niceeval 0.6.1 → 0.7.0

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 (378) hide show
  1. package/INDEX.md +23 -23
  2. package/README.zh.md +6 -6
  3. package/dist/agents/types.d.ts +69 -7
  4. package/dist/context/types.d.ts +32 -12
  5. package/dist/i18n/en.d.ts +54 -0
  6. package/dist/i18n/zh-CN.d.ts +57 -3
  7. package/dist/o11y/types.d.ts +16 -2
  8. package/dist/report/aggregate.d.ts +32 -24
  9. package/dist/report/aggregate.js +158 -50
  10. package/dist/report/built-in/index.d.ts +2 -0
  11. package/dist/report/built-in/index.js +8 -0
  12. package/dist/report/components.d.ts +93 -160
  13. package/dist/report/components.js +377 -114
  14. package/dist/report/compute.d.ts +87 -81
  15. package/dist/report/compute.js +597 -417
  16. package/dist/report/flag.d.ts +32 -6
  17. package/dist/report/flag.js +92 -4
  18. package/dist/report/format.d.ts +19 -11
  19. package/dist/report/format.js +30 -13
  20. package/dist/report/index.d.ts +16 -16
  21. package/dist/report/index.js +20 -21
  22. package/dist/report/load.js +3 -2
  23. package/dist/report/locale.d.ts +57 -33
  24. package/dist/report/locale.js +122 -56
  25. package/dist/report/metrics.d.ts +23 -4
  26. package/dist/report/metrics.js +110 -25
  27. package/dist/report/primitives.d.ts +48 -15
  28. package/dist/report/primitives.js +135 -26
  29. package/dist/report/react/AttemptList.d.ts +9 -7
  30. package/dist/report/react/AttemptList.js +17 -10
  31. package/dist/report/react/DeltaTable.js +19 -18
  32. package/dist/report/react/EvalList.d.ts +4 -4
  33. package/dist/report/react/EvalList.js +0 -0
  34. package/dist/report/react/ExperimentComparison.d.ts +10 -0
  35. package/dist/report/react/ExperimentComparison.js +12 -0
  36. package/dist/report/react/ExperimentList.d.ts +4 -3
  37. package/dist/report/react/ExperimentList.js +17 -18
  38. package/dist/report/react/MetricBars.js +5 -4
  39. package/dist/report/react/MetricLine.js +12 -5
  40. package/dist/report/react/MetricMatrix.js +1 -1
  41. package/dist/report/react/MetricScatter.js +59 -28
  42. package/dist/report/react/MetricTable.js +2 -12
  43. package/dist/report/react/ScopeSummary.d.ts +10 -0
  44. package/dist/report/react/ScopeSummary.js +17 -0
  45. package/dist/report/react/Scoreboard.js +6 -6
  46. package/dist/report/react/cell.js +2 -2
  47. package/dist/report/react/chart-math.d.ts +23 -6
  48. package/dist/report/react/chart-math.js +71 -19
  49. package/dist/report/react/fixtures.d.ts +5 -9
  50. package/dist/report/react/fixtures.js +110 -147
  51. package/dist/report/react/index.d.ts +15 -5
  52. package/dist/report/react/index.js +18 -7
  53. package/dist/report/report.d.ts +137 -16
  54. package/dist/report/report.js +259 -28
  55. package/dist/report/text/faces.d.ts +17 -19
  56. package/dist/report/text/faces.js +253 -184
  57. package/dist/report/text/plot.js +1 -1
  58. package/dist/report/text/table.js +38 -7
  59. package/dist/report/tree.d.ts +90 -40
  60. package/dist/report/tree.js +252 -94
  61. package/dist/report/types.d.ts +247 -284
  62. package/dist/report/types.js +4 -3
  63. package/dist/report/web.d.ts +21 -5
  64. package/dist/report/web.js +42 -16
  65. package/dist/results/select.d.ts +38 -16
  66. package/dist/results/select.js +73 -25
  67. package/dist/results/types.d.ts +49 -14
  68. package/dist/runner/feedback/sink.d.ts +110 -0
  69. package/dist/runner/types.d.ts +513 -22
  70. package/dist/sandbox/docker.d.ts +23 -2
  71. package/dist/sandbox/e2b.d.ts +15 -1
  72. package/dist/sandbox/errors.d.ts +30 -3
  73. package/dist/sandbox/io-retry.d.ts +17 -0
  74. package/dist/sandbox/registry.d.ts +2 -0
  75. package/dist/sandbox/resolve.d.ts +18 -5
  76. package/dist/sandbox/retry.d.ts +11 -1
  77. package/dist/sandbox/types.d.ts +39 -5
  78. package/dist/sandbox/vercel.d.ts +7 -1
  79. package/dist/scoring/coverage.d.ts +30 -0
  80. package/dist/scoring/display.d.ts +21 -0
  81. package/dist/scoring/display.js +120 -0
  82. package/dist/scoring/types.d.ts +103 -20
  83. package/dist/shared/aggregate.d.ts +4 -2
  84. package/dist/shared/aggregate.js +8 -7
  85. package/dist/shared/types.d.ts +28 -0
  86. package/dist/tty-line.d.ts +0 -4
  87. package/dist/util.d.ts +23 -0
  88. package/docs-site/zh/README.md +44 -0
  89. package/docs-site/zh/examples/ai-agent-application.mdx +63 -0
  90. package/docs-site/zh/examples/coding-agent-extensions.mdx +57 -0
  91. package/docs-site/zh/examples/index.mdx +50 -0
  92. package/docs-site/zh/{concepts → explanation}/adapter.mdx +31 -13
  93. package/docs-site/zh/{concepts → explanation}/assert.mdx +7 -7
  94. package/docs-site/zh/{concepts → explanation}/drive.mdx +8 -8
