@post-print/agent-test 1.1.9 → 2.0.1

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 (300) hide show
  1. package/README.md +18 -66
  2. package/dist/cli.d.ts +2 -64
  3. package/dist/cli.d.ts.map +1 -1
  4. package/dist/cli.js +26 -691
  5. package/dist/cli.js.map +1 -1
  6. package/dist/host-login.d.ts +1 -26
  7. package/dist/host-login.d.ts.map +1 -1
  8. package/dist/host-login.js +1 -92
  9. package/dist/host-login.js.map +1 -1
  10. package/dist/index.d.ts +6 -28
  11. package/dist/index.d.ts.map +1 -1
  12. package/dist/index.js +4 -26
  13. package/dist/index.js.map +1 -1
  14. package/dist/sdk/cli.d.ts +3 -0
  15. package/dist/sdk/cli.d.ts.map +1 -0
  16. package/dist/sdk/cli.js +55 -0
  17. package/dist/sdk/cli.js.map +1 -0
  18. package/dist/sdk/definitions.d.ts +22 -0
  19. package/dist/sdk/definitions.d.ts.map +1 -0
  20. package/dist/sdk/definitions.js +58 -0
  21. package/dist/sdk/definitions.js.map +1 -0
  22. package/dist/sdk/execution-reporter.d.ts +18 -0
  23. package/dist/sdk/execution-reporter.d.ts.map +1 -0
  24. package/dist/sdk/execution-reporter.js +171 -0
  25. package/dist/sdk/execution-reporter.js.map +1 -0
  26. package/dist/sdk/execution-runner.d.ts +14 -0
  27. package/dist/sdk/execution-runner.d.ts.map +1 -0
  28. package/dist/sdk/execution-runner.js +80 -0
  29. package/dist/sdk/execution-runner.js.map +1 -0
  30. package/dist/sdk/execution-store.d.ts +92 -0
  31. package/dist/sdk/execution-store.d.ts.map +1 -0
  32. package/dist/sdk/execution-store.js +357 -0
  33. package/dist/sdk/execution-store.js.map +1 -0
  34. package/dist/sdk/expect.d.ts +31 -0
  35. package/dist/sdk/expect.d.ts.map +1 -0
  36. package/dist/sdk/expect.js +96 -0
  37. package/dist/sdk/expect.js.map +1 -0
  38. package/dist/sdk/judge.d.ts +16 -0
  39. package/dist/sdk/judge.d.ts.map +1 -0
  40. package/dist/sdk/judge.js +82 -0
  41. package/dist/sdk/judge.js.map +1 -0
  42. package/dist/sdk/metrics.d.ts +3 -0
  43. package/dist/sdk/metrics.d.ts.map +1 -0
  44. package/dist/sdk/metrics.js +24 -0
  45. package/dist/sdk/metrics.js.map +1 -0
  46. package/dist/sdk/playwright-cli.d.ts +2 -0
  47. package/dist/sdk/playwright-cli.d.ts.map +1 -0
  48. package/dist/sdk/playwright-cli.js +6 -0
  49. package/dist/sdk/playwright-cli.js.map +1 -0
  50. package/dist/sdk/reporter.d.ts +12 -0
  51. package/dist/sdk/reporter.d.ts.map +1 -0
  52. package/dist/sdk/reporter.js +98 -0
  53. package/dist/sdk/reporter.js.map +1 -0
  54. package/dist/sdk/runtime.d.ts +33 -0
  55. package/dist/sdk/runtime.d.ts.map +1 -0
  56. package/dist/sdk/runtime.js +197 -0
  57. package/dist/sdk/runtime.js.map +1 -0
  58. package/dist/sdk/test.d.ts +36 -0
  59. package/dist/sdk/test.d.ts.map +1 -0
  60. package/dist/sdk/test.js +133 -0
  61. package/dist/sdk/test.js.map +1 -0
  62. package/dist/sdk/types.d.ts +76 -0
  63. package/dist/sdk/types.d.ts.map +1 -0
  64. package/dist/sdk/types.js.map +1 -0
  65. package/dist/sdk/usage.d.ts +10 -0
  66. package/dist/sdk/usage.d.ts.map +1 -0
  67. package/dist/sdk/usage.js +13 -0
  68. package/dist/sdk/usage.js.map +1 -0
  69. package/dist/sdk/viewer.d.ts +5 -0
  70. package/dist/sdk/viewer.d.ts.map +1 -0
  71. package/dist/sdk/viewer.js +74 -0
  72. package/dist/sdk/viewer.js.map +1 -0
  73. package/dist/sdk/workspace.d.ts +33 -0
  74. package/dist/sdk/workspace.d.ts.map +1 -0
  75. package/dist/sdk/workspace.js +100 -0
  76. package/dist/sdk/workspace.js.map +1 -0
  77. package/dist/viewer/bootstrap.d.ts +10 -0
  78. package/dist/viewer/bootstrap.d.ts.map +1 -0
  79. package/dist/viewer/bootstrap.js +2 -0
  80. package/dist/viewer/bootstrap.js.map +1 -0
  81. package/dist/viewer/client-entry.d.ts +2 -0
  82. package/dist/viewer/client-entry.d.ts.map +1 -0
  83. package/dist/viewer/client-entry.js +31 -0
  84. package/dist/viewer/client-entry.js.map +1 -0
  85. package/dist/viewer/generated-api.d.ts +15 -0
  86. package/dist/viewer/generated-api.d.ts.map +1 -0
  87. package/dist/viewer/generated-api.js +42 -0
  88. package/dist/viewer/generated-api.js.map +1 -0
  89. package/dist/viewer/generated-contracts.d.ts +13 -0
  90. package/dist/viewer/generated-contracts.d.ts.map +1 -0
  91. package/dist/viewer/generated-contracts.js +12 -0
  92. package/dist/viewer/generated-contracts.js.map +1 -0
  93. package/dist/viewer/main.js +64 -3
  94. package/dist/viewer/main.js.map +1 -1
  95. package/dist/viewer/page.d.ts +2 -4
  96. package/dist/viewer/page.d.ts.map +1 -1
  97. package/dist/viewer/page.js +18 -194
  98. package/dist/viewer/page.js.map +1 -1
  99. package/dist/viewer/presentation.d.ts +8 -0
  100. package/dist/viewer/presentation.d.ts.map +1 -0
  101. package/dist/viewer/presentation.js +45 -0
  102. package/dist/viewer/presentation.js.map +1 -0
  103. package/dist/viewer/protocol.d.ts +21 -0
  104. package/dist/viewer/protocol.d.ts.map +1 -0
  105. package/dist/viewer/protocol.js +6 -0
  106. package/dist/viewer/protocol.js.map +1 -0
  107. package/dist/viewer/server.d.ts +7 -11
  108. package/dist/viewer/server.d.ts.map +1 -1
  109. package/dist/viewer/server.js +314 -175
  110. package/dist/viewer/server.js.map +1 -1
  111. package/dist/viewer/styles.d.ts +2 -0
  112. package/dist/viewer/styles.d.ts.map +1 -0
  113. package/dist/viewer/styles.js +97 -0
  114. package/dist/viewer/styles.js.map +1 -0
  115. package/dist/viewer/stylex.css +1322 -0
  116. package/dist/viewer/test-catalog.d.ts +43 -0
  117. package/dist/viewer/test-catalog.d.ts.map +1 -0
  118. package/dist/viewer/test-catalog.js +27 -0
  119. package/dist/viewer/test-catalog.js.map +1 -0
  120. package/dist/viewer/v2-client.d.ts +7 -0
