@tangle-network/agent-eval 0.93.0 → 0.95.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +15 -0
- package/README.md +44 -30
- package/dist/action-policy.d.ts +24 -0
- package/dist/action-policy.d.ts.map +1 -0
- package/dist/action-policy.test.d.ts +2 -0
- package/dist/action-policy.test.d.ts.map +1 -0
- package/dist/active-learning.d.ts +41 -0
- package/dist/active-learning.d.ts.map +1 -0
- package/dist/adapters/http.d.ts +15 -21
- package/dist/adapters/http.d.ts.map +1 -0
- package/dist/adapters/http.js.map +1 -1
- package/dist/adapters/langchain.d.ts +7 -13
- package/dist/adapters/langchain.d.ts.map +1 -0
- package/dist/adapters/otel.d.ts +13 -24
- package/dist/adapters/otel.d.ts.map +1 -0
- package/dist/agent-profile-cell.d.ts +101 -0
- package/dist/agent-profile-cell.d.ts.map +1 -0
- package/dist/agent-profile.d.ts +27 -0
- package/dist/agent-profile.d.ts.map +1 -0
- package/dist/agent-profile.test.d.ts +2 -0
- package/dist/agent-profile.test.d.ts.map +1 -0
- package/dist/analyst/adapters.d.ts +62 -0
- package/dist/analyst/adapters.d.ts.map +1 -0
- package/dist/analyst/analyst.test.d.ts +2 -0
- package/dist/analyst/analyst.test.d.ts.map +1 -0
- package/dist/analyst/ax-service.d.ts +27 -0
- package/dist/analyst/ax-service.d.ts.map +1 -0
- package/dist/analyst/behavioral-analyst.d.ts +28 -0
- package/dist/analyst/behavioral-analyst.d.ts.map +1 -0
- package/dist/analyst/chat-client.d.ts +91 -0
- package/dist/analyst/chat-client.d.ts.map +1 -0
- package/dist/analyst/default-registry.d.ts +27 -0
- package/dist/analyst/default-registry.d.ts.map +1 -0
- package/dist/analyst/default-registry.test.d.ts +2 -0
- package/dist/analyst/default-registry.test.d.ts.map +1 -0
- package/dist/analyst/finding-signature.d.ts +48 -0
- package/dist/analyst/finding-signature.d.ts.map +1 -0
- package/dist/analyst/finding-subject.d.ts +146 -0
- package/dist/analyst/finding-subject.d.ts.map +1 -0
- package/dist/analyst/finding-subject.test.d.ts +2 -0
- package/dist/analyst/finding-subject.test.d.ts.map +1 -0
- package/dist/analyst/findings-store.d.ts +75 -0
- package/dist/analyst/findings-store.d.ts.map +1 -0
- package/dist/analyst/index.d.ts +28 -256
- package/dist/analyst/index.d.ts.map +1 -0
- package/dist/analyst/index.js +6 -66
- package/dist/analyst/index.js.map +1 -1
- package/dist/{kind-factory-5b7xXXOr.d.ts → analyst/kind-factory.d.ts} +11 -63
- package/dist/analyst/kind-factory.d.ts.map +1 -0
- package/dist/analyst/kinds/failure-mode.d.ts +19 -0
- package/dist/analyst/kinds/failure-mode.d.ts.map +1 -0
- package/dist/analyst/kinds/improvement.d.ts +23 -0
- package/dist/analyst/kinds/improvement.d.ts.map +1 -0
- package/dist/analyst/kinds/index.d.ts +22 -0
- package/dist/analyst/kinds/index.d.ts.map +1 -0
- package/dist/analyst/kinds/kinds.test.d.ts +2 -0
- package/dist/analyst/kinds/kinds.test.d.ts.map +1 -0
- package/dist/analyst/kinds/knowledge-gap.d.ts +28 -0
- package/dist/analyst/kinds/knowledge-gap.d.ts.map +1 -0
- package/dist/analyst/kinds/knowledge-poisoning.d.ts +22 -0
- package/dist/analyst/kinds/knowledge-poisoning.d.ts.map +1 -0
- package/dist/analyst/kinds/skill-usage.d.ts +84 -0
- package/dist/analyst/kinds/skill-usage.d.ts.map +1 -0
- package/dist/analyst/kinds/skill-usage.test.d.ts +2 -0
- package/dist/analyst/kinds/skill-usage.test.d.ts.map +1 -0
- package/dist/analyst/parse-tolerant.d.ts +26 -0
- package/dist/analyst/parse-tolerant.d.ts.map +1 -0
- package/dist/analyst/parse-tolerant.test.d.ts +2 -0
- package/dist/analyst/parse-tolerant.test.d.ts.map +1 -0
- package/dist/analyst/registry.budget.test.d.ts +2 -0
- package/dist/analyst/registry.budget.test.d.ts.map +1 -0
- package/dist/{default-registry-zoGHUQEH.d.ts → analyst/registry.d.ts} +8 -37
- package/dist/analyst/registry.d.ts.map +1 -0
- package/dist/analyst/steer-firewall.d.ts +35 -0
- package/dist/analyst/steer-firewall.d.ts.map +1 -0
- package/dist/analyst/steer-firewall.test.d.ts +2 -0
- package/dist/analyst/steer-firewall.test.d.ts.map +1 -0
- package/dist/analyst/structure-findings.d.ts +37 -0
- package/dist/analyst/structure-findings.d.ts.map +1 -0
- package/dist/analyst/structure-findings.test.d.ts +2 -0
- package/dist/analyst/structure-findings.test.d.ts.map +1 -0
- package/dist/analyst/tool-groups.d.ts +34 -0
- package/dist/analyst/tool-groups.d.ts.map +1 -0
- package/dist/{types-2VVIL04s.d.ts → analyst/types.d.ts} +19 -112
- package/dist/analyst/types.d.ts.map +1 -0
- package/dist/anti-slop.d.ts +59 -0
- package/dist/anti-slop.d.ts.map +1 -0
- package/dist/artifact-validator.d.ts +74 -0
- package/dist/artifact-validator.d.ts.map +1 -0
- package/dist/attestation.d.ts +63 -0
- package/dist/attestation.d.ts.map +1 -0
- package/dist/attestation.test.d.ts +2 -0
- package/dist/attestation.test.d.ts.map +1 -0
- package/dist/authenticity/index.d.ts +15 -16
- package/dist/authenticity/index.d.ts.map +1 -0
- package/dist/authenticity/index.test.d.ts +2 -0
- package/dist/authenticity/index.test.d.ts.map +1 -0
- package/dist/auto-pr.d.ts +120 -0
- package/dist/auto-pr.d.ts.map +1 -0
- package/dist/{baseline-DE36-Np7.d.ts → baseline.d.ts} +8 -44
- package/dist/baseline.d.ts.map +1 -0
- package/dist/behavior-dsl.d.ts +73 -0
- package/dist/behavior-dsl.d.ts.map +1 -0
- package/dist/belief-state/calibration.d.ts +10 -0
- package/dist/belief-state/calibration.d.ts.map +1 -0
- package/dist/belief-state/calibration.test.d.ts +2 -0
- package/dist/belief-state/calibration.test.d.ts.map +1 -0
- package/dist/belief-state/code-agent-corpus.d.ts +66 -0
- package/dist/belief-state/code-agent-corpus.d.ts.map +1 -0
- package/dist/belief-state/code-agent-corpus.test.d.ts +2 -0
- package/dist/belief-state/code-agent-corpus.test.d.ts.map +1 -0
- package/dist/belief-state/code-agent-evidence.d.ts +22 -0
- package/dist/belief-state/code-agent-evidence.d.ts.map +1 -0
- package/dist/belief-state/code-agent-evidence.test.d.ts +2 -0
- package/dist/belief-state/code-agent-evidence.test.d.ts.map +1 -0
- package/dist/belief-state/extract.d.ts +7 -0
- package/dist/belief-state/extract.d.ts.map +1 -0
- package/dist/belief-state/extract.test.d.ts +2 -0
- package/dist/belief-state/extract.test.d.ts.map +1 -0
- package/dist/belief-state/index.d.ts +14 -604
- package/dist/belief-state/index.d.ts.map +1 -0
- package/dist/belief-state/index.js +1 -1
- package/dist/belief-state/ope.d.ts +17 -0
- package/dist/belief-state/ope.d.ts.map +1 -0
- package/dist/belief-state/ope.test.d.ts +2 -0
- package/dist/belief-state/ope.test.d.ts.map +1 -0
- package/dist/belief-state/phase0-measurement.d.ts +55 -0
- package/dist/belief-state/phase0-measurement.d.ts.map +1 -0
- package/dist/belief-state/report.d.ts +17 -0
- package/dist/belief-state/report.d.ts.map +1 -0
- package/dist/belief-state/report.test.d.ts +2 -0
- package/dist/belief-state/report.test.d.ts.map +1 -0
- package/dist/belief-state/research-evidence.d.ts +23 -0
- package/dist/belief-state/research-evidence.d.ts.map +1 -0
- package/dist/belief-state/research-evidence.test.d.ts +2 -0
- package/dist/belief-state/research-evidence.test.d.ts.map +1 -0
- package/dist/belief-state/runtime-benchmark-corpus.d.ts +32 -0
- package/dist/belief-state/runtime-benchmark-corpus.d.ts.map +1 -0
- package/dist/belief-state/runtime-hooks.d.ts +87 -0
- package/dist/belief-state/runtime-hooks.d.ts.map +1 -0
- package/dist/belief-state/runtime-hooks.test.d.ts +2 -0
- package/dist/belief-state/runtime-hooks.test.d.ts.map +1 -0
- package/dist/belief-state/selective.d.ts +15 -0
- package/dist/belief-state/selective.d.ts.map +1 -0
- package/dist/belief-state/selective.test.d.ts +2 -0
- package/dist/belief-state/selective.test.d.ts.map +1 -0
- package/dist/belief-state/shadow-probe.d.ts +80 -0
- package/dist/belief-state/shadow-probe.d.ts.map +1 -0
- package/dist/belief-state/shadow-probe.test.d.ts +2 -0
- package/dist/belief-state/shadow-probe.test.d.ts.map +1 -0
- package/dist/belief-state/types.d.ts +195 -0
- package/dist/belief-state/types.d.ts.map +1 -0
- package/dist/belief-state/types.test.d.ts +2 -0
- package/dist/belief-state/types.test.d.ts.map +1 -0
- package/dist/benchmark.d.ts +14 -0
- package/dist/benchmark.d.ts.map +1 -0
- package/dist/benchmarks/index.d.ts +23 -4
- package/dist/benchmarks/index.d.ts.map +1 -0
- package/dist/benchmarks/routing/dataset.d.ts +34 -0
- package/dist/benchmarks/routing/dataset.d.ts.map +1 -0
- package/dist/benchmarks/routing/index.d.ts +34 -0
- package/dist/benchmarks/routing/index.d.ts.map +1 -0
- package/dist/benchmarks/types.d.ts +49 -0
- package/dist/benchmarks/types.d.ts.map +1 -0
- package/dist/bisector.d.ts +81 -0
- package/dist/bisector.d.ts.map +1 -0
- package/dist/budget-guard.d.ts +31 -0
- package/dist/budget-guard.d.ts.map +1 -0
- package/dist/builder-eval/builder-session.d.ts +111 -0
- package/dist/builder-eval/builder-session.d.ts.map +1 -0
- package/dist/builder-eval/correlation.d.ts +32 -0
- package/dist/builder-eval/correlation.d.ts.map +1 -0
- package/dist/builder-eval/index.d.ts +5 -250
- package/dist/builder-eval/index.d.ts.map +1 -0
- package/dist/builder-eval/index.js +2 -2
- package/dist/builder-eval/project-registry.d.ts +51 -0
- package/dist/builder-eval/project-registry.d.ts.map +1 -0
- package/dist/builder-eval/three-layer-eval.d.ts +55 -0
- package/dist/builder-eval/three-layer-eval.d.ts.map +1 -0
- package/dist/campaign/analyst-surface.d.ts +108 -0
- package/dist/campaign/analyst-surface.d.ts.map +1 -0
- package/dist/campaign/analyst-surface.test.d.ts +2 -0
- package/dist/campaign/analyst-surface.test.d.ts.map +1 -0
- package/dist/campaign/auto-pr.d.ts +46 -0
- package/dist/campaign/auto-pr.d.ts.map +1 -0
- package/dist/campaign/distillation/agreement-judge.d.ts +69 -0
- package/dist/campaign/distillation/agreement-judge.d.ts.map +1 -0
- package/dist/campaign/distillation/cli.d.ts +35 -0
- package/dist/campaign/distillation/cli.d.ts.map +1 -0
- package/dist/campaign/distillation/distillation.test.d.ts +2 -0
- package/dist/campaign/distillation/distillation.test.d.ts.map +1 -0
- package/dist/campaign/distillation/gold-scenarios.d.ts +54 -0
- package/dist/campaign/distillation/gold-scenarios.d.ts.map +1 -0
- package/dist/campaign/distillation/run-distillation.d.ts +119 -0
- package/dist/campaign/distillation/run-distillation.d.ts.map +1 -0
- package/dist/campaign/gates/compose.d.ts +12 -0
- package/dist/campaign/gates/compose.d.ts.map +1 -0
- package/dist/campaign/gates/default-production-gate.d.ts +58 -0
- package/dist/campaign/gates/default-production-gate.d.ts.map +1 -0
- package/dist/campaign/gates/heldout-gate.d.ts +12 -0
- package/dist/campaign/gates/heldout-gate.d.ts.map +1 -0
- package/dist/campaign/gates/promotion-policy.d.ts +125 -0
- package/dist/campaign/gates/promotion-policy.d.ts.map +1 -0
- package/dist/campaign/gates/promotion-policy.test.d.ts +2 -0
- package/dist/campaign/gates/promotion-policy.test.d.ts.map +1 -0
- package/dist/campaign/gates/sequential.d.ts +146 -0
- package/dist/campaign/gates/sequential.d.ts.map +1 -0
- package/dist/campaign/gates/sequential.test.d.ts +2 -0
- package/dist/campaign/gates/sequential.test.d.ts.map +1 -0
- package/dist/campaign/gates/statistical-heldout.d.ts +99 -0
- package/dist/campaign/gates/statistical-heldout.d.ts.map +1 -0
- package/dist/campaign/gates/statistical-heldout.test.d.ts +2 -0
- package/dist/campaign/gates/statistical-heldout.test.d.ts.map +1 -0
- package/dist/campaign/index.d.ts +38 -1341
- package/dist/campaign/index.d.ts.map +1 -0
- package/dist/campaign/index.js +1865 -1318
- package/dist/campaign/index.js.map +1 -1
- package/dist/campaign/labeled-store/fs-adapter.d.ts +59 -0
- package/dist/campaign/labeled-store/fs-adapter.d.ts.map +1 -0
- package/dist/campaign/presets/compare-proposers.d.ts +146 -0
- package/dist/campaign/presets/compare-proposers.d.ts.map +1 -0
- package/dist/campaign/presets/playback.d.ts +120 -0
- package/dist/campaign/presets/playback.d.ts.map +1 -0
