@citeark/agent 0.3.20

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 (347) hide show
  1. package/LICENSE +202 -0
  2. package/README.md +128 -0
  3. package/data/dataset-source-registry.v1.json +300 -0
  4. package/dist/arkgraph/boot.js +6 -0
  5. package/dist/arkgraph/index.html +1 -0
  6. package/dist/arkgraph/viewer.css +1 -0
  7. package/dist/arkgraph/viewer.en.css +1 -0
  8. package/dist/arkgraph/viewer.en.js +49 -0
  9. package/dist/arkgraph/viewer.en.js.LEGAL.txt +56 -0
  10. package/dist/arkgraph/viewer.js +49 -0
  11. package/dist/arkgraph/viewer.js.LEGAL.txt +56 -0
  12. package/docker/claude-code/Dockerfile +97 -0
  13. package/docker/claude-code/codex-pro-relay.mjs +466 -0
  14. package/docker/claude-code/runtime-contract-check.mjs +79 -0
  15. package/docs/arkgraph-reading.md +79 -0
  16. package/docs/configuration.md +100 -0
  17. package/docs/integration.md +92 -0
  18. package/docs/maturity-plan.md +27 -0
  19. package/docs/npm-release.md +44 -0
  20. package/docs/paper-reading.md +40 -0
  21. package/docs/research-plan-granularity.md +27 -0
  22. package/docs/terminal.md +49 -0
  23. package/examples/toy-evaluation/compile-task.json +27 -0
  24. package/examples/toy-evaluation/paper.md +5 -0
  25. package/examples/toy-evaluation/repository/README.md +9 -0
  26. package/examples/toy-evaluation/repository/checkpoint.json +4 -0
  27. package/examples/toy-evaluation/repository/evaluate.py +17 -0
  28. package/examples/toy-evaluation/task.json +81 -0
  29. package/package.json +59 -0
  30. package/prompts/compile-research.md +58 -0
  31. package/prompts/execute-contract.md +72 -0
  32. package/prompts/execute-workspace-simple.md +51 -0
  33. package/prompts/execute-workspace.md +34 -0
  34. package/prompts/prepare-reproduction.md +82 -0
  35. package/prompts/repair-research.md +45 -0
  36. package/protocol/CAP.md +129 -0
  37. package/protocol/LICENSE +12 -0
  38. package/protocol/MAPPINGS.md +72 -0
  39. package/protocol/README.md +38 -0
  40. package/protocol/conformance-v2.0-alpha.1.json +36 -0
  41. package/protocol/examples/arkgraph/checkpoint-evaluation.json +309 -0
  42. package/protocol/examples/arkgraph/fixtures.mjs +49 -0
  43. package/protocol/examples/arkgraph/paper-free.json +291 -0
  44. package/protocol/examples/arkgraph/partial-failure.json +344 -0
  45. package/protocol/examples/arkgraph/training-evaluation.json +443 -0
  46. package/protocol/profiles/agent-trace.md +16 -0
  47. package/protocol/profiles/computational-run.md +16 -0
  48. package/protocol/profiles/core.md +15 -0
  49. package/protocol/profiles/public-bundle.md +18 -0
  50. package/protocol/profiles/reproduction.md +29 -0
  51. package/protocol/profiles/research-compilation.md +44 -0
  52. package/protocol/profiles/research-plan.md +39 -0
  53. package/protocol/profiles/restricted-evidence.md +15 -0
  54. package/runtime/bootstrap-autodl-runtime.sh +314 -0
  55. package/runtime/create-runtime-venv.sh +41 -0
  56. package/runtime/install-local-cpu-runtime.sh +23 -0
  57. package/runtime/install-scientific-runtime.sh +153 -0
  58. package/runtime/mineru/parse.py +62 -0
  59. package/runtime/mineru/requirements.txt +4 -0
  60. package/runtime/requirements-baseline.txt +38 -0
  61. package/schemas/cap/v2/activity.schema.json +47 -0
  62. package/schemas/cap/v2/agent.schema.json +32 -0
  63. package/schemas/cap/v2/assertion.schema.json +110 -0
  64. package/schemas/cap/v2/descriptor.schema.json +243 -0
  65. package/schemas/cap/v2/entity.schema.json +64 -0
  66. package/schemas/cap/v2/manifest.schema.json +67 -0
  67. package/schemas/cap/v2/relation.schema.json +82 -0
  68. package/schemas/compute-catalog.schema.json +63 -0
  69. package/schemas/compute-decision.schema.json +27 -0