  121. package/dist/viewer/v2-client.d.ts.map +1 -0
  122. package/dist/viewer/v2-client.js +1862 -0
  123. package/dist/viewer/v2-client.js.map +1 -0
  124. package/package.json +19 -9
  125. package/dist/check.d.ts +0 -38
  126. package/dist/check.d.ts.map +0 -1
  127. package/dist/check.js +0 -169
  128. package/dist/check.js.map +0 -1
  129. package/dist/command-allowlist.d.ts +0 -9
  130. package/dist/command-allowlist.d.ts.map +0 -1
  131. package/dist/command-allowlist.js +0 -29
  132. package/dist/command-allowlist.js.map +0 -1
  133. package/dist/compare-scenario.d.ts +0 -65
  134. package/dist/compare-scenario.d.ts.map +0 -1
  135. package/dist/compare-scenario.js +0 -328
  136. package/dist/compare-scenario.js.map +0 -1
  137. package/dist/debug-bundle.d.ts +0 -81
  138. package/dist/debug-bundle.d.ts.map +0 -1
  139. package/dist/debug-bundle.js +0 -392
  140. package/dist/debug-bundle.js.map +0 -1
  141. package/dist/discover-suites.d.ts +0 -2
  142. package/dist/discover-suites.d.ts.map +0 -1
  143. package/dist/discover-suites.js +0 -19
  144. package/dist/discover-suites.js.map +0 -1
  145. package/dist/doctor.d.ts +0 -24
  146. package/dist/doctor.d.ts.map +0 -1
  147. package/dist/doctor.js +0 -272
  148. package/dist/doctor.js.map +0 -1
  149. package/dist/expect.d.ts +0 -59
  150. package/dist/expect.d.ts.map +0 -1
  151. package/dist/expect.js +0 -518
  152. package/dist/expect.js.map +0 -1
  153. package/dist/failures.d.ts +0 -4
  154. package/dist/failures.d.ts.map +0 -1
  155. package/dist/failures.js +0 -7
  156. package/dist/failures.js.map +0 -1
  157. package/dist/host-log.d.ts +0 -17
  158. package/dist/host-log.d.ts.map +0 -1
  159. package/dist/host-log.js +0 -154
  160. package/dist/host-log.js.map +0 -1
  161. package/dist/hosts.d.ts +0 -22
  162. package/dist/hosts.d.ts.map +0 -1
  163. package/dist/hosts.js +0 -84
  164. package/dist/hosts.js.map +0 -1
  165. package/dist/html-report.d.ts +0 -18
  166. package/dist/html-report.d.ts.map +0 -1
  167. package/dist/html-report.js +0 -1276
  168. package/dist/html-report.js.map +0 -1
  169. package/dist/live-isolation.d.ts +0 -82
  170. package/dist/live-isolation.d.ts.map +0 -1
  171. package/dist/live-isolation.js +0 -399
  172. package/dist/live-isolation.js.map +0 -1
  173. package/dist/live-timeout.d.ts +0 -11
  174. package/dist/live-timeout.d.ts.map +0 -1
  175. package/dist/live-timeout.js +0 -29
  176. package/dist/live-timeout.js.map +0 -1
  177. package/dist/load-adapters.d.ts +0 -13
  178. package/dist/load-adapters.d.ts.map +0 -1
  179. package/dist/load-adapters.js +0 -76
  180. package/dist/load-adapters.js.map +0 -1
  181. package/dist/load-suite.d.ts +0 -22
  182. package/dist/load-suite.d.ts.map +0 -1
  183. package/dist/load-suite.js +0 -154
  184. package/dist/load-suite.js.map +0 -1
  185. package/dist/mcp-config.d.ts +0 -13
  186. package/dist/mcp-config.d.ts.map +0 -1
  187. package/dist/mcp-config.js +0 -121
  188. package/dist/mcp-config.js.map +0 -1
  189. package/dist/preflight.d.ts +0 -3
  190. package/dist/preflight.d.ts.map +0 -1
  191. package/dist/preflight.js +0 -17
  192. package/dist/preflight.js.map +0 -1
  193. package/dist/progress.d.ts +0 -24
  194. package/dist/progress.d.ts.map +0 -1
  195. package/dist/progress.js +0 -111
  196. package/dist/progress.js.map +0 -1
  197. package/dist/record-trace.d.ts +0 -64
  198. package/dist/record-trace.d.ts.map +0 -1
  199. package/dist/record-trace.js +0 -124
  200. package/dist/record-trace.js.map +0 -1
  201. package/dist/report-preview-server.d.ts +0 -2
  202. package/dist/report-preview-server.d.ts.map +0 -1
  203. package/dist/report-preview-server.js +0 -24
  204. package/dist/report-preview-server.js.map +0 -1
  205. package/dist/report-preview.d.ts +0 -35
  206. package/dist/report-preview.d.ts.map +0 -1
  207. package/dist/report-preview.js +0 -205
  208. package/dist/report-preview.js.map +0 -1
  209. package/dist/run-suite.d.ts +0 -139
  210. package/dist/run-suite.d.ts.map +0 -1
  211. package/dist/run-suite.js +0 -1688
  212. package/dist/run-suite.js.map +0 -1
  213. package/dist/scenario-finalizer.d.ts +0 -27
  214. package/dist/scenario-finalizer.d.ts.map +0 -1
  215. package/dist/scenario-finalizer.js +0 -39
  216. package/dist/scenario-finalizer.js.map +0 -1
  217. package/dist/scenario-retry.d.ts +0 -15
  218. package/dist/scenario-retry.d.ts.map +0 -1
  219. package/dist/scenario-retry.js +0 -52
  220. package/dist/scenario-retry.js.map +0 -1
  221. package/dist/scenario-seed.d.ts +0 -13
  222. package/dist/scenario-seed.d.ts.map +0 -1
  223. package/dist/scenario-seed.js +0 -72
  224. package/dist/scenario-seed.js.map +0 -1
  225. package/dist/scenario-story.d.ts +0 -41
  226. package/dist/scenario-story.d.ts.map +0 -1
  227. package/dist/scenario-story.js +0 -459
  228. package/dist/scenario-story.js.map +0 -1
  229. package/dist/scenario-usage.d.ts +0 -15
  230. package/dist/scenario-usage.d.ts.map +0 -1
  231. package/dist/scenario-usage.js +0 -18
  232. package/dist/scenario-usage.js.map +0 -1
  233. package/dist/suite-summary.d.ts +0 -16
  234. package/dist/suite-summary.d.ts.map +0 -1
  235. package/dist/suite-summary.js +0 -218
  236. package/dist/suite-summary.js.map +0 -1
  237. package/dist/theme.d.ts +0 -96
  238. package/dist/theme.d.ts.map +0 -1
  239. package/dist/theme.js +0 -446
  240. package/dist/theme.js.map +0 -1
  241. package/dist/types.d.ts +0 -364
  242. package/dist/types.d.ts.map +0 -1
  243. package/dist/types.js.map +0 -1
  244. package/dist/validate-seeds.d.ts +0 -20
  245. package/dist/validate-seeds.d.ts.map +0 -1
  246. package/dist/validate-seeds.js +0 -72
