@claude-flow/cli 3.10.36 → 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.
- package/.claude/agents/analysis/analyze-code-quality.md +178 -178
- package/.claude/agents/analysis/code-analyzer.md +209 -209
- package/.claude/agents/analysis/code-review/analyze-code-quality.md +178 -178
- package/.claude/agents/architecture/arch-system-design.md +156 -156
- package/.claude/agents/architecture/system-design/arch-system-design.md +154 -154
- package/.claude/agents/browser/browser-agent.yaml +182 -182
- package/.claude/agents/consensus/byzantine-coordinator.md +62 -62
- package/.claude/agents/consensus/crdt-synchronizer.md +996 -996
- package/.claude/agents/consensus/gossip-coordinator.md +62 -62
- package/.claude/agents/consensus/performance-benchmarker.md +850 -850
- package/.claude/agents/consensus/quorum-manager.md +822 -822
- package/.claude/agents/consensus/raft-manager.md +62 -62
- package/.claude/agents/consensus/security-manager.md +621 -621
- package/.claude/agents/core/planner.md +374 -374
- package/.claude/agents/custom/test-long-runner.md +44 -44
- package/.claude/agents/data/data-ml-model.md +444 -444
- package/.claude/agents/data/ml/data-ml-model.md +192 -192
- package/.claude/agents/development/backend/dev-backend-api.md +141 -141
- package/.claude/agents/development/dev-backend-api.md +344 -344
- package/.claude/agents/devops/ci-cd/ops-cicd-github.md +163 -163
- package/.claude/agents/devops/ops-cicd-github.md +164 -164
- package/.claude/agents/documentation/api-docs/docs-api-openapi.md +173 -173
- package/.claude/agents/documentation/docs-api-openapi.md +354 -354
- package/.claude/agents/flow-nexus/app-store.md +87 -87
- package/.claude/agents/flow-nexus/authentication.md +68 -68
- package/.claude/agents/flow-nexus/challenges.md +80 -80
- package/.claude/agents/flow-nexus/neural-network.md +87 -87
- package/.claude/agents/flow-nexus/payments.md +82 -82
- package/.claude/agents/flow-nexus/sandbox.md +75 -75
- package/.claude/agents/flow-nexus/swarm.md +75 -75
- package/.claude/agents/flow-nexus/user-tools.md +95 -95
- package/.claude/agents/flow-nexus/workflow.md +83 -83
- package/.claude/agents/github/code-review-swarm.md +377 -377
- package/.claude/agents/github/github-modes.md +172 -172
- package/.claude/agents/github/issue-tracker.md +575 -575
- package/.claude/agents/github/multi-repo-swarm.md +552 -552
- package/.claude/agents/github/pr-manager.md +437 -437
- package/.claude/agents/github/project-board-sync.md +508 -508
- package/.claude/agents/github/release-manager.md +604 -604
- package/.claude/agents/github/release-swarm.md +582 -582
- package/.claude/agents/github/repo-architect.md +397 -397
- package/.claude/agents/github/swarm-issue.md +572 -572
- package/.claude/agents/github/swarm-pr.md +427 -427
- package/.claude/agents/github/sync-coordinator.md +451 -451
- package/.claude/agents/github/workflow-automation.md +902 -902
- package/.claude/agents/goal/agent.md +815 -815
- package/.claude/agents/optimization/benchmark-suite.md +664 -664
- package/.claude/agents/optimization/load-balancer.md +430 -430
- package/.claude/agents/optimization/performance-monitor.md +671 -671
- package/.claude/agents/optimization/resource-allocator.md +673 -673
- package/.claude/agents/optimization/topology-optimizer.md +807 -807
- package/.claude/agents/payments/agentic-payments.md +126 -126
- package/.claude/agents/sona/sona-learning-optimizer.md +74 -74
- package/.claude/agents/sparc/architecture.md +698 -698
- package/.claude/agents/sparc/pseudocode.md +519 -519
- package/.claude/agents/sparc/refinement.md +801 -801
- package/.claude/agents/sparc/specification.md +477 -477
- package/.claude/agents/specialized/mobile/spec-mobile-react-native.md +224 -224
- package/.claude/agents/specialized/spec-mobile-react-native.md +226 -226
- package/.claude/agents/sublinear/consensus-coordinator.md +337 -337
- package/.claude/agents/sublinear/matrix-optimizer.md +184 -184
- package/.claude/agents/sublinear/pagerank-analyzer.md +298 -298
- package/.claude/agents/sublinear/performance-optimizer.md +367 -367
- package/.claude/agents/sublinear/trading-predictor.md +245 -245
- package/.claude/agents/swarm/adaptive-coordinator.md +1126 -1126
- package/.claude/agents/swarm/hierarchical-coordinator.md +709 -709
- package/.claude/agents/swarm/mesh-coordinator.md +962 -962
- package/.claude/agents/templates/automation-smart-agent.md +204 -204
- package/.claude/agents/templates/base-template-generator.md +289 -289
- package/.claude/agents/templates/coordinator-swarm-init.md +89 -89
- package/.claude/agents/templates/github-pr-manager.md +176 -176
- package/.claude/agents/templates/implementer-sparc-coder.md +258 -258
- package/.claude/agents/templates/memory-coordinator.md +186 -186
- package/.claude/agents/templates/orchestrator-task.md +138 -138
- package/.claude/agents/templates/performance-analyzer.md +198 -198
- package/.claude/agents/templates/sparc-coordinator.md +513 -513
- package/.claude/agents/testing/production-validator.md +394 -394
- package/.claude/agents/testing/tdd-london-swarm.md +243 -243
- package/.claude/agents/v3/aidefence-guardian.md +282 -282
- package/.claude/agents/v3/claims-authorizer.md +208 -208
- package/.claude/agents/v3/collective-intelligence-coordinator.md +993 -993
- package/.claude/agents/v3/ddd-domain-expert.md +220 -220
- package/.claude/agents/v3/injection-analyst.md +236 -236
- package/.claude/agents/v3/performance-engineer.md +1233 -1233
- package/.claude/agents/v3/pii-detector.md +151 -151
- package/.claude/agents/v3/reasoningbank-learner.md +213 -213
- package/.claude/agents/v3/security-architect-aidefence.md +410 -410
- package/.claude/agents/v3/security-architect.md +867 -867
- package/.claude/agents/v3/swarm-memory-manager.md +157 -157
- package/.claude/agents/v3/v3-integration-architect.md +205 -205
- package/.claude/commands/agents/README.md +50 -50
- package/.claude/commands/agents/agent-capabilities.md +140 -140
- package/.claude/commands/agents/agent-coordination.md +28 -28
- package/.claude/commands/agents/agent-spawning.md +28 -28
- package/.claude/commands/agents/agent-types.md +216 -216
- package/.claude/commands/agents/health.md +139 -139
- package/.claude/commands/agents/list.md +100 -100
- package/.claude/commands/agents/logs.md +130 -130
- package/.claude/commands/agents/metrics.md +122 -122
- package/.claude/commands/agents/pool.md +127 -127
- package/.claude/commands/agents/spawn.md +140 -140
- package/.claude/commands/agents/status.md +115 -115
- package/.claude/commands/agents/stop.md +102 -102
- package/.claude/commands/analysis/COMMAND_COMPLIANCE_REPORT.md +53 -53
- package/.claude/commands/analysis/README.md +9 -9
- package/.claude/commands/analysis/bottleneck-detect.md +162 -162
- package/.claude/commands/analysis/performance-bottlenecks.md +58 -58
- package/.claude/commands/analysis/performance-report.md +25 -25
- package/.claude/commands/analysis/token-efficiency.md +44 -44
- package/.claude/commands/analysis/token-usage.md +25 -25
- package/.claude/commands/automation/README.md +9 -9
- package/.claude/commands/automation/auto-agent.md +122 -122
- package/.claude/commands/automation/self-healing.md +105 -105
- package/.claude/commands/automation/session-memory.md +89 -89
- package/.claude/commands/automation/smart-agents.md +72 -72
- package/.claude/commands/automation/smart-spawn.md +25 -25
- package/.claude/commands/automation/workflow-select.md +25 -25
- package/.claude/commands/claude-flow-help.md +103 -103
- package/.claude/commands/claude-flow-memory.md +107 -107
- package/.claude/commands/claude-flow-swarm.md +205 -205
- package/.claude/commands/coordination/README.md +9 -9
- package/.claude/commands/coordination/agent-spawn.md +25 -25
- package/.claude/commands/coordination/init.md +44 -44
- package/.claude/commands/coordination/orchestrate.md +43 -43
- package/.claude/commands/coordination/spawn.md +45 -45
- package/.claude/commands/coordination/swarm-init.md +85 -85
