minimal-agent 0.12.0 → 0.14.0

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 (472) hide show
  1. package/README.md +6 -5
  2. package/package.json +62 -5
  3. package/plugins/computer_use/.claude-plugin/plugin.json +1 -1
  4. package/plugins/cron-runner/src/commands.js +4 -4
  5. package/plugins/cron-runner/src/commands.ts +4 -4
  6. package/plugins/workflow-runner/.claude-plugin/plugin.json +2 -2
  7. package/plugins/workflow-runner/commands/workflow.md +57 -90
  8. package/plugins/workflow-runner/commands/workflows.md +4 -4
  9. package/plugins/workflow-runner/plugin.js +24 -13
  10. package/plugins/workflow-runner/plugin.ts +28 -13
  11. package/plugins/workflow-runner/src/discovery.js +160 -0
  12. package/plugins/workflow-runner/src/discovery.ts +194 -0
  13. package/plugins/workflow-runner/src/goalAnchor.js +53 -0
  14. package/plugins/workflow-runner/src/goalAnchor.ts +59 -0
  15. package/plugins/workflow-runner/src/index.js +204 -185
  16. package/plugins/workflow-runner/src/index.ts +249 -224
  17. package/plugins/workflow-runner/src/inputCoercion.js +40 -0
  18. package/plugins/workflow-runner/src/inputCoercion.ts +43 -0
  19. package/plugins/workflow-runner/src/journal.js +224 -0
  20. package/plugins/workflow-runner/src/journal.ts +274 -0
  21. package/plugins/workflow-runner/src/jsonSchema.js +132 -0
  22. package/plugins/workflow-runner/src/jsonSchema.ts +162 -0
  23. package/plugins/workflow-runner/src/moduleResolution.js +64 -0
  24. package/plugins/workflow-runner/src/moduleResolution.ts +71 -0
  25. package/plugins/workflow-runner/src/resourceInputs.js +220 -0
  26. package/plugins/workflow-runner/src/resourceInputs.ts +300 -0
  27. package/plugins/workflow-runner/src/runner.js +404 -467
  28. package/plugins/workflow-runner/src/runner.ts +491 -535
  29. package/plugins/workflow-runner/src/runtime/config.js +46 -0
  30. package/plugins/workflow-runner/src/runtime/config.ts +69 -0
  31. package/plugins/workflow-runner/src/runtime/context.js +1304 -0
  32. package/plugins/workflow-runner/src/runtime/context.ts +1786 -0
  33. package/plugins/workflow-runner/src/runtime/exec.js +34 -0
  34. package/plugins/workflow-runner/src/runtime/exec.ts +53 -0
  35. package/plugins/workflow-runner/src/runtime/flow.js +8 -0
  36. package/plugins/workflow-runner/src/runtime/flow.ts +12 -0
  37. package/plugins/workflow-runner/src/runtime/fs.js +27 -0
  38. package/plugins/workflow-runner/src/runtime/fs.ts +41 -0
  39. package/plugins/workflow-runner/src/runtime/model.js +6 -0
  40. package/plugins/workflow-runner/src/runtime/model.ts +8 -0
  41. package/plugins/workflow-runner/src/runtime/output.js +7 -0
  42. package/plugins/workflow-runner/src/runtime/output.ts +21 -0
  43. package/plugins/workflow-runner/src/runtime/scheduler.js +141 -0
  44. package/plugins/workflow-runner/src/runtime/scheduler.ts +166 -0
  45. package/plugins/workflow-runner/src/runtime/state.js +16 -0
  46. package/plugins/workflow-runner/src/runtime/state.ts +19 -0
  47. package/plugins/workflow-runner/src/runtime/tools.js +3 -0
  48. package/plugins/workflow-runner/src/runtime/tools.ts +6 -0
  49. package/plugins/workflow-runner/src/sandbox.js +141 -0
  50. package/plugins/workflow-runner/src/sandbox.ts +182 -0
  51. package/plugins/workflow-runner/src/scriptGuard.d.mts +38 -0
  52. package/plugins/workflow-runner/src/scriptGuard.mjs +449 -0
  53. package/plugins/workflow-runner/src/taskIdentity.js +172 -152
  54. package/plugins/workflow-runner/src/taskIdentity.ts +233 -200
  55. package/plugins/workflow-runner/src/toolRuntime.js +29 -8
  56. package/plugins/workflow-runner/src/toolRuntime.ts +36 -11
  57. package/plugins/workflow-runner/src/types.js +7 -105
  58. package/plugins/workflow-runner/src/types.ts +155 -520
  59. package/plugins/workflow-runner/src/util.js +47 -0
  60. package/plugins/workflow-runner/src/util.ts +51 -0
  61. package/plugins/workflow-runner/src/workflowContext.js +20 -0
  62. package/plugins/workflow-runner/src/workflowContext.ts +58 -0
  63. package/plugins/workflow-runner/src/workflowDependencies.js +36 -0
  64. package/plugins/workflow-runner/src/workflowDependencies.ts +59 -0
  65. package/plugins/workflow-runner/src/workflowFingerprint.js +118 -0
  66. package/plugins/workflow-runner/src/workflowFingerprint.ts +122 -0
  67. package/plugins/workflow-runner/src/workflowState.js +542 -469
  68. package/plugins/workflow-runner/src/workflowState.ts +639 -589
  69. package/plugins/workflow-runner/test/api.test.ts +223 -0
  70. package/plugins/workflow-runner/test/apiCapabilities.test.ts +614 -0
  71. package/plugins/workflow-runner/test/apiConcurrencyGate.test.ts +104 -0
  72. package/plugins/workflow-runner/test/apiEcho.test.ts +210 -0
  73. package/plugins/workflow-runner/test/apiReadText.test.ts +178 -0
  74. package/plugins/workflow-runner/test/apiResourceLedger.test.ts +184 -0
  75. package/plugins/workflow-runner/test/childWorkflow.test.ts +304 -0
  76. package/plugins/workflow-runner/test/discovery.test.ts +234 -0
  77. package/plugins/workflow-runner/test/goalAnchor.test.ts +106 -0
  78. package/plugins/workflow-runner/test/goalController.test.ts +317 -0
  79. package/plugins/workflow-runner/test/journalLanes.test.ts +198 -0
  80. package/plugins/workflow-runner/test/listResources.test.ts +106 -0
  81. package/plugins/workflow-runner/test/resourceInputs.test.ts +294 -0
  82. package/plugins/workflow-runner/test/sandbox.test.ts +126 -0
  83. package/plugins/workflow-runner/test/scheduler.test.ts +70 -0
  84. package/plugins/workflow-runner/test/scriptGuard.test.ts +276 -0
  85. package/plugins/workflow-runner/test/skillDocSync.test.ts +78 -0
  86. package/plugins/workflow-runner/test/taskIdentity.test.ts +288 -179
  87. package/plugins/workflow-runner/test/workflowFingerprint.test.ts +106 -0
  88. package/plugins/workflow-runner/test/workflowState.test.ts +429 -0
  89. package/skills/pptx/examples/assets/cover-bg.svg +23 -0
  90. package/skills/pptx/examples/assets/icon-spark.svg +5 -0
  91. package/skills/skill-creator/scripts/run_eval.py +3 -2
  92. package/skills/ui-design/SKILL.md +5 -2
  93. package/skills/ui-design/references/layout-composition.md +1 -1
  94. package/skills/workflow-creator/SKILL.md +344 -0
  95. package/skills/workflow-creator/assets/template/workflow.mjs +48 -0
  96. package/skills/workflow-creator/references/api.md +216 -0
