@piagent/platform 1.6.0 → 1.7.1

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 (736) hide show
  1. package/CHANGELOG.md +104 -0
  2. package/README.md +24 -15
  3. package/README.vi.md +36 -2
  4. package/adapters/common/assurance-producers.json +109 -0
  5. package/adapters/common/structured-assurance-facts.mjs +153 -0
  6. package/adapters/node-typescript/contract-families.json +2811 -0
  7. package/adapters/node-typescript/contract-families.md +90 -0
  8. package/benchmarks/production-v1/grade.mjs +20 -0
  9. package/benchmarks/production-v1/spend-control.v1.json +27 -3
  10. package/benchmarks/production-v1/suite.json +5 -1
  11. package/benchmarks/production-v2/grade.mjs +892 -0
  12. package/benchmarks/production-v2/project/README.md +7 -0
  13. package/benchmarks/production-v2/project/config/service.json +4 -0
  14. package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
  15. package/benchmarks/production-v2/project/docs/ops.md +3 -0
  16. package/benchmarks/production-v2/project/logs/incident.log +2 -0
  17. package/benchmarks/production-v2/project/package.json +11 -0
  18. package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
  19. package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
  20. package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
  21. package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
  22. package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
  23. package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
  24. package/benchmarks/production-v2/project/src/data/csv.js +3 -0
  25. package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
  26. package/benchmarks/production-v2/project/src/data/migration.js +9 -0
  27. package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
  28. package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
  29. package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
  30. package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
  31. package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
  32. package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
  33. package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
  34. package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
  35. package/benchmarks/production-v2/project/src/platform/args.js +10 -0
  36. package/benchmarks/production-v2/project/src/platform/config.js +7 -0
  37. package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
  38. package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
  39. package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
  40. package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
  41. package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
  42. package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
  43. package/benchmarks/production-v2/project/vendor/README.md +1 -0
  44. package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
  45. package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
  46. package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
  47. package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
  48. package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
  49. package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
  50. package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
  51. package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
  52. package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
  53. package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
  54. package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
  55. package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
  56. package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
  57. package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
  58. package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
  59. package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
  60. package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
  61. package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
  62. package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
  63. package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
  64. package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
  65. package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
  66. package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
  67. package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
  68. package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
  69. package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
  70. package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
  71. package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
  72. package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
  73. package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
  74. package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
  75. package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
  76. package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
  77. package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
  78. package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
  79. package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
  80. package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
  81. package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
  82. package/benchmarks/production-v2/spend-control.v1.json +115 -0
  83. package/benchmarks/production-v2/suite.json +1009 -0
  84. package/benchmarks/production-v2/variant.mjs +169 -0
  85. package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
  86. package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
  87. package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
  88. package/benchmarks/production-v3/grade.mjs +1017 -0
  89. package/benchmarks/production-v3/project/README.md +7 -0
  90. package/benchmarks/production-v3/project/config/service.json +4 -0
  91. package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
  92. package/benchmarks/production-v3/project/docs/ops.md +3 -0
  93. package/benchmarks/production-v3/project/logs/incident.log +2 -0
  94. package/benchmarks/production-v3/project/package.json +11 -0
  95. package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
  96. package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
  97. package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
  98. package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
  99. package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
  100. package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
  101. package/benchmarks/production-v3/project/src/data/csv.js +3 -0
  102. package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
  103. package/benchmarks/production-v3/project/src/data/migration.js +9 -0
  104. package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
  105. package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
  106. package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
  107. package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
  108. package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
  109. package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
  110. package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
  111. package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
  112. package/benchmarks/production-v3/project/src/platform/args.js +10 -0
  113. package/benchmarks/production-v3/project/src/platform/config.js +7 -0
  114. package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
  115. package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
  116. package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
  117. package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
  118. package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
  119. package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
  120. package/benchmarks/production-v3/project/vendor/README.md +1 -0
  121. package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
  122. package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
  123. package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
  124. package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
  125. package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
  126. package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
  127. package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
  128. package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
  129. package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
  130. package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
  131. package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
  132. package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
  133. package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
  134. package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
  135. package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
  136. package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
  137. package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
  138. package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
  139. package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
  140. package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
  141. package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
  142. package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
  143. package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
  144. package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
  145. package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
  146. package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
  147. package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
  148. package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
  149. package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
  150. package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
  151. package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
  152. package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
  153. package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
  154. package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
  155. package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
  156. package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
  157. package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
  158. package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
  159. package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
  160. package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
  161. package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
  162. package/benchmarks/production-v3/spend-control.v1.json +121 -0
  163. package/benchmarks/production-v3/suite.json +1010 -0
  164. package/benchmarks/production-v3/variant.mjs +178 -0
  165. package/catalog/capabilities.json +16 -16
  166. package/docs/authority-manifest.md +1 -1
  167. package/docs/auto-delegation-policy.md +38 -73
  168. package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
  169. package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
  170. package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
  171. package/docs/benchmark-public-contract-addendum-v2.md +85 -0
  172. package/docs/benchmark-public-contract-addendum-v3.md +116 -0
  173. package/docs/command-reference-vietnamese.md +30 -35
  174. package/docs/context-engine.md +5 -3
  175. package/docs/context-window-policy.md +37 -1
  176. package/docs/distribution-standard.md +9 -9
  177. package/docs/en/architecture.md +26 -13