- package/.claude/commands/coordination/task-orchestrate.md +25 -25
- package/.claude/commands/github/README.md +11 -11
- package/.claude/commands/github/code-review-swarm.md +513 -513
- package/.claude/commands/github/code-review.md +25 -25
- package/.claude/commands/github/github-modes.md +146 -146
- package/.claude/commands/github/github-swarm.md +121 -121
- package/.claude/commands/github/issue-tracker.md +291 -291
- package/.claude/commands/github/issue-triage.md +25 -25
- package/.claude/commands/github/multi-repo-swarm.md +518 -518
- package/.claude/commands/github/pr-enhance.md +26 -26
- package/.claude/commands/github/pr-manager.md +169 -169
- package/.claude/commands/github/project-board-sync.md +470 -470
- package/.claude/commands/github/release-manager.md +339 -339
- package/.claude/commands/github/release-swarm.md +543 -543
- package/.claude/commands/github/repo-analyze.md +25 -25
- package/.claude/commands/github/repo-architect.md +366 -366
- package/.claude/commands/github/swarm-issue.md +484 -484
- package/.claude/commands/github/swarm-pr.md +287 -287
- package/.claude/commands/github/sync-coordinator.md +302 -302
- package/.claude/commands/github/workflow-automation.md +441 -441
- package/.claude/commands/hive-mind/README.md +17 -17
- package/.claude/commands/hive-mind/hive-mind-consensus.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-init.md +18 -18
- package/.claude/commands/hive-mind/hive-mind-memory.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-metrics.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-resume.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-sessions.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-spawn.md +21 -21
- package/.claude/commands/hive-mind/hive-mind-status.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-stop.md +8 -8
- package/.claude/commands/hive-mind/hive-mind-wizard.md +8 -8
- package/.claude/commands/hive-mind/hive-mind.md +27 -27
- package/.claude/commands/hooks/README.md +11 -11
- package/.claude/commands/hooks/overview.md +57 -57
- package/.claude/commands/hooks/post-edit.md +117 -117
- package/.claude/commands/hooks/post-task.md +112 -112
- package/.claude/commands/hooks/pre-edit.md +113 -113
- package/.claude/commands/hooks/pre-task.md +111 -111
- package/.claude/commands/hooks/session-end.md +118 -118
- package/.claude/commands/hooks/setup.md +102 -102
- package/.claude/commands/memory/README.md +9 -9
- package/.claude/commands/memory/memory-persist.md +25 -25
- package/.claude/commands/memory/memory-search.md +25 -25
- package/.claude/commands/memory/memory-usage.md +25 -25
- package/.claude/commands/memory/neural.md +47 -47
- package/.claude/commands/monitoring/README.md +9 -9
- package/.claude/commands/monitoring/agent-metrics.md +25 -25
- package/.claude/commands/monitoring/agents.md +44 -44
- package/.claude/commands/monitoring/real-time-view.md +25 -25
- package/.claude/commands/monitoring/status.md +46 -46
- package/.claude/commands/monitoring/swarm-monitor.md +25 -25
- package/.claude/commands/optimization/README.md +9 -9
- package/.claude/commands/optimization/auto-topology.md +61 -61
- package/.claude/commands/optimization/cache-manage.md +25 -25
- package/.claude/commands/optimization/parallel-execute.md +25 -25
- package/.claude/commands/optimization/parallel-execution.md +49 -49
- package/.claude/commands/optimization/topology-optimize.md +25 -25
- package/.claude/commands/pair/README.md +260 -260
- package/.claude/commands/pair/commands.md +545 -545
- package/.claude/commands/pair/config.md +509 -509
- package/.claude/commands/pair/examples.md +511 -511
- package/.claude/commands/pair/modes.md +347 -347
- package/.claude/commands/pair/session.md +406 -406
- package/.claude/commands/pair/start.md +208 -208
- package/.claude/commands/sparc/analyzer.md +51 -51
- package/.claude/commands/sparc/architect.md +53 -53
- package/.claude/commands/sparc/ask.md +97 -97
- package/.claude/commands/sparc/batch-executor.md +54 -54
- package/.claude/commands/sparc/code.md +89 -89
- package/.claude/commands/sparc/coder.md +54 -54
- package/.claude/commands/sparc/debug.md +83 -83
- package/.claude/commands/sparc/debugger.md +54 -54
- package/.claude/commands/sparc/designer.md +53 -53
- package/.claude/commands/sparc/devops.md +109 -109
- package/.claude/commands/sparc/docs-writer.md +80 -80
- package/.claude/commands/sparc/documenter.md +54 -54
- package/.claude/commands/sparc/innovator.md +54 -54
- package/.claude/commands/sparc/integration.md +83 -83
- package/.claude/commands/sparc/mcp.md +117 -117
- package/.claude/commands/sparc/memory-manager.md +54 -54
- package/.claude/commands/sparc/optimizer.md +54 -54
- package/.claude/commands/sparc/orchestrator.md +131 -131
- package/.claude/commands/sparc/post-deployment-monitoring-mode.md +83 -83
- package/.claude/commands/sparc/refinement-optimization-mode.md +83 -83
- package/.claude/commands/sparc/researcher.md +54 -54
- package/.claude/commands/sparc/reviewer.md +54 -54
- package/.claude/commands/sparc/security-review.md +80 -80
- package/.claude/commands/sparc/sparc-modes.md +174 -174
- package/.claude/commands/sparc/sparc.md +111 -111
- package/.claude/commands/sparc/spec-pseudocode.md +80 -80
- package/.claude/commands/sparc/supabase-admin.md +348 -348
- package/.claude/commands/sparc/swarm-coordinator.md +54 -54
- package/.claude/commands/sparc/tdd.md +54 -54
- package/.claude/commands/sparc/tester.md +54 -54
- package/.claude/commands/sparc/tutorial.md +79 -79
- package/.claude/commands/sparc/workflow-manager.md +54 -54
- package/.claude/commands/sparc.md +166 -166
- package/.claude/commands/stream-chain/pipeline.md +120 -120
- package/.claude/commands/stream-chain/run.md +69 -69
- package/.claude/commands/swarm/README.md +15 -15
- package/.claude/commands/swarm/analysis.md +95 -95
- package/.claude/commands/swarm/development.md +96 -96
- package/.claude/commands/swarm/examples.md +168 -168
- package/.claude/commands/swarm/maintenance.md +102 -102
- package/.claude/commands/swarm/optimization.md +117 -117
- package/.claude/commands/swarm/research.md +136 -136
- package/.claude/commands/swarm/swarm-analysis.md +8 -8
- package/.claude/commands/swarm/swarm-background.md +8 -8
- package/.claude/commands/swarm/swarm-init.md +19 -19
- package/.claude/commands/swarm/swarm-modes.md +8 -8
- package/.claude/commands/swarm/swarm-monitor.md +8 -8
- package/.claude/commands/swarm/swarm-spawn.md +19 -19
- package/.claude/commands/swarm/swarm-status.md +8 -8
- package/.claude/commands/swarm/swarm-strategies.md +8 -8
- package/.claude/commands/swarm/swarm.md +87 -87
- package/.claude/commands/swarm/testing.md +131 -131
- package/.claude/commands/training/README.md +9 -9
- package/.claude/commands/training/model-update.md +25 -25
- package/.claude/commands/training/neural-patterns.md +107 -107
- package/.claude/commands/training/neural-train.md +75 -75
- package/.claude/commands/training/pattern-learn.md +25 -25
- package/.claude/commands/training/specialization.md +62 -62
- package/.claude/commands/truth/start.md +142 -142
- package/.claude/commands/verify/check.md +49 -49
- package/.claude/commands/verify/start.md +127 -127
- package/.claude/commands/workflows/README.md +9 -9
- package/.claude/commands/workflows/development.md +77 -77
- package/.claude/commands/workflows/research.md +62 -62
- package/.claude/commands/workflows/workflow-create.md +25 -25
- package/.claude/commands/workflows/workflow-execute.md +25 -25
- package/.claude/commands/workflows/workflow-export.md +25 -25
- package/.claude/helpers/README.md +96 -96
- package/.claude/helpers/adr-compliance.sh +186 -186
- package/.claude/helpers/auto-commit.sh +178 -178
- package/.claude/helpers/auto-memory-hook.mjs +368 -368
- package/.claude/helpers/checkpoint-manager.sh +251 -251
- package/.claude/helpers/daemon-manager.sh +252 -252
- package/.claude/helpers/ddd-tracker.sh +144 -144
- package/.claude/helpers/github-safe.js +156 -156
- package/.claude/helpers/github-setup.sh +45 -45