  247. package/dist/validate-seeds.js.map +0 -1
  248. package/dist/validate-suite.d.ts +0 -24
  249. package/dist/validate-suite.d.ts.map +0 -1
  250. package/dist/validate-suite.js +0 -542
  251. package/dist/validate-suite.js.map +0 -1
  252. package/dist/viewer/catalog.d.ts +0 -72
  253. package/dist/viewer/catalog.d.ts.map +0 -1
  254. package/dist/viewer/catalog.js +0 -154
  255. package/dist/viewer/catalog.js.map +0 -1
  256. package/dist/viewer/cli.d.ts +0 -17
  257. package/dist/viewer/cli.d.ts.map +0 -1
  258. package/dist/viewer/cli.js +0 -38
  259. package/dist/viewer/cli.js.map +0 -1
  260. package/dist/viewer/client.d.ts +0 -7
  261. package/dist/viewer/client.d.ts.map +0 -1
  262. package/dist/viewer/client.js +0 -14
  263. package/dist/viewer/client.js.map +0 -1
  264. package/dist/viewer/compare-metrics.d.ts +0 -30
  265. package/dist/viewer/compare-metrics.d.ts.map +0 -1
  266. package/dist/viewer/compare-metrics.js +0 -101
  267. package/dist/viewer/compare-metrics.js.map +0 -1
  268. package/dist/viewer/context-files.d.ts +0 -8
  269. package/dist/viewer/context-files.d.ts.map +0 -1
  270. package/dist/viewer/context-files.js +0 -57
  271. package/dist/viewer/context-files.js.map +0 -1
  272. package/dist/viewer/emit.d.ts +0 -7
  273. package/dist/viewer/emit.d.ts.map +0 -1
  274. package/dist/viewer/emit.js +0 -25
  275. package/dist/viewer/emit.js.map +0 -1
  276. package/dist/viewer/events.d.ts +0 -118
  277. package/dist/viewer/events.d.ts.map +0 -1
  278. package/dist/viewer/events.js +0 -51
  279. package/dist/viewer/events.js.map +0 -1
  280. package/dist/viewer/live-runner.d.ts +0 -27
  281. package/dist/viewer/live-runner.d.ts.map +0 -1
  282. package/dist/viewer/live-runner.js +0 -166
  283. package/dist/viewer/live-runner.js.map +0 -1
  284. package/dist/viewer/run-controller.d.ts +0 -41
  285. package/dist/viewer/run-controller.d.ts.map +0 -1
  286. package/dist/viewer/run-controller.js +0 -426
  287. package/dist/viewer/run-controller.js.map +0 -1
  288. package/dist/viewer/session-server.d.ts +0 -2
  289. package/dist/viewer/session-server.d.ts.map +0 -1
  290. package/dist/viewer/session-server.js +0 -71
  291. package/dist/viewer/session-server.js.map +0 -1
  292. package/dist/warnings.d.ts +0 -3
  293. package/dist/warnings.d.ts.map +0 -1
  294. package/dist/warnings.js +0 -13
  295. package/dist/warnings.js.map +0 -1
  296. package/dist/worker-pool.d.ts +0 -8
  297. package/dist/worker-pool.d.ts.map +0 -1
  298. package/dist/worker-pool.js +0 -26
  299. package/dist/worker-pool.js.map +0 -1
  300. /package/dist/{types.js → sdk/types.js} +0 -0
@@ -1,27 +0,0 @@
1
- import type { HostAuthMode } from "@post-print/agent-harness";
2
- import { type SpawnLiveScenarioOptions, type SpawnLiveScenarioResult } from "../live-isolation.js";
3
- import type { ViewerJob } from "./catalog.js";
4
- import type { ViewerRunner } from "./run-controller.js";
5
- export interface LiveViewerRunnerOptions {
6
- /** CLI entrypoint to use for isolated children when the viewer runs in a detached server. */
7
- cliPath?: string;
8
- cwd: string;
9
- suitesDir: string;
10
- rubricsDir?: string;
11
- judge?: boolean;
12
- timeoutMs?: number;
13
- authMode?: HostAuthMode;
14
- adapterModules?: string[];
15
- worktree?: boolean;
16
- keepRecordings?: boolean;
17
- allowUserInput?: boolean;
18
- debug?: boolean;
19
- debugDir?: string;
20
- /** Override host auth probe. Tests inject this. */
21
- missingAuth?: (host: ViewerJob["host"]) => string | undefined;
22
- /** Override isolated spawn. Tests inject this. */
23
- spawnLiveScenario?: (options: SpawnLiveScenarioOptions) => Promise<SpawnLiveScenarioResult>;
24
- }
25
- /** Spawn one isolated child per viewer job. Forward NDJSON events to the UI. */
26
- export declare function createLiveViewerRunner(options: LiveViewerRunnerOptions): ViewerRunner;
27
- //# sourceMappingURL=live-runner.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"live-runner.d.ts","sourceRoot":"","sources":["../../src/viewer/live-runner.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,2BAA2B,CAAC;AAG9D,OAAO,EAEN,KAAK,wBAAwB,EAC7B,KAAK,uBAAuB,EAG5B,MAAM,sBAAsB,CAAC;AAO9B,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,cAAc,CAAC;AAE9C,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AAExD,MAAM,WAAW,uBAAuB;IACvC,6FAA6F;IAC7F,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,GAAG,EAAE,MAAM,CAAC;IACZ,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,cAAc,CAAC,EAAE,OAAO,CAAC;IACzB,KAAK,CAAC,EAAE,OAAO,CAAC;IAChB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,mDAAmD;IACnD,WAAW,CAAC,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,MAAM,CAAC,KAAK,MAAM,GAAG,SAAS,CAAC;IAC9D,kDAAkD;IAClD,iBAAiB,CAAC,EAAE,CAAC,OAAO,EAAE,wBAAwB,KAAK,OAAO,CAAC,uBAAuB,CAAC,CAAC;CAC5F;AAWD,gFAAgF;AAChF,wBAAgB,sBAAsB,CAAC,OAAO,EAAE,uBAAuB,GAAG,YAAY,CAuJrF"}
@@ -1,166 +0,0 @@
1
- import { missingAgentAuth } from "../doctor.js";
2
- import { killActiveLiveChildren, spawnLiveScenario, subprocessFailureMessage, } from "../live-isolation.js";
3
- import { createLiveStagingSessionId, getStagingResultPath, loadStagingResult, } from "../record-trace.js";
4
- import { finalizeScheduledCompareScenario, finalizeScheduledScenario } from "../run-suite.js";
5
- function envelope(job) {
6
- return {
7
- suite: job.suite,
8
- scenario: job.scenario,
9
- host: job.host,
10
- ...(job.arm ? { arm: job.arm } : {}),
11
- };
12
- }
13
- /** Spawn one isolated child per viewer job. Forward NDJSON events to the UI. */
14
- export function createLiveViewerRunner(options) {
15
- return {
16