  70. package/schemas/execution-contract.schema.json +1024 -0
  71. package/schemas/research-card.schema.json +30 -0
  72. package/schemas/research-inventory-draft.schema.json +366 -0
  73. package/schemas/research.schema.json +1044 -0
  74. package/schemas/result.schema.json +173 -0
  75. package/schemas/verification-policy.schema.json +47 -0
  76. package/schemas/verified-conclusion.schema.json +58 -0
  77. package/schemas/workspace-summary.schema.json +24 -0
  78. package/scripts/build-arkgraph-view.mjs +12 -0
  79. package/scripts/check-execution-feasibility.mjs +24 -0
  80. package/scripts/check-syntax.mjs +15 -0
  81. package/scripts/deterministic-asset-preparation.py +438 -0
  82. package/scripts/package-cap.mjs +23 -0
  83. package/scripts/package-local-agent.mjs +23 -0
  84. package/scripts/preview-arkgraph.mjs +25 -0
  85. package/scripts/replay-research-compiler-candidate.mjs +134 -0
  86. package/scripts/review-compiler-sources.mjs +44 -0
  87. package/scripts/run-asset-preparation.sh +17 -0
  88. package/scripts/run-research-plan.mjs +98 -0
  89. package/scripts/validate-asset-preparation.py +290 -0
  90. package/scripts/verify-local-runtime.mjs +57 -0
  91. package/scripts/verify-npm-package.mjs +57 -0
  92. package/src/adapters/paper2agent.mjs +107 -0
  93. package/src/assets/cache.mjs +159 -0
  94. package/src/assets/compute.mjs +98 -0
  95. package/src/assets/executor.mjs +145 -0
  96. package/src/assets/lifecycle.mjs +213 -0
  97. package/src/assets/manifest.mjs +242 -0
  98. package/src/assets/opportunistic-preparation.mjs +81 -0
  99. package/src/assets/plan.mjs +411 -0
  100. package/src/assets/prompts.mjs +29 -0
  101. package/src/assets/public-asset-probe.mjs +525 -0
  102. package/src/assets/qualification.mjs +119 -0
  103. package/src/assets/readiness.mjs +130 -0
  104. package/src/assets/reproduction-admission.mjs +355 -0
  105. package/src/assets/requirements.mjs +152 -0
  106. package/src/assets/source-grounding.mjs +341 -0
  107. package/src/assets/source-policy.mjs +118 -0
  108. package/src/autodl/client.mjs +260 -0
  109. package/src/autodl/ssh.mjs +380 -0
  110. package/src/autodl/tools.mjs +129 -0
  111. package/src/cap/redaction.mjs +38 -0
  112. package/src/cap/v2/archive.mjs +152 -0
  113. package/src/cap/v2/attestation.mjs +204 -0
  114. package/src/cap/v2/canonical-json.mjs +114 -0
  115. package/src/cap/v2/compilation-artifact.mjs +240 -0
  116. package/src/cap/v2/core.mjs +282 -0
  117. package/src/cap/v2/measurement-assessment-records.mjs +23 -0
  118. package/src/cap/v2/pipeline-artifact.mjs +922 -0
  119. package/src/cap/v2/read.mjs +41 -0
  120. package/src/cap/v2/reassessment-artifact.mjs +383 -0
  121. package/src/cap/v2/research-artifact.mjs +231 -0
  122. package/src/cap/v2/research-map-records.mjs +46 -0
  123. package/src/cap/v2/research-object-records.mjs +163 -0
  124. package/src/cap/v2/research-records.mjs +187 -0
  125. package/src/cap/v2/verify.mjs +642 -0
  126. package/src/cli.mjs +1146 -0
  127. package/src/compute/autodl-pro-compiler.mjs +347 -0
  128. package/src/compute/autodl-pro-executor.mjs +459 -0
  129. package/src/compute/autodl-pro-job.mjs +843 -0
  130. package/src/compute/autodl-pro-network.mjs +295 -0
  131. package/src/compute/autodl-pro-remote.mjs +810 -0
  132. package/src/compute/autodl-pro-staging.mjs +117 -0
  133. package/src/compute/campaign.mjs +110 -0
  134. package/src/compute/catalog.mjs +123 -0
  135. package/src/compute/checkpoint-protocol.mjs +154 -0
  136. package/src/compute/codex-account-lock.mjs +111 -0
  137. package/src/compute/codex-account-session.mjs +107 -0
  138. package/src/compute/compiler-profile.mjs +38 -0
  139. package/src/compute/compiler-router.mjs +23 -0
  140. package/src/compute/coordinator-recovery.mjs +210 -0