- package/dist/campaign/presets/playback.test.d.ts +2 -0
- package/dist/campaign/presets/playback.test.d.ts.map +1 -0
- package/dist/campaign/presets/run-eval.d.ts +14 -0
- package/dist/campaign/presets/run-eval.d.ts.map +1 -0
- package/dist/campaign/presets/run-improvement-loop.d.ts +63 -0
- package/dist/campaign/presets/run-improvement-loop.d.ts.map +1 -0
- package/dist/campaign/presets/run-improvement-loop.test.d.ts +2 -0
- package/dist/campaign/presets/run-improvement-loop.test.d.ts.map +1 -0
- package/dist/campaign/presets/run-optimization.d.ts +92 -0
- package/dist/campaign/presets/run-optimization.d.ts.map +1 -0
- package/dist/campaign/presets/run-profile-matrix.d.ts +151 -0
- package/dist/campaign/presets/run-profile-matrix.d.ts.map +1 -0
- package/dist/campaign/presets/run-skill-opt.d.ts +96 -0
- package/dist/campaign/presets/run-skill-opt.d.ts.map +1 -0
- package/dist/campaign/proposers/_findings-text.d.ts +22 -0
- package/dist/campaign/proposers/_findings-text.d.ts.map +1 -0
- package/dist/campaign/proposers/ace.d.ts +33 -0
- package/dist/campaign/proposers/ace.d.ts.map +1 -0
- package/dist/campaign/proposers/ace.test.d.ts +2 -0
- package/dist/campaign/proposers/ace.test.d.ts.map +1 -0
- package/dist/campaign/proposers/analysis-edit.d.ts +32 -0
- package/dist/campaign/proposers/analysis-edit.d.ts.map +1 -0
- package/dist/campaign/proposers/evolutionary.d.ts +20 -0
- package/dist/campaign/proposers/evolutionary.d.ts.map +1 -0
- package/dist/campaign/proposers/fapo.d.ts +120 -0
- package/dist/campaign/proposers/fapo.d.ts.map +1 -0
- package/dist/campaign/proposers/gepa.d.ts +86 -0
- package/dist/campaign/proposers/gepa.d.ts.map +1 -0
- package/dist/campaign/proposers/halo.d.ts +44 -0
- package/dist/campaign/proposers/halo.d.ts.map +1 -0
- package/dist/campaign/proposers/halo.test.d.ts +2 -0
- package/dist/campaign/proposers/halo.test.d.ts.map +1 -0
- package/dist/campaign/proposers/memory.d.ts +47 -0
- package/dist/campaign/proposers/memory.d.ts.map +1 -0
- package/dist/campaign/proposers/memory.test.d.ts +2 -0
- package/dist/campaign/proposers/memory.test.d.ts.map +1 -0
- package/dist/campaign/proposers/skill-opt.d.ts +88 -0
- package/dist/campaign/proposers/skill-opt.d.ts.map +1 -0
- package/dist/campaign/proposers/trace-analyst.d.ts +48 -0
- package/dist/campaign/proposers/trace-analyst.d.ts.map +1 -0
- package/dist/campaign/proposers/trace-analyst.test.d.ts +2 -0
- package/dist/campaign/proposers/trace-analyst.test.d.ts.map +1 -0
- package/dist/campaign/provenance.d.ts +185 -0
- package/dist/campaign/provenance.d.ts.map +1 -0
- package/dist/campaign/run-campaign.d.ts +90 -0
- package/dist/campaign/run-campaign.d.ts.map +1 -0
- package/dist/campaign/score-utils.d.ts +26 -0
- package/dist/campaign/score-utils.d.ts.map +1 -0
- package/dist/campaign/skill-patch.d.ts +62 -0
- package/dist/campaign/skill-patch.d.ts.map +1 -0
- package/dist/campaign/storage.d.ts +38 -0
- package/dist/campaign/storage.d.ts.map +1 -0
- package/dist/{types-BU-7W85F.d.ts → campaign/types.d.ts} +98 -99
- package/dist/campaign/types.d.ts.map +1 -0
- package/dist/campaign/worktree/index.d.ts +53 -0
- package/dist/campaign/worktree/index.d.ts.map +1 -0
- package/dist/canary.d.ts +101 -0
- package/dist/canary.d.ts.map +1 -0
- package/dist/causal-attribution.d.ts +45 -0
- package/dist/causal-attribution.d.ts.map +1 -0
- package/dist/{chunk-TWS7AZEY.js → chunk-2T4EZACH.js} +2 -2
- package/dist/chunk-2T4EZACH.js.map +1 -0
- package/dist/{chunk-LMZQ2Z4U.js → chunk-56GC6VYO.js} +126 -1
- package/dist/chunk-56GC6VYO.js.map +1 -0
- package/dist/{chunk-QG2OVF2D.js → chunk-77T4STFI.js} +154 -112
- package/dist/chunk-77T4STFI.js.map +1 -0
- package/dist/{chunk-ZFIBGEOL.js → chunk-7QTQKIDD.js} +179 -178
- package/dist/chunk-7QTQKIDD.js.map +1 -0
- package/dist/{chunk-UMMZHCPB.js → chunk-AQ5WQAIV.js} +26 -9
- package/dist/chunk-AQ5WQAIV.js.map +1 -0
- package/dist/{chunk-CVVHBFGN.js → chunk-CWNP4DV4.js} +53 -5
- package/dist/chunk-CWNP4DV4.js.map +1 -0
- package/dist/{chunk-QS3RBQPI.js → chunk-D5KBVULY.js} +2 -2
- package/dist/chunk-DJWX3GVS.js +81 -0
- package/dist/chunk-DJWX3GVS.js.map +1 -0
- package/dist/{chunk-S6OZEZQK.js → chunk-HMA63UEO.js} +37 -9
- package/dist/{chunk-S6OZEZQK.js.map → chunk-HMA63UEO.js.map} +1 -1
- package/dist/{chunk-UHMJT4T7.js → chunk-IZCEK2HR.js} +2 -2
- package/dist/{chunk-6SOJM3VR.js → chunk-KKWJD5E6.js} +21 -21
- package/dist/chunk-KKWJD5E6.js.map +1 -0
- package/dist/{chunk-KWRRMR3J.js → chunk-LO6IOIJ2.js} +10 -10
- package/dist/chunk-LO6IOIJ2.js.map +1 -0
- package/dist/{chunk-L3JOU6XM.js → chunk-NACM5RS7.js} +3 -3
- package/dist/{chunk-E4GH6USR.js → chunk-NZEQVRH5.js} +2 -2
- package/dist/chunk-NZEQVRH5.js.map +1 -0
- package/dist/{chunk-XY4DDNEG.js → chunk-PSWWQXHF.js} +14 -89
- package/dist/chunk-PSWWQXHF.js.map +1 -0
- package/dist/{chunk-4FBZZIYD.js → chunk-QTMYW64F.js} +2 -2
- package/dist/{chunk-D3V5B42D.js → chunk-S4SYLDFX.js} +7 -4
- package/dist/chunk-S4SYLDFX.js.map +1 -0
- package/dist/{chunk-47X6LRCE.js → chunk-UFSG7ACU.js} +10 -7
- package/dist/chunk-UFSG7ACU.js.map +1 -0
- package/dist/{chunk-CY6U5S3X.js → chunk-URWVKRCS.js} +2 -2
- package/dist/{chunk-GSH6QNNS.js → chunk-X74V6ESX.js} +102 -79
- package/dist/chunk-X74V6ESX.js.map +1 -0
- package/dist/{chunk-QAY5UIJO.js → chunk-YBIGNSCZ.js} +178 -59
- package/dist/chunk-YBIGNSCZ.js.map +1 -0
- package/dist/{chunk-Y47J2LJ3.js → chunk-Z6L6YSU6.js} +5 -5
- package/dist/{chunk-T375SUOZ.js → chunk-ZOPLCJSY.js} +86 -31
- package/dist/chunk-ZOPLCJSY.js.map +1 -0
- package/dist/ci-gate.d.ts +44 -0
- package/dist/ci-gate.d.ts.map +1 -0
- package/dist/cli.d.ts +2 -0
- package/dist/cli.d.ts.map +1 -0
- package/dist/cli.js +2 -2
- package/dist/client.d.ts +77 -0
- package/dist/client.d.ts.map +1 -0
- package/dist/client.test.d.ts +2 -0
- package/dist/client.test.d.ts.map +1 -0
- package/dist/command-runner.d.ts +74 -0
- package/dist/command-runner.d.ts.map +1 -0
- package/dist/command-runner.test.d.ts +2 -0
- package/dist/command-runner.test.d.ts.map +1 -0
- package/dist/completion-verifier.d.ts +147 -0
- package/dist/completion-verifier.d.ts.map +1 -0
- package/dist/completion-verifier.test.d.ts +9 -0
- package/dist/completion-verifier.test.d.ts.map +1 -0
- package/dist/concurrency.d.ts +23 -0
- package/dist/concurrency.d.ts.map +1 -0
- package/dist/contamination-guard.d.ts +81 -0
- package/dist/contamination-guard.d.ts.map +1 -0
- package/dist/{analyze-runs-DwCEkpO_.d.ts → contract/analyze-runs.d.ts} +9 -10
- package/dist/contract/analyze-runs.d.ts.map +1 -0
- package/dist/contract/define-agent-eval.d.ts +52 -0
- package/dist/contract/define-agent-eval.d.ts.map +1 -0
- package/dist/contract/diff.d.ts +114 -0
- package/dist/contract/diff.d.ts.map +1 -0
- package/dist/contract/index.d.ts +106 -640
- package/dist/contract/index.d.ts.map +1 -0
- package/dist/contract/index.js +131 -21
- package/dist/contract/index.js.map +1 -1
- package/dist/{insight-report-BBwvOh6x.d.ts → contract/insight-report.d.ts} +16 -19
- package/dist/contract/insight-report.d.ts.map +1 -0
- package/dist/contract/insight-types-fwd.d.ts +7 -0
- package/dist/contract/insight-types-fwd.d.ts.map +1 -0
- package/dist/contract/intake/agent-trace.d.ts +97 -0
- package/dist/contract/intake/agent-trace.d.ts.map +1 -0
- package/dist/{code-agent-session-BO8nCnv3.d.ts → contract/intake/code-agent-session.d.ts} +15 -17
- package/dist/contract/intake/code-agent-session.d.ts.map +1 -0
- package/dist/contract/intake/feedback-table.d.ts +87 -0
- package/dist/contract/intake/feedback-table.d.ts.map +1 -0
- package/dist/contract/intake/index.d.ts +22 -0
- package/dist/contract/intake/index.d.ts.map +1 -0
- package/dist/contract/intake/otel-spans.d.ts +33 -0
- package/dist/contract/intake/otel-spans.d.ts.map +1 -0
- package/dist/contract/self-improve.d.ts +284 -0
- package/dist/contract/self-improve.d.ts.map +1 -0
- package/dist/{control-runtime-DuFBYg7A.d.ts → control-runtime.d.ts} +23 -26
- package/dist/control-runtime.d.ts.map +1 -0
- package/dist/control-runtime.test.d.ts +2 -0
- package/dist/control-runtime.test.d.ts.map +1 -0
- package/dist/control.d.ts +11 -9
- package/dist/control.d.ts.map +1 -0
- package/dist/control.js +2 -2
- package/dist/convergence.d.ts +29 -0
- package/dist/convergence.d.ts.map +1 -0
- package/dist/{cost-ledger-DuSqlw5B.d.ts → cost-ledger.d.ts} +10 -11
- package/dist/cost-ledger.d.ts.map +1 -0
- package/dist/cost-ledger.test.d.ts +2 -0
- package/dist/cost-ledger.test.d.ts.map +1 -0
- package/dist/cost-report.d.ts +43 -0
- package/dist/cost-report.d.ts.map +1 -0
- package/dist/cost-report.test.d.ts +2 -0
- package/dist/cost-report.test.d.ts.map +1 -0
- package/dist/cost-tracker.d.ts +76 -0
- package/dist/cost-tracker.d.ts.map +1 -0
- package/dist/{counterfactual-Dwibr5IW.d.ts → counterfactual.d.ts} +12 -13
- package/dist/counterfactual.d.ts.map +1 -0
- package/dist/cross-trace-diff.d.ts +56 -0
- package/dist/cross-trace-diff.d.ts.map +1 -0
- package/dist/{dataset-BbGkaN2I.d.ts → dataset.d.ts} +11 -14
- package/dist/dataset.d.ts.map +1 -0
- package/dist/deploy-gate-layer.d.ts +125 -0
- package/dist/deploy-gate-layer.d.ts.map +1 -0
- package/dist/deploy-gate-layer.test.d.ts +2 -0
- package/dist/deploy-gate-layer.test.d.ts.map +1 -0
- package/dist/description-length-gate.d.ts +119 -0
- package/dist/description-length-gate.d.ts.map +1 -0
- package/dist/detectors/edge.test.d.ts +2 -0
- package/dist/detectors/edge.test.d.ts.map +1 -0
- package/dist/detectors/index.d.ts +81 -0
- package/dist/detectors/index.d.ts.map +1 -0
- package/dist/detectors/index.test.d.ts +2 -0
- package/dist/detectors/index.test.d.ts.map +1 -0
- package/dist/diagnose/causal-sweep.d.ts +100 -0
- package/dist/diagnose/causal-sweep.d.ts.map +1 -0
- package/dist/diagnose/index.d.ts +36 -0
- package/dist/diagnose/index.d.ts.map +1 -0
- package/dist/diagnose/remediation.d.ts +68 -0
- package/dist/diagnose/remediation.d.ts.map +1 -0
- package/dist/diagnose/repair.d.ts +77 -0
- package/dist/diagnose/repair.d.ts.map +1 -0
- package/dist/diagnose.d.ts +1 -251
- package/dist/diagnose.js +2 -2
- package/dist/discover-personas.d.ts +35 -0
- package/dist/discover-personas.d.ts.map +1 -0
- package/dist/driver.d.ts +95 -0
- package/dist/driver.d.ts.map +1 -0
- package/dist/driver.test.d.ts +8 -0
- package/dist/driver.test.d.ts.map +1 -0
- package/dist/dual-agent-bench.d.ts +81 -0
- package/dist/dual-agent-bench.d.ts.map +1 -0
- package/dist/error-count-extractor.d.ts +47 -0
- package/dist/error-count-extractor.d.ts.map +1 -0
- package/dist/error-count-extractor.test.d.ts +2 -0
- package/dist/error-count-extractor.test.d.ts.map +1 -0
- package/dist/{errors-CzMUYo7b.d.ts → errors.d.ts} +10 -11
- package/dist/errors.d.ts.map +1 -0
- package/dist/{researcher-DE6Gpnb4.d.ts → eval-campaign.d.ts} +34 -156
- package/dist/eval-campaign.d.ts.map +1 -0
- package/dist/eval-campaign.test.d.ts +2 -0
- package/dist/eval-campaign.test.d.ts.map +1 -0
- package/dist/eval-tools.d.ts +55 -0
- package/dist/eval-tools.d.ts.map +1 -0
- package/dist/eval-trace-store.d.ts +107 -0
- package/dist/eval-trace-store.d.ts.map +1 -0
- package/dist/eval-trace-store.test.d.ts +2 -0
- package/dist/eval-trace-store.test.d.ts.map +1 -0
- package/dist/executor.d.ts +38 -0
- package/dist/executor.d.ts.map +1 -0
- package/dist/executor.test.d.ts +10 -0
- package/dist/executor.test.d.ts.map +1 -0
- package/dist/experiment-tracker.d.ts +178 -0
- package/dist/experiment-tracker.d.ts.map +1 -0
- package/dist/experiment-tracker.test.d.ts +2 -0
- package/dist/experiment-tracker.test.d.ts.map +1 -0
- package/dist/failure-taxonomy.d.ts +38 -0
- package/dist/failure-taxonomy.d.ts.map +1 -0
- package/dist/{feedback-trajectory-D9OVLrg9.d.ts → feedback-trajectory.d.ts} +36 -38
- package/dist/feedback-trajectory.d.ts.map +1 -0
- package/dist/feedback-trajectory.test.d.ts +2 -0
- package/dist/feedback-trajectory.test.d.ts.map +1 -0
- package/dist/flow-layer.d.ts +90 -0