- finalizeScenario: (input, emit) => finalizeScheduledScenario({
17
- ...input,
18
- cwd: options.cwd,
19
- judge: options.judge,
20
- onJudgeEvent: (event) => {
21
- const cell = { suite: input.suite, scenario: input.scenario.name, host: input.host };
22
- if (event.type === "criterion_started") {
23
- emit({ type: "judge_started", ...cell, id: event.id, question: event.question });
24
- }
25
- else if (event.type === "text") {
26
- emit({
27
- type: "judge_text",
28
- ...cell,
29
- id: event.id,
30
- question: event.question,
31
- text: event.text,
32
- });
33
- }
34
- else {
35
- emit({
36
- type: "judge",
37
- ...cell,
38
- verdicts: [
39
- {
40
- id: event.verdict.id,
41
- question: event.question,
42
- pass: event.verdict.pass,
43
- rationale: event.verdict.rationale,
44
- evidence: event.verdict.evidence,
45
- },
46
- ],
47
- });
48
- }
49
- },
50
- }),
51
- finalizeCompare: (input) => finalizeScheduledCompareScenario({
52
- ...input,
53
- cwd: options.cwd,
54
- judge: options.judge,
55
- }),
56
- async runJob(job, emit, signal) {
57
- const cell = envelope(job);
58
- const emitFailure = (message) => {
59
- emit({ type: "error", message, ...cell });
60
- emit({
61
- type: "cell_finished",
62
- ...cell,
63
- passed: false,
64
- durationMs: 0,
65
- failures: [{ matcher: "liveScenario", message }],
66
- });
67
- emit({
68
- type: "scenario_result",
69
- ...cell,
70
- result: {
71
- suite: job.suite,
72
- scenario: job.scenario,
73
- prompt: job.prompt,
74
- passed: false,
75
- durationMs: 0,
76
- failures: [{ matcher: "liveScenario", message, category: "agent_runtime" }],
77
- },
78
- });
79
- };
80
- emit({ type: "cell_started", ...cell });
81
- emit({ type: "status", text: "Starting isolated child.", ...cell });
82
- const missing = (options.missingAuth ?? missingAgentAuth)(job.host);
83
- if (missing) {
84
- emitFailure(missing);
85
- return;
86
- }
87
- if (signal.aborted) {
88
- return;
89
- }
90
- const onAbort = () => {
91
- killActiveLiveChildren();
92
- };
93
- signal.addEventListener("abort", onAbort, { once: true });
94
- const spawn = options.spawnLiveScenario ?? spawnLiveScenario;
95
- let finished = false;
96
- let deferredScenarioResult;
97
- const stagingSessionId = createLiveStagingSessionId();
98
- try {
99
- const result = await spawn({
100
- cliPath: options.cliPath,
101
- cwd: options.cwd,
102
- suiteName: job.suite,
103
- scenarioName: job.scenario,
104
- suitesDir: options.suitesDir,
105
- rubricsDir: options.rubricsDir,
106
- suiteFilter: job.suite,
107
- host: job.host,
108
- compareArm: job.arm,
109
- stagingSessionId,
110
- judge: job.arm ? false : options.judge,
111
- timeoutMs: options.timeoutMs,
112
- noTimeout: options.timeoutMs === 0,
113
- authMode: options.authMode,
114
- adapterModules: options.adapterModules,
115
- worktree: options.worktree,
116
- keepRecordings: options.keepRecordings,
117
- allowUserInput: options.allowUserInput,
118
- debug: options.debug,
119
- debugDir: options.debugDir,
120
- signal,
121
- onViewerEvent: (event) => {
122
- if (event.type === "cell_finished") {
123
- finished = true;
124
- }
125
- if (job.arm && event.type === "scenario_result") {
126
- deferredScenarioResult = event;
127
- return;
128
- }
129
- if (signal.aborted) {
130
- return;
131
- }
132
- emit(event);
133
- },
134
- });
135
- if (signal.aborted) {
136
- return;
137
- }
138
- if (job.arm && deferredScenarioResult) {
139
- const sidecar = await loadStagingResult(getStagingResultPath(stagingSessionId, job.suite, job.scenario, job.arm));
140
- if (sidecar?.judgeWorkspace) {
141
- Object.defineProperty(deferredScenarioResult.result, "judgeWorkspace", {
142
- value: sidecar.judgeWorkspace,
143
- enumerable: false,
144
- });
145
- }
146
- emit(deferredScenarioResult);
147
- }
148
- if (!finished && result.exitCode !== 0) {
149
- const message = subprocessFailureMessage(result.exitCode, result.stderr);
150
- emitFailure(message);
151
- }
152
- }
153
- catch (error) {
154
- if (signal.aborted) {
155
- return;
156
- }
157
- const message = error instanceof Error ? error.message : String(error);
158
- emitFailure(message);
159
- }
160
- finally {
161
- signal.removeEventListener("abort", onAbort);
162
- }
163
- },
164
- };
165
- }
166
- //# sourceMappingURL=live-runner.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"live-runner.js","sourceRoot":"","sources":["../../src/viewer/live-runner.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,gBAAgB,EAAE,MAAM,cAAc,CAAC;AAChD,OAAO,EACN,sBAAsB,EAGtB,iBAAiB,EACjB,wBAAwB,GACxB,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EACN,0BAA0B,EAC1B,oBAAoB,EACpB,iBAAiB,GACjB,MAAM,oBAAoB,CAAC;AAC5B,OAAO,EAAE,gCAAgC,EAAE,yBAAyB,EAAE,MAAM,iBAAiB,CAAC;AA0B9F,SAAS,QAAQ,CAAC,GAAc;IAC/B,OAAO;QACN,KAAK,EAAE,GAAG,CAAC,KAAK;QAChB,QAAQ,EAAE,GAAG,CAAC,QAAQ;QACtB,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACpC,CAAC;AACH,CAAC;AAED,gFAAgF;AAChF,MAAM,UAAU,sBAAsB,CAAC,OAAgC;IACtE,OAAO;QACN,gBAAgB,EAAE,CAAC,KAAK,EAAE,IAAI,EAAE,EAAE,CACjC,yBAAyB,CAAC;YACzB,GAAG,KAAK;YACR,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,KAAK,EAAE,OAAO,CAAC,KAAK;YACpB,YAAY,EAAE,CAAC,KAAK,EAAE,EAAE;gBACvB,MAAM,IAAI,GAAG,EAAE,KAAK,EAAE,KAAK,CAAC,KAAK,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,CAAC,IAAI,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,EAAE,CAAC;gBACrF,IAAI,KAAK,CAAC,IAAI,KAAK,mBAAmB,EAAE,CAAC;oBACxC,IAAI,CAAC,EAAE,IAAI,EAAE,eAAe,EAAE,GAAG,IAAI,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,CAAC,CAAC;gBAClF,CAAC;qBAAM,IAAI,KAAK,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;oBAClC,IAAI,CAAC;wBACJ,IAAI,EAAE,YAAY;wBAClB,GAAG,IAAI;wBACP,EAAE,EAAE,KAAK,CAAC,EAAE;wBACZ,QAAQ,EAAE,KAAK,CAAC,QAAQ;wBACxB,IAAI,EAAE,KAAK,CAAC,IAAI;qBAChB,CAAC,CAAC;gBACJ,CAAC;qBAAM,CAAC;oBACP,IAAI,CAAC;wBACJ,IAAI,EAAE,OAAO;wBACb,GAAG,IAAI;wBACP,QAAQ,EAAE;4BACT;gCACC,EAAE,EAAE,KAAK,CAAC,OAAO,CAAC,EAAE;gCACpB,QAAQ,EAAE,KAAK,CAAC,QAAQ;gCACxB,IAAI,EAAE,KAAK,CAAC,OAAO,CAAC,IAAI;gCACxB,SAAS,EAAE,KAAK,CAAC,OAAO,CAAC,SAAS;gCAClC,QAAQ,EAAE,KAAK,CAAC,OAAO,CAAC,QAAQ;6BAChC;yBACD;qBACD,CAAC,CAAC;gBACJ,CAAC;YACF,CAAC;SACD,CAAC;QACH,eAAe,EAAE,CAAC,KAAK,EAAE,EAAE,CAC1B,gCAAgC,CAAC;YAChC,GAAG,KAAK;YACR,GAAG,EAAE,OAAO,CAAC,GAAG;YAChB,KAAK,EAAE,OAAO,CAAC,KAAK;SACpB,CAAC;QACH,KAAK,CAAC,MAAM,CAAC,GAAG,EAAE,IAAI,EAAE,MAAM;YAC7B,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,CAAC;YAC3B,MAAM,WAAW,GAAG,CAAC,OAAe,EAAQ,EAAE;gBAC7C,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;gBAC1C,IAAI,CAAC;oBACJ,IAAI,EAAE,eAAe;oBACrB,GAAG,IAAI;oBACP,MAAM,EAAE,KAAK;oBACb,UAAU,EAAE,CAAC;oBACb,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,CAAC;iBAChD,CAAC,CAAC;gBACH,IAAI,CAAC;oBACJ,IAAI,EAAE,iBAAiB;oBACvB,GAAG,IAAI;oBACP,MAAM,EAAE;wBACP,KAAK,EAAE,GAAG,CAAC,KAAK;wBAChB,QAAQ,EAAE,GAAG,CAAC,QAAQ;wBACtB,MAAM,EAAE,GAAG,CAAC,MAAM;wBAClB,MAAM,EAAE,KAAK;wBACb,UAAU,EAAE,CAAC;wBACb,QAAQ,EAAE,CAAC,EAAE,OAAO,EAAE,cAAc,EAAE,OAAO,EAAE,QAAQ,EAAE,eAAe,EAAE,CAAC;qBAC3E;iBACD,CAAC,CAAC;YACJ,CAAC,CAAC;YACF,IAAI,CAAC,EAAE,IAAI,EAAE,cAAc,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;YACxC,IAAI,CAAC,EAAE,IAAI,EAAE,QAAQ,EAAE,IAAI,EAAE,0BAA0B,EAAE,GAAG,IAAI,EAAE,CAAC,CAAC;YACpE,MAAM,OAAO,GAAG,CAAC,OAAO,CAAC,WAAW,IAAI,gBAAgB,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;YACpE,IAAI,OAAO,EAAE,CAAC;gBACb,WAAW,CAAC,OAAO,CAAC,CAAC;gBACrB,OAAO;YACR,CAAC;YACD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;gBACpB,OAAO;YACR,CAAC;YACD,MAAM,OAAO,GAAG,GAAS,EAAE;gBAC1B,sBAAsB,EAAE,CAAC;YAC1B,CAAC,CAAC;YACF,MAAM,CAAC,gBAAgB,CAAC,OAAO,EAAE,OAAO,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;YAC1D,MAAM,KAAK,GAAG,OAAO,CAAC,iBAAiB,IAAI,iBAAiB,CAAC;YAC7D,IAAI,QAAQ,GAAG,KAAK,CAAC;YACrB,IAAI,sBAAqF,CAAC;YAC1F,MAAM,gBAAgB,GAAG,0BAA0B,EAAE,CAAC;YACtD,IAAI,CAAC;gBACJ,MAAM,MAAM,GAAG,MAAM,KAAK,CAAC;oBAC1B,OAAO,EAAE,OAAO,CAAC,OAAO;oBACxB,GAAG,EAAE,OAAO,CAAC,GAAG;oBAChB,SAAS,EAAE,GAAG,CAAC,KAAK;oBACpB,YAAY,EAAE,GAAG,CAAC,QAAQ;oBAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,UAAU,EAAE,OAAO,CAAC,UAAU;oBAC9B,WAAW,EAAE,GAAG,CAAC,KAAK;oBACtB,IAAI,EAAE,GAAG,CAAC,IAAI;oBACd,UAAU,EAAE,GAAG,CAAC,GAAG;oBACnB,gBAAgB;oBAChB,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK;oBACtC,SAAS,EAAE,OAAO,CAAC,SAAS;oBAC5B,SAAS,EAAE,OAAO,CAAC,SAAS,KAAK,CAAC;oBAClC,QAAQ,EAAE,OAAO,CAAC,QAAQ;oBAC1B,cAAc,EAAE,OAAO,CAAC,cAAc;oBACtC,QAAQ,EAAE,OAAO,CAAC,QAAQ;oBAC1B,cAAc,EAAE,OAAO,CAAC,cAAc;oBACtC,cAAc,EAAE,OAAO,CAAC,cAAc;oBACtC,KAAK,EAAE,OAAO,CAAC,KAAK;oBACpB,QAAQ,EAAE,OAAO,CAAC,QAAQ;oBAC1B,MAAM;oBACN,aAAa,EAAE,CAAC,KAAkB,EAAE,EAAE;wBACrC,IAAI,KAAK,CAAC,IAAI,KAAK,eAAe,EAAE,CAAC;4BACpC,QAAQ,GAAG,IAAI,CAAC;wBACjB,CAAC;wBACD,IAAI,GAAG,CAAC,GAAG,IAAI,KAAK,CAAC,IAAI,KAAK,iBAAiB,EAAE,CAAC;4BACjD,sBAAsB,GAAG,KAAK,CAAC;4BAC/B,OAAO;wBACR,CAAC;wBACD,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;4BACpB,OAAO;wBACR,CAAC;wBACD,IAAI,CAAC,KAAK,CAAC,CAAC;oBACb,CAAC;iBACD,CAAC,CAAC;gBACH,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;oBACpB,OAAO;gBACR,CAAC;gBACD,IAAI,GAAG,CAAC,GAAG,IAAI,sBAAsB,EAAE,CAAC;oBACvC,MAAM,OAAO,GAAG,MAAM,iBAAiB,CACtC,oBAAoB,CAAC,gBAAgB,EAAE,GAAG,CAAC,KAAK,EAAE,GAAG,CAAC,QAAQ,EAAE,GAAG,CAAC,GAAG,CAAC,CACxE,CAAC;oBACF,IAAI,OAAO,EAAE,cAAc,EAAE,CAAC;wBAC7B,MAAM,CAAC,cAAc,CAAC,sBAAsB,CAAC,MAAM,EAAE,gBAAgB,EAAE;4BACtE,KAAK,EAAE,OAAO,CAAC,cAAc;4BAC7B,UAAU,EAAE,KAAK;yBACjB,CAAC,CAAC;oBACJ,CAAC;oBACD,IAAI,CAAC,sBAAsB,CAAC,CAAC;gBAC9B,CAAC;gBACD,IAAI,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;oBACxC,MAAM,OAAO,GAAG,wBAAwB,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;oBACzE,WAAW,CAAC,OAAO,CAAC,CAAC;gBACtB,CAAC;YACF,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBAChB,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;oBACpB,OAAO;gBACR,CAAC;gBACD,MAAM,OAAO,GAAG,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;gBACvE,WAAW,CAAC,OAAO,CAAC,CAAC;YACtB,CAAC;oBAAS,CAAC;gBACV,MAAM,CAAC,mBAAmB,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;YAC9C,CAAC;QACF,CAAC;KACD,CAAC;AACH,CAAC"}
@@ -1,41 +0,0 @@
1
- import { type AgentHost } from "@post-print/agent-harness";
2
- import type { ScenarioResult } from "../types.js";
3
- import { type ViewerCatalog, type ViewerCatalogScenario, type ViewerJob, type ViewerRunRequest } from "./catalog.js";
4
- import type { ViewerEvent, ViewerRunRecord } from "./events.js";
5
- export declare const VIEWER_MAX_PARALLEL_AGENTS = 4;
6
- export interface ViewerRunner {
7
- runJob(job: ViewerJob, emit: (event: ViewerEvent) => void, signal: AbortSignal): Promise<void>;
8
- finalizeScenario?(input: {
9
- suite: string;
10
- scenario: ViewerCatalogScenario;
11
- host: AgentHost;
12
- result: ScenarioResult;
13
- }, emit: (event: ViewerEvent) => void): Promise<ScenarioResult>;
14
- finalizeCompare?(input: {
15
- suite: string;
16
- scenario: ViewerCatalogScenario;
17
- host: AgentHost;
18
- armResults: Map<string, ScenarioResult>;
19
- }): Promise<ScenarioResult>;
20
- }
21
- export interface ViewerRunHandle {
22
- runId: string;
23
- }
24
- export interface ViewerRunController {
25
- start(request: ViewerRunRequest): ViewerRunHandle;
26
- maxParallelAgents(): number;
27
- cancel(runId: string): boolean;
28
- history(runId: string): ViewerEvent[] | undefined;
29
- subscribe(runId: string, listener: (event: ViewerEvent) => void): () => void;
30
- activeRunId(): string | undefined;
31
- runs(): ViewerRunRecord[];
32