  141. package/src/compute/executor-router.mjs +29 -0
  142. package/src/compute/gcp-batch-compiler.mjs +685 -0
  143. package/src/compute/gcp-batch-executor.mjs +1215 -0
  144. package/src/compute/gcp-batch-failure.mjs +92 -0
  145. package/src/compute/gcp-batch-job.mjs +527 -0
  146. package/src/compute/gcp-batch-lifecycle.mjs +81 -0
  147. package/src/compute/gcp-checkpoint-worker.mjs +1633 -0
  148. package/src/compute/local-codex-compiler.mjs +52 -0
  149. package/src/compute/measurement-hardware.mjs +128 -0
  150. package/src/compute/remote-attempt.mjs +226 -0
  151. package/src/compute/requirements.mjs +124 -0
  152. package/src/compute/research-phases.mjs +48 -0
  153. package/src/compute/scheduler.mjs +452 -0
  154. package/src/compute/shared-workloads.mjs +26 -0
  155. package/src/compute/stage-archive.mjs +79 -0
  156. package/src/contracts/campaign-contract.mjs +52 -0
  157. package/src/contracts/execution-contract.mjs +819 -0
  158. package/src/contracts/execution-mode.mjs +19 -0
  159. package/src/contracts/execution-timeouts.mjs +45 -0
  160. package/src/contracts/execution-workload.mjs +68 -0
  161. package/src/contracts/preflight-schema.mjs +25 -0
  162. package/src/contracts/public-contract.mjs +63 -0
  163. package/src/contracts/subject-tags.mjs +31 -0
  164. package/src/dashboard/data.mjs +898 -0
  165. package/src/dashboard/server.mjs +79 -0
  166. package/src/dashboard/static/dashboard.css +366 -0
  167. package/src/dashboard/static/dashboard.js +560 -0
  168. package/src/dashboard/static/index.html +85 -0
  169. package/src/deployment/community-policy.mjs +9 -0
  170. package/src/deployment/environment.mjs +112 -0
  171. package/src/deployment/guided.mjs +98 -0
  172. package/src/deployment/handoff.mjs +102 -0
  173. package/src/deployment/local-contract.mjs +31 -0
  174. package/src/deployment/local.mjs +100 -0
  175. package/src/deployment/prepare.mjs +46 -0
  176. package/src/deployment/recipe.mjs +108 -0
  177. package/src/deployment/supplement.mjs +51 -0
  178. package/src/deployment/terminal.mjs +43 -0
  179. package/src/diagnosis/renderer.mjs +75 -0
  180. package/src/diagnosis/target-failure.mjs +46 -0
  181. package/src/evidence/parser-registry.mjs +54 -0
  182. package/src/evidence/parsers/fasttext-classification.mjs +82 -0
  183. package/src/evidence/parsers/json-scalar.mjs +96 -0
  184. package/src/evidence/parsers/simcse-senteval.mjs +104 -0
  185. package/src/evidence/parsers/starspace-classification.mjs +78 -0
  186. package/src/evidence/registry.mjs +147 -0
  187. package/src/execution/runner-audit.mjs +473 -0
  188. package/src/gcp/auth.mjs +106 -0
  189. package/src/gcp/batch-client.mjs +120 -0
  190. package/src/gcp/resource-discovery.mjs +177 -0
  191. package/src/gcp/rest.mjs +82 -0
  192. package/src/gcp/secret-manager.mjs +34 -0
  193. package/src/gcp/signed-url.mjs +133 -0
  194. package/src/gcp/storage.mjs +220 -0
  195. package/src/graph/command.mjs +41 -0
  196. package/src/graph/execution.mjs +97 -0
  197. package/src/graph/model.mjs +37 -0
  198. package/src/graph/presentation.mjs +110 -0
  199. package/src/graph/query.mjs +159 -0
  200. package/src/graph/research-relations.mjs +69 -0
  201. package/src/graph/source-page.mjs +12 -0
  202. package/src/graph/source-preview.mjs +34 -0
  203. package/src/graph/validate.mjs +76 -0
  204. package/src/job.mjs +496 -0
  205. package/src/network/autodl-routing-proxy.mjs +462 -0
  206. package/src/network/egress-proxy.mjs +158 -0
  207. package/src/observability/event-contract.mjs +230 -0
  208. package/src/observability/pipeline-monitor.mjs +166 -0
  209. package/src/pipeline/orchestrator.mjs +1281 -0
  210. package/src/pipeline/recovery-error.mjs +11 -0