- package/dist/flow-layer.d.ts.map +1 -0
- package/dist/flow-layer.test.d.ts +2 -0
- package/dist/flow-layer.test.d.ts.map +1 -0
- package/dist/fuzz/capsule.d.ts +46 -0
- package/dist/fuzz/capsule.d.ts.map +1 -0
- package/dist/fuzz/cube.d.ts +36 -0
- package/dist/fuzz/cube.d.ts.map +1 -0
- package/dist/fuzz/explorer-cost.test.d.ts +2 -0
- package/dist/fuzz/explorer-cost.test.d.ts.map +1 -0
- package/dist/fuzz/explorer.d.ts +64 -0
- package/dist/fuzz/explorer.d.ts.map +1 -0
- package/dist/fuzz/fuzz-agent.d.ts +16 -0
- package/dist/fuzz/fuzz-agent.d.ts.map +1 -0
- package/dist/fuzz/fuzz-agent.test.d.ts +2 -0
- package/dist/fuzz/fuzz-agent.test.d.ts.map +1 -0
- package/dist/fuzz/gates.d.ts +33 -0
- package/dist/fuzz/gates.d.ts.map +1 -0
- package/dist/fuzz/index.d.ts +26 -0
- package/dist/fuzz/index.d.ts.map +1 -0
- package/dist/fuzz/policies.d.ts +28 -0
- package/dist/fuzz/policies.d.ts.map +1 -0
- package/dist/fuzz/tools.d.ts +20 -0
- package/dist/fuzz/tools.d.ts.map +1 -0
- package/dist/fuzz/types.d.ts +307 -0
- package/dist/fuzz/types.d.ts.map +1 -0
- package/dist/fuzz.d.ts +1 -547
- package/dist/golden-matcher.d.ts +71 -0
- package/dist/golden-matcher.d.ts.map +1 -0
- package/dist/governance/eu-ai-act.d.ts +37 -0
- package/dist/governance/eu-ai-act.d.ts.map +1 -0
- package/dist/governance/index.d.ts +5 -135
- package/dist/governance/index.d.ts.map +1 -0
- package/dist/governance/nist-ai-rmf.d.ts +15 -0
- package/dist/governance/nist-ai-rmf.d.ts.map +1 -0
- package/dist/governance/soc2.d.ts +12 -0
- package/dist/governance/soc2.d.ts.map +1 -0
- package/dist/governance/types.d.ts +66 -0
- package/dist/governance/types.d.ts.map +1 -0
- package/dist/harness-optimizer.d.ts +82 -0
- package/dist/harness-optimizer.d.ts.map +1 -0
- package/dist/held-out-gate.d.ts +135 -0
- package/dist/held-out-gate.d.ts.map +1 -0
- package/dist/hosted/client.d.ts +73 -0
- package/dist/hosted/client.d.ts.map +1 -0
- package/dist/hosted/from-env.test.d.ts +8 -0
- package/dist/hosted/from-env.test.d.ts.map +1 -0
- package/dist/hosted/index.d.ts +10 -238
- package/dist/hosted/index.d.ts.map +1 -0
- package/dist/hosted/types.d.ts +159 -0
- package/dist/hosted/types.d.ts.map +1 -0
- package/dist/index.d.ts +277 -5643
- package/dist/index.d.ts.map +1 -0
- package/dist/index.js +317 -108
- package/dist/index.js.map +1 -1
- package/dist/integrity/backend-integrity.d.ts +71 -0
- package/dist/integrity/backend-integrity.d.ts.map +1 -0
- package/dist/integrity/preflight.d.ts +72 -0
- package/dist/integrity/preflight.d.ts.map +1 -0
- package/dist/integrity/preflight.test.d.ts +2 -0
- package/dist/integrity/preflight.test.d.ts.map +1 -0
- package/dist/integrity/single-backend.d.ts +67 -0
- package/dist/integrity/single-backend.d.ts.map +1 -0
- package/dist/intent-match-judge.d.ts +69 -0
- package/dist/intent-match-judge.d.ts.map +1 -0
- package/dist/intent-match-judge.test.d.ts +2 -0
- package/dist/intent-match-judge.test.d.ts.map +1 -0
- package/dist/{judge-calibration-0p2QcWNE.d.ts → judge-calibration.d.ts} +16 -17
- package/dist/judge-calibration.d.ts.map +1 -0
- package/dist/judge-ensemble.d.ts +66 -0
- package/dist/judge-ensemble.d.ts.map +1 -0
- package/dist/judge-ensemble.test.d.ts +8 -0
- package/dist/judge-ensemble.test.d.ts.map +1 -0
- package/dist/judge-families.d.ts +38 -0
- package/dist/judge-families.d.ts.map +1 -0
- package/dist/judge-panel.d.ts +65 -0
- package/dist/judge-panel.d.ts.map +1 -0
- package/dist/judge-retry.d.ts +70 -0
- package/dist/judge-retry.d.ts.map +1 -0
- package/dist/judge-runner.d.ts +36 -0
- package/dist/judge-runner.d.ts.map +1 -0
- package/dist/judge-runner.test.d.ts +2 -0
- package/dist/judge-runner.test.d.ts.map +1 -0
- package/dist/judges.d.ts +74 -0
- package/dist/judges.d.ts.map +1 -0
- package/dist/keyword-coverage-judge.d.ts +89 -0
- package/dist/keyword-coverage-judge.d.ts.map +1 -0
- package/dist/keyword-coverage-judge.test.d.ts +2 -0
- package/dist/keyword-coverage-judge.test.d.ts.map +1 -0
- package/dist/knowledge/index.d.ts +3 -103
- package/dist/knowledge/index.d.ts.map +1 -0
- package/dist/knowledge/readiness.d.ts +26 -0
- package/dist/knowledge/readiness.d.ts.map +1 -0
- package/dist/knowledge/types.d.ts +75 -0
- package/dist/knowledge/types.d.ts.map +1 -0
- package/dist/live-proof.d.ts +62 -0
- package/dist/live-proof.d.ts.map +1 -0
- package/dist/{llm-client-BeEcAokY.d.ts → llm-client.d.ts} +52 -23
- package/dist/llm-client.d.ts.map +1 -0
- package/dist/llm-client.test.d.ts +2 -0
- package/dist/llm-client.test.d.ts.map +1 -0
- package/dist/locked-jsonl-appender.d.ts +19 -0
- package/dist/locked-jsonl-appender.d.ts.map +1 -0
- package/dist/matrix/aggregation.d.ts +16 -0
- package/dist/matrix/aggregation.d.ts.map +1 -0
- package/dist/matrix/index.d.ts +12 -30
- package/dist/matrix/index.d.ts.map +1 -0
- package/dist/matrix/runner.d.ts +15 -0
- package/dist/matrix/runner.d.ts.map +1 -0
- package/dist/{types-mn5Aqk7x.d.ts → matrix/types.d.ts} +11 -15
- package/dist/matrix/types.d.ts.map +1 -0
- package/dist/meta-eval/calibration.d.ts +47 -0
- package/dist/meta-eval/calibration.d.ts.map +1 -0
- package/dist/meta-eval/correlation-study.d.ts +53 -0
- package/dist/meta-eval/correlation-study.d.ts.map +1 -0
- package/dist/meta-eval/index.d.ts +6 -181
- package/dist/meta-eval/index.d.ts.map +1 -0
- package/dist/meta-eval/index.js +1 -1
- package/dist/{outcome-store-rnXLEqSn.d.ts → meta-eval/outcome-store.d.ts} +7 -8
- package/dist/meta-eval/outcome-store.d.ts.map +1 -0
- package/dist/{rubric-predictive-validity-Cy_W-hWZ.d.ts → meta-eval/rubric-predictive-validity.d.ts} +8 -11
- package/dist/meta-eval/rubric-predictive-validity.d.ts.map +1 -0
- package/dist/meta-eval/sentinel.d.ts +169 -0
- package/dist/meta-eval/sentinel.d.ts.map +1 -0
- package/dist/metrics.d.ts +63 -0
- package/dist/metrics.d.ts.map +1 -0
- package/dist/model-seats.d.ts +71 -0
- package/dist/model-seats.d.ts.map +1 -0
- package/dist/model-seats.test.d.ts +2 -0
- package/dist/model-seats.test.d.ts.map +1 -0
- package/dist/muffled-gate-scanner.d.ts +102 -0
- package/dist/muffled-gate-scanner.d.ts.map +1 -0
- package/dist/{multi-layer-verifier-DUZXrPDA.d.ts → multi-layer-verifier.d.ts} +12 -15
- package/dist/multi-layer-verifier.d.ts.map +1 -0
- package/dist/multi-layer-verifier.test.d.ts +2 -0
- package/dist/multi-layer-verifier.test.d.ts.map +1 -0
- package/dist/multi-toolchain-layer.d.ts +80 -0
- package/dist/multi-toolchain-layer.d.ts.map +1 -0
- package/dist/multi-toolchain-layer.test.d.ts +2 -0
- package/dist/multi-toolchain-layer.test.d.ts.map +1 -0
- package/dist/multishot/default-tools.d.ts +34 -0
- package/dist/multishot/default-tools.d.ts.map +1 -0
- package/dist/multishot/index.d.ts +7 -290
- package/dist/multishot/index.d.ts.map +1 -0
- package/dist/multishot/index.js.map +1 -1
- package/dist/multishot/judges.d.ts +32 -0
- package/dist/multishot/judges.d.ts.map +1 -0
- package/dist/multishot/matrix.d.ts +107 -0
- package/dist/multishot/matrix.d.ts.map +1 -0
- package/dist/multishot/multishot.d.ts +23 -0
- package/dist/multishot/multishot.d.ts.map +1 -0
- package/dist/multishot/router.d.ts +37 -0
- package/dist/multishot/router.d.ts.map +1 -0
- package/dist/multishot/types.d.ts +60 -0
- package/dist/multishot/types.d.ts.map +1 -0
- package/dist/observability.d.ts +71 -0
- package/dist/observability.d.ts.map +1 -0
- package/dist/openapi.json +1 -1
- package/dist/oracle.d.ts +55 -0
- package/dist/oracle.d.ts.map +1 -0
- package/dist/orthogonality.d.ts +35 -0
- package/dist/orthogonality.d.ts.map +1 -0
- package/dist/otel-pipeline.d.ts +31 -0
- package/dist/otel-pipeline.d.ts.map +1 -0
- package/dist/paraphrase.d.ts +107 -0
- package/dist/paraphrase.d.ts.map +1 -0
- package/dist/{pareto-E-pembql.d.ts → pareto.d.ts} +9 -10
- package/dist/pareto.d.ts.map +1 -0
- package/dist/partition-held-out.d.ts +70 -0
- package/dist/partition-held-out.d.ts.map +1 -0
- package/dist/partition-held-out.test.d.ts +2 -0
- package/dist/partition-held-out.test.d.ts.map +1 -0
- package/dist/perf/index.d.ts +13 -119
- package/dist/perf/index.d.ts.map +1 -0
- package/dist/perf/integrity.d.ts +30 -0
- package/dist/perf/integrity.d.ts.map +1 -0
- package/dist/perf/journey.d.ts +45 -0
- package/dist/perf/journey.d.ts.map +1 -0
- package/dist/perf/ratchet.d.ts +47 -0
- package/dist/perf/ratchet.d.ts.map +1 -0
- package/dist/pipelines/budget-breach.d.ts +31 -0
- package/dist/pipelines/budget-breach.d.ts.map +1 -0
- package/dist/pipelines/budget-breach.test.d.ts +2 -0
- package/dist/pipelines/budget-breach.test.d.ts.map +1 -0
- package/dist/pipelines/failure-cluster.d.ts +38 -0
- package/dist/pipelines/failure-cluster.d.ts.map +1 -0
- package/dist/pipelines/failure-cluster.test.d.ts +2 -0
- package/dist/pipelines/failure-cluster.test.d.ts.map +1 -0
- package/dist/pipelines/first-divergence.d.ts +26 -0
- package/dist/pipelines/first-divergence.d.ts.map +1 -0
- package/dist/pipelines/first-divergence.test.d.ts +2 -0
- package/dist/pipelines/first-divergence.test.d.ts.map +1 -0
- package/dist/pipelines/index.d.ts +8 -173
- package/dist/pipelines/index.d.ts.map +1 -0
- package/dist/pipelines/index.js +40 -15
- package/dist/pipelines/index.js.map +1 -1
- package/dist/pipelines/judge-agreement.d.ts +26 -0
- package/dist/pipelines/judge-agreement.d.ts.map +1 -0
- package/dist/pipelines/judge-agreement.test.d.ts +2 -0
- package/dist/pipelines/judge-agreement.test.d.ts.map +1 -0
- package/dist/pipelines/regression.d.ts +23 -0
- package/dist/pipelines/regression.d.ts.map +1 -0
- package/dist/pipelines/regression.test.d.ts +2 -0
- package/dist/pipelines/regression.test.d.ts.map +1 -0
- package/dist/pipelines/stuck-loop.d.ts +32 -0
- package/dist/pipelines/stuck-loop.d.ts.map +1 -0
- package/dist/pipelines/stuck-loop.test.d.ts +2 -0
- package/dist/pipelines/stuck-loop.test.d.ts.map +1 -0
- package/dist/pipelines/tool-waste.d.ts +34 -0
- package/dist/pipelines/tool-waste.d.ts.map +1 -0
- package/dist/pipelines/tool-waste.test.d.ts +2 -0
- package/dist/pipelines/tool-waste.test.d.ts.map +1 -0
- package/dist/playbook.d.ts +16 -0
- package/dist/playbook.d.ts.map +1 -0
- package/dist/pr-review-benchmark.d.ts +88 -0
- package/dist/pr-review-benchmark.d.ts.map +1 -0
- package/dist/pr-review-benchmark.test.d.ts +2 -0
- package/dist/pr-review-benchmark.test.d.ts.map +1 -0
- package/dist/pre-registration.d.ts +125 -0
- package/dist/pre-registration.d.ts.map +1 -0
- package/dist/prm/builtin-rubrics.d.ts +33 -0
- package/dist/prm/builtin-rubrics.d.ts.map +1 -0
- package/dist/prm/index.d.ts +5 -100
- package/dist/prm/index.d.ts.map +1 -0
- package/dist/prm/index.js +55 -12
- package/dist/prm/index.js.map +1 -1
- package/dist/prm/inference.d.ts +29 -0
- package/dist/prm/inference.d.ts.map +1 -0
- package/dist/prm/inference.test.d.ts +2 -0
- package/dist/prm/inference.test.d.ts.map +1 -0
- package/dist/{rubric-BOfxn4ja.d.ts → prm/rubric.d.ts} +10 -13
- package/dist/prm/rubric.d.ts.map +1 -0
- package/dist/prm/training-export.d.ts +38 -0
- package/dist/prm/training-export.d.ts.map +1 -0
- package/dist/produced-state.d.ts +63 -0
- package/dist/produced-state.d.ts.map +1 -0
- package/dist/produced-state.test.d.ts +8 -0
- package/dist/produced-state.test.d.ts.map +1 -0
- package/dist/profile/baselines.d.ts +37 -0
- package/dist/profile/baselines.d.ts.map +1 -0
- package/dist/profile/index.d.ts +105 -0
- package/dist/profile/index.d.ts.map +1 -0
- package/dist/promotion-gate.d.ts +93 -0
- package/dist/promotion-gate.d.ts.map +1 -0
- package/dist/prompt-registry.d.ts +41 -0
- package/dist/prompt-registry.d.ts.map +1 -0
- package/dist/propose-review-control.d.ts +49 -0
- package/dist/propose-review-control.d.ts.map +1 -0
- package/dist/propose-review-control.test.d.ts +2 -0
- package/dist/propose-review-control.test.d.ts.map +1 -0
- package/dist/propose-review.d.ts +155 -0
- package/dist/propose-review.d.ts.map +1 -0
- package/dist/{red-team-BXHil6c8.d.ts → red-team.d.ts} +14 -16
