAAC,CAAC;AAE3G,MAAM,CAAC,MAAM,uBAAuB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC9C,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;SACtB,QAAQ,CAAC,kFAAkF,CAAC;CAChG,CAAC,CAAC,QAAQ,CAAC,sEAAsE,CAAC,CAAC;AAEpF,MAAM,CAAC,MAAM,4BAA4B,GAAG,CAAC,CAAC,MAAM,CAAC;IACnD,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,cAAc,CAAC;SAC5B,QAAQ,CAAC,+EAA+E,CAAC;IAC5F,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC3B,QAAQ,CAAC,qGAAqG,CAAC;SAC/G,OAAO,CAAC,GAAG,CAAC;IACf,GAAG,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC1B,QAAQ,CAAC,kHAAkH,CAAC;CAChI,CAAC,CAAC,QAAQ,CAAC,0EAA0E,CAAC,CAAC;AAExF,MAAM,CAAC,MAAM,iBAAiB,GAAG,CAAC,CAAC,kBAAkB,CAAC,MAAM,EAAE;IAC5D,uBAAuB;IACvB,4BAA4B;CAC7B,CAAC,CAAC,QAAQ,CAAC,0DAA0D,CAAC,CAAC;AAExE,MAAM,CAAC,MAAM,yBAAyB,GAAG,CAAC,CAAC,MAAM,CAAC;IAChD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAChC,QAAQ,CAAC,iEAAiE,CAAC;SAC3E,QAAQ,EAAE;IACb,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC7B,QAAQ,CAAC,4FAA4F,CAAC;SACtG,QAAQ,EAAE;IACb,gBAAgB,EAAE,CAAC,CAAC,KAAK,CACvB,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,0FAA0F,CAAC,CAC9H;SACE,QAAQ,CAAC,4DAA4D,CAAC;SACtE,OAAO,CAAC,EAAE,CAAC;IACd,YAAY,EAAE,CAAC,CAAC,KAAK,CACnB,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,8EAA8E,CAAC,CAClH;SACE,QAAQ,CAAC,4EAA4E,CAAC;SACtF,OAAO,CAAC,EAAE,CAAC;IACd,UAAU,EAAE,CAAC,CAAC,KAAK,CACjB,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,6EAA6E,CAAC,CACjH;SACE,QAAQ,CAAC,iEAAiE,CAAC;SAC3E,OAAO,CAAC,EAAE,CAAC;IACd,cAAc,EAAE,CAAC,CAAC,KAAK,CACrB,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,oDAAoD,CAAC,CACxF;SACE,QAAQ,CAAC,qFAAqF,CAAC;SAC/F,OAAO,CAAC,EAAE,CAAC;CACf,CAAC,CAAC,QAAQ,CAAC,6FAA6F,CAAC,CAAC;AAE3G,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAC,MAAM,CAAC;IAC1C,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,mBAAmB,EAAE,gCAAgC,CAAC;SAC/E,QAAQ,CAAC,8EAA8E,CAAC;IAC3F,IAAI,EAAE,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;SACtB,QAAQ,CAAC,8FAA8F,CAAC;IAC3G,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE;SAC3B,QAAQ,CAAC,6EAA6E,CAAC;SACvF,QAAQ,EAAE;IACb,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC7B,QAAQ,CAAC,6GAA6G,CAAC;IAC1H,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;SAC5B,QAAQ,CAAC,8FAA8F,CAAC;SACxG,QAAQ,EAAE;IACb,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE;SAClC,QAAQ,CAAC,4GAA4G,CAAC;SACtH,QAAQ,EAAE;IACb,KAAK,EAAE,eAAe;SACnB,QAAQ,CAAC,2CAA2C,CAAC;SACrD,OAAO,CAAC,EAAE,CAAC;IACd,OAAO,EAAE,iBAAiB;SACvB,QAAQ,CAAC,uEAAuE,CAAC;SACjF,OAAO,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,CAAC;IAC9B,MAAM,EAAE,yBAAyB;SAC9B,QAAQ,CAAC,oEAAoE,CAAC;SAC9E,OAAO,CAAC;QACP,gBAAgB,EAAE,EAAE;QACpB,YAAY,EAAE,EAAE;QAChB,UAAU,EAAE,EAAE;QACd,cAAc,EAAE,EAAE;KACnB,CAAC;IACJ,MAAM,EAAE,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC;SAC7B,QAAQ,CAAC,oFAAoF,CAAC;SAC9F,OAAO,CAAC,EAAE,CAAC;IACd,MAAM,EAAE,CAAC,CAAC,KAAK,CACb,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,iEAAiE,CAAC,CACrG;SACE,QAAQ,CAAC,4FAA4F,CAAC;SACtG,OAAO,CAAC,EAAE,CAAC;IACd,IAAI,EAAE,CAAC,CAAC,KAAK,CACX,CAAC,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,uDAAuD,CAAC,CAC3F;SACE,QAAQ,CAAC,yEAAyE,CAAC;SACnF,OAAO,CAAC,EAAE,CAAC;CACf,CAAC,CAAC,QAAQ,CAAC,sGAAsG,CAAC,CAAC;AAEpH,MAAM,CAAC,MAAM,cAAc,GAAG,mBAAmB,CAAC"}
@@ -0,0 +1,6 @@
1
+ import type { TraceEvent } from '../types.js';
2
+ import type { EvalTraceMetrics } from './schema.js';