  178. package/docs/languages.json +4 -4
  179. package/docs/operator-manual-vietnamese.md +26 -46
  180. package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
  181. package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
  182. package/docs/piagent-core-improvement-plan.md +528 -0
  183. package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
  184. package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
  185. package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
  186. package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
  187. package/docs/piagent-open-session-wait.md +19 -0
  188. package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
  189. package/docs/publishing-for-teams.md +7 -0
  190. package/docs/quality-benchmark.md +299 -23
  191. package/docs/quickstart-vietnamese.md +2 -2
  192. package/docs/release-install-policy.md +32 -9
  193. package/docs/research/harness-next-evidence-review.md +113 -0
  194. package/docs/runtime-harness-standard.md +3 -3
  195. package/docs/security-threat-model.md +1 -1
  196. package/docs/subagent-orchestration-capabilities.md +19 -51
  197. package/docs/subagents-and-multiagent.md +21 -47
  198. package/docs/task-implementation-contract.md +3 -3
  199. package/docs/task-lifecycle-vietnamese.md +1 -1
  200. package/docs/team-onboarding.md +7 -7
  201. package/docs/third-party-neutrality.md +7 -0
  202. package/docs/usage-observability.md +1 -1
  203. package/docs/vi/architecture.md +26 -13
  204. package/docs-site/architecture.html +1 -1
  205. package/docs-site/benchmark.html +89 -95
  206. package/docs-site/commands.html +1 -1
  207. package/docs-site/content/benchmark.html +88 -94
  208. package/docs-site/content/en/benchmark.html +134 -90
  209. package/docs-site/content/en/index.html +19 -19
  210. package/docs-site/content/en/runtime.html +5 -5
  211. package/docs-site/content/en/whats-new.html +23 -1
  212. package/docs-site/content/en/workflows.html +2 -2
  213. package/docs-site/content/index.html +19 -19
  214. package/docs-site/content/runtime.html +6 -6
  215. package/docs-site/content/whats-new.html +23 -1
  216. package/docs-site/content/workflows.html +2 -2
  217. package/docs-site/context-engine.html +1 -1
  218. package/docs-site/en/architecture.html +1 -1
  219. package/docs-site/en/benchmark.html +135 -91
  220. package/docs-site/en/commands.html +1 -1
  221. package/docs-site/en/context-engine.html +1 -1
  222. package/docs-site/en/index.html +20 -20
  223. package/docs-site/en/mcp-approval.html +1 -1
  224. package/docs-site/en/mcp-auth.html +1 -1
  225. package/docs-site/en/mcp-servers.html +1 -1
  226. package/docs-site/en/mcp.html +1 -1
  227. package/docs-site/en/permissions.html +1 -1
  228. package/docs-site/en/profiles.html +1 -1
  229. package/docs-site/en/quickstart.html +1 -1
  230. package/docs-site/en/runtime.html +6 -6
  231. package/docs-site/en/scope.html +1 -1
  232. package/docs-site/en/security.html +1 -1
  233. package/docs-site/en/team.html +1 -1
  234. package/docs-site/en/whats-new.html +24 -2
  235. package/docs-site/en/workflows.html +3 -3
  236. package/docs-site/index.html +20 -20
  237. package/docs-site/mcp-approval.html +1 -1
  238. package/docs-site/mcp-auth.html +1 -1
  239. package/docs-site/mcp-servers.html +1 -1
  240. package/docs-site/mcp.html +1 -1
  241. package/docs-site/permissions.html +1 -1
  242. package/docs-site/profiles.html +1 -1
  243. package/docs-site/quickstart.html +1 -1
  244. package/docs-site/runtime.html +7 -7
  245. package/docs-site/scope.html +1 -1
  246. package/docs-site/security.html +1 -1
  247. package/docs-site/team.html +1 -1
  248. package/docs-site/whats-new.html +24 -2
  249. package/docs-site/workflows.html +3 -3
  250. package/evals/architecture-conformance-v1/lane.json +18 -0
  251. package/evals/architecture-conformance-v1/runner.mjs +77 -0
  252. package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
  253. package/evals/fixtures/approved-host-contracts.valid.json +43 -0
  254. package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
  255. package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
  256. package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
  257. package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
  258. package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
  259. package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
  260. package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
  261. package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
  262. package/evals/fixtures/composite-criterion.invalid.json +10 -0
  263. package/evals/fixtures/composite-criterion.valid.json +33 -0
  264. package/evals/fixtures/contract-family-library.invalid.json +42 -0
  265. package/evals/fixtures/contract-family-library.valid.json +41 -0
  266. package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
  267. package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
  268. package/evals/fixtures/helper-request.invalid.json +1 -1
  269. package/evals/fixtures/helper-request.valid.json +1 -1
  270. package/evals/fixtures/host-contract-plan.invalid.json +42 -0
  271. package/evals/fixtures/host-contract-plan.valid.json +41 -0
  272. package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
  273. package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
  274. package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
  275. package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
  276. package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
  277. package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
  278. package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
  279. package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
  280. package/evals/fs4-readiness-matrix.v1.json +2 -2
  281. package/evals/fs5-artifact-archive/README.md +32 -0
  282. package/evals/fs5-artifact-archive/manifest.json +148 -0
  283. package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
  284. package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
  285. package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
  286. package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
  287. package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
  288. package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
  289. package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
  290. package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
  291. package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
  292. package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
  293. package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
  294. package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
  295. package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
  296. package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
  297. package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
  298. package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
  299. package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
  300. package/evals/fs5-causal-arm.v1.json +10 -6
  301. package/evals/fs5-pilot-protocol.v1.json +3 -3
  302. package/evals/fs5-pilot-protocol.v2.json +8 -8
  303. package/evals/fs5-pilot-protocol.v3.json +8 -8
  304. package/evals/fs5-pilot-protocol.v4.json +7 -7
  305. package/evals/fs5-pilot-protocol.v5.json +10 -8
  306. package/evals/harness-next/development-corpus.mjs +106 -0
  307. package/evals/harness-next/resource-budget-probe.mjs +38 -0
  308. package/evals/harness-next/run-calibration.mjs +83 -0
  309. package/evals/harness-next/run-resource-probe.mjs +55 -0
  310. package/evals/harness-next/selected-development-corpus.mjs +27 -0
  311. package/evals/harness-next.v1.json +51 -0
  312. package/evals/long-horizon-v1/README.md +19 -0
  313. package/evals/long-horizon-v1/runner.mjs +116 -48
  314. package/evals/long-horizon-v1/worker.mjs +77 -30
  315. package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
  316. package/evals/private-holdout-v1/README.md +5 -2
  317. package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
  318. package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
  319. package/evals/runtime-conformance-v1/README.md +43 -0
  320. package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
  321. package/evals/runtime-conformance-v1/lane.json +57 -0
  322. package/evals/runtime-conformance-v1/paid-canary.json +31 -0
  323. package/evals/runtime-conformance-v1/runner.mjs +456 -0
  324. package/package.json +9 -4
  325. package/packages/piagent-core/README.md +15 -6
  326. package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
  327. package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
  328. package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
  329. package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
  330. package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
  331. package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
  332. package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
  333. package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
  334. package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
  335. package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
  336. package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
  337. package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
  338. package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
  339. package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
  340. package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
  341. package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
  342. package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
  343. package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
  344. package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
  345. package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
  346. package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
  347. package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
  348. package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
  349. package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
  350. package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
  351. package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
  352. package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
  353. package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
  354. package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
  355. package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
  356. package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
  357. package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
  358. package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
  359. package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
  360. package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
  361. package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
  362. package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
  363. package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
  364. package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
  365. package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
  366. package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
  367. package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
  368. package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
  369. package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
  370. package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
  371. package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
