audrey 1.1.2 → 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 +101 -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 +4 -38
  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
@@ -1,190 +0,0 @@
1
- import { cpSync, existsSync, mkdirSync, readFileSync, rmSync, writeFileSync } from 'node:fs';
2
- import { createHash } from 'node:crypto';
3
- import { basename, isAbsolute, join, relative, resolve } from 'node:path';
4
- import { writeGuardBenchConformanceCard } from './create-conformance-card.mjs';
5
- import { validateGuardBenchArtifacts } from './validate-guardbench-artifacts.mjs';
6
- import { publicPath } from './public-paths.mjs';
7
-
8
- const REQUIRED_ARTIFACTS = [
9
- 'guardbench-manifest.json',
10
- 'guardbench-summary.json',
11
- 'guardbench-raw.json',
12
- 'guardbench-conformance-card.json',
13
- ];
14
- const OPTIONAL_ARTIFACTS = ['external-run-metadata.json'];
15
- const SCHEMA_FILES = [
16
- 'guardbench-adapter-registry.schema.json',
17
- 'guardbench-adapter-self-test.schema.json',
18
- 'guardbench-external-dry-run.schema.json',
19
- 'guardbench-external-evidence.schema.json',
20
- 'guardbench-publication-verification.schema.json',
21
- 'guardbench-manifest.schema.json',
22
- 'guardbench-summary.schema.json',
23
- 'guardbench-raw.schema.json',
24
- 'guardbench-external-run.schema.json',
25
- 'guardbench-conformance-card.schema.json',
26
- 'guardbench-leaderboard.schema.json',
27
- 'guardbench-submission-manifest.schema.json',
28
- ];
29
-
30
- function readJson(path) {
31
- return JSON.parse(readFileSync(path, 'utf-8'));
32
- }
33
-
34
- function sha256File(path) {
35
- return createHash('sha256').update(readFileSync(path)).digest('hex');
36
- }
37
-
38
- function copyFileInto(sourceDir, outDir, file) {
39
- const source = join(sourceDir, file);
40
- if (!existsSync(source)) return null;
41
- const target = join(outDir, file);
42
- cpSync(source, target);
43
- return target;
44
- }
45
-
46
- export function bundleRelativeFilePath(path, root) {
47
- const relativePath = relative(root, path);
48
- if (!relativePath || relativePath.startsWith('..') || isAbsolute(relativePath)) {
49
- throw new Error(`Cannot add file outside GuardBench submission bundle: ${path}`);
50
- }
51
- return relativePath.replaceAll('\\', '/');
52
- }
53
-
54
- function fileRecord(path, root) {
55
- return {
56
- path: bundleRelativeFilePath(path, root),
57
- bytes: readFileSync(path).byteLength,
58
- sha256: sha256File(path),
59
- };
60
- }
61
-
62
- export function writeGuardBenchSubmissionBundle(options = {}) {
63
- const sourceDir = resolve(options.dir ?? 'benchmarks/output');
64
- const outDir = resolve(options.outDir ?? join(sourceDir, 'submission-bundle'));
65
- const schemasDir = resolve(options.schemasDir ?? 'benchmarks/schemas');
66
-
67
- writeGuardBenchConformanceCard({ dir: sourceDir });
68
- const sourceValidation = validateGuardBenchArtifacts({ dir: sourceDir, schemasDir });
69
- if (!sourceValidation.ok) {
70
- throw new Error(
71
- `Cannot create GuardBench submission bundle from invalid artifacts: ${sourceValidation.failures.join('; ')}`,
72
- );
73
- }
74
-
75