  97. package/skills/workflow-creator/references/migration.md +103 -0
  98. package/skills/workflow-creator/references/patterns.md +779 -0
  99. package/skills/workflow-creator/scripts/list-resources.mjs +205 -0
  100. package/skills/workflow-creator/scripts/scaffold.mjs +148 -0
  101. package/skills/workflow-creator/scripts/validate.mjs +275 -0
  102. package/skills/workflow-creator/scripts/workflow-creator.test.mjs +82 -0
  103. package/src/advisor/advisorCoordinator.js +503 -0
  104. package/src/advisor/advisorCoordinator.ts +657 -0
  105. package/src/cli/configMenu.tsx +0 -1
  106. package/src/cli/print.js +44 -6
  107. package/src/cli/print.ts +73 -4
  108. package/src/cli/streamJson.js +5 -6
  109. package/src/cli/streamJson.ts +8 -4
  110. package/src/cli/wizards/advisorWizard.js +4 -2
  111. package/src/cli/wizards/advisorWizard.tsx +8 -2
  112. package/src/cli/wizards/contextAdvisorWizard.js +5 -4
  113. package/src/cli/wizards/contextAdvisorWizard.tsx +7 -6
  114. package/src/collaboration/taskFlow.js +115 -0
  115. package/src/collaboration/taskFlow.ts +169 -0
  116. package/src/config/configDefaults.js +41 -1
  117. package/src/config/configDefaults.ts +43 -1
  118. package/src/config/configDiagnostics.ts +2 -2
  119. package/src/config/configFile.js +40 -9
  120. package/src/config/configFile.ts +60 -11
  121. package/src/config/configService.js +443 -74
  122. package/src/config/configService.ts +723 -142
  123. package/src/config/numericLimits.js +22 -0
  124. package/src/config/numericLimits.ts +24 -0
  125. package/src/config.js +1 -0
  126. package/src/config.ts +3 -2
  127. package/src/context/budgetPlan.js +1 -1
  128. package/src/context/budgetPlan.ts +1 -1
  129. package/src/context/compact.js +7 -2
  130. package/src/context/compact.ts +9 -2
  131. package/src/context/microCompactLite.js +212 -13
  132. package/src/context/microCompactLite.ts +261 -12
  133. package/src/context/sessionRegistry.js +8 -1
  134. package/src/context/sessionRegistry.ts +18 -5
  135. package/src/context/toolResultArchive.js +72 -0
  136. package/src/context/toolResultArchive.ts +95 -0
  137. package/src/goal/errors.js +8 -0
  138. package/src/goal/errors.ts +18 -0
  139. package/src/goal/goalRuntime.js +344 -0
  140. package/src/goal/goalRuntime.ts +466 -0
  141. package/src/goal/goalStore.js +200 -0
  142. package/src/goal/goalStore.ts +219 -0
  143. package/src/goal/index.js +4 -0
  144. package/src/goal/index.ts +4 -0
  145. package/src/goal/statusMapping.js +71 -0
  146. package/src/goal/statusMapping.ts +95 -0
  147. package/src/goal/types.js +7 -0
  148. package/src/goal/types.ts +179 -0
  149. package/src/goal/workflowController.js +306 -0
  150. package/src/goal/workflowController.ts +360 -0
  151. package/src/harness/evidenceLedger.js +2 -0
  152. package/src/harness/evidenceLedger.ts +1 -0
  153. package/src/harness/stallGuard.js +9 -3
  154. package/src/harness/stallGuard.ts +7 -3
  155. package/src/images/imageResolver.js +161 -0
  156. package/src/images/imageResolver.ts +189 -0
  157. package/src/infra/verification.js +56 -20
  158. package/src/infra/verification.ts +87 -20
  159. package/src/llm/client.js +79 -21
  160. package/src/llm/client.ts +182 -35
  161. package/src/llm/gatewayClient.js +201 -0
  162. package/src/llm/gatewayClient.ts +299 -0
  163. package/src/llm/messageProvenance.js +46 -0
  164. package/src/llm/messageProvenance.ts +58 -0
  165. package/src/llm/modelProvider.js +1 -0
  166. package/src/llm/modelProvider.ts +22 -0
  167. package/src/llm/providers/anthropicFormat.js +108 -1
  168. package/src/llm/providers/anthropicFormat.ts +115 -4
  169. package/src/llm/providers/openaiFormat.js +9 -0
  170. package/src/llm/providers/openaiFormat.ts +12 -3
  171. package/src/loop.js +319 -54
  172. package/src/loop.ts +373 -57
  173. package/src/main.js +4 -1
  174. package/src/main.tsx +4 -1
  175. package/src/managedExecution/index.js +2 -0
  176. package/src/managedExecution/index.ts +2 -0
  177. package/src/managedExecution/registry.js +15 -0
  178. package/src/managedExecution/registry.ts +20 -0
  179. package/src/managedExecution/types.js +1 -0
  180. package/src/managedExecution/types.ts +86 -0
  181. package/src/modelRef.js +9 -0
  182. package/src/modelRef.ts +31 -0
  183. package/src/multimodal.js +14 -28
  184. package/src/multimodal.ts +17 -33
  185. package/src/net/lanDiscovery.js +65 -0
  186. package/src/net/lanDiscovery.ts +82 -0
  187. package/src/plugin-sdk.js +28 -3
  188. package/src/plugin-sdk.ts +39 -6
  189. package/src/plugins/pluginApi.ts +16 -1
  190. package/src/plugins/pluginLoader.js +4 -0
  191. package/src/plugins/pluginLoader.ts +6 -0
  192. package/src/plugins/pluginRunner.js +152 -9
  193. package/src/plugins/pluginRunner.ts +186 -8
  194. package/src/plugins/slashFallback.js +113 -0
  195. package/src/plugins/slashFallback.ts +137 -0
  196. package/src/prompts/policyRegistry.js +15 -8
  197. package/src/prompts/policyRegistry.ts +18 -8
  198. package/src/prompts/promptAssembly.js +103 -12
  199. package/src/prompts/promptAssembly.ts +114 -12
  200. package/src/prompts/skillList.js +21 -3
  201. package/src/prompts/skillList.ts +27 -2
  202. package/src/prompts/stableSystemPrompt.js +18 -0
  203. package/src/prompts/stableSystemPrompt.ts +33 -0
  204. package/src/runtime/agentRuntimeContext.js +8 -2
  205. package/src/runtime/agentRuntimeContext.ts +17 -4
  206. package/src/session/runAgentTurn.js +78 -12
  207. package/src/session/runAgentTurn.ts +105 -23
  208. package/src/session/runTurn.js +10 -9
  209. package/src/session/runTurn.ts +17 -8
  210. package/src/session/sessionState.ts +2 -0
  211. package/src/team/auth.js +186 -0
  212. package/src/team/auth.ts +298 -0
  213. package/src/team/cli.js +105 -0
  214. package/src/team/cli.ts +122 -0
  215. package/src/team/onlineToolContract.js +64 -0
  216. package/src/team/onlineToolContract.ts +109 -0
  217. package/src/team/toolPolicy.js +36 -0
  218. package/src/team/toolPolicy.ts +75 -0
  219. package/src/tools/advisor/advisor.js +148 -37
  220. package/src/tools/advisor/advisor.ts +161 -38
  221. package/src/tools/advisor/advisorDecision.js +71 -0
  222. package/src/tools/advisor/advisorDecision.ts +85 -0
  223. package/src/tools/agent/agent.js +18 -5
  224. package/src/tools/agent/agent.ts +21 -7
  225. package/src/tools/agent/agentControl.js +246 -0
