audrey 1.1.3 → 1.2.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (285) hide show
  1. package/CHANGELOG.md +93 -0
  2. package/README.md +39 -12
  3. package/dist/mcp-server/codex-hooks-probe.d.ts +22 -0
  4. package/dist/mcp-server/codex-hooks-probe.d.ts.map +1 -0
  5. package/dist/mcp-server/codex-hooks-probe.js +160 -0
  6. package/dist/mcp-server/codex-hooks-probe.js.map +1 -0
  7. package/dist/mcp-server/config.d.ts +26 -2
  8. package/dist/mcp-server/config.d.ts.map +1 -1
  9. package/dist/mcp-server/config.js +52 -20
  10. package/dist/mcp-server/config.js.map +1 -1
  11. package/dist/mcp-server/hooks.d.ts +28 -0
  12. package/dist/mcp-server/hooks.d.ts.map +1 -1
  13. package/dist/mcp-server/hooks.js +91 -2
  14. package/dist/mcp-server/hooks.js.map +1 -1
  15. package/dist/mcp-server/index.d.ts +242 -2
  16. package/dist/mcp-server/index.d.ts.map +1 -1
  17. package/dist/mcp-server/index.js +806 -240
  18. package/dist/mcp-server/index.js.map +1 -1
  19. package/dist/mcp-server/tool-schemas.d.ts +32 -2
  20. package/dist/mcp-server/tool-schemas.d.ts.map +1 -1
  21. package/dist/mcp-server/tool-schemas.js +55 -0
  22. package/dist/mcp-server/tool-schemas.js.map +1 -1
  23. package/dist/src/audrey.d.ts +14 -2
  24. package/dist/src/audrey.d.ts.map +1 -1
  25. package/dist/src/audrey.js +152 -28
  26. package/dist/src/audrey.js.map +1 -1
  27. package/dist/src/autopilot.d.ts +22 -0
  28. package/dist/src/autopilot.d.ts.map +1 -1
  29. package/dist/src/autopilot.js +271 -50
  30. package/dist/src/autopilot.js.map +1 -1
  31. package/dist/src/capsule.d.ts +23 -0
  32. package/dist/src/capsule.d.ts.map +1 -1
  33. package/dist/src/capsule.js +56 -11
  34. package/dist/src/capsule.js.map +1 -1
  35. package/dist/src/confidence.d.ts +9 -1
  36. package/dist/src/confidence.d.ts.map +1 -1
  37. package/dist/src/confidence.js +16 -2
  38. package/dist/src/confidence.js.map +1 -1
  39. package/dist/src/consolidate.d.ts +4 -0
  40. package/dist/src/consolidate.d.ts.map +1 -1
  41. package/dist/src/consolidate.js +206 -34
  42. package/dist/src/consolidate.js.map +1 -1
  43. package/dist/src/controller.d.ts +17 -1
  44. package/dist/src/controller.d.ts.map +1 -1
  45. package/dist/src/controller.js +244 -128
  46. package/dist/src/controller.js.map +1 -1
  47. package/dist/src/db.d.ts.map +1 -1
  48. package/dist/src/db.js +233 -56
  49. package/dist/src/db.js.map +1 -1
  50. package/dist/src/encode.d.ts +2 -1
  51. package/dist/src/encode.d.ts.map +1 -1
  52. package/dist/src/encode.js +41 -4
  53. package/dist/src/encode.js.map +1 -1
  54. package/dist/src/events.d.ts +48 -1
  55. package/dist/src/events.d.ts.map +1 -1
  56. package/dist/src/events.js +160 -62
  57. package/dist/src/events.js.map +1 -1
  58. package/dist/src/export.d.ts.map +1 -1
  59. package/dist/src/export.js +4 -4
  60. package/dist/src/export.js.map +1 -1
  61. package/dist/src/grounding.d.ts +111 -0
  62. package/dist/src/grounding.d.ts.map +1 -0
  63. package/dist/src/grounding.js +259 -0
  64. package/dist/src/grounding.js.map +1 -0
  65. package/dist/src/hybrid-recall.d.ts +16 -7
  66. package/dist/src/hybrid-recall.d.ts.map +1 -1
  67. package/dist/src/hybrid-recall.js +22 -11
  68. package/dist/src/hybrid-recall.js.map +1 -1
  69. package/dist/src/import.d.ts +10 -0
  70. package/dist/src/import.d.ts.map +1 -1
  71. package/dist/src/import.js +122 -22
  72. package/dist/src/import.js.map +1 -1
  73. package/dist/src/llm.d.ts.map +1 -1
  74. package/dist/src/llm.js +16 -0
  75. package/dist/src/llm.js.map +1 -1
  76. package/dist/src/preflight.d.ts.map +1 -1
  77. package/dist/src/preflight.js +89 -47
  78. package/dist/src/preflight.js.map +1 -1
  79. package/dist/src/project.d.ts +4 -1
  80. package/dist/src/project.d.ts.map +1 -1
  81. package/dist/src/project.js +65 -5
  82. package/dist/src/project.js.map +1 -1
  83. package/dist/src/recall.d.ts.map +1 -1
  84. package/dist/src/recall.js +138 -13
  85. package/dist/src/recall.js.map +1 -1
  86. package/dist/src/redact.d.ts +1 -1
  87. package/dist/src/redact.d.ts.map +1 -1
