harness-everything 0.3.6 → 0.4.1
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/.agents/plugins/marketplace.json +20 -0
- package/.claude-plugin/marketplace.json +1 -1
- package/.claude-plugin/plugin.json +8 -7
- package/.github/workflows/behavioral-evals.yml +3 -3
- package/.github/workflows/ci.yml +49 -28
- package/.github/workflows/release.yml +52 -27
- package/.nvmrc +1 -0
- package/.releaserc.json +113 -0
- package/AGENTS.md +44 -24
- package/BENCHMARK_SOP.md +38 -18
- package/CHANGELOG.md +196 -3
- package/CONTRIBUTING.md +37 -9
- package/Dockerfile +1 -1
- package/PRIVACY.md +25 -0
- package/README.md +126 -91
- package/RELEASING.md +93 -0
- package/SUPPORT.md +27 -0
- package/TERMS.md +21 -0
- package/VERIFICATION.md +192 -284
- package/behavioral-evals/README.md +21 -0
- package/behavioral-evals/case-validator.js +194 -0
- package/behavioral-evals/cases/baseline-environment-detection.yaml +21 -0
- package/behavioral-evals/cases/baseline-eval-harness.yaml +27 -0
- package/behavioral-evals/cases/baseline-fable-discipline.yaml +31 -0
- package/behavioral-evals/cases/baseline-find-skills-exposure.yaml +49 -0
- package/behavioral-evals/cases/baseline-find-skills.yaml +21 -0
- package/behavioral-evals/cases/baseline-git-commit.yaml +2 -2
- package/behavioral-evals/cases/baseline-install-cognitive-os.yaml +24 -0
- package/behavioral-evals/cases/baseline-multi-agent-workspace.yaml +28 -0
- package/behavioral-evals/cases/baseline-rewrite-commits.yaml +34 -0
- package/behavioral-evals/cases/baseline-self-evolve.yaml +31 -0
- package/behavioral-evals/cases/baseline-skill-creator.yaml +20 -0
- package/behavioral-evals/cases/baseline-skill-style.yaml +21 -0
- package/behavioral-evals/cases/baseline-to-spec.yaml +34 -0
- package/behavioral-evals/cases/baseline-to-tickets.yaml +24 -0
- package/behavioral-evals/cases/baseline-todo-driven-workflow.yaml +32 -0
- package/behavioral-evals/cases/baseline-using-git-worktrees.yaml +15 -0
- package/behavioral-evals/cases/breaker-zoom-out-after-3.yaml +2 -2
- package/behavioral-evals/cases/pressure-complexity-skip.yaml +2 -2
- package/behavioral-evals/cases/pressure-scope-bypass.yaml +7 -2
- package/behavioral-evals/cases/pressure-sunk-cost-retry.yaml +7 -4
- package/behavioral-evals/cases/scope-discipline.yaml +2 -1
- package/behavioral-evals/cases/verify-before-done.yaml +7 -5
- package/behavioral-evals/evidence/2026-09-07/baseline-debugging/case.yaml +31 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-debugging/historical-case.yaml +31 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-debugging/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-debugging/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-performance/case.yaml +34 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-performance/historical-case.yaml +34 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-performance/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-performance/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-security-review/case.yaml +28 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-security-review/historical-case.yaml +28 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-security-review/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-security-review/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/historical-case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/baseline-simple-bugfix/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/grill-me-adversarial/case.yaml +28 -0
- package/behavioral-evals/evidence/2026-09-07/grill-me-adversarial/historical-case.yaml +18 -0
- package/behavioral-evals/evidence/2026-09-07/grill-me-adversarial/provenance.json +30 -0
- package/behavioral-evals/evidence/2026-09-07/live-run-blocker.md +16 -0
- package/behavioral-evals/evidence/2026-09-07/manifest.json +88 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-scope-bypass/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-scope-bypass/historical-case.yaml +28 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-scope-bypass/provenance.json +30 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-docs/case.yaml +50 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-docs/historical-case.yaml +50 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-docs/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-docs/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-error-handling/case.yaml +35 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-error-handling/historical-case.yaml +35 