3
+ export declare function analyzeTraceFiles(paths: string[]): EvalTraceMetrics;
4
+ export declare function analyzeTrace(trace: TraceEvent[]): EvalTraceMetrics;
5
+ export declare function readTraceFile(path: string): TraceEvent[];
6
+ //# sourceMappingURL=trace-analyzer.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"trace-analyzer.d.ts","sourceRoot":"","sources":["../../../../src/core/eval/trace-analyzer.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AAC9C,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,aAAa,CAAC;AAKpD,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,gBAAgB,CAEnE;AAED,wBAAgB,YAAY,CAAC,KAAK,EAAE,UAAU,EAAE,GAAG,gBAAgB,CAiFlE;AAED,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,UAAU,EAAE,CAMxD"}
@@ -0,0 +1,106 @@
1
+ import { existsSync, readFileSync } from 'node:fs';
2
+ const MUTATION_TOOLS = new Set(['edit_file', 'delete_file', 'move_file', 'run_shell_mutate']);
3
+ const READ_OR_SEARCH_TOOLS = new Set(['read_file', 'list_files', 'search_files', 'run_shell_inspect']);
4
+ export function analyzeTraceFiles(paths) {
5
+ return analyzeTrace(paths.flatMap(readTraceFile));
6
+ }
7
+ export function analyzeTrace(trace) {
8
+ const toolsByName = {};
9
+ const readOrSearchBeforeMutation = [];
10
+ let assistantTurns = 0;
11
+ let toolCalls = 0;
12
+ let toolResults = 0;
13
+ let mutations = 0;
14
+ let approvalsRequested = 0;
15
+ let approvalsResolved = 0;
16
+ let toolErrors = 0;
17
+ let verificationCommandsAfterMutation = 0;
18
+ const verificationCommandDetails = [];
19
+ let firstMutationStep;
20
+ let outcome;
21
+ let summary;
22
+ for (const event of trace) {
23
+ if (event.type === 'assistant.turn') {
24
+ assistantTurns++;
25
+ for (const call of event.toolCalls ?? []) {
26
+ toolCalls++;
27
+ toolsByName[call.tool] = (toolsByName[call.tool] ?? 0) + 1;
28
+ if (READ_OR_SEARCH_TOOLS.has(call.tool) && firstMutationStep === undefined) {
29
+ readOrSearchBeforeMutation.push(summarizeToolInput(call.tool, call.input));
30
+ }
31
+ if (MUTATION_TOOLS.has(call.tool)) {
32
+ mutations++;
33
+ firstMutationStep ??= event.step;
34
+ }
35
+ if (firstMutationStep !== undefined
36
+ && (call.tool === 'run_shell_mutate' || call.tool === 'run_shell_inspect')
37
+ && isVerificationCommand(call.input)) {
38
+ verificationCommandsAfterMutation++;
39
+ verificationCommandDetails.push(summarizeToolInput(call.tool, call.input));
40
+ }
41
+ }
42
+ continue;
43
+ }
44
+ if (event.type === 'tool.result') {
45
+ toolResults++;
46
+ if (!event.result.ok) {
47
+ toolErrors++;
48
+ }
49
+ }
50
+ if (event.type === 'tool.approval_requested') {
51
+ approvalsRequested++;
52
+ }
53
+ if (event.type === 'tool.approval_resolved') {
54
+ approvalsResolved++;
55
+ }
56
+ if (event.type === 'run.finished') {
57
+ outcome = event.outcome;
58
+ summary = event.summary;
59
+ }
60
+ }
61
+ return {
62
+ assistantTurns,
63
+ toolCalls,
64
+ toolResults,
65
+ mutations,
66
+ approvalsRequested,
67
+ approvalsResolved,
68
+ toolErrors,
69
+ verificationCommandsAfterMutation,
70
+ verificationCommandDetails: [...new Set(verificationCommandDetails)],
71
+ firstMutationStep,
72
+ outcome,
73
+ summary,
74
+ toolsByName,
75
+ readOrSearchBeforeMutation: [...new Set(readOrSearchBeforeMutation)],
76
+ };
77
+ }
78
+ export function readTraceFile(path) {
79