- run(runId: string): ViewerRunRecord | undefined;
33
- }
34
- export declare function followViewerRun(controller: ViewerRunController, runId: string, onEvent: (event: ViewerEvent) => void): (() => void) | undefined;
35
- export declare function createViewerRunController(options: {
36
- catalog: ViewerCatalog;
37
- runner: ViewerRunner;
38
- maxParallelAgents?: number;
39
- initialRuns?: ViewerRunRecord[];
40
- }): ViewerRunController;
41
- //# sourceMappingURL=run-controller.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"run-controller.d.ts","sourceRoot":"","sources":["../../src/viewer/run-controller.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,KAAK,SAAS,EAAiB,MAAM,2BAA2B,CAAC;AAK1E,OAAO,KAAK,EAAE,cAAc,EAAkB,MAAM,aAAa,CAAC;AAElE,OAAO,EAEN,KAAK,aAAa,EAClB,KAAK,qBAAqB,EAC1B,KAAK,SAAS,EACd,KAAK,gBAAgB,EACrB,MAAM,cAAc,CAAC;AACtB,OAAO,KAAK,EAAE,WAAW,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAEhE,eAAO,MAAM,0BAA0B,IAAyB,CAAC;AAEjE,MAAM,WAAW,YAAY;IAC5B,MAAM,CAAC,GAAG,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,EAAE,MAAM,EAAE,WAAW,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC/F,gBAAgB,CAAC,CAChB,KAAK,EAAE;QACN,KAAK,EAAE,MAAM,CAAC;QACd,QAAQ,EAAE,qBAAqB,CAAC;QAChC,IAAI,EAAE,SAAS,CAAC;QAChB,MAAM,EAAE,cAAc,CAAC;KACvB,EACD,IAAI,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,GAChC,OAAO,CAAC,cAAc,CAAC,CAAC;IAC3B,eAAe,CAAC,CAAC,KAAK,EAAE;QACvB,KAAK,EAAE,MAAM,CAAC;QACd,QAAQ,EAAE,qBAAqB,CAAC;QAChC,IAAI,EAAE,SAAS,CAAC;QAChB,UAAU,EAAE,GAAG,CAAC,MAAM,EAAE,cAAc,CAAC,CAAC;KACxC,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;CAC5B;AAED,MAAM,WAAW,eAAe;IAC/B,KAAK,EAAE,MAAM,CAAC;CACd;AAqBD,MAAM,WAAW,mBAAmB;IACnC,KAAK,CAAC,OAAO,EAAE,gBAAgB,GAAG,eAAe,CAAC;IAClD,iBAAiB,IAAI,MAAM,CAAC;IAC5B,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC;IAC/B,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,WAAW,EAAE,GAAG,SAAS,CAAC;IAClD,SAAS,CAAC,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,GAAG,MAAM,IAAI,CAAC;IAC7E,WAAW,IAAI,MAAM,GAAG,SAAS,CAAC;IAClC,IAAI,IAAI,eAAe,EAAE,CAAC;IAC1B,GAAG,CAAC,KAAK,EAAE,MAAM,GAAG,eAAe,GAAG,SAAS,CAAC;CAChD;AAED,wBAAgB,eAAe,CAC9B,UAAU,EAAE,mBAAmB,EAC/B,KAAK,EAAE,MAAM,EACb,OAAO,EAAE,CAAC,KAAK,EAAE,WAAW,KAAK,IAAI,GACnC,CAAC,MAAM,IAAI,CAAC,GAAG,SAAS,CAW1B;AAED,wBAAgB,yBAAyB,CAAC,OAAO,EAAE;IAClD,OAAO,EAAE,aAAa,CAAC;IACvB,MAAM,EAAE,YAAY,CAAC;IACrB,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,WAAW,CAAC,EAAE,eAAe,EAAE,CAAC;CAChC,GAAG,mBAAmB,CA4GtB"}
@@ -1,426 +0,0 @@
1
- import { sumUsageParts } from "@post-print/agent-harness";
2
- import { finalizeCompareOutcome } from "../compare-scenario.js";
3
- import { buildScenarioStory } from "../scenario-story.js";
4
- import { buildScenarioResultUsage } from "../scenario-usage.js";
5
- import { DEFAULT_VIEWER_WORKERS, runWorkerPool } from "../worker-pool.js";
6
- import { expandViewerJobs, } from "./catalog.js";
7
- export const VIEWER_MAX_PARALLEL_AGENTS = DEFAULT_VIEWER_WORKERS;
8
- export function followViewerRun(controller, runId, onEvent) {
9
- if (!controller.history(runId))
10
- return undefined;
11
- let index = 0;
12
- const flush = () => {
13
- const history = controller.history(runId);
14
- if (!history)
15
- return;
16
- while (index < history.length)
17
- onEvent(history[index++]);
18
- };
19
- const unsubscribe = controller.subscribe(runId, flush);
20
- flush();
21
- return unsubscribe;
22
- }
23
- export function createViewerRunController(options) {
24
- const maxParallel = options.maxParallelAgents ?? VIEWER_MAX_PARALLEL_AGENTS;
25
- const sessions = new Map();
26
- let activeId;
27
- for (const record of options.initialRuns ?? []) {
28
- sessions.set(record.id, {
29
- record: structuredClone(record),
30
- events: [],
31
- listeners: new Set(),
32
- done: Promise.resolve(),
33
- armResults: new Map(),
34
- finalizing: new Set(),
35
- });
36
- }
37
- const emit = (target, event) => {
38
- target.events.push(event);
39
- if (event.type === "scenario_result") {
40
- addScenarioResult(target, event, options.catalog);
41
- }
42
- for (const listener of target.listeners)
43
- listener(event);
44
- };
45
- return {
46
- activeRunId: () => activeId,
47
- maxParallelAgents: () => maxParallel,
48
- runs: () => [...sessions.values()].map(({ record }) => structuredClone(record)),
49
- run: (runId) => {
50
- const record = sessions.get(runId)?.record;
51
- return record ? structuredClone(record) : undefined;
52
- },
53
- start(request) {
54
- if (activeId)
55
- throw new Error("A viewer run is already in progress");
56
- if (request.workers !== undefined &&
57
- (!Number.isInteger(request.workers) || request.workers < 1 || request.workers > maxParallel)) {
58
- throw new Error(`workers must be an integer 1-${maxParallel}`);
59
- }
60
- const jobs = expandViewerJobs(options.catalog, request);
61
- if (jobs.length === 0)
62
- throw new Error("No matching scenarios to run");
63
- const runId = `run-${Date.now()}`;
64
- const abort = new AbortController();
65
- const current = {
66
- record: {
67
- id: runId,
68
- request: structuredClone(request),
69
- status: "running",
70
- startedAt: new Date().toISOString(),
71
- reports: [],
72
- },
73
- events: [],
74
- listeners: new Set(),
75
- abort,
76
- done: Promise.resolve(),
77
- armResults: new Map(),
78
- finalizing: new Set(),
79
- };
80
- sessions.set(runId, current);
81
- activeId = runId;
82
- emit(current, { type: "run_started", runId });
83
- current.done = runJobBatches({
84
- jobs,
85
- catalog: options.catalog,
86
- hosts: request.hosts ?? options.catalog.defaultSelectedHosts,
87
- parallelHosts: request.parallelHosts === true,
88
- maxParallel: request.workers ?? maxParallel,
89
