@claude-flow/cli 3.10.35 → 3.10.37

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 (515) hide show
  1. package/.claude/agents/analysis/analyze-code-quality.md +178 -178
  2. package/.claude/agents/analysis/code-analyzer.md +209 -209
  3. package/.claude/agents/analysis/code-review/analyze-code-quality.md +178 -178
  4. package/.claude/agents/architecture/arch-system-design.md +156 -156
  5. package/.claude/agents/architecture/system-design/arch-system-design.md +154 -154
  6. package/.claude/agents/browser/browser-agent.yaml +182 -182
  7. package/.claude/agents/consensus/byzantine-coordinator.md +62 -62
  8. package/.claude/agents/consensus/crdt-synchronizer.md +996 -996
  9. package/.claude/agents/consensus/gossip-coordinator.md +62 -62
  10. package/.claude/agents/consensus/performance-benchmarker.md +850 -850
  11. package/.claude/agents/consensus/quorum-manager.md +822 -822
  12. package/.claude/agents/consensus/raft-manager.md +62 -62
  13. package/.claude/agents/consensus/security-manager.md +621 -621
  14. package/.claude/agents/core/planner.md +374 -374
  15. package/.claude/agents/custom/test-long-runner.md +44 -44
  16. package/.claude/agents/data/data-ml-model.md +444 -444
  17. package/.claude/agents/data/ml/data-ml-model.md +192 -192
  18. package/.claude/agents/development/backend/dev-backend-api.md +141 -141
  19. package/.claude/agents/development/dev-backend-api.md +344 -344
  20. package/.claude/agents/devops/ci-cd/ops-cicd-github.md +163 -163
  21. package/.claude/agents/devops/ops-cicd-github.md +164 -164
  22. package/.claude/agents/documentation/api-docs/docs-api-openapi.md +173 -173
  23. package/.claude/agents/documentation/docs-api-openapi.md +354 -354
  24. package/.claude/agents/flow-nexus/app-store.md +87 -87
  25. package/.claude/agents/flow-nexus/authentication.md +68 -68
  26. package/.claude/agents/flow-nexus/challenges.md +80 -80
  27. package/.claude/agents/flow-nexus/neural-network.md +87 -87
  28. package/.claude/agents/flow-nexus/payments.md +82 -82
  29. package/.claude/agents/flow-nexus/sandbox.md +75 -75
  30. package/.claude/agents/flow-nexus/swarm.md +75 -75
  31. package/.claude/agents/flow-nexus/user-tools.md +95 -95
  32. package/.claude/agents/flow-nexus/workflow.md +83 -83
  33. package/.claude/agents/github/code-review-swarm.md +377 -377
  34. package/.claude/agents/github/github-modes.md +172 -172
  35. package/.claude/agents/github/issue-tracker.md +575 -575
  36. package/.claude/agents/github/multi-repo-swarm.md +552 -552
  37. package/.claude/agents/github/pr-manager.md +437 -437
  38. package/.claude/agents/github/project-board-sync.md +508 -508
  39. package/.claude/agents/github/release-manager.md +604 -604
  40. package/.claude/agents/github/release-swarm.md +582 -582
  41. package/.claude/agents/github/repo-architect.md +397 -397
  42. package/.claude/agents/github/swarm-issue.md +572 -572
  43. package/.claude/agents/github/swarm-pr.md +427 -427
  44. package/.claude/agents/github/sync-coordinator.md +451 -451
  45. package/.claude/agents/github/workflow-automation.md +902 -902
  46. package/.claude/agents/goal/agent.md +815 -815
  47. package/.claude/agents/optimization/benchmark-suite.md +664 -664
  48. package/.claude/agents/optimization/load-balancer.md +430 -430
  49. package/.claude/agents/optimization/performance-monitor.md +671 -671
  50. package/.claude/agents/optimization/resource-allocator.md +673 -673
  51. package/.claude/agents/optimization/topology-optimizer.md +807 -807
  52. package/.claude/agents/payments/agentic-payments.md +126 -126
  53. package/.claude/agents/sona/sona-learning-optimizer.md +74 -74
  54. package/.claude/agents/sparc/architecture.md +698 -698
  55. package/.claude/agents/sparc/pseudocode.md +519 -519
  56. package/.claude/agents/sparc/refinement.md +801 -801
  57. package/.claude/agents/sparc/specification.md +477 -477
  58. package/.claude/agents/specialized/mobile/spec-mobile-react-native.md +224 -224
  59. package/.claude/agents/specialized/spec-mobile-react-native.md +226 -226
  60. package/.claude/agents/sublinear/consensus-coordinator.md +337 -337
  61. package/.claude/agents/sublinear/matrix-optimizer.md +184 -184
  62. package/.claude/agents/sublinear/pagerank-analyzer.md +298 -298
  63. package/.claude/agents/sublinear/performance-optimizer.md +367 -367
  64. package/.claude/agents/sublinear/trading-predictor.md +245 -245
  65. package/.claude/agents/swarm/adaptive-coordinator.md +1126 -1126
  66. package/.claude/agents/swarm/hierarchical-coordinator.md +709 -709
  67. package/.claude/agents/swarm/mesh-coordinator.md +962 -962
  68. package/.claude/agents/templates/automation-smart-agent.md +204 -204
  69. package/.claude/agents/templates/base-template-generator.md +289 -289
  70. package/.claude/agents/templates/coordinator-swarm-init.md +89 -89
  71. package/.claude/agents/templates/github-pr-manager.md +176 -176
  72. package/.claude/agents/templates/implementer-sparc-coder.md +258 -258
  73. package/.claude/agents/templates/memory-coordinator.md +186 -186
  74. package/.claude/agents/templates/orchestrator-task.md +138 -138
  75. package/.claude/agents/templates/performance-analyzer.md +198 -198
  76. package/.claude/agents/templates/sparc-coordinator.md +513 -513
  77. package/.claude/agents/testing/production-validator.md +394 -394
  78. package/.claude/agents/testing/tdd-london-swarm.md +243 -243
  79. package/.claude/agents/v3/aidefence-guardian.md +282 -282
  80. package/.claude/agents/v3/claims-authorizer.md +208 -208
  81. package/.claude/agents/v3/collective-intelligence-coordinator.md +993 -993
  82. package/.claude/agents/v3/ddd-domain-expert.md +220 -220
  83. package/.claude/agents/v3/injection-analyst.md +236 -236
  84. package/.claude/agents/v3/performance-engineer.md +1233 -1233
  85. package/.claude/agents/v3/pii-detector.md +151 -151
  86. package/.claude/agents/v3/reasoningbank-learner.md +213 -213
  87. package/.claude/agents/v3/security-architect-aidefence.md +410 -410
  88. package/.claude/agents/v3/security-architect.md +867 -867
  89. package/.claude/agents/v3/swarm-memory-manager.md +157 -157
  90. package/.claude/agents/v3/v3-integration-architect.md +205 -205
  91. package/.claude/commands/agents/README.md +50 -50
  92. package/.claude/commands/agents/agent-capabilities.md +140 -140
  93. package/.claude/commands/agents/agent-coordination.md +28 -28
  94. package/.claude/commands/agents/agent-spawning.md +28 -28
  95. package/.claude/commands/agents/agent-types.md +216 -216
  96. package/.claude/commands/agents/health.md +139 -139
  97. package/.claude/commands/agents/list.md +100 -100
  98. package/.claude/commands/agents/logs.md +130 -130
  99. package/.claude/commands/agents/metrics.md +122 -122
  100. package/.claude/commands/agents/pool.md +127 -127
  101. package/.claude/commands/agents/spawn.md +140 -140
  102. package/.claude/commands/agents/status.md +115 -115
  103. package/.claude/commands/agents/stop.md +102 -102
  104. package/.claude/commands/analysis/COMMAND_COMPLIANCE_REPORT.md +53 -53
  105. package/.claude/commands/analysis/README.md +9 -9
  106. package/.claude/commands/analysis/bottleneck-detect.md +162 -162
  107. package/.claude/commands/analysis/performance-bottlenecks.md +58 -58
  108. package/.claude/commands/analysis/performance-report.md +25 -25
  109. package/.claude/commands/analysis/token-efficiency.md +44 -44
  110. package/.claude/commands/analysis/token-usage.md +25 -25
  111. package/.claude/commands/automation/README.md +9 -9
  112. package/.claude/commands/automation/auto-agent.md +122 -122
  113. package/.claude/commands/automation/self-healing.md +105 -105
  114. package/.claude/commands/automation/session-memory.md +89 -89
  115. package/.claude/commands/automation/smart-agents.md +72 -72
  116. package/.claude/commands/automation/smart-spawn.md +25 -25
  117. package/.claude/commands/automation/workflow-select.md +25 -25
  118. package/.claude/commands/claude-flow-help.md +103 -103
  119. package/.claude/commands/claude-flow-memory.md +107 -107
  120. package/.claude/commands/claude-flow-swarm.md +205 -205
  121. package/.claude/commands/coordination/README.md +9 -9
  122. package/.claude/commands/coordination/agent-spawn.md +25 -25
  123. package/.claude/commands/coordination/init.md +44 -44
  124. package/.claude/commands/coordination/orchestrate.md +43 -43
  125. package/.claude/commands/coordination/spawn.md +45 -45
  126. package/.claude/commands/coordination/swarm-init.md +85 -85
  127. package/.claude/commands/coordination/task-orchestrate.md +25 -25
  128. package/.claude/commands/github/README.md +11 -11
  129. package/.claude/commands/github/code-review-swarm.md +513 -513
  130. package/.claude/commands/github/code-review.md +25 -25
  131. package/.claude/commands/github/github-modes.md +146 -146
  132. package/.claude/commands/github/github-swarm.md +121 -121
  133. package/.claude/commands/github/issue-tracker.md +291 -291
  134. package/.claude/commands/github/issue-triage.md +25 -25