  95. package/docs-site/zh/{concepts → explanation}/evals.mdx +4 -4
  96. package/docs-site/zh/{concepts → explanation}/experiment.mdx +8 -8
  97. package/docs-site/zh/{concepts → explanation}/hitl.mdx +8 -8
  98. package/docs-site/zh/{concepts → explanation}/judge.mdx +5 -5
  99. package/docs-site/zh/{concepts → explanation}/overview.mdx +11 -11
  100. package/docs-site/zh/{guides → explanation}/runner.mdx +18 -8
  101. package/docs-site/zh/{concepts → explanation}/tier.mdx +6 -6
  102. package/docs-site/zh/{guides → how-to}/agent-feedback-loop.mdx +35 -33
  103. package/docs-site/zh/{guides → how-to}/authoring.mdx +35 -2
  104. package/docs-site/zh/{guides → how-to}/ci-integration.mdx +23 -12
  105. package/docs-site/zh/{guides → how-to}/connect-otel.mdx +6 -6
  106. package/docs-site/zh/{guides → how-to}/connect-your-agent.mdx +47 -21
  107. package/docs-site/zh/{guides → how-to}/custom-reports.mdx +34 -39
  108. package/docs-site/zh/{guides → how-to}/dataset-fanout.mdx +25 -3
  109. package/docs-site/zh/{guides → how-to}/experiments.mdx +12 -5
  110. package/docs-site/zh/how-to/publish-report.mdx +105 -0
  111. package/docs-site/zh/{guides → how-to}/reporters.mdx +2 -2
  112. package/docs-site/zh/{guides → how-to}/sandbox-agent.mdx +56 -7
  113. package/docs-site/zh/how-to/sandbox-providers.mdx +350 -0
  114. package/docs-site/zh/{guides → how-to}/scoring-guide.mdx +4 -4
  115. package/docs-site/zh/{guides → how-to}/viewing-results.mdx +82 -39
  116. package/docs-site/zh/{guides → how-to}/write-experiment.mdx +6 -4
  117. package/docs-site/zh/{guides → how-to}/write-send.mdx +30 -14
  118. package/docs-site/zh/index.mdx +24 -26
  119. package/docs-site/zh/introduction.mdx +8 -8
  120. package/docs-site/zh/reference/builtin-agents.mdx +32 -5
  121. package/docs-site/zh/reference/capabilities.mdx +8 -8
  122. package/docs-site/zh/reference/cli.mdx +40 -12
  123. package/docs-site/zh/reference/define-agent.mdx +58 -5
  124. package/docs-site/zh/reference/define-config.mdx +1 -1
  125. package/docs-site/zh/reference/define-eval.mdx +42 -9
  126. package/docs-site/zh/reference/events.mdx +3 -3
  127. package/docs-site/zh/reference/expect.mdx +26 -1
  128. package/docs-site/zh/{guides → reference}/official-adapters.mdx +32 -8
  129. package/docs-site/zh/{guides → reference}/report-components.mdx +45 -33
  130. package/docs-site/zh/{guides → reference}/results-data.mdx +21 -13
  131. package/docs-site/zh/troubleshooting/debug-sandbox.mdx +57 -0
  132. package/docs-site/zh/troubleshooting/debugging.mdx +212 -0
  133. package/docs-site/zh/{quickstart.mdx → tutorials/quickstart.mdx} +5 -17
  134. package/package.json +10 -2
  135. package/src/agents/ai-sdk-otel.test.ts +1 -0
  136. package/src/agents/ai-sdk.test.ts +3 -0
  137. package/src/agents/ai-sdk.ts +3 -0
  138. package/src/agents/bub-install-spec.test.ts +34 -0
  139. package/src/agents/bub-install-spec.ts +32 -0
  140. package/src/agents/bub.ts +31 -32
  141. package/src/agents/claude-code.test.ts +130 -9
  142. package/src/agents/claude-code.ts +76 -4
  143. package/src/agents/codex.test.ts +189 -40
  144. package/src/agents/codex.ts +155 -14
  145. package/src/agents/coding-cli-versions.test.ts +15 -0
  146. package/src/agents/coding-cli-versions.ts +3 -0
  147. package/src/agents/index.ts +13 -2
  148. package/src/agents/langgraph.test.ts +204 -0
  149. package/src/agents/langgraph.ts +495 -0
  150. package/src/agents/marketplace.ts +85 -0
  151. package/src/agents/native-config.test.ts +179 -0
  152. package/src/agents/native-config.ts +267 -0
  153. package/src/agents/openai-compat.test.ts +1 -0
  154. package/src/agents/openai-compat.ts +1 -1
  155. package/src/agents/openclaw.test.ts +31 -0
  156. package/src/agents/openclaw.ts +171 -0
  157. package/src/agents/plugin-config.test.ts +1 -0
  158. package/src/agents/sdk-streams.test.ts +79 -0
  159. package/src/agents/sdk-streams.ts +55 -10
  160. package/src/agents/skills.test.ts +1 -0
  161. package/src/agents/streaming.test.ts +3 -9
  162. package/src/agents/streaming.ts +2 -2
  163. package/src/agents/types.ts +71 -8
  164. package/src/agents/ui-message-stream.test.ts +3 -0
  165. package/src/cli.ts +446 -124
  166. package/src/context/context.test.ts +51 -12
  167. package/src/context/context.ts +162 -30
  168. package/src/context/session.test.ts +2 -1
  169. package/src/context/session.ts +115 -7
  170. package/src/context/types.ts +30 -12
  171. package/src/define.test.ts +13 -8
  172. package/src/define.ts +25 -4
  173. package/src/expect/index.ts +53 -23
  174. package/src/i18n/en.ts +81 -17
  175. package/src/i18n/zh-CN.ts +80 -17
  176. package/src/o11y/cost.test.ts +1 -0