  372. package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
  373. package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
  374. package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
  375. package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
  376. package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
  377. package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
  378. package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
  379. package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
  380. package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
  381. package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
  382. package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
  383. package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
  384. package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
  385. package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
  386. package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
  387. package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
  388. package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
  389. package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
  390. package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
  391. package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
  392. package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
  393. package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
  394. package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
  395. package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
  396. package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
  397. package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
  398. package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
  399. package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
  400. package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
  401. package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
  402. package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
  403. package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
  404. package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
  405. package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
  406. package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
  407. package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
  408. package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
  409. package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
  410. package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
  411. package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
  412. package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
  413. package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
  414. package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
  415. package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
  416. package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
  417. package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
  418. package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
  419. package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
  420. package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
  421. package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
  422. package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
  423. package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
  424. package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
  425. package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
  426. package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
  427. package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
  428. package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
  429. package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
  430. package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
  431. package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
  432. package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
  433. package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
  434. package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
  435. package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
  436. package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
  437. package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
  438. package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
  439. package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
  440. package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
  441. package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
  442. package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
  443. package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
  444. package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
  445. package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
  446. package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
  447. package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
  448. package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
  449. package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
  450. package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
  451. package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
  452. package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
  453. package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
  454. package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
  455. package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
  456. package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
  457. package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
  458. package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
  459. package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
  460. package/packages/piagent-core/extensions/context-engine.js +12 -17
  461. package/packages/piagent-core/extensions/context-index-policy.js +4 -0
  462. package/packages/piagent-core/extensions/context-query-paths.js +57 -0
  463. package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
  464. package/packages/piagent-core/extensions/criterion-graph.js +4 -1
  465. package/packages/piagent-core/extensions/failure-types.ts +2 -2
  466. package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
  467. package/packages/piagent-core/extensions/guard-types.ts +9 -1
  468. package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
  469. package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
  470. package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
  471. package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
  472. package/packages/piagent-core/extensions/task-contract-view.js +4 -4
  473. package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
  474. package/packages/piagent-core/extensions/task-journal.js +48 -3
  475. package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
  476. package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
  477. package/packages/piagent-core/extensions/task-state.js +140 -43
  478. package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
  479. package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
  480. package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
  481. package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
  482. package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
  483. package/packages/piagent-core/extensions/workspace-revision.js +48 -0
  484. package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
  485. package/packages/piagent-core/package.json +3 -3
  486. package/packages/piagent-core/policies/base-policy.json +43 -2
  487. package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
  488. package/packages/piagent-core/prompts/be-to-fe.md +2 -8
  489. package/packages/piagent-core/prompts/plan.md +1 -1
  490. package/packages/piagent-core/prompts/platform-improve.md +2 -6
  491. package/packages/piagent-core/prompts/review.md +2 -7
  492. package/packages/piagent-core/prompts/scout.md +1 -1
  493. package/packages/piagent-core/prompts/task.md +4 -3
  494. package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
  495. package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
  496. package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
  497. package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
  498. package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
  499. package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
  500. package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
  501. package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
  502. package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
  503. package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
  504. package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
  505. package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
  506. package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
  507. package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
  508. package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
  509. package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
  510. package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
  511. package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
  512. package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
  513. package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
  514. package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
  515. package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
  516. package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
  517. package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
  518. package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
  519. package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
  520. package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
  521. package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
  522. package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
  523. package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
  524. package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
  525. package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
  526. package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
  527. package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
  528. package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
  529. package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
  530. package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
  531. package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
  532. package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
  533. package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
  534. package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
  535. package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
  536. package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
  537. package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
  538. package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
  539. package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
  540. package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
  541. package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
  542. package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
  543. package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
  544. package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