  88. package/dist/src/redact.js +185 -5
  89. package/dist/src/redact.js.map +1 -1
  90. package/dist/src/routes.d.ts.map +1 -1
  91. package/dist/src/routes.js +91 -18
  92. package/dist/src/routes.js.map +1 -1
  93. package/dist/src/trust.d.ts +35 -0
  94. package/dist/src/trust.d.ts.map +1 -0
  95. package/dist/src/trust.js +77 -0
  96. package/dist/src/trust.js.map +1 -0
  97. package/dist/src/types.d.ts +24 -0
  98. package/dist/src/types.d.ts.map +1 -1
  99. package/package.json +3 -37
  100. package/benchmarks/adapter-kit.mjs +0 -20
  101. package/benchmarks/adapter-self-test.mjs +0 -170
  102. package/benchmarks/adapters/example-allow.mjs +0 -31
  103. package/benchmarks/adapters/mem0-platform.mjs +0 -274
  104. package/benchmarks/adapters/registry.json +0 -51
  105. package/benchmarks/adapters/zep-cloud.mjs +0 -304
  106. package/benchmarks/baselines.js +0 -174
  107. package/benchmarks/build-leaderboard.mjs +0 -183
  108. package/benchmarks/cases.js +0 -549
  109. package/benchmarks/create-conformance-card.mjs +0 -148
  110. package/benchmarks/create-submission-bundle.mjs +0 -190
  111. package/benchmarks/dry-run-external-adapters.mjs +0 -177
  112. package/benchmarks/guardbench.js +0 -1265
  113. package/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +0 -50
  114. package/benchmarks/output/external/guardbench-external-dry-run.json +0 -69
  115. package/benchmarks/output/external/guardbench-external-evidence.json +0 -56
  116. package/benchmarks/output/guardbench-conformance-card.json +0 -63
  117. package/benchmarks/output/guardbench-manifest.json +0 -414
  118. package/benchmarks/output/guardbench-raw.json +0 -1279
  119. package/benchmarks/output/guardbench-summary.json +0 -2123
  120. package/benchmarks/output/leaderboard/guardbench-leaderboard.json +0 -93
  121. package/benchmarks/output/leaderboard/guardbench-leaderboard.md +0 -7
  122. package/benchmarks/output/submission-bundle/guardbench-conformance-card.json +0 -63
  123. package/benchmarks/output/submission-bundle/guardbench-manifest.json +0 -414
  124. package/benchmarks/output/submission-bundle/guardbench-raw.json +0 -1279
  125. package/benchmarks/output/submission-bundle/guardbench-summary.json +0 -2123
  126. package/benchmarks/output/submission-bundle/schemas/guardbench-adapter-registry.schema.json +0 -69
  127. package/benchmarks/output/submission-bundle/schemas/guardbench-adapter-self-test.schema.json +0 -156
  128. package/benchmarks/output/submission-bundle/schemas/guardbench-conformance-card.schema.json +0 -184
  129. package/benchmarks/output/submission-bundle/schemas/guardbench-external-dry-run.schema.json +0 -74
  130. package/benchmarks/output/submission-bundle/schemas/guardbench-external-evidence.schema.json +0 -108
  131. package/benchmarks/output/submission-bundle/schemas/guardbench-external-run.schema.json +0 -160
  132. package/benchmarks/output/submission-bundle/schemas/guardbench-leaderboard.schema.json +0 -179
  133. package/benchmarks/output/submission-bundle/schemas/guardbench-manifest.schema.json +0 -213
  134. package/benchmarks/output/submission-bundle/schemas/guardbench-publication-verification.schema.json +0 -47
  135. package/benchmarks/output/submission-bundle/schemas/guardbench-raw.schema.json +0 -184
  136. package/benchmarks/output/submission-bundle/schemas/guardbench-submission-manifest.schema.json +0 -151
  137. package/benchmarks/output/submission-bundle/schemas/guardbench-summary.schema.json +0 -249
  138. package/benchmarks/output/submission-bundle/submission-manifest.json +0 -131
  139. package/benchmarks/output/submission-bundle/validation-report.json +0 -31
  140. package/benchmarks/output/summary.json +0 -2354
  141. package/benchmarks/perf-snapshot.js +0 -307
  142. package/benchmarks/perf.bench.js +0 -169
  143. package/benchmarks/public-paths.mjs +0 -97
  144. package/benchmarks/reference-results.js +0 -75
  145. package/benchmarks/report.js +0 -280
  146. package/benchmarks/run-external-guardbench.mjs +0 -303
  147. package/benchmarks/run.js +0 -735
  148. package/benchmarks/schemas/guardbench-adapter-registry.schema.json +0 -69