  135. package/.claude/commands/github/multi-repo-swarm.md +518 -518
  136. package/.claude/commands/github/pr-enhance.md +26 -26
  137. package/.claude/commands/github/pr-manager.md +169 -169
  138. package/.claude/commands/github/project-board-sync.md +470 -470
  139. package/.claude/commands/github/release-manager.md +339 -339
  140. package/.claude/commands/github/release-swarm.md +543 -543
  141. package/.claude/commands/github/repo-analyze.md +25 -25
  142. package/.claude/commands/github/repo-architect.md +366 -366
  143. package/.claude/commands/github/swarm-issue.md +484 -484
  144. package/.claude/commands/github/swarm-pr.md +287 -287
  145. package/.claude/commands/github/sync-coordinator.md +302 -302
  146. package/.claude/commands/github/workflow-automation.md +441 -441
  147. package/.claude/commands/hive-mind/README.md +17 -17
  148. package/.claude/commands/hive-mind/hive-mind-consensus.md +8 -8
  149. package/.claude/commands/hive-mind/hive-mind-init.md +18 -18
  150. package/.claude/commands/hive-mind/hive-mind-memory.md +8 -8
  151. package/.claude/commands/hive-mind/hive-mind-metrics.md +8 -8
  152. package/.claude/commands/hive-mind/hive-mind-resume.md +8 -8
  153. package/.claude/commands/hive-mind/hive-mind-sessions.md +8 -8
  154. package/.claude/commands/hive-mind/hive-mind-spawn.md +21 -21
  155. package/.claude/commands/hive-mind/hive-mind-status.md +8 -8
  156. package/.claude/commands/hive-mind/hive-mind-stop.md +8 -8
  157. package/.claude/commands/hive-mind/hive-mind-wizard.md +8 -8
  158. package/.claude/commands/hive-mind/hive-mind.md +27 -27
  159. package/.claude/commands/hooks/README.md +11 -11
  160. package/.claude/commands/hooks/overview.md +57 -57
  161. package/.claude/commands/hooks/post-edit.md +117 -117
  162. package/.claude/commands/hooks/post-task.md +112 -112
  163. package/.claude/commands/hooks/pre-edit.md +113 -113
  164. package/.claude/commands/hooks/pre-task.md +111 -111
  165. package/.claude/commands/hooks/session-end.md +118 -118
  166. package/.claude/commands/hooks/setup.md +102 -102
  167. package/.claude/commands/memory/README.md +9 -9
  168. package/.claude/commands/memory/memory-persist.md +25 -25
  169. package/.claude/commands/memory/memory-search.md +25 -25
  170. package/.claude/commands/memory/memory-usage.md +25 -25
  171. package/.claude/commands/memory/neural.md +47 -47
  172. package/.claude/commands/monitoring/README.md +9 -9
  173. package/.claude/commands/monitoring/agent-metrics.md +25 -25
  174. package/.claude/commands/monitoring/agents.md +44 -44
  175. package/.claude/commands/monitoring/real-time-view.md +25 -25
  176. package/.claude/commands/monitoring/status.md +46 -46
  177. package/.claude/commands/monitoring/swarm-monitor.md +25 -25
  178. package/.claude/commands/optimization/README.md +9 -9
  179. package/.claude/commands/optimization/auto-topology.md +61 -61
  180. package/.claude/commands/optimization/cache-manage.md +25 -25
  181. package/.claude/commands/optimization/parallel-execute.md +25 -25
  182. package/.claude/commands/optimization/parallel-execution.md +49 -49
  183. package/.claude/commands/optimization/topology-optimize.md +25 -25
  184. package/.claude/commands/pair/README.md +260 -260
  185. package/.claude/commands/pair/commands.md +545 -545
  186. package/.claude/commands/pair/config.md +509 -509
  187. package/.claude/commands/pair/examples.md +511 -511
  188. package/.claude/commands/pair/modes.md +347 -347
  189. package/.claude/commands/pair/session.md +406 -406
  190. package/.claude/commands/pair/start.md +208 -208
  191. package/.claude/commands/sparc/analyzer.md +51 -51
  192. package/.claude/commands/sparc/architect.md +53 -53
  193. package/.claude/commands/sparc/ask.md +97 -97
  194. package/.claude/commands/sparc/batch-executor.md +54 -54
  195. package/.claude/commands/sparc/code.md +89 -89
  196. package/.claude/commands/sparc/coder.md +54 -54
  197. package/.claude/commands/sparc/debug.md +83 -83
  198. package/.claude/commands/sparc/debugger.md +54 -54
  199. package/.claude/commands/sparc/designer.md +53 -53
  200. package/.claude/commands/sparc/devops.md +109 -109
  201. package/.claude/commands/sparc/docs-writer.md +80 -80
  202. package/.claude/commands/sparc/documenter.md +54 -54
  203. package/.claude/commands/sparc/innovator.md +54 -54
  204. package/.claude/commands/sparc/integration.md +83 -83
  205. package/.claude/commands/sparc/mcp.md +117 -117
  206. package/.claude/commands/sparc/memory-manager.md +54 -54
  207. package/.claude/commands/sparc/optimizer.md +54 -54
  208. package/.claude/commands/sparc/orchestrator.md +131 -131
  209. package/.claude/commands/sparc/post-deployment-monitoring-mode.md +83 -83
  210. package/.claude/commands/sparc/refinement-optimization-mode.md +83 -83
  211. package/.claude/commands/sparc/researcher.md +54 -54
  212. package/.claude/commands/sparc/reviewer.md +54 -54
  213. package/.claude/commands/sparc/security-review.md +80 -80
  214. package/.claude/commands/sparc/sparc-modes.md +174 -174
  215. package/.claude/commands/sparc/sparc.md +111 -111
  216. package/.claude/commands/sparc/spec-pseudocode.md +80 -80
  217. package/.claude/commands/sparc/supabase-admin.md +348 -348
  218. package/.claude/commands/sparc/swarm-coordinator.md +54 -54
  219. package/.claude/commands/sparc/tdd.md +54 -54
  220. package/.claude/commands/sparc/tester.md +54 -54
  221. package/.claude/commands/sparc/tutorial.md +79 -79
  222. package/.claude/commands/sparc/workflow-manager.md +54 -54
  223. package/.claude/commands/sparc.md +166 -166
  224. package/.claude/commands/stream-chain/pipeline.md +120 -120
  225. package/.claude/commands/stream-chain/run.md +69 -69
  226. package/.claude/commands/swarm/README.md +15 -15
  227. package/.claude/commands/swarm/analysis.md +95 -95
  228. package/.claude/commands/swarm/development.md +96 -96
  229. package/.claude/commands/swarm/examples.md +168 -168
  230. package/.claude/commands/swarm/maintenance.md +102 -102
  231. package/.claude/commands/swarm/optimization.md +117 -117
  232. package/.claude/commands/swarm/research.md +136 -136
  233. package/.claude/commands/swarm/swarm-analysis.md +8 -8
  234. package/.claude/commands/swarm/swarm-background.md +8 -8
  235. package/.claude/commands/swarm/swarm-init.md +19 -19
  236. package/.claude/commands/swarm/swarm-modes.md +8 -8
  237. package/.claude/commands/swarm/swarm-monitor.md +8 -8
  238. package/.claude/commands/swarm/swarm-spawn.md +19 -19
  239. package/.claude/commands/swarm/swarm-status.md +8 -8
  240. package/.claude/commands/swarm/swarm-strategies.md +8 -8
  241. package/.claude/commands/swarm/swarm.md +87 -87
  242. package/.claude/commands/swarm/testing.md +131 -131
  243. package/.claude/commands/training/README.md +9 -9
  244. package/.claude/commands/training/model-update.md +25 -25
  245. package/.claude/commands/training/neural-patterns.md +107 -107
  246. package/.claude/commands/training/neural-train.md +75 -75
  247. package/.claude/commands/training/pattern-learn.md +25 -25
  248. package/.claude/commands/training/specialization.md +62 -62
  249. package/.claude/commands/truth/start.md +142 -142
  250. package/.claude/commands/verify/check.md +49 -49
  251. package/.claude/commands/verify/start.md +127 -127
  252. package/.claude/commands/workflows/README.md +9 -9
  253. package/.claude/commands/workflows/development.md +77 -77
  254. package/.claude/commands/workflows/research.md +62 -62
  255. package/.claude/commands/workflows/workflow-create.md +25 -25
  256. package/.claude/commands/workflows/workflow-execute.md +25 -25
  257. package/.claude/commands/workflows/workflow-export.md +25 -25
  258. package/.claude/helpers/README.md +96 -96
  259. package/.claude/helpers/adr-compliance.sh +186 -186
  260. package/.claude/helpers/auto-commit.sh +178 -178
  261. package/.claude/helpers/auto-memory-hook.mjs +368 -368
  262. package/.claude/helpers/checkpoint-manager.sh +251 -251
  263. package/.claude/helpers/daemon-manager.sh +252 -252
  264. package/.claude/helpers/ddd-tracker.sh +144 -144
  265. package/.claude/helpers/github-safe.js +156 -156
  266. package/.claude/helpers/github-setup.sh +45 -45
  267. package/.claude/helpers/guidance-hook.sh +13 -13
  268. package/.claude/helpers/guidance-hooks.sh +102 -102
  269. package/.claude/helpers/health-monitor.sh +108 -108
  270. package/.claude/helpers/hook-handler.cjs +286 -286
  271. package/.claude/helpers/intelligence.cjs +1031 -1031
  272. package/.claude/helpers/learning-hooks.sh +329 -329
  273. package/.claude/helpers/learning-optimizer.sh +127 -127
  274. package/.claude/helpers/learning-service.mjs +1144 -1144
  275. package/.claude/helpers/memory.js +83 -83
  276. package/.claude/helpers/metrics-db.mjs +488 -488
  277. package/.claude/helpers/pattern-consolidator.sh +86 -86
  278. package/.claude/helpers/perf-worker.sh +160 -160
  279. package/.claude/helpers/post-commit +16 -16
  280. package/.claude/helpers/pre-commit +26 -26
  281. package/.claude/helpers/quick-start.sh +19 -19