  545. package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
  546. package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
  547. package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
  548. package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
  549. package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
  550. package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
  551. package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
  552. package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
  553. package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
  554. package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
  555. package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
  556. package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
  557. package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
  558. package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
  559. package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
  560. package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
  561. package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
  562. package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
  563. package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
  564. package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
  565. package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
  566. package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
  567. package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
  568. package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
  569. package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
  570. package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
  571. package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
  572. package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
  573. package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
  574. package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
  575. package/packages/piagent-core/runtime/session/usage.ts +4 -3
  576. package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
  577. package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
  578. package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
  579. package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
  580. package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
  581. package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
  582. package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
  583. package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
  584. package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
  585. package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
  586. package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
  587. package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
  588. package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
  589. package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
  590. package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
  591. package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
  592. package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
  593. package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
  594. package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
  595. package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
  596. package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
  597. package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
  598. package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
  599. package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
  600. package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
  601. package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
  602. package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
  603. package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
  604. package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
  605. package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
  606. package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
  607. package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
  608. package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
  609. package/packages/piagent-core/subagents/piagent-planner.md +6 -4
  610. package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
  611. package/packages/piagent-core/subagents/piagent-scout.md +4 -2
  612. package/packages/piagent-core/subagents/piagent-worker.md +6 -4
  613. package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
  614. package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
  615. package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
  616. package/packages/piagent-webui/dist/client/index.html +2 -2
  617. package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
  618. package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
  619. package/packages/piagent-webui/extension/review-controller.ts +29 -6
  620. package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
  621. package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
  622. package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
  623. package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
  624. package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
  625. package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
  626. package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
  627. package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
  628. package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
  629. package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
  630. package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
  631. package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
  632. package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
  633. package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
  634. package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
  635. package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
  636. package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
  637. package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
  638. package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
  639. package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
  640. package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
  641. package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
  642. package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
  643. package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
  644. package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
  645. package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
  646. package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
  647. package/packages/piagent-webui/server/loopback-server.ts +9 -7
  648. package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
  649. package/packages/piagent-webui/server/transcript-projection.ts +150 -11
  650. package/packages/piagent-webui/shared/message-correlation.ts +25 -0
  651. package/schemas/approved-host-contracts.schema.json +973 -0
  652. package/schemas/benchmark-attempt-outcome.schema.json +303 -0
  653. package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
  654. package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
  655. package/schemas/composite-criterion.schema.json +788 -0
  656. package/schemas/contract-family-library.schema.json +28 -0
  657. package/schemas/contract-selection-recipe.schema.json +31 -0
  658. package/schemas/failure-evidence.schema.json +2 -2
  659. package/schemas/helper-request.schema.json +16 -4
  660. package/schemas/host-contract-plan.schema.json +25 -0
  661. package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
  662. package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
  663. package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
  664. package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
  665. package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
  666. package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
  667. package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
  668. package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
  669. package/schemas/project-profile.schema.json +1 -1
  670. package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
  671. package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
  672. package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
  673. package/schemas/public-contracts/revocation-v1.schema.json +87 -0
  674. package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
  675. package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
  676. package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
  677. package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
  678. package/schemas/task-contract.schema.json +45 -0
  679. package/scripts/approve-independent-verification.mjs +57 -0
  680. package/scripts/benchmark-arm-observer.mjs +320 -0
  681. package/scripts/benchmark-budget-ipc.mjs +34 -0
  682. package/scripts/benchmark-budget-launcher.mjs +91 -0
  683. package/scripts/benchmark-budget-processes.mjs +133 -0
  684. package/scripts/benchmark-budget-publication.mjs +228 -0
  685. package/scripts/benchmark-budget-runtime.mjs +136 -0
  686. package/scripts/benchmark-candidate-ablation.mjs +265 -0
  687. package/scripts/benchmark-codex-journey.mjs +406 -0
  688. package/scripts/benchmark-independent-verification.mjs +628 -0
  689. package/scripts/benchmark-journey-outcome.mjs +92 -0
  690. package/scripts/benchmark-production-v3-human-review.mjs +192 -0
  691. package/scripts/benchmark-public-input-catalog.mjs +158 -0
  692. package/scripts/benchmark-registered-plan-author.mjs +355 -0
  693. package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
  694. package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
  695. package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
  696. package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
  697. package/scripts/benchmark-runner-completion.mjs +133 -0
  698. package/scripts/benchmark-runner-configuration.mjs +159 -0
  699. package/scripts/benchmark-runner-core.mjs +224 -387
  700. package/scripts/benchmark-runner-finalization.mjs +144 -27
  701. package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
  702. package/scripts/benchmark-runner-schedule.mjs +256 -0
  703. package/scripts/benchmark-runner-support.mjs +136 -4
  704. package/scripts/benchmark-runner.mjs +95 -18
  705. package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
  706. package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
  707. package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
  708. package/scripts/benchmark-scoped-pi-router.mjs +196 -0
  709. package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
  710. package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
  711. package/scripts/benchmark-scoped-session-custody.mjs +599 -0
  712. package/scripts/benchmark-scoped-session-factory.mjs +342 -0
  713. package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
  714. package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
  715. package/scripts/benchmark-session-evaluator.mjs +169 -0
  716. package/scripts/benchmark-session-provider-boundary.mjs +62 -0
  717. package/scripts/benchmark-session-usage.mjs +28 -0
  718. package/scripts/benchmark-session.mjs +343 -50
  719. package/scripts/benchmark-webui-journey.mjs +678 -0
  720. package/scripts/check-public-wording.mjs +46 -0