- rmSync(outDir, { recursive: true, force: true });
76
- mkdirSync(join(outDir, 'schemas'), { recursive: true });
77
-
78
- const copied = [];
79
- for (const file of [...REQUIRED_ARTIFACTS, ...OPTIONAL_ARTIFACTS]) {
80
- const target = copyFileInto(sourceDir, outDir, file);
81
- if (target) copied.push(target);
82
- }
83
- for (const file of SCHEMA_FILES) {
84
- const target = join(outDir, 'schemas', file);
85
- cpSync(join(schemasDir, file), target);
86
- copied.push(target);
87
- }
88
-
89
- const bundleValidation = validateGuardBenchArtifacts({
90
- dir: outDir,
91
- schemasDir: join(outDir, 'schemas'),
92
- });
93
- const validationReportPath = join(outDir, 'validation-report.json');
94
- writeFileSync(
95
- validationReportPath,
96
- `${JSON.stringify(
97
- {
98
- generatedAt: new Date().toISOString(),
99
- sourceValidation,
100
- bundleValidation,
101
- },
102
- null,
103
- 2,
104
- )}\n`,
105
- 'utf-8',
106
- );
107
- copied.push(validationReportPath);
108
-
109
- const card = readJson(join(outDir, 'guardbench-conformance-card.json'));
110
- const manifestPath = join(outDir, 'submission-manifest.json');
111
- const manifest = {
112
- schemaVersion: '1.0.0',
113
- suite: 'GuardBench submission bundle',
114
- generatedAt: new Date().toISOString(),
115
- sourceDir: publicPath(sourceDir),
116
- subject: card.subject,
117
- score: card.score,
118
- conformance: card.conformance,
119
- validation: bundleValidation,
120
- files: copied
121
- .map(path => fileRecord(path, outDir))
122
- .sort((a, b) => a.path.localeCompare(b.path)),
123
- };
124
- writeFileSync(manifestPath, `${JSON.stringify(manifest, null, 2)}\n`, 'utf-8');
125
-
126
- const finalFiles = copied
127
- .map(path => fileRecord(path, outDir))
128
- .sort((a, b) => a.path.localeCompare(b.path));
129
- manifest.files = finalFiles;
130
- writeFileSync(manifestPath, `${JSON.stringify(manifest, null, 2)}\n`, 'utf-8');
131
-
132
- return {
133
- outDir,
134
- manifestPath,
135
- validation: bundleValidation,
136
- subject: card.subject,
137
- files: finalFiles.map(record => record.path),
138
- };
139
- }
140
-
141
- function parseArgs(argv = process.argv.slice(2)) {
142
- const args = {
143
- dir: 'benchmarks/output',
144
- outDir: null,
145
- schemasDir: 'benchmarks/schemas',
146
- json: false,
147
- };
148
- for (let i = 0; i < argv.length; i++) {
149
- const token = argv[i];
150
- if ((token === '--dir' || token === '--source-dir') && argv[i + 1]) args.dir = argv[++i];
151
- else if (token === '--out-dir' && argv[i + 1]) args.outDir = argv[++i];
152
- else if (token === '--schemas-dir' && argv[i + 1]) args.schemasDir = argv[++i];
153
- else if (token === '--json') args.json = true;
154
- else if (token === '--help' || token === '-h') args.help = true;
155
- else throw new Error(`Unknown argument: ${token}`);
156
- }
157
- return args;
158
- }
159
-
160
- function usage() {
161
- return [
162
- 'Usage: node benchmarks/create-submission-bundle.mjs [--dir benchmarks/output] [--out-dir <dir>] [--json]',
163
- '',
164
- 'Creates a portable GuardBench submission bundle containing artifacts, schemas,',
165
- 'a conformance card, validation report, and submission manifest.',