  282. package/.claude/helpers/router.js +105 -105
  283. package/.claude/helpers/security-scanner.sh +127 -127
  284. package/.claude/helpers/session.js +135 -135
  285. package/.claude/helpers/setup-mcp.sh +18 -18
  286. package/.claude/helpers/standard-checkpoint-hooks.sh +189 -189
  287. package/.claude/helpers/statusline-hook.sh +21 -21
  288. package/.claude/helpers/statusline.cjs +604 -604
  289. package/.claude/helpers/statusline.js +352 -352
  290. package/.claude/helpers/swarm-comms.sh +353 -353
  291. package/.claude/helpers/swarm-hooks.sh +761 -761
  292. package/.claude/helpers/swarm-monitor.sh +210 -210
  293. package/.claude/helpers/sync-v3-metrics.sh +245 -245
  294. package/.claude/helpers/update-v3-progress.sh +165 -165
  295. package/.claude/helpers/v3-quick-status.sh +57 -57
  296. package/.claude/helpers/v3.sh +110 -110
  297. package/.claude/helpers/validate-v3-config.sh +215 -215
  298. package/.claude/helpers/worker-manager.sh +170 -170
  299. package/.claude/settings.json +182 -182
  300. package/.claude/skills/agentdb-advanced/SKILL.md +550 -550
  301. package/.claude/skills/agentdb-learning/SKILL.md +545 -545
  302. package/.claude/skills/agentdb-memory-patterns/SKILL.md +339 -339
  303. package/.claude/skills/agentdb-optimization/SKILL.md +509 -509
  304. package/.claude/skills/agentdb-vector-search/SKILL.md +339 -339
  305. package/.claude/skills/browser/SKILL.md +204 -204
  306. package/.claude/skills/dual-mode/README.md +71 -71
  307. package/.claude/skills/dual-mode/dual-collect.md +103 -103
  308. package/.claude/skills/dual-mode/dual-coordinate.md +85 -85
  309. package/.claude/skills/dual-mode/dual-spawn.md +81 -81
  310. package/.claude/skills/flow-nexus-neural/SKILL.md +727 -727
  311. package/.claude/skills/flow-nexus-platform/SKILL.md +1154 -1154
  312. package/.claude/skills/flow-nexus-swarm/SKILL.md +604 -604
  313. package/.claude/skills/github-code-review/SKILL.md +1125 -1125
  314. package/.claude/skills/github-multi-repo/SKILL.md +862 -862
  315. package/.claude/skills/github-project-management/SKILL.md +1262 -1262
  316. package/.claude/skills/github-release-management/SKILL.md +1064 -1064
  317. package/.claude/skills/github-workflow-automation/SKILL.md +1047 -1047
  318. package/.claude/skills/hooks-automation/SKILL.md +1201 -1201
  319. package/.claude/skills/pair-programming/SKILL.md +1202 -1202
  320. package/.claude/skills/reasoningbank-agentdb/SKILL.md +446 -446
  321. package/.claude/skills/reasoningbank-intelligence/SKILL.md +201 -201
  322. package/.claude/skills/skill-builder/SKILL.md +910 -910
  323. package/.claude/skills/sparc-methodology/SKILL.md +1106 -1106
  324. package/.claude/skills/stream-chain/SKILL.md +560 -560
  325. package/.claude/skills/swarm-advanced/SKILL.md +970 -970
  326. package/.claude/skills/swarm-orchestration/SKILL.md +179 -179
  327. package/.claude/skills/v3-cli-modernization/SKILL.md +871 -871
  328. package/.claude/skills/v3-core-implementation/SKILL.md +796 -796
  329. package/.claude/skills/v3-ddd-architecture/SKILL.md +441 -441
  330. package/.claude/skills/v3-integration-deep/SKILL.md +240 -240
  331. package/.claude/skills/v3-mcp-optimization/SKILL.md +776 -776
  332. package/.claude/skills/v3-memory-unification/SKILL.md +173 -173
  333. package/.claude/skills/v3-performance-optimization/SKILL.md +389 -389
  334. package/.claude/skills/v3-security-overhaul/SKILL.md +81 -81
  335. package/.claude/skills/v3-swarm-coordination/SKILL.md +339 -339
  336. package/.claude/skills/verification-quality/SKILL.md +691 -691
  337. package/README.md +416 -416
  338. package/bin/cli.js +314 -286
  339. package/bin/mcp-server.js +224 -224
  340. package/bin/preinstall.cjs +2 -2
  341. package/dist/src/benchmarks/gaia-critic.js +24 -24
  342. package/dist/src/commands/completions.js +409 -409
  343. package/dist/src/commands/daemon.js +44 -44
  344. package/dist/src/commands/embeddings.js +29 -29
  345. package/dist/src/commands/embeddings.js.map +1 -1
  346. package/dist/src/commands/hive-mind.js +97 -97
  347. package/dist/src/commands/hooks.js +9 -9
  348. package/dist/src/commands/ruvector/backup.js +23 -23
  349. package/dist/src/commands/ruvector/benchmark.js +31 -31
  350. package/dist/src/commands/ruvector/import.js +14 -14
  351. package/dist/src/commands/ruvector/init.js +115 -115
  352. package/dist/src/commands/ruvector/migrate.js +99 -99
  353. package/dist/src/commands/ruvector/optimize.js +51 -51
  354. package/dist/src/commands/ruvector/setup.js +624 -624
  355. package/dist/src/commands/ruvector/status.js +38 -38
  356. package/dist/src/init/claudemd-generator.js +227 -227
  357. package/dist/src/init/executor.js +453 -453
  358. package/dist/src/init/helpers-generator.js +727 -727
  359. package/dist/src/init/statusline-generator.js +570 -570
  360. package/dist/src/log-filters.d.ts +21 -8
  361. package/dist/src/log-filters.d.ts.map +1 -1
  362. package/dist/src/log-filters.js +45 -8
  363. package/dist/src/log-filters.js.map +1 -1
  364. package/dist/src/mcp-tools/agentdb-tools.js +15 -15
  365. package/dist/src/memory/graph-edge-writer.js +18 -18
  366. package/dist/src/memory/memory-bridge.js +84 -84
  367. package/dist/src/memory/memory-initializer.js +393 -393
  368. package/dist/src/memory/rabitq-index.js +5 -5
  369. package/dist/src/runtime/headless.js +28 -28
  370. package/dist/src/ruvector/flash-attention.d.ts +195 -0
  371. package/dist/src/ruvector/flash-attention.d.ts.map +1 -0
  372. package/dist/src/ruvector/flash-attention.js +643 -0
  373. package/dist/src/ruvector/flash-attention.js.map +1 -0
  374. package/dist/src/ruvector/moe-router.d.ts +206 -0
  375. package/dist/src/ruvector/moe-router.d.ts.map +1 -0
  376. package/dist/src/ruvector/moe-router.js +626 -0
  377. package/dist/src/ruvector/moe-router.js.map +1 -0
  378. package/dist/src/services/event-stream.d.ts +25 -0
  379. package/dist/src/services/event-stream.js +27 -0
  380. package/dist/src/services/headless-worker-executor.js +84 -84
  381. package/dist/src/services/loop-worker-runner.d.ts +16 -0
  382. package/dist/src/services/loop-worker-runner.js +34 -0
  383. package/dist/src/services/runtime-capabilities.d.ts +22 -0
  384. package/dist/src/services/runtime-capabilities.js +45 -0
  385. package/dist/src/transfer/deploy-seraphine.js +23 -23
  386. package/dist/tsconfig.tsbuildinfo +1 -1
  387. package/package.json +134 -134
  388. package/scripts/beir-bootstrap-significance.mjs +162 -162
  389. package/scripts/bench-rvagent.mjs +312 -312
  390. package/scripts/benchmark-codemods.mjs +127 -127
  391. package/scripts/benchmark-cross-repo.mjs +206 -206
  392. package/scripts/benchmark-pretrained-retrieval.mjs +305 -305
  393. package/scripts/benchmark-self-learning.mjs +241 -241
  394. package/scripts/benchmark-trajectory-mrr.mjs +191 -191
  395. package/scripts/deploy-ipfs-node.sh +153 -153
  396. package/scripts/grid-search-retrieval.mjs +216 -216
  397. package/scripts/postinstall.cjs +153 -153
  398. package/scripts/pretrain-from-github.mjs +358 -358
  399. package/scripts/publish-registry.ts +345 -345
  400. package/scripts/publish.sh +57 -57
  401. package/scripts/run-beir-bge.mjs +324 -336
  402. package/scripts/run-beir-hybrid.mjs +261 -263
  403. package/scripts/run-beir-lucene-bm25.mjs +154 -154
  404. package/scripts/run-beir-nfcorpus.mjs +231 -231
  405. package/scripts/run-beir-rrf-ablation.mjs +221 -221
  406. package/scripts/setup-ipfs-registry.md +366 -366
  407. package/dist/src/benchmarks/capability-tasks.d.ts +0 -126
  408. package/dist/src/benchmarks/capability-tasks.d.ts.map +0 -1
  409. package/dist/src/benchmarks/capability-tasks.js +0 -147
  410. package/dist/src/benchmarks/capability-tasks.js.map +0 -1
  411. package/dist/src/benchmarks/gaia-agent-contrastive.d.ts +0 -53
  412. package/dist/src/benchmarks/gaia-agent-contrastive.d.ts.map +0 -1
  413. package/dist/src/benchmarks/gaia-agent-contrastive.js +0 -177
  414. package/dist/src/benchmarks/gaia-agent-contrastive.js.map +0 -1
  415. package/dist/src/benchmarks/gaia-agent-gemini.d.ts +0 -51
  416. package/dist/src/benchmarks/gaia-agent-gemini.d.ts.map +0 -1
  417. package/dist/src/benchmarks/gaia-agent-gemini.js +0 -352
  418. package/dist/src/benchmarks/gaia-agent-gemini.js.map +0 -1
  419. package/dist/src/benchmarks/gaia-claude-p.d.ts +0 -105
  420. package/dist/src/benchmarks/gaia-claude-p.d.ts.map +0 -1
  421. package/dist/src/benchmarks/gaia-claude-p.js +0 -269
  422. package/dist/src/benchmarks/gaia-claude-p.js.map +0 -1
  423. package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts +0 -20
  424. package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts.map +0 -1
  425. package/dist/src/benchmarks/gaia-claude-p.smoke.js +0 -168
  426. package/dist/src/benchmarks/gaia-claude-p.smoke.js.map +0 -1
  427. package/dist/src/benchmarks/gaia-codeagent-runner.py +0 -556
  428. package/dist/src/benchmarks/gaia-codeagent.d.ts +0 -129
  429. package/dist/src/benchmarks/gaia-codeagent.d.ts.map +0 -1