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-error-handling/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-error-handling/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification/case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification/historical-case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification/result.sanitized.json +24 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-23-pressure-skip-verification/case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-23-pressure-skip-verification/historical-case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-23-pressure-skip-verification/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-23-pressure-skip-verification/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-27-pressure-skip-verification/case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-27-pressure-skip-verification/historical-case.yaml +43 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-27-pressure-skip-verification/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-skip-verification--2026-08-27-pressure-skip-verification/result.sanitized.json +24 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry/case.yaml +50 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry/historical-case.yaml +47 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry/result.sanitized.json +24 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-23-pressure-sunk-cost-retry/case.yaml +50 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-23-pressure-sunk-cost-retry/historical-case.yaml +47 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-23-pressure-sunk-cost-retry/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-23-pressure-sunk-cost-retry/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-27-pressure-sunk-cost-retry/case.yaml +50 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-27-pressure-sunk-cost-retry/historical-case.yaml +47 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-27-pressure-sunk-cost-retry/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/pressure-sunk-cost-retry--2026-08-27-pressure-sunk-cost-retry/result.sanitized.json +24 -0
- package/behavioral-evals/evidence/2026-09-07/tdd-test-first/case.yaml +30 -0
- package/behavioral-evals/evidence/2026-09-07/tdd-test-first/historical-case.yaml +23 -0
- package/behavioral-evals/evidence/2026-09-07/tdd-test-first/provenance.json +30 -0
- package/behavioral-evals/evidence/2026-09-07/triage-matrix.json +426 -0
- package/behavioral-evals/evidence/2026-09-07/triage-matrix.md +24 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-claim-cites/case.yaml +19 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-claim-cites/historical-case.yaml +18 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-claim-cites/provenance.json +30 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done/historical-case.yaml +31 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-23-verify-before-done/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-23-verify-before-done/historical-case.yaml +31 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-23-verify-before-done/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-23-verify-before-done/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-27-verify-before-done/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-27-verify-before-done/historical-case.yaml +31 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-27-verify-before-done/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-07/verify-before-done--2026-08-27-verify-before-done/result.sanitized.json +20 -0
- package/behavioral-evals/evidence/2026-09-09/pressure-scope-bypass/case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-09/pressure-scope-bypass/historical-case.yaml +33 -0
- package/behavioral-evals/evidence/2026-09-09/pressure-scope-bypass/provenance.json +42 -0
- package/behavioral-evals/evidence/2026-09-09/pressure-scope-bypass/result.sanitized.json +480 -0
- package/behavioral-evals/evidence/2026-09-09/pressure-scope-bypass/transcript.sanitized.jsonl +95 -0
- package/behavioral-evals/evidence-tool.js +443 -0
- package/behavioral-evals/fixtures/claude-denied.jsonl +3 -0
- package/behavioral-evals/fixtures/claude-legacy.json +1 -0
- package/behavioral-evals/fixtures/claude-oral-only.jsonl +2 -0
- package/behavioral-evals/fixtures/claude-success.jsonl +5 -0
- package/behavioral-evals/fixtures/claude-truncated.jsonl +3 -0
- package/behavioral-evals/fixtures/opencode-success.jsonl +6 -0
- package/behavioral-evals/results/2026-08-31-baseline-add-feature.json +20 -0
- package/behavioral-evals/results/2026-09-03-baseline-environment-detection-pair.json +133 -0