  226. package/src/tools/agent/agentControl.ts +284 -0
  227. package/src/tools/agent/agentTaskService.js +40 -1
  228. package/src/tools/agent/agentTaskService.ts +44 -1
  229. package/src/tools/agent/agentTypes.js +14 -2
  230. package/src/tools/agent/agentTypes.ts +15 -2
  231. package/src/tools/applypatch/applyPatch.js +362 -0
  232. package/src/tools/applypatch/applyPatch.ts +428 -0
  233. package/src/tools/bash/bash.js +70 -18
  234. package/src/tools/bash/bash.ts +75 -20
  235. package/src/tools/capabilities.js +10 -5
  236. package/src/tools/capabilities.ts +10 -5
  237. package/src/tools/computeruse/computerIntent.js +32 -0
  238. package/src/tools/computeruse/computerIntent.ts +46 -0
  239. package/src/tools/computeruse/telemetryLog.js +12 -0
  240. package/src/tools/computeruse/telemetryLog.ts +16 -0
  241. package/src/tools/computeruse/vision/telemetry.js +1 -0
  242. package/src/tools/computeruse/vision/telemetry.ts +1 -0
  243. package/src/tools/effectiveToolSet.js +4 -0
  244. package/src/tools/effectiveToolSet.ts +4 -0
  245. package/src/tools/index.js +21 -2
  246. package/src/tools/index.ts +33 -2
  247. package/src/tools/powershell/powershell.js +18 -1
  248. package/src/tools/powershell/powershell.ts +18 -2
  249. package/src/tools/presentresult/presentresult.js +6 -15
  250. package/src/tools/presentresult/presentresult.ts +6 -15
  251. package/src/tools/read/read.js +69 -11
  252. package/src/tools/read/read.ts +70 -11
  253. package/src/tools/shared/fileUtils.js +1 -2
  254. package/src/tools/shared/fileUtils.ts +1 -2
  255. package/src/tools/shared/permanentDeleteGuard.js +34 -3
  256. package/src/tools/shared/permanentDeleteGuard.ts +43 -6
  257. package/src/tools/shared/toolOutputStore.js +23 -1
  258. package/src/tools/shared/toolOutputStore.ts +22 -1
  259. package/src/tools/skill/skill.js +9 -3
  260. package/src/tools/skill/skill.ts +8 -4
  261. package/src/tools/toolPromptGuides.js +117 -0
  262. package/src/tools/toolPromptGuides.ts +137 -0
  263. package/src/tools/webbrowser/agentBrowserPath.js +19 -9
  264. package/src/tools/webbrowser/agentBrowserPath.ts +70 -28
  265. package/src/tools/webbrowser/browser.js +27 -11
  266. package/src/tools/webbrowser/browser.ts +68 -24
  267. package/src/tools/webbrowser/browserAutomation.js +329 -0
  268. package/src/tools/webbrowser/browserAutomation.ts +467 -0
  269. package/src/tools/webbrowser/browserAxState.js +97 -0
  270. package/src/tools/webbrowser/browserAxState.ts +142 -0
  271. package/src/tools/webbrowser/browserBatch.js +1578 -0
  272. package/src/tools/webbrowser/browserBatch.ts +2382 -0
  273. package/src/tools/webbrowser/browserCdp.js +208 -0
  274. package/src/tools/webbrowser/browserCdp.ts +255 -0
  275. package/src/tools/webbrowser/browserConditions.js +24 -0
  276. package/src/tools/webbrowser/browserConditions.ts +45 -0
  277. package/src/tools/webbrowser/browserContent.js +281 -0
  278. package/src/tools/webbrowser/browserContent.ts +438 -0
  279. package/src/tools/webbrowser/browserControl.js +131 -30
  280. package/src/tools/webbrowser/browserControl.ts +210 -50
  281. package/src/tools/webbrowser/browserDaemon.js +22 -14
  282. package/src/tools/webbrowser/browserDaemon.ts +54 -21
  283. package/src/tools/webbrowser/browserFrames.js +47 -0
  284. package/src/tools/webbrowser/browserFrames.ts +54 -0
  285. package/src/tools/webbrowser/browserInteractionResolver.js +17 -0
  286. package/src/tools/webbrowser/browserInteractionResolver.ts +40 -0
  287. package/src/tools/webbrowser/browserLocator.js +140 -0
  288. package/src/tools/webbrowser/browserLocator.ts +194 -0
  289. package/src/tools/webbrowser/browserOutcome.js +10 -17
  290. package/src/tools/webbrowser/browserOutcome.ts +48 -31
  291. package/src/tools/webbrowser/browserRecovery.js +58 -0
  292. package/src/tools/webbrowser/browserRecovery.ts +83 -0
  293. package/src/tools/webbrowser/browserRunRegistry.js +35 -3
  294. package/src/tools/webbrowser/browserRunRegistry.ts +91 -15
  295. package/src/tools/webbrowser/browserService.js +28 -2
  296. package/src/tools/webbrowser/browserService.ts +49 -9
  297. package/src/tools/webbrowser/browserStream.js +132 -33
  298. package/src/tools/webbrowser/browserStream.ts +224 -50
  299. package/src/tools/webbrowser/browserTabs.js +48 -15
  300. package/src/tools/webbrowser/browserTabs.ts +178 -96
  301. package/src/tools/webbrowser/webbrowser.js +683 -144
  302. package/src/tools/webbrowser/webbrowser.ts +979 -236
  303. package/src/tools/webfetch/preapproved.js +0 -2
  304. package/src/tools/webfetch/preapproved.ts +0 -2
  305. package/src/tools/websearch/websearch.js +129 -8
  306. package/src/tools/websearch/websearch.ts +138 -6
  307. package/src/types.ts +450 -42
  308. package/src/ui/App.js +377 -58
  309. package/src/ui/App.tsx +576 -111
  310. package/src/ui/LiveArea.js +14 -8
  311. package/src/ui/LiveArea.tsx +14 -10
  312. package/src/ui/MessageList.js +63 -20
  313. package/src/ui/MessageList.tsx +101 -16
  314. package/src/ui/ModelPicker.js +7 -4
  315. package/src/ui/ModelPicker.tsx +55 -8
  316. package/src/ui/hooks/useChat.js +152 -163
  317. package/src/ui/hooks/useChat.ts +396 -334
  318. package/src/ui/loginCommand.js +56 -0
  319. package/src/ui/loginCommand.ts +63 -0
  320. package/src/ui/toolProgress.js +41 -0
  321. package/src/ui/toolProgress.ts +44 -0
  322. package/src/web/browserBridge.js +20 -2
  323. package/src/web/browserBridge.ts +20 -2
  324. package/src/web/protocol.ts +1897 -194
  325. package/workflows/adaptive-research/workflow.mjs +86 -0
  326. package/workflows/e2e-capabilities/workflow.mjs +102 -0
  327. package/workflows/e2e-llm/workflow.mjs +52 -0
  328. package/workflows/e2e-smoke/workflow.mjs +59 -0
  329. package/workflows/e2e-vision-schema/README.md +48 -0
  330. package/workflows/e2e-vision-schema/workflow.mjs +149 -0
  331. package/workflows/e2e-write-greet/workflow.mjs +37 -0
  332. package/workflows/file-census/README.md +14 -0
  333. package/workflows/file-census/scripts/.gitkeep +2 -0
  334. package/workflows/file-census/workflow.mjs +110 -0
  335. package/workflows/goal/README.md +166 -0
  336. package/workflows/goal/prompts/audit.md +57 -0
  337. package/workflows/goal/prompts/contract.md +56 -0
  338. package/workflows/goal/prompts/iterate.md +37 -0
  339. package/workflows/goal/workflow.mjs +501 -0