  430. package/dist/src/benchmarks/gaia-codeagent.js +0 -590
  431. package/dist/src/benchmarks/gaia-codeagent.js.map +0 -1
  432. package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts +0 -33
  433. package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts.map +0 -1
  434. package/dist/src/benchmarks/gaia-codeagent.smoke.js +0 -241
  435. package/dist/src/benchmarks/gaia-codeagent.smoke.js.map +0 -1
  436. package/dist/src/benchmarks/gaia-dag-pilot.d.ts +0 -22
  437. package/dist/src/benchmarks/gaia-dag-pilot.d.ts.map +0 -1
  438. package/dist/src/benchmarks/gaia-dag-pilot.js +0 -158
  439. package/dist/src/benchmarks/gaia-dag-pilot.js.map +0 -1
  440. package/dist/src/benchmarks/gaia-dag.d.ts +0 -110
  441. package/dist/src/benchmarks/gaia-dag.d.ts.map +0 -1
  442. package/dist/src/benchmarks/gaia-dag.js +0 -658
  443. package/dist/src/benchmarks/gaia-dag.js.map +0 -1
  444. package/dist/src/benchmarks/gaia-dag.smoke.d.ts +0 -21
  445. package/dist/src/benchmarks/gaia-dag.smoke.d.ts.map +0 -1
  446. package/dist/src/benchmarks/gaia-dag.smoke.js +0 -290
  447. package/dist/src/benchmarks/gaia-dag.smoke.js.map +0 -1
  448. package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts +0 -17
  449. package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts.map +0 -1
  450. package/dist/src/benchmarks/gaia-ensemble-pilot.js +0 -106
  451. package/dist/src/benchmarks/gaia-ensemble-pilot.js.map +0 -1
  452. package/dist/src/benchmarks/gaia-ensemble.d.ts +0 -103
  453. package/dist/src/benchmarks/gaia-ensemble.d.ts.map +0 -1
  454. package/dist/src/benchmarks/gaia-ensemble.js +0 -462
  455. package/dist/src/benchmarks/gaia-ensemble.js.map +0 -1
  456. package/dist/src/benchmarks/gaia-gemini-pilot.d.ts +0 -17
  457. package/dist/src/benchmarks/gaia-gemini-pilot.d.ts.map +0 -1
  458. package/dist/src/benchmarks/gaia-gemini-pilot.js +0 -168
  459. package/dist/src/benchmarks/gaia-gemini-pilot.js.map +0 -1
  460. package/dist/src/benchmarks/gaia-mode-router.d.ts +0 -67
  461. package/dist/src/benchmarks/gaia-mode-router.d.ts.map +0 -1
  462. package/dist/src/benchmarks/gaia-mode-router.js +0 -126
  463. package/dist/src/benchmarks/gaia-mode-router.js.map +0 -1
  464. package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts +0 -15
  465. package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts.map +0 -1
  466. package/dist/src/benchmarks/gaia-mode-router.smoke.js +0 -131
  467. package/dist/src/benchmarks/gaia-mode-router.smoke.js.map +0 -1
  468. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts +0 -24
  469. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts.map +0 -1
  470. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js +0 -276
  471. package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js.map +0 -1
  472. package/dist/src/benchmarks/gaia-tools/image_describe.d.ts +0 -45
  473. package/dist/src/benchmarks/gaia-tools/image_describe.d.ts.map +0 -1
  474. package/dist/src/benchmarks/gaia-tools/image_describe.js +0 -222
  475. package/dist/src/benchmarks/gaia-tools/image_describe.js.map +0 -1
  476. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts +0 -24
  477. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts.map +0 -1
  478. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js +0 -169
  479. package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js.map +0 -1
  480. package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts +0 -27
  481. package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts.map +0 -1
  482. package/dist/src/benchmarks/gaia-tools/pdf_read.js +0 -226
  483. package/dist/src/benchmarks/gaia-tools/pdf_read.js.map +0 -1
  484. package/dist/src/benchmarks/gaia-tools/python_exec.d.ts +0 -31
  485. package/dist/src/benchmarks/gaia-tools/python_exec.d.ts.map +0 -1
  486. package/dist/src/benchmarks/gaia-tools/python_exec.js +0 -94
  487. package/dist/src/benchmarks/gaia-tools/python_exec.js.map +0 -1
  488. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts +0 -13
  489. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts.map +0 -1
  490. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js +0 -98
  491. package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js.map +0 -1
  492. package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts +0 -52
  493. package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts.map +0 -1
  494. package/dist/src/benchmarks/gaia-tools/visit_webpage.js +0 -304
  495. package/dist/src/benchmarks/gaia-tools/visit_webpage.js.map +0 -1
  496. package/dist/src/benchmarks/gaia-tools/web_browse.d.ts +0 -61
  497. package/dist/src/benchmarks/gaia-tools/web_browse.d.ts.map +0 -1
  498. package/dist/src/benchmarks/gaia-tools/web_browse.js +0 -217
  499. package/dist/src/benchmarks/gaia-tools/web_browse.js.map +0 -1
  500. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts +0 -26
  501. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts.map +0 -1
  502. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js +0 -152
  503. package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js.map +0 -1
  504. package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts +0 -24
  505. package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts.map +0 -1
  506. package/dist/src/benchmarks/gaia-tools/web_search.smoke.js +0 -311
  507. package/dist/src/benchmarks/gaia-tools/web_search.smoke.js.map +0 -1
  508. package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts +0 -25
  509. package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts.map +0 -1
  510. package/dist/src/benchmarks/gaia-visit-webpage.smoke.js +0 -304
  511. package/dist/src/benchmarks/gaia-visit-webpage.smoke.js.map +0 -1
  512. package/dist/src/commands/performance-capability.d.ts +0 -30
  513. package/dist/src/commands/performance-capability.d.ts.map +0 -1
  514. package/dist/src/commands/performance-capability.js +0 -349
  515. package/dist/src/commands/performance-capability.js.map +0 -1
@@ -1,126 +0,0 @@
1
- export declare const BUILTIN_CAPABILITY_TASKS: {
2
- readonly version: "1.3";
3
- readonly description: "Text-only agent capability benchmark — verifiable multi-step reasoning tasks scoreable without tool use. Format inspired by GAIA / SWE-bench / GSM8K. The fixture mixes EASY (regression-floor) and HARD (model-gradient) questions so the pass rate has signal across Haiku → Sonnet → Opus. Real GAIA (web browsing, attachments, HF dataset) remains future work.";
4
- readonly answerFormat: "Each question requires the model to reply with the answer wrapped in <answer>...</answer> tags. The harness extracts the tag contents and checks against `expected` per `matchMode`. Per-task `maxTokens` overrides the default cap.";
5
- readonly tasks: readonly [{
6
- readonly id: "math-prime";
7
- readonly category: "easy:reasoning";
8
- readonly prompt: "What is the smallest 3-digit prime number that does not contain the digit 7?";
9
- readonly expected: "101";
10
- readonly matchMode: "exact";
11
- readonly maxTokens: 192;
12
- }, {
13
- readonly id: "logic-syllogism";
14
- readonly category: "easy:reasoning";
15
- readonly prompt: "All routers in tier 1 cost less than $0.001 per call. The Booster router is in tier 1. The Sonnet router costs $0.003 per call. Is the Sonnet router in tier 1? Answer with just \"yes\" or \"no\".";
16
- readonly expected: "no";
17
- readonly matchMode: "exact";
18
- readonly maxTokens: 160;
19
- }, {
20
- readonly id: "regex-match";
21
- readonly category: "easy:code-reasoning";
22
- readonly prompt: "Given the regex /^([a-z]+)-(\\d+)$/ and the input string 'pattern-1779526376', what is the value of capture group 2?";
23
- readonly expected: "1779526376";
24
- readonly matchMode: "exact";
25
- readonly maxTokens: 192;
26
- }, {
27
- readonly id: "gsm8k-trip";
28
- readonly category: "hard:gsm8k-style";
29
- readonly prompt: "A delivery van starts a route with 240 packages. At stop A it drops off 1/4 of its current load and picks up 6 new packages. At stop B it drops off 1/3 of its current load and picks up 4 new packages. At stop C it drops off half of its current load. How many packages does the van have after stop C? Answer with the integer.";
30
- readonly expected: "64";
31
- readonly matchMode: "exact";
32
- readonly maxTokens: 256;
33
- }, {
34
- readonly id: "gsm8k-discount";
35
- readonly category: "hard:gsm8k-style";
36
- readonly prompt: "A store sells 3 widgets and 2 sprockets for $23. It also sells 2 widgets and 4 sprockets for $26. What is the price of one widget? Answer with the integer dollar amount only.";
37
- readonly expected: "5";
38
- readonly matchMode: "exact";
39
- readonly maxTokens: 256;
40
- }, {
41
- readonly id: "code-trace";
42
- readonly category: "hard:code-trace";
43
- readonly prompt: "Consider this JavaScript code:\n```\nconst counts = new Map();\nfor (const c of 'abracadabra') {\n counts.set(c, (counts.get(c) ?? 0) + 1);\n}\nlet maxK = '', maxV = 0;\nfor (const [k, v] of counts) {\n if (v > maxV || (v === maxV && k < maxK)) { maxK = k; maxV = v; }\n}\nconsole.log(`${maxK}:${maxV}`);\n```\nWhat does it print? Answer with just the printed string.";