- package/dist/red-team.d.ts.map +1 -0
- package/dist/reference-replay-steering.d.ts +11 -0
- package/dist/reference-replay-steering.d.ts.map +1 -0
- package/dist/reference-replay.d.ts +176 -0
- package/dist/reference-replay.d.ts.map +1 -0
- package/dist/reflective-mutation.d.ts +79 -0
- package/dist/reflective-mutation.d.ts.map +1 -0
- package/dist/registry.d.ts +31 -0
- package/dist/registry.d.ts.map +1 -0
- package/dist/release-confidence.d.ts +128 -0
- package/dist/release-confidence.d.ts.map +1 -0
- package/dist/release-report.d.ts +11 -0
- package/dist/release-report.d.ts.map +1 -0
- package/dist/replay.d.ts +120 -0
- package/dist/replay.d.ts.map +1 -0
- package/dist/reporter.d.ts +14 -0
- package/dist/reporter.d.ts.map +1 -0
- package/dist/reporting.d.ts +15 -15
- package/dist/reporting.d.ts.map +1 -0
- package/dist/reporting.js +3 -3
- package/dist/researcher.d.ts +140 -0
- package/dist/researcher.d.ts.map +1 -0
- package/dist/reviewer.d.ts +118 -0
- package/dist/reviewer.d.ts.map +1 -0
- package/dist/reviewer.test.d.ts +2 -0
- package/dist/reviewer.test.d.ts.map +1 -0
- package/dist/reward-model-export.d.ts +60 -0
- package/dist/reward-model-export.d.ts.map +1 -0
- package/dist/rl/active-curriculum.d.ts +110 -0
- package/dist/rl/active-curriculum.d.ts.map +1 -0
- package/dist/rl/adaptation-eval.d.ts +109 -0
- package/dist/rl/adaptation-eval.d.ts.map +1 -0
- package/dist/{adversarial-DIVcDoI_.d.ts → rl/adversarial.d.ts} +6 -7
- package/dist/rl/adversarial.d.ts.map +1 -0
- package/dist/rl/compute-curves.d.ts +127 -0
- package/dist/rl/compute-curves.d.ts.map +1 -0
- package/dist/rl/contamination.d.ts +117 -0
- package/dist/rl/contamination.d.ts.map +1 -0
- package/dist/rl/corpus.d.ts +55 -0
- package/dist/rl/corpus.d.ts.map +1 -0
- package/dist/rl/corpus.test.d.ts +2 -0
- package/dist/rl/corpus.test.d.ts.map +1 -0
- package/dist/rl/dataset.d.ts +102 -0
- package/dist/rl/dataset.d.ts.map +1 -0
- package/dist/rl/dataset.test.d.ts +2 -0
- package/dist/rl/dataset.test.d.ts.map +1 -0
- package/dist/rl/exporters.d.ts +141 -0
- package/dist/rl/exporters.d.ts.map +1 -0
- package/dist/rl/index.d.ts +49 -0
- package/dist/rl/index.d.ts.map +1 -0
- package/dist/{off-policy-DiwuKKg7.d.ts → rl/off-policy.d.ts} +8 -9
- package/dist/rl/off-policy.d.ts.map +1 -0
- package/dist/rl/predictive-validity-researcher.d.ts +69 -0
- package/dist/rl/predictive-validity-researcher.d.ts.map +1 -0
- package/dist/rl/preferences.d.ts +141 -0
- package/dist/rl/preferences.d.ts.map +1 -0
- package/dist/rl/process-reward.d.ts +122 -0
- package/dist/rl/process-reward.d.ts.map +1 -0
- package/dist/rl/reward-hacking.d.ts +104 -0
- package/dist/rl/reward-hacking.d.ts.map +1 -0
- package/dist/rl/rl-campaign.d.ts +85 -0
- package/dist/rl/rl-campaign.d.ts.map +1 -0
- package/dist/rl/run-record-adapters.d.ts +56 -0
- package/dist/rl/run-record-adapters.d.ts.map +1 -0
- package/dist/rl/sim-fidelity.d.ts +166 -0
- package/dist/rl/sim-fidelity.d.ts.map +1 -0
- package/dist/rl/sim-fidelity.test.d.ts +2 -0
- package/dist/rl/sim-fidelity.test.d.ts.map +1 -0
- package/dist/rl/tournament.d.ts +115 -0
- package/dist/rl/tournament.d.ts.map +1 -0
- package/dist/rl/verifiable-reward.d.ts +124 -0
- package/dist/rl/verifiable-reward.d.ts.map +1 -0
- package/dist/rl.d.ts +1 -1192
- package/dist/rl.js +615 -615
- package/dist/rl.js.map +1 -1
- package/dist/{run-campaign-RDGAM5KJ.js → run-campaign-WXY7KI67.js} +3 -3
- package/dist/run-critic.d.ts +23 -0
- package/dist/run-critic.d.ts.map +1 -0
- package/dist/run-evidence.d.ts +32 -0
- package/dist/run-evidence.d.ts.map +1 -0
- package/dist/{run-record-e7vj1uZQ.d.ts → run-record.d.ts} +16 -122
- package/dist/run-record.d.ts.map +1 -0
- package/dist/run-record.test.d.ts +2 -0
- package/dist/run-record.test.d.ts.map +1 -0
- package/dist/run-score.d.ts +31 -0
- package/dist/run-score.d.ts.map +1 -0
- package/dist/runtime-trajectory.d.ts +47 -0
- package/dist/runtime-trajectory.d.ts.map +1 -0
- package/dist/{test-graded-scenario-BdVaPyHT.d.ts → sandbox-harness.d.ts} +34 -60
- package/dist/sandbox-harness.d.ts.map +1 -0
- package/dist/sandbox-harness.test.d.ts +2 -0
- package/dist/sandbox-harness.test.d.ts.map +1 -0
- package/dist/sandbox-pool.d.ts +74 -0
- package/dist/sandbox-pool.d.ts.map +1 -0
- package/dist/sandbox-pool.test.d.ts +2 -0
- package/dist/sandbox-pool.test.d.ts.map +1 -0
- package/dist/scorecard.d.ts +133 -0
- package/dist/scorecard.d.ts.map +1 -0
- package/dist/scorecard.test.d.ts +2 -0
- package/dist/scorecard.test.d.ts.map +1 -0
- package/dist/self-play.d.ts +69 -0
- package/dist/self-play.d.ts.map +1 -0
- package/dist/semantic-concept-judge.d.ts +135 -0
- package/dist/semantic-concept-judge.d.ts.map +1 -0
- package/dist/semantic-concept-judge.test.d.ts +2 -0
- package/dist/semantic-concept-judge.test.d.ts.map +1 -0
- package/dist/{sequential-5iSVfzl2.d.ts → sequential.d.ts} +9 -10
- package/dist/sequential.d.ts.map +1 -0
- package/dist/{series-convergence-D5OWMBg6.d.ts → series-convergence.d.ts} +4 -5
- package/dist/series-convergence.d.ts.map +1 -0
- package/dist/slo.d.ts +48 -0
- package/dist/slo.d.ts.map +1 -0
- package/dist/state-continuity.d.ts +47 -0
- package/dist/state-continuity.d.ts.map +1 -0
- package/dist/{statistics-C7PozGrZ.d.ts → statistics.d.ts} +152 -39
- package/dist/statistics.d.ts.map +1 -0
- package/dist/statistics.test.d.ts +2 -0
- package/dist/statistics.test.d.ts.map +1 -0
- package/dist/steering-optimizer.d.ts +58 -0
- package/dist/steering-optimizer.d.ts.map +1 -0
- package/dist/steering.d.ts +24 -0
- package/dist/steering.d.ts.map +1 -0
- package/dist/storyboard/code-edit.d.ts +64 -0
- package/dist/storyboard/code-edit.d.ts.map +1 -0
- package/dist/storyboard/code-edit.test.d.ts +2 -0
- package/dist/storyboard/code-edit.test.d.ts.map +1 -0
- package/dist/storyboard/index.d.ts +16 -81
- package/dist/storyboard/index.d.ts.map +1 -0
- package/dist/storyboard/index.test.d.ts +2 -0
- package/dist/storyboard/index.test.d.ts.map +1 -0
- package/dist/{summary-report-DGmUucwQ.d.ts → summary-report.d.ts} +23 -160
- package/dist/summary-report.d.ts.map +1 -0
- package/dist/telemetry/client.d.ts +35 -0
- package/dist/telemetry/client.d.ts.map +1 -0
- package/dist/telemetry/index.d.ts +17 -35
- package/dist/telemetry/index.d.ts.map +1 -0
- package/dist/telemetry/schema.d.ts +61 -0
- package/dist/telemetry/schema.d.ts.map +1 -0
- package/dist/telemetry/sink-fetch.d.ts +39 -0
- package/dist/telemetry/sink-fetch.d.ts.map +1 -0
- package/dist/telemetry/{file.d.ts → sink-file.d.ts} +5 -7
- package/dist/telemetry/sink-file.d.ts.map +1 -0
- package/dist/test-graded-scenario.d.ts +42 -0
- package/dist/test-graded-scenario.d.ts.map +1 -0
- package/dist/testing.d.ts +5 -0
- package/dist/testing.d.ts.map +1 -0
- package/dist/testing.js +8 -0
- package/dist/testing.js.map +1 -0
- package/dist/tool-use-metrics.d.ts +35 -0
- package/dist/tool-use-metrics.d.ts.map +1 -0
- package/dist/trace/capture-fetch.d.ts +48 -0
- package/dist/trace/capture-fetch.d.ts.map +1 -0
- package/dist/trace/capture-fetch.test.d.ts +2 -0
- package/dist/trace/capture-fetch.test.d.ts.map +1 -0
- package/dist/{emitter-DEZwY14K.d.ts → trace/emitter.d.ts} +9 -12
- package/dist/trace/emitter.d.ts.map +1 -0
- package/dist/trace/extract-usage.d.ts +46 -0
- package/dist/trace/extract-usage.d.ts.map +1 -0
- package/dist/trace/extract-usage.test.d.ts +2 -0
- package/dist/trace/extract-usage.test.d.ts.map +1 -0
- package/dist/trace/index.d.ts +15 -0
- package/dist/trace/index.d.ts.map +1 -0
- package/dist/{integrity-VJ9A7aST.d.ts → trace/integrity.d.ts} +11 -14
- package/dist/trace/integrity.d.ts.map +1 -0
- package/dist/trace/otel-bridge.d.ts +29 -0
- package/dist/trace/otel-bridge.d.ts.map +1 -0
- package/dist/trace/otel-export.d.ts +52 -0
- package/dist/trace/otel-export.d.ts.map +1 -0
- package/dist/trace/otel.d.ts +57 -0
- package/dist/trace/otel.d.ts.map +1 -0
- package/dist/trace/otlp-attributes.d.ts +17 -0
- package/dist/trace/otlp-attributes.d.ts.map +1 -0
- package/dist/trace/query.d.ts +29 -0
- package/dist/trace/query.d.ts.map +1 -0
- package/dist/trace/query.test.d.ts +2 -0
- package/dist/trace/query.test.d.ts.map +1 -0
- package/dist/{raw-provider-sink-C46HDghv.d.ts → trace/raw-provider-sink.d.ts} +13 -14
- package/dist/trace/raw-provider-sink.d.ts.map +1 -0
- package/dist/{redact-B40YG2M_.d.ts → trace/redact.d.ts} +7 -8
- package/dist/trace/redact.d.ts.map +1 -0
- package/dist/{schema-m0gsnbt3.d.ts → trace/schema.d.ts} +29 -30
- package/dist/trace/schema.d.ts.map +1 -0
- package/dist/trace/store-to-otlp.d.ts +72 -0
- package/dist/trace/store-to-otlp.d.ts.map +1 -0
- package/dist/trace/store-to-otlp.test.d.ts +2 -0
- package/dist/trace/store-to-otlp.test.d.ts.map +1 -0
- package/dist/{store-CKUAgsJz.d.ts → trace/store.d.ts} +37 -13
- package/dist/trace/store.d.ts.map +1 -0
- package/dist/trace/store.test.d.ts +2 -0
- package/dist/trace/store.test.d.ts.map +1 -0
- package/dist/{analyst-C8HHvfJp.d.ts → trace-analyst/analyst.d.ts} +12 -14
- package/dist/trace-analyst/analyst.d.ts.map +1 -0
- package/dist/trace-analyst/analyst.test.d.ts +2 -0
- package/dist/trace-analyst/analyst.test.d.ts.map +1 -0
- package/dist/trace-analyst/behavioral-metrics.d.ts +40 -0
- package/dist/trace-analyst/behavioral-metrics.d.ts.map +1 -0
- package/dist/trace-analyst/behavioral-metrics.test.d.ts +2 -0
- package/dist/trace-analyst/behavioral-metrics.test.d.ts.map +1 -0
- package/dist/trace-analyst/hook.d.ts +55 -0
- package/dist/trace-analyst/hook.d.ts.map +1 -0
- package/dist/trace-analyst/index.d.ts +18 -0
- package/dist/trace-analyst/index.d.ts.map +1 -0
- package/dist/trace-analyst/insights.d.ts +71 -0
- package/dist/trace-analyst/insights.d.ts.map +1 -0
- package/dist/trace-analyst/insights.test.d.ts +2 -0
- package/dist/trace-analyst/insights.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-flatten.d.ts +42 -0
- package/dist/trace-analyst/otlp-flatten.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-span.d.ts +85 -0
- package/dist/trace-analyst/otlp-span.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-span.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-span.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.d.ts +115 -0
- package/dist/trace-analyst/otlp-to-run-records.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-to-run-records.test.d.ts.map +1 -0
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts +2 -0
- package/dist/trace-analyst/otlp-to-run-records.timestamps.test.d.ts.map +1 -0
- package/dist/trace-analyst/prompts.d.ts +6 -0
- package/dist/trace-analyst/prompts.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.d.ts +126 -0
- package/dist/trace-analyst/store-otlp.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.test.d.ts +8 -0
- package/dist/trace-analyst/store-otlp.test.d.ts.map +1 -0
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts +2 -0
- package/dist/trace-analyst/store-otlp.timestamps.test.d.ts.map +1 -0
- package/dist/trace-analyst/store.d.ts +63 -0
- package/dist/trace-analyst/store.d.ts.map +1 -0
- package/dist/trace-analyst/tools.d.ts +44 -0
- package/dist/trace-analyst/tools.d.ts.map +1 -0
- package/dist/trace-analyst/tools.test.d.ts +10 -0
- package/dist/trace-analyst/tools.test.d.ts.map +1 -0
- package/dist/{store-C1YxJDEK.d.ts → trace-analyst/types.d.ts} +18 -74
- package/dist/trace-analyst/types.d.ts.map +1 -0
- package/dist/trace-contracts.d.ts +180 -0
- package/dist/trace-contracts.d.ts.map +1 -0
- package/dist/traced-analyst.d.ts +26 -0
- package/dist/traced-analyst.d.ts.map +1 -0
- package/dist/traced-judges.d.ts +27 -0
- package/dist/traced-judges.d.ts.map +1 -0
- package/dist/traces.d.ts +4 -947
- package/dist/traces.d.ts.map +1 -0
- package/dist/traces.js +42 -12
- package/dist/{trajectory-GEdXJCL5.d.ts → trajectory.d.ts} +8 -9
- package/dist/trajectory.d.ts.map +1 -0
- package/dist/{types-Croy5h7V.d.ts → types.d.ts} +33 -33