  149. package/benchmarks/schemas/guardbench-adapter-self-test.schema.json +0 -156
  150. package/benchmarks/schemas/guardbench-conformance-card.schema.json +0 -184
  151. package/benchmarks/schemas/guardbench-external-dry-run.schema.json +0 -74
  152. package/benchmarks/schemas/guardbench-external-evidence.schema.json +0 -108
  153. package/benchmarks/schemas/guardbench-external-run.schema.json +0 -160
  154. package/benchmarks/schemas/guardbench-leaderboard.schema.json +0 -179
  155. package/benchmarks/schemas/guardbench-manifest.schema.json +0 -213
  156. package/benchmarks/schemas/guardbench-publication-verification.schema.json +0 -47
  157. package/benchmarks/schemas/guardbench-raw.schema.json +0 -184
  158. package/benchmarks/schemas/guardbench-submission-manifest.schema.json +0 -151
  159. package/benchmarks/schemas/guardbench-summary.schema.json +0 -249
  160. package/benchmarks/snapshots/perf-0.22.2.json +0 -123
  161. package/benchmarks/snapshots/perf-0.23.0.json +0 -123
  162. package/benchmarks/validate-adapter-module.mjs +0 -107
  163. package/benchmarks/validate-adapter-registry.mjs +0 -145
  164. package/benchmarks/validate-adapter-self-test.mjs +0 -96
  165. package/benchmarks/validate-guardbench-artifacts.mjs +0 -400
  166. package/benchmarks/verify-external-evidence.mjs +0 -351
  167. package/benchmarks/verify-publication-artifacts.mjs +0 -309
  168. package/benchmarks/verify-submission-bundle.mjs +0 -172
  169. package/docs/AUDREY_PAPER_OUTLINE.md +0 -178
  170. package/docs/MEMORY_BENCHMARKING.md +0 -59
  171. package/docs/PRODUCTION_BACKLOG.md +0 -321
  172. package/docs/assets/audrey-feature-grid.jpg +0 -0
  173. package/docs/assets/audrey-logo.svg +0 -45
  174. package/docs/assets/audrey-wordmark.png +0 -0
  175. package/docs/paper/00-master.md +0 -48
  176. package/docs/paper/01-introduction.md +0 -27
  177. package/docs/paper/02-related-work.md +0 -47
  178. package/docs/paper/03-problem-definition.md +0 -108
  179. package/docs/paper/04-design.md +0 -164
  180. package/docs/paper/05-guardbench-spec.md +0 -412
  181. package/docs/paper/06-implementation.md +0 -115
  182. package/docs/paper/07-evaluation.md +0 -168
  183. package/docs/paper/08-discussion-limitations.md +0 -61
  184. package/docs/paper/09-conclusion.md +0 -11
  185. package/docs/paper/SUBMISSION_README.md +0 -162
  186. package/docs/paper/appendix-a-demo-transcript.md +0 -114
  187. package/docs/paper/arxiv-compile-report.schema.json +0 -116
  188. package/docs/paper/arxiv-source.schema.json +0 -61
  189. package/docs/paper/audrey-paper-v1.md +0 -1108
  190. package/docs/paper/browser-launch-plan.json +0 -209
  191. package/docs/paper/browser-launch-plan.schema.json +0 -100
  192. package/docs/paper/browser-launch-results.json +0 -86
  193. package/docs/paper/browser-launch-results.schema.json +0 -66
  194. package/docs/paper/claim-register.json +0 -138
  195. package/docs/paper/claim-register.schema.json +0 -81
  196. package/docs/paper/evidence-ledger.md +0 -103
  197. package/docs/paper/output/arxiv/README-arxiv.txt +0 -8
  198. package/docs/paper/output/arxiv/arxiv-manifest.json +0 -41
  199. package/docs/paper/output/arxiv/main.tex +0 -951
  200. package/docs/paper/output/arxiv/references.bib +0 -222
  201. package/docs/paper/output/arxiv-compile-report.json +0 -24
  202. package/docs/paper/output/submission-bundle/LICENSE +0 -21
  203. package/docs/paper/output/submission-bundle/README.md +0 -440
  204. package/docs/paper/output/submission-bundle/benchmarks/output/adapter-self-test/guardbench-adapter-self-test.json +0 -50
  205. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-dry-run.json +0 -69
  206. package/docs/paper/output/submission-bundle/benchmarks/output/external/guardbench-external-evidence.json +0 -56
  207. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-conformance-card.json +0 -63
  208. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-manifest.json +0 -414
  209. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-raw.json +0 -1279