44
- readonly expected: "a:5";
45
- readonly matchMode: "exact";
46
- readonly maxTokens: 192;
47
- }, {
48
- readonly id: "hard-graph-shortest";
49
- readonly category: "hard:graph-reasoning";
50
- readonly prompt: "A directed graph has these weighted edges: A→B(3), A→C(7), B→C(2), B→D(5), C→D(1), C→E(4), D→E(2). What is the cost of the shortest path from A to E? Answer with the integer.";
51
- readonly expected: "8";
52
- readonly matchMode: "exact";
53
- readonly maxTokens: 192;
54
- }, {
55
- readonly id: "hard-probability";
56
- readonly category: "hard:probability";
57
- readonly prompt: "A bag contains 5 red, 3 blue, and 2 green balls. Two balls are drawn without replacement. What is the probability that both balls are the same color? Express as a fraction in lowest terms in the form a/b (e.g. 3/10).";
58
- readonly expected: "14/45";
59
- readonly matchMode: "exact";
60
- readonly maxTokens: 256;
61
- }, {
62
- readonly id: "expert-marble-inverse";
63
- readonly category: "expert:inverse-arithmetic";
64
- readonly prompt: "A bag of marbles is split as follows: 40% are given to Alice, then 25% of the REMAINING marbles are given to Bob, then half of the remaining marbles (after Alice and Bob) are given to Carol. Carol receives exactly 18 marbles. How many marbles were in the bag originally? Answer with the integer.";
65
- readonly expected: "80";
66
- readonly matchMode: "exact";
67
- readonly maxTokens: 512;
68
- }, {
69
- readonly id: "expert-crt";
70
- readonly category: "expert:number-theory";
71
- readonly prompt: "Find the smallest positive integer n such that all three of these hold simultaneously: n mod 7 = 3, n mod 9 = 4, n mod 11 = 5. Answer with just the integer.";
72
- readonly expected: "346";
73
- readonly matchMode: "exact";
74
- readonly maxTokens: 768;
75
- }, {
76
- readonly id: "expert-bayes";
77
- readonly category: "expert:bayesian";
78
- readonly prompt: "A medical test has 95% sensitivity (true positive rate) and 90% specificity (true negative rate). In the screened population, 1% of people have the disease. A patient tests positive. What is the probability the patient actually has the disease, rounded to the nearest whole percent? Answer with just the integer percentage (no '%' sign).";
79
- readonly expected: "9";
80
- readonly matchMode: "exact";
81
- readonly maxTokens: 512;
82
- }, {
83
- readonly id: "expert-banana";
84
- readonly category: "expert:combinatorics";
85
- readonly prompt: "In how many distinct arrangements of the letters of the word BANANA do no two N's appear next to each other? Answer with the integer.";
86
- readonly expected: "40";
87
- readonly matchMode: "exact";
88
- readonly maxTokens: 512;
89
- }, {
90
- readonly id: "expert-rectangle";
91
- readonly category: "expert:diophantine";
92
- readonly prompt: "A rectangle has positive integer side lengths. Its perimeter (in linear units) is numerically equal to its area (in square units). Two rectangles that are rotations of each other (e.g. 3x6 and 6x3) count as the same rectangle. What is the sum of the areas of all distinct such rectangles? Answer with the integer.";
93
- readonly expected: "34";
94
- readonly matchMode: "exact";
95
- readonly maxTokens: 512;
96
- }, {
97
- readonly id: "expert-dice-reroll";
98
- readonly category: "expert:expected-value";
99
- readonly prompt: "You roll a fair 6-sided die. If the result is a 6, you reroll exactly once and take the new result. Otherwise, you keep the original result. What is the expected value of your final number? Express the answer as a fraction in lowest terms in the form a/b (no spaces, no surrounding text).";
100
- readonly expected: "37/12";
101
- readonly matchMode: "exact";
102
- readonly maxTokens: 512;
103
- }, {
104
- readonly id: "sonnet-killer-knights";
105
- readonly category: "sonnet-killer:logic-puzzle";
106
- readonly prompt: "On an island, knights always tell the truth and knaves always lie. You meet four people named Alice, Bob, Carol, and Dan. They make the following statements: Alice says 'Bob and Carol are different types (one is a knight, the other is a knave).' Bob says 'Alice is a knave.' Carol says 'Dan is a knave.' Dan says 'Carol is a knave.' How many knaves are among the four people? Answer with just the integer.";
107
- readonly expected: "2";
108
- readonly matchMode: "exact";
109
- readonly maxTokens: 768;
110
- }, {
111
- readonly id: "sonnet-killer-hofstadter";
112
- readonly category: "sonnet-killer:recursive-sequence";
113
- readonly prompt: "A sequence is defined on the positive integers by f(1) = 1 and, for every n > 1, f(n) = f(n - f(n-1)) + 1. Compute f(10). Answer with just the integer.";
114
- readonly expected: "4";
115
- readonly matchMode: "exact";
116
- readonly maxTokens: 768;
117
- }, {
118
- readonly id: "sonnet-killer-modexp";
119
- readonly category: "sonnet-killer:number-theory";
120
- readonly prompt: "What are the last two digits of 7 raised to the 2026th power? Answer with exactly the two-digit number (e.g. '07' if it's seven, '49' if it's forty-nine).";
121
- readonly expected: "49";
122
- readonly matchMode: "exact";
123
- readonly maxTokens: 512;
124
- }];
125
- };
126
- //# sourceMappingURL=capability-tasks.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"capability-tasks.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/capability-tasks.ts"],"names":[],"mappings":"AAIA,eAAO,MAAM,wBAAwB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8I3B,CAAC"}
@@ -1,147 +0,0 @@
1
- // Auto-generated from src/benchmarks/capability-tasks.json — keep in sync.
2
- // This module exists so the fixture is bundled into dist/ by tsc (JSON files
3
- // are not copied by tsc, and the published CLI ships only dist/).
4
- export const BUILTIN_CAPABILITY_TASKS = {
5
- "version": "1.3",
6
- "description": "Text-only agent capability benchmark — verifiable multi-step reasoning tasks scoreable without tool use. Format inspired by GAIA / SWE-bench / GSM8K. The fixture mixes EASY (regression-floor) and HARD (model-gradient) questions so the pass rate has signal across Haiku → Sonnet → Opus. Real GAIA (web browsing, attachments, HF dataset) remains future work.",
7
- "answerFormat": "Each question requires the model to reply with the answer wrapped in <answer>...</answer> tags. The harness extracts the tag contents and checks against `expected` per `matchMode`. Per-task `maxTokens` overrides the default cap.",
8
- "tasks": [
9
- {
10
- "id": "math-prime",
11
- "category": "easy:reasoning",
12
- "prompt": "What is the smallest 3-digit prime number that does not contain the digit 7?",
13
- "expected": "101",
14
- "matchMode": "exact",
15
- "maxTokens": 192
16
- },
17
- {
18
- "id": "logic-syllogism",
19
- "category": "easy:reasoning",
20
- "prompt": "All routers in tier 1 cost less than $0.001 per call. The Booster router is in tier 1. The Sonnet router costs $0.003 per call. Is the Sonnet router in tier 1? Answer with just \"yes\" or \"no\".",
21
- "expected": "no",
22
- "matchMode": "exact",
23
- "maxTokens": 160
24
- },
25
- {
26
- "id": "regex-match",
27
- "category": "easy:code-reasoning",
28
- "prompt": "Given the regex /^([a-z]+)-(\\d+)$/ and the input string 'pattern-1779526376', what is the value of capture group 2?",
29
- "expected": "1779526376",
30
- "matchMode": "exact",
31
- "maxTokens": 192
32
- },
33
- {
34
- "id": "gsm8k-trip",
35
- "category": "hard:gsm8k-style",
36
- "prompt": "A delivery van starts a route with 240 packages. At stop A it drops off 1/4 of its current load and picks up 6 new packages. At stop B it drops off 1/3 of its current load and picks up 4 new packages. At stop C it drops off half of its current load. How many packages does the van have after stop C? Answer with the integer.",
37
- "expected": "64",
38
- "matchMode": "exact",
39
- "maxTokens": 256
40
- },
41
- {
42
- "id": "gsm8k-discount",
43
- "category": "hard:gsm8k-style",
44
- "prompt": "A store sells 3 widgets and 2 sprockets for $23. It also sells 2 widgets and 4 sprockets for $26. What is the price of one widget? Answer with the integer dollar amount only.",
45
- "expected": "5",
46
- "matchMode": "exact",
47
- "maxTokens": 256
48
- },
49
- {
50
- "id": "code-trace",
51
- "category": "hard:code-trace",
52
- "prompt": "Consider this JavaScript code:\n```\nconst counts = new Map();\nfor (const c of 'abracadabra') {\n counts.set(c, (counts.get(c) ?? 0) + 1);\n}\nlet maxK = '', maxV = 0;\nfor (const [k, v] of counts) {\n if (v > maxV || (v === maxV && k < maxK)) { maxK = k; maxV = v; }\n}\nconsole.log(`${maxK}:${maxV}`);\n```\nWhat does it print? Answer with just the printed string.",
53
- "expected": "a:5",
54
- "matchMode": "exact",
55
- "maxTokens": 192
56
- },
57
- {
58
- "id": "hard-graph-shortest",
59
- "category": "hard:graph-reasoning",
60
- "prompt": "A directed graph has these weighted edges: A→B(3), A→C(7), B→C(2), B→D(5), C→D(1), C→E(4), D→E(2). What is the cost of the shortest path from A to E? Answer with the integer.",