  177. package/src/o11y/execution-tree.test.ts +1 -20
  178. package/src/o11y/otlp/mappers/claude-code.test.ts +1 -0
  179. package/src/o11y/otlp/parse.test.ts +1 -0
  180. package/src/o11y/otlp/turn-otel.test.ts +1 -0
  181. package/src/o11y/parsers/bub.test.ts +1 -0
  182. package/src/o11y/parsers/claude-code.test.ts +1 -34
  183. package/src/o11y/parsers/openclaw.test.ts +154 -0
  184. package/src/o11y/parsers/openclaw.ts +310 -0
  185. package/src/o11y/prices.json +746 -311
  186. package/src/o11y/tool-names.test.ts +1 -0
  187. package/src/o11y/types.ts +16 -2
  188. package/src/report/aggregate.ts +178 -61
  189. package/src/report/built-in/index.tsx +9 -0
  190. package/src/report/components.tsx +625 -279
  191. package/src/report/compute.ts +723 -491
  192. package/src/report/dual-render.test.tsx +741 -1024
  193. package/src/report/flag.ts +104 -12
  194. package/src/report/format.ts +32 -12
  195. package/src/report/index.ts +119 -46
  196. package/src/report/load.ts +3 -2
  197. package/src/report/locale.ts +136 -65
  198. package/src/report/metrics.ts +108 -25
  199. package/src/report/primitives.tsx +196 -45
  200. package/src/report/react/AttemptList.tsx +30 -43
  201. package/src/report/react/DeltaTable.tsx +63 -45
  202. package/src/report/react/EvalList.tsx +0 -0
  203. package/src/report/react/ExperimentComparison.tsx +73 -0
  204. package/src/report/react/ExperimentList.tsx +50 -32
  205. package/src/report/react/MetricBars.tsx +5 -4
  206. package/src/report/react/MetricLine.tsx +13 -8
  207. package/src/report/react/MetricMatrix.tsx +2 -2
  208. package/src/report/react/MetricScatter.tsx +86 -34
  209. package/src/report/react/MetricTable.tsx +4 -76
  210. package/src/report/react/ScopeSummary.tsx +86 -0
  211. package/src/report/react/Scoreboard.tsx +28 -10
  212. package/src/report/react/cell.tsx +2 -2
  213. package/src/report/react/chart-math.test.ts +85 -0
  214. package/src/report/react/chart-math.ts +101 -22
  215. package/src/report/react/enhance.js +89 -5
  216. package/src/report/react/fixtures.ts +114 -154
  217. package/src/report/react/index.tsx +24 -39
  218. package/src/report/react/render.test.tsx +138 -158
  219. package/src/report/react/styles.css +243 -82
  220. package/src/report/report.test.ts +779 -841
  221. package/src/report/report.ts +423 -41
  222. package/src/report/text/faces.ts +290 -193
  223. package/src/report/text/plot.ts +1 -1
  224. package/src/report/text/table.ts +44 -7
  225. package/src/report/tree.ts +362 -104
  226. package/src/report/types.ts +261 -271
  227. package/src/report/web.ts +63 -20
  228. package/src/results/annotated-source.test.ts +62 -9
  229. package/src/results/annotated-source.ts +64 -6
  230. package/src/results/attempt-evidence.test.ts +13 -11
  231. package/src/results/attempt-evidence.ts +20 -13
  232. package/src/results/attempt-source.ts +6 -3
  233. package/src/results/copy.ts +150 -60
  234. package/src/results/host-equivalence.test.ts +34 -20
  235. package/src/results/index.ts +12 -4
  236. package/src/results/locator.test.ts +1 -22
  237. package/src/results/open.ts +15 -5
  238. package/src/results/publish.ts +149 -0
  239. package/src/results/results.test.ts +89 -54
  240. package/src/results/select.ts +104 -34
  241. package/src/results/truncate.ts +90 -0
  242. package/src/results/types.ts +43 -14
  243. package/src/results/writer.ts +31 -13
  244. package/src/runner/attempt.test.ts +138 -7
  245. package/src/runner/attempt.ts +603 -104
  246. package/src/runner/discover.test.ts +47 -0
  247. package/src/runner/discover.ts +36 -2
  248. package/src/runner/eval-source.test.ts +1 -27
  249. package/src/runner/feedback/agent.test.ts +504 -0
  250. package/src/runner/feedback/agent.ts +409 -0
  251. package/src/runner/feedback/ci.test.ts +562 -0
  252. package/src/runner/feedback/ci.ts +401 -0
  253. package/src/runner/feedback/coordinator.test.ts +317 -0
  254. package/src/runner/feedback/coordinator.ts +397 -0
  255. package/src/runner/feedback/failure.ts +40 -0
  256. package/src/runner/feedback/human.test.ts +616 -0
  257. package/src/runner/feedback/human.ts +535 -0
  258. package/src/runner/feedback/index.ts +66 -0
  259. package/src/runner/feedback/io.ts +78 -0
  260. package/src/runner/feedback/profile.test.ts +50 -0
  261. package/src/runner/feedback/profile.ts +58 -0
  262. package/src/runner/feedback/reducer.test.ts +395 -0
  263. package/src/runner/feedback/reducer.ts +260 -0
  264. package/src/runner/feedback/renderer.ts +82 -0
  265. package/src/runner/feedback/sink.ts +203 -0
  266. package/src/runner/feedback/testing.ts +106 -0
  267. package/src/runner/ledger.test.ts +230 -0
  268. package/src/runner/ledger.ts +329 -0