- package/.claude/helpers/guidance-hook.sh +13 -13
- package/.claude/helpers/guidance-hooks.sh +102 -102
- package/.claude/helpers/health-monitor.sh +108 -108
- package/.claude/helpers/hook-handler.cjs +286 -286
- package/.claude/helpers/intelligence.cjs +1031 -1031
- package/.claude/helpers/learning-hooks.sh +329 -329
- package/.claude/helpers/learning-optimizer.sh +127 -127
- package/.claude/helpers/learning-service.mjs +1144 -1144
- package/.claude/helpers/memory.js +83 -83
- package/.claude/helpers/metrics-db.mjs +488 -488
- package/.claude/helpers/pattern-consolidator.sh +86 -86
- package/.claude/helpers/perf-worker.sh +160 -160
- package/.claude/helpers/post-commit +16 -16
- package/.claude/helpers/pre-commit +26 -26
- package/.claude/helpers/quick-start.sh +19 -19
- package/.claude/helpers/router.js +105 -105
- package/.claude/helpers/security-scanner.sh +127 -127
- package/.claude/helpers/session.js +135 -135
- package/.claude/helpers/setup-mcp.sh +18 -18
- package/.claude/helpers/standard-checkpoint-hooks.sh +189 -189
- package/.claude/helpers/statusline-hook.sh +21 -21
- package/.claude/helpers/statusline.cjs +604 -604
- package/.claude/helpers/statusline.js +352 -352
- package/.claude/helpers/swarm-comms.sh +353 -353
- package/.claude/helpers/swarm-hooks.sh +761 -761
- package/.claude/helpers/swarm-monitor.sh +210 -210
- package/.claude/helpers/sync-v3-metrics.sh +245 -245
- package/.claude/helpers/update-v3-progress.sh +165 -165
- package/.claude/helpers/v3-quick-status.sh +57 -57
- package/.claude/helpers/v3.sh +110 -110
- package/.claude/helpers/validate-v3-config.sh +215 -215
- package/.claude/helpers/worker-manager.sh +170 -170
- package/.claude/settings.json +182 -182
- package/.claude/skills/agentdb-advanced/SKILL.md +550 -550
- package/.claude/skills/agentdb-learning/SKILL.md +545 -545
- package/.claude/skills/agentdb-memory-patterns/SKILL.md +339 -339
- package/.claude/skills/agentdb-optimization/SKILL.md +509 -509
- package/.claude/skills/agentdb-vector-search/SKILL.md +339 -339
- package/.claude/skills/browser/SKILL.md +204 -204
- package/.claude/skills/dual-mode/README.md +71 -71
- package/.claude/skills/dual-mode/dual-collect.md +103 -103
- package/.claude/skills/dual-mode/dual-coordinate.md +85 -85
- package/.claude/skills/dual-mode/dual-spawn.md +81 -81
- package/.claude/skills/flow-nexus-neural/SKILL.md +727 -727
- package/.claude/skills/flow-nexus-platform/SKILL.md +1154 -1154
- package/.claude/skills/flow-nexus-swarm/SKILL.md +604 -604
- package/.claude/skills/github-code-review/SKILL.md +1125 -1125
- package/.claude/skills/github-multi-repo/SKILL.md +862 -862
- package/.claude/skills/github-project-management/SKILL.md +1262 -1262
- package/.claude/skills/github-release-management/SKILL.md +1064 -1064
- package/.claude/skills/github-workflow-automation/SKILL.md +1047 -1047
- package/.claude/skills/hooks-automation/SKILL.md +1201 -1201
- package/.claude/skills/pair-programming/SKILL.md +1202 -1202
- package/.claude/skills/reasoningbank-agentdb/SKILL.md +446 -446
- package/.claude/skills/reasoningbank-intelligence/SKILL.md +201 -201
- package/.claude/skills/skill-builder/SKILL.md +910 -910
- package/.claude/skills/sparc-methodology/SKILL.md +1106 -1106
- package/.claude/skills/stream-chain/SKILL.md +560 -560
- package/.claude/skills/swarm-advanced/SKILL.md +970 -970
- package/.claude/skills/swarm-orchestration/SKILL.md +179 -179
- package/.claude/skills/v3-cli-modernization/SKILL.md +871 -871
- package/.claude/skills/v3-core-implementation/SKILL.md +796 -796
- package/.claude/skills/v3-ddd-architecture/SKILL.md +441 -441
- package/.claude/skills/v3-integration-deep/SKILL.md +240 -240
- package/.claude/skills/v3-mcp-optimization/SKILL.md +776 -776
- package/.claude/skills/v3-memory-unification/SKILL.md +173 -173
- package/.claude/skills/v3-performance-optimization/SKILL.md +389 -389
- package/.claude/skills/v3-security-overhaul/SKILL.md +81 -81
- package/.claude/skills/v3-swarm-coordination/SKILL.md +339 -339
- package/.claude/skills/verification-quality/SKILL.md +691 -691
- package/README.md +416 -416
- package/bin/cli.js +314 -314
- package/bin/mcp-server.js +224 -224
- package/bin/preinstall.cjs +2 -2
- package/dist/src/benchmarks/gaia-critic.js +24 -24
- package/dist/src/commands/completions.js +409 -409
- package/dist/src/commands/daemon.js +44 -44
- package/dist/src/commands/embeddings.js +29 -29
- package/dist/src/commands/embeddings.js.map +1 -1
- package/dist/src/commands/hive-mind.js +97 -97
- package/dist/src/commands/hooks.js +9 -9
- package/dist/src/commands/ruvector/backup.js +23 -23
- package/dist/src/commands/ruvector/benchmark.js +31 -31
- package/dist/src/commands/ruvector/import.js +14 -14
- package/dist/src/commands/ruvector/init.js +115 -115
- package/dist/src/commands/ruvector/migrate.js +99 -99
- package/dist/src/commands/ruvector/optimize.js +51 -51
- package/dist/src/commands/ruvector/setup.js +624 -624
- package/dist/src/commands/ruvector/status.js +38 -38
- package/dist/src/init/claudemd-generator.js +227 -227
- package/dist/src/init/executor.js +453 -453
- package/dist/src/init/helpers-generator.js +727 -727
- package/dist/src/init/statusline-generator.js +570 -570
- package/dist/src/mcp-tools/agentdb-tools.js +15 -15
- package/dist/src/memory/graph-edge-writer.js +18 -18
- package/dist/src/memory/memory-bridge.js +84 -84
- package/dist/src/memory/memory-initializer.js +393 -393
- package/dist/src/memory/rabitq-index.js +5 -5
- package/dist/src/runtime/headless.js +28 -28
- package/dist/src/ruvector/flash-attention.d.ts +195 -0
- package/dist/src/ruvector/flash-attention.d.ts.map +1 -0
- package/dist/src/ruvector/flash-attention.js +643 -0
- package/dist/src/ruvector/flash-attention.js.map +1 -0
- package/dist/src/ruvector/moe-router.d.ts +206 -0
- package/dist/src/ruvector/moe-router.d.ts.map +1 -0
- package/dist/src/ruvector/moe-router.js +626 -0
- package/dist/src/ruvector/moe-router.js.map +1 -0
- package/dist/src/services/event-stream.d.ts +25 -0
- package/dist/src/services/event-stream.js +27 -0
- package/dist/src/services/headless-worker-executor.js +84 -84
- package/dist/src/services/loop-worker-runner.d.ts +16 -0
- package/dist/src/services/loop-worker-runner.js +34 -0
- package/dist/src/services/runtime-capabilities.d.ts +22 -0
- package/dist/src/services/runtime-capabilities.js +45 -0
- package/dist/src/transfer/deploy-seraphine.js +23 -23
- package/dist/tsconfig.tsbuildinfo +1 -1
- package/package.json +134 -134
- package/scripts/beir-bootstrap-significance.mjs +162 -162
- package/scripts/bench-rvagent.mjs +312 -312
- package/scripts/benchmark-codemods.mjs +127 -127
- package/scripts/benchmark-cross-repo.mjs +206 -206
- package/scripts/benchmark-pretrained-retrieval.mjs +305 -305
- package/scripts/benchmark-self-learning.mjs +241 -241
- package/scripts/benchmark-trajectory-mrr.mjs +191 -191
- package/scripts/deploy-ipfs-node.sh +153 -153
- package/scripts/grid-search-retrieval.mjs +216 -216
- package/scripts/postinstall.cjs +153 -153
- package/scripts/pretrain-from-github.mjs +358 -358
- package/scripts/publish-registry.ts +345 -345
- package/scripts/publish.sh +57 -57
- package/scripts/run-beir-bge.mjs +324 -336
- package/scripts/run-beir-hybrid.mjs +261 -263
- package/scripts/run-beir-lucene-bm25.mjs +154 -154
- package/scripts/run-beir-nfcorpus.mjs +231 -231
- package/scripts/run-beir-rrf-ablation.mjs +221 -221
- package/scripts/setup-ipfs-registry.md +366 -366
- package/dist/src/benchmarks/capability-tasks.d.ts +0 -126
- package/dist/src/benchmarks/capability-tasks.d.ts.map +0 -1
- package/dist/src/benchmarks/capability-tasks.js +0 -147
- package/dist/src/benchmarks/capability-tasks.js.map +0 -1
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts +0 -53
- package/dist/src/benchmarks/gaia-agent-contrastive.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-agent-contrastive.js +0 -177
- package/dist/src/benchmarks/gaia-agent-contrastive.js.map +0 -1
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts +0 -51