  340. package/workflows/illustrate/README.md +90 -0
  341. package/workflows/illustrate/assets/engines.json +20 -0
  342. package/workflows/illustrate/assets/extract-baidu.js +21 -0
  343. package/workflows/illustrate/assets/extract-bing.js +33 -0
  344. package/workflows/illustrate/assets/extract-google.js +31 -0
  345. package/workflows/illustrate/workflow.mjs +301 -0
  346. package/workflows/plan/workflow.mjs +389 -0
  347. package/workflows/question-crops/README.md +35 -0
  348. package/workflows/question-crops/prompts/inventory.md +23 -0
  349. package/workflows/question-crops/scripts/image-ops.mjs +305 -0
  350. package/workflows/question-crops/scripts/image-ops.test.mjs +95 -0
  351. package/workflows/question-crops/workflow.mjs +172 -0
  352. package/workflows/ralph-loop/workflow.mjs +162 -0
  353. package/workflows/research/DESIGN.md +449 -0
  354. package/workflows/research/README.md +74 -0
  355. package/workflows/research/prompts/s0-goal.md +43 -0
  356. package/workflows/research/prompts/s1-early-exit.md +22 -0
  357. package/workflows/research/prompts/s1-map.md +58 -0
  358. package/workflows/research/prompts/s2-options.md +53 -0
  359. package/workflows/research/prompts/s3-deep-dive.md +56 -0
  360. package/workflows/research/prompts/s4-verify.md +74 -0
  361. package/workflows/research/prompts/s5-synthesis.md +81 -0
  362. package/workflows/research/prompts/s5b-fill-gaps.md +32 -0
  363. package/workflows/research/prompts/s6-package.md +32 -0
  364. package/workflows/research/scripts/build-report.mjs +98 -0
  365. package/workflows/research/workflow.mjs +288 -0
  366. package/workflows/wechat-reading-shots/prompts/collect-params.md +16 -0
  367. package/workflows/wechat-reading-shots/prompts/evaluate.md +20 -0
  368. package/workflows/wechat-reading-shots/prompts/open-book.md +28 -0
  369. package/workflows/wechat-reading-shots/prompts/read-page.md +13 -0
  370. package/workflows/wechat-reading-shots/prompts/turn-batch.md +24 -0
  371. package/workflows/wechat-reading-shots/test/control-flow.test.ts +329 -0
  372. package/workflows/wechat-reading-shots/workflow.mjs +266 -0
  373. package/workflows/wrong-question-docx/README.md +55 -0
  374. package/workflows/wrong-question-docx/prompts/figure-qa.md +8 -0
  375. package/workflows/wrong-question-docx/prompts/transcribe.md +25 -0
  376. package/workflows/wrong-question-docx/scripts/workbook.mjs +857 -0
  377. package/workflows/wrong-question-docx/test/control-flow.test.ts +165 -0
  378. package/workflows/wrong-question-docx/test/workbook.test.mjs +303 -0
  379. package/workflows/wrong-question-docx/workflow.mjs +485 -0
  380. package/plugins/workflow-runner/src/agentCommon.js +0 -354
  381. package/plugins/workflow-runner/src/agentCommon.ts +0 -435
  382. package/plugins/workflow-runner/src/controlFlow/branch.js +0 -40
  383. package/plugins/workflow-runner/src/controlFlow/branch.ts +0 -46
  384. package/plugins/workflow-runner/src/controlFlow/loop.js +0 -211
  385. package/plugins/workflow-runner/src/controlFlow/loop.ts +0 -248
  386. package/plugins/workflow-runner/src/controlFlow/pause.js +0 -129
  387. package/plugins/workflow-runner/src/controlFlow/pause.ts +0 -141
  388. package/plugins/workflow-runner/src/controlFlow/signal.js +0 -21
  389. package/plugins/workflow-runner/src/controlFlow/signal.ts +0 -29
  390. package/plugins/workflow-runner/src/controlFlow/types.js +0 -18
  391. package/plugins/workflow-runner/src/controlFlow/types.ts +0 -63
  392. package/plugins/workflow-runner/src/dslManifest.js +0 -72
  393. package/plugins/workflow-runner/src/dslManifest.ts +0 -127
  394. package/plugins/workflow-runner/src/expressions.js +0 -618
  395. package/plugins/workflow-runner/src/expressions.ts +0 -622
  396. package/plugins/workflow-runner/src/loader.js +0 -573
  397. package/plugins/workflow-runner/src/loader.ts +0 -721
  398. package/plugins/workflow-runner/src/multimodal.js +0 -38
  399. package/plugins/workflow-runner/src/multimodal.ts +0 -64
  400. package/plugins/workflow-runner/src/recovery.js +0 -119
  401. package/plugins/workflow-runner/src/recovery.ts +0 -149
  402. package/plugins/workflow-runner/src/stepExecutors/agent.js +0 -173
  403. package/plugins/workflow-runner/src/stepExecutors/agent.ts +0 -234
  404. package/plugins/workflow-runner/src/stepExecutors/assert.js +0 -41
  405. package/plugins/workflow-runner/src/stepExecutors/assert.ts +0 -50
  406. package/plugins/workflow-runner/src/stepExecutors/echo.js +0 -31
  407. package/plugins/workflow-runner/src/stepExecutors/echo.ts +0 -38
  408. package/plugins/workflow-runner/src/stepExecutors/llm.js +0 -99
  409. package/plugins/workflow-runner/src/stepExecutors/llm.ts +0 -133
  410. package/plugins/workflow-runner/src/stepExecutors/set.js +0 -49
  411. package/plugins/workflow-runner/src/stepExecutors/set.ts +0 -56
  412. package/plugins/workflow-runner/src/stepExecutors/skill.js +0 -111
  413. package/plugins/workflow-runner/src/stepExecutors/skill.ts +0 -140
  414. package/plugins/workflow-runner/src/stepExecutors/tool.js +0 -208
  415. package/plugins/workflow-runner/src/stepExecutors/tool.ts +0 -256
  416. package/plugins/workflow-runner/src/stepExecutors/verify.js +0 -58
  417. package/plugins/workflow-runner/src/stepExecutors/verify.ts +0 -68
  418. package/plugins/workflow-runner/src/stepPipeline.js +0 -277
  419. package/plugins/workflow-runner/src/stepPipeline.ts +0 -366
  420. package/plugins/workflow-runner/test/agent.test.ts +0 -587
  421. package/plugins/workflow-runner/test/agentLlmError.test.ts +0 -373
  422. package/plugins/workflow-runner/test/bashBlacklist.sweep.test.ts +0 -65
  423. package/plugins/workflow-runner/test/captureMisbind.test.ts +0 -99
  424. package/plugins/workflow-runner/test/cli.e2e.test.ts +0 -147
  425. package/plugins/workflow-runner/test/context-files.test.ts +0 -320
  426. package/plugins/workflow-runner/test/e2e.test.ts +0 -370
  427. package/plugins/workflow-runner/test/engineRobustness.test.ts +0 -226
  428. package/plugins/workflow-runner/test/expressions.test.ts +0 -397
  429. package/plugins/workflow-runner/test/foundation.test.ts +0 -126
  430. package/plugins/workflow-runner/test/graceful.test.ts +0 -139
  431. package/plugins/workflow-runner/test/greedyInputs.load.test.ts +0 -47
  432. package/plugins/workflow-runner/test/hostCapability.test.ts +0 -108