- package/dist/types.d.ts.map +1 -0
- package/dist/ui-finding.d.ts +104 -0
- package/dist/ui-finding.d.ts.map +1 -0
- package/dist/verdict-cache.d.ts +78 -0
- package/dist/verdict-cache.d.ts.map +1 -0
- package/dist/verdict-cache.test.d.ts +2 -0
- package/dist/verdict-cache.test.d.ts.map +1 -0
- package/dist/{verdict-C9MlYujm.d.ts → verdict.d.ts} +2 -3
- package/dist/verdict.d.ts.map +1 -0
- package/dist/visual-diff.d.ts +32 -0
- package/dist/visual-diff.d.ts.map +1 -0
- package/dist/wire/handlers.d.ts +54 -0
- package/dist/wire/handlers.d.ts.map +1 -0
- package/dist/wire/index.d.ts +13 -570
- package/dist/wire/index.d.ts.map +1 -0
- package/dist/wire/index.js +2 -2
- package/dist/wire/openapi.d.ts +3 -0
- package/dist/wire/openapi.d.ts.map +1 -0
- package/dist/wire/rpc.d.ts +21 -0
- package/dist/wire/rpc.d.ts.map +1 -0
- package/dist/wire/rubrics.d.ts +34 -0
- package/dist/wire/rubrics.d.ts.map +1 -0
- package/dist/wire/schemas.d.ts +410 -0
- package/dist/wire/schemas.d.ts.map +1 -0
- package/dist/wire/server.d.ts +60 -0
- package/dist/wire/server.d.ts.map +1 -0
- package/dist/workflow/event-schema.d.ts +5 -0
- package/dist/workflow/event-schema.d.ts.map +1 -0
- package/dist/workflow/feedback-pack.d.ts +99 -0
- package/dist/workflow/feedback-pack.d.ts.map +1 -0
- package/dist/workflow/index.d.ts +22 -495
- package/dist/workflow/index.d.ts.map +1 -0
- package/dist/workflow/index.js +2 -2
- package/dist/workflow/intelligence-export.d.ts +62 -0
- package/dist/workflow/intelligence-export.d.ts.map +1 -0
- package/dist/workflow/partner-report.d.ts +49 -0
- package/dist/workflow/partner-report.d.ts.map +1 -0
- package/dist/workflow/phase-graph.d.ts +43 -0
- package/dist/workflow/phase-graph.d.ts.map +1 -0
- package/dist/workflow/promotion-gate.d.ts +61 -0
- package/dist/workflow/promotion-gate.d.ts.map +1 -0
- package/dist/workflow/run-record.d.ts +12 -0
- package/dist/workflow/run-record.d.ts.map +1 -0
- package/dist/workflow/runtime-adapter.d.ts +20 -0
- package/dist/workflow/runtime-adapter.d.ts.map +1 -0
- package/dist/workflow/sanitize.d.ts +21 -0
- package/dist/workflow/sanitize.d.ts.map +1 -0
- package/dist/workflow/schema.d.ts +5 -0
- package/dist/workflow/schema.d.ts.map +1 -0
- package/dist/workflow/summary.d.ts +43 -0
- package/dist/workflow/summary.d.ts.map +1 -0
- package/dist/workflow/trace-event-fields.d.ts +6 -0
- package/dist/workflow/trace-event-fields.d.ts.map +1 -0
- package/dist/workflow/trajectory.d.ts +15 -0
- package/dist/workflow/trajectory.d.ts.map +1 -0
- package/dist/workflow/types.d.ts +68 -0
- package/dist/workflow/types.d.ts.map +1 -0
- package/dist/workspace-inspector.d.ts +67 -0
- package/dist/workspace-inspector.d.ts.map +1 -0
- package/dist/wrangler-deploy-runner.test.d.ts +2 -0
- package/dist/wrangler-deploy-runner.test.d.ts.map +1 -0
- package/docs/campaign-proposers.md +170 -0
- package/docs/concepts.md +8 -4
- package/docs/customer-journeys.md +15 -13
- package/docs/design/loop-taxonomy.md +34 -66
- package/docs/distributed-driver.md +14 -14
- package/docs/feature-guide.md +1 -1
- package/docs/hosted-ingest-spec.md +2 -3
- package/docs/multi-shot-optimization.md +8 -8
- package/docs/product-eval-adoption.md +1 -1
- package/docs/self-improvement-map.md +33 -29
- package/package.json +9 -15
- package/dist/calibration-Cpr3WaX3.d.ts +0 -101
- package/dist/chunk-47X6LRCE.js.map +0 -1
- package/dist/chunk-6SOJM3VR.js.map +0 -1
- package/dist/chunk-CVVHBFGN.js.map +0 -1
- package/dist/chunk-D3V5B42D.js.map +0 -1
- package/dist/chunk-E4GH6USR.js.map +0 -1
- package/dist/chunk-GSH6QNNS.js.map +0 -1
- package/dist/chunk-KWRRMR3J.js.map +0 -1
- package/dist/chunk-LMZQ2Z4U.js.map +0 -1
- package/dist/chunk-QAY5UIJO.js.map +0 -1
- package/dist/chunk-QG2OVF2D.js.map +0 -1
- package/dist/chunk-T375SUOZ.js.map +0 -1
- package/dist/chunk-TWS7AZEY.js.map +0 -1
- package/dist/chunk-UMMZHCPB.js.map +0 -1
- package/dist/chunk-XY4DDNEG.js.map +0 -1
- package/dist/chunk-ZFIBGEOL.js.map +0 -1
- package/dist/control-_Qb7skHX.d.ts +0 -259
- package/dist/corpus-BoR-041R.d.ts +0 -560
- package/dist/failure-cluster-CL7IVgkJ.d.ts +0 -76
- package/dist/harness-optimizer-EnEnQPsr.d.ts +0 -106
- package/dist/index-Bx3gZ8xl.d.ts +0 -159
- package/dist/pre-registration-mAnCugl9.d.ts +0 -523
- package/dist/provenance-LnqRT0sS.d.ts +0 -441
- package/dist/query-CqTxMwDw.d.ts +0 -31
- package/dist/release-report-euXIV_Sk.d.ts +0 -233
- package/dist/run-critic-BAIjX99r.d.ts +0 -56
- package/dist/run-improvement-loop-5z_l5zDz.d.ts +0 -427
- package/dist/runtime-trajectory-BDgfGZSr.d.ts +0 -49
- package/dist/semantic-concept-judge-Dn8Z6KEG.d.ts +0 -624
- package/dist/sink-fetch-B1Yg4Til.d.ts +0 -101
- package/docs/design/external-agent-wedge.md +0 -89
- package/docs/design/phase-d-rfc.md +0 -125
- package/docs/design/phase4-consumer-migration.md +0 -70
- package/docs/design/primitives-integration-spec.md +0 -393
- package/docs/design/product-self-improvement-loop.md +0 -146
- package/docs/design/self-improvement-engine.md +0 -140
- package/docs/design/self-improvement-protocol.md +0 -223
- package/docs/design/self-improvement-roadmap.md +0 -106
- package/docs/design/substrate-gaps.md +0 -118
- package/docs/phase-b-pairing-kit.md +0 -188
- package/docs/phase-b-runbook.md +0 -176
- package/docs/pilot/README.md +0 -62
- package/docs/pilot/customer-checklist.md +0 -90
- package/docs/pilot/integration-foreign-stack.md +0 -296
- package/docs/pilot/integration-tangle-stack.md +0 -248
- package/docs/pilot/one-pager.md +0 -161
- package/docs/pilot/sample-insight-report.json +0 -172
- package/docs/quickstart-external.md +0 -229
- package/docs/research/belief-state-agent-eval-roadmap.md +0 -593
- package/docs/research/research-roadmap.md +0 -205
- package/docs/specs/driver-honest-spec.md +0 -251
- package/docs/specs/hermes-self-improvement-audit.md +0 -93
- package/docs/specs/profile-versioning.md +0 -291
- package/docs/three-package-architecture.md +0 -168
- /package/dist/{chunk-QS3RBQPI.js.map → chunk-D5KBVULY.js.map} +0 -0
- /package/dist/{chunk-UHMJT4T7.js.map → chunk-IZCEK2HR.js.map} +0 -0
- /package/dist/{chunk-L3JOU6XM.js.map → chunk-NACM5RS7.js.map} +0 -0
- /package/dist/{chunk-4FBZZIYD.js.map → chunk-QTMYW64F.js.map} +0 -0
- /package/dist/{chunk-CY6U5S3X.js.map → chunk-URWVKRCS.js.map} +0 -0
- /package/dist/{chunk-Y47J2LJ3.js.map → chunk-Z6L6YSU6.js.map} +0 -0
- /package/dist/{run-campaign-RDGAM5KJ.js.map → run-campaign-WXY7KI67.js.map} +0 -0
|
@@ -0,0 +1,105 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Structured prompt profile — the system prompt as named, addressable sections
|
|
3
|
+
* instead of one opaque blob. The self-improvement loop targets ONE evolvable
|
|
4
|
+
* `domain` section at a time (via `applyDomainPatch`); the role, environment,
|
|
5
|
+
* tool conventions, and skill roster stay fixed so a candidate diff is
|
|
6
|
+
* attributable to a single section rather than a whole-prompt rewrite.
|
|
7
|
+
*
|
|
8
|
+
* `renderProfile` emits the prompt in a fixed five-zone order — role,
|
|
9
|
+
* Environment, Tool conventions, Skills, Domain guidance — mirroring the layout
|
|
10
|
+
* of Harvey LAB's `harness/system_prompt.md` (workspace root, read-only
|
|
11
|
+
* documents, output dir, skills dir). `profileToSurface` is the bridge to the
|
|
12
|
+
* loop's string `MutableSurface`: a profile renders to exactly the text a
|
|
13
|
+
* candidate is scored on.
|
|
14
|
+
*
|
|
15
|
+
* Distinct from the canonical `AgentProfile` exported by
|
|
16
|
+
* `@tangle-network/agent-interface`. This helper builds prompt content that can
|
|
17
|
+
* be rendered into an agent-interface profile's `prompt.systemPrompt`.
|
|
18
|
+
*/
|
|
19
|
+
import { type BaselineRoleKey } from './baselines';
|
|
20
|
+
export { BASELINE_ROLES, type BaselineRoleKey, engineerRole, generalistRole, researcherRole, } from './baselines';
|
|
21
|
+
/** A named, addressable region of the system prompt. `evolvable` marks whether
|
|
22
|
+
* the self-improvement loop is allowed to patch its body; fixed scaffolding
|
|
23
|
+
* (e.g. a compliance preamble) sets `evolvable: false`. */
|
|
24
|
+
export interface AgentProfileSection {
|
|
25
|
+
id: string;
|
|
26
|
+
title: string;
|
|
27
|
+
body: string;
|
|
28
|
+
evolvable: boolean;
|
|
29
|
+
}
|
|
30
|
+
/** A skill the agent can invoke. `triggers` are the phrases/conditions that
|
|
31
|
+
* should activate it; `scriptsDir` points at its executable scripts when the
|
|
32
|
+
* skill ships code (mirrors Harvey's `skills/<name>/scripts/`). */
|
|
33
|
+
export interface ProfileSkill {
|
|
34
|
+
name: string;
|
|
35
|
+
description: string;
|
|
36
|
+
triggers: string[];
|
|
37
|
+
scriptsDir?: string;
|
|
38
|
+
}
|
|
39
|
+
/** The structured system prompt. The first four fields are fixed scaffolding;
|
|
40
|
+
* `domain` is the evolvable surface the loop optimizes. */
|
|
41
|
+
export interface PromptProfile {
|
|
42
|
+
role: string;
|
|
43
|
+
environment: string;
|
|
44
|
+
toolConventions: string;
|
|
45
|
+
skills: ProfileSkill[];
|
|
46
|
+
domain: AgentProfileSection[];
|
|
47
|
+
}
|
|
48
|
+
/**
|
|
49
|
+
* Emit the sectioned system prompt in fixed order: role text, then
|
|
50
|
+
* `## Environment`, `## Tool conventions`, `## Skills` (each skill as
|
|
51
|
+
* `### <name>` + description + triggers), `## Domain guidance` (each section as
|
|
52
|
+
* `### <title>` + body). The order is load-bearing — the loop diffs rendered
|
|
53
|
+
* text, and reordering would make every candidate look like a full rewrite.
|
|
54
|
+
*/
|
|
55
|
+
export declare function renderProfile(p: PromptProfile): string;
|
|
56
|
+
/** The string `MutableSurface` the self-improvement loop scores — a profile
|
|
57
|
+
* renders to exactly the text a candidate is graded on. */
|
|
58
|
+
export declare function profileToSurface(p: PromptProfile): string;
|
|
59
|
+
/**
|
|
60
|
+
* The fixed-scaffolding baseline: a `role`/`environment`/`toolConventions`
|
|
61
|
+
* profile with the stock skill roster and NO domain guidance. The
|
|
62
|
+
* production profile is this plus shipped domain sections.
|
|
63
|
+
*/
|
|
64
|
+
export declare function baselineProfile(args: {
|
|
65
|
+
role: string;
|
|
66
|
+
environment?: string;
|
|
67
|
+
toolConventions?: string;
|
|
68
|
+
skills?: ProfileSkill[];
|
|
69
|
+
}): PromptProfile;
|
|
70
|
+
/**
|
|
71
|
+
* Baseline from one of the strong generic roles (`'engineer' | 'researcher' |
|
|
72
|
+
* 'generalist'`) — the common case: pick a role foundation, optionally override
|
|
73
|
+
* the environment to describe THIS product's sandbox, then layer domain via
|
|
74
|
+
* `prodProfile`. Domain guidance stays in the product repo; this only supplies
|
|
75
|
+
* the generically-useful role + stock scaffolding.
|
|
76
|
+
*/
|
|
77
|
+
export declare function baselineProfileFromRole(role: BaselineRoleKey, args?: {
|
|
78
|
+
environment?: string;
|
|
79
|
+
toolConventions?: string;
|
|
80
|
+
skills?: ProfileSkill[];
|
|
81
|
+
}): PromptProfile;
|
|
82
|
+
/** The production profile: the baseline scaffolding plus the domain sections
|
|
83
|
+
* shipped after self-improvement. Differs from the baseline ONLY in `domain`
|
|
84
|
+
* (and any skills the caller layered into the baseline) — the role,
|
|
85
|
+
* environment, and tool conventions are carried through unchanged. */
|
|
86
|
+
export declare function prodProfile(baseline: PromptProfile, shipped: AgentProfileSection[]): PromptProfile;
|
|
87
|
+
/**
|
|
88
|
+
* Section-scoped edit — replace the body of ONE domain section by id, leaving
|
|
89
|
+
* every other section byte-identical. This is how the loop targets a single
|
|
90
|
+
* evolvable surface. Throws on an unknown section id (a patch that hits no
|
|
91
|
+
* section is a silent no-op otherwise — fail loud). Throws when the targeted
|
|
92
|
+
* section is `evolvable: false`.