- package/dist/src/benchmarks/gaia-agent-gemini.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-agent-gemini.js +0 -352
- package/dist/src/benchmarks/gaia-agent-gemini.js.map +0 -1
- package/dist/src/benchmarks/gaia-claude-p.d.ts +0 -105
- package/dist/src/benchmarks/gaia-claude-p.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-claude-p.js +0 -269
- package/dist/src/benchmarks/gaia-claude-p.js.map +0 -1
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts +0 -20
- package/dist/src/benchmarks/gaia-claude-p.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-claude-p.smoke.js +0 -168
- package/dist/src/benchmarks/gaia-claude-p.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-codeagent-runner.py +0 -556
- package/dist/src/benchmarks/gaia-codeagent.d.ts +0 -129
- package/dist/src/benchmarks/gaia-codeagent.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-codeagent.js +0 -590
- package/dist/src/benchmarks/gaia-codeagent.js.map +0 -1
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts +0 -33
- package/dist/src/benchmarks/gaia-codeagent.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-codeagent.smoke.js +0 -241
- package/dist/src/benchmarks/gaia-codeagent.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts +0 -22
- package/dist/src/benchmarks/gaia-dag-pilot.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-dag-pilot.js +0 -158
- package/dist/src/benchmarks/gaia-dag-pilot.js.map +0 -1
- package/dist/src/benchmarks/gaia-dag.d.ts +0 -110
- package/dist/src/benchmarks/gaia-dag.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-dag.js +0 -658
- package/dist/src/benchmarks/gaia-dag.js.map +0 -1
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts +0 -21
- package/dist/src/benchmarks/gaia-dag.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-dag.smoke.js +0 -290
- package/dist/src/benchmarks/gaia-dag.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts +0 -17
- package/dist/src/benchmarks/gaia-ensemble-pilot.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-ensemble-pilot.js +0 -106
- package/dist/src/benchmarks/gaia-ensemble-pilot.js.map +0 -1
- package/dist/src/benchmarks/gaia-ensemble.d.ts +0 -103
- package/dist/src/benchmarks/gaia-ensemble.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-ensemble.js +0 -462
- package/dist/src/benchmarks/gaia-ensemble.js.map +0 -1
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts +0 -17
- package/dist/src/benchmarks/gaia-gemini-pilot.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-gemini-pilot.js +0 -168
- package/dist/src/benchmarks/gaia-gemini-pilot.js.map +0 -1
- package/dist/src/benchmarks/gaia-mode-router.d.ts +0 -67
- package/dist/src/benchmarks/gaia-mode-router.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-mode-router.js +0 -126
- package/dist/src/benchmarks/gaia-mode-router.js.map +0 -1
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts +0 -15
- package/dist/src/benchmarks/gaia-mode-router.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-mode-router.smoke.js +0 -131
- package/dist/src/benchmarks/gaia-mode-router.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts +0 -24
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js +0 -276
- package/dist/src/benchmarks/gaia-tools/grounded_query.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts +0 -45
- package/dist/src/benchmarks/gaia-tools/image_describe.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/image_describe.js +0 -222
- package/dist/src/benchmarks/gaia-tools/image_describe.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts +0 -24
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js +0 -169
- package/dist/src/benchmarks/gaia-tools/image_describe.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts +0 -27
- package/dist/src/benchmarks/gaia-tools/pdf_read.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/pdf_read.js +0 -226
- package/dist/src/benchmarks/gaia-tools/pdf_read.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts +0 -31
- package/dist/src/benchmarks/gaia-tools/python_exec.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/python_exec.js +0 -94
- package/dist/src/benchmarks/gaia-tools/python_exec.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts +0 -13
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js +0 -98
- package/dist/src/benchmarks/gaia-tools/python_exec.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts +0 -52
- package/dist/src/benchmarks/gaia-tools/visit_webpage.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js +0 -304
- package/dist/src/benchmarks/gaia-tools/visit_webpage.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts +0 -61
- package/dist/src/benchmarks/gaia-tools/web_browse.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_browse.js +0 -217
- package/dist/src/benchmarks/gaia-tools/web_browse.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts +0 -26
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js +0 -152
- package/dist/src/benchmarks/gaia-tools/web_browse.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts +0 -24
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js +0 -311
- package/dist/src/benchmarks/gaia-tools/web_search.smoke.js.map +0 -1
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts +0 -25
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.d.ts.map +0 -1
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js +0 -304
- package/dist/src/benchmarks/gaia-visit-webpage.smoke.js.map +0 -1
- package/dist/src/commands/performance-capability.d.ts +0 -30
- package/dist/src/commands/performance-capability.d.ts.map +0 -1
- package/dist/src/commands/performance-capability.js +0 -349
- package/dist/src/commands/performance-capability.js.map +0 -1
|
@@ -1,352 +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 { execSync } from 'node:child_process';
|
|
25
|
-
import * as fs from 'node:fs';
|
|
26
|
-
import * as path from 'node:path';
|
|
27
|
-
import { createDefaultToolCatalogue, } from './gaia-tools/index.js';
|
|
28
|
-
// ---------------------------------------------------------------------------
|
|
29
|
-
// Constants
|
|
30
|
-
// ---------------------------------------------------------------------------
|
|
31
|
-
const GEMINI_API_BASE = 'https://generativelanguage.googleapis.com/v1beta/models';
|
|
32
|
-
export const DEFAULT_GEMINI_MODEL = 'gemini-2.5-pro';
|
|
33
|
-
const DEFAULT_MAX_TURNS = 12;
|
|
34
|
-
const DEFAULT_MAX_TOKENS_PER_TURN = 8192;
|
|
35
|
-
const DEFAULT_PER_TURN_TIMEOUT_MS = 120_000;
|
|
36
|
-
/** Gemini 2.5 Pro pricing (USD per million tokens, ≤200k bracket). */
|
|
37
|
-
const GEMINI_INPUT_COST_PER_M = 1.25;
|
|
38
|
-
const GEMINI_OUTPUT_COST_PER_M = 10.00;
|
|
39
|
-
const FINAL_ANSWER_RE = /FINAL_ANSWER:\s*(.+)/i;
|
|
40
|
-
// ---------------------------------------------------------------------------
|
|
41
|
-
// API key resolution
|
|
42
|
-
// ---------------------------------------------------------------------------
|
|
43
|
-
export function resolveGeminiApiKey(apiKey) {
|
|
44
|
-
if (apiKey && apiKey.trim())
|
|
45
|
-
return apiKey.trim();
|
|
46
|
-
const envKey = process.env['GOOGLE_AI_API_KEY'];
|
|
47
|
-
if (envKey && envKey.trim())
|
|
48
|
-
return envKey.trim();
|
|
49
|
-
try {
|
|
50
|
-
const out = execSync('gcloud secrets versions access latest --secret=GOOGLE_AI_API_KEY --project=ruv-dev 2>/dev/null', { encoding: 'utf-8', timeout: 10_000 }).trim();
|
|
51
|
-
if (out)
|
|
52
|
-
return out;
|
|
53
|
-
}
|
|
54
|
-
catch {
|
|
55
|
-
/* fall through */
|
|
56
|
-
}
|
|
57
|
-
throw new Error('GOOGLE_AI_API_KEY not found. Set the env var or store in GCP Secret Manager.');
|
|
58
|
-
}
|
|
59
|
-
function toGeminiType(anthropicType) {
|
|
60
|
-
const map = {
|
|
61
|
-
string: 'STRING',
|
|
62
|
-
number: 'NUMBER',
|
|
63
|
-
integer: 'INTEGER',
|
|
64
|
-
boolean: 'BOOLEAN',
|
|
65
|
-
array: 'ARRAY',
|
|
66
|
-
object: 'OBJECT',
|
|
67
|
-
};
|
|
68
|
-
return map[anthropicType.toLowerCase()] ?? 'STRING';