  433. package/plugins/workflow-runner/test/illustrate.load.test.ts +0 -183
  434. package/plugins/workflow-runner/test/jsonPayload.test.ts +0 -83
  435. package/plugins/workflow-runner/test/loader.test.ts +0 -617
  436. package/plugins/workflow-runner/test/maxTurnsRetryDelay.test.ts +0 -33
  437. package/plugins/workflow-runner/test/multimodal.test.ts +0 -342
  438. package/plugins/workflow-runner/test/output-schema.test.ts +0 -213
  439. package/plugins/workflow-runner/test/parseWorkflowArgs.test.ts +0 -87
  440. package/plugins/workflow-runner/test/pluginRunner.isolation.test.ts +0 -269
  441. package/plugins/workflow-runner/test/ppt.load.test.ts +0 -1220
  442. package/plugins/workflow-runner/test/ralphLoop.e2e.test.ts +0 -220
  443. package/plugins/workflow-runner/test/recovery.test.ts +0 -265
  444. package/plugins/workflow-runner/test/runner.test.ts +0 -911
  445. package/plugins/workflow-runner/test/runnerPreflight.test.ts +0 -379
  446. package/plugins/workflow-runner/test/skillTurnDone.test.ts +0 -71
  447. package/plugins/workflow-runner/test/stateLifecycle.test.ts +0 -150
  448. package/plugins/workflow-runner/test/stepPipeline.test.ts +0 -114
  449. package/plugins/workflow-runner/test/taskArbitration.test.ts +0 -338
  450. package/plugins/workflow-runner/test/taskLifecycle.test.ts +0 -270
  451. package/plugins/workflow-runner/test/toolArtifactOverwrite.test.ts +0 -221
  452. package/plugins/workflow-runner/test/toolCapabilityBoundary.test.ts +0 -29
  453. package/plugins/workflow-runner/test/toolRuntimeIntegration.test.ts +0 -168
  454. package/plugins/workflow-runner/test/webppt.load.test.ts +0 -250
  455. package/plugins/workflow-runner/test/word.load.test.ts +0 -208
  456. package/plugins/workflow-runner/test/workflows.smoke.test.ts +0 -76
  457. package/plugins/workflow-runner/test/yamlGuardConditions.test.ts +0 -207
  458. package/src/advisorState.js +0 -22
  459. package/src/cli/configWizard.js +0 -293
  460. package/src/tools/toolProgressBridge.js +0 -48
  461. package/workflows/adaptive-research.yaml +0 -50
  462. package/workflows/book-review-short.yaml +0 -125
  463. package/workflows/e2e-write-greet.yaml +0 -31
  464. package/workflows/illustrate.yaml +0 -227
  465. package/workflows/plan.yaml +0 -470
  466. package/workflows/ppt.yaml +0 -1670
  467. package/workflows/ralph-loop.yaml +0 -199
  468. package/workflows/schema.json +0 -176
  469. package/workflows/webppt.yaml +0 -496
  470. package/workflows/wechat-reading-shots.yaml +0 -315
  471. package/workflows/word.yaml +0 -720
  472. package/workflows/youtube-shorts.yaml +0 -202
@@ -0,0 +1,36 @@
1
+ import { ONLINE_TOOL_CATEGORY_BY_NAME, onlineToolCategory as sharedOnlineToolCategory, } from './onlineToolContract.js';
2
+ const TOOL_SECURITY_METADATA = Object.fromEntries(Object.entries(ONLINE_TOOL_CATEGORY_BY_NAME).map(([name, category]) => [
3
+ name,
4
+ {
5
+ name,
6
+ category,
7
+ mayTransmitLocalData: category === 'workspace' || category === 'computer',
8
+ },
9
+ ]));
10
+ export function onlineToolCategory(name) {
11
+ return sharedOnlineToolCategory(name);
12
+ }
13
+ export function toolSecurityMetadata(name) {
14
+ return TOOL_SECURITY_METADATA[name];
15
+ }
16
+ /** Applies the remote model's maximum category boundary before schemas leave
17
+ * the device. Local providers retain their original complete tool set. */
18
+ export function filterToolsForProvider(tools, provider) {
19
+ if (provider.source !== 'gateway')
20
+ return [...tools];
21
+ return filterToolsForCategories(tools,
22
+ // The server maximum is the non-negotiable upper bound. The compatibility
23
+ // field is only a fallback for old descriptors that lack the explicit
24
+ // max/default split; it must never override the maximum with defaults.
25
+ provider.gatewayMaxToolCategories ?? provider.gatewayToolCategories);
26
+ }
27
+ /** Applies the per-Session switch inside the server/model upper bound. */
28
+ export function filterToolsForCategories(tools, categories) {
29
+ if (!categories)
30
+ return [...tools];
31
+ const allowed = new Set(categories);
32
+ return tools.filter((tool) => {
33
+ const category = onlineToolCategory(tool.name);
34
+ return category !== undefined && allowed.has(category);
35
+ });
36
+ }
@@ -0,0 +1,75 @@
1
+ import type { Provider, Tool } from '../types.js';
2
+ import {
3
+ ONLINE_TOOL_CATEGORY_BY_NAME,
4
+ type OnlineToolCategory,
5
+ onlineToolCategory as sharedOnlineToolCategory,
6
+ } from './onlineToolContract.js';
7
+
8
+ /**
9
+ * Categories shared with minimal-server. They are a server-enforced upper
10
+ * bound when a remote model is selected. Unknown/dynamic tools fail closed:
11
+ * a plugin cannot silently gain remote-model visibility just by registering a
12
+ * new tool name.
13
+ */
14
+ export type { OnlineToolCategory } from './onlineToolContract.js';
15
+
16
+ export interface ToolSecurityMetadata {
17
+ name: string;
18
+ category: OnlineToolCategory;
19
+ /** Reserved for the privacy UI and future per-tool redaction policy. */
20
+ mayTransmitLocalData: boolean;
21
+ }
22
+
23
+ const TOOL_SECURITY_METADATA: Readonly<Record<string, ToolSecurityMetadata>> =
24
+ Object.fromEntries(
25
+ Object.entries(ONLINE_TOOL_CATEGORY_BY_NAME).map(([name, category]) => [
26
+ name,
27
+ {
28
+ name,
29
+ category,
30
+ mayTransmitLocalData:
31
+ category === 'workspace' || category === 'computer',
32
+ },
33
+ ]),
34
+ ) as Readonly<Record<string, ToolSecurityMetadata>>;
35
+
36
+ export function onlineToolCategory(
37
+ name: string,
38
+ ): OnlineToolCategory | undefined {
39
+ return sharedOnlineToolCategory(name);
40
+ }
41
+
42
+ export function toolSecurityMetadata(
43
+ name: string,
44
+ ): ToolSecurityMetadata | undefined {
45
+ return TOOL_SECURITY_METADATA[name];
46
+ }
47
+
48
+ /** Applies the remote model's maximum category boundary before schemas leave
49
+ * the device. Local providers retain their original complete tool set. */
50
+ export function filterToolsForProvider(
51
+ tools: readonly Tool[],
52
+ provider: Provider,
53
+ ): Tool[] {
54
+ if (provider.source !== 'gateway') return [...tools];
55
+ return filterToolsForCategories(
56
+ tools,
57
+ // The server maximum is the non-negotiable upper bound. The compatibility
58
+ // field is only a fallback for old descriptors that lack the explicit
59
+ // max/default split; it must never override the maximum with defaults.