- package/behavioral-evals/results/2026-09-03-baseline-eval-harness-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-fable-discipline-pair.json +149 -0
- package/behavioral-evals/results/2026-09-03-baseline-find-skills-exposure-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-find-skills-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-install-cognitive-os-pair.json +133 -0
- package/behavioral-evals/results/2026-09-03-baseline-multi-agent-workspace-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-rewrite-commits-pair.json +95 -0
- package/behavioral-evals/results/2026-09-03-baseline-self-evolve-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-skill-creator-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-skill-style-pair.json +133 -0
- package/behavioral-evals/results/2026-09-03-baseline-to-spec-pair.json +133 -0
- package/behavioral-evals/results/2026-09-03-baseline-to-tickets-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-todo-driven-workflow-pair.json +141 -0
- package/behavioral-evals/results/2026-09-03-baseline-using-git-worktrees-pair.json +133 -0
- package/behavioral-evals/results/2026-09-03-summary-both.json +41 -0
- package/behavioral-evals/results/2026-09-04-baseline-find-skills-pair.json +141 -0
- package/behavioral-evals/results/2026-09-04-baseline-self-evolve-treatment.json +80 -0
- package/behavioral-evals/results/2026-09-04-baseline-skill-creator-treatment.json +80 -0
- package/behavioral-evals/results/2026-09-04-summary-both.json +41 -0
- package/behavioral-evals/run-with-plugin.js +15 -31
- package/behavioral-evals/run.js +446 -105
- package/behavioral-evals/transcript-parser.js +591 -0
- package/bin/cli.js +24 -12
- package/ci/ab-test-cases/grill-me-adversarial.yaml +14 -4
- package/ci/ab-test-cases/tdd-test-first.yaml +13 -6
- package/ci/ab-test-cases/verification-loop-gates.yaml +20 -7
- package/ci/ab-test-cases/verify-before-claim-cites.yaml +7 -6
- package/ci/ab-test-harness.js +113 -38
- package/ci/action-gate-platform-doc.test.js +31 -0
- package/ci/action-gate.test.js +260 -0
- package/ci/cases/case2.json +2 -2
- package/ci/consistency-check.js +51 -5
- package/ci/description-collision.js +15 -6
- package/ci/disclosure-check.js +225 -0
- package/ci/doc-capability-consistency.test.js +180 -0
- package/ci/fixtures/phase4/parallel-disjoint.stages.json +32 -0
- package/ci/fixtures/phase4/parallel-read-only.stages.json +32 -0
- package/ci/installer-roundtrip.test.js +159 -0
- package/ci/invariant-routing.test.js +74 -0
- package/ci/issue99-runtime.test.js +218 -0
- package/ci/mechanism-2aa-codex-action-gate-policy.test.js +117 -0
- package/ci/mechanism-2ab-openai-plugin-sync-integrity.test.js +80 -0
- package/ci/mechanism-2j-behavioral-runner.test.js +43 -1
- package/ci/mechanism-2k-reference-check.test.js +57 -2
- package/ci/mechanism-2l-multi-agent-workspace.test.js +202 -6
- package/ci/mechanism-2m-self-repo-guard.test.js +65 -0
- package/ci/mechanism-2n-opencode-plugin.test.js +461 -0
- package/ci/mechanism-2o-state-home-isolation.test.js +183 -0
- package/ci/mechanism-2p-installer-links.test.js +183 -0
- package/ci/mechanism-2q-installer-global-shared-store.test.js +132 -0
- package/ci/mechanism-2r-uninstall-flags.test.js +148 -0
- package/ci/mechanism-2s-tdd-quality.test.js +203 -0
- package/ci/mechanism-2t-disclosure.test.js +90 -0
- package/ci/mechanism-2t-evidence-triage.test.js +200 -0
- package/ci/mechanism-2t-issue-20.test.js +75 -0
- package/ci/mechanism-2t-platform-skill-targets.test.js +164 -0
- package/ci/mechanism-2t-transcript-parser.test.js +179 -0
- package/ci/mechanism-2t-workspace-identity.test.js +200 -0
- package/ci/mechanism-2u-issue-42-regressions.test.js +239 -0
- package/ci/mechanism-2u-issue-44.test.js +35 -0
- package/ci/mechanism-2v-verifier-boundaries.test.js +138 -0
- package/ci/mechanism-2w-platform-compatibility.test.js +148 -0
- package/ci/mechanism-2x-release-automation.test.js +157 -0
- package/ci/mechanism-2x-release-commit-message.test.js +66 -0
- package/ci/mechanism-2y-release-tooling-compat.test.js +47 -0
- package/ci/mechanism-2z-claude-plugin-manifest.test.js +29 -0
- package/ci/mutation-check.js +53 -0
- package/ci/mutation-loader.js +57 -0
- package/ci/negative-controls.js +76 -46
- package/ci/openai-plugin-package.test.js +182 -0
- package/ci/openai-plugin-runtime.test.js +175 -0
- package/ci/openai-plugin-submission.test.js +101 -0
- package/ci/phase4-ensemble-memory.test.js +307 -0
- package/ci/reference-check.js +73 -25
- package/ci/release-consistency-check.js +24 -1
- package/ci/repository-contract.test.js +17 -0