  210. package/docs/paper/output/submission-bundle/benchmarks/output/guardbench-summary.json +0 -2123
  211. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.json +0 -93
  212. package/docs/paper/output/submission-bundle/benchmarks/output/leaderboard/guardbench-leaderboard.md +0 -7
  213. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/submission-manifest.json +0 -131
  214. package/docs/paper/output/submission-bundle/benchmarks/output/submission-bundle/validation-report.json +0 -31
  215. package/docs/paper/output/submission-bundle/benchmarks/output/summary.json +0 -2354
  216. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-adapter-registry.schema.json +0 -69
  217. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-adapter-self-test.schema.json +0 -156
  218. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-conformance-card.schema.json +0 -184
  219. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-dry-run.schema.json +0 -74
  220. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-evidence.schema.json +0 -108
  221. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-external-run.schema.json +0 -160
  222. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-leaderboard.schema.json +0 -179
  223. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-manifest.schema.json +0 -213
  224. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-publication-verification.schema.json +0 -47
  225. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-raw.schema.json +0 -184
  226. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-submission-manifest.schema.json +0 -151
  227. package/docs/paper/output/submission-bundle/benchmarks/schemas/guardbench-summary.schema.json +0 -249
  228. package/docs/paper/output/submission-bundle/docs/AUDREY_PAPER_OUTLINE.md +0 -178
  229. package/docs/paper/output/submission-bundle/docs/paper/00-master.md +0 -48
  230. package/docs/paper/output/submission-bundle/docs/paper/01-introduction.md +0 -27
  231. package/docs/paper/output/submission-bundle/docs/paper/02-related-work.md +0 -47
  232. package/docs/paper/output/submission-bundle/docs/paper/03-problem-definition.md +0 -108
  233. package/docs/paper/output/submission-bundle/docs/paper/04-design.md +0 -164
  234. package/docs/paper/output/submission-bundle/docs/paper/05-guardbench-spec.md +0 -412
  235. package/docs/paper/output/submission-bundle/docs/paper/06-implementation.md +0 -115
  236. package/docs/paper/output/submission-bundle/docs/paper/07-evaluation.md +0 -168
  237. package/docs/paper/output/submission-bundle/docs/paper/08-discussion-limitations.md +0 -61
  238. package/docs/paper/output/submission-bundle/docs/paper/09-conclusion.md +0 -11
  239. package/docs/paper/output/submission-bundle/docs/paper/SUBMISSION_README.md +0 -162
  240. package/docs/paper/output/submission-bundle/docs/paper/appendix-a-demo-transcript.md +0 -114
  241. package/docs/paper/output/submission-bundle/docs/paper/arxiv-compile-report.schema.json +0 -116
  242. package/docs/paper/output/submission-bundle/docs/paper/arxiv-source.schema.json +0 -61
  243. package/docs/paper/output/submission-bundle/docs/paper/audrey-paper-v1.md +0 -1108
  244. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-plan.json +0 -209
  245. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-plan.schema.json +0 -100
  246. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-results.json +0 -86
  247. package/docs/paper/output/submission-bundle/docs/paper/browser-launch-results.schema.json +0 -66
  248. package/docs/paper/output/submission-bundle/docs/paper/claim-register.json +0 -138
  249. package/docs/paper/output/submission-bundle/docs/paper/claim-register.schema.json +0 -81
  250. package/docs/paper/output/submission-bundle/docs/paper/evidence-ledger.md +0 -103
  251. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/README-arxiv.txt +0 -8
  252. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/arxiv-manifest.json +0 -41
  253. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/main.tex +0 -951
  254. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv/references.bib +0 -222