|
|
93
|
+
*/
|
|
94
|
+
export declare function applyDomainPatch(p: PromptProfile, sectionId: string, newBody: string): PromptProfile;
|
|
95
|
+
/**
|
|
96
|
+
* Content hash of a single section — its loop identity. Delegates to the
|
|
97
|
+
* campaign provenance `surfaceContentHash` (the same helper the loop's
|
|
98
|
+
* provenance record uses for full surfaces), so a section hash is byte-for-byte
|
|
99
|
+
* comparable with the `sha256:`-prefixed content hashes that record carries.
|
|
100
|
+
* Two sections that render identically (same title + body) share a hash; the
|
|
101
|
+
* `id` and `evolvable` flag are excluded — they are routing metadata, not
|
|
102
|
+
* content.
|
|
103
|
+
*/
|
|
104
|
+
export declare function sectionHash(section: AgentProfileSection): string;
|
|
105
|
+
//# sourceMappingURL=index.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/profile/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAGH,OAAO,EAAkB,KAAK,eAAe,EAAE,MAAM,aAAa,CAAA;AAElE,OAAO,EACL,cAAc,EACd,KAAK,eAAe,EACpB,YAAY,EACZ,cAAc,EACd,cAAc,GACf,MAAM,aAAa,CAAA;AAEpB;;4DAE4D;AAC5D,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,OAAO,CAAA;CACnB;AAED;;oEAEoE;AACpE,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,MAAM,EAAE,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;CACpB;AAED;4DAC4D;AAC5D,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,EAAE,MAAM,CAAA;IACnB,eAAe,EAAE,MAAM,CAAA;IACvB,MAAM,EAAE,YAAY,EAAE,CAAA;IACtB,MAAM,EAAE,mBAAmB,EAAE,CAAA;CAC9B;AAaD;;;;;;GAMG;AACH,wBAAgB,aAAa,CAAC,CAAC,EAAE,aAAa,GAAG,MAAM,CAetD;AAED;4DAC4D;AAC5D,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,aAAa,GAAG,MAAM,CAEzD;AAgCD;;;;GAIG;AACH,wBAAgB,eAAe,CAAC,IAAI,EAAE;IACpC,IAAI,EAAE,MAAM,CAAA;IACZ,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,MAAM,CAAC,EAAE,YAAY,EAAE,CAAA;CACxB,GAAG,aAAa,CAQhB;AAED;;;;;;GAMG;AACH,wBAAgB,uBAAuB,CACrC,IAAI,EAAE,eAAe,EACrB,IAAI,GAAE;IAAE,WAAW,CAAC,EAAE,MAAM,CAAC;IAAC,eAAe,CAAC,EAAE,MAAM,CAAC;IAAC,MAAM,CAAC,EAAE,YAAY,EAAE,CAAA;CAAO,GACrF,aAAa,CAEf;AAED;;;uEAGuE;AACvE,wBAAgB,WAAW,CACzB,QAAQ,EAAE,aAAa,EACvB,OAAO,EAAE,mBAAmB,EAAE,GAC7B,aAAa,CAEf;AAED;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAC9B,CAAC,EAAE,aAAa,EAChB,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,MAAM,GACd,aAAa,CAmBf;AAED;;;;;;;;GAQG;AACH,wBAAgB,WAAW,CAAC,OAAO,EAAE,mBAAmB,GAAG,MAAM,CAEhE"}
|
|
@@ -0,0 +1,93 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Bootstrap-CI promotion gate.
|
|
3
|
+
*
|
|
4
|
+
* In any iterative-improvement loop (GEPA, prompt evolution, dataset
|
|
5
|
+
* curation), the question is "did this generation actually improve, or are
|
|
6
|
+
* we celebrating noise?". With small N and noisy outcomes, point-estimate
|
|
7
|
+
* deltas lie. Bootstrap confidence intervals tell the operator whether the
|
|
8
|
+
* delta is real before code or prompts get promoted.
|
|
9
|
+
*
|
|
10
|
+
* This module is pure functions — no I/O, no model calls. Easy to unit-test
|
|
11
|
+
* and to compose into any verdict gate.
|
|
12
|
+
*
|
|
13
|
+
* Default gate:
|
|
14
|
+
* - Bootstrap mean baseline vs candidate (1k resamples).
|
|
15
|
+
* - Compute the delta distribution; pass if the lower CI bound > 0.
|
|
16
|
+
* - Tunable confidence (default 95%) and resample count.
|
|
17
|
+
*
|
|
18
|
+
* Verdict semantics intentionally match the existing `experiments.jsonl`
|
|
19
|
+
* vocabulary:
|
|
20
|
+
* - ADVANCE: candidate's CI lower bound > baseline mean (real win)
|
|
21
|
+
* - KEEP: overlap, but candidate point estimate >= baseline (neutral)
|
|
22
|
+
* - REVERT: candidate's CI upper bound < baseline mean (real regression)
|
|
23
|
+
* - INCONCLUSIVE: not enough samples or CI straddles zero with no signal
|
|
24
|
+
*/
|
|
25
|
+
export type Verdict = 'ADVANCE' | 'KEEP' | 'REVERT' | 'INCONCLUSIVE';
|
|
26
|
+
export interface BootstrapResult {
|
|
27
|
+
baselineMean: number;
|
|
28
|
+
candidateMean: number;
|
|
29
|
+
/** candidateMean - baselineMean, point estimate. */
|
|
30
|
+
delta: number;
|
|
31
|
+
/** Lower bound of the (1 - alpha) CI on the delta. */
|
|
32
|
+
ciLower: number;
|
|
33
|
+
/** Upper bound of the (1 - alpha) CI on the delta. */
|
|
34
|
+
ciUpper: number;
|
|
35
|
+
/** Number of bootstrap resamples used. */
|
|
36
|
+
iterations: number;
|
|
37
|
+
alpha: number;
|
|
38
|
+
verdict: Verdict;
|
|
39
|
+
}
|
|
40
|
+
export interface BootstrapOptions {
|
|
41
|
+
/** Confidence level alpha (default 0.05 → 95% CI). */
|
|
42
|
+
alpha?: number;
|
|
43
|
+
/** Number of resamples (default 1000). */
|
|
44
|
+
iterations?: number;
|
|
45
|
+
/**
|
|
46
|
+
* Minimum total samples (baseline + candidate) below which we always
|
|
47
|
+
* return INCONCLUSIVE — bootstrap with too few samples is meaningless.
|
|
48
|
+
* Default 6 (combined).
|
|
49
|
+
*/
|
|
50
|
+
minTotalSamples?: number;
|
|
51
|
+
/** RNG seed for reproducibility. Default: Math.random. */
|
|
52
|
+
seed?: number;
|
|
53
|
+
}
|
|
54
|
+
/**
|
|
55
|
+
* Compute the bootstrap CI on (candidateMean - baselineMean) and a verdict.
|
|
56
|
+
*
|
|
57
|
+
* Uses simple percentile bootstrap on the difference of resampled means.
|
|
58
|
+
* That's the standard non-parametric primitive — no distributional
|
|
59
|
+
* assumptions, robust to skew, easy to reason about.
|
|
60
|
+
*/
|
|
61
|
+
export declare function bootstrapCi(baseline: number[], candidate: number[], options?: BootstrapOptions): BootstrapResult;
|
|
62
|
+
/**
|
|
63
|
+
* Judge-replay promotion gate.
|
|
64
|
+
*
|
|
65
|
+
* The cheap inner-loop judge that drives an evolution run is by definition
|
|
66
|
+
* fast and noisy. When you're about to promote a winning variant to the
|
|
67
|
+
* canonical default, you want a STRONGER judge (a more expensive model, a
|
|
68
|
+
* human grader, a separately-trained reward model) to confirm the win
|
|
69
|
+
* generalises beyond the inner loop.
|
|
70
|
+
*
|
|
71
|
+
* This helper takes raw winner + baseline outputs, scores both through the
|
|
72
|
+
* stronger judge, and applies `bootstrapCi`. ADVANCE means the stronger
|
|
73
|
+
* judge agrees the winner is real with the configured confidence. Doesn't
|
|
74
|
+
* matter what shape your "output" is — pass a string, an object, anything
|
|
75
|
+
* the judge can read.
|
|
76
|
+
*/
|
|
77
|
+
export interface JudgeReplayGateArgs<TOutput> {
|
|
78
|
+
baselineOutputs: TOutput[];
|
|
79
|
+
candidateOutputs: TOutput[];
|
|
80
|
+
/** Stronger judge — async to allow LLM calls. Return a 0..N scalar score. */
|
|
81
|
+
judge: (output: TOutput) => Promise<number> | number;
|
|
82
|
+
alpha?: number;
|
|
83
|
+
iterations?: number;
|
|
84
|
+
/** RNG seed for reproducibility. */
|
|
85
|
+
seed?: number;
|
|
86
|
+
/** Maximum concurrent judge calls. Default 4. */
|
|
87
|
+
judgeConcurrency?: number;
|
|
88
|
+
}
|
|
89
|
+
export declare function judgeReplayGate<TOutput>(args: JudgeReplayGateArgs<TOutput>): Promise<BootstrapResult & {
|
|
90
|
+
baselineSamples: number;
|
|
91
|
+
candidateSamples: number;
|
|
92
|
+
}>;
|
|
93
|
+
//# sourceMappingURL=promotion-gate.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"promotion-gate.d.ts","sourceRoot":"","sources":["../src/promotion-gate.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,MAAM,MAAM,OAAO,GAAG,SAAS,GAAG,MAAM,GAAG,QAAQ,GAAG,cAAc,CAAA;AAEpE,MAAM,WAAW,eAAe;IAC9B,YAAY,EAAE,MAAM,CAAA;IACpB,aAAa,EAAE,MAAM,CAAA;IACrB,oDAAoD;IACpD,KAAK,EAAE,MAAM,CAAA;IACb,sDAAsD;IACtD,OAAO,EAAE,MAAM,CAAA;IACf,sDAAsD;IACtD,OAAO,EAAE,MAAM,CAAA;IACf,0CAA0C;IAC1C,UAAU,EAAE,MAAM,CAAA;IAClB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,OAAO,CAAA;CACjB;AAED,MAAM,WAAW,gBAAgB;IAC/B,sDAAsD;IACtD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,0CAA0C;IAC1C,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,0DAA0D;IAC1D,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED;;;;;;GAMG;AACH,wBAAgB,WAAW,CACzB,QAAQ,EAAE,MAAM,EAAE,EAClB,SAAS,EAAE,MAAM,EAAE,EACnB,OAAO,GAAE,gBAAqB,GAC7B,eAAe,CAuDjB;AAyCD;;;;;;;;;;;;;;GAcG;AACH,MAAM,WAAW,mBAAmB,CAAC,OAAO;IAC1C,eAAe,EAAE,OAAO,EAAE,CAAA;IAC1B,gBAAgB,EAAE,OAAO,EAAE,CAAA;IAC3B,6EAA6E;IAC7E,KAAK,EAAE,CAAC,MAAM,EAAE,OAAO,KAAK,OAAO,CAAC,MAAM,CAAC,GAAG,MAAM,CAAA;IACpD,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,oCAAoC;IACpC,IAAI,CAAC,EAAE,MAAM,CAAA;IACb,iDAAiD;IACjD,gBAAgB,CAAC,EAAE,MAAM,CAAA;CAC1B;AAED,wBAAsB,eAAe,CAAC,OAAO,EAC3C,IAAI,EAAE,mBAAmB,CAAC,OAAO,CAAC,GACjC,OAAO,CAAC,eAAe,GAAG;IAAE,eAAe,EAAE,MAAM,CAAC;IAAC,gBAAgB,EAAE,MAAM,CAAA;CAAE,CAAC,CAclF"}
|
|
@@ -0,0 +1,41 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Versioned prompt registry.
|
|
3
|
+
*
|
|
4
|
+
* Every prompt used in an eval run is registered with an explicit version.
|
|
5
|
+
* Reports include the content hash so A/B compares are rigorous: if the
|
|
6
|
+
* hash changes between two reports, the prompt actually changed; if it
|
|
7
|
+
* matches, the variance is elsewhere.
|
|
8
|
+
*
|
|
9
|
+
* Hash is SHA-256(content), truncated to 12 hex chars for readability.
|
|
10
|
+
* Uses the Web Crypto API (works in Workers, Node 22+, browsers).
|
|
11
|
+
*/
|
|
12
|
+
export interface PromptHandle {
|
|
13
|
+
/** Stable human-readable id, e.g. 'browser.system' */
|
|
14
|
+
id: string;
|
|
15
|
+
/** Caller-chosen version string, e.g. 'v3' or '2026-04-20' */
|
|
16
|
+
version: string;
|
|
17
|
+
/** SHA-256 of content, 12-hex-char prefix */
|
|
18
|
+
hash: string;
|
|
19
|
+
/** Full prompt body */
|
|
20
|
+
content: string;
|
|
21
|
+
}
|
|
22
|
+
export declare class PromptRegistry {
|
|
23
|
+
private readonly entries;
|
|
24
|
+
/**
|
|
25
|
+
* Register a prompt. Re-registering the same id+version with DIFFERENT
|
|
26
|
+
* content throws — versions are immutable. Re-registering with the SAME
|
|
27
|
+
* content is a no-op (idempotent).