  269. package/src/runner/report.test.ts +128 -3
  270. package/src/runner/report.ts +33 -9
  271. package/src/runner/reporters/artifacts.ts +8 -2
  272. package/src/runner/reporters/braintrust.test.ts +8 -7
  273. package/src/runner/reporters/braintrust.ts +9 -2
  274. package/src/runner/reporters/index.ts +2 -2
  275. package/src/runner/reporters/json.test.ts +162 -0
  276. package/src/runner/reporters/json.ts +35 -8
  277. package/src/runner/reporters/shared.ts +1 -5
  278. package/src/runner/run.test.ts +760 -3
  279. package/src/runner/run.ts +243 -37
  280. package/src/runner/sandbox-prep.ts +3 -42
  281. package/src/runner/timing.ts +158 -0
  282. package/src/runner/types.ts +518 -22
  283. package/src/sandbox/checkpoint.test.ts +55 -0
  284. package/src/sandbox/checkpoint.ts +29 -8
  285. package/src/sandbox/cli-commands.ts +407 -0
  286. package/src/sandbox/docker.ts +115 -16
  287. package/src/sandbox/e2b-agent-template.test.ts +56 -0
  288. package/src/sandbox/e2b-agent-template.ts +94 -0
  289. package/src/sandbox/e2b.ts +74 -9
  290. package/src/sandbox/errors.ts +111 -4
  291. package/src/sandbox/index.ts +2 -0
  292. package/src/sandbox/io-retry.test.ts +58 -0
  293. package/src/sandbox/io-retry.ts +45 -0
  294. package/src/sandbox/keep-registry.test.ts +86 -0
  295. package/src/sandbox/keep-registry.ts +142 -0
  296. package/src/sandbox/keep.ts +178 -0
  297. package/src/sandbox/paths.test.ts +1 -0
  298. package/src/sandbox/paths.ts +19 -8
  299. package/src/sandbox/registry.ts +20 -3
  300. package/src/sandbox/resolve.ts +76 -11
  301. package/src/sandbox/retry.test.ts +70 -0
  302. package/src/sandbox/retry.ts +46 -4
  303. package/src/sandbox/types.ts +44 -6
  304. package/src/sandbox/vercel.ts +43 -20
  305. package/src/scoring/collector.ts +60 -17
  306. package/src/scoring/coverage.ts +95 -0
  307. package/src/scoring/diff.ts +81 -0
  308. package/src/scoring/display.test.ts +121 -0
  309. package/src/scoring/display.ts +133 -0
  310. package/src/scoring/evidence.test.ts +189 -0
  311. package/src/scoring/judge.test.ts +142 -0
  312. package/src/scoring/judge.ts +15 -18
  313. package/src/scoring/scoped.ts +217 -50
  314. package/src/scoring/types.ts +117 -20
  315. package/src/scoring/verdict.ts +16 -4
  316. package/src/shared/aggregate.ts +8 -6
  317. package/src/shared/types.ts +31 -0
  318. package/src/show/compose.ts +50 -67
  319. package/src/show/index.ts +127 -56
  320. package/src/show/render.ts +662 -131
  321. package/src/show/report-host.test.ts +188 -0
  322. package/src/show/report-host.ts +375 -0
  323. package/src/show/show.test.ts +320 -54
  324. package/src/tty-line.ts +8 -26
  325. package/src/util.test.ts +1 -0
  326. package/src/util.ts +41 -0
  327. package/src/view/app/App.test.tsx +69 -0
  328. package/src/view/app/App.tsx +144 -48
  329. package/src/view/app/components/AttemptModal.tsx +423 -11
  330. package/src/view/app/components/CodeView.tsx +41 -14
  331. package/src/view/app/components/CopyControls.tsx +2 -2
  332. package/src/view/app/i18n.ts +37 -17
  333. package/src/view/app/lib/attempt-route.test.ts +1 -0
  334. package/src/view/app/lib/verdict.ts +7 -9
  335. package/src/view/app/main.tsx +13 -8
  336. package/src/view/app/pages/{RunsPage.tsx → AttemptsPage.tsx} +6 -6
  337. package/src/view/app/types.ts +4 -1
  338. package/src/view/artifact-serving.test.ts +2 -1
  339. package/src/view/client-dist/app.css +1 -1
  340. package/src/view/client-dist/app.js +17 -17
  341. package/src/view/data.test.ts +10 -3
  342. package/src/view/data.ts +155 -49
  343. package/src/view/index.ts +56 -41
  344. package/src/view/server.ts +37 -15
  345. package/src/view/shared/types.ts +34 -5
  346. package/src/view/styles.css +227 -0
  347. package/src/view/view-report.test.ts +167 -62
  348. package/dist/report/built-ins/experiment-comparison.d.ts +0 -1
  349. package/dist/report/built-ins/experiment-comparison.js +0 -13
  350. package/dist/report/built-ins/index.d.ts +0 -1
  351. package/dist/report/built-ins/index.js +0 -2
  352. package/dist/report/react/GroupSummary.d.ts +0 -8
  353. package/dist/report/react/GroupSummary.js +0 -8
  354. package/dist/report/react/RunOverview.d.ts +0 -8
  355. package/dist/report/react/RunOverview.js +0 -12
  356. package/docs-site/zh/example/ai-agent-application.mdx +0 -152
  357. package/docs-site/zh/example/claude-code-codex-plugin.mdx +0 -167
  358. package/docs-site/zh/example/claude-code-codex-skill.mdx +0 -152
  359. package/docs-site/zh/example/showcase.mdx +0 -39
  360. package/docs-site/zh/guides/publish-report.mdx +0 -91
  361. package/docs-site/zh/guides/sandbox-providers.mdx +0 -102