  211. package/src/pipeline/replay.mjs +304 -0
  212. package/src/pipeline/shared-execution.mjs +115 -0
  213. package/src/pipeline/stage-checkpoint.mjs +86 -0
  214. package/src/pipeline/stage-recovery.mjs +101 -0
  215. package/src/pipeline/targets.mjs +110 -0
  216. package/src/process.mjs +143 -0
  217. package/src/protocol.mjs +312 -0
  218. package/src/provider/codex-account.mjs +44 -0
  219. package/src/provider/codex-completion.mjs +49 -0
  220. package/src/provider/completion.mjs +292 -0
  221. package/src/provider/model-client.mjs +44 -0
  222. package/src/provider/model-route.mjs +29 -0
  223. package/src/provider/openrouter-readiness.mjs +189 -0
  224. package/src/provider/reader-bridge.mjs +35 -0
  225. package/src/provider/relay.mjs +263 -0
  226. package/src/provider/runtime-auth.mjs +40 -0
  227. package/src/public/cap.d.mts +90 -0
  228. package/src/public/cap.mjs +12 -0
  229. package/src/public/contracts.d.mts +2 -0
  230. package/src/public/host.mjs +171 -0
  231. package/src/public/operations.d.mts +11 -0
  232. package/src/public/presentation.d.mts +4 -0
  233. package/src/records/views.mjs +26 -0
  234. package/src/remote/command.mjs +178 -0
  235. package/src/remote/ssh.mjs +59 -0
  236. package/src/repository-origin.mjs +81 -0
  237. package/src/reproduction/evidence-feedback.mjs +96 -0
  238. package/src/reproduction/incomplete-initialization.mjs +25 -0
  239. package/src/reproduction/lifecycle.mjs +253 -0
  240. package/src/reproduction/plan.mjs +132 -0
  241. package/src/reproduction/prompts.mjs +70 -0
  242. package/src/reproduction/runner.mjs +188 -0
  243. package/src/reproduction/summary.mjs +130 -0
  244. package/src/reproduction/workspace-mode.mjs +7 -0
  245. package/src/research/automatic-admission.mjs +156 -0
  246. package/src/research/compiler-coverage.mjs +85 -0
  247. package/src/research/compiler-failure.mjs +24 -0
  248. package/src/research/compiler-normalization-guards.mjs +112 -0
  249. package/src/research/compiler-repair.mjs +3 -0
  250. package/src/research/compiler.mjs +853 -0
  251. package/src/research/continuation-selection.mjs +26 -0
  252. package/src/research/execution-graph-context.mjs +43 -0
  253. package/src/research/experiment-importance.mjs +15 -0
  254. package/src/research/inventory-handoff.mjs +104 -0
  255. package/src/research/inventory-revisions.mjs +32 -0
  256. package/src/research/mineru-local.mjs +73 -0
  257. package/src/research/paper-command.mjs +19 -0
  258. package/src/research/paper-markdown.mjs +180 -0
  259. package/src/research/paper-source-map.mjs +69 -0
  260. package/src/research/planning-policy.mjs +88 -0
  261. package/src/research/reference-materials.mjs +11 -0
  262. package/src/research/reproduction-scope.mjs +30 -0
  263. package/src/research/research-map.mjs +94 -0
  264. package/src/research/research-objects.mjs +88 -0
  265. package/src/research/source-discovery.mjs +646 -0
  266. package/src/research/source-observations.mjs +75 -0
  267. package/src/research/source-review-cli-mcp.mjs +26 -0
  268. package/src/research/source-review-input.mjs +209 -0
  269. package/src/research/source-review-local-codex.mjs +36 -0
  270. package/src/research/source-review-model.mjs +70 -0
  271. package/src/research/source-review.mjs +173 -0
  272. package/src/research/structure.mjs +3163 -0
  273. package/src/research-card/renderer.mjs +277 -0
  274. package/src/research-card/verified-conclusion.mjs +143 -0
  275. package/src/results/output-registry.mjs +183 -0
  276. package/src/runtime/claude-code.mjs +52 -0
  277. package/src/runtime/codex-capacity-retry.mjs +87 -0
  278. package/src/runtime/codex.mjs +64 -0
  279. package/src/runtime/config.mjs +157 -0
  280. package/src/runtime/final-output.mjs +40 -0
  281. package/src/runtime/index.mjs +21 -0