  721. package/scripts/configure-subagents.sh +42 -34
  722. package/scripts/fs4-readiness-evaluation.mjs +7 -1
  723. package/scripts/helper-evaluation.mjs +1 -1
  724. package/scripts/phase-tools-evaluation.mjs +1 -0
  725. package/scripts/pi-auto.sh +7 -1
  726. package/scripts/pi-usage-history.mjs +201 -11
  727. package/scripts/piagent-cli.mjs +8 -0
  728. package/scripts/private-holdout-readiness.mjs +7 -1
  729. package/scripts/public-evaluation-exposure.mjs +140 -0
  730. package/scripts/select-independent-verification.mjs +62 -0
  731. package/scripts/team-doctor.sh +1 -1
  732. package/scripts/typescript-loader.mjs +135 -1
  733. package/scripts/update-global.mjs +6 -7
  734. package/scripts/verify-local.sh +41 -55
  735. package/scripts/verify-release-identity.mjs +5 -3
  736. package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
@@ -0,0 +1,337 @@
1
+ import { createHash, createHmac, createSecretKey, randomBytes, timingSafeEqual } from "node:crypto";
2
+ import fs from "node:fs";
3
+ import path from "node:path";
4
+ import { discoverRuntimeIntegrityFiles } from "../capabilities/runtime-integrity.js";
5
+ import { compileCompositeContract } from "./acceptance-composite-contract.js";
6
+ import { compileIndependentContract } from "./acceptance-independent-contract.js";
7
+ import { openAcceptanceEvidenceStore } from "./acceptance-evidence-store.js";
8
+ import { validateModulePaths } from "./acceptance-executor/module-graph.mjs";
9
+ import { validateContractSelection } from "./acceptance-contract-selection.js";
10
+ import { expectedNodeProfile } from "./acceptance-executor/node-profile.mjs";
11
+
12
+ export const HOST_CONTRACT_CONFIGURATION_VERSION = "approved-host-contracts-v1";
13
+ export const HOST_CONTRACT_SET_VERSION = "approved-host-contract-set-v1";
14
+ export const NODE_HOST_CONTRACT_CONFIGURATION_VERSION = "approved-host-contracts-v2";
15
+ export const NODE_HOST_CONTRACT_SET_VERSION = "approved-host-contract-set-v2";
16
+ const HASH = /^[a-f0-9]{64}$/;
17
+ const hash = (value) => createHash("sha256").update(value).digest("hex");
18
+ const freeze = (value) => { if (value && typeof value === "object") { Object.values(value).forEach(freeze); Object.freeze(value); } return value; };
19
+ const exact = (value, fields) => value && typeof value === "object" && !Array.isArray(value)
20
+ && Object.keys(value).length === fields.length && fields.every((field) => Object.hasOwn(value, field));
21
+ const COMPOSITE_PLAN_CONTEXT_VERSION = "composite-plan-context-v1";
22
+ const MATERIAL_PATH = /^(?!\/)(?!.*\/\/)(?!.*\/$)(?!.*(?:^|\/)(?:\.{1,2}|\.git|\.pi|node_modules)(?:\/|$))[^\\\0-\x1f\x7f:%?#]+$/;
23
+
24
+ function parseStrictCompositeDeclarations(raw) {
25
+ if (typeof raw !== "string" || !raw.isWellFormed() || Buffer.byteLength(raw) < 1 || Buffer.byteLength(raw) > 1024 * 1024) {
26
+ throw new TypeError("Invalid composite declaration bytes");
27
+ }
28
+ let value;
29
+ try { value = JSON.parse(raw); } catch { throw new TypeError("Invalid composite declaration JSON"); }
30
+ const stack = [];
31
+ for (const match of raw.matchAll(/"(?:\\.|[^"\\])*"|[{}\[\],:]/g)) {
32
+ const token = match[0], current = stack.at(-1);
33
+ if (token === "{") stack.push({ keys: new Set(), key: true });
34
+ else if (token === "[") stack.push({ keys: null });
35
+ else if (token === "}" || token === "]") stack.pop();
36
+ else if (current?.keys && token === ",") current.key = true;
37
+ else if (current?.keys && token === ":") current.key = false;
38
+ else if (current?.keys && current.key && token.startsWith('"')) {
39
+ const key = JSON.parse(token);
40
+ if (current.keys.has(key)) throw new TypeError("Duplicate composite declaration key");
41
+ current.keys.add(key);
42
+ }
43
+ if (stack.length > 20) throw new TypeError("Composite declaration nesting exceeds bound");
44
+ }
45
+ return value;
46
+ }
47
+
48
+ function validateCompositeHostContract(contract, backend) {
49
+ if (!exact(contract, ["route", "criterionId", "criterionHash", "maxAttempts", "planContext"])
50
+ || contract.route !== "composite" || typeof contract.criterionId !== "string" || !/^[a-z][a-z0-9._-]{0,63}$/.test(contract.criterionId)
51
+ || !HASH.test(contract.criterionHash) || !Number.isSafeInteger(contract.maxAttempts) || contract.maxAttempts < 1 || contract.maxAttempts > 8
52
+ || !exact(contract.planContext, ["version", "contractText", "declarationsText", "identity", "materialBindings"])
53
+ || contract.planContext.version !== COMPOSITE_PLAN_CONTEXT_VERSION || typeof contract.planContext.contractText !== "string"
54
+ || !contract.planContext.contractText.isWellFormed() || Buffer.byteLength(contract.planContext.contractText) < 1
55
+ || Buffer.byteLength(contract.planContext.contractText) > 1024 * 1024) throw new TypeError("Invalid approved composite criterion contract");
56
+ const declarations = parseStrictCompositeDeclarations(contract.planContext.declarationsText);
57
+ const compiled = compileCompositeContract(contract.planContext.contractText, declarations);
58
+ if (compiled.criterionBinding !== "matched" || compiled.contract.criterionId !== contract.criterionId
59
+ || compiled.contract.criterionHash !== contract.criterionHash
60
+ || compiled.codePlans.some(plan => JSON.stringify(plan.profile) !== JSON.stringify(backend?.profile))) {
61
+ throw new TypeError("Approved composite criterion binding mismatch");
62
+ }
63
+ const identity = contract.planContext.identity;
64
+ if (!exact(identity, ["suiteDigest", "configDigest", "armDigest"])
65
+ || ![identity.suiteDigest, identity.configDigest, identity.armDigest].every(value => HASH.test(value))) {
66
+ throw new TypeError("Invalid approved composite identity");
67
+ }
68
+ const bindings = contract.planContext.materialBindings;
69
+ if (!Array.isArray(bindings) || bindings.length > 32 || bindings.length !== declarations.materials.length) {
70
+ throw new TypeError("Invalid approved composite material bindings");
71
+ }
72
+ const declared = new Map(declarations.materials.map(material => [material.id, material])), ids = new Set(), paths = new Set();
73
+ for (const binding of bindings) {
74
+ if (!exact(binding, ["id", "mode", "relativePath", "sha256"]) || !declared.has(binding.id) || ids.has(binding.id)
75
+ || !["frozen", "current", "protected"].includes(binding.mode) || typeof binding.relativePath !== "string"
76
+ || binding.relativePath.length > 1024 || !MATERIAL_PATH.test(binding.relativePath) || paths.has(binding.relativePath)
77
+ || binding.mode === "frozen" !== HASH.test(binding.sha256 ?? "")
78
+ || binding.mode !== "frozen" && binding.sha256 !== null
79
+ || binding.mode === "protected" && declared.get(binding.id).byteLength !== 0) {
80
+ throw new TypeError("Invalid approved composite material binding");
81
+ }
82
+ ids.add(binding.id); paths.add(binding.relativePath);
83
+ }
84
+ return compiled;
85
+ }
86
+
87
+ export function installedContractVerifierDigest(root) {
88
+ const canonical = fs.realpathSync.native(root), digest = createHash("sha256");
89
+ digest.update(HOST_CONTRACT_CONFIGURATION_VERSION);
90
+ const files = discoverRuntimeIntegrityFiles(canonical);
91
+ const familyLibrary = "adapters/node-typescript/contract-families.json";
92
+ if (fs.existsSync(path.join(canonical, familyLibrary)) && !files.includes(familyLibrary)) files.push(familyLibrary);
93
+ for (const required of ["packages/piagent-core/extensions/acceptance-authenticated-admission.js", "packages/piagent-core/extensions/acceptance-durable-execution.js",
94
+ "packages/piagent-core/extensions/acceptance-host-configuration.js"]) if (!files.includes(required)) files.push(required);
95
+ for (const relative of files.sort()) {
96
+ const file = path.join(canonical, relative), stat = fs.lstatSync(file);
97
+ if (!stat.isFile() || stat.isSymbolicLink() || stat.size > 16 * 1024 * 1024) throw new Error("Unsafe installed verifier file");
98
+ const bytes = fs.readFileSync(file);
99
+ digest.update(JSON.stringify([relative, bytes.length])); digest.update(bytes);
100
+ }
101
+ return digest.digest("hex");
102
+ }
103
+
104
+ /** Structural/semantic validation only; a payload is not authenticated authority. */
105
+ /** Request-set native-only steps grant no criterion receipts. Unknown requests
106
+ * and mutation-capable tasks cannot borrow another request's authority. */
107
+ export function independentRequestAdmissionBlock(request, task) {
108
+ if (!request) return "independent host approval does not cover the current operator request";
109
+ if (request.nativeOnly === true && (task.changeMode !== "read-only" || task.mutationPolicy !== "forbidden")) {
110
+ return "native-only host approval requires a read-only mutation-forbidden task";
111
+ }
112
+ return undefined;
113
+ }
114
+
115
+ export function validateHostContractPayload(payload) {
116
+ if ([HOST_CONTRACT_SET_VERSION, NODE_HOST_CONTRACT_SET_VERSION].includes(payload?.version)) {
117
+ if (!exact(payload, ["version", "projectId", "verifierDigest", "plans"])
118
+ || !HASH.test(payload.projectId) || !HASH.test(payload.verifierDigest)) throw new TypeError("Invalid host contract set");
119
+ validatePlanList(payload.plans, payload.version === NODE_HOST_CONTRACT_SET_VERSION ? 3 : 1);
120
+ return payload;
121
+ }
122
+ const schemaVersion = payload?.version === NODE_HOST_CONTRACT_CONFIGURATION_VERSION ? 3
123
+ : payload?.version === HOST_CONTRACT_CONFIGURATION_VERSION ? 1 : null;
124
+ if (!exact(payload, ["version", "projectId", "operatorRequestDigest", "verifierDigest", "backend", "contracts"])
125
+ || schemaVersion === null || !HASH.test(payload.projectId)
126
+ || !HASH.test(payload.verifierDigest)) throw new TypeError("Invalid host contract approval");
127
+ validateContractBody(payload, schemaVersion);
128
+ return payload;
129
+ }
130
+
131
+ /** The same plan validation used by the operator CLI, without filesystem writes. */
132
+ export function validateHostContractPlan(plan) {
133
+ if ([2, 4].includes(plan?.schemaVersion)) {
134
+ if (!exact(plan, ["schemaVersion", "plans"])) throw new TypeError("Invalid host contract plan set");
135
+ validatePlanList(plan.plans, plan.schemaVersion === 4 ? 3 : 1);
136
+ return plan;
137
+ }
138
+ if (Object.hasOwn(plan ?? {}, "nativeOnly")) throw new TypeError("Native-only steps require an explicit request set");
139
+ validateSinglePlan(plan);
140
+ return plan;
141
+ }
142
+
143
+ function validateSinglePlan(plan) {
144
+ if (!exact(plan, ["schemaVersion", "operatorRequestDigest", "backend", "contracts",
145
+ ...(Object.hasOwn(plan ?? {}, "nativeOnly") ? ["nativeOnly"] : [])]) || ![1, 3].includes(plan.schemaVersion)) {
146
+ throw new TypeError("Invalid host contract plan");
147
+ }
148
+ validateContractBody(plan, plan.schemaVersion);
149
+ }
150
+
151
+ function validatePlanList(plans, schemaVersion) {
152
+ if (!Array.isArray(plans) || plans.length < 1 || plans.length > 32) throw new TypeError("Invalid host contract plan count");
153
+ const requests = new Set();
154
+ for (const plan of plans) {
155
+ validateSinglePlan(plan);
156
+ if (plan.schemaVersion !== schemaVersion) throw new TypeError("Mixed host contract plan versions");
157
+ if (requests.has(plan.operatorRequestDigest)) throw new TypeError("Ambiguous approved operator request");
158
+ requests.add(plan.operatorRequestDigest);
159
+ }
160
+ }
161
+
162
+ function validateContractBody(payload, schemaVersion = 1) {
163
+ const nodeProfile = schemaVersion === 3;
164
+ const hasDockerCommand = Object.hasOwn(payload.backend ?? {}, "dockerCommand");
165
+ const backendFields = nodeProfile
166
+ ? ["imageId", "dockerSocket", ...(hasDockerCommand ? ["dockerCommand"] : []), "timeoutMs", "profile"]
167
+ : ["imageId", "dockerSocket", "timeoutMs"];
168