166
- ].join('\n');
167
- }
168
-
169
- async function main() {
170
- const args = parseArgs();
171
- if (args.help) {
172
- console.log(usage());
173
- return;
174
- }
175
- const result = writeGuardBenchSubmissionBundle(args);
176
- if (args.json) console.log(JSON.stringify(result, null, 2));
177
- else {
178
- console.log(`GuardBench submission bundle: ${result.outDir}`);
179
- console.log(`Subject: ${result.subject.name}`);
180
- console.log(`Validation: ${result.validation.ok ? 'passed' : 'failed'}`);
181
- }
182
- if (!result.validation.ok) process.exit(1);
183
- }
184
-
185
- if (process.argv[1] && resolve(process.argv[1]).endsWith(basename(import.meta.url))) {
186
- main().catch(error => {
187
- console.error(error.stack ?? error.message);
188
- process.exit(1);
189
- });
190
- }
@@ -1,177 +0,0 @@
1
- import { mkdirSync, readFileSync, writeFileSync } from 'node:fs';
2
- import { dirname, join, resolve } from 'node:path';
3
- import { fileURLToPath } from 'node:url';
4
- import {
5
- buildExternalGuardBenchRun,
6
- writeExternalRunMetadata,
7
- } from './run-external-guardbench.mjs';
8
- import { validateAdapterRegistry } from './validate-adapter-registry.mjs';
9
- import { validateSchema } from './validate-guardbench-artifacts.mjs';
10
- import { publicCommand, publicPath } from './public-paths.mjs';
11
-
12
- const DEFAULT_REGISTRY = 'benchmarks/adapters/registry.json';
13
- const DEFAULT_OUT_ROOT = 'benchmarks/output/external';
14
- const DEFAULT_OUT = 'benchmarks/output/external/guardbench-external-dry-run.json';
15
- const DEFAULT_SCHEMA = 'benchmarks/schemas/guardbench-external-dry-run.schema.json';
16
-
17
- export function parseExternalDryRunArgs(argv = process.argv.slice(2)) {
18
- const args = {
19
- registry: DEFAULT_REGISTRY,
20
- outRoot: DEFAULT_OUT_ROOT,
21
- out: DEFAULT_OUT,
22
- includeCredentialFree: false,
23
- json: false,
24
- noWrite: false,
25
- };
26
-
27
- for (let i = 0; i < argv.length; i++) {
28
- const token = argv[i];
29
- if (token === '--registry' && argv[i + 1]) args.registry = argv[++i];
30
- else if (token === '--out-root' && argv[i + 1]) args.outRoot = argv[++i];
31
- else if (token === '--out' && argv[i + 1]) args.out = argv[++i];
32
- else if (token === '--include-credential-free') args.includeCredentialFree = true;
33
- else if (token === '--json') args.json = true;
34
- else if (token === '--no-write') args.noWrite = true;
35
- else if (token === '--help' || token === '-h') args.help = true;
36
- else throw new Error(`Unknown argument: ${token}`);
37
- }
38
-
39
- return args;
40
- }
41
-
42
- function usage() {
43
- return `Usage: node benchmarks/dry-run-external-adapters.mjs [options]
44
-
45
- Options:
46
- --registry <path> Adapter registry JSON. Default: ${DEFAULT_REGISTRY}.
47
- --out-root <path> Root directory for dry-run metadata. Default: ${DEFAULT_OUT_ROOT}.
48
- --out <path> Matrix JSON report path. Default: ${DEFAULT_OUT}.
49
- --include-credential-free Include credential-free registry adapters.
50
- --json Print the machine-readable dry-run matrix.
51
- --no-write Do not write the matrix JSON report.