  362. package/src/report/built-in-user-parity.test.tsx +0 -640
  363. package/src/report/built-ins/experiment-comparison.tsx +0 -19
  364. package/src/report/built-ins/index.ts +0 -2
  365. package/src/report/react/GroupSummary.tsx +0 -66
  366. package/src/report/react/RunOverview.tsx +0 -109
  367. package/src/runner/reporters/console.ts +0 -70
  368. package/src/runner/reporters/live.test.ts +0 -56
  369. package/src/runner/reporters/live.ts +0 -247
  370. package/src/runner/reporters/quiet.test.ts +0 -66
  371. package/src/runner/reporters/quiet.ts +0 -49
  372. package/src/runner/reporters/table.ts +0 -277
  373. /package/docs-site/zh/{example/tier1-ai-sdk-v7.mdx → examples/integrations/ai-sdk-v7.mdx} +0 -0
  374. /package/docs-site/zh/{example/tier1-claude-sdk.mdx → examples/integrations/claude-sdk.mdx} +0 -0
  375. /package/docs-site/zh/{example/tier1-codex-sdk.mdx → examples/integrations/codex-sdk.mdx} +0 -0
  376. /package/docs-site/zh/{example/tier1-langgraph.mdx → examples/integrations/langgraph.mdx} +0 -0
  377. /package/docs-site/zh/{example/tier1-pi-sdk.mdx → examples/integrations/pi-sdk.mdx} +0 -0
  378. /package/docs-site/zh/{guides → how-to}/fixtures.mdx +0 -0
@@ -0,0 +1,154 @@
1
+ // cases: docs/engineering/unit-tests/adapters/cases.md
2
+ import { describe, expect, it } from "vitest";
3
+
4
+ import { parseOpenClawTranscript, parseOpenClaw, parseOpenClawRunJson } from "./openclaw.ts";
5
+
6
+ function jsonl(lines: unknown[]): string {
7
+ return lines.map((l) => JSON.stringify(l)).join("\n");
8
+ }
9
+
10
+ describe("parseOpenClawTranscript", () => {
11
+ it("空 / 无内容 → 空事件流,parseSuccess true", () => {
12
+ expect(parseOpenClawTranscript(undefined)).toEqual({
13
+ events: [],
14
+ usage: { inputTokens: 0, outputTokens: 0 },
15
+ compactions: 0,
16
+ parseSuccess: true,
17
+ });
18
+ expect(parseOpenClawTranscript(" \n ").events).toEqual([]);
19
+ });
20
+
21
+ it("assistant 的 text/thinking/toolCall parts + toolResult 按 call ID 配对,usage 逐消息累加", () => {
22
+ const raw = jsonl([
23
+ { role: "user", content: "改一下配置" },
24
+ {
25
+ role: "assistant",
26
+ content: [
27
+ { type: "thinking", thinking: "先看文件" },
28
+ { type: "text", text: "我来改" },
29
+ { type: "toolCall", id: "tc_1", name: "read", arguments: { path: "a.ts" } },
30
+ ],
31
+ usage: { input: 100, output: 20, cacheRead: 5, cacheWrite: 3, cost: { total: 0.01 } },
32
+ },
33
+ { role: "toolResult", toolCallId: "tc_1", content: [{ type: "text", text: "file body" }], isError: false },
34
+ {
35
+ role: "assistant",
36
+ content: [{ type: "text", text: "改好了" }],
37
+ usage: { input: 50, output: 10, cacheRead: 0, cacheWrite: 0, cost: { total: 0.002 } },
38
+ },
39
+ ]);
40
+
41
+ const parsed = parseOpenClawTranscript(raw);
42
+ expect(parsed.events).toEqual([
43
+ { type: "message", role: "user", text: "改一下配置" },
44
+ { type: "thinking", text: "先看文件" },
45
+ { type: "message", role: "assistant", text: "我来改" },
46
+ { type: "action.called", callId: "tc_1", name: "read", input: { path: "a.ts" }, tool: "file_read" },
47
+ { type: "action.result", callId: "tc_1", output: "file body", status: "completed" },
48
+ { type: "message", role: "assistant", text: "改好了" },
49
+ ]);
50
+ expect(parsed.usage).toEqual({
51
+ inputTokens: 150,
52
+ outputTokens: 30,
53
+ cacheReadTokens: 5,
54
+ cacheWriteTokens: 3,
55
+ requests: 2,
56
+ costUSD: 0.012,
57
+ });
58
+ expect(parsed.parseSuccess).toBe(true);
59
+ });
60
+
61
+ it("{ type: 'message', message: {...} } 包装条目与 snake_case 变体也认", () => {
62
+ const raw = jsonl([
63
+ {
64
+ type: "message",
65
+ message: {
66
+ role: "assistant",
67
+ content: [{ type: "tool_call", tool_call_id: "c9", toolName: "exec", input: { cmd: "ls" } }],
68
+ },
69
+ },
70
+ { type: "message", message: { role: "tool", tool_call_id: "c9", output: "a.txt", is_error: true } },
71
+ ]);
72
+ expect(parseOpenClaw(raw)).toEqual([
73
+ { type: "action.called", callId: "c9", name: "exec", input: { cmd: "ls" }, tool: "shell" },
74
+ { type: "action.result", callId: "c9", output: "a.txt", status: "failed" },
75
+ ]);
76
+ });
77
+
78
+ it("compaction 条目计数,error 条目落 error 事件", () => {
79
+ const parsed = parseOpenClawTranscript(
80
+ jsonl([
81
+ { type: "compaction", reason: "context-limit" },
82
+ { type: "error", message: "boom" },
83
+ ]),
84
+ );
85
+ expect(parsed.events).toEqual([
86
+ { type: "compaction", reason: "context-limit" },
87
+ { type: "error", message: "boom" },
88
+ ]);
89
+ expect(parsed.compactions).toBe(1);
90
+ });
91
+
92
+ it("坏 JSON 行不中断解析,标 parseSuccess: false 但保留其余行的事件", () => {
93
+ const raw = `{"role":"assistant","content":[{"type":"text","text":"hi"}]}\nnot-json\n`;
94
+ const parsed = parseOpenClawTranscript(raw);
95