60
+ provider.gatewayMaxToolCategories ?? provider.gatewayToolCategories,
61
+ );
62
+ }
63
+
64
+ /** Applies the per-Session switch inside the server/model upper bound. */
65
+ export function filterToolsForCategories(
66
+ tools: readonly Tool[],
67
+ categories?: readonly string[],
68
+ ): Tool[] {
69
+ if (!categories) return [...tools];
70
+ const allowed = new Set(categories as OnlineToolCategory[]);
71
+ return tools.filter((tool) => {
72
+ const category = onlineToolCategory(tool.name);
73
+ return category !== undefined && allowed.has(category);
74
+ });
75
+ }
@@ -25,6 +25,7 @@
25
25
  import { z } from 'zod';
26
26
  import { configService } from '../../config/configService.js';
27
27
  import { CONFIG_LIMITS, normalizeLimited } from '../../config/numericLimits.js';
28
+ import { DEFAULT_ADVISOR_CONFIG } from '../../config/configDefaults.js';
28
29
  import { countMessagesTokens } from '../../context/tokenCounter.js';
29
30
  import { chat } from '../../llm/client.js';
30
31
  import { derivePromptCacheContext } from '../../llm/promptCache.js';
@@ -34,19 +35,21 @@ import { DEFAULT_MAX_RESULT_SIZE_CHARS } from '../types.js';
34
35
  const inputSchema = z.object({
35
36
  note: z
36
37
  .string()
37
- .max(2000, 'note 太长(>2000 字)')
38
+ // 上限只做护栏,不做实用约束:note 是顾问唯一能看到的「模型主动强调的证据」,
39
+ // 卡太低(曾经 2000 字符)会让完整方案+关键代码原文根本塞不进去。真正的预算
40
+ // 闸门在 buildAdvisorMessages -> fitAdvisorBudget(inputTokenBudget 配置 +
41
+ // 顾问 contextWindow 推导),note 超预算只会挤掉历史消息,不会炸请求。
42
+ .max(100_000, 'note 太长(>100000 字符)')
38
43
  .optional()
39
- .describe('可选。大多数情况下不需要填——你此前说过的方案/诊断已经在会话历史里,顾问会自动看到。' +
44
+ .describe('可选。大多数情况下不需要填--你此前说过的方案/诊断已经在会话历史里,顾问会自动看到。' +
40
45
  '只有需要额外强调、历史里没体现的点时才填一句话。'),
41
46
  });
42
47
  const parameters = toToolParameters(inputSchema);
43
48
  // ---------------- 2. Description(给 LLM 看的工具说明) ----------------
44
- const description = `- Consults an independent, separately-configured second LLM ("顾问模型" / advisor) for a second opinion on your current plan, diagnosis, or progress.
45
- - Call with no arguments in most cases — the advisor receives a bounded, relevant view of the current session (the original request, recent discussion, current plan/diagnosis, and key tool results). Only pass \`note\` if something important is not otherwise clear.
46
- - Exception: if your current plan/diagnosis relies on things you saw in images (screenshots, photos, diagrams), summarize the key visual facts in \`note\` in 1-2 sentences (what each image shows and what you concluded from it) — the advisor may not be able to see the images themselves, and any visual judgment you never wrote down as text does not exist for it.
47
- - Call this: (1) after a plan/diagnosis has taken shape but before you commit to it (write code, declare a conclusion); (2) when you're stuck — the same error keeps recurring, your fixes aren't converging, or the user has pushed back and you're reconsidering direction; (3) before declaring a multi-step task complete and reporting back to the user.
48
- - The advisor has no tool access and does a single read-only pass over the transcript — it cannot verify anything itself, only reason about what it's shown.
49
- - If this tool errors (advisor not configured in shared settings, disabled, manual-only, or a call failure), that's not a signal to stop — treat it as "no second opinion available this time" and continue with your own judgment. Do not block or retry in a loop.`;
49
+ const description = `Consult an independently configured, read-only advisor model for a second opinion.
50
+ The advisor receives a bounded view of the current session. Usually omit note; use it only for decisive evidence or visual facts that are absent or easy to miss in history.
51
+ The result includes a full consultationId. After a successful automatic consultation, record the main Agent's evidence-based accept/partial/reject judgment with AdvisorDecision.
52
+ The advisor is neither a fact source nor an authority. If it is unavailable or fails, follow the returned degradation guidance and do not retry in a loop.`;
50
53
  // ---------------- 3. 顾问 persona system prompt ----------------
51
54
  const ADVISOR_SYSTEM_PROMPT = `你是一位经验丰富的全栈开发专家技术顾问,为正在独立执行任务的另一个 AI 助手提供第二意见评审。
52
55
  你的职责:技术方案评估、代码评审、疑难问题的根因定位与解决方向判断。
@@ -54,6 +57,14 @@ const ADVISOR_SYSTEM_PROMPT = `你是一位经验丰富的全栈开发专家技
54
57
  你会看到该助手当前会话中受预算约束的相关记录——包括首要需求、近期交流、它的思考过程、当前方案
55
58
  和关键工具结果与执行结果,以及它现在想让你重点关注的内容(如果有)。
56
59
 
60
+ 重要:你**没有工具访问权,无法自行验证任何事实**。你只是提供一个不同视角的怀疑和推理,
61
+ 不能把自己的推测当成确定结论。每条指出的问题都要明确标注:
62
+ - 【有证据】:基于记录里明确出现的代码/测试结果/事实
63
+ - 【推断】:基于已有信息的合理推断,但未被证实
64
+ - 【存疑】:信息不足,需要执行助手回去核实
65
+ 不要过度自信,不要给出确定性语气的未验证判断——执行助手会对你的意见逐条做验证,
66
+ 夸大其词会被判定为误报。
67
+
57
68
  审视时请做到:
58
69
  - 像资深工程师做方案评审/代码评审一样直接、具体——发现问题就明说,不要无条件附和或说套话。
59
70
  - 优先检查:方案是否遗漏了风险或边界情况、诊断是否真的找到了根因还是只是让症状消失、
@@ -66,7 +77,16 @@ const ADVISOR_SYSTEM_PROMPT = `你是一位经验丰富的全栈开发专家技
66
77
  或换个方式绕开,都要点出来。
67
78
  - 结论必须基于你在历史记录里看到的实际证据,不要脱离上下文给泛泛的通用建议;
68
79
  被省略的记录(尤其工具结果原文)所涉及的判断,明说你无法核实。
69
- - 篇幅精炼:只讲最关键的 1-3 点,不要复述历史、不要客套开场白。`;
80
+ - 篇幅精炼:只讲最关键的 1-3 点,不要复述历史、不要客套开场白。
81
+
82
+ 关于「带着反证回来」的那一次咨询(助手被要求这样做,不是在质疑你):
83
+ 当它说「我实测是 X,你上次建议 Y」时,它手里的是一手证据,你手里的是推理——
84
+ 默认它的观测为真,然后只回答一个问题:**哪个约束更硬**。
85
+ - 你上次漏看了它给出的证据 → 直接认,并说清结论怎么变。
86
+ - 它的观测其实没测到你担心的那一层(测了被改的函数,而真实调用路径根本走不到)
87
+ → 指出这个错位,并说要测什么才算数。
88
+ - 两者都成立、只是适用范围不同 → 说清各自的边界,别硬选一个。
89
+ 不要为了显得一致而坚持原判,也不要因为被反驳就整个收回——两种都会让它失去判断依据。`;
70
90
  // ---------------- 4. 消息构造(纯函数,单独导出可测) ----------------
71
91
  /**
72
92
  * 把主循环的 history 转成喂给顾问模型的 messages。
@@ -75,8 +95,19 @@ const ADVISOR_SYSTEM_PROMPT = `你是一位经验丰富的全栈开发专家技
75
95