61
- "expected": "8",
62
- "matchMode": "exact",
63
- "maxTokens": 192
64
- },
65
- {
66
- "id": "hard-probability",
67
- "category": "hard:probability",
68
- "prompt": "A bag contains 5 red, 3 blue, and 2 green balls. Two balls are drawn without replacement. What is the probability that both balls are the same color? Express as a fraction in lowest terms in the form a/b (e.g. 3/10).",
69
- "expected": "14/45",
70
- "matchMode": "exact",
71
- "maxTokens": 256
72
- },
73
- {
74
- "id": "expert-marble-inverse",
75
- "category": "expert:inverse-arithmetic",
76
- "prompt": "A bag of marbles is split as follows: 40% are given to Alice, then 25% of the REMAINING marbles are given to Bob, then half of the remaining marbles (after Alice and Bob) are given to Carol. Carol receives exactly 18 marbles. How many marbles were in the bag originally? Answer with the integer.",
77
- "expected": "80",
78
- "matchMode": "exact",
79
- "maxTokens": 512
80
- },
81
- {
82
- "id": "expert-crt",
83
- "category": "expert:number-theory",
84
- "prompt": "Find the smallest positive integer n such that all three of these hold simultaneously: n mod 7 = 3, n mod 9 = 4, n mod 11 = 5. Answer with just the integer.",
85
- "expected": "346",
86
- "matchMode": "exact",
87
- "maxTokens": 768
88
- },
89
- {
90
- "id": "expert-bayes",
91
- "category": "expert:bayesian",
92
- "prompt": "A medical test has 95% sensitivity (true positive rate) and 90% specificity (true negative rate). In the screened population, 1% of people have the disease. A patient tests positive. What is the probability the patient actually has the disease, rounded to the nearest whole percent? Answer with just the integer percentage (no '%' sign).",
93
- "expected": "9",
94
- "matchMode": "exact",
95
- "maxTokens": 512
96
- },
97
- {
98
- "id": "expert-banana",
99
- "category": "expert:combinatorics",
100
- "prompt": "In how many distinct arrangements of the letters of the word BANANA do no two N's appear next to each other? Answer with the integer.",
101
- "expected": "40",
102
- "matchMode": "exact",
103
- "maxTokens": 512
104
- },
105
- {
106
- "id": "expert-rectangle",
107
- "category": "expert:diophantine",
108
- "prompt": "A rectangle has positive integer side lengths. Its perimeter (in linear units) is numerically equal to its area (in square units). Two rectangles that are rotations of each other (e.g. 3x6 and 6x3) count as the same rectangle. What is the sum of the areas of all distinct such rectangles? Answer with the integer.",
109
- "expected": "34",
110
- "matchMode": "exact",
111
- "maxTokens": 512
112
- },
113
- {
114
- "id": "expert-dice-reroll",
115
- "category": "expert:expected-value",
116
- "prompt": "You roll a fair 6-sided die. If the result is a 6, you reroll exactly once and take the new result. Otherwise, you keep the original result. What is the expected value of your final number? Express the answer as a fraction in lowest terms in the form a/b (no spaces, no surrounding text).",
117
- "expected": "37/12",
118
- "matchMode": "exact",
119
- "maxTokens": 512
120
- },
121
- {
122
- "id": "sonnet-killer-knights",
123
- "category": "sonnet-killer:logic-puzzle",
124
- "prompt": "On an island, knights always tell the truth and knaves always lie. You meet four people named Alice, Bob, Carol, and Dan. They make the following statements: Alice says 'Bob and Carol are different types (one is a knight, the other is a knave).' Bob says 'Alice is a knave.' Carol says 'Dan is a knave.' Dan says 'Carol is a knave.' How many knaves are among the four people? Answer with just the integer.",
125
- "expected": "2",
126
- "matchMode": "exact",
127
- "maxTokens": 768
128
- },
129
- {
130
- "id": "sonnet-killer-hofstadter",
131
- "category": "sonnet-killer:recursive-sequence",
132
- "prompt": "A sequence is defined on the positive integers by f(1) = 1 and, for every n > 1, f(n) = f(n - f(n-1)) + 1. Compute f(10). Answer with just the integer.",
133
- "expected": "4",
134
- "matchMode": "exact",
135
- "maxTokens": 768
136
- },
137
- {
138
- "id": "sonnet-killer-modexp",
139
- "category": "sonnet-killer:number-theory",
140
- "prompt": "What are the last two digits of 7 raised to the 2026th power? Answer with exactly the two-digit number (e.g. '07' if it's seven, '49' if it's forty-nine).",
141
- "expected": "49",
142
- "matchMode": "exact",
143
- "maxTokens": 512
144
- }
145
- ]
146
- };
147
- //# sourceMappingURL=capability-tasks.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"capability-tasks.js","sourceRoot":"","sources":["../../../src/benchmarks/capability-tasks.ts"],"names":[],"mappings":"AAAA,2EAA2E;AAC3E,6EAA6E;AAC7E,kEAAkE;AAElE,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,SAAS,EAAE,KAAK;IAChB,aAAa,EAAE,sWAAsW;IACrX,cAAc,EAAE,sOAAsO;IACtP,OAAO,EAAE;QACP;YACE,IAAI,EAAE,YAAY;YAClB,UAAU,EAAE,gBAAgB;YAC5B,QAAQ,EAAE,8EAA8E;YACxF,UAAU,EAAE,KAAK;YACjB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,iBAAiB;YACvB,UAAU,EAAE,gBAAgB;YAC5B,QAAQ,EAAE,qMAAqM;YAC/M,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,aAAa;YACnB,UAAU,EAAE,qBAAqB;YACjC,QAAQ,EAAE,sHAAsH;YAChI,UAAU,EAAE,YAAY;YACxB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,YAAY;YAClB,UAAU,EAAE,kBAAkB;YAC9B,QAAQ,EAAE,sUAAsU;YAChV,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,gBAAgB;YACtB,UAAU,EAAE,kBAAkB;YAC9B,QAAQ,EAAE,gLAAgL;YAC1L,UAAU,EAAE,GAAG;YACf,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,YAAY;YAClB,UAAU,EAAE,iBAAiB;YAC7B,QAAQ,EAAE,mXAAmX;YAC7X,UAAU,EAAE,KAAK;YACjB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,qBAAqB;YAC3B,UAAU,EAAE,sBAAsB;YAClC,QAAQ,EAAE,gLAAgL;YAC1L,UAAU,EAAE,GAAG;YACf,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,kBAAkB;YACxB,UAAU,EAAE,kBAAkB;YAC9B,QAAQ,EAAE,0NAA0N;YACpO,UAAU,EAAE,OAAO;YACnB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,uBAAuB;YAC7B,UAAU,EAAE,2BAA2B;YACvC,QAAQ,EAAE,ySAAyS;YACnT,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,YAAY;YAClB,UAAU,EAAE,sBAAsB;YAClC,QAAQ,EAAE,8JAA8J;YACxK,UAAU,EAAE,KAAK;YACjB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,cAAc;YACpB,UAAU,EAAE,iBAAiB;YAC7B,QAAQ,EAAE,mVAAmV;YAC7V,UAAU,EAAE,GAAG;YACf,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,eAAe;YACrB,UAAU,EAAE,sBAAsB;YAClC,QAAQ,EAAE,uIAAuI;YACjJ,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,kBAAkB;YACxB,UAAU,EAAE,oBAAoB;YAChC,QAAQ,EAAE,2TAA2T;YACrU,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,oBAAoB;YAC1B,UAAU,EAAE,uBAAuB;YACnC,QAAQ,EAAE,kSAAkS;YAC5S,UAAU,EAAE,OAAO;YACnB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,uBAAuB;YAC7B,UAAU,EAAE,4BAA4B;YACxC,QAAQ,EAAE,uZAAuZ;YACja,UAAU,EAAE,GAAG;YACf,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,0BAA0B;YAChC,UAAU,EAAE,kCAAkC;YAC9C,QAAQ,EAAE,yJAAyJ;YACnK,UAAU,EAAE,GAAG;YACf,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;QACD;YACE,IAAI,EAAE,sBAAsB;YAC5B,UAAU,EAAE,6BAA6B;YACzC,QAAQ,EAAE,4JAA4J;YACtK,UAAU,EAAE,IAAI;YAChB,WAAW,EAAE,OAAO;YACpB,WAAW,EAAE,GAAG;SACjB;KACF;CACO,CAAC"}
@@ -1,53 +0,0 @@
1
- /**
2
- * GAIA Agent Contrastive Wrapper — iter 49.5
3
- *
4
- * Thin wrapper around runGaiaAgent that adds three ruflo intelligence hooks:
5
- * 1. PRE: memory_search_unified for relevant patterns
6
- * 2. DURING: trajectory recording (start/step/end) via memory store
7
- * 3. POST: memory_store of question+answer+verdict for future recall
8
- *
9
- * Purpose: measure whether ruflo's AgentDB intelligence (5,930 entries,
10
- * 26,490 patterns) provides a measurable lift on GAIA L1 when injected
11
- * as system context before the agent loop.
12
- *
13
- * Design constraints:
14
- * - Does NOT change the underlying agent loop logic
15
- * - Does NOT enable ADR-135 tracks (critic, decomposition, voting, MoE, KG)
16
- * - Pure addition: memory before, trajectory during, store after
17
- * - Graceful fallback: any hook failure is logged but does NOT fail the question
18
- *
19
- * Refs: ADR-133, iter 49.5, #2156
20
- */
21
- import type { GaiaQuestion } from './gaia-loader.js';
22
- import { type GaiaAgentOptions, type GaiaAgentResult } from './gaia-agent.js';
23
- export interface ContrastiveStats {
24
- memorySearchCalls: number;
25
- memorySearchHits: number;
26
- memoryStoreCount: number;
27
- trajectoryCount: number;
28
- totalPatternsInjected: number;
29
- }
30
- export interface ContrastiveAgentResult extends GaiaAgentResult {
31
- /** Number of memory patterns injected as context for this question. */
32
- patternsInjected?: number;
33
- /** Whether a trajectory was recorded for this question. */
34
- trajectoryRecorded?: boolean;
35
- /** Whether the question+answer was stored in memory after completion. */
36
- memoryStored?: boolean;
37
- }
38
- /**
39
- * Wrap runGaiaAgent with ruflo intelligence hooks.
40
- *
41
- * Hook points:
42
- * 1. memory_search before the agent loop (context injection)
43
- * 2. trajectory-start/end records via memory store
44
- * 3. memory_store after the final answer
45
- *
46
- * The underlying agent loop is called unchanged.