+ if (!existsSync(path)) {
80
+ return [];
81
+ }
82
+ const parsed = JSON.parse(readFileSync(path, 'utf8'));
83
+ return Array.isArray(parsed) ? parsed.filter(isTraceEvent) : [];
84
+ }
85
+ function isTraceEvent(value) {
86
+ return Boolean(value && typeof value === 'object' && !Array.isArray(value) && typeof value.type === 'string');
87
+ }
88
+ function summarizeToolInput(tool, input) {
89
+ const object = input && typeof input === 'object' && !Array.isArray(input) ? input : undefined;
90
+ const path = object && typeof object.path === 'string' ? object.path : undefined;
91
+ const query = object && typeof object.query === 'string' ? object.query : undefined;
92
+ const command = object && typeof object.command === 'string' ? object.command : undefined;
93
+ return `${tool}:${path ?? query ?? command ?? ''}`.slice(0, 160);
94
+ }
95
+ function isVerificationCommand(input) {
96
+ if (!input || typeof input !== 'object' || Array.isArray(input)) {
97
+ return false;
98
+ }
99
+ const command = input.command;
100
+ if (typeof command !== 'string') {
101
+ return false;
102
+ }
103
+ return /\b(yarn|npm|pnpm|vitest|jest|mocha|tsc|eslint|cargo|go|pytest|python|ruff)\b/.test(command)
104
+ && /\b(test|build|lint|typecheck|check|vitest|tsc|pytest|ruff)\b/.test(command);
105
+ }
106
+ //# sourceMappingURL=trace-analyzer.js.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"trace-analyzer.js","sourceRoot":"","sources":["../../../../src/core/eval/trace-analyzer.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,SAAS,CAAC;AAInD,MAAM,cAAc,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,aAAa,EAAE,WAAW,EAAE,kBAAkB,CAAC,CAAC,CAAC;AAC9F,MAAM,oBAAoB,GAAG,IAAI,GAAG,CAAC,CAAC,WAAW,EAAE,YAAY,EAAE,cAAc,EAAE,mBAAmB,CAAC,CAAC,CAAC;AAEvG,MAAM,UAAU,iBAAiB,CAAC,KAAe;IAC/C,OAAO,YAAY,CAAC,KAAK,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC,CAAC;AACpD,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,KAAmB;IAC9C,MAAM,WAAW,GAA2B,EAAE,CAAC;IAC/C,MAAM,0BAA0B,GAAa,EAAE,CAAC;IAChD,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,SAAS,GAAG,CAAC,CAAC;IAClB,IAAI,kBAAkB,GAAG,CAAC,CAAC;IAC3B,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,IAAI,UAAU,GAAG,CAAC,CAAC;IACnB,IAAI,iCAAiC,GAAG,CAAC,CAAC;IAC1C,MAAM,0BAA0B,GAAa,EAAE,CAAC;IAChD,IAAI,iBAAqC,CAAC;IAC1C,IAAI,OAA2B,CAAC;IAChC,IAAI,OAA2B,CAAC;IAEhC,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,IAAI,KAAK,CAAC,IAAI,KAAK,gBAAgB,EAAE,CAAC;YACpC,cAAc,EAAE,CAAC;YACjB,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,SAAS,IAAI,EAAE,EAAE,CAAC;gBACzC,SAAS,EAAE,CAAC;gBACZ,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;gBAE3D,IAAI,oBAAoB,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,iBAAiB,KAAK,SAAS,EAAE,CAAC;oBAC3E,0BAA0B,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;gBAC7E,CAAC;gBAED,IAAI,cAAc,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;oBAClC,SAAS,EAAE,CAAC;oBACZ,iBAAiB,KAAK,KAAK,CAAC,IAAI,CAAC;gBACnC,CAAC;gBAED,IACE,iBAAiB,KAAK,SAAS;uBAC5B,CAAC,IAAI,CAAC,IAAI,KAAK,kBAAkB,IAAI,IAAI,CAAC,IAAI,KAAK,mBAAmB,CAAC;uBACvE,qBAAqB,CAAC,IAAI,CAAC,KAAK,CAAC,EACpC,CAAC;oBACD,iCAAiC,EAAE,CAAC;oBACpC,0BAA0B,CAAC,IAAI,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC;gBAC7E,CAAC;YACH,CAAC;YACD,SAAS;QACX,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,aAAa,EAAE,CAAC;YACjC,WAAW,EAAE,CAAC;YACd,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