  255. package/docs/paper/output/submission-bundle/docs/paper/output/arxiv-compile-report.json +0 -24
  256. package/docs/paper/output/submission-bundle/docs/paper/paper-submission-bundle.schema.json +0 -70
  257. package/docs/paper/output/submission-bundle/docs/paper/publication-pack.json +0 -81
  258. package/docs/paper/output/submission-bundle/docs/paper/publication-pack.schema.json +0 -60
  259. package/docs/paper/output/submission-bundle/docs/paper/references.bib +0 -222
  260. package/docs/paper/output/submission-bundle/package.json +0 -235
  261. package/docs/paper/output/submission-bundle/paper-submission-manifest.json +0 -379
  262. package/docs/paper/paper-submission-bundle.schema.json +0 -70
  263. package/docs/paper/publication-pack.json +0 -81
  264. package/docs/paper/publication-pack.schema.json +0 -60
  265. package/docs/paper/references.bib +0 -222
  266. package/scripts/audit-release-completion.mjs +0 -440
  267. package/scripts/create-arxiv-source.mjs +0 -368
  268. package/scripts/create-paper-submission-bundle.mjs +0 -214
  269. package/scripts/finalize-release.mjs +0 -601
  270. package/scripts/prepare-release-cut.mjs +0 -277
  271. package/scripts/publish-release-bundle.mjs +0 -252
  272. package/scripts/publish-release-github-api.mjs +0 -494
  273. package/scripts/run-vitest.mjs +0 -34
  274. package/scripts/smoke-cli.js +0 -92
  275. package/scripts/sync-paper-artifacts.mjs +0 -112
  276. package/scripts/verify-arxiv-compile.mjs +0 -476
  277. package/scripts/verify-arxiv-source.mjs +0 -224
  278. package/scripts/verify-browser-launch-plan.mjs +0 -254
  279. package/scripts/verify-browser-launch-results.mjs +0 -303
  280. package/scripts/verify-paper-artifacts.mjs +0 -798
  281. package/scripts/verify-paper-claims.mjs +0 -254
  282. package/scripts/verify-paper-submission-bundle.mjs +0 -218
  283. package/scripts/verify-publication-pack.mjs +0 -210
  284. package/scripts/verify-python-package.py +0 -201
  285. package/scripts/verify-release-readiness.mjs +0 -920
@@ -1,400 +0,0 @@
1
- import { existsSync, readFileSync } from 'node:fs';
2
- import { createHash } from 'node:crypto';
3
- import { basename, join, resolve } from 'node:path';
4
- import { fileURLToPath } from 'node:url';
5
- import { publicPath } from './public-paths.mjs';
6
-
7
- const DEFAULT_DIR = 'benchmarks/output';
8
- const DEFAULT_SEEDED_SECRET = 'sk-guardbench-secret-0000000000000000000000000000';
9
- const SCHEMA_FILES = {
10
- manifest: 'guardbench-manifest.schema.json',
11
- summary: 'guardbench-summary.schema.json',
12
- raw: 'guardbench-raw.schema.json',
13
- externalRun: 'guardbench-external-run.schema.json',
14
- conformanceCard: 'guardbench-conformance-card.schema.json',
15
- };
16
- const ARTIFACT_FILES = {
17
- manifest: 'guardbench-manifest.json',
18
- summary: 'guardbench-summary.json',
19
- raw: 'guardbench-raw.json',
20
- };
21
- const OPTIONAL_ARTIFACT_FILES = {
22
- externalRun: 'external-run-metadata.json',
23
- conformanceCard: 'guardbench-conformance-card.json',
24
- };
25
-
26
- function parseArgs(argv = process.argv.slice(2)) {
27
- const args = {
28
- dir: DEFAULT_DIR,
29
- schemasDir: 'benchmarks/schemas',
30
- seededSecrets: [DEFAULT_SEEDED_SECRET],
31
- json: false,
32
- };
33
-
34
- for (let i = 0; i < argv.length; i++) {
35
- const token = argv[i];
36
- if ((token === '--dir' || token === '--out-dir') && argv[i + 1]) args.dir = argv[++i];
37
- else if (token === '--schemas-dir' && argv[i + 1]) args.schemasDir = argv[++i];
38
- else if (token === '--seeded-secret' && argv[i + 1]) args.seededSecrets.push(argv[++i]);
39
- else if (token === '--no-default-secret') args.seededSecrets = [];
40
- else if (token === '--json') args.json = true;
41
- else if (token === '--help') {
42
- return { ...args, help: true };
43
- }
44
- }
45
-
46
- return args;
47
- }
48
-
49
- function usage() {
50
- return [
51
- 'Usage: node benchmarks/validate-guardbench-artifacts.mjs [--dir benchmarks/output] [--json]',
52
- '',
53
- 'Validates guardbench-manifest.json, guardbench-summary.json, and',
54