47
- */
48
- export declare function runGaiaAgentContrastive(question: GaiaQuestion, options: GaiaAgentOptions, stats: ContrastiveStats): Promise<ContrastiveAgentResult>;
49
- /**
50
- * Create a fresh ContrastiveStats counter object.
51
- */
52
- export declare function createContrastiveStats(): ContrastiveStats;
53
- //# sourceMappingURL=gaia-agent-contrastive.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"gaia-agent-contrastive.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-agent-contrastive.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AACrD,OAAO,EAAgB,KAAK,gBAAgB,EAAE,KAAK,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAM5F,MAAM,WAAW,gBAAgB;IAC/B,iBAAiB,EAAE,MAAM,CAAC;IAC1B,gBAAgB,EAAE,MAAM,CAAC;IACzB,gBAAgB,EAAE,MAAM,CAAC;IACzB,eAAe,EAAE,MAAM,CAAC;IACxB,qBAAqB,EAAE,MAAM,CAAC;CAC/B;AAED,MAAM,WAAW,sBAAuB,SAAQ,eAAe;IAC7D,uEAAuE;IACvE,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,2DAA2D;IAC3D,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B,yEAAyE;IACzE,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AA2FD;;;;;;;;;GASG;AACH,wBAAsB,uBAAuB,CAC3C,QAAQ,EAAE,YAAY,EACtB,OAAO,EAAE,gBAAgB,EACzB,KAAK,EAAE,gBAAgB,GACtB,OAAO,CAAC,sBAAsB,CAAC,CA+DjC;AAED;;GAEG;AACH,wBAAgB,sBAAsB,IAAI,gBAAgB,CAQzD"}
@@ -1,177 +0,0 @@
1
- /**
2
- * GAIA Agent Contrastive Wrapper — iter 49.5
3
- *
4
- * Thin wrapper around runGaiaAgent that adds three ruflo intelligence hooks:
5
- * 1. PRE: memory_search_unified for relevant patterns
6
- * 2. DURING: trajectory recording (start/step/end) via memory store
7
- * 3. POST: memory_store of question+answer+verdict for future recall
8
- *
9
- * Purpose: measure whether ruflo's AgentDB intelligence (5,930 entries,
10
- * 26,490 patterns) provides a measurable lift on GAIA L1 when injected
11
- * as system context before the agent loop.
12
- *
13
- * Design constraints:
14
- * - Does NOT change the underlying agent loop logic
15
- * - Does NOT enable ADR-135 tracks (critic, decomposition, voting, MoE, KG)
16
- * - Pure addition: memory before, trajectory during, store after
17
- * - Graceful fallback: any hook failure is logged but does NOT fail the question
18
- *
19
- * Refs: ADR-133, iter 49.5, #2156
20
- */
21
- import { execSync } from 'node:child_process';
22
- import { runGaiaAgent } from './gaia-agent.js';
23
- // ---------------------------------------------------------------------------
24
- // Ruflo CLI helpers (best-effort — never throw)
25
- // ---------------------------------------------------------------------------
26
- const CLI_TIMEOUT_MS = 10_000;
27
- const GAIA_NAMESPACE = 'gaia-l1-questions';
28
- /**
29
- * Run memory search via CLI and return up to `limit` result summaries.
30
- * Returns empty array on any error (graceful fallback).
31
- */
32
- function memorySearch(query, limit) {
33
- try {
34
- const raw = execSync(`npx @claude-flow/cli@latest memory search --query ${JSON.stringify(query)} --limit ${limit}`, { encoding: 'utf-8', timeout: CLI_TIMEOUT_MS, stdio: ['pipe', 'pipe', 'pipe'] });
35
- // Parse table rows: extract Preview column (5th column in pipe-separated table)
36
- const rows = raw
37
- .split('\n')
38
- .filter((line) => line.startsWith('|') && !line.includes('Key') && !line.includes('---'))
39
- .map((line) => {
40
- const cols = line.split('|').map((c) => c.trim()).filter(Boolean);
41
- // cols: [Key, Score, Namespace, Preview]
42
- return cols[3] ?? '';
43
- })
44
- .filter((preview) => preview.length > 0 && !preview.includes('Preview'));
45
- return rows.slice(0, limit);
46
- }
47
- catch {
48
- return [];
49
- }
50
- }
51
- /**
52
- * Store a key/value in memory under the gaia namespace.
53
- * Best-effort — swallows errors.
54
- */
55
- function memoryStore(key, value) {
56
- try {
57
- const valueStr = JSON.stringify(value);
58
- execSync(`npx @claude-flow/cli@latest memory store --key ${JSON.stringify(key)} --value ${JSON.stringify(valueStr)} --namespace ${GAIA_NAMESPACE}`, { encoding: 'utf-8', timeout: CLI_TIMEOUT_MS, stdio: ['pipe', 'pipe', 'pipe'] });
59
- return true;
60
- }
61
- catch {
62
- return false;
63
- }
64
- }
65
- /**
66
- * Record a trajectory event (start/step/end) via memory store.
67
- * This is a lightweight proxy — the MCP trajectory tools are not available
68
- * from inside the benchmark harness. We persist the trajectory as memory
69
- * entries so they can be recalled later.
70
- */
71
- function trajectoryRecord(type, data) {
72
- try {
73
- const key = `traj-gaia-l1-${type}-${Date.now()}`;
74
- execSync(`npx @claude-flow/cli@latest memory store --key ${JSON.stringify(key)} --value ${JSON.stringify(JSON.stringify(data))} --namespace trajectories`, { encoding: 'utf-8', timeout: CLI_TIMEOUT_MS, stdio: ['pipe', 'pipe', 'pipe'] });
75
- }
76
- catch {
77
- // Best-effort — trajectory recording never blocks progress
78
- }
79
- }
80
- // ---------------------------------------------------------------------------
81
- // System prompt injection
82
- // ---------------------------------------------------------------------------
83
- /**
84
- * Build a context prefix to prepend to the system prompt when patterns
85
- * are found in memory. Returns empty string when patterns is empty.
86
- */
87
- function buildMemoryContextPrefix(patterns) {
88
- if (patterns.length === 0)
89
- return '';
90
- return ('Relevant patterns from prior ruflo work:\n' +
91
- patterns.map((p) => `- ${p.slice(0, 120)}`).join('\n') +
92
- '\n\n');
93
- }
94
- // ---------------------------------------------------------------------------
95
- // Contrastive wrapper
96
- // ---------------------------------------------------------------------------
97
- /**
98
- * Wrap runGaiaAgent with ruflo intelligence hooks.
99
- *
100
- * Hook points:
101
- * 1. memory_search before the agent loop (context injection)
102
- * 2. trajectory-start/end records via memory store
103
- * 3. memory_store after the final answer
104
- *
105
- * The underlying agent loop is called unchanged.
106
- */
107
- export async function runGaiaAgentContrastive(question, options, stats) {
108
- const questionId = question.task_id;
109
- // ---- 1. PRE: memory_search for relevant patterns ----------------------
110
- stats.memorySearchCalls++;
111
- const patterns = memorySearch(question.question, 3);
112
- const patternsInjected = patterns.length;
113
- if (patterns.length > 0) {
114
- stats.memorySearchHits++;
115
- stats.totalPatternsInjected += patterns.length;
116
- }
117
- // ---- 2. TRAJECTORY: start recording -----------------------------------
118
- trajectoryRecord('start', {
119
- task: 'gaia-l1',
120
- question_id: questionId,
121
- question_snippet: question.question.slice(0, 80),
122
- startedAt: new Date().toISOString(),
123
- patternsInjected,
124
- });
125
- // ---- Build patched options with memory context injected ---------------
126
- // We inject context by wrapping the question text (system prompt cannot be
127
- // overridden via GaiaAgentOptions today; instead we prepend context to
128
- // the question text as a user-visible hint). This is intentionally minimal.
129
- const contextPrefix = buildMemoryContextPrefix(patterns);
130
- const patchedQuestion = contextPrefix
131
- ? { ...question, question: contextPrefix + question.question }
132
- : question;
133
- // ---- Run the actual agent loop (unchanged) ----------------------------
134
- const result = await runGaiaAgent(patchedQuestion, options);
135
- // ---- 3. POST: store question+answer+verdict ---------------------------
136
- const questionKey = `gaia-l1-q${questionId}-${Date.now()}`;
137
- const stored = memoryStore(questionKey, {
138
- question: question.question,
139
- answer: result.finalAnswer,
140
- questionId,
141
- model: options.model ?? 'claude-sonnet-4-6',
142
- turns: result.turns,
143
- timedOut: result.timedOut ?? false,
144
- storedAt: new Date().toISOString(),
145
- });
146
- if (stored)
147
- stats.memoryStoreCount++;
148
- // ---- TRAJECTORY: end recording ----------------------------------------
149
- trajectoryRecord('end', {
150
- task: 'gaia-l1',
151
- question_id: questionId,
152
- answer: result.finalAnswer,
153
- turns: result.turns,
154
- timedOut: result.timedOut ?? false,
155
- endedAt: new Date().toISOString(),
156
- });
157
- stats.trajectoryCount++;
158
- return {
159
- ...result,
160
- patternsInjected,
161
- trajectoryRecorded: true,
162
- memoryStored: stored,
163
- };
164
- }
165
- /**
166
- * Create a fresh ContrastiveStats counter object.