+ if (Object.hasOwn(payload.backend ?? {}, "startupAllowanceMs")) backendFields.push("startupAllowanceMs");
169
+ if (!/^operator-request-v1:[a-f0-9]{64}$/.test(payload.operatorRequestDigest)
170
+ || !exact(payload.backend, backendFields)
171
+ || !/^sha256:[a-f0-9]{64}$/.test(payload.backend.imageId) || typeof payload.backend.dockerSocket !== "string"
172
+ || !path.isAbsolute(payload.backend.dockerSocket) || payload.backend.dockerSocket.includes("\0")
173
+ || hasDockerCommand && (!nodeProfile
174
+ || JSON.stringify(Object.keys(payload.backend.dockerCommand ?? {})) !== JSON.stringify(["path", "sha256"])
175
+ || typeof payload.backend.dockerCommand.path !== "string"
176
+ || !path.isAbsolute(payload.backend.dockerCommand.path)
177
+ || path.normalize(payload.backend.dockerCommand.path) !== payload.backend.dockerCommand.path
178
+ || payload.backend.dockerCommand.path.includes("\0")
179
+ || !HASH.test(String(payload.backend.dockerCommand.sha256 ?? "")))
180
+ || Object.hasOwn(payload.backend, "startupAllowanceMs") && (!Number.isSafeInteger(payload.backend.startupAllowanceMs)
181
+ || payload.backend.startupAllowanceMs < 0 || payload.backend.startupAllowanceMs > 60000)
182
+ || !Number.isSafeInteger(payload.backend.timeoutMs) || payload.backend.timeoutMs < 25 || payload.backend.timeoutMs > 30000
183
+ || nodeProfile && JSON.stringify(payload.backend.profile) !== JSON.stringify(expectedNodeProfile())
184
+ || !Array.isArray(payload.contracts) || payload.contracts.length > 12
185
+ || (Object.hasOwn(payload, "nativeOnly")
186
+ ? !nodeProfile || payload.nativeOnly !== true || payload.contracts.length !== 0
187
+ : payload.contracts.length < 1)) throw new TypeError("Invalid host contract approval");
188
+ const ids = new Set();
189
+ for (const contract of payload.contracts) {
190
+ if (nodeProfile && contract?.route === "composite") {
191
+ validateCompositeHostContract(contract, payload.backend);
192
+ if (ids.has(contract.criterionId)) throw new TypeError("Invalid approved criterion contract");
193
+ ids.add(contract.criterionId);
194
+ continue;
195
+ }
196
+ if (!exact(contract, [...(nodeProfile ? ["route"] : []), "criterionId", "criterionHash", "sourcePath", "exportName", "maxAttempts", "checks",
197
+ ...(Object.hasOwn(contract, "modulePaths") ? ["modulePaths"] : []), ...(Object.hasOwn(contract, "selection") ? ["selection"] : [])])
198
+ || typeof contract.criterionId !== "string" || !/^[a-z0-9][a-z0-9:._-]{0,79}$/.test(contract.criterionId)
199
+ || ids.has(contract.criterionId) || !HASH.test(contract.criterionHash) || typeof contract.sourcePath !== "string"
200
+ || !contract.sourcePath || path.isAbsolute(contract.sourcePath) || contract.sourcePath.includes("\\") || contract.sourcePath.includes("\0")
201
+ || contract.sourcePath.split("/").some((part) => ["", ".", "..", ".git", ".pi", "node_modules"].includes(part))
202
+ || nodeProfile && contract.route !== "code"
203
+ || !Number.isSafeInteger(contract.maxAttempts) || contract.maxAttempts < 1 || contract.maxAttempts > 8) throw new TypeError("Invalid approved criterion contract");
204
+ ids.add(contract.criterionId);
205
+ if (Object.hasOwn(contract, "selection")) validateContractSelection(contract.selection, contract.criterionHash);
206
+ if (Object.hasOwn(contract, "modulePaths")) validateModulePaths(contract.sourcePath, contract.modulePaths);
207
+ compileIndependentContract(JSON.stringify({ schemaVersion: nodeProfile ? 2 : 1, ...(nodeProfile ? { profile: payload.backend.profile } : {}),
208
+ source: "export const placeholder = 0;", exportName: contract.exportName, checks: contract.checks }));
209
+ }
210
+ }
211
+
212
+ function privateDirectory(directory, projectRoot) {
213
+ const root = fs.realpathSync.native(projectRoot);
214
+ if (!path.isAbsolute(directory) || fs.realpathSync.native(directory) !== directory) throw new Error("Authority path must be canonical");
215
+ const relative = path.relative(root, directory);
216
+ if (!relative || (!relative.startsWith(`..${path.sep}`) && relative !== ".." && !path.isAbsolute(relative))) throw new Error("Authority must be outside the candidate project");
217
+ const stat = fs.lstatSync(directory);
218
+ if (!stat.isDirectory() || stat.uid !== process.getuid() || (stat.mode & 0o077) !== 0) throw new Error("Authority directory must be private and host-owned");
219
+ }
220
+
221
+ function readPrivate(file, maximum) {
222
+ const fd = fs.openSync(file, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
223
+ try {
224
+ const stat = fs.fstatSync(fd);
225
+ if (!stat.isFile() || stat.nlink !== 1 || stat.uid !== process.getuid() || (stat.mode & 0o077) !== 0
226
+ || stat.size < 1 || stat.size > maximum) throw new Error("Invalid private authority file");
227
+ const bytes = fs.readFileSync(fd);
228
+ if (bytes.length !== stat.size) throw new Error("Authority changed while reading");
229
+ return bytes;
230
+ } finally { fs.closeSync(fd); }
231
+ }
232
+
233
+ const signatureVersion = payload => [NODE_HOST_CONTRACT_CONFIGURATION_VERSION, NODE_HOST_CONTRACT_SET_VERSION].includes(payload?.version)
234
+ ? NODE_HOST_CONTRACT_CONFIGURATION_VERSION : HOST_CONTRACT_CONFIGURATION_VERSION;
235
+ const signature = (key, configPath, payload) => createHmac("sha256", key)
236
+ .update(`${signatureVersion(payload)}\0${configPath}\0${JSON.stringify(payload)}`).digest("hex");
237
+
238
+ /** Read-only preview; validation creates no key, receipt, or execution. */
239
+ export function prepareHostContractApproval({ projectRoot, installedRoot, operatorRequestDigest, backend, contracts, plans }) {
240
+ if (plans !== undefined && [operatorRequestDigest, backend, contracts].some(value => value !== undefined)) {
241
+ throw new TypeError("Cannot mix a single approved request with a plan set");
242
+ }
243
+ const identity = { projectId: hash(fs.realpathSync.native(projectRoot)), verifierDigest: installedContractVerifierDigest(installedRoot) };
244
+ const nodeProfile = plans === undefined ? backend?.profile !== undefined : plans[0]?.schemaVersion === 3;
245
+ const text = JSON.stringify(plans === undefined
246
+ ? { version: nodeProfile ? NODE_HOST_CONTRACT_CONFIGURATION_VERSION : HOST_CONTRACT_CONFIGURATION_VERSION, ...identity, operatorRequestDigest, backend, contracts }
247
+ : { version: nodeProfile ? NODE_HOST_CONTRACT_SET_VERSION : HOST_CONTRACT_SET_VERSION, ...identity, plans });
248
+ if (Buffer.byteLength(text) > 2 * 1024 * 1024) throw new TypeError("Host approval is too large");
249
+ const payload = JSON.parse(text);
250
+ validateHostContractPayload(payload);
251
+ return freeze(payload);
252
+ }
253
+
254
+ /** Explicit operator setup only. A new directory is required; nothing is overwritten. */
255
+ export function writeHostContractApproval({ directory, projectRoot, installedRoot, operatorRequestDigest, backend, contracts, plans, approved = false }) {
256
+ if (approved !== true || typeof directory !== "string" || !path.isAbsolute(directory)
257
+ || path.resolve(directory) !== directory || fs.realpathSync.native(path.dirname(directory)) !== path.dirname(directory)) throw new Error("Explicit canonical host approval is required");
258
+ const root = fs.realpathSync.native(projectRoot), relative = path.relative(root, directory);
259
+ if (!relative || (!relative.startsWith(`..${path.sep}`) && relative !== ".." && !path.isAbsolute(relative))) throw new Error("Authority must be outside the candidate project");
260
+ const payload = prepareHostContractApproval({ projectRoot, installedRoot, operatorRequestDigest, backend, contracts, plans });
261
+ const configPath = path.join(directory, "approval.json"), bytes = randomBytes(32), key = createSecretKey(bytes);
262
+ const envelope = JSON.stringify({ payload, signature: signature(key, configPath, payload) });
263
+ if (Buffer.byteLength(envelope) > 2 * 1024 * 1024) throw new Error("Host approval is too large");
264
+ fs.mkdirSync(directory, { mode: 0o700 });
265
+ privateDirectory(directory, projectRoot);
266
+ function write(name, content) {
267
+ const fd = fs.openSync(path.join(directory, name), fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL | fs.constants.O_NOFOLLOW, 0o600);
268
+ try { fs.writeFileSync(fd, content); fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
269
+ }
270
+ try { write("authority.key", bytes); write("approval.json", envelope); }
271
+ finally { bytes.fill(0); }
272
+ const fd = fs.openSync(directory, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
273
+ try { fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
274
+ return Object.freeze({ configPath, verifierDigest: payload.verifierDigest });
275
+ }
276
+
277
+ export function openHostContractConfiguration({ configPath, projectRoot, installedRoot }) {
278
+ const directory = path.dirname(configPath);
279
+ privateDirectory(directory, projectRoot);
280
+ if (path.basename(configPath) !== "approval.json" || path.resolve(configPath) !== configPath) throw new Error("Invalid host approval path");
281
+ const bytes = readPrivate(path.join(directory, "authority.key"), 32);
282
+ if (bytes.length !== 32) throw new Error("Invalid host authority key");
283
+ const key = createSecretKey(bytes); bytes.fill(0);
284
+ const text = readPrivate(configPath, 2 * 1024 * 1024).toString("utf8"), envelope = JSON.parse(text);
285
+ if (!exact(envelope, ["payload", "signature"]) || !HASH.test(envelope.signature)
286
+ || !timingSafeEqual(Buffer.from(signature(key, configPath, envelope.payload), "hex"), Buffer.from(envelope.signature, "hex"))) throw new Error("Host approval is unauthenticated");
287
+ validateHostContractPayload(envelope.payload);
288
+ if (envelope.payload.projectId !== hash(fs.realpathSync.native(projectRoot)) || envelope.payload.verifierDigest !== installedContractVerifierDigest(installedRoot)) {
289
+ throw new Error("Host approval project or installed verifier changed");
290
+ }
291
+ const store = openAcceptanceEvidenceStore({ filePath: path.join(directory, "evidence.sqlite"), projectRoot, key });
292
+ const payload = freeze(envelope.payload);
293
+ const setVersion = [HOST_CONTRACT_SET_VERSION, NODE_HOST_CONTRACT_SET_VERSION].includes(payload.version);
294
+ const requests = setVersion ? payload.plans.map(plan => freeze({
295
+ version: payload.version === NODE_HOST_CONTRACT_SET_VERSION ? NODE_HOST_CONTRACT_CONFIGURATION_VERSION : HOST_CONTRACT_CONFIGURATION_VERSION,
296
+ projectId: payload.projectId, verifierDigest: payload.verifierDigest,
297
+ operatorRequestDigest: plan.operatorRequestDigest, backend: plan.backend, contracts: plan.contracts,
298
+ ...(plan.nativeOnly === true ? { nativeOnly: true } : {})
299
+ })) : [payload];
300
+ function withCompositeAuthority({ contract, binding, create }) {
301
+ const owner = payload.contracts?.includes(contract) ? payload : requests.find(request => request.contracts.includes(contract));
302
+ if (!owner) {
303
+ throw new Error("Composite contract is not owned by this host approval");
304
+ }
305
+ if (contract.route !== "composite" || typeof create !== "function") throw new TypeError("Invalid composite authority request");
306
+ const compiled = validateCompositeHostContract(contract, owner.backend), plan = contract.planContext;
307
+ if (binding.suiteDigest !== plan.identity.suiteDigest || binding.configDigest !== plan.identity.configDigest
308
+ || binding.armDigest !== plan.identity.armDigest || binding.projectId !== payload.projectId
309
+ || binding.criterionId !== contract.criterionId || binding.criterionHash !== contract.criterionHash) {
310
+ throw new Error("Composite authority binding does not match the approved plan");
311
+ }
312
+ // The KeyObject is a same-process capability handed only to the trusted
313
+ // runtime callback. It is never returned, serialized, or placed in task data.