52
- `;
53
- }
54
-
55
- function readJson(path) {
56
- return JSON.parse(readFileSync(path, 'utf-8'));
57
- }
58
-
59
- export function validateExternalAdapterDryRunMatrix(matrix, options = {}) {
60
- const schema = options.schemaObject ?? readJson(options.schema ?? DEFAULT_SCHEMA);
61
- return validateSchema(matrix, schema, 'guardbench-external-dry-run');
62
- }
63
-
64
- export function writeExternalAdapterDryRunMatrix(matrix, out = DEFAULT_OUT) {
65
- const outPath = resolve(out);
66
- mkdirSync(dirname(outPath), { recursive: true });
67
- writeFileSync(outPath, `${JSON.stringify(matrix, null, 2)}\n`, 'utf-8');
68
- return outPath;
69
- }
70
-
71
- export async function buildExternalAdapterDryRunMatrix(options = {}) {
72
- const registryPath = options.registry ?? DEFAULT_REGISTRY;
73
- const outRoot = resolve(options.outRoot ?? DEFAULT_OUT_ROOT);
74
- const registryValidation = await validateAdapterRegistry({ registry: registryPath });
75
- const registry = readJson(registryPath);
76
- const adapters = registry.adapters.filter(
77
- adapter =>
78
- adapter.status === 'external-system' &&
79
- (options.includeCredentialFree || adapter.credentialMode === 'runtime-env'),
80
- );
81
- const rows = [];
82
- const failures = [];
83
-
84
- if (!registryValidation.ok) {
85
- failures.push(...registryValidation.failures.map(failure => `registry: ${failure}`));
86
- }
87
-
88
- for (const adapter of adapters) {
89
- const run = buildExternalGuardBenchRun(
90
- {
91
- adapter: adapter.id,
92
- outDir: join(outRoot, adapter.id),
93
- check: true,
94
- json: true,
95
- },
96
- options.env ?? process.env,
97
- );
98
- const metadata = {
99
- suite: 'GuardBench external adapter run',
100
- startedAt: new Date().toISOString(),
101
- adapter: run.adapter,
102
- adapterPath: run.adapterPath,
103
- outDir: run.outDir,
104
- requiredEnv: run.requiredEnv,
105
- missingEnv: run.missingEnv,
106
- command: run.command,
107
- validationCommand: run.validationCommand,
108
- dryRun: true,
109
- status: run.missingEnv.length ? 'dry-run-missing-env' : 'dry-run-ready',
110
- };
111
- const metadataPath = writeExternalRunMetadata(run.outDir, metadata);
112
- rows.push({
113
- id: adapter.id,
114
- name: adapter.name,
115
- credentialMode: adapter.credentialMode,
116
- requiredEnv: run.requiredEnv,
117
- missingEnv: run.missingEnv,
118
- status: metadata.status,
119
- command: publicCommand(run.command),
120
- validationCommand: publicCommand(run.validationCommand),
121
- metadataPath: publicPath(metadataPath),
122
- });
123
- }
124
-
125
- const matrix = {
126
- schemaVersion: '1.0.0',
127
- suite: 'GuardBench external adapter dry-run matrix',
128
- generatedAt: new Date().toISOString(),
129
- ok: failures.length === 0,
130
- registry: publicPath(resolve(registryPath)),
131
- outRoot: publicPath(outRoot),
132
- adapters: rows,
133
- failures,
134
- };
135
- const schemaFailures = validateExternalAdapterDryRunMatrix(matrix, options);
136
- if (schemaFailures.length > 0) {
137
- throw new Error(
138
- `GuardBench external adapter dry-run schema validation failed: ${schemaFailures.join('; ')}`,
139
- );
140
- }
141
- return matrix;
142
- }
143
-
144
- async function main() {
145
- const args = parseExternalDryRunArgs();
146
- if (args.help) {
147
- console.log(usage());
148
- return;
149
- }
150
-
151
- const matrix = await buildExternalAdapterDryRunMatrix(args);
152
- const outPath = args.noWrite ? null : writeExternalAdapterDryRunMatrix(matrix, args.out);
153
- if (args.json) {
154
- console.log(JSON.stringify(matrix, null, 2));
155
- } else if (matrix.ok) {
156
- console.log(
157
- `GuardBench external adapter dry-run matrix passed: ${matrix.adapters.length} adapter(s)`,
158
- );
159
- for (const row of matrix.adapters) {
160
- const missing = row.missingEnv.length ? `missing ${row.missingEnv.join(', ')}` : 'ready';
161
- console.log(`- ${row.id}: ${missing}; metadata ${row.metadataPath}`);
162
- }
163
- if (outPath) console.log(`Matrix report: ${outPath}`);
164
- } else {
165
- console.error('GuardBench external adapter dry-run matrix failed:');
166
- for (const failure of matrix.failures) console.error(`- ${failure}`);
167
- }
168
-
169
- if (!matrix.ok) process.exit(1);
170
- }
171
-
172
- if (process.argv[1] && resolve(process.argv[1]) === fileURLToPath(import.meta.url)) {
173
- main().catch(error => {
174
- console.error(error.stack ?? error.message);
175
- process.exit(1);
176
- });
177
- }