- 'guardbench-raw.json against the published GuardBench JSON schemas.',
55
- '',
56
- 'Options:',
57
- ' --dir <path> Directory containing GuardBench output artifacts.',
58
- ' --schemas-dir <path> Directory containing GuardBench schema files.',
59
- ' --seeded-secret <value> Additional seeded raw secret that must not appear.',
60
- ' --no-default-secret Do not check the built-in GuardBench redaction probe.',
61
- ' --json Print a machine-readable validation report.',
62
- ].join('\n');
63
- }
64
-
65
- function readText(path) {
66
- if (!existsSync(path)) throw new Error(`Missing required file: ${path}`);
67
- return readFileSync(path, 'utf-8');
68
- }
69
-
70
- function readJson(path) {
71
- return JSON.parse(readText(path));
72
- }
73
-
74
- function sha256File(path) {
75
- return createHash('sha256').update(readFileSync(path)).digest('hex');
76
- }
77
-
78
- export function computeGuardBenchArtifactHashes(dir, files = Object.values(ARTIFACT_FILES)) {
79
- const resolvedDir = resolve(dir);
80
- return Object.fromEntries(files.map(file => [file, sha256File(join(resolvedDir, file))]));
81
- }
82
-
83
- function typeOf(value) {
84
- if (Array.isArray(value)) return 'array';
85
- if (value === null) return 'null';
86
- return typeof value;
87
- }
88
-
89
- export function validateSchema(value, schema, label, root = schema) {
90
- const errors = [];
91
-
92
- function validate(current, currentSchema, path) {
93
- if (currentSchema.$ref) {
94
- const refPath = currentSchema.$ref.replace(/^#\//, '').split('/');
95
- const resolved = refPath.reduce((node, key) => node?.[key], root);
96
- if (!resolved) {
97
- errors.push(`${path}: unresolved schema ref ${currentSchema.$ref}`);
98
- return;
99
- }
100
- validate(current, resolved, path);
101
- return;
102
- }
103
-
104
- if (currentSchema.anyOf) {
105
- const nested = currentSchema.anyOf.map(option => {
106
- const before = errors.length;
107
- validate(current, option, path);
108
- return errors.splice(before);
109
- });
110
- if (!nested.some(group => group.length === 0)) {
111
- errors.push(`${path}: did not match any allowed schema`);
112
- }
113
- return;
114
- }
115
-
116
- if (currentSchema.const !== undefined && current !== currentSchema.const) {
117
- errors.push(`${path}: expected constant ${currentSchema.const}`);
118
- }
119
- if (currentSchema.enum && !currentSchema.enum.includes(current)) {
120
- errors.push(`${path}: expected one of ${currentSchema.enum.join(', ')}`);
121
- }
122
- if (currentSchema.type === 'integer') {
123
- if (typeof current !== 'number' || !Number.isInteger(current)) {
124
- errors.push(`${path}: expected integer, got ${typeOf(current)}`);
125
- return;
126
- }
127
- } else if (currentSchema.type) {
128
- const actual = typeOf(current);
129
- if (actual !== currentSchema.type) {
130
- errors.push(`${path}: expected ${currentSchema.type}, got ${actual}`);
131
- return;
132
- }
133
- }
134
- if (currentSchema.minLength != null && String(current).length < currentSchema.minLength) {
135
- errors.push(`${path}: shorter than minLength ${currentSchema.minLength}`);
136
- }
137
- if (
138
- currentSchema.pattern &&
139
- typeof current === 'string' &&
140
- !new RegExp(currentSchema.pattern).test(current)
141
- ) {
142
- errors.push(`${path}: does not match ${currentSchema.pattern}`);
143
- }
144
- if (
145
- currentSchema.minimum != null &&
146
- typeof current === 'number' &&
147
- current < currentSchema.minimum
148
- ) {
149
- errors.push(`${path}: below minimum ${currentSchema.minimum}`);
150
- }
151
- if (
152
- currentSchema.maximum != null &&
153
- typeof current === 'number' &&
154
- current > currentSchema.maximum
155
- ) {
156
- errors.push(`${path}: above maximum ${currentSchema.maximum}`);
157
- }
158
-
159
- if (currentSchema.type === 'array') {
160
- if (currentSchema.minItems != null && current.length < currentSchema.minItems) {
161
- errors.push(`${path}: expected at least ${currentSchema.minItems} items`);
162
- }
163
- if (currentSchema.items) {
164
- current.forEach((item, index) => validate(item, currentSchema.items, `${path}[${index}]`));
165
- }