  282. package/src/runtime/local-codex.mjs +74 -0
  283. package/src/runtime/opencode.mjs +95 -0
  284. package/src/runtime/prompt.mjs +13 -0
  285. package/src/sandbox/docker.mjs +363 -0
  286. package/src/settings/command.mjs +297 -0
  287. package/src/settings/store.mjs +119 -0
  288. package/src/telemetry/pricing.mjs +68 -0
  289. package/src/telemetry/usage.mjs +265 -0
  290. package/src/terminal/events.mjs +97 -0
  291. package/src/terminal/input.mjs +40 -0
  292. package/src/terminal/plain.mjs +40 -0
  293. package/src/terminal/remote-stream.mjs +22 -0
  294. package/src/terminal/screen.mjs +214 -0
  295. package/src/terminal/transcript.mjs +69 -0
  296. package/src/util.mjs +107 -0
  297. package/src/verification/ai-assessor.mjs +534 -0
  298. package/src/verification/claim-evaluator.mjs +242 -0
  299. package/src/verification/evidence-context.mjs +165 -0
  300. package/src/verification/evidence-reader.mjs +95 -0
  301. package/src/verification/integrity.mjs +570 -0
  302. package/src/verification/tolerance.mjs +32 -0
  303. package/src/workloads/cpu-research-preparation.mjs +56 -0
  304. package/src/workloads/definition.mjs +74 -0
  305. package/src/workloads/phase-aware-reproduction.mjs +46 -0
  306. package/src/workloads/reproduction.mjs +85 -0
  307. package/src/workspace/command.mjs +242 -0
  308. package/src/workspace/control.mjs +49 -0
  309. package/src/workspace/entry.mjs +28 -0
  310. package/src/workspace/input.mjs +93 -0
  311. package/src/workspace/interactive.mjs +94 -0
  312. package/src/workspace/jobs.mjs +418 -0
  313. package/src/workspace/session.mjs +97 -0
  314. package/src/workspace/worker.mjs +137 -0
  315. package/ui/arkgraph/ambient-motion.mjs +10 -0
  316. package/ui/arkgraph/app.jsx +153 -0
  317. package/ui/arkgraph/boot.js +6 -0
  318. package/ui/arkgraph/camera-motion.mjs +20 -0
  319. package/ui/arkgraph/context-reveal.mjs +39 -0
  320. package/ui/arkgraph/details.css +3 -0
  321. package/ui/arkgraph/entry.jsx +28 -0
  322. package/ui/arkgraph/experiment-curves.mjs +17 -0
  323. package/ui/arkgraph/experiment-selection.mjs +15 -0
  324. package/ui/arkgraph/experiment-style.css +26 -0
  325. package/ui/arkgraph/experiment-ui.jsx +32 -0
  326. package/ui/arkgraph/frame.html +1 -0
  327. package/ui/arkgraph/graph-gestures.mjs +62 -0
  328. package/ui/arkgraph/label-layout.mjs +57 -0
  329. package/ui/arkgraph/locales/en.json +229 -0
  330. package/ui/arkgraph/locales/source-types.json +15 -0
  331. package/ui/arkgraph/localization-build.mjs +27 -0
  332. package/ui/arkgraph/material-build.mjs +23 -0
  333. package/ui/arkgraph/material-colors.mjs +39 -0
  334. package/ui/arkgraph/material-style.css +15 -0
  335. package/ui/arkgraph/open-graph.jsx +326 -0
  336. package/ui/arkgraph/outline.jsx +49 -0
  337. package/ui/arkgraph/package-lock.json +888 -0
  338. package/ui/arkgraph/package.json +17 -0
  339. package/ui/arkgraph/reading-layout.mjs +130 -0
  340. package/ui/arkgraph/reading-presentation.mjs +73 -0
  341. package/ui/arkgraph/record-detail.css +51 -0
  342. package/ui/arkgraph/record-details.jsx +29 -0
  343. package/ui/arkgraph/research-types.mjs +31 -0
  344. package/ui/arkgraph/selection-mark.jsx +6 -0
  345. package/ui/arkgraph/soft-spine.mjs +26 -0
  346. package/ui/arkgraph/steering-style.css +187 -0
  347. package/ui/arkgraph/style.css +272 -0
@@ -0,0 +1,347 @@
1
+ import { recordResearchCleanup } from '../workspace/control.mjs';
2
+ import { restoreRunRecovery } from './coordinator-recovery.mjs';
3
+ import { mkdir, mkdtemp, readFile, rm, writeFile } from "node:fs/promises";
4
+ import os from "node:os";
5
+ import path from "node:path";
6
+
7
+ import { ensureFinalWorkspaceCapture, updateRunState } from "../job.mjs";
8