|
|
69
|
-
}
|
|
70
|
-
function translateToolDef(def) {
|
|
71
|
-
const props = {};
|
|
72
|
-
for (const [key, val] of Object.entries(def.input_schema.properties)) {
|
|
73
|
-
props[key] = { type: toGeminiType(val.type), description: val.description };
|
|
74
|
-
}
|
|
75
|
-
return {
|
|
76
|
-
name: def.name,
|
|
77
|
-
description: def.description,
|
|
78
|
-
parameters: {
|
|
79
|
-
type: 'OBJECT',
|
|
80
|
-
properties: props,
|
|
81
|
-
required: def.input_schema.required ?? [],
|
|
82
|
-
},
|
|
83
|
-
};
|
|
84
|
-
}
|
|
85
|
-
// ---------------------------------------------------------------------------
|
|
86
|
-
// Single Gemini API call
|
|
87
|
-
// ---------------------------------------------------------------------------
|
|
88
|
-
async function callGemini(apiKey, model, contents, systemInstruction, functionDeclarations, maxTokens, timeoutMs) {
|
|
89
|
-
const url = `${GEMINI_API_BASE}/${model}:generateContent?key=${encodeURIComponent(apiKey)}`;
|
|
90
|
-
const requestBody = {
|
|
91
|
-
contents,
|
|
92
|
-
systemInstruction: { parts: [{ text: systemInstruction }] },
|
|
93
|
-
generationConfig: {
|
|
94
|
-
maxOutputTokens: maxTokens,
|
|
95
|
-
thinkingConfig: { thinkingBudget: 8192 },
|
|
96
|
-
},
|
|
97
|
-
};
|
|
98
|
-
if (functionDeclarations.length > 0) {
|
|
99
|
-
requestBody['tools'] = [{ functionDeclarations }];
|
|
100
|
-
requestBody['toolConfig'] = { functionCallingConfig: { mode: 'AUTO' } };
|
|
101
|
-
}
|
|
102
|
-
const controller = new AbortController();
|
|
103
|
-
const timer = setTimeout(() => controller.abort(), timeoutMs);
|
|
104
|
-
let res;
|
|
105
|
-
try {
|
|
106
|
-
res = await fetch(url, {
|
|
107
|
-
method: 'POST',
|
|
108
|
-
headers: { 'Content-Type': 'application/json' },
|
|
109
|
-
body: JSON.stringify(requestBody),
|
|
110
|
-
signal: controller.signal,
|
|
111
|
-
});
|
|
112
|
-
}
|
|
113
|
-
finally {
|
|
114
|
-
clearTimeout(timer);
|
|
115
|
-
}
|
|
116
|
-
if (!res.ok) {
|
|
117
|
-
const errText = await res.text().catch(() => '<unreadable>');
|
|
118
|
-
throw new Error(`Gemini API ${res.status}: ${errText.slice(0, 400)}`);
|
|
119
|
-
}
|
|
120
|
-
return (await res.json());
|
|
121
|
-
}
|
|
122
|
-
// ---------------------------------------------------------------------------
|
|
123
|
-
// Build initial contents (handles image attachments inline)
|
|
124
|
-
// ---------------------------------------------------------------------------
|
|
125
|
-
function buildInitialContents(question) {
|
|
126
|
-
const questionText = question.question;
|
|
127
|
-
const parts = [];
|
|
128
|
-
if (question.file_path) {
|
|
129
|
-
const ext = path.extname(question.file_path).toLowerCase();
|
|
130
|
-
const imageExts = ['.png', '.jpg', '.jpeg', '.gif', '.webp'];
|
|
131
|
-
if (imageExts.includes(ext)) {
|
|
132
|
-
try {
|
|
133
|
-
const buf = fs.readFileSync(question.file_path);
|
|
134
|
-
const mediaTypeMap = {
|
|
135
|
-
'.png': 'image/png', '.jpg': 'image/jpeg', '.jpeg': 'image/jpeg',
|
|
136
|
-
'.gif': 'image/gif', '.webp': 'image/webp',
|
|
137
|
-
};
|
|
138
|
-
parts.push({ text: questionText });
|
|
139
|
-
// Gemini inline image part
|
|
140
|
-
parts.push({
|
|
141
|
-
inlineData: {
|
|
142
|
-
mimeType: mediaTypeMap[ext] ?? 'image/png',
|
|
143
|
-
data: buf.toString('base64'),
|
|
144
|
-
},
|
|
145
|
-
});
|
|
146
|
-
}
|
|
147
|
-
catch {
|
|
148
|
-
parts.push({
|
|
149
|
-
text: questionText + `\n\nAttached file: ${question.file_path}. Call file_read to read it.`,
|
|
150
|
-
});
|
|
151
|
-
}
|
|
152
|
-
}
|
|
153
|
-
else {
|
|
154
|
-
parts.push({
|
|
155
|
-
text: questionText + `\n\nThis question has an attached file. Call file_read with path="${question.file_path}" to read it, then answer.`,
|
|
156
|
-
});
|
|
157
|
-
}
|
|
158
|
-
}
|
|
159
|
-
else {
|
|
160
|
-
parts.push({ text: questionText });
|
|
161
|
-
}
|
|
162
|
-
return [{ role: 'user', parts }];
|
|
163
|
-
}
|
|
164
|
-
// ---------------------------------------------------------------------------
|
|
165
|
-
// Extract FINAL_ANSWER from model response text
|
|
166
|
-
// ---------------------------------------------------------------------------
|
|
167
|
-
function extractFinalAnswer(candidate) {
|
|
168
|
-
if (!candidate?.content?.parts)
|
|
169
|
-
return null;
|
|
170
|
-
for (const part of candidate.content.parts) {
|
|
171
|
-
if (part.thought)
|
|
172
|
-
continue;
|
|
173
|
-
if (part.text) {
|
|
174
|
-
const match = FINAL_ANSWER_RE.exec(part.text);
|
|
175
|
-
if (match?.[1])
|
|
176
|
-
return match[1].trim();
|
|
177
|
-
}
|
|
178
|
-
}
|
|
179
|
-
return null;
|
|
180
|
-
}
|
|
181
|
-
// ---------------------------------------------------------------------------
|
|
182
|
-
// Check if response has any functionCall parts
|
|
183
|
-
// ---------------------------------------------------------------------------
|
|
184
|
-
function getFunctionCalls(candidate) {
|
|
185
|
-
if (!candidate?.content?.parts)
|
|
186
|
-
return [];
|
|
187
|
-
return candidate.content.parts
|
|
188
|
-
.filter((p) => !!p.functionCall)
|
|
189
|
-
.map((p) => p.functionCall);
|
|
190
|
-
}
|
|
191
|
-
// ---------------------------------------------------------------------------
|
|
192
|
-
// Main agent loop
|
|
193
|
-
// ---------------------------------------------------------------------------
|
|
194
|
-
export async function runGeminiAgent(question, options = {}) {
|
|
195
|
-
const { model = DEFAULT_GEMINI_MODEL, maxTurns = DEFAULT_MAX_TURNS, maxTokensPerTurn = DEFAULT_MAX_TOKENS_PER_TURN, perTurnTimeoutMs = DEFAULT_PER_TURN_TIMEOUT_MS, apiKey: suppliedKey, catalogue: suppliedCatalogue, } = options;
|
|
196
|
-
const wallStart = Date.now();
|
|
197
|
-
const apiKey = resolveGeminiApiKey(suppliedKey);
|
|
198
|
-
const catalogue = suppliedCatalogue ?? createDefaultToolCatalogue();
|
|
199
|
-
const functionDeclarations = catalogue.map((t) => translateToolDef(t.definition));
|
|
200
|
-
const systemInstruction = [
|
|
201
|
-
'You are a precise question-answering agent. Answer the user\'s question using the tools available.',
|
|
202
|
-
'',
|
|
203
|
-
'RULES:',
|
|
204
|
-
'1. Use tools when you need information you do not have with certainty.',
|
|
205
|
-
'2. When you have a final answer, output it on its own line in this EXACT format:',
|
|
206
|
-
' FINAL_ANSWER: <your answer here>',
|
|
207
|
-
'3. Keep answers concise. For numbers, give just the number. For names, give just the name.',
|
|
208
|
-
'4. Do not include units unless the question specifically asks for them.',
|
|
209
|
-
'5. MANDATORY: You MUST ALWAYS end your final response with a FINAL_ANSWER line.',
|
|
210
|
-
'6. If the question text appears garbled or reversed, try to interpret it before concluding you cannot answer.',
|
|
211
|
-
].join('\n');
|
|
212
|
-
const toolCallsByName = {};
|
|
213
|
-
let totalInputTokens = 0;
|
|
214
|
-
let totalOutputTokens = 0;
|
|
215
|
-
let totalThinkingTokens = 0;
|
|
216
|
-
const contents = buildInitialContents(question);
|
|
217
|
-
let turns = 0;
|
|
218
|
-
for (let turn = 0; turn < maxTurns; turn++) {
|
|
219
|
-
turns = turn + 1;
|
|
220
|
-
let resp;
|
|
221
|
-
try {
|
|
222
|
-
resp = await callGemini(apiKey, model, contents, systemInstruction, functionDeclarations, maxTokensPerTurn, perTurnTimeoutMs);
|
|
223
|
-
}
|
|
224
|
-
catch (err) {
|
|
225
|
-
return {
|
|
226
|
-
questionId: question.task_id,
|
|
227
|
-
finalAnswer: null,
|
|
228
|
-
turns,
|
|
229
|
-
toolCallsByName,
|
|
230
|
-
totalInputTokens,
|
|
231
|
-
totalOutputTokens,
|
|
232
|
-
totalThinkingTokens,
|
|
233
|
-
wallMs: Date.now() - wallStart,
|
|
234
|
-
estimatedCostUsd: estimateCost(totalInputTokens, totalOutputTokens),
|
|
235
|
-
error: err instanceof Error ? err.message : String(err),
|
|
236
|
-
};
|
|
237
|
-
}
|
|
238
|
-
if (resp.error) {
|
|
239
|
-
return {
|
|
240
|
-
questionId: question.task_id,
|
|
241
|
-
finalAnswer: null,
|
|
242
|
-
turns,
|
|
243
|
-
toolCallsByName,
|
|
244
|
-