167
- */
168
- export function createContrastiveStats() {
169
- return {
170
- memorySearchCalls: 0,
171
- memorySearchHits: 0,
172
- memoryStoreCount: 0,
173
- trajectoryCount: 0,
174
- totalPatternsInjected: 0,
175
- };
176
- }
177
- //# sourceMappingURL=gaia-agent-contrastive.js.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"gaia-agent-contrastive.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-agent-contrastive.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAE9C,OAAO,EAAE,YAAY,EAA+C,MAAM,iBAAiB,CAAC;AAuB5F,8EAA8E;AAC9E,gDAAgD;AAChD,8EAA8E;AAE9E,MAAM,cAAc,GAAG,MAAM,CAAC;AAC9B,MAAM,cAAc,GAAG,mBAAmB,CAAC;AAE3C;;;GAGG;AACH,SAAS,YAAY,CAAC,KAAa,EAAE,KAAa;IAChD,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,QAAQ,CAClB,qDAAqD,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,YAAY,KAAK,EAAE,EAC7F,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAChF,CAAC;QACF,gFAAgF;QAChF,MAAM,IAAI,GAAG,GAAG;aACb,KAAK,CAAC,IAAI,CAAC;aACX,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,UAAU,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;aACxF,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACZ,MAAM,IAAI,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;YAClE,yCAAyC;YACzC,OAAO,IAAI,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACvB,CAAC,CAAC;aACD,MAAM,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,CAAC;QAC3E,OAAO,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC;IAC9B,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,SAAS,WAAW,CAAC,GAAW,EAAE,KAAa;IAC7C,IAAI,CAAC;QACH,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC;QACvC,QAAQ,CACN,kDAAkD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,gBAAgB,cAAc,EAAE,EACzI,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAChF,CAAC;QACF,OAAO,IAAI,CAAC;IACd,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,KAAK,CAAC;IACf,CAAC;AACH,CAAC;AAED;;;;;GAKG;AACH,SAAS,gBAAgB,CAAC,IAAqB,EAAE,IAAY;IAC3D,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,gBAAgB,IAAI,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;QACjD,QAAQ,CACN,kDAAkD,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,YAAY,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,CAAC,2BAA2B,EAChJ,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,cAAc,EAAE,KAAK,EAAE,CAAC,MAAM,EAAE,MAAM,EAAE,MAAM,CAAC,EAAE,CAChF,CAAC;IACJ,CAAC;IAAC,MAAM,CAAC;QACP,2DAA2D;IAC7D,CAAC;AACH,CAAC;AAED,8EAA8E;AAC9E,0BAA0B;AAC1B,8EAA8E;AAE9E;;;GAGG;AACH,SAAS,wBAAwB,CAAC,QAAkB;IAClD,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IACrC,OAAO,CACL,4CAA4C;QAC5C,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC;QACtD,MAAM,CACP,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,sBAAsB;AACtB,8EAA8E;AAE9E;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,uBAAuB,CAC3C,QAAsB,EACtB,OAAyB,EACzB,KAAuB;IAEvB,MAAM,UAAU,GAAG,QAAQ,CAAC,OAAO,CAAC;IAEpC,0EAA0E;IAC1E,KAAK,CAAC,iBAAiB,EAAE,CAAC;IAC1B,MAAM,QAAQ,GAAG,YAAY,CAAC,QAAQ,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IACpD,MAAM,gBAAgB,GAAG,QAAQ,CAAC,MAAM,CAAC;IACzC,IAAI,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACxB,KAAK,CAAC,gBAAgB,EAAE,CAAC;QACzB,KAAK,CAAC,qBAAqB,IAAI,QAAQ,CAAC,MAAM,CAAC;IACjD,CAAC;IAED,0EAA0E;IAC1E,gBAAgB,CAAC,OAAO,EAAE;QACxB,IAAI,EAAE,SAAS;QACf,WAAW,EAAE,UAAU;QACvB,gBAAgB,EAAE,QAAQ,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC;QAChD,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;QACnC,gBAAgB;KACjB,CAAC,CAAC;IAEH,0EAA0E;IAC1E,2EAA2E;IAC3E,uEAAuE;IACvE,4EAA4E;IAC5E,MAAM,aAAa,GAAG,wBAAwB,CAAC,QAAQ,CAAC,CAAC;IACzD,MAAM,eAAe,GAAiB,aAAa;QACjD,CAAC,CAAC,EAAE,GAAG,QAAQ,EAAE,QAAQ,EAAE,aAAa,GAAG,QAAQ,CAAC,QAAQ,EAAE;QAC9D,CAAC,CAAC,QAAQ,CAAC;IAEb,0EAA0E;IAC1E,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,eAAe,EAAE,OAAO,CAAC,CAAC;IAE5D,0EAA0E;IAC1E,MAAM,WAAW,GAAG,YAAY,UAAU,IAAI,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC;IAC3D,MAAM,MAAM,GAAG,WAAW,CAAC,WAAW,EAAE;QACtC,QAAQ,EAAE,QAAQ,CAAC,QAAQ;QAC3B,MAAM,EAAE,MAAM,CAAC,WAAW;QAC1B,UAAU;QACV,KAAK,EAAE,OAAO,CAAC,KAAK,IAAI,mBAAmB;QAC3C,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,QAAQ,EAAE,MAAM,CAAC,QAAQ,IAAI,KAAK;QAClC,QAAQ,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACnC,CAAC,CAAC;IACH,IAAI,MAAM;QAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;IAErC,0EAA0E;IAC1E,gBAAgB,CAAC,KAAK,EAAE;QACtB,IAAI,EAAE,SAAS;QACf,WAAW,EAAE,UAAU;QACvB,MAAM,EAAE,MAAM,CAAC,WAAW;QAC1B,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,QAAQ,EAAE,MAAM,CAAC,QAAQ,IAAI,KAAK;QAClC,OAAO,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KAClC,CAAC,CAAC;IACH,KAAK,CAAC,eAAe,EAAE,CAAC;IAExB,OAAO;QACL,GAAG,MAAM;QACT,gBAAgB;QAChB,kBAAkB,EAAE,IAAI;QACxB,YAAY,EAAE,MAAM;KACrB,CAAC;AACJ,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,sBAAsB;IACpC,OAAO;QACL,iBAAiB,EAAE,CAAC;QACpB,gBAAgB,EAAE,CAAC;QACnB,gBAAgB,EAAE,CAAC;QACnB,eAAe,EAAE,CAAC;QAClB,qBAAqB,EAAE,CAAC;KACzB,CAAC;AACJ,CAAC"}
@@ -1,51 +0,0 @@
1
- /**
2
- * GAIA Agent — Gemini 2.5 Pro Thinking adapter (Phase 2 pilot)
3
- *
4
- * Adapts the existing GAIA tool catalogue to Google's generateContent API,
5
- * translating Anthropic tool_use ↔ Gemini functionCall/functionResponse.
6
- *
7
- * Loop algorithm mirrors gaia-agent.ts:
8
- * 1. Build initial contents array with system instruction + first user turn.
9
- * 2. Call Gemini generateContent with functionDeclarations.
10
- * 3. On functionCall parts: execute tools, append functionResponse, repeat.
11
- * 4. On text-only response: scan for FINAL_ANSWER pattern.
12
- * 5. On maxTurns: return timedOut result.
13
- *
14
- * Gemini 2.5 Pro pricing (2026-05-28, per million tokens):
15
- * Input: $1.25 (≤200k tokens), $2.50 (>200k)
16
- * Output: $10.00 (≤200k tokens), $15.00 (>200k)
17
- * Thinking tokens are billed as output tokens.
18
- *
19
- * Cost gate for pilot: ≤$0.12 per question average.
20
- * Expected per-question cost at ~10k input + ~2k output: ~$0.032 — well within gate.
21
- *
22
- * Refs: ADR-133, ADR-135, #2156 Phase 2
23
- */
24
- import { GaiaQuestion } from './gaia-loader.js';
25
- import { GaiaToolCatalogue } from './gaia-tools/index.js';
26
- export declare const DEFAULT_GEMINI_MODEL = "gemini-2.5-pro";
27
- export interface GeminiAgentResult {
28
- questionId: string;
29
- finalAnswer: string | null;
30
- turns: number;
31
- toolCallsByName: Record<string, number>;
32
- totalInputTokens: number;
33
- totalOutputTokens: number;
34
- totalThinkingTokens: number;
35
- wallMs: number;
36
- estimatedCostUsd: number;
37
- timedOut?: boolean;
38
- error?: string;
39
- }
40
- export interface GeminiAgentOptions {
41
- model?: string;
42
- maxTurns?: number;
43
- maxTokensPerTurn?: number;
44
- perTurnTimeoutMs?: number;
45
- apiKey?: string;
46
- catalogue?: GaiaToolCatalogue;
47
- }
48
- export declare function resolveGeminiApiKey(apiKey?: string): string;
49
- export declare function runGeminiAgent(question: GaiaQuestion, options?: GeminiAgentOptions): Promise<GeminiAgentResult>;
50
- export declare function isGeminiAnswerCorrect(modelAnswer: string, expected: string): boolean;
51
- //# sourceMappingURL=gaia-agent-gemini.d.ts.map
@@ -1 +0,0 @@
1
- {"version":3,"file":"gaia-agent-gemini.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-agent-gemini.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAKH,OAAO,EACL,YAAY,EACb,MAAM,kBAAkB,CAAC;AAC1B,OAAO,EAEL,iBAAiB,EAElB,MAAM,uBAAuB,CAAC;AAO/B,eAAO,MAAM,oBAAoB,mBAAmB,CAAC;AAerD,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,eAAe,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACxC,gBAAgB,EAAE,MAAM,CAAC;IACzB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,mBAAmB,EAAE,MAAM,CAAC;IAC5B,MAAM,EAAE,MAAM,CAAC;IACf,gBAAgB,EAAE,MAAM,CAAC;IACzB,QAAQ,CAAC,EAAE,OAAO,CAAC;IACnB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,kBAAkB;IACjC,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,gBAAgB,CAAC,EAAE,MAAM,CAAC;IAC1B,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,SAAS,CAAC,EAAE,iBAAiB,CAAC;CAC/B;AAMD,wBAAgB,mBAAmB,CAAC,MAAM,CAAC,EAAE,MAAM,GAAG,MAAM,CAmB3D;AAiND,wBAAsB,cAAc,CAClC,QAAQ,EAAE,YAAY,EACtB,OAAO,GAAE,kBAAuB,GAC/B,OAAO,CAAC,iBAAiB,CAAC,CA0J5B;AAaD,wBAAgB,qBAAqB,CAAC,WAAW,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,GAAG,OAAO,CAgBpF"}