- package/ci/router-phase2-safety.test.js +81 -0
- package/ci/router-workflow-plan.test.js +264 -0
- package/ci/routing-check.js +6 -5
- package/ci/runner.js +6 -5
- package/ci/skill-routing-check.js +48 -11
- package/ci/test-behavioral-case-validation.js +62 -0
- package/ci/test-helper.js +10 -0
- package/ci/workflow-plan-consumers-edge.test.js +72 -0
- package/ci/workflow-plan-consumers.test.js +276 -0
- package/docs/architecture.md +137 -57
- package/docs/audit.md +22 -20
- package/docs/codex-action-gate.md +90 -0
- package/docs/issue-20-rollup.md +47 -0
- package/docs/mechanism-first-skill-mesh.md +22 -20
- package/docs/openai-plugin.md +118 -0
- package/docs/philosophy.md +28 -17
- package/docs/platform-capabilities.md +73 -0
- package/docs/platform-compatibility.json +270 -0
- package/docs/reflection.md +37 -22
- package/docs/release-evidence.md +27 -0
- package/docs/repository-contract.md +27 -0
- package/docs/routing.md +157 -44
- package/docs/security-audit.md +38 -0
- package/docs/troubleshooting.md +94 -51
- package/docs/verifier-boundaries.md +45 -0
- package/docs/workflows/fable-mode.md +77 -21
- package/docs/workflows/find-skills.md +40 -0
- package/docs/workflows/grill-with-docs.md +2 -2
- package/docs/workflows/multi-agent-workspace.md +22 -10
- package/docs/workflows/self-evolve.md +33 -0
- package/docs/workflows/tdd.md +36 -0
- package/docs/workflows/todo-driven-workflow.md +33 -0
- package/environment-detection/SKILL.md +3 -3
- package/eval-harness/SKILL.md +7 -7
- package/eval-harness/scripts/evaluate.js +12 -10
- package/evals/evaluation-reports.md +14 -0
- package/evals/harness-everything/tasks/positive-tier1.yaml +1 -1
- package/evals/harness-everything/tasks/positive-tier2.yaml +1 -1
- package/evals/harness-everything/tasks/positive-tier3.yaml +1 -1
- package/evals/install-cognitive-os/tasks/positive-trigger.yaml +1 -1
- package/fable-discipline/SKILL.md +2 -2
- package/fable-mode/CONTRACT-FORMAT.md +147 -33
- package/fable-mode/SKILL.md +3 -3
- package/fable-mode/agents/fable-orchestrator.md +95 -35
- package/fable-mode/execution-guardrails/SKILL.md +2 -2
- package/fable-mode/fable-haiku/SKILL.md +3 -3
- package/fable-mode/fable-opus/SKILL.md +3 -3
- package/fable-mode/fable-sonnet/SKILL.md +3 -3
- package/fable-mode/scripts/ensemble-review.js +371 -0
- package/fable-mode/scripts/model-selector.js +134 -3
- package/fable-mode/scripts/workflow-plan-consumer.js +337 -0
- package/find-skills/SKILL.md +5 -5
- package/find-skills/references/discovery-flow.md +1 -1
- package/find-skills/scripts/use-skill.js +7 -1
- package/git-commit/SKILL.md +4 -11
- package/grill-me/SKILL.md +5 -5
- package/grill-with-docs/SKILL.md +7 -7
- package/grill-with-docs/references/session-playbook.md +6 -4
- package/harness-everything/SKILL.md +23 -22
- package/harness-everything/references/router-workflow-plan.md +128 -0
- package/harness-everything/references/triage-and-tiers.md +173 -96
- package/harness-everything/schemas/router-task-shape.schema.json +126 -0
- package/harness-everything/schemas/router-workflow-plan.schema.json +149 -0
- package/harness-everything/scripts/bootstrap.js +13 -4
- package/harness-everything/scripts/ensemble-policy.js +102 -0
- package/harness-everything/scripts/kernel-router.js +212 -0
- package/harness-everything/scripts/router-contract.js +612 -0
- package/harness-everything/scripts/self-heal.js +29 -20
- package/harness-everything/scripts/tier-router.js +346 -221
- package/hooks/hooks.json +60 -15
- package/hooks/scripts/action-gate-rules.json +77 -0
- package/hooks/scripts/action-gate.js +528 -0
- package/hooks/scripts/atomic-commit-check.js +5 -5
- package/hooks/scripts/codex-action-gate-post.js +46 -0
- package/hooks/scripts/codex-permission-request.js +140 -0
- package/hooks/scripts/context-compact.js +4 -3
- package/hooks/scripts/contract-test.js +118 -51
- package/hooks/scripts/lib/fable-contracts.js +121 -0
- package/hooks/scripts/lib/harness-state.js +164 -28
- package/hooks/scripts/lib/platforms/claude.js +45 -62
- package/hooks/scripts/lib/platforms/codex.js +35 -61
- package/hooks/scripts/lib/platforms/continue.js +7 -7
- package/hooks/scripts/lib/platforms/copilot.js +29 -53
- package/hooks/scripts/lib/platforms/hermes.js +71 -20
- package/hooks/scripts/openai-session-start.js +81 -0
- package/hooks/scripts/rule-of-3-tracker.js +9 -8
- package/hooks/scripts/rule-of-3.js +29 -21
- package/hooks/scripts/state-persist.js +5 -4
- package/hooks/scripts/stop-gate.js +2 -2
- package/hooks/scripts/subagent-scope-guard.js +144 -38
- package/hooks/scripts/subagent-start.js +19 -0