314
+ return create(Object.freeze({ key, directory, projectRoot, compiled,
315
+ contractText: plan.contractText, declarations: parseStrictCompositeDeclarations(plan.declarationsText),
316
+ maxAttempts: contract.maxAttempts }));
317
+ }
318
+ function withCompositeRecovery(create) {
319
+ if (typeof create !== "function" || !requests.some(request => request.contracts.some(contract => contract.route === "composite"))) {
320
+ throw new TypeError("Invalid composite recovery authority request");
321
+ }
322
+ return create(Object.freeze({ key, directory, projectRoot, payload }));
323
+ }
324
+ const configuration = { payload, store,
325
+ forRequest(operatorRequestDigest) { return requests.find(request => request.operatorRequestDigest === operatorRequestDigest) ?? null; },
326
+ isCurrent() {
327
+ try {
328
+ privateDirectory(directory, projectRoot);
329
+ const currentKey = readPrivate(path.join(directory, "authority.key"), 32);
330
+ const sameKey = currentKey.length === 32 && createSecretKey(currentKey).equals(key); currentKey.fill(0);
331
+ return sameKey && readPrivate(configPath, 2 * 1024 * 1024).toString("utf8") === text
332
+ && installedContractVerifierDigest(installedRoot) === envelope.payload.verifierDigest;
333
+ } catch { return false; }
334
+ }, withCompositeAuthority, withCompositeRecovery, close: () => store.close()
335
+ };
336
+ return Object.freeze(configuration);
337
+ }
@@ -0,0 +1,25 @@
1
+ import { hasExplicitInputRejection, malformedIdentifierContractText } from "./acceptance-boundary-guidance.js";
2
+
3
+ function normalizedText(value) {
4
+ return String(value ?? "").normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase().replace(/\s+/g, " ");
5
+ }
6
+
7
+ function uniqueStrings(values) {
8
+ return [...new Set(values.filter((item) => typeof item === "string" && item.trim()).map((item) => item.trim()))];
9
+ }
10
+
11
+ /** Pair one malformed-input criterion only with the smallest related identifier declaration. */
12
+ export function malformedIdentifierCriterionText(primary, task = {}, excluded = new Set()) {
13
+ const direct = malformedIdentifierContractText(primary);
14
+ if (direct) return direct;
15
+ if (!hasExplicitInputRejection(primary)) return primary;
16
+ const candidates = uniqueStrings([task?.summary, task?.expectedOutput, ...(task?.acceptanceCriteria ?? [])])
17
+ .filter((candidate) => candidate !== primary && !excluded.has(candidate));
18
+ for (const candidate of candidates) {
19
+ const candidateContract = malformedIdentifierContractText(candidate);
20
+ if (candidateContract && !normalizedText(candidate).includes(normalizedText(primary))) continue;
21
+ const related = malformedIdentifierContractText(`${primary}\n${candidate}`);
22
+ if (related) return related;
23
+ }
24
+ return primary;
25
+ }
@@ -0,0 +1,209 @@
1
+ import { createHash } from "node:crypto";
2
+ import { isDeepStrictEqual } from "node:util";
3
+ import { runIsolatedContract } from "./acceptance-isolated-executor.js";
4
+ import { parseRequest, validateValue, MAX_STRING_LENGTH } from "./acceptance-executor/protocol.mjs";
5
+ import { canonicalValue } from "./acceptance-executor/values.mjs";
6
+ import { CASE_CAPABILITY_FIELDS, OBSERVATION_CAPABILITY_FIELDS, callbackIds, validateCallbackTrace, validateReturnIdentity, validateErrorObservation } from "./acceptance-executor/capabilities.mjs";
7
+ import { validateReferenceIdentity } from "./acceptance-executor/reference-identity.mjs";
8
+ import { validateDeclaredApi, observeDeclaredApi } from "./acceptance-declared-api.js";
9
+
10
+ export const INDEPENDENT_CONTRACT_VERSION = "bounded-module-contract-comparison-v6";
11
+ export const NODE_INDEPENDENT_CONTRACT_VERSION = "bounded-node-profile-contract-comparison-v3";
12
+ const ID = /^[a-zA-Z0-9][a-zA-Z0-9:._-]{0,159}$/;
13
+ const hash = (value) => createHash("sha256").update(value).digest("hex");
14
+ const ERROR_CLASSES = ["TypeError", "RangeError", "SyntaxError", "ReferenceError", "EvalError", "URIError", "Error", "non-error"];
15
+
16
+ function shape(value, keys, required = keys) {
17
+ if (!value || typeof value !== "object" || Array.isArray(value) || Object.getPrototypeOf(value) !== Object.prototype
18
+ || Object.keys(value).some((key) => !keys.includes(key)) || required.some((key) => !Object.hasOwn(value, key))) {
19
+ throw new TypeError("Invalid independent contract object");
20
+ }
21
+ }
22
+
23
+ function freeze(value) {
24
+ if (value && typeof value === "object" && !Object.isFrozen(value)) {
25
+ for (const child of Object.values(value)) freeze(child);
26
+ Object.freeze(value);
27
+ }
28
+ return value;
29
+ }
30
+
31
+ function validateExpected(expected, item, nodeProfile = false) {
32
+ const { args } = item;
33
+ shape(expected, ["outcome", "value", "errorClass", "clockReads", "dateArgsAfter", "argsAfter", "publicApi", ...OBSERVATION_CAPABILITY_FIELDS], ["outcome"]);
34
+ if (Object.hasOwn(expected, "publicApi")) {
35
+ if (!nodeProfile) throw new TypeError("Declared API requires the explicit Node contract profile");
36
+ validateDeclaredApi(expected.publicApi);
37
+ }
38
+ if (nodeProfile && item.invocation?.kind === "construct" && !["constructed", "throw"].includes(expected.outcome)) {
39
+ throw new TypeError("Invalid constructor expectation");
40
+ }
41
+ if (expected.outcome === "return") {
42
+ validateValue(expected.value, false, new Set(), { typedBytes: nodeProfile });
43
+ if (Object.hasOwn(expected, "errorClass")) throw new TypeError("Conflicting expected result");
44
+ } else if (expected.outcome === "throw") {
45
+ if (!ERROR_CLASSES.includes(expected.errorClass) || Object.hasOwn(expected, "value")) throw new TypeError("Invalid expected exception");
46
+ } else if (nodeProfile && expected.outcome === "constructed") {
47
+ if (item.invocation?.kind !== "construct" || Object.hasOwn(expected, "value") || Object.hasOwn(expected, "errorClass")) {
48
+ throw new TypeError("Invalid expected construction");
49
+ }
50
+ } else throw new TypeError("Invalid expected outcome");
51
+ if (Object.hasOwn(expected, "clockReads") && (!Number.isSafeInteger(expected.clockReads) || expected.clockReads < 0)) {
52
+ throw new TypeError("Invalid expected clock reads");
53
+ }
54
+ if (Object.hasOwn(expected, "dateArgsAfter")) {
55
+ const indexes = args.flatMap((arg, index) => arg.type === "date" ? [index] : []);
56
+ if (!Array.isArray(expected.dateArgsAfter) || expected.dateArgsAfter.length !== indexes.length) throw new TypeError("Incomplete expected Date state");
57
+ expected.dateArgsAfter.forEach((arg, index) => {
58
+ shape(arg, ["index", "value"]);
59
+ if (arg.index !== indexes[index]) throw new TypeError("Wrong expected Date argument");
60
+ validateValue({ type: "number", value: arg.value }, false);
61
+ });
62
+ }
63
+ if (Object.hasOwn(expected, "argsAfter")) {
64
+ if (!item.observeArgs || !Array.isArray(expected.argsAfter) || expected.argsAfter.length !== args.length) throw new TypeError("Incomplete expected argument state");
65
+ expected.argsAfter.forEach((arg) => validateValue(arg, true, callbackIds(item), { typedBytes: nodeProfile }));
66
+ }
67
+ if (item.callbacks) validateCallbackTrace(expected.callbackTrace, item, { nodeProfile });
68
+ else if (Object.hasOwn(expected, "callbackTrace")) throw new TypeError("Unexpected expected callback trace");
69
+ if (item.referencePairs || Object.hasOwn(expected, "referenceIdentity")) validateReferenceIdentity(expected.referenceIdentity, item);
70
+ if (item.observeIdentity && expected.outcome === "return" || Object.hasOwn(expected, "returnIdentity")) {
71
+ if (expected.outcome !== "return") throw new TypeError("Return identity on an exception");
72
+ validateReturnIdentity(expected.returnIdentity, item);
73
+ }
74
+ if (item.observeError && expected.outcome === "throw" || Object.hasOwn(expected, "errorObservation")) {
75
+ if (expected.outcome !== "throw") throw new TypeError("Error observation on a return");
76
+ validateErrorObservation(expected.errorObservation, item, { nodeProfile });
77
+ }
78
+ if (item.observeErrorMessage && expected.outcome === "throw" || Object.hasOwn(expected, "errorMessage")) {
79
+ if (!item.observeErrorMessage || expected.outcome !== "throw") throw new TypeError("Unexpected expected error message");
80
+ shape(expected.errorMessage, ["includes", "ignoreCase"]);
81
+ if (typeof expected.errorMessage.includes !== "string" || !expected.errorMessage.includes.length
82
+ || expected.errorMessage.includes.length > MAX_STRING_LENGTH || typeof expected.errorMessage.ignoreCase !== "boolean") {
83
+ throw new TypeError("Invalid expected error message");
84
+ }
85
+ }
86
+ }
87
+
88
+ /** Compile a HOST-OWNED, independently specified plan, not model-authored proof. */
89
+ export function compileIndependentContract(planText) {
90
+ if (typeof planText !== "string" || Buffer.byteLength(planText) > 1024 * 1024) throw new TypeError("Contract plan size exceeded");
91
+ const plan = JSON.parse(planText);
92
+ const nodeProfile = plan?.schemaVersion === 2;
93
+ shape(plan, ["schemaVersion", "profile", "source", "exportName", "checks", "moduleGraph"],
94
+ nodeProfile ? ["schemaVersion", "profile", "source", "exportName", "checks"] : ["schemaVersion", "source", "exportName", "checks"]);
95
+ if (!nodeProfile && plan.schemaVersion !== 1 || !Array.isArray(plan.checks) || plan.checks.length < 1 || plan.checks.length > 256) {
96
+ throw new TypeError("Invalid contract checks");
97
+ }
98
+ const ids = new Set();
99
+ const cases = [];
100
+ for (const check of plan.checks) {
101
+ shape(check, ["id", "cases"]);
102
+ if (typeof check.id !== "string" || !ID.test(check.id) || ids.has(check.id) || !Array.isArray(check.cases)
103
+ || check.cases.length < 1 || check.cases.length > 256) throw new TypeError("Invalid independent check");
104
+ ids.add(check.id);
105
+ for (const item of check.cases) {
106
+ shape(item, ["id", "args", "clock", "expected", "sequence", "exportName", "reset", "observeArgs", "invocation", ...CASE_CAPABILITY_FIELDS],
107
+ nodeProfile ? ["id", "args", "expected", "invocation"] : ["id", "args", "expected"]);
108
+ const { expected, ...input } = item;
109
+ cases.push(input);
110
+ if (cases.length > 256) throw new TypeError("Too many independent cases");
111
+ }
112
+ }
113
+ const requestText = JSON.stringify({ schemaVersion: nodeProfile ? 2 : 1, ...(nodeProfile ? { profile: plan.profile } : {}),
114
+ source: plan.source, exportName: plan.exportName, cases,
115
+ ...(Object.hasOwn(plan, "moduleGraph") ? { moduleGraph: plan.moduleGraph } : {}) });
116
+ parseRequest(requestText); // Validate all guest inputs before validating dependent expected state.