;gBACrB,UAAU,EAAE,CAAC;YACf,CAAC;QACH,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,yBAAyB,EAAE,CAAC;YAC7C,kBAAkB,EAAE,CAAC;QACvB,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,wBAAwB,EAAE,CAAC;YAC5C,iBAAiB,EAAE,CAAC;QACtB,CAAC;QAED,IAAI,KAAK,CAAC,IAAI,KAAK,cAAc,EAAE,CAAC;YAClC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;YACxB,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,OAAO;QACL,cAAc;QACd,SAAS;QACT,WAAW;QACX,SAAS;QACT,kBAAkB;QAClB,iBAAiB;QACjB,UAAU;QACV,iCAAiC;QACjC,0BAA0B,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,0BAA0B,CAAC,CAAC;QACpE,iBAAiB;QACjB,OAAO;QACP,OAAO;QACP,WAAW;QACX,0BAA0B,EAAE,CAAC,GAAG,IAAI,GAAG,CAAC,0BAA0B,CAAC,CAAC;KACrE,CAAC;AACJ,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,IAAY;IACxC,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CAAC;QACtB,OAAO,EAAE,CAAC;IACZ,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAAY,CAAC;IACjE,OAAO,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;AAClE,CAAC;AAED,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO,OAAO,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,IAAI,OAAQ,KAA4B,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC;AACxI,CAAC;AAED,SAAS,kBAAkB,CAAC,IAAY,EAAE,KAAc;IACtD,MAAM,MAAM,GAAG,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,KAAgC,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1H,MAAM,IAAI,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS,CAAC;IACjF,MAAM,KAAK,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,SAAS,CAAC;IACpF,MAAM,OAAO,GAAG,MAAM,IAAI,OAAO,MAAM,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,SAAS,CAAC;IAC1F,OAAO,GAAG,IAAI,IAAI,IAAI,IAAI,KAAK,IAAI,OAAO,IAAI,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC;AACnE,CAAC;AAED,SAAS,qBAAqB,CAAC,KAAc;IAC3C,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC,EAAE,CAAC;QAChE,OAAO,KAAK,CAAC;IACf,CAAC;IACD,MAAM,OAAO,GAAI,KAA+B,CAAC,OAAO,CAAC;IACzD,IAAI,OAAO,OAAO,KAAK,QAAQ,EAAE,CAAC;QAChC,OAAO,KAAK,CAAC;IACf,CAAC;IACD,OAAO,8EAA8E,CAAC,IAAI,CAAC,OAAO,CAAC;WAC9F,8DAA8D,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;AACpF,CAAC"}
@@ -0,0 +1,14 @@
1
+ import type { EvalProgressReporter } from './progress.js';
2
+ import type { AgentEvalCase, EvalCheckResult, EvalRunResult } from './schema.js';
3
+ export type PreparedEvalWorkspace = {
4
+ workspaceRoot: string;
5
+ setupResults: EvalCheckResult[];
6
+ fixture: EvalRunResult['fixture'];
7
+ };
8
+ export declare function prepareEvalWorkspace(args: {
9
+ testCase: AgentEvalCase;
10
+ repoRoot: string;
11
+ workRoot?: string;
12
+ progress?: EvalProgressReporter;
13
+ }): Promise<PreparedEvalWorkspace>;
14
+ //# sourceMappingURL=workspace-fixture.d.ts.map
@@ -0,0 +1 @@
1
+ {"version":3,"file":"workspace-fixture.d.ts","sourceRoot":"","sources":["../../../../src/core/eval/workspace-fixture.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,EAAE,oBAAoB,EAAE,MAAM,eAAe,CAAC;AAC1D,OAAO,KAAK,EAAE,aAAa,EAAE,eAAe,EAAE,aAAa,EAAE,MAAM,aAAa,CAAC;AAEjF,MAAM,MAAM,qBAAqB,GAAG;IAClC,aAAa,EAAE,MAAM,CAAC;IACtB,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,OAAO,EAAE,aAAa,CAAC,SAAS,CAAC,CAAC;CACnC,CAAC;AAEF,wBAAsB,oBAAoB,CAAC,IAAI,EAAE;IAC/C,QAAQ,EAAE,aAAa,CAAC;IACxB,QAAQ,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,oBAAoB,CAAC;CACjC,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA8BjC"}