+ import {
9
+ acceptCompilerDraft,
10
+ compilerPrompt,
11
+ normalizeCompilationTask,
12
+ prepareCompilationBundle,
13
+ } from "../research/compiler.mjs";
14
+ import { agentInvocationPromptArgument, buildAgentInvocation } from "../runtime/index.mjs";
15
+ import { publicAgentRuntime, resolveAgentRuntime } from "../runtime/config.mjs";
16
+ import {
17
+ CODEX_FINAL_MESSAGE_PATH,
18
+ recoverJsonFromFinalMessage,
19
+ } from "../runtime/final-output.mjs";
20
+ import { readAgentUsage } from "../telemetry/usage.mjs";
21
+ import {
22
+ CiteArkError,
23
+ createRunId,
24
+ pathExists,
25
+ readJson,
26
+ writeJson,
27
+ } from "../util.mjs";
28
+ import { remoteProviderSession } from "./gcp-batch-job.mjs";
29
+ import {
30
+ classifyProviderActionRequired,
31
+ describeRemoteAgentFailure,
32
+ } from "./remote-attempt.mjs";
33
+ import {
34
+ acquireAutoDlProLease,
35
+ autoDlInstanceRequest,
36
+ resolveAutoDlProConfig,
37
+ } from "./autodl-pro-job.mjs";
38
+ import {
39
+ createAutoDlStageArchive,
40
+ executeAutoDlRemoteAttempt,
41
+ } from "./autodl-pro-remote.mjs";
42
+
43
+ export async function executeAutoDlProCompilation({
44
+ taskPath,
45
+ runsDirectory,
46
+ runId = createRunId(),
47
+ agentRuntime,
48
+ computeProfile,
49
+ executorOptions = {},
50
+ dryRun = false,
51
+ progress = () => {},
52
+ sourceReviewer,
53
+ prepareMarkdown,
54
+ validateResearch = async () => [],
55
+ }) {
56
+ assertCompilerProfile(computeProfile);
57
+ const absoluteTaskPath = path.resolve(taskPath);
58
+ const taskDirectory = path.dirname(absoluteTaskPath);
59
+ const task = normalizeCompilationTask(await readJson(absoluteTaskPath));
60
+ const runtimeAgent = resolveAgentRuntime({
61
+ taskAgent: task.agent,
62
+ ...agentRuntime,
63
+ requireApi: false,
64
+ });
65
+ task.environment = {
66
+ ...task.environment,
67
+ image: computeProfile.image,
68
+ cpus: computeProfile.cpuCores,
69
+ memoryGb: computeProfile.memoryGb,
70
+ gpu: computeProfile.accelerator === "gpu" ? "all" : "none",
71
+ };
72
+ const bundle = await restoreRunRecovery({ runsDirectory, runId }) ?? await prepareCompilationBundle({
73
+ task,
74
+ taskDirectory,
75
+ runsDirectory,
76
+ runId,
77
+ prepareMarkdown,
78
+ });
79
+ bundle.runtimeAgent = runtimeAgent;
80
+ bundle.computeDecision = {
81
+ schemaVersion: "0.1",
82
+ selectedProfile: structuredClone(computeProfile),
83
+ };
84
+ bundle.executionEnvironment = structuredClone(task.environment);
85
+ bundle.sandboxProvider = "autodl-pro";
86
+ await updateRunState(bundle, {
87
+ runtime: runtimeAgent.runtime,
88
+ agent: publicAgentRuntime(runtimeAgent),
89
+ sandbox: "autodl-pro",
90
+ image: task.environment.image,
91
+ executionEnvironment: bundle.executionEnvironment,
92
+ });
93
+
94
+ const instructions = await readFile(
95
+ path.join(bundle.directories.input, "agent-instructions.md"),
96
+ "utf8",
97
+ );
98
+ const autoDlOptions = executorOptions.autoDl ?? {};
99
+ const config = resolveAutoDlProConfig({
100
+ bundle,
101
+ options: autoDlOptions,
102
+ requireToken: !dryRun,
103
+ });
104
+ bundle.sourceReviewProviderSecret = config.providerSecret;
105
+ bundle.paperBudget = autoDlOptions.paperBudget;
106
+ const instanceRequest = autoDlInstanceRequest(config, bundle.runId);
107
+ await writeJson(
108
+ path.join(bundle.directories.execution, "autodl-instance-request.json"),
109
+ { schemaVersion: "0.1", executor: "autodl-pro", instanceRequest },
110
+ );
111
+ if (dryRun) {
112
+ await updateRunState(bundle, { status: "prepared" });
113
+ return {
114
+ dryRun: true,
115
+ bundle,
116
+ instanceRequest,
117
+ cloud: publicAutoDlExecution(config, null, []),
118
+ };
119
+ }
120
+
121
+ const temporaryDirectory = await mkdtemp(
122
+ path.join(os.tmpdir(), "citeark-autodl-compiler-"),