166
- }
167
-
168
- if (currentSchema.type === 'object') {
169
- for (const required of currentSchema.required ?? []) {
170
- if (!Object.hasOwn(current, required))
171
- errors.push(`${path}: missing required property ${required}`);
172
- }
173
- if (currentSchema.additionalProperties === false) {
174
- for (const key of Object.keys(current)) {
175
- if (!Object.hasOwn(currentSchema.properties ?? {}, key)) {
176
- errors.push(`${path}: unexpected property ${key}`);
177
- }
178
- }
179
- }
180
- for (const [key, propertySchema] of Object.entries(currentSchema.properties ?? {})) {
181
- if (Object.hasOwn(current, key)) validate(current[key], propertySchema, `${path}.${key}`);
182
- }
183
- }
184
- }
185
-
186
- validate(value, schema, label);
187
- return errors;
188
- }
189
-
190
- function stableJson(value) {
191
- if (Array.isArray(value)) return `[${value.map(stableJson).join(',')}]`;
192
- if (value && typeof value === 'object') {
193
- return `{${Object.keys(value)
194
- .sort()
195
- .map(key => `${JSON.stringify(key)}:${stableJson(value[key])}`)
196
- .join(',')}}`;
197
- }
198
- return JSON.stringify(value);
199
- }
200
-
201
- function assertSameJson(actual, expected, label, failures) {
202
- if (stableJson(actual) !== stableJson(expected)) {
203
- failures.push(`${label}: cross-artifact mismatch`);
204
- }
205
- }
206
-
207
- export function validateGuardBenchArtifacts(options = {}) {
208
- const dir = resolve(options.dir ?? DEFAULT_DIR);
209
- const schemasDir = resolve(options.schemasDir ?? 'benchmarks/schemas');
210
- const seededSecrets = options.seededSecrets ?? [DEFAULT_SEEDED_SECRET];
211
- const failures = [];
212
- const artifacts = {};
213
- const schemas = {};
214
- const artifactPaths = {};
215
- const optionalArtifacts = {};
216
-
217
- for (const [key, file] of Object.entries(ARTIFACT_FILES)) {
218
- artifactPaths[key] = join(dir, file);
219
- try {
220
- artifacts[key] = readJson(artifactPaths[key]);
221
- } catch (error) {
222
- failures.push(error.message);
223
- }
224
- }
225
-
226
- for (const [key, file] of Object.entries(SCHEMA_FILES)) {
227
- try {
228
- schemas[key] = readJson(join(schemasDir, file));
229
- } catch (error) {
230
- failures.push(error.message);
231
- }
232
- }
233
-
234
- if (failures.length === 0) {
235
- for (const key of Object.keys(ARTIFACT_FILES)) {
236
- for (const error of validateSchema(artifacts[key], schemas[key], `guardbench-${key}`)) {
237
- failures.push(`${basename(artifactPaths[key])}: ${error}`);
238
- }
239
- }
240
- for (const [key, file] of Object.entries(OPTIONAL_ARTIFACT_FILES)) {
241
- const path = join(dir, file);
242
- if (!existsSync(path)) continue;
243
- artifactPaths[key] = path;
244
- try {
245
- optionalArtifacts[key] = readJson(path);
246
- } catch (error) {
247
- failures.push(error.message);
248
- continue;
249
- }
250
- for (const error of validateSchema(
251
- optionalArtifacts[key],
252
- schemas[key],
253
- `guardbench-${key}`,
254
- )) {
255
- failures.push(`${basename(path)}: ${error}`);
256
- }
257
- }
258
-
259
- const externalRun = optionalArtifacts.externalRun;
260
- if (externalRun?.artifactHashes) {
261
- const currentHashes = computeGuardBenchArtifactHashes(dir);
262
- for (const [file, expectedHash] of Object.entries(externalRun.artifactHashes)) {
263
- if (!Object.hasOwn(currentHashes, file)) {
264
- failures.push(`external-run-metadata.json: artifactHashes includes unknown file ${file}`);
265
- } else if (currentHashes[file] !== expectedHash) {
266
- failures.push(
267
- `external-run-metadata.json: artifactHashes.${file} does not match current artifact`,
268
- );
269
- }
270
- }
271
- for (const file of Object.values(ARTIFACT_FILES)) {
272
- if (!Object.hasOwn(externalRun.artifactHashes, file)) {
273
- failures.push(`external-run-metadata.json: artifactHashes missing ${file}`);
274
- }
275
- }
276
- }
277
- const conformanceCard = optionalArtifacts.conformanceCard;
278
- if (conformanceCard) {
279
- const currentHashes = computeGuardBenchArtifactHashes(dir);
280
- for (const [file, expectedHash] of Object.entries(
281