- signal: abort.signal,
90
- runner: options.runner,
91
- forward: (event) => emit(current, event),
92
- afterJob: () => finalizeScheduledCompares(current, options.catalog, options.runner, emit),
93
- })
94
- .catch((error) => emit(current, {
95
- type: "error",
96
- message: error instanceof Error ? error.message : String(error),
97
- }))
98
- .finally(() => {
99
- const totals = summarizeRun(current.record.reports, current.events);
100
- current.record.status = abort.signal.aborted ? "cancelled" : "completed";
101
- current.record.finishedAt = new Date().toISOString();
102
- emit(current, { type: "run_finished", runId, status: current.record.status, ...totals });
103
- if (activeId === runId)
104
- activeId = undefined;
105
- });
106
- return { runId };
107
- },
108
- cancel(runId) {
109
- const target = sessions.get(runId);
110
- if (!target?.abort || target.record.status !== "running")
111
- return false;
112
- target.record.status = "cancelling";
113
- target.abort.abort();
114
- return true;
115
- },
116
- history(runId) {
117
- const target = sessions.get(runId);
118
- return target ? [...target.events] : undefined;
119
- },
120
- subscribe(runId, listener) {
121
- const target = sessions.get(runId);
122
- if (!target)
123
- return () => undefined;
124
- target.listeners.add(listener);
125
- return () => {
126
- target.listeners.delete(listener);
127
- };
128
- },
129
- };
130
- }
131
- function addScenarioResult(session, event, catalog) {
132
- const report = ensureReport(session.record, event.suite, event.host);
133
- const scenario = catalog.suites
134
- .find((suite) => suite.name === event.suite)
135
- ?.scenarios.find((item) => item.name === event.scenario);
136
- if (!event.arm || !scenario?.compare?.length) {
137
- upsertReportResult(report, event.result);
138
- return;
139
- }
140
- const key = `${event.suite}::${event.scenario}::${event.host}`;
141
- const group = session.armResults.get(key) ?? {
142
- suite: event.suite,
143
- scenario: event.scenario,
144
- host: event.host,
145
- results: new Map(),
146
- judgeWorkspaces: new Map(),
147
- };
148
- const judgeWorkspace = event.result.judgeWorkspace;
149
- if (judgeWorkspace)
150
- group.judgeWorkspaces.set(event.arm, judgeWorkspace);
151
- group.results.set(event.arm, event.result);
152
- session.armResults.set(key, group);
153
- }
154
- function buildViewerCompareResult(suite, scenarioName, scenario, stored) {
155
- const arms = (scenario.compare ?? []).flatMap((arm) => {
156
- const result = stored.get(arm.id);
157
- return result
158
- ? [
159
- {
160
- id: arm.id,
161
- label: arm.label,
162
- description: arm.description,
163
- prompt: arm.prompt ?? scenario.prompt,
164
- trace: result.trace,
165
- durationMs: result.durationMs,
166
- passed: !result.failures.some((failure) => failure.category === "rubric_miss"),
167
- failures: result.failures,
168
- judgeVerdicts: result.judgeVerdicts,
169
- },
170
- ]
171
- : [];
172
- });
173
- const { compare, failures } = finalizeCompareOutcome(arms, scenario.gates);
174
- const passed = failures.length === 0;
175
- const result = {
176
- suite,
177
- scenario: scenarioName,
178
- description: scenario.description,
179
- prompt: scenario.prompt,
180
- passed,
181
- failures,
182
- durationMs: Math.max(...arms.map((arm) => arm.durationMs ?? 0)),
183
- compare,
184
- story: buildScenarioStory({
185
- rubric: scenario.rubric,
186
- trace: arms[0]?.trace,
187
- passed,
188
- failures,
189
- compare: {
190
- arms: arms.map((arm) => ({
191
- id: arm.id,
192
- label: arm.label,
193
- description: arm.description,
194
- prompt: arm.prompt,
195
- trace: arm.trace,
196
- passed: arm.passed,
197
- failures: arm.failures,
198
- judgeVerdicts: arm.judgeVerdicts,
199
- })),
200
- gates: scenario.gates,
201
- },
202
- }),
203
- ...buildScenarioResultUsage({
204
- agentUsage: sumUsageParts(arms.map((arm) => arm.trace?.usage)),
205
- }),
206
- };
207
- return result;
208
- }
209
- async function finalizeScheduledCompares(session, catalog, runner, emit) {
210
- for (const [key, group] of session.armResults) {
211
- if (session.abort?.signal.aborted)
212
- return;
213
- if (session.finalizing.has(key))
214
- continue;
215
- const scenario = catalog.suites
216
- .find((suite) => suite.name === group.suite)
217
- ?.scenarios.find((item) => item.name === group.scenario);
218
- if (!scenario?.compare?.length || group.results.size < scenario.compare.length)
219
- continue;
220
- session.finalizing.add(key);
221
- emit(session, {
222
- type: "cell_started",
223
- suite: group.suite,
224
- scenario: group.scenario,
225
- host: group.host,
226
- });
227
- emit(session, {
228
- type: "scenario_finalizing",
229
- suite: group.suite,
230
- scenario: group.scenario,
231
- host: group.host,
232
- });
233
- emit(session, {
234
- type: "status",
235
- suite: group.suite,
236
- scenario: group.scenario,
237
- host: group.host,
238
- text: "Finalizing comparison and judge.",
239
- });
240
- let result;
241
- try {
242
- const armResults = new Map();
243
- for (const [armId, armResult] of group.results) {
244
- const resultWithWorkspace = structuredClone(armResult);
245
- const workspace = group.judgeWorkspaces.get(armId);
246
- if (workspace) {
247
- Object.defineProperty(resultWithWorkspace, "judgeWorkspace", {
248
- value: workspace,
249
- enumerable: false,
250
- });
251
- }
252
- armResults.set(armId, resultWithWorkspace);
253
- }
254
- result = runner.finalizeCompare
255
- ? await runner.finalizeCompare({
256
- suite: group.suite,
257
- scenario,
258
- host: group.host,
259
- armResults,
260
- })
261
- : buildViewerCompareResult(group.suite, group.scenario, scenario, group.results);