123
+ );
124
+ const records = [];
125
+ let validationIssues = ["research.json 尚未生成"];
126
+ let normalizationWarnings = [];
127
+ let normalizedResearch = null;
128
+ let blockingFailure = null;
129
+ let lease = null;
130
+ let successful = false;
131
+ await mkdir(path.join(bundle.directories.execution, "prompts"), { recursive: true });
132
+ try {
133
+ await updateRunState(bundle, {
134
+ status: "running",
135
+ startedAt: new Date().toISOString(),
136
+ });
137
+ await autoDlOptions.assertActive?.();
138
+ lease = await acquireAutoDlProLease({
139
+ config,
140
+ runId: bundle.runId,
141
+ clients: autoDlOptions.clients,
142
+ progress,
143
+ });
144
+ await autoDlOptions.assertActive?.();
145
+ const maxAttempts = 1 + runtimeAgent.validationRetries;
146
+ for (let attempt = 1; attempt <= maxAttempts; attempt += 1) {
147
+ const invocation = await buildAgentInvocation({
148
+ bundle,
149
+ attempt,
150
+ prompt: compilerPrompt(attempt, validationIssues, {
151
+ continuation: bundle.runtimeTask?.continuation,
152
+ repairBaseline: bundle.repairBaselineResearch,
153
+ }),
154
+ instructions,
155
+ providerSession: remoteProviderSession(runtimeAgent),
156
+ });
157
+ const resumeInvocation = await buildAgentInvocation({
158
+ bundle,
159
+ attempt: 2,
160
+ prompt: [
161
+ "Continue from the restored compilation workspace. Inspect preserved artifacts and apply the current task below; a prior research.json may still contain the rejected draft.",
162
+ compilerPrompt(attempt, validationIssues, {
163
+ continuation: bundle.runtimeTask?.continuation,
164
+ repairBaseline: bundle.repairBaselineResearch,
165
+ }),
166
+ ].join("\n\n"),
167
+ instructions,
168
+ providerSession: remoteProviderSession(runtimeAgent),
169
+ });
170
+ await writeFile(
171
+ path.join(bundle.directories.execution, "prompts", `attempt-${attempt}.txt`),
172
+ `${agentInvocationPromptArgument(invocation)}\n`,
173
+ );
174
+ await writeFile(
175
+ path.join(bundle.directories.execution, "prompts", `attempt-${attempt}-resume.txt`),
176
+ `${agentInvocationPromptArgument(resumeInvocation)}\n`,
177
+ );
178
+ const stageArchive = path.join(temporaryDirectory, `compile-stage-a${attempt}.tar.gz`);
179
+ const resultArchive = path.join(temporaryDirectory, `compile-result-a${attempt}.tar.gz`);
180
+ await createAutoDlStageArchive(bundle, stageArchive);
181
+ const remote = await executeAutoDlRemoteAttempt({
182
+ bundle,
183
+ lease,
184
+ config,
185
+ invocation,
186
+ resumeInvocation,
187
+ attempt,
188
+ stageArchive,
189
+ resultArchive,
190
+ checkpointPolicy: config.checkpointing,
191
+ restoreCheckpointOnStart: attempt > 1,
192
+ reuseCompletedResultOnRestore: false,
193
+ completionRelativePath: "output/research.json",
194
+ completionHasEvidence: false,
195
+ secretManager: autoDlOptions.secretManager,
196
+ stageTransfer: autoDlOptions.stageTransfer,
197
+ stageClients: autoDlOptions.clients,
198
+ assertActive: autoDlOptions.assertActive,
199
+ progress,
200
+ });
201
+ records.push(remote.record);
202
+ const researchPath = path.join(bundle.directories.output, "research.json");
203
+ await recoverJsonFromFinalMessage({
204
+ finalMessagePath: hostRuntimePath(bundle, CODEX_FINAL_MESSAGE_PATH),
205
+ outputPath: researchPath,
206
+ });
207
+ const remoteFailure = await describeRemoteAgentFailure(bundle, attempt);
208
+ const providerFailure = classifyProviderActionRequired(remoteFailure?.message);
209
+ if (providerFailure) {
210
+ blockingFailure = providerFailure;
211
+ validationIssues = [providerFailure.message];
212
+ break;
213
+ }
214
+ // A syntactically valid compiler file can still be an unfinished draft.
215
+ // Transport publication errors retain their independent recovery path.