+ expect(parsed.parseSuccess).toBe(false);
96
+ expect(parsed.events).toEqual([{ type: "message", role: "assistant", text: "hi" }]);
97
+ });
98
+
99
+ it("字符串形态的 arguments 先按 JSON 解;解不开原样保留", () => {
100
+ const raw = jsonl([
101
+ {
102
+ role: "assistant",
103
+ content: [{ type: "toolCall", id: "c1", name: "write", arguments: '{"path":"b.ts"}' }],
104
+ },
105
+ ]);
106
+ expect(parseOpenClaw(raw)).toEqual([
107
+ { type: "action.called", callId: "c1", name: "write", input: { path: "b.ts" }, tool: "file_write" },
108
+ ]);
109
+ });
110
+ });
111
+
112
+ describe("parseOpenClawRunJson", () => {
113
+ it("整段 pretty-print JSON 封包:text / sessionId / usage / failed", () => {
114
+ const stdout = JSON.stringify(
115
+ {
116
+ sessionId: "sess-42",
117
+ result: { text: "done" },
118
+ usage: { input_tokens: 10, output_tokens: 3 },
119
+ status: "ok",
120
+ },
121
+ null,
122
+ 2,
123
+ );
124
+ expect(parseOpenClawRunJson(stdout)).toEqual({
125
+ text: "done",
126
+ sessionId: "sess-42",
127
+ usage: { inputTokens: 10, outputTokens: 3 },
128
+ failed: false,
129
+ });
130
+ });
131
+
132
+ it("混日志行的 stdout:取最后一个完整 JSON 对象;payloads[].text 拼接", () => {
133
+ const stdout = [
134
+ "starting agent...",
135
+ JSON.stringify({ payloads: [{ text: "第一段" }, { text: "第二段" }], sessionKey: "k1" }),
136
+ ].join("\n");
137
+ expect(parseOpenClawRunJson(stdout)).toEqual({
138
+ text: "第一段\n第二段",
139
+ sessionId: "k1",
140
+ failed: false,
141
+ });
142
+ });
143
+
144
+ it("error 字段 / status=error / success=false → failed", () => {
145
+ expect(parseOpenClawRunJson(JSON.stringify({ error: "rate limited" })).failed).toBe(true);
146
+ expect(parseOpenClawRunJson(JSON.stringify({ status: "error" })).failed).toBe(true);
147
+ expect(parseOpenClawRunJson(JSON.stringify({ success: false })).failed).toBe(true);
148
+ });
149
+
150
+ it("解析不出 JSON → 空摘要,failed 交给调用方 exitCode", () => {
151
+ expect(parseOpenClawRunJson("plain text output")).toEqual({ failed: false });
152
+ expect(parseOpenClawRunJson(undefined)).toEqual({ failed: false });
153
+ });
154
+ });
@@ -0,0 +1,310 @@
1
+ // OpenClaw transcript / `agent --json` 解析器。OpenClaw 方言只住这里,不进 core
2
+ // (契约见 docs/feature/adapters/sdk/openclaw/README.md:字段事实以真实 CLI 与 transcript
3
+ // fixture 固定为准;fixture 未证明的形状这里只做防御式解析,认不出的行标 parseSuccess=false,
4
+ // 绝不从最终文本猜测工具轨迹)。
5
+ //
6
+ // 行为轨两个来源(collection.md 的通道优先级):
7
+ // 1. session transcript JSONL(~/.openclaw/agents/**/sessions/*.jsonl,pi-agent 系
8
+ // 消息格式:role/content parts,toolCall/toolResult 按 call ID 配对)→ parseOpenClawTranscript;
9
+ // 2. `openclaw agent --json` 的 stdout 结果封包(最终回复、session key、失败与 usage 摘要,
10
+ // 无完整工具轨迹)→ parseOpenClawRunJson,只在 transcript 拿不到时兜底。
11
+
12
+ import type { StreamEvent, Usage, ToolName, JsonValue } from "../../types.ts";
13
+ import type { ParsedTranscript } from "./index.ts";
14
+ import { GENERIC_VERB_ALIASES, normalizeToolName as normalizeShared } from "../tool-names.ts";
15
+
16
+ /**
17
+ * OpenClaw 特有工具别名(CLI 自己的 transcript 词汇,不会撞用户域名,裸动词可 opt-in)
18
+ * + 通用裸动词基表。
19
+ */
20
+ export const OPENCLAW_TOOL_ALIASES: Record<string, ToolName> = {
21
+ ...GENERIC_VERB_ALIASES,
22
+ read: "file_read",
23
+ write: "file_write",
24
+ edit: "file_edit",
25
+ process: "shell",
26
+ browser: "web_fetch",
27
+ };
28
+
29
+ function normalizeToolName(name: string): ToolName {
30
+ return normalizeShared(name, OPENCLAW_TOOL_ALIASES);
31
+ }
32
+
33
+ function get(obj: unknown, key: string): unknown {
34
+ return obj && typeof obj === "object" ? (obj as Record<string, unknown>)[key] : undefined;
35
+ }
36
+
37
+ function str(v: unknown): string | undefined {
38
+ return typeof v === "string" && v ? v : undefined;
39
+ }
40
+
41
+ function num(obj: unknown, ...keys: string[]): number {
42
+ for (const k of keys) {
43
+ const v = get(obj, k);
44
+ if (typeof v === "number" && Number.isFinite(v)) return v;
45
+ }
46
+ return 0;
47
+ }
48
+
49
+ function coerceArgs(value: unknown): JsonValue {
50
+ if (typeof value === "string") {
51
+ try {
52
+ return JSON.parse(value) as JsonValue;
53
+ } catch {
54
+ return value;
55
+ }
56
+ }
57
+ return (value ?? {}) as JsonValue;
58
+ }
59
+
60
+ /** content parts / 字符串 → 纯文本(text 类 part 拼接)。 */
61
+ function extractText(content: unknown): string {
62
+ if (typeof content === "string") return content;
63
+ if (Array.isArray(content)) {
64
+ const parts: string[] = [];
65
+ for (const b of content) {
66
+ if (typeof b === "string") parts.push(b);
67
+ else {
68
+ const t = get(b, "text") ?? get(b, "content");
69
+ if (typeof t === "string" && t) parts.push(t);
70
+ }
71
+ }
72
+ return parts.join("\n");
73
+ }
74
+ return "";
75
+ }
76
+
77
+ /**
78