  * 和 TUI 的 ephemeral 提示消息(两者都是 role:'system'),换成顾问自己的 persona。
76
96
  */
77
97
  const MAX_TOOL_RESULT_CHARS = 12_000;
98
+ /**
99
+ * 工具调用参数与思维链的截断上限——与 `fitAdvisorBudget` 的预算裁剪是两回事:
100
+ * 那个只在总量超预算时才生效、按消息整条丢弃;这两个是每条消息内部无条件应用的
101
+ * 字段级上限,不看当前预算是否宽裕。思维链截得比正文更狠:重思考模型的
102
+ * reasoning 常比 content 长几倍,不设限会把预算吃光、挤掉更该保留的工具结果原文。
103
+ * 未发现过当前取值导致顾问漏判的实例,因此只命名、不改值——真出现具体案例
104
+ * 再调,不做没有证据支撑的预防性放宽。
105
+ */
106
+ const ADVISOR_TOOL_ARGS_TRUNCATE_CHARS = 2_000;
107
+ const ADVISOR_REASONING_TRUNCATE_CHARS = 2_000;
78
108
  /** 归一化后的工具结果前缀;裁剪统计要靠它反查,两处必须共用同一常量。 */
79
109
  const TOOL_RESULT_PREFIX = '[工具结果 ';
110
+ const CONTEXT_PACKET_PREFIX = '[Advisor 结构化上下文包]';
80
111
  function truncateMiddle(value, limit = MAX_TOOL_RESULT_CHARS) {
81
112
  if (value.length <= limit)
82
113
  return value;
@@ -161,18 +192,16 @@ function normalizeAdvisorMessage(message, supportsVision, allowImages, imageBudg
161
192
  return { role: 'user', content: `${TOOL_RESULT_PREFIX}${message.tool_call_id}]\n${truncateMiddle(message.content)}` };
162
193
  }
163
194
  if (message.role === 'assistant') {
164
- const calls = message.tool_calls?.map((call) => `[调用工具 ${call.function.name}] ${truncateMiddle(call.function.arguments, 2000)}`).join('\n');
195
+ const calls = message.tool_calls?.map((call) => `[调用工具 ${call.function.name}] ${truncateMiddle(call.function.arguments, ADVISOR_TOOL_ARGS_TRUNCATE_CHARS)}`).join('\n');
165
196
  // 思维链一并送给顾问:评审「方案对不对」很大程度上要看**它是怎么想到这个方案的**——
166
197
  // 推理里的跳跃(把推断当已验证、跳过一个没排除的可能)在最终结论文本里往往看不出来。
167
- // 截断比正文更狠(2000):重思考模型的 reasoning 常比 content 长几倍,不设限会把
168
- // 预算吃光、反而挤掉更该保留的工具结果原文。
169
198
  // reasoning_details(结构化数组)不传:结构因 provider 而异,既无法可靠脱敏也无法
170
199
  // 稳定转成文本,传过去只会变成噪声。
171
200
  const thinking = typeof message.reasoning_content === 'string' && message.reasoning_content.length > 0
172
201
  ? message.reasoning_content
173
202
  : (typeof message.reasoning === 'string' ? message.reasoning : '');
174
203
  const thinkingLine = thinking.length > 0
175
- ? `[思考过程]\n${truncateMiddle(thinking, 2000)}`
204
+ ? `[思考过程]\n${truncateMiddle(thinking, ADVISOR_REASONING_TRUNCATE_CHARS)}`
176
205
  : '';
177
206
  return {
178
207
  role: 'assistant',
@@ -195,22 +224,34 @@ function fitAdvisorBudget(messages, inputBudget) {
195
224
  const system = messages[0];
196
225
  const closing = messages[messages.length - 1];
197
226
  const body = messages.slice(1, -1);
198
- const firstUserIndex = body.findIndex((message) => message.role === 'user');
227
+ const packetIndex = body.findIndex((message) => message.role === 'user'
228
+ && typeof message.content === 'string'
229
+ && message.content.startsWith(CONTEXT_PACKET_PREFIX));
230
+ const firstUserIndex = body.findIndex((message, index) => index !== packetIndex && message.role === 'user');
199
231
  const selected = [];
200
232
  let used = countMessagesTokens([system, closing]);
201
- if (firstUserIndex >= 0) {
202
- const first = body[firstUserIndex];
203
- selected.push(first);
204
- used += countMessagesTokens([first]);
233
+ for (const index of [packetIndex, firstUserIndex]) {
234
+ if (index < 0)
235
+ continue;
236
+ const pinned = body[index];
237
+ selected.push(pinned);
238
+ used += countMessagesTokens([pinned]);
205
239
  }
240
+ // 从最近往前收,**装不下就停**(break,不是 continue)。
241
+ //
242
+ // ⚠️ 曾经是 continue:一条大消息放不下就跳过它、继续往前捡更小的旧消息。
243
+ // 结果是给顾问一段**中间有洞**的对话 —— 而洞恰好开在信息量最大的地方(长工具
244
+ // 结果、完整方案文本),偏偏下面那句省略提示还写着「保留最初需求和最近相关记录」,
245
+ // 顾问于是把两条本不相邻的消息读成前后相接,在残缺的因果链上给出自信结论。
246
+ // 宁可少给,不可给错:连续的尾部 + 如实说明从哪里开始省略。
206
247
  const recent = [];
207
248
  for (let index = body.length - 1; index >= 0; index--) {
208
- if (index === firstUserIndex)
249
+ if (index === firstUserIndex || index === packetIndex)
209
250
  continue;
210
251
  const message = body[index];
211
252
  const cost = countMessagesTokens([message]);
212
253
  if (used + cost > inputBudget)
213
- continue;
254
+ break;
214
255
  recent.unshift(message);
215
256
  used += cost;
216
257
  }
@@ -248,12 +289,18 @@ export function buildAdvisorMessages(history, note, options = {}) {
248
289
  .join('\n');
249
290
  const messages = [
250
291
  { role: 'system', content: ADVISOR_SYSTEM_PROMPT },
292
+ ...(options.contextPacket
293
+ ? [{
294
+ role: 'user',
295
+ content: `${CONTEXT_PACKET_PREFIX}\n${JSON.stringify(options.contextPacket, null, 2)}`,
296
+ }]
297
+ : []),
251
298
  ...nonSystem,
252
299
  { role: 'user', content: closing },
253
300
  ];
254
301
  const contextWindow = options.contextWindow ?? 128_000;
255
302
  const maxTokens = options.maxTokens ?? 4096;
256
- const configuredBudget = normalizeLimited(options.inputTokenBudget, CONFIG_LIMITS.advisorInputTokenBudget, 128_000);
303
+ const configuredBudget = normalizeLimited(options.inputTokenBudget, CONFIG_LIMITS.advisorInputTokenBudget, DEFAULT_ADVISOR_CONFIG.inputTokenBudget);
257
304
  const providerBudget = Math.max(CONFIG_LIMITS.advisorInputTokenBudget.min, contextWindow - maxTokens - 4096);
258
305
  const inputBudget = Math.min(configuredBudget, providerBudget);
259
306
  return fitAdvisorBudget(messages, inputBudget);
@@ -282,34 +329,66 @@ function classifyAdvisorError(e) {
282
329
  * `chatFn: typeof chat = chat` 模式),不需要 mock.module。
283
330
  */
284
331
  export async function runAdvisor(input, signal, ctx, chatFn = chat, runtime = {}) {
332
+ const consultation = ctx?.advisorInvocation === 'agent'
333
+ ? ctx.advisorRuntime?.beginConsultation()
334
+ : undefined;
335
+ const finish = (result) => {
336
+ if (!consultation || !ctx?.advisorRuntime)
337
+ return result;
338
+ const metadata = ctx.advisorRuntime.finishConsultation(consultation, result.ok);