totalInputTokens,
|
|
245
|
-
totalOutputTokens,
|
|
246
|
-
totalThinkingTokens,
|
|
247
|
-
wallMs: Date.now() - wallStart,
|
|
248
|
-
estimatedCostUsd: estimateCost(totalInputTokens, totalOutputTokens),
|
|
249
|
-
error: `Gemini error ${resp.error.code}: ${resp.error.message}`,
|
|
250
|
-
};
|
|
251
|
-
}
|
|
252
|
-
// Accumulate token counts
|
|
253
|
-
const usage = resp.usageMetadata ?? {};
|
|
254
|
-
totalInputTokens += usage.promptTokenCount ?? 0;
|
|
255
|
-
totalOutputTokens += usage.candidatesTokenCount ?? 0;
|
|
256
|
-
totalThinkingTokens += usage.thoughtsTokenCount ?? 0;
|
|
257
|
-
const candidate = resp.candidates?.[0];
|
|
258
|
-
const functionCalls = getFunctionCalls(candidate);
|
|
259
|
-
if (functionCalls.length > 0) {
|
|
260
|
-
// Append model turn (with functionCall parts)
|
|
261
|
-
if (candidate?.content) {
|
|
262
|
-
contents.push({ role: 'model', parts: candidate.content.parts });
|
|
263
|
-
}
|
|
264
|
-
// Execute all function calls and build functionResponse parts
|
|
265
|
-
const responseParts = await Promise.all(functionCalls.map(async (call) => {
|
|
266
|
-
toolCallsByName[call.name] = (toolCallsByName[call.name] ?? 0) + 1;
|
|
267
|
-
const tool = catalogue.find((t) => t.name === call.name);
|
|
268
|
-
if (!tool) {
|
|
269
|
-
return {
|
|
270
|
-
functionResponse: {
|
|
271
|
-
name: call.name,
|
|
272
|
-
response: { error: `Unknown tool "${call.name}"` },
|
|
273
|
-
},
|
|
274
|
-
};
|
|
275
|
-
}
|
|
276
|
-
try {
|
|
277
|
-
const output = await tool.execute(call.args);
|
|
278
|
-
return {
|
|
279
|
-
functionResponse: {
|
|
280
|
-
name: call.name,
|
|
281
|
-
response: { output },
|
|
282
|
-
},
|
|
283
|
-
};
|
|
284
|
-
}
|
|
285
|
-
catch (err) {
|
|
286
|
-
return {
|
|
287
|
-
functionResponse: {
|
|
288
|
-
name: call.name,
|
|
289
|
-
response: { error: err instanceof Error ? err.message : String(err) },
|
|
290
|
-
},
|
|
291
|
-
};
|
|
292
|
-
}
|
|
293
|
-
}));
|
|
294
|
-
contents.push({ role: 'user', parts: responseParts });
|
|
295
|
-
continue;
|
|
296
|
-
}
|
|
297
|
-
// No function calls — extract final answer
|
|
298
|
-
const finalAnswer = extractFinalAnswer(candidate);
|
|
299
|
-
return {
|
|
300
|
-
questionId: question.task_id,
|
|
301
|
-
finalAnswer,
|
|
302
|
-
turns,
|
|
303
|
-
toolCallsByName,
|
|
304
|
-
totalInputTokens,
|
|
305
|
-
totalOutputTokens,
|
|
306
|
-
totalThinkingTokens,
|
|
307
|
-
wallMs: Date.now() - wallStart,
|
|
308
|
-
estimatedCostUsd: estimateCost(totalInputTokens, totalOutputTokens + totalThinkingTokens),
|
|
309
|
-
};
|
|
310
|
-
}
|
|
311
|
-
// Exhausted maxTurns
|
|
312
|
-
return {
|
|
313
|
-
questionId: question.task_id,
|
|
314
|
-
finalAnswer: null,
|
|
315
|
-
turns,
|
|
316
|
-
toolCallsByName,
|
|
317
|
-
totalInputTokens,
|
|
318
|
-
totalOutputTokens,
|
|
319
|
-
totalThinkingTokens,
|
|
320
|
-
wallMs: Date.now() - wallStart,
|
|
321
|
-
estimatedCostUsd: estimateCost(totalInputTokens, totalOutputTokens + totalThinkingTokens),
|
|
322
|
-
timedOut: true,
|
|
323
|
-
};
|
|
324
|
-
}
|
|
325
|
-
function estimateCost(inputTokens, outputTokens) {
|
|
326
|
-
return ((inputTokens / 1_000_000) * GEMINI_INPUT_COST_PER_M +
|
|
327
|
-
(outputTokens / 1_000_000) * GEMINI_OUTPUT_COST_PER_M);
|
|
328
|
-
}
|
|
329
|
-
// ---------------------------------------------------------------------------
|
|
330
|
-
// Answer matching (same as gaia-agent.ts)
|
|
331
|
-
// ---------------------------------------------------------------------------
|
|
332
|
-
export function isGeminiAnswerCorrect(modelAnswer, expected) {
|
|
333
|
-
if (!modelAnswer)
|
|
334
|
-
return false;
|
|
335
|
-
const norm = (s) => s.trim().toLowerCase();
|
|
336
|
-
const normModel = norm(modelAnswer);
|
|
337
|
-
const normExpected = norm(expected);
|
|
338
|
-
if (normModel === normExpected)
|
|
339
|
-
return true;
|
|
340
|
-
if (normModel.includes(normExpected))
|
|
341
|
-
return true;
|
|
342
|
-
if (normExpected.includes(normModel))
|
|
343
|
-
return true;
|
|
344
|
-
const numModel = parseFloat(normModel.replace(/[^0-9.\-]/g, ''));
|
|
345
|
-
const numExpected = parseFloat(normExpected.replace(/[^0-9.\-]/g, ''));
|
|
346
|
-
if (!Number.isNaN(numModel) && !Number.isNaN(numExpected) &&
|
|
347
|
-
numExpected !== 0 &&
|
|
348
|
-
Math.abs((numModel - numExpected) / numExpected) < 0.01)
|
|
349
|
-
return true;
|
|
350
|
-
return false;
|
|
351
|
-
}
|
|
352
|
-
//# sourceMappingURL=gaia-agent-gemini.js.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"gaia-agent-gemini.js","sourceRoot":"","sources":["../../../src/benchmarks/gaia-agent-gemini.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;GAsBG;AAEH,OAAO,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAC9C,OAAO,KAAK,EAAE,MAAM,SAAS,CAAC;AAC9B,OAAO,KAAK,IAAI,MAAM,WAAW,CAAC;AAIlC,OAAO,EACL,0BAA0B,GAG3B,MAAM,uBAAuB,CAAC;AAE/B,8EAA8E;AAC9E,YAAY;AACZ,8EAA8E;AAE9E,MAAM,eAAe,GAAG,yDAAyD,CAAC;AAClF,MAAM,CAAC,MAAM,oBAAoB,GAAG,gBAAgB,CAAC;AACrD,MAAM,iBAAiB,GAAG,EAAE,CAAC;AAC7B,MAAM,2BAA2B,GAAG,IAAI,CAAC;AACzC,MAAM,2BAA2B,GAAG,OAAO,CAAC;AAE5C,sEAAsE;AACtE,MAAM,uBAAuB,GAAG,IAAI,CAAC;AACrC,MAAM,wBAAwB,GAAG,KAAK,CAAC;AAEvC,MAAM,eAAe,GAAG,uBAAuB,CAAC;AA6BhD,8EAA8E;AAC9E,qBAAqB;AACrB,8EAA8E;AAE9E,MAAM,UAAU,mBAAmB,CAAC,MAAe;IACjD,IAAI,MAAM,IAAI,MAAM,CAAC,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;IAElD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;IAChD,IAAI,MAAM,IAAI,MAAM,CAAC,IAAI,EAAE;QAAE,OAAO,MAAM,CAAC,IAAI,EAAE,CAAC;IAElD,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,QAAQ,CAClB,gGAAgG,EAChG,EAAE,QAAQ,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,CACvC,CAAC,IAAI,EAAE,CAAC;QACT,IAAI,GAAG;YAAE,OAAO,GAAG,CAAC;IACtB,CAAC;IAAC,MAAM,CAAC;QACP,kBAAkB;IACpB,CAAC;IAED,MAAM,IAAI,KAAK,CACb,8EAA8E,CAC/E,CAAC;AACJ,CAAC;AAgBD,SAAS,YAAY,CAAC,aAAqB;IACzC,MAAM,GAAG,GAA2B;QAClC,MAAM,EAAE,QAAQ;QAChB,MAAM,EAAE,QAAQ;QAChB,OAAO,EAAE,SAAS;QAClB,OAAO,EAAE,SAAS;QAClB,KAAK,EAAE,OAAO;QACd,MAAM,EAAE,QAAQ;KACjB,CAAC;IACF,OAAO,GAAG,CAAC,aAAa,CAAC,WAAW,EAAE,CAAC,IAAI,QAAQ,CAAC;AACtD,CAAC;AAED,SAAS,gBAAgB,CAAC,GAAmB;IAC3C,MAAM,KAAK,GAA2D,EAAE,CAAC;IACzE,KAAK,MAAM,CAAC,GAAG,EAAE,GAAG,CAAC,IAAI,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,YAAY,CAAC,UAAU,CAAC,EAAE,CAAC;QACrE,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,EAAE,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,WAAW,EAAE,GAAG,CAAC,WAAW,EAAE,CAAC;IAC9E,CAAC;IACD,OAAO;QACL,IAAI,EAAE,GAAG,CAAC,IAAI;QACd,WAAW,EAAE,GAAG,CAAC,WAAW;QAC5B,UAAU,EAAE;YACV,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE,KAAK;YACjB,QAAQ,EAAE,GAAG,CAAC,YAAY,CAAC,QAAQ,IAAI,EAAE;SAC1C;KACF,CAAC;AACJ,CAAC;AAoCD,8EAA8E;AAC9E,yBAAyB;AACzB,8EAA8E;AAE9E,KAAK,UAAU,UAAU,CACvB,MAAc,EACd,KAAa,EACb,QAAyB,EACzB,iBAAyB,EACzB,oBAAiD,EACjD,SAAiB,EACjB,SAAiB;IAEjB,MAAM,GAAG,GAAG,GAAG,eAAe,IAAI,KAAK,wBAAwB,kBAAkB,CAAC,MAAM,CAAC,EAAE,CAAC;IAE5F,MAAM,WAAW,GAA4B;QAC3C,QAAQ;QACR,iBAAiB,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,iBAAiB,EAAE,CAAC,EAAE;QAC3D,gBAAgB,EAAE;YAChB,eAAe,EAAE,SAAS;YAC1B,cAAc,EAAE,EAAE,cAAc,EAAE,IAAI,EAAE;SACzC;KACF,CAAC;IAEF,IAAI,oBAAoB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACpC,WAAW,CAAC,OAAO,CAAC,GAAG,CAAC,EAAE,oBAAoB,EAAE,CAAC,CAAC;QAClD,WAAW,CAAC,YAAY,CAAC,GAAG,EAAE,qBAAqB,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,CAAC;IAC1E,CAAC;IAED,MAAM,UAAU,GAAG,IAAI,eAAe,EAAE,CAAC;IACzC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,UAAU,CAAC,KAAK,EAAE,EAAE,SAAS,CAAC,CAAC;IAE9D,IAAI,GAAa,CAAC;IAClB,IAAI,CAAC;QACH,GAAG,GAAG,MAAM,KAAK,CAAC,GAAG,EAAE;YACrB,MAAM,EAAE,MAAM;YACd,OAAO,EAAE,EAAE,cAAc,EAAE,kBAAkB,EAAE;YAC/C,IAAI,EAAE,IAAI,CAAC,SAAS,CAAC,WAAW,CAAC;YACjC,MAAM,EAAE,UAAU,CAAC,MAAM;SAC1B,CAAC,CAAC;IACL,CAAC;YAAS,CAAC;QACT,YAAY,CAAC,KAAK,CAAC,CAAC;IACtB,CAAC;IAED,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;QACZ,MAAM,OAAO,GAAG,MAAM,GAAG,CAAC,IAAI,EAAE,CAAC,KAAK,CAAC,GAAG,EAAE,CAAC,cAAc,CAAC,CAAC;QAC7D,MAAM,IAAI,KAAK,CAAC,cAAc,GAAG,CAAC,MAAM,KAAK,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;IACxE,CAAC;IAED,