- conformanceCard.integrity?.artifactHashes ?? {},
282
- )) {
283
- if (!Object.hasOwn(currentHashes, file)) {
284
- failures.push(
285
- `guardbench-conformance-card.json: integrity.artifactHashes includes unknown file ${file}`,
286
- );
287
- } else if (currentHashes[file] !== expectedHash) {
288
- failures.push(
289
- `guardbench-conformance-card.json: integrity.artifactHashes.${file} does not match current artifact`,
290
- );
291
- }
292
- }
293
- if (conformanceCard.manifestVersion !== artifacts.manifest.manifestVersion) {
294
- failures.push(
295
- 'guardbench-conformance-card.json: manifestVersion does not match guardbench-manifest.json',
296
- );
297
- }
298
- if (conformanceCard.suiteId !== artifacts.manifest.suiteId) {
299
- failures.push(
300
- 'guardbench-conformance-card.json: suiteId does not match guardbench-manifest.json',
301
- );
302
- }
303
- if (
304
- !artifacts.summary.systemSummaries?.some(
305
- row => row.system === conformanceCard.subject?.name,
306
- )
307
- ) {
308
- failures.push(
309
- 'guardbench-conformance-card.json: subject.name is not present in guardbench-summary.json',
310
- );
311
- }
312
- }
313
-
314
- assertSameJson(
315
- artifacts.summary.manifest,
316
- artifacts.manifest,
317
- 'summary.manifest vs guardbench-manifest.json',
318
- failures,
319
- );
320
- assertSameJson(
321
- artifacts.summary.cases,
322
- artifacts.raw.cases,
323
- 'summary.cases vs raw.cases',
324
- failures,
325
- );
326
- assertSameJson(
327
- artifacts.summary.provenance,
328
- artifacts.raw.provenance,
329
- 'summary.provenance vs raw.provenance',
330
- failures,
331
- );
332
- if (artifacts.summary.generatedAt !== artifacts.raw.generatedAt) {
333
- failures.push('summary.generatedAt vs raw.generatedAt: cross-artifact mismatch');
334
- }
335
- if (artifacts.manifest.manifestVersion !== artifacts.raw.manifestVersion) {
336
- failures.push('manifest.manifestVersion vs raw.manifestVersion: cross-artifact mismatch');
337
- }
338
-
339
- if (artifacts.summary.artifactRedactionSweep?.passed !== true) {
340
- failures.push('guardbench-summary.json: artifactRedactionSweep did not pass');
341
- }
342
- if (artifacts.raw.artifactRedactionSweep?.passed !== true) {
343
- failures.push('guardbench-raw.json: artifactRedactionSweep did not pass');
344
- }
345
-
346
- const artifactText = Object.values(artifacts)
347
- .map(value => JSON.stringify(value))
348
- .join('\n');
349
- for (const secret of seededSecrets) {
350
- if (secret && artifactText.includes(secret)) {
351
- failures.push(`raw seeded secret leaked into GuardBench artifacts: ${secret}`);
352
- }
353
- }
354
- const manifestText = JSON.stringify(artifacts.manifest);
355
- if (!manifestText.includes('seededSecretRefs')) {
356
- failures.push('guardbench-manifest.json: missing seededSecretRefs');
357
- }
358
- if (manifestText.includes('"seededSecrets"')) {
359
- failures.push('guardbench-manifest.json: contains seededSecrets');
360
- }
361
- }
362
-
363
- return {
364
- ok: failures.length === 0,
365
- dir: publicPath(dir),
366
- schemasDir: publicPath(schemasDir),
367
- files: Object.values(ARTIFACT_FILES),
368
- optionalFiles: Object.values(OPTIONAL_ARTIFACT_FILES).filter(file =>
369
- existsSync(join(dir, file)),
370
- ),
371
- failures,
372
- };
373
- }
374
-
375
- async function main() {
376
- const args = parseArgs();
377
- if (args.help) {
378
- console.log(usage());
379
- return;
380
- }
381
-
382
- const report = validateGuardBenchArtifacts(args);
383
- if (args.json) {
384
- console.log(JSON.stringify(report, null, 2));
385
- } else if (report.ok) {
386
- console.log(`GuardBench artifact validation passed: ${report.dir}`);
387
- } else {
388
- console.error('GuardBench artifact validation failed:');
389
- for (const failure of report.failures) console.error(`- ${failure}`);
390
- }
391
-
392
- if (!report.ok) process.exit(1);
393
- }
394
-
395
- if (process.argv[1] && resolve(process.argv[1]) === fileURLToPath(import.meta.url)) {
396
- main().catch(error => {
397
- console.error(error.stack ?? error.message);
398
- process.exit(1);
399
- });
400
- }