117
+ // Every failing step can retain its full replay prefix. Bound the worst-case
118
+ // repeated input bytes before execution, not after a large receipt is built.
119
+ let sequence, prefixBytes = 2, historyBytes = 0;
120
+ for (const item of cases) {
121
+ if (!item.sequence || item.sequence !== sequence) prefixBytes = 2;
122
+ sequence = item.sequence;
123
+ if (!sequence) continue;
124
+ prefixBytes += Buffer.byteLength(JSON.stringify(item)) + 1;
125
+ historyBytes += prefixBytes;
126
+ if (historyBytes > 2 * 1024 * 1024) throw new TypeError("Counterexample history budget exceeded");
127
+ }
128
+ for (const check of plan.checks) for (const item of check.cases) validateExpected(item.expected, item, nodeProfile);
129
+ const version = nodeProfile ? NODE_INDEPENDENT_CONTRACT_VERSION : INDEPENDENT_CONTRACT_VERSION;
130
+ return freeze({ version, plan, requestText, planDigest: hash(JSON.stringify([version, plan])) });
131
+ }
132
+
133
+ function matches(expected, observed) {
134
+ if (expected.outcome !== observed.outcome) return false;
135
+ if (Object.hasOwn(expected, "publicApi") && !isDeepStrictEqual(expected.publicApi, observed.publicApi)) return false;
136
+ if (expected.outcome === "return" && !isDeepStrictEqual(canonicalValue(expected.value), canonicalValue(observed.value))) return false;
137
+ if (expected.outcome === "throw" && expected.errorClass !== observed.errorClass) return false;
138
+ if (Object.hasOwn(expected, "errorMessage")) {
139
+ if (typeof observed.errorMessage !== "string" || observed.errorMessage.length > MAX_STRING_LENGTH) return false;
140
+ // Literal text only, with ECMAScript /i semantics. No plan-supplied pattern
141
+ // syntax, flags, callbacks or expectations enter the guest.
142
+ const literal = expected.errorMessage.includes.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
143
+ if (!new RegExp(literal, expected.errorMessage.ignoreCase ? "i" : "").test(observed.errorMessage)) return false;
144
+ }
145
+ if (Object.hasOwn(expected, "clockReads") && expected.clockReads !== observed.clockReads) return false;
146
+ if (Object.hasOwn(expected, "dateArgsAfter") && !isDeepStrictEqual(expected.dateArgsAfter, observed.dateArgsAfter)) return false;
147
+ if (Object.hasOwn(expected, "argsAfter") && !isDeepStrictEqual(expected.argsAfter.map(canonicalValue), observed.argsAfter?.map(canonicalValue))) return false;
148
+ const canonicalTrace = trace => trace?.map(call => call.event === "call" ? { ...call, args: call.args.map(canonicalValue) } : call);
149
+ if (Object.hasOwn(expected, "callbackTrace") && !isDeepStrictEqual(canonicalTrace(expected.callbackTrace), canonicalTrace(observed.callbackTrace))) return false;
150
+ if (Object.hasOwn(expected, "returnIdentity") && !isDeepStrictEqual(expected.returnIdentity, observed.returnIdentity)) return false;
151
+ if (Object.hasOwn(expected, "referenceIdentity") && !isDeepStrictEqual(expected.referenceIdentity, observed.referenceIdentity)) return false;
152
+ if (Object.hasOwn(expected, "errorObservation") && (expected.errorObservation.identity !== observed.errorObservation?.identity
153
+ || !observed.errorObservation || !isDeepStrictEqual(canonicalValue(expected.errorObservation.properties), canonicalValue(observed.errorObservation.properties)))) return false;
154
+ return true;
155
+ }
156
+
157
+ /**
158
+ * Actual observations come only from this invocation of the isolated adapter.
159
+ * Expected outputs remain on the host; guest output cannot select a verdict.
160
+ * This returns diagnostic evidence, NOT an authenticated durable receipt or
161
+ * completion authority. Runtime contract approval and snapshot binding remain
162
+ * required at integration. Unsupported execution is never a passed case.
163
+ */
164
+ export async function runIndependentContract({ planText, ...backend } = {}) {
165
+ const compiled = compileIndependentContract(planText);
166
+ const execution = await runIsolatedContract({ ...backend, ...(compiled.plan.schemaVersion === 2 ? { profile: compiled.plan.profile } : {}), requestText: compiled.requestText });
167
+ return compareIndependentExecution(compiled, execution);
168
+ }
169
+
170
+ /** Host comparison only: this pure projection does not certify an execution. */
171
+ export function compareIndependentExecution(compiled, execution) {
172
+ const counterexamples = [];
173
+ const observations = new Map((execution.observation?.cases ?? []).map((item) => [item.id, item]));
174
+ const api = compiled.plan.checks.some(check => check.cases.some(item => Object.hasOwn(item.expected, "publicApi")))
175
+ ? observeDeclaredApi(compiled.plan.source) : null;
176
+ let sequence, history = [];
177
+ const checks = compiled.plan.checks.map((check) => {
178
+ let caseCount = 0, incomplete = false, error = false, counterexampleRef;
179
+ for (const item of check.cases) {
180
+ const { expected: _expected, ...input } = item;
181
+ if (!item.sequence || item.sequence !== sequence) history = [];
182
+ sequence = item.sequence; history.push(input);
183
+ let observed = observations.get(item.id);
184
+ if (!observed || observed.outcome === "unsupported") { incomplete = true; continue; }
185
+ if (observed.outcome === "error") { error = true; continue; }
186
+ if (Object.hasOwn(item.expected, "publicApi")) {
187
+ if (api?.status !== "observed") { incomplete = true; continue; }
188
+ observed = { ...observed, publicApi: api.value };
189
+ }
190
+ caseCount += 1;
191
+ if (!matches(item.expected, observed)) {
192
+ const evidence = { version: compiled.version ?? INDEPENDENT_CONTRACT_VERSION, planDigest: compiled.planDigest,
193
+ runId: execution.runId, sourceDigest: execution.sourceDigest, imageId: execution.imageId,
194
+ checkId: check.id, input: { ...input, ...(sequence ? { prefix: history.slice() } : {}) },
195
+ expected: item.expected, observed };
196
+ const digest = hash(JSON.stringify(evidence));
197
+ counterexamples.push({ digest, evidence });
198
+ counterexampleRef ??= digest;
199
+ }
200
+ }
201
+ const status = error ? "error" : counterexampleRef ? "fail" : incomplete ? "unknown" : "pass";
202
+ return { id: check.id, status, caseCount, ...(status === "fail" ? { counterexampleRef } : {}) };
203
+ });
204
+ const statuses = checks.map((check) => check.status);
205
+ const verdict = execution.status !== "completed" || !execution.cleanupConfirmed || statuses.includes("error") ? "error"
206
+ : statuses.includes("fail") ? "fail" : statuses.includes("unknown") ? "unknown" : "pass";
207
+ return freeze({ version: compiled.version ?? INDEPENDENT_CONTRACT_VERSION, planDigest: compiled.planDigest,
208
+ verdict, execution, checks, counterexamples });
209
+ }