216
+ if (remoteFailure && remoteFailure.failureCategory !== "publication") {
217
+ blockingFailure = { code: remoteFailure.failureCode ?? "compiler.agent_incomplete",
218
+ failureCategory: remoteFailure.failureCategory, retryable: remoteFailure.retryable };
219
+ validationIssues = [remoteFailure.message];
220
+ break;
221
+ }
222
+ try {
223
+ if (!(await pathExists(researchPath))) {
224
+ throw new CiteArkError("缺少 /job/output/research.json");
225
+ }
226
+ const accepted = await acceptCompilerDraft({ bundle, draft: await readJson(researchPath), validateResearch, sourceReviewer });
227
+ normalizedResearch = accepted.research;
228
+ normalizationWarnings = accepted.normalizationWarnings;
229
+ validationIssues = [];
230
+ break;
231
+ } catch (error) {
232
+ validationIssues = [
233
+ ...(remoteFailure ? [remoteFailure.message] : []),
234
+ error instanceof Error ? error.message : String(error),
235
+ ];
236
+ if (error.failureCode === "compiler.source_review_unavailable") { blockingFailure = { code: error.failureCode, retryable: false }; break; }
237
+ if (attempt < maxAttempts) {
238
+ progress(`Research Compiler 结果校验失败,将在同一 AutoDL 实例继续:${validationIssues.join(";")}`);
239
+ }
240
+ }
241
+ }
242
+
243
+ await ensureFinalWorkspaceCapture(bundle);
244
+ const usage = await readAgentUsage(
245
+ path.join(bundle.directories.execution, "trace.jsonl"),
246
+ { runtimeHome: bundle.directories.runtimeHome, model: runtimeAgent.model },
247
+ );
248
+ const status = validationIssues.length ? "invalid_result" : "completed";
249
+ await updateRunState(bundle, {
250
+ status,
251
+ finishedAt: new Date().toISOString(),
252
+ validationIssues,
253
+ normalizationWarnings,
254
+ output: normalizedResearch ? "output/research.normalized.json" : null,
255
+ sandboxImageIdentity: {
256
+ id: config.imageUuid,
257
+ repositoryDigests: [],
258
+ },
259
+ cloudExecution: publicAutoDlExecution(config, lease, records),
260
+ usage,
261
+ });
262
+ if (validationIssues.length || !normalizedResearch) {
263
+ throw new CiteArkError(
264
+ `研究编译失败:\n- ${validationIssues.join("\n- ")}`,
265
+ {
266
+ exitCode: 2,
267
+ retryable: blockingFailure?.retryable,
268
+ failureCode: blockingFailure?.code,
269
+ failureCategory: blockingFailure?.failureCategory,
270
+ recoveryAction: blockingFailure?.recoveryAction,
271
+ },
272
+ );
273
+ }
274
+ successful = true;
275
+ return {
276
+ dryRun: false,
277
+ bundle,
278
+ research: normalizedResearch,
279
+ normalizationWarnings,
280
+ runs: records,
281
+ };
282
+ } catch (error) {
283
+ if (bundle.state.status === "running") {
284
+ await updateRunState(bundle, {
285
+ status: "failed",
286
+ finishedAt: new Date().toISOString(),
287
+ error: error instanceof Error ? error.message : String(error),
288
+ cloudExecution: publicAutoDlExecution(config, lease, records),
289
+ });
290
+ }
291
+ throw error;
292
+ } finally {
293
+ if (lease) {
294
+ const cleanup = await lease.close({ successful }).catch(error => ({ released: false, error: error.message }));
295
+ await recordResearchCleanup({ provider: 'autodl', resource: lease.instanceUuid, status: cleanup?.released ? 'released' : 'failed', ...(cleanup?.error ? { error: cleanup.error } : {}) });
296
+ }
297
+ await rm(temporaryDirectory, { recursive: true, force: true });
298
+ }
299
+ }
300
+
301
+ function publicAutoDlExecution(config, lease, records) {
302
+ const runSeconds = records.reduce(
303
+ (total, record) => total + (Number(record.durationSeconds) || 0),
304
+ 0,
305
+ );
306
+ const hourlyCny = lease?.observedHourlyCny ?? config.hourlyCny;
307
+ return {
308
+ provider: "autodl",
309
+ service: "container-instance-pro",
310
+ instance: lease?.publicRecord() ?? null,
311
+ gpuSpecUuid: lease?.config.gpuSpecUuid ?? config.gpuSpecUuid,
312
+ imageUuid: config.imageUuid,
313
+ inputStaging: {
314
+ preferred: config.stagingBucket && config.stageSigningServiceAccount
315
+ ? "gcs-signed-url"
316
+ : "sftp",
317
+ fallback: "sftp",
318
+ signedUrlExpiresSeconds: config.stageSignedUrlExpiresSeconds,
319
+ },
320
+ credentialInjection: {
321
+ mode: "secret-manager-to-ephemeral-ssh-environment",
322
+ valueRecorded: false,
323
+ },
324
+ computeUsage: {
325
+ runSeconds,
326
+ observedHourlyCny: hourlyCny ?? null,
327
+ estimatedCostCny: typeof hourlyCny === "number"
328
+ ? Number(((runSeconds / 3600) * hourlyCny).toFixed(6))
329
+ : null,
330
+ invoiceReconciled: false,
331
+ },
332
+ runs: structuredClone(records),
333
+ };
334
+ }
335
+
336
+ function hostRuntimePath(bundle, containerPath) {
337
+ return path.join(bundle.directories.run, containerPath.replace(/^\/job\//, ""));
338
+ }
339
+
340
+ function assertCompilerProfile(profile) {
341
+ if (!profile || profile.executor !== "autodl-pro") {
342
+ throw new CiteArkError("AutoDL Research Compiler 必须使用 autodl-pro 资源规格");
343
+ }
344
+ if (typeof profile.image !== "string" || !profile.image) {
345
+ throw new CiteArkError("AutoDL Research Compiler 资源规格必须声明私有镜像标识");
346
+ }
347
+ }