339
+ return {
340
+ ...result,
341
+ ...(result.ok
342
+ ? {
343
+ content: `consultationId: ${metadata.consultationId}\n调用 AdvisorDecision 时必须原样使用上面的完整 consultationId。\n\n${result.content}`,
344
+ }
345
+ : {}),
346
+ metadata,
347
+ };
348
+ };
285
349
  const advisor = await configService.advisor();
286
350
  if (!advisor) {
287
- return {
351
+ return finish({
288
352
  ok: false,
289
353
  error: '顾问模型未配置。请在设置 → 模型中完成专家顾问配置。' +
290
- '可继续凭自己判断推进,不必等待顾问。',
291
- };
354
+ '本回合后续不要再次尝试调用 Advisor;可继续凭自己判断推进,不必等待顾问。',
355
+ });
292
356
  }
293
357
  if (!advisor.enabled) {
294
- return {
358
+ return finish({
295
359
  ok: false,
296
- error: '顾问模型当前已关闭(/advisor on 开启)。可继续凭自己判断推进,不必等待顾问。',
297
- };
360
+ error: '顾问模型当前已关闭(/advisor on 开启)。' +
361
+ '本回合后续不要再次尝试调用 Advisor;可继续凭自己判断推进,不必等待顾问。',
362
+ });
298
363
  }
299
364
  if (advisor.mode === 'manual' && ctx?.advisorInvocation !== 'manual') {
300
- return { ok: false, error: '顾问模型当前为“仅手动”模式;请使用 /advisor ask 发起咨询。' };
365
+ return finish({
366
+ ok: false,
367
+ error: '顾问模型当前为"仅手动"模式;请使用 /advisor ask 发起咨询。' +
368
+ '本回合后续不要再次尝试自动调用 Advisor。',
369
+ });
301
370
  }
302
- if (ctx?.advisorCalls && ctx.advisorCalls.used >= ctx.advisorCalls.max) {
371
+ // 复核豁免:有一个已成功、还没被 AdvisorDecision 结清的咨询时,这次调用是对它
372
+ // 的复核而不是一次新检查点的咨询——工具说明书要求"顾问意见与证据冲突时再问一次",
373
+ // 若复核也照常计费/受配额拦截,短任务(配额恒为 1)会在第二次调用直接撞墙,
374
+ // 而配额耗尽的处理是把这次咨询标记失败并**顺带结清 #pending 指向的检查点**——
375
+ // 模型想认真复核,反而因为差一点配额把本该把关的高风险检查点悄悄放行了。
376
+ // 豁免只覆盖第一次复核:协调器按 checkpoint 记录成功咨询次数,完成一次复核后
377
+ // canReconcileConsultation 会变为 false,后续调用重新受正常配额约束。
378
+ const isReconcile = ctx?.advisorInvocation === 'agent'
379
+ && Boolean(ctx.advisorRuntime?.pendingConsultationId)
380
+ && ctx.advisorRuntime?.canReconcileConsultation === true;
381
+ if (!isReconcile && ctx?.advisorCalls && ctx.advisorCalls.used >= ctx.advisorCalls.max) {
303
382
  // ⚠️ 必须说明「配额会增长」:配额随本回合工作量上调(loop.ts::advisorQuotaFor),
304
383
  // 若这里只说「已达上限」,模型会当成本回合的硬上限而不再尝试 —— 那样长回合里
305
384
  // 「批次做完再复核一次」的配额永远不会被用到,增长机制等于没做。
306
- return {
385
+ return finish({
307
386
  ok: false,
308
387
  error: `本回合顾问咨询配额暂时用完(当前 ${ctx.advisorCalls.max} 次)。` +
309
388
  '配额会随本回合的工作量增长,继续推进一段后可以再次咨询;现在请按已有意见继续。',
310
- };
389
+ });
311
390
  }
312
- if (ctx?.advisorCalls)
391
+ if (!isReconcile && ctx?.advisorCalls)
313
392
  ctx.advisorCalls.used++;
314
393
  const provider = advisor.provider;
315
394
  const snapshot = await configService.getSnapshot();
@@ -329,6 +408,9 @@ export async function runAdvisor(input, signal, ctx, chatFn = chat, runtime = {}
329
408
  contextWindow: provider.contextWindow,
330
409
  maxTokens: provider.maxTokens,
331
410
  inputTokenBudget: advisor.inputTokenBudget,
411
+ ...(ctx?.advisorRuntime
412
+ ? { contextPacket: ctx.advisorRuntime.contextPacket(input.note) }
413
+ : {}),
332
414
  });
333
415
  // Advisor 是可降级的第二意见:使用独立、可配置的总时限,不影响主模型的流式时限。
334
416
  // 默认 180 秒,直接使用所选模型的上下文窗口配置。
@@ -390,22 +472,25 @@ export async function runAdvisor(input, signal, ctx, chatFn = chat, runtime = {}
390
472
  contextWindow: provider.contextWindow,
391
473
  maxTokens: provider.maxTokens,
392
474
  inputTokenBudget: advisor.inputTokenBudget,
475
+ ...(ctx?.advisorRuntime
476
+ ? { contextPacket: ctx.advisorRuntime.contextPacket(input.note) }
477
+ : {}),
393
478
  }));
394
479
  degradedWithoutImages = true;
395
480
  }
396
481
  }
397
482
  catch (e) {
398
483
  if (timeoutController.signal.aborted && !signal?.aborted) {
399
- return {
484
+ return finish({
400
485
  ok: false,
401
486
  error: `顾问模型响应超过 ${Math.round(timeoutMs / 1000)}s 未完成,已跳过。` +
402
487
  '可继续凭自己判断推进,不必等待顾问。',
403
- };
488
+ });
404
489
  }
405
- return {
490
+ return finish({
406
491
  ok: false,
407
492
  error: `${classifyAdvisorError(e)} 可继续凭自己判断推进,不必等待顾问。`,
408
- };
493
+ });
409
494
  }
410
495
  finally {
411
496
  clearTimeout(timer);
@@ -419,7 +504,7 @@ export async function runAdvisor(input, signal, ctx, chatFn = chat, runtime = {}
419
504
  content.slice(0, DEFAULT_MAX_RESULT_SIZE_CHARS) +
420
505
  `\n\n... (输出超过 ${DEFAULT_MAX_RESULT_SIZE_CHARS} 字符,已截断)`;
421
506
  }
422
- return { ok: true, content };
507
+ return finish({ ok: true, content });
423
508
  }
424
509
  // 无宿主上下文的调用视为用户手动咨询;生产主循环使用 callWithContext。
425
510
  async function call(input, signal) {
@@ -435,6 +520,32 @@ export const advisorTool = {
435
520
  isConcurrencySafe: false,
436
521
  maxResultSizeChars: DEFAULT_MAX_RESULT_SIZE_CHARS,
437
522
  interruptBehavior: () => 'cancel',
523
+ /**
524
+ * 动态可用性探测:未配置 / 已关闭 / 仅手动模式 → 工具不暴露给模型。
525
+ * 对齐 ComputerUse / WebBrowser 的模式:关闭时工具直接从 LLM 视野消失,
526
+ * 避免"看到工具但调不了→反复尝试"的流畅度问题。
527
+ *
528
+ * signature 随 enabled/mode/modelId 变化,确保配置改动后工具集签名更新、
529
+ * 模型下一轮拿到的是新工具列表。
530
+ */
531
+ resolveAvailability() {
532
+ // 走同步探测(读内存快照),不触发磁盘 IO;
533
+ // resolveAvailability 是工具集构造阶段的同步调用,不能 await。
534
+ const avail = configService.advisorAvailabilitySync();
535
+ // ⚠️ 安全兜底:如果 snapshot 尚未初始化(极端情况:工具集构造发生在任何
536
+ // configService 异步调用之前),宁可返回 available=true 让工具显示出来
537
+ // (调用时 runAdvisor 会做真正的 fresh reload 校验),也不能误判为
538
+ // "未配置" 导致 Advisor 从工具列表里消失——少一个工具比多一个工具严重得多。
539
+ if (avail.configured && !avail.enabled) {
540
+ return { available: false, signature: 'disabled' };
541
+ }
542
+ if (avail.configured && avail.mode === 'manual') {
543
+ return { available: false, signature: 'manual-only' };
544
+ }
545
+ // 未初始化 / 已配置且可用 → 都显示工具(未配置时调用会有明确错误提示)
546
+ // modelId 进签名:不同顾问模型 = 不同能力 = 工具语义不同。
547
+ return { available: true, signature: avail.modelId ?? 'initializing' };
548
+ },
438
549
  call,
439
550
  callWithContext: runAdvisor,
440
551
  };