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAAkC,CAAC;AAC7D,CAAC;AAED,8EAA8E;AAC9E,4DAA4D;AAC5D,8EAA8E;AAE9E,SAAS,oBAAoB,CAAC,QAAsB;IAClD,MAAM,YAAY,GAAG,QAAQ,CAAC,QAAQ,CAAC;IACvC,MAAM,KAAK,GAAiB,EAAE,CAAC;IAE/B,IAAI,QAAQ,CAAC,SAAS,EAAE,CAAC;QACvB,MAAM,GAAG,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC,WAAW,EAAE,CAAC;QAC3D,MAAM,SAAS,GAAG,CAAC,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;QAE7D,IAAI,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YAC5B,IAAI,CAAC;gBACH,MAAM,GAAG,GAAG,EAAE,CAAC,YAAY,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;gBAChD,MAAM,YAAY,GAA2B;oBAC3C,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,OAAO,EAAE,YAAY;oBAChE,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,YAAY;iBAC3C,CAAC;gBACF,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;gBACnC,2BAA2B;gBAC1B,KAAmD,CAAC,IAAI,CAAC;oBACxD,UAAU,EAAE;wBACV,QAAQ,EAAE,YAAY,CAAC,GAAG,CAAC,IAAI,WAAW;wBAC1C,IAAI,EAAE,GAAG,CAAC,QAAQ,CAAC,QAAQ,CAAC;qBAC7B;iBACF,CAAC,CAAC;YACL,CAAC;YAAC,MAAM,CAAC;gBACP,KAAK,CAAC,IAAI,CAAC;oBACT,IAAI,EAAE,YAAY,GAAG,sBAAsB,QAAQ,CAAC,SAAS,8BAA8B;iBAC5F,CAAC,CAAC;YACL,CAAC;QACH,CAAC;aAAM,CAAC;YACN,KAAK,CAAC,IAAI,CAAC;gBACT,IAAI,EAAE,YAAY,GAAG,qEAAqE,QAAQ,CAAC,SAAS,4BAA4B;aACzI,CAAC,CAAC;QACL,CAAC;IACH,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,YAAY,EAAE,CAAC,CAAC;IACrC,CAAC;IAED,OAAO,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC,CAAC;AACnC,CAAC;AAED,8EAA8E;AAC9E,gDAAgD;AAChD,8EAA8E;AAE9E,SAAS,kBAAkB,CAAC,SAAsC;IAChE,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,KAAK;QAAE,OAAO,IAAI,CAAC;IAC5C,KAAK,MAAM,IAAI,IAAI,SAAS,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC;QAC3C,IAAI,IAAI,CAAC,OAAO;YAAE,SAAS;QAC3B,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC;YACd,MAAM,KAAK,GAAG,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;YAC9C,IAAI,KAAK,EAAE,CAAC,CAAC,CAAC;gBAAE,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;QACzC,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,8EAA8E;AAC9E,+CAA+C;AAC/C,8EAA8E;AAE9E,SAAS,gBAAgB,CACvB,SAAsC;IAEtC,IAAI,CAAC,SAAS,EAAE,OAAO,EAAE,KAAK;QAAE,OAAO,EAAE,CAAC;IAC1C,OAAO,SAAS,CAAC,OAAO,CAAC,KAAK;SAC3B,MAAM,CAAC,CAAC,CAAC,EAA+E,EAAE,CACzF,CAAC,CAAC,CAAC,CAAC,YAAY,CACjB;SACA,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC;AAChC,CAAC;AAED,8EAA8E;AAC9E,kBAAkB;AAClB,8EAA8E;AAE9E,MAAM,CAAC,KAAK,UAAU,cAAc,CAClC,QAAsB,EACtB,UAA8B,EAAE;IAEhC,MAAM,EACJ,KAAK,GAAG,oBAAoB,EAC5B,QAAQ,GAAG,iBAAiB,EAC5B,gBAAgB,GAAG,2BAA2B,EAC9C,gBAAgB,GAAG,2BAA2B,EAC9C,MAAM,EAAE,WAAW,EACnB,SAAS,EAAE,iBAAiB,GAC7B,GAAG,OAAO,CAAC;IAEZ,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,MAAM,MAAM,GAAG,mBAAmB,CAAC,WAAW,CAAC,CAAC;IAChD,MAAM,SAAS,GAAG,iBAAiB,IAAI,0BAA0B,EAAE,CAAC;IACpE,MAAM,oBAAoB,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,gBAAgB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC;IAElF,MAAM,iBAAiB,GAAG;QACxB,oGAAoG;QACpG,EAAE;QACF,QAAQ;QACR,wEAAwE;QACxE,kFAAkF;QAClF,qCAAqC;QACrC,4FAA4F;QAC5F,yEAAyE;QACzE,iFAAiF;QACjF,+GAA+G;KAChH,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEb,MAAM,eAAe,GAA2B,EAAE,CAAC;IACnD,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,IAAI,iBAAiB,GAAG,CAAC,CAAC;IAC1B,IAAI,mBAAmB,GAAG,CAAC,CAAC;IAE5B,MAAM,QAAQ,GAAoB,oBAAoB,CAAC,QAAQ,CAAC,CAAC;IACjE,IAAI,KAAK,GAAG,CAAC,CAAC;IAEd,KAAK,IAAI,IAAI,GAAG,CAAC,EAAE,IAAI,GAAG,QAAQ,EAAE,IAAI,EAAE,EAAE,CAAC;QAC3C,KAAK,GAAG,IAAI,GAAG,CAAC,CAAC;QAEjB,IAAI,IAAmC,CAAC;QACxC,IAAI,CAAC;YACH,IAAI,GAAG,MAAM,UAAU,CACrB,MAAM,EAAE,KAAK,EAAE,QAAQ,EAAE,iBAAiB,EAC1C,oBAAoB,EAAE,gBAAgB,EAAE,gBAAgB,CACzD,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO;gBACL,UAAU,EAAE,QAAQ,CAAC,OAAO;gBAC5B,WAAW,EAAE,IAAI;gBACjB,KAAK;gBACL,eAAe;gBACf,gBAAgB;gBAChB,iBAAiB;gBACjB,mBAAmB;gBACnB,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBAC9B,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAE,iBAAiB,CAAC;gBACnE,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;aACxD,CAAC;QACJ,CAAC;QAED,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;YACf,OAAO;gBACL,UAAU,EAAE,QAAQ,CAAC,OAAO;gBAC5B,WAAW,EAAE,IAAI;gBACjB,KAAK;gBACL,eAAe;gBACf,gBAAgB;gBAChB,iBAAiB;gBACjB,mBAAmB;gBACnB,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;gBAC9B,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAE,iBAAiB,CAAC;gBACnE,KAAK,EAAE,gBAAgB,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,IAAI,CAAC,KAAK,CAAC,OAAO,EAAE;aAChE,CAAC;QACJ,CAAC;QAED,0BAA0B;QAC1B,MAAM,KAAK,GAAG,IAAI,CAAC,aAAa,IAAI,EAAE,CAAC;QACvC,gBAAgB,IAAI,KAAK,CAAC,gBAAgB,IAAI,CAAC,CAAC;QAChD,iBAAiB,IAAI,KAAK,CAAC,oBAAoB,IAAI,CAAC,CAAC;QACrD,mBAAmB,IAAI,KAAK,CAAC,kBAAkB,IAAI,CAAC,CAAC;QAErD,MAAM,SAAS,GAAG,IAAI,CAAC,UAAU,EAAE,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,aAAa,GAAG,gBAAgB,CAAC,SAAS,CAAC,CAAC;QAElD,IAAI,aAAa,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC7B,8CAA8C;YAC9C,IAAI,SAAS,EAAE,OAAO,EAAE,CAAC;gBACvB,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,KAAK,EAAE,SAAS,CAAC,OAAO,CAAC,KAAK,EAAE,CAAC,CAAC;YACnE,CAAC;YAED,8DAA8D;YAC9D,MAAM,aAAa,GAAiB,MAAM,OAAO,CAAC,GAAG,CACnD,aAAa,CAAC,GAAG,CAAC,KAAK,EAAE,IAAI,EAAuB,EAAE;gBACpD,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,eAAe,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC;gBACnE,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,IAAI,CAAC,CAAC;gBACzD,IAAI,CAAC,IAAI,EAAE,CAAC;oBACV,OAAO;wBACL,gBAAgB,EAAE;4BAChB,IAAI,EAAE,IAAI,CAAC,IAAI;4BACf,QAAQ,EAAE,EAAE,KAAK,EAAE,iBAAiB,IAAI,CAAC,IAAI,GAAG,EAAE;yBACnD;qBACF,CAAC;gBACJ,CAAC;gBACD,IAAI,CAAC;oBACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;oBAC7C,OAAO;wBACL,gBAAgB,EAAE;4BAChB,IAAI,EAAE,IAAI,CAAC,IAAI;4BACf,QAAQ,EAAE,EAAE,MAAM,EAAE;yBACrB;qBACF,CAAC;gBACJ,CAAC;gBAAC,OAAO,GAAG,EAAE,CAAC;oBACb,OAAO;wBACL,gBAAgB,EAAE;4BAChB,IAAI,EAAE,IAAI,CAAC,IAAI;4BACf,QAAQ,EAAE,EAAE,KAAK,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,EAAE;yBACtE;qBACF,CAAC;gBACJ,CAAC;YACH,CAAC,CAAC,CACH,CAAC;YAEF,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,aAAa,EAAE,CAAC,CAAC;YACtD,SAAS;QACX,CAAC;QAED,2CAA2C;QAC3C,MAAM,WAAW,GAAG,kBAAkB,CAAC,SAAS,CAAC,CAAC;QAClD,OAAO;YACL,UAAU,EAAE,QAAQ,CAAC,OAAO;YAC5B,WAAW;YACX,KAAK;YACL,eAAe;YACf,gBAAgB;YAChB,iBAAiB;YACjB,mBAAmB;YACnB,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;YAC9B,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAE,iBAAiB,GAAG,mBAAmB,CAAC;SAC1F,CAAC;IACJ,CAAC;IAED,qBAAqB;IACrB,OAAO;QACL,UAAU,EAAE,QAAQ,CAAC,OAAO;QAC5B,WAAW,EAAE,IAAI;QACjB,KAAK;QACL,eAAe;QACf,gBAAgB;QAChB,iBAAiB;QACjB,mBAAmB;QACnB,MAAM,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;QAC9B,gBAAgB,EAAE,YAAY,CAAC,gBAAgB,EAAE,iBAAiB,GAAG,mBAAmB,CAAC;QACzF,QAAQ,EAAE,IAAI;KACf,CAAC;AACJ,CAAC;AAED,SAAS,YAAY,CAAC,WAAmB,EAAE,YAAoB;IAC7D,OAAO,CACL,CAAC,WAAW,GAAG,SAAS,CAAC,GAAG,uBAAuB;QACnD,CAAC,YAAY,GAAG,SAAS,CAAC,GAAG,wBAAwB,CACtD,CAAC;AACJ,CAAC;AAED,8EAA8E;AAC9E,0CAA0C;AAC1C,8EAA8E;AAE9E,MAAM,UAAU,qBAAqB,CAAC,WAAmB,EAAE,QAAgB;IACzE,IAAI,CAAC,WAAW;QAAE,OAAO,KAAK,CAAC;IAC/B,MAAM,IAAI,GAAG,CAAC,CAAS,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IACnD,MAAM,SAAS,GAAG,IAAI,CAAC,WAAW,CAAC,CAAC;IACpC,MAAM,YAAY,GAAG,IAAI,CAAC,QAAQ,CAAC,CAAC;IACpC,IAAI,SAAS,KAAK,YAAY;QAAE,OAAO,IAAI,CAAC;IAC5C,IAAI,SAAS,CAAC,QAAQ,CAAC,YAAY,CAAC;QAAE,OAAO,IAAI,CAAC;IAClD,IAAI,YAAY,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,OAAO,IAAI,CAAC;IAClD,MAAM,QAAQ,GAAG,UAAU,CAAC,SAAS,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAC;IACjE,MAAM,WAAW,GAAG,UAAU,CAAC,YAAY,CAAC,OAAO,CAAC,YAAY,EAAE,EAAE,CAAC,CAAC,CAAC;IACvE,IACE,CAAC,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,CAAC;QACrD,WAAW,KAAK,CAAC;QACjB,IAAI,CAAC,GAAG,CAAC,CAAC,QAAQ,GAAG,WAAW,CAAC,GAAG,WAAW,CAAC,GAAG,IAAI;QACvD,OAAO,IAAI,CAAC;IACd,OAAO,KAAK,CAAC;AACf,CAAC"}
|
|
@@ -1,105 +0,0 @@
|
|
|
1
|
-
/**
|
|
2
|
-
* GAIA Claude-p Wrapper — iter 54 (#2156)
|
|
3
|
-
*
|
|
4
|
-
* Delegates each GAIA question to `claude -p` (Claude Code headless mode),
|
|
5
|
-
* which gives us WebSearch, WebFetch, Read (multimodal incl. PDF/DOCX/images),
|
|
6
|
-
* and Bash (Python execution) for free — the same tools HAL uses.