262
- }
263
- catch (error) {
264
- result = buildViewerCompareResult(group.suite, group.scenario, scenario, group.results);
265
- result.failures.push({
266
- matcher: "compareFinalize",
267
- message: error instanceof Error ? error.message : String(error),
268
- category: "agent_runtime",
269
- });
270
- result.passed = false;
271
- }
272
- if (session.abort?.signal.aborted)
273
- return;
274
- emit(session, {
275
- type: "cell_finished",
276
- suite: group.suite,
277
- scenario: group.scenario,
278
- host: group.host,
279
- passed: result.passed,
280
- durationMs: result.durationMs,
281
- failures: result.failures.map((failure) => ({
282
- matcher: failure.matcher,
283
- message: failure.message,
284
- })),
285
- });
286
- emit(session, {
287
- type: "scenario_result",
288
- suite: group.suite,
289
- scenario: group.scenario,
290
- host: group.host,
291
- result,
292
- });
293
- }
294
- }
295
- function ensureReport(record, suite, host) {
296
- let report = record.reports.find((item) => item.suite === suite && item.host === host);
297
- if (!report) {
298
- report = { suite, host, passed: 0, failed: 0, skipped: 0, results: [] };
299
- record.reports.push(report);
300
- }
301
- return report;
302
- }
303
- function upsertReportResult(report, result) {
304
- const index = report.results.findIndex((item) => item.scenario === result.scenario);
305
- if (index < 0)
306
- report.results.push(structuredClone(result));
307
- else
308
- report.results[index] = structuredClone(result);
309
- report.passed = report.results.filter((item) => item.passed && !item.skipped).length;
310
- report.failed = report.results.filter((item) => !item.passed && !item.skipped).length;
311
- report.skipped = report.results.filter((item) => item.skipped).length;
312
- }
313
- function summarizeReports(reports) {
314
- return reports.reduce((total, report) => ({
315
- passed: total.passed + report.passed,
316
- failed: total.failed + report.failed,
317
- skipped: total.skipped + report.skipped,
318
- }), { passed: 0, failed: 0, skipped: 0 });
319
- }
320
- function summarizeRun(reports, events) {
321
- if (events.some((event) => event.type === "scenario_result"))
322
- return summarizeReports(reports);
323
- return events.reduce((total, event) => {
324
- if (event.type !== "cell_finished")
325
- return total;
326
- if (event.skipped)
327
- total.skipped += 1;
328
- else if (event.passed)
329
- total.passed += 1;
330
- else
331
- total.failed += 1;
332
- return total;
333
- }, { passed: 0, failed: 0, skipped: 0 });
334
- }
335
- async function runJobBatches(options) {
336
- const batches = options.parallelHosts
337
- ? [options.jobs]
338
- : options.hosts
339
- .map((host) => options.jobs.filter((job) => job.host === host))
340
- .filter((batch) => batch.length > 0);
341
- for (const batch of batches) {
342
- if (options.signal.aborted)
343
- return;
344
- await runWorkerPool(batch, options.maxParallel, async (job) => {
345
- try {
346
- let completedResult;
347
- await options.runner.runJob(job, (event) => {
348
- if (event.type === "scenario_result" && !job.arm) {
349
- completedResult = event.result;
350
- return;
351
- }
352
- options.forward(event);
353
- }, options.signal);
354
- if (!completedResult ||
355
- job.arm ||
356
- !options.runner.finalizeScenario ||
357
- options.signal.aborted) {
358
- if (completedResult && !options.signal.aborted)
359
- options.forward({ type: "scenario_result", ...job, result: completedResult });
360
- return;
361
- }
362
- const scenario = options.catalog.suites
363
- .find((suite) => suite.name === job.suite)
364
- ?.scenarios.find((item) => item.name === job.scenario);
365
- if (!scenario || scenario.compare?.length) {
366
- options.forward({ type: "scenario_result", ...job, result: completedResult });
367
- return;
368
- }
369
- options.forward({ type: "scenario_finalizing", ...job });
370
- const finalized = await options.runner.finalizeScenario({ suite: job.suite, scenario, host: job.host, result: completedResult }, options.forward);
371
- // The child emits an agent-only completion before parent-process judging.
372
- // Replace that provisional status so the cell cannot remain green when
373
- // judging later fails or becomes unavailable.
374
- options.forward({
375
- type: "cell_finished",
376
- ...job,
377
- passed: finalized.passed,
378
- durationMs: finalized.durationMs,
379
- failures: finalized.failures.map((failure) => ({
380
- matcher: failure.matcher,
381
- message: failure.message,
382
- })),
383
- });
384
- options.forward({ type: "scenario_result", ...job, result: finalized });
385
- }
386
- catch (error) {
387
- if (options.signal.aborted)
388
- return;
389
- const message = error instanceof Error ? error.message : String(error);
390
- const envelope = {
391
- suite: job.suite,
392
- scenario: job.scenario,
393
- host: job.host,
394
- ...(job.arm ? { arm: job.arm } : {}),
395
- };
396
- const failures = [
397
- { matcher: "liveScenario", message, category: "agent_runtime" },
398
- ];
399
- options.forward({ type: "error", ...envelope, message });
400
- options.forward({
401
- type: "cell_finished",
402
- ...envelope,
403
- passed: false,
404
- failures,
405
- durationMs: 0,
406
- });
407
- options.forward({
408
- type: "scenario_result",
409
- ...envelope,
410
- result: {
411
- suite: job.suite,
412
- scenario: job.scenario,
413
- prompt: job.prompt,
414
- passed: false,
415
- failures,
416
- durationMs: 0,
417
- },
418
- });
419
- }
420
- finally {
421
- await options.afterJob();
422
- }
423
- }, options.signal);
424
- }
425
- }
426
- //# sourceMappingURL=run-controller.js.map