+ * OpenClaw session transcript(JSONL,pi-agent 系消息格式)→ 标准事件流 + 用量 + 压缩计数。
79
+ * assistant 消息的 text / thinking / toolCall parts → message / thinking / action.called,
80
+ * toolResult 消息按 toolCallId 配对成 action.result(isError → failed),消息级 usage 逐条
81
+ * 累加,compaction 条目计入压缩次数。认不出 / 解析失败的行标 `parseSuccess: false` 并跳过,
82
+ * 不猜测——transcript 不完整时负断言不可信(契约见 sdk/openclaw/README.md)。
83
+ */
84
+ export function parseOpenClawTranscript(raw: string | undefined): ParsedTranscript {
85
+ const events: StreamEvent[] = [];
86
+ let inputTokens = 0;
87
+ let outputTokens = 0;
88
+ let cacheReadTokens = 0;
89
+ let cacheWriteTokens = 0;
90
+ let costUSD = 0;
91
+ let requests = 0;
92
+ let compactions = 0;
93
+ let parseSuccess = true;
94
+
95
+ if (!raw || !raw.trim()) {
96
+ return { events, usage: { inputTokens: 0, outputTokens: 0 }, compactions: 0, parseSuccess: true };
97
+ }
98
+
99
+ let synth = 0;
100
+ const nextSynthId = (): string => `oc_${++synth}`;
101
+
102
+ const addUsage = (usage: unknown): void => {
103
+ if (!usage || typeof usage !== "object") return;
104
+ // pi 系简写(input/output/cacheRead/cacheWrite/cost.total)与 snake_case 变体都认。
105
+ const input = num(usage, "input", "input_tokens", "inputTokens", "prompt_tokens");
106
+ const output = num(usage, "output", "output_tokens", "outputTokens", "completion_tokens");
107
+ const cacheRead = num(usage, "cacheRead", "cache_read_input_tokens", "cacheReadTokens");
108
+ const cacheWrite = num(usage, "cacheWrite", "cache_creation_input_tokens", "cacheWriteTokens");
109
+ const cost = num(usage, "cost") || num(get(usage, "cost"), "total");
110
+ if (input === 0 && output === 0 && cacheRead === 0 && cacheWrite === 0 && cost === 0) return;
111
+ inputTokens += input;
112
+ outputTokens += output;
113
+ cacheReadTokens += cacheRead;
114
+ cacheWriteTokens += cacheWrite;
115
+ costUSD += cost;
116
+ requests += 1;
117
+ };
118
+
119
+ for (const line of raw.split("\n")) {
120
+ const trimmed = line.trim();
121
+ if (!trimmed) continue;
122
+
123
+ let entry: unknown;
124
+ try {
125
+ entry = JSON.parse(trimmed);
126
+ } catch {
127
+ parseSuccess = false;
128
+ continue;
129
+ }
130
+
131
+ try {
132
+ // 会话条目可能带 { type: "message", message: {...} } 包装,也可能就是消息本体。
133
+ const wrappedMessage = get(entry, "message");
134
+ const entryType = get(entry, "type");
135
+ const msg =
136
+ wrappedMessage && typeof wrappedMessage === "object" ? wrappedMessage : entry;
137
+ const role = get(msg, "role");
138
+
139
+ if (entryType === "compaction" || entryType === "compact" || get(entry, "kind") === "compaction") {
140
+ compactions += 1;
141
+ events.push({ type: "compaction", reason: str(get(entry, "reason")) });
142
+ continue;
143
+ }
144
+ if (entryType === "error" && role === undefined) {
145
+ events.push({ type: "error", message: String(get(entry, "message") ?? get(entry, "error") ?? "error") });
146
+ continue;
147
+ }
148
+
149
+ if (role === "assistant") {
150
+ addUsage(get(msg, "usage"));
151
+ const content = get(msg, "content");
152
+ if (typeof content === "string") {
153
+ if (content) events.push({ type: "message", role: "assistant", text: content });
154
+ continue;
155
+ }
156
+ if (!Array.isArray(content)) continue;
157
+ for (const part of content) {
158
+ const partType = get(part, "type");
159
+ if (partType === "text") {
160
+ const text = str(get(part, "text"));
161
+ if (text) events.push({ type: "message", role: "assistant", text });
162
+ } else if (partType === "thinking" || partType === "reasoning") {
163
+ const text = str(get(part, "thinking")) ?? str(get(part, "reasoning")) ?? str(get(part, "text"));
164
+ if (text) events.push({ type: "thinking", text });
165
+ } else if (partType === "toolCall" || partType === "tool_call" || partType === "tool-call" || partType === "toolUse") {
166
+ const name = String(get(part, "name") ?? get(part, "toolName") ?? "unknown");
167
+ const callId = str(get(part, "id")) ?? str(get(part, "toolCallId")) ?? str(get(part, "tool_call_id")) ?? nextSynthId();
168
+ events.push({
169
+ type: "action.called",
170
+ callId,
171
+ name,
172
+ input: coerceArgs(get(part, "arguments") ?? get(part, "args") ?? get(part, "input")),
173
+ tool: normalizeToolName(name),
174
+ });
175
+ }
176
+ }
177
+ continue;
178
+ }
179
+
180
+ if (role === "user") {
181
+ const text = extractText(get(msg, "content") ?? get(msg, "text"));
182
+ if (text) events.push({ type: "message", role: "user", text });
183
+ continue;
184
+ }
185
+
186
+ if (role === "toolResult" || role === "tool") {
187
+ const callId =
188
+ str(get(msg, "toolCallId")) ?? str(get(msg, "tool_call_id")) ?? str(get(msg, "id")) ?? nextSynthId();
189