|
|
7
|
-
*
|
|
8
|
-
* Why this approach over a native TS CodeAgent:
|
|
9
|
-
* - HAL gaps vs ruflo were: visit_webpage, file reading (PDF/DOCX/XLSX/images),
|
|
10
|
-
* Python execution. Claude Code's built-in tools solve ALL of these.
|
|
11
|
-
* - No wheel-reinvention: battle-tested tool infra, native multimodal, proper
|
|
12
|
-
* tool-budget management, Anthropic WebSearch API.
|
|
13
|
-
* - Baseline: 24/53 (45.3%). Target: ≥45/53 to surpass HAL's 82.07%.
|
|
14
|
-
*
|
|
15
|
-
* SECURITY NOTE on --dangerously-skip-permissions:
|
|
16
|
-
* This flag is ONLY used inside the GAIA harness context, which is a sandboxed
|
|
17
|
-
* benchmark evaluation environment. GAIA questions have no real-world
|
|
18
|
-
* side effects — they are read-only research questions. The flag lets Claude Code
|
|
19
|
-
* use its tools (WebSearch, WebFetch, Read, Bash) without per-tool permission
|
|
20
|
-
* prompts, which is required for unattended benchmark execution. It MUST NOT
|
|
21
|
-
* be used in production workflows where Claude Code could affect real systems.
|
|
22
|
-
*
|
|
23
|
-
* JSON output format from `claude -p --output-format json`:
|
|
24
|
-
* {
|
|
25
|
-
* type: "result",
|
|
26
|
-
* subtype: "success" | "error_max_budget_usd" | ...,
|
|
27
|
-
* is_error: boolean,
|
|
28
|
-
* result: string, // final assistant message text
|
|
29
|
-
* total_cost_usd: number,
|
|
30
|
-
* duration_ms: number,
|
|
31
|
-
* num_turns: number,
|
|
32
|
-
* ...
|
|
33
|
-
* }
|
|
34
|
-
*
|
|
35
|
-
* Refs: ADR-138 (reference, NOT implemented), iter 54, #2156
|
|
36
|
-
*/
|
|
37
|
-
import type { GaiaQuestion } from './gaia-loader.js';
|
|
38
|
-
/** Default model for claude -p GAIA runs. Sonnet for quality parity with HAL. */
|
|
39
|
-
export declare const CLAUDE_P_DEFAULT_MODEL = "claude-sonnet-4-6";
|
|
40
|
-
/** Per-question budget cap (USD). HAL uses Sonnet 4.5 so $0.30 headroom is safe. */
|
|
41
|
-
export declare const CLAUDE_P_PER_QUESTION_BUDGET_USD = 0.3;
|
|
42
|
-
/** Subprocess timeout: 5 minutes per question. */
|
|
43
|
-
export declare const CLAUDE_P_TIMEOUT_MS: number;
|
|
44
|
-
export interface ClaudePResult {
|
|
45
|
-
/** The extracted answer, or null if extraction failed. */
|
|
46
|
-
finalAnswer: string | null;
|
|
47
|
-
/** Raw result text from claude -p. */
|
|
48
|
-
rawResult: string;
|
|
49
|
-
/** Whether claude -p exited with an error. */
|
|
50
|
-
isError: boolean;
|
|
51
|
-
/** claude -p's reported error message (if any). */
|
|
52
|
-
errorMessage?: string;
|
|
53
|
-
/** Actual cost reported by claude -p. */
|
|
54
|
-
costUsd: number;
|
|
55
|
-
/** Wall-clock time in ms. */
|
|
56
|
-
wallMs: number;
|
|
57
|
-
/** Number of turns claude -p used. */
|
|
58
|
-
numTurns: number;
|
|
59
|
-
/** claude -p stop reason. */
|
|
60
|
-
stopReason?: string;
|
|
61
|
-
}
|
|
62
|
-
export interface ClaudePOptions {
|
|
63
|
-
/** Model ID (default: CLAUDE_P_DEFAULT_MODEL). */
|
|
64
|
-
model?: string;
|
|
65
|
-
/** Per-question budget cap in USD (default: CLAUDE_P_PER_QUESTION_BUDGET_USD). */
|
|
66
|
-
budgetUsd?: number;
|
|
67
|
-
/** Timeout in ms (default: CLAUDE_P_TIMEOUT_MS). */
|
|
68
|
-
timeoutMs?: number;
|
|
69
|
-
/** Absolute path to the claude binary (default: resolved from $PATH). */
|
|
70
|
-
claudeBin?: string;
|
|
71
|
-
}
|
|
72
|
-
/**
|
|
73
|
-
* Build the prompt sent to claude -p for a GAIA question.
|
|
74
|
-
*
|
|
75
|
-
* Includes the question text, optional attachment path, and precise instructions
|
|
76
|
-
* for using available tools and producing FINAL_ANSWER: in the expected format.
|
|
77
|
-
*/
|
|
78
|
-
export declare function buildClaudePPrompt(question: GaiaQuestion): string;
|
|
79
|
-
/**
|
|
80
|
-
* Run a single GAIA question via `claude -p` headless mode.
|
|
81
|
-
*
|
|
82
|
-
* Spawns a subprocess, captures JSON output, extracts the final answer.
|
|
83
|
-
*/
|
|
84
|
-
export declare function runGaiaQuestionViaClaudeP(question: GaiaQuestion, options?: ClaudePOptions): Promise<ClaudePResult>;
|
|
85
|
-
/**
|
|
86
|
-
* Extract the FINAL_ANSWER value from claude -p's result text.
|
|
87
|
-
*
|
|
88
|
-
* Primary: regex match on `FINAL_ANSWER: <value>`
|
|
89
|
-
* Fallback: last non-empty line if no FINAL_ANSWER marker found.
|
|
90
|
-
*/
|
|
91
|
-
export declare function extractFinalAnswer(text: string): string | null;
|
|
92
|
-
export interface ClaudePBatchOptions extends ClaudePOptions {
|
|
93
|
-
/** Max parallel questions (default: 2 — claude -p uses significant local resources). */
|
|
94
|
-
concurrency?: number;
|
|
95
|
-
/** Callback for per-question progress logging. */
|
|
96
|
-
onProgress?: (idx: number, total: number, questionId: string, answer: string | null, costUsd: number) => void;
|
|
97
|
-
}
|
|
98
|
-
/**
|
|
99
|
-
* Run a batch of GAIA questions through the claude -p wrapper.
|
|
100
|
-
*
|
|
101
|
-
* Concurrency is limited (default 2) because each claude -p subprocess
|
|
102
|
-
* is heavyweight — it starts a full Claude Code session with LSP etc.
|
|
103
|
-
*/
|
|
104
|
-
export declare function runGaiaQuestionsBatchViaClaudeP(questions: GaiaQuestion[], options?: ClaudePBatchOptions): Promise<ClaudePResult[]>;
|
|
105
|
-
//# sourceMappingURL=gaia-claude-p.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"gaia-claude-p.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/gaia-claude-p.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAmCG;AAIH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAMrD,iFAAiF;AACjF,eAAO,MAAM,sBAAsB,sBAAsB,CAAC;AAE1D,oFAAoF;AACpF,eAAO,MAAM,gCAAgC,MAAO,CAAC;AAErD,kDAAkD;AAClD,eAAO,MAAM,mBAAmB,QAAgB,CAAC;AASjD,MAAM,WAAW,aAAa;IAC5B,0DAA0D;IAC1D,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,sCAAsC;IACtC,SAAS,EAAE,MAAM,CAAC;IAClB,8CAA8C;IAC9C,OAAO,EAAE,OAAO,CAAC;IACjB,mDAAmD;IACnD,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,yCAAyC;IACzC,OAAO,EAAE,MAAM,CAAC;IAChB,6BAA6B;IAC7B,MAAM,EAAE,MAAM,CAAC;IACf,sCAAsC;IACtC,QAAQ,EAAE,MAAM,CAAC;IACjB,6BAA6B;IAC7B,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,cAAc;IAC7B,kDAAkD;IAClD,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,kFAAkF;IAClF,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,oDAAoD;IACpD,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,yEAAyE;IACzE,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAMD;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,QAAQ,EAAE,YAAY,GAAG,MAAM,CAiCjE;AAMD;;;;GAIG;AACH,wBAAsB,yBAAyB,CAC7C,QAAQ,EAAE,YAAY,EACtB,OAAO,GAAE,cAAmB,GAC3B,OAAO,CAAC,aAAa,CAAC,CAsFxB;AA4ED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAiB9D;AAMD,MAAM,WAAW,mBAAoB,SAAQ,cAAc;IACzD,wFAAwF;IACxF,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,kDAAkD;IAClD,UAAU,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,GAAG,IAAI,EAAE,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CAC/G;AAED;;;;;GAKG;AACH,wBAAsB,+BAA+B,CACnD,SAAS,EAAE,YAAY,EAAE,EACzB,OAAO,GAAE,mBAAwB,GAChC,OAAO,CAAC,aAAa,EAAE,CAAC,CAqB1B"}
|