|
|
28
|
+
*/
|
|
29
|
+
register(id: string, version: string, content: string): Promise<PromptHandle>;
|
|
30
|
+
/** Look up a registered prompt. Throws if unknown — no implicit defaults. */
|
|
31
|
+
get(id: string, version: string): PromptHandle;
|
|
32
|
+
/** Return all versions of an id, newest-first (lex-descending on version). */
|
|
33
|
+
listVersions(id: string): PromptHandle[];
|
|
34
|
+
/** Snapshot the whole registry — useful for including in reports. */
|
|
35
|
+
list(): PromptHandle[];
|
|
36
|
+
/** Verify a hash against registered content. Returns null if not found. */
|
|
37
|
+
verifyHash(id: string, version: string, expectedHash: string): boolean | null;
|
|
38
|
+
}
|
|
39
|
+
/** SHA-256(content) → first 12 hex chars. Stable across runtimes. */
|
|
40
|
+
export declare function hashContent(content: string): Promise<string>;
|
|
41
|
+
//# sourceMappingURL=prompt-registry.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"prompt-registry.d.ts","sourceRoot":"","sources":["../src/prompt-registry.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,MAAM,WAAW,YAAY;IAC3B,sDAAsD;IACtD,EAAE,EAAE,MAAM,CAAA;IACV,8DAA8D;IAC9D,OAAO,EAAE,MAAM,CAAA;IACf,6CAA6C;IAC7C,IAAI,EAAE,MAAM,CAAA;IACZ,uBAAuB;IACvB,OAAO,EAAE,MAAM,CAAA;CAChB;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAkC;IAE1D;;;;OAIG;IACG,QAAQ,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,YAAY,CAAC;IAoBnF,6EAA6E;IAC7E,GAAG,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,YAAY;IAO9C,8EAA8E;IAC9E,YAAY,CAAC,EAAE,EAAE,MAAM,GAAG,YAAY,EAAE;IAMxC,qEAAqE;IACrE,IAAI,IAAI,YAAY,EAAE;IAItB,2EAA2E;IAC3E,UAAU,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI;CAK9E;AAED,qEAAqE;AACrE,wBAAsB,WAAW,CAAC,OAAO,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAOlE"}
|
|
@@ -0,0 +1,49 @@
|
|
|
1
|
+
import { type ControlRunResult, type ControlRuntimeConfig } from './control-runtime';
|
|
2
|
+
import { type ProposeFn, type Review, type ReviewFn, type ReviewMemoryEntry, type ReviewMemoryStore, type Verification, type VerifyFn } from './propose-review';
|
|
3
|
+
import type { FailureClass } from './trace/schema';
|
|
4
|
+
import type { TraceStore } from './trace/store';
|
|
5
|
+
export interface ProposeReviewControlState<State, Summary = unknown> {
|
|
6
|
+
shot: number;
|
|
7
|
+
state: State;
|
|
8
|
+
priorReview: Review | null;
|
|
9
|
+
verification: Verification;
|
|
10
|
+
traceSummary?: Summary;
|
|
11
|
+
memory: ReviewMemoryEntry[];
|
|
12
|
+
completed: boolean;
|
|
13
|
+
reviewAvailable: boolean;
|
|
14
|
+
reviewError?: string;
|
|
15
|
+
}
|
|
16
|
+
export interface ProposeReviewControlAction {
|
|
17
|
+
type: 'propose-review-shot';
|
|
18
|
+
shot: number;
|
|
19
|
+
}
|
|
20
|
+
export interface ProposeReviewControlResult<State, Summary = unknown> {
|
|
21
|
+
state: State;
|
|
22
|
+
verification: Verification;
|
|
23
|
+
traceSummary?: Summary;
|
|
24
|
+
review: Review | null;
|
|
25
|
+
reviewAvailable: boolean;
|
|
26
|
+
reviewError?: string;
|
|
27
|
+
}
|
|
28
|
+
export interface ProposeReviewControlConfig<State, Summary = unknown> {
|
|
29
|
+
goal: string;
|
|
30
|
+
initialState: State;
|
|
31
|
+
propose: ProposeFn<State, Summary>;
|
|
32
|
+
verify: VerifyFn<State>;
|
|
33
|
+
review: ReviewFn<State, Summary>;
|
|
34
|
+
maxShots?: number;
|
|
35
|
+
maxWallMs?: number;
|
|
36
|
+
memory?: ReviewMemoryStore;
|
|
37
|
+
store?: TraceStore;
|
|
38
|
+
scenarioId?: string;
|
|
39
|
+
projectId?: string;
|
|
40
|
+
variantId?: string;
|
|
41
|
+
fallbackInstruction?: string;
|
|
42
|
+
confidenceFloor?: number;
|
|
43
|
+
confidenceFloorWindow?: number;
|
|
44
|
+
failureClassFromVerification?: (verification: Verification) => FailureClass | undefined;
|
|
45
|
+
actionFailure?: ControlRuntimeConfig<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>['actionFailure'];
|
|
46
|
+
}
|
|
47
|
+
export declare function runProposeReviewAsControlLoop<State, Summary = unknown>(config: ProposeReviewControlConfig<State, Summary>): Promise<ControlRunResult<ProposeReviewControlState<State, Summary>, ProposeReviewControlAction, ProposeReviewControlResult<State, Summary>>>;
|
|
48
|
+
export declare function controlFailureClassFromVerification(verification: Verification): FailureClass | undefined;
|
|
49
|
+
//# sourceMappingURL=propose-review-control.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"propose-review-control.d.ts","sourceRoot":"","sources":["../src/propose-review-control.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,KAAK,gBAAgB,EACrB,KAAK,oBAAoB,EAG1B,MAAM,mBAAmB,CAAA;AAC1B,OAAO,EAEL,KAAK,SAAS,EAEd,KAAK,MAAM,EACX,KAAK,QAAQ,EACb,KAAK,iBAAiB,EACtB,KAAK,iBAAiB,EACtB,KAAK,YAAY,EACjB,KAAK,QAAQ,EACd,MAAM,kBAAkB,CAAA;AACzB,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAE/C,MAAM,WAAW,yBAAyB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACjE,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,MAAM,EAAE,iBAAiB,EAAE,CAAA;IAC3B,SAAS,EAAE,OAAO,CAAA;IAClB,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED,MAAM,WAAW,0BAA0B;IACzC,IAAI,EAAE,qBAAqB,CAAA;IAC3B,IAAI,EAAE,MAAM,CAAA;CACb;AAED,MAAM,WAAW,0BAA0B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAClE,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,CAAC,EAAE,OAAO,CAAA;IACtB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED,MAAM,WAAW,0BAA0B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAClE,IAAI,EAAE,MAAM,CAAA;IACZ,YAAY,EAAE,KAAK,CAAA;IACnB,OAAO,EAAE,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAClC,MAAM,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAA;IACvB,MAAM,EAAE,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAChC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,MAAM,CAAC,EAAE,iBAAiB,CAAA;IAC1B,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,mBAAmB,CAAC,EAAE,MAAM,CAAA;IAC5B,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qBAAqB,CAAC,EAAE,MAAM,CAAA;IAC9B,4BAA4B,CAAC,EAAE,CAAC,YAAY,EAAE,YAAY,KAAK,YAAY,GAAG,SAAS,CAAA;IACvF,aAAa,CAAC,EAAE,oBAAoB,CAClC,yBAAyB,CAAC,KAAK,EAAE,OAAO,CAAC,EACzC,0BAA0B,EAC1B,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,CAC3C,CAAC,eAAe,CAAC,CAAA;CACnB;AAKD,wBAAsB,6BAA6B,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EAC1E,MAAM,EAAE,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,GACjD,OAAO,CACR,gBAAgB,CACd,yBAAyB,CAAC,KAAK,EAAE,OAAO,CAAC,EACzC,0BAA0B,EAC1B,0BAA0B,CAAC,KAAK,EAAE,OAAO,CAAC,CAC3C,CACF,CAoKA;AAED,wBAAgB,mCAAmC,CACjD,YAAY,EAAE,YAAY,GACzB,YAAY,GAAG,SAAS,CAG1B"}
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"propose-review-control.test.d.ts","sourceRoot":"","sources":["../src/propose-review-control.test.ts"],"names":[],"mappings":""}
|
|
@@ -0,0 +1,155 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Propose / Verify / Review — the core multi-shot primitive.
|
|
3
|
+
*
|
|
4
|
+
* shot N: propose(state, priorReview) → new state
|
|
5
|
+
* verify(state) → pass/fail, optional layers
|
|
6
|
+
* review(state, verification, memory) → observations + next-shot
|
|
7
|
+
* instruction + shouldContinue
|
|
8
|
+
* memory.append(entry)
|
|
9
|
+
*
|
|
10
|
+
* Roles are strictly separated:
|
|
11
|
+
*
|
|
12
|
+
* - The WORKER is whatever the caller wraps in `propose`. It is
|
|
13
|
+
* stateful — caller owns its resume/session mechanism.
|
|
14
|
+
* - The VERIFIER grades the state. It produces the ground truth.
|
|
15
|
+
* The reviewer cannot overturn or downgrade a verification layer.
|
|
16
|
+
* - The REVIEWER is stateless per call. Its continuity is the
|
|
17
|
+
* `ReviewMemoryStore` — durable JSONL by default, or any store
|
|
18
|
+
* implementing the interface. It reads memory + trace summary +
|
|
19
|
+
* verification and directs the NEXT proposer shot.
|
|
20
|
+
*
|
|
21
|
+
* This shape is load-bearing. The reviewer never grades; the verifier
|
|
22
|
+
* never directs. Two processes, two prompts, two concerns — which is
|
|
23
|
+
* what keeps the loop from confirmation-biasing itself into "all
|
|
24
|
+
* passed" when it didn't.
|
|
25
|
+
*
|
|
26
|
+
* Short-circuits and soft-fails are both first-class:
|
|
27
|
+
* - verify.pass === true → reviewer LLM call is skipped, memory
|
|
28
|
+
* records a success entry, loop exits.
|
|
29
|
+
* - review throws → the shot still counts; the loop uses the
|
|
30
|
+
* last-known instruction (or `fallbackInstruction`) for the next
|
|
31
|
+
* propose call. A transient reviewer failure must NEVER abort a
|
|
32
|
+
* valid arc.
|
|
33
|
+
*
|
|
34
|
+
* Composable: `propose` itself can be another `runProposeReview` call.
|
|
35
|
+
* That's the dogfooding path — a harness built on this primitive is in
|
|
36
|
+
* turn evaluable by it.
|
|
37
|
+
*/
|
|
38
|
+
import { TraceEmitter } from './trace/emitter';
|
|
39
|
+
import type { FailureClass } from './trace/schema';
|
|
40
|
+
import type { TraceStore } from './trace/store';
|
|
41
|
+
export interface Verification {
|
|
42
|
+
pass: boolean;
|
|
43
|
+
score?: number;
|
|
44
|
+
failingLayers?: string[];
|
|
45
|
+
details?: unknown;
|
|
46
|
+
}
|
|
47
|
+
export interface Review {
|
|
48
|
+
observations: string;
|
|
49
|
+
diagnosis: string;
|
|
50
|
+
nextShotInstruction: string;
|
|
51
|
+
shouldContinue: boolean;
|
|
52
|
+
confidence: number;
|
|
53
|
+
}
|
|
54
|
+
export interface ReviewMemoryEntry extends Review {
|
|
55
|
+
shot: number;
|
|
56
|
+
timestamp: number;
|
|
57
|
+
verification: {
|
|
58
|
+
pass: boolean;
|
|
59
|
+
score?: number;
|
|
60
|
+
failingLayers?: string[];
|
|
61
|
+
};
|
|
62
|
+
}
|
|
63
|
+
export interface ProposeInput<State> {
|
|
64
|
+
shot: number;
|
|
65
|
+
goal: string;
|
|
66
|
+
state: State;
|
|
67
|
+
priorReview: Review | null;
|
|
68
|
+
abortSignal: AbortSignal;
|
|
69
|
+
emitter?: TraceEmitter;
|
|
70
|
+
}
|
|
71
|
+
export interface ProposeOutput<State, Summary = unknown> {
|
|
72
|
+
state: State;
|
|
73
|
+
traceSummary?: Summary;
|
|
74
|
+
}
|
|
75
|
+
export interface ReviewInput<State, Summary = unknown> {
|
|
76
|
+
shot: number;
|
|
77
|
+
goal: string;
|
|
78
|
+
state: State;
|
|
79
|
+
verification: Verification;
|
|
80
|
+
traceSummary: Summary | undefined;
|
|
81
|
+
memory: ReviewMemoryEntry[];
|
|
82
|
+
}
|
|
83
|
+
export type ProposeFn<State, Summary = unknown> = (input: ProposeInput<State>) => Promise<ProposeOutput<State, Summary>>;
|
|
84
|
+
export type VerifyFn<State> = (state: State) => Promise<Verification>;
|
|
85
|
+
export type ReviewFn<State, Summary = unknown> = (input: ReviewInput<State, Summary>) => Promise<Review>;
|
|
86
|
+
export interface ReviewMemoryStore {
|
|
87
|
+
load(): Promise<ReviewMemoryEntry[]>;
|
|
88
|
+
append(entry: ReviewMemoryEntry): Promise<void>;
|
|
89
|
+
}
|
|
90
|
+
export interface ProposeReviewConfig<State, Summary = unknown> {
|
|
91
|
+
goal: string;
|
|
92
|
+
initialState: State;
|
|
93
|
+
propose: ProposeFn<State, Summary>;
|
|
94
|
+
verify: VerifyFn<State>;
|
|
95
|
+
review: ReviewFn<State, Summary>;
|
|
96
|
+
/** Hard shot cap. Default 10. */
|
|
97
|
+
maxShots?: number;
|
|
98
|
+
/** Wall-clock cap in ms. Default 10 min. */
|
|
99
|
+
maxWallMs?: number;
|
|
100
|
+
/**
|
|
101
|
+
* If the reviewer returns confidence ≤ floor on `confidenceFloorWindow`
|
|
102
|
+
* consecutive shots, terminate early. Default floor 0.3, window 2.
|
|
103
|
+
* Set window to 0 or floor to <0 to disable.
|
|
104
|
+
*/
|
|
105
|
+
confidenceFloor?: number;
|
|
106
|
+
confidenceFloorWindow?: number;
|
|
107
|
+
/** Defaults to an in-memory store if omitted. */
|
|
108
|
+
memory?: ReviewMemoryStore;
|
|
109
|
+
/** If provided, emit a Run + per-shot spans. */
|
|
110
|
+
store?: TraceStore;
|
|
111
|
+
scenarioId?: string;
|
|
112
|
+
projectId?: string;
|
|
113
|
+
variantId?: string;
|
|
114
|
+
/**
|
|
115
|
+
* Used when the reviewer soft-fails on shot 1 (no prior instruction to
|
|
116
|
+
* fall back to). Default is a generic "inspect failures and fix".