+ const isError = get(msg, "isError") === true || get(msg, "is_error") === true;
190
+ const output = get(msg, "output") ?? get(msg, "result") ?? get(msg, "content");
191
+ events.push({
192
+ type: "action.result",
193
+ callId,
194
+ output: (typeof output === "string" ? output : (extractText(output) || output)) as JsonValue,
195
+ status: isError ? "failed" : "completed",
196
+ });
197
+ continue;
198
+ }
199
+ // 其它条目(system prompt、订阅元数据等):无对应 StreamEvent。
200
+ } catch {
201
+ parseSuccess = false;
202
+ }
203
+ }
204
+
205
+ const usage: Usage = { inputTokens, outputTokens };
206
+ if (cacheReadTokens > 0) usage.cacheReadTokens = cacheReadTokens;
207
+ if (cacheWriteTokens > 0) usage.cacheWriteTokens = cacheWriteTokens;
208
+ if (requests > 0) usage.requests = requests;
209
+ if (costUSD > 0) usage.costUSD = costUSD;
210
+
211
+ return { events, usage, compactions, parseSuccess };
212
+ }
213
+
214
+ /** 便捷形态:只要 StreamEvent[]。 */
215
+ export function parseOpenClaw(raw: string | undefined): StreamEvent[] {
216
+ return parseOpenClawTranscript(raw).events;
217
+ }
218
+
219
+ /** `openclaw agent --json` 结果封包里抠出的摘要(无完整工具轨迹,只兜底)。 */
220
+ export interface OpenClawRunJson {
221
+ /** 最终回复文本(result.text / payloads[].text / reply 变体)。 */
222
+ text?: string;
223
+ /** 本次 run 的 session key(首轮 capture、后续 resume 用)。 */
224
+ sessionId?: string;
225
+ /** 封包报告的用量;没报就省略,不编造。 */
226
+ usage?: Usage;
227
+ /** 封包自报失败(error 字段 / status=error|failed / success=false)。 */
228
+ failed: boolean;
229
+ }
230
+
231
+ /**
232
+ * `openclaw agent --json` 的 stdout → 结果摘要。整段 stdout 先按单个 JSON 文档解析
233
+ * (含 pretty-print 多行),失败再逐行扫最后一个 JSON 对象(混了日志行的场景)。
234
+ * 完全解析不出时返回空摘要(`failed: false`)——失败判定交给调用方的 exitCode。
235
+ */
236
+ export function parseOpenClawRunJson(stdout: string | undefined): OpenClawRunJson {
237
+ const doc = parseJsonDocument(stdout);
238
+ if (!doc) return { failed: false };
239
+
240
+ const result = get(doc, "result");
241
+ const textFromPayloads = (payloads: unknown): string | undefined => {
242
+ if (!Array.isArray(payloads)) return undefined;
243
+ const texts = payloads.map((p) => str(get(p, "text"))).filter((t): t is string => !!t);
244
+ return texts.length ? texts.join("\n") : undefined;
245
+ };
246
+ const text =
247
+ str(get(result, "text")) ??
248
+ textFromPayloads(get(result, "payloads")) ??
249
+ textFromPayloads(get(doc, "payloads")) ??
250
+ str(get(doc, "text")) ??
251
+ str(get(doc, "reply")) ??
252
+ str(get(doc, "message"));
253
+
254
+ const sessionId =
255
+ str(get(doc, "sessionId")) ??
256
+ str(get(doc, "session_id")) ??
257
+ str(get(doc, "sessionKey")) ??
258
+ str(get(result, "sessionId")) ??
259
+ str(get(result, "session_id"));
260
+
261
+ const rawUsage = get(doc, "usage") ?? get(result, "usage") ?? get(get(doc, "meta"), "usage");
262
+ let usage: Usage | undefined;
263
+ if (rawUsage && typeof rawUsage === "object") {
264
+ const input = num(rawUsage, "input", "input_tokens", "inputTokens", "prompt_tokens");
265
+ const output = num(rawUsage, "output", "output_tokens", "outputTokens", "completion_tokens");
266
+ if (input > 0 || output > 0) {
267
+ usage = { inputTokens: input, outputTokens: output };
268
+ const cacheRead = num(rawUsage, "cacheRead", "cache_read_input_tokens", "cacheReadTokens");
269
+ if (cacheRead > 0) usage.cacheReadTokens = cacheRead;
270
+ const cost = num(rawUsage, "cost") || num(get(rawUsage, "cost"), "total");
271
+ if (cost > 0) usage.costUSD = cost;
272
+ }
273
+ }
274
+
275
+ const status = get(doc, "status");
276
+ const failed =
277
+ get(doc, "error") != null ||
278
+ get(doc, "success") === false ||
279
+ (typeof status === "string" && /^(error|failed)$/i.test(status));
280
+
281
+ return {
282
+ ...(text !== undefined ? { text } : {}),
283
+ ...(sessionId !== undefined ? { sessionId } : {}),
284
+ ...(usage !== undefined ? { usage } : {}),
285
+ failed,
286
+ };
287
+ }
288
+
289
+ function parseJsonDocument(stdout: string | undefined): Record<string, unknown> | undefined {
290
+ if (!stdout || !stdout.trim()) return undefined;
291
+ const trimmed = stdout.trim();
292
+ try {
293
+ const doc = JSON.parse(trimmed) as unknown;
294
+ if (doc && typeof doc === "object" && !Array.isArray(doc)) return doc as Record<string, unknown>;
295
+ } catch {
296
+ // 混日志行:逐行找最后一个完整 JSON 对象
297
+ }
298
+ const lines = trimmed.split("\n").reverse();
299
+ for (const line of lines) {
300
+ const l = line.trim();
301
+ if (!l.startsWith("{") || !l.endsWith("}")) continue;
302
+ try {
303
+ const doc = JSON.parse(l) as unknown;
304
+ if (doc && typeof doc === "object") return doc as Record<string, unknown>;
305
+ } catch {
306
+ // 继续往上找
307
+ }
308
+ }
309
+ return undefined;
310
+ }