|
|
117
|
+
*/
|
|
118
|
+
fallbackInstruction?: string;
|
|
119
|
+
}
|
|
120
|
+
export interface ProposeReviewShot<State, Summary = unknown> {
|
|
121
|
+
shot: number;
|
|
122
|
+
state: State;
|
|
123
|
+
verification: Verification;
|
|
124
|
+
traceSummary: Summary | undefined;
|
|
125
|
+
review: Review;
|
|
126
|
+
reviewAvailable: boolean;
|
|
127
|
+
reviewError?: string;
|
|
128
|
+
durationMs: number;
|
|
129
|
+
}
|
|
130
|
+
export interface ProposeReviewReport<State, Summary = unknown> {
|
|
131
|
+
runId: string | null;
|
|
132
|
+
completed: boolean;
|
|
133
|
+
shots: ProposeReviewShot<State, Summary>[];
|
|
134
|
+
finalState: State;
|
|
135
|
+
finalVerification: Verification;
|
|
136
|
+
failureClass?: FailureClass;
|
|
137
|
+
wallMs: number;
|
|
138
|
+
score: number;
|
|
139
|
+
}
|
|
140
|
+
export declare function inMemoryReviewStore(initial?: ReviewMemoryEntry[]): ReviewMemoryStore;
|
|
141
|
+
export declare function jsonlReviewStore(path: string): ReviewMemoryStore;
|
|
142
|
+
export declare function runProposeReview<State, Summary = unknown>(config: ProposeReviewConfig<State, Summary>): Promise<ProposeReviewReport<State, Summary>>;
|
|
143
|
+
export type LlmJsonCall = (req: {
|
|
144
|
+
system: string;
|
|
145
|
+
user: string;
|
|
146
|
+
}) => Promise<unknown>;
|
|
147
|
+
export interface LlmReviewerConfig<State, Summary = unknown> {
|
|
148
|
+
callJson: LlmJsonCall;
|
|
149
|
+
renderState?: (state: State) => string;
|
|
150
|
+
renderTraceSummary?: (summary: Summary | undefined) => string;
|
|
151
|
+
/** Appended to the default system prompt. */
|
|
152
|
+
systemPromptAddendum?: string;
|
|
153
|
+
}
|
|
154
|
+
export declare function createLlmReviewer<State, Summary = unknown>(cfg: LlmReviewerConfig<State, Summary>): ReviewFn<State, Summary>;
|
|
155
|
+
//# sourceMappingURL=propose-review.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"propose-review.d.ts","sourceRoot":"","sources":["../src/propose-review.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAoCG;AAIH,OAAO,EAAmB,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAC/D,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAClD,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAI/C,MAAM,WAAW,YAAY;IAC3B,IAAI,EAAE,OAAO,CAAA;IACb,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;IACxB,OAAO,CAAC,EAAE,OAAO,CAAA;CAClB;AAED,MAAM,WAAW,MAAM;IACrB,YAAY,EAAE,MAAM,CAAA;IACpB,SAAS,EAAE,MAAM,CAAA;IACjB,mBAAmB,EAAE,MAAM,CAAA;IAC3B,cAAc,EAAE,OAAO,CAAA;IACvB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,iBAAkB,SAAQ,MAAM;IAC/C,IAAI,EAAE,MAAM,CAAA;IACZ,SAAS,EAAE,MAAM,CAAA;IACjB,YAAY,EAAE;QACZ,IAAI,EAAE,OAAO,CAAA;QACb,KAAK,CAAC,EAAE,MAAM,CAAA;QACd,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;KACzB,CAAA;CACF;AAED,MAAM,WAAW,YAAY,CAAC,KAAK;IACjC,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,WAAW,EAAE,MAAM,GAAG,IAAI,CAAA;IAC1B,WAAW,EAAE,WAAW,CAAA;IACxB,OAAO,CAAC,EAAE,YAAY,CAAA;CACvB;AAED,MAAM,WAAW,aAAa,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACrD,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,CAAC,EAAE,OAAO,CAAA;CACvB;AAED,MAAM,WAAW,WAAW,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACnD,IAAI,EAAE,MAAM,CAAA;IACZ,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,EAAE,OAAO,GAAG,SAAS,CAAA;IACjC,MAAM,EAAE,iBAAiB,EAAE,CAAA;CAC5B;AAED,MAAM,MAAM,SAAS,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,IAAI,CAChD,KAAK,EAAE,YAAY,CAAC,KAAK,CAAC,KACvB,OAAO,CAAC,aAAa,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CAAA;AAE3C,MAAM,MAAM,QAAQ,CAAC,KAAK,IAAI,CAAC,KAAK,EAAE,KAAK,KAAK,OAAO,CAAC,YAAY,CAAC,CAAA;AAErE,MAAM,MAAM,QAAQ,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,IAAI,CAC/C,KAAK,EAAE,WAAW,CAAC,KAAK,EAAE,OAAO,CAAC,KAC/B,OAAO,CAAC,MAAM,CAAC,CAAA;AAEpB,MAAM,WAAW,iBAAiB;IAChC,IAAI,IAAI,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;IACpC,MAAM,CAAC,KAAK,EAAE,iBAAiB,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;CAChD;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAC3D,IAAI,EAAE,MAAM,CAAA;IACZ,YAAY,EAAE,KAAK,CAAA;IACnB,OAAO,EAAE,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAClC,MAAM,EAAE,QAAQ,CAAC,KAAK,CAAC,CAAA;IACvB,MAAM,EAAE,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;IAChC,iCAAiC;IACjC,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,4CAA4C;IAC5C,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;;OAIG;IACH,eAAe,CAAC,EAAE,MAAM,CAAA;IACxB,qBAAqB,CAAC,EAAE,MAAM,CAAA;IAC9B,iDAAiD;IACjD,MAAM,CAAC,EAAE,iBAAiB,CAAA;IAC1B,gDAAgD;IAChD,KAAK,CAAC,EAAE,UAAU,CAAA;IAClB,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;OAGG;IACH,mBAAmB,CAAC,EAAE,MAAM,CAAA;CAC7B;AAED,MAAM,WAAW,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACzD,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,KAAK,CAAA;IACZ,YAAY,EAAE,YAAY,CAAA;IAC1B,YAAY,EAAE,OAAO,GAAG,SAAS,CAAA;IACjC,MAAM,EAAE,MAAM,CAAA;IACd,eAAe,EAAE,OAAO,CAAA;IACxB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,UAAU,EAAE,MAAM,CAAA;CACnB;AAED,MAAM,WAAW,mBAAmB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IAC3D,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,SAAS,EAAE,OAAO,CAAA;IAClB,KAAK,EAAE,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,EAAE,CAAA;IAC1C,UAAU,EAAE,KAAK,CAAA;IACjB,iBAAiB,EAAE,YAAY,CAAA;IAC/B,YAAY,CAAC,EAAE,YAAY,CAAA;IAC3B,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;CACd;AAID,wBAAgB,mBAAmB,CAAC,OAAO,GAAE,iBAAiB,EAAO,GAAG,iBAAiB,CAUxF;AAED,wBAAgB,gBAAgB,CAAC,IAAI,EAAE,MAAM,GAAG,iBAAiB,CAuBhE;AAOD,wBAAsB,gBAAgB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EAC7D,MAAM,EAAE,mBAAmB,CAAC,KAAK,EAAE,OAAO,CAAC,GAC1C,OAAO,CAAC,mBAAmB,CAAC,KAAK,EAAE,OAAO,CAAC,CAAC,CA6M9C;AAID,MAAM,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,KAAK,OAAO,CAAC,OAAO,CAAC,CAAA;AAErF,MAAM,WAAW,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO;IACzD,QAAQ,EAAE,WAAW,CAAA;IACrB,WAAW,CAAC,EAAE,CAAC,KAAK,EAAE,KAAK,KAAK,MAAM,CAAA;IACtC,kBAAkB,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,GAAG,SAAS,KAAK,MAAM,CAAA;IAC7D,6CAA6C;IAC7C,oBAAoB,CAAC,EAAE,MAAM,CAAA;CAC9B;AAOD,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,OAAO,EACxD,GAAG,EAAE,iBAAiB,CAAC,KAAK,EAAE,OAAO,CAAC,GACrC,QAAQ,CAAC,KAAK,EAAE,OAAO,CAAC,CA6D1B"}
|
|
@@ -1,6 +1,3 @@
|
|
|
1
|
-
import { a as DatasetScenario, b as Dataset } from './dataset-BbGkaN2I.js';
|
|
2
|
-
import { T as TraceStore } from './store-CKUAgsJz.js';
|
|
3
|
-
|
|
4
1
|
/**
|
|
5
2
|
* Red-team battery — adversarial scenario corpus with per-category
|
|
6
3
|
* scorers.
|
|
@@ -13,9 +10,11 @@ import { T as TraceStore } from './store-CKUAgsJz.js';
|
|
|
13
10
|
* The built-in corpus is intentionally conservative — 40 cases
|
|
14
11
|
* covering the most-seen attack patterns. Extend via `extendCorpus`.
|
|
15
12
|
*/
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
13
|
+
import type { DatasetScenario } from './dataset';
|
|
14
|
+
import { Dataset } from './dataset';
|
|
15
|
+
import type { TraceStore } from './trace/store';
|
|
16
|
+
export type RedTeamCategory = 'prompt_injection_direct' | 'prompt_injection_indirect' | 'jailbreak_persona' | 'jailbreak_dan' | 'pii_leak' | 'permission_escalation' | 'data_exfil' | 'policy_override';
|
|
17
|
+
export interface RedTeamPayload {
|
|
19
18
|
category: RedTeamCategory;
|
|
20
19
|
/** The hostile input the agent sees. */
|
|
21
20
|
input: string;
|
|
@@ -29,35 +28,34 @@ interface RedTeamPayload {
|
|
|
29
28
|
/** Tool names the agent MUST NOT call for this scenario. */
|
|
30
29
|
forbiddenTools?: string[];
|
|
31
30
|
}
|
|
32
|
-
interface RedTeamCase extends DatasetScenario {
|
|
31
|
+
export interface RedTeamCase extends DatasetScenario {
|
|
33
32
|
payload: RedTeamPayload;
|
|
34
33
|
}
|
|
35
|
-
interface RedTeamFinding {
|
|
34
|
+
export interface RedTeamFinding {
|
|
36
35
|
scenarioId: string;
|
|
37
36
|
category: RedTeamCategory;
|
|
38
37
|
passed: boolean;
|
|
39
38
|
reason: string;
|
|
40
39
|
evidence?: string;
|
|
41
40
|
}
|
|
42
|
-
interface RedTeamReport {
|
|
41
|
+
export interface RedTeamReport {
|
|
43
42
|
findings: RedTeamFinding[];
|
|
44
43
|
passRateByCategory: Record<RedTeamCategory, number>;
|
|
45
44
|
overallPassRate: number;
|
|
46
45
|
}
|
|
47
46
|
/** Ship a canonical, small corpus. Consumers extend via `extendCorpus`. */
|
|
48
|
-
declare const DEFAULT_RED_TEAM_CORPUS: RedTeamCase[];
|
|
49
|
-
declare function redTeamDataset(extraCases?: RedTeamCase[]): Dataset;
|
|
47
|
+
export declare const DEFAULT_RED_TEAM_CORPUS: RedTeamCase[];
|
|
48
|
+
export declare function redTeamDataset(extraCases?: RedTeamCase[]): Dataset;
|
|
50
49
|
/**
|
|
51
50
|
* Score a single agent response against the red-team case.
|
|
52
51
|
* `toolCalls` should be the names of tools the agent invoked during the run.
|
|
53
52
|
*/
|
|
54
|
-
declare function scoreRedTeamOutput(output: string, toolCalls: string[], rtCase: RedTeamCase): RedTeamFinding;
|
|
53
|
+
export declare function scoreRedTeamOutput(output: string, toolCalls: string[], rtCase: RedTeamCase): RedTeamFinding;
|
|
55
54
|
/** Aggregate red-team findings into per-category pass rates. */
|
|
56
|
-
declare function redTeamReport(findings: RedTeamFinding[]): RedTeamReport;
|
|
55
|
+
export declare function redTeamReport(findings: RedTeamFinding[]): RedTeamReport;
|
|
57
56
|
/**
|
|
58
57
|
* Extract the tool-call names from a corpus run — convenience for the
|
|
59
58
|
* common pipeline (run the scenario → score the run).
|
|
60
59
|
*/
|
|
61
|
-
declare function toolNamesForRun(store: TraceStore, runId: string): Promise<string[]>;
|
|
62
|
-
|
|
63
|
-
export { DEFAULT_RED_TEAM_CORPUS as D, type RedTeamCase as R, type RedTeamCategory as a, type RedTeamFinding as b, type RedTeamPayload as c, type RedTeamReport as d, redTeamReport as e, redTeamDataset as r, scoreRedTeamOutput as s, toolNamesForRun as t };
|
|
60
|
+
export declare function toolNamesForRun(store: TraceStore, runId: string): Promise<string[]>;
|
|
61
|
+
//# sourceMappingURL=red-team.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"red-team.d.ts","sourceRoot":"","sources":["../src/red-team.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAA;AAChD,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAA;AAGnC,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,eAAe,CAAA;AAE/C,MAAM,MAAM,eAAe,GACvB,yBAAyB,GACzB,2BAA2B,GAC3B,mBAAmB,GACnB,eAAe,GACf,UAAU,GACV,uBAAuB,GACvB,YAAY,GACZ,iBAAiB,CAAA;AAErB,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,eAAe,CAAA;IACzB,wCAAwC;IACxC,KAAK,EAAE,MAAM,CAAA;IACb;;;OAGG;IACH,QAAQ,EAAE,QAAQ,GAAG,QAAQ,CAAA;IAC7B,mDAAmD;IACnD,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC3B,4DAA4D;IAC5D,cAAc,CAAC,EAAE,MAAM,EAAE,CAAA;CAC1B;AAED,MAAM,WAAW,WAAY,SAAQ,eAAe;IAClD,OAAO,EAAE,cAAc,CAAA;CACxB;AAED,MAAM,WAAW,cAAc;IAC7B,UAAU,EAAE,MAAM,CAAA;IAClB,QAAQ,EAAE,eAAe,CAAA;IACzB,MAAM,EAAE,OAAO,CAAA;IACf,MAAM,EAAE,MAAM,CAAA;IACd,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED,MAAM,WAAW,aAAa;IAC5B,QAAQ,EAAE,cAAc,EAAE,CAAA;IAC1B,kBAAkB,EAAE,MAAM,CAAC,eAAe,EAAE,MAAM,CAAC,CAAA;IACnD,eAAe,EAAE,MAAM,CAAA;CACxB;AAUD,2EAA2E;AAC3E,eAAO,MAAM,uBAAuB,EAAE,WAAW,EA2FhD,CAAA;AAED,wBAAgB,cAAc,CAAC,UAAU,GAAE,WAAW,EAAO,GAAG,OAAO,CAYtE;AAED;;;GAGG;AACH,wBAAgB,kBAAkB,CAChC,MAAM,EAAE,MAAM,EACd,SAAS,EAAE,MAAM,EAAE,EACnB,MAAM,EAAE,WAAW,GAClB,cAAc,CA8DhB;AAED,gEAAgE;AAChE,wBAAgB,aAAa,CAAC,QAAQ,EAAE,cAAc,EAAE,GAAG,aAAa,CAevE;AAED;;;GAGG;AACH,wBAAsB,eAAe,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAGzF"}
|
|
@@ -0,0 +1,11 @@
|
|
|
1
|
+
import type { ReferenceReplayCaseRun, ReferenceReplayRun, ReferenceReplayScenarioScore } from './reference-replay';
|
|
2
|
+
import type { RunScore } from './run-score';
|
|
3
|
+
import type { SteeringBundle } from './steering';
|
|
4
|
+
import type { SteeringOptimizationRow } from './steering-optimizer';
|
|
5
|
+
export interface ReferenceReplaySteeringRowsOptions<Input = unknown> {
|
|
6
|
+
bundleForRun?: (run: ReferenceReplayRun<Input>) => SteeringBundle;
|
|
7
|
+
scoreForCase?: (caseRun: ReferenceReplayCaseRun<Input>, run: ReferenceReplayRun<Input>) => RunScore;
|
|
8
|
+
}
|
|
9
|
+
export declare function referenceReplayRunsToSteeringRows<Input = unknown>(runs: ReferenceReplayRun<Input>[], options?: ReferenceReplaySteeringRowsOptions<Input>): SteeringOptimizationRow[];
|
|
10
|
+
export declare function referenceReplayScenarioToRunScore(scenarioScore: ReferenceReplayScenarioScore, durationMs?: number): RunScore;
|
|
11
|
+
//# sourceMappingURL=reference-replay-steering.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"reference-replay-steering.d.ts","sourceRoot":"","sources":["../src/reference-replay-steering.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,sBAAsB,EACtB,kBAAkB,EAClB,4BAA4B,EAC7B,MAAM,oBAAoB,CAAA;AAC3B,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAA;AAC3C,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAChD,OAAO,KAAK,EAAE,uBAAuB,EAAE,MAAM,sBAAsB,CAAA;AAEnE,MAAM,WAAW,kCAAkC,CAAC,KAAK,GAAG,OAAO;IACjE,YAAY,CAAC,EAAE,CAAC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,KAAK,cAAc,CAAA;IACjE,YAAY,CAAC,EAAE,CACb,OAAO,EAAE,sBAAsB,CAAC,KAAK,CAAC,EACtC,GAAG,EAAE,kBAAkB,CAAC,KAAK,CAAC,KAC3B,QAAQ,CAAA;CACd;AAED,wBAAgB,iCAAiC,CAAC,KAAK,GAAG,OAAO,EAC/D,IAAI,EAAE,kBAAkB,CAAC,KAAK,CAAC,EAAE,EACjC,OAAO,GAAE,kCAAkC,CAAC,KAAK,CAAM,GACtD,uBAAuB,EAAE,CAoC3B;AAED,wBAAgB,iCAAiC,CAC/C,aAAa,EAAE,4BAA4B,EAC3C,UAAU,SAAI,GACb,QAAQ,CAuBV"}
|