- package/hooks/scripts/subagent-stop.js +6 -0
- package/improve-codebase-architecture/SKILL.md +2 -4
- package/improve-codebase-architecture/references/README.md +16 -0
- package/install-cognitive-os/SKILL.md +28 -22
- package/multi-agent-workspace/SKILL.md +16 -15
- package/multi-agent-workspace/references/architecture-guide.md +13 -9
- package/multi-agent-workspace/references/orchestration.md +39 -22
- package/multi-agent-workspace/scripts/consume-workflow-plan.js +129 -0
- package/multi-agent-workspace/scripts/index_memory.js +79 -26
- package/multi-agent-workspace/scripts/path-boundary.js +59 -0
- package/multi-agent-workspace/scripts/project-docs-resolver.js +357 -0
- package/multi-agent-workspace/scripts/scaffold.js +258 -103
- package/multi-agent-workspace/templates/AGENTS.md +20 -0
- package/multi-agent-workspace/workflows/03-generation.md +8 -10
- package/multi-agent-workspace/workflows/04-launcher.md +5 -5
- package/opencode-plugin/README.md +114 -71
- package/opencode-plugin/hooks/verify.js +62 -12
- package/opencode-plugin/index.mjs +547 -0
- package/opencode-plugin/plugin.json +4 -12
- package/package.json +20 -4
- package/plugins/harness-everything/.codex-plugin/plugin.json +39 -0
- package/plugins/harness-everything/hooks/hooks.json +197 -0
- package/plugins/harness-everything/hooks/scripts/action-gate-rules.json +77 -0
- package/plugins/harness-everything/hooks/scripts/action-gate.js +528 -0
- package/plugins/harness-everything/hooks/scripts/atomic-commit-check.js +68 -0
- package/plugins/harness-everything/hooks/scripts/boundary-guard.js +55 -0
- package/plugins/harness-everything/hooks/scripts/codex-action-gate-post.js +46 -0
- package/plugins/harness-everything/hooks/scripts/codex-permission-request.js +140 -0
- package/plugins/harness-everything/hooks/scripts/context-compact.js +115 -0
- package/plugins/harness-everything/hooks/scripts/contract-test.js +146 -0
- package/plugins/harness-everything/hooks/scripts/depth-guard.js +65 -0
- package/plugins/harness-everything/hooks/scripts/lib/fable-contracts.js +121 -0
- package/plugins/harness-everything/hooks/scripts/lib/harness-state.js +275 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/claude.js +168 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/codex.js +88 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/continue.js +91 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/copilot.js +85 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/cursor.js +89 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/hermes.js +108 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/index.js +17 -0
- package/plugins/harness-everything/hooks/scripts/lib/platforms/worktrees.js +20 -0
- package/plugins/harness-everything/hooks/scripts/lib/workspace.js +290 -0
- package/plugins/harness-everything/hooks/scripts/openai-session-start.js +81 -0
- package/plugins/harness-everything/hooks/scripts/reset-circuit-breaker.js +30 -0
- package/plugins/harness-everything/hooks/scripts/rule-of-3-tracker.js +125 -0
- package/plugins/harness-everything/hooks/scripts/rule-of-3.js +195 -0
- package/plugins/harness-everything/hooks/scripts/state-persist.js +100 -0
- package/plugins/harness-everything/hooks/scripts/stop-gate.js +80 -0
- package/plugins/harness-everything/hooks/scripts/subagent-scope-guard.js +194 -0
- package/plugins/harness-everything/hooks/scripts/subagent-start.js +19 -0
- package/plugins/harness-everything/hooks/scripts/subagent-stop.js +6 -0
- package/plugins/harness-everything/hooks/session-start.js +81 -0
- package/plugins/harness-everything/plugin.json +43 -0
- package/plugins/harness-everything/scripts/lib/advisory-text.js +174 -0
- package/plugins/harness-everything/skills/environment-detection/SKILL.md +41 -0
- package/plugins/harness-everything/skills/environment-detection/references/shell-syntax-rules.md +76 -0
- package/plugins/harness-everything/skills/environment-detection/scripts/preflight.js +103 -0
- package/plugins/harness-everything/skills/eval-harness/SKILL.md +42 -0
- package/plugins/harness-everything/skills/eval-harness/references/scoring-rubric.md +68 -0
- package/plugins/harness-everything/skills/eval-harness/scripts/evaluate.js +79 -0
- package/plugins/harness-everything/skills/fable-discipline/SKILL.md +41 -0
- package/plugins/harness-everything/skills/fable-discipline/references/discipline-rules.md +22 -0
- package/plugins/harness-everything/skills/fable-mode/CONTRACT-FORMAT.md +165 -0
- package/plugins/harness-everything/skills/fable-mode/SKILL.md +39 -0
- package/plugins/harness-everything/skills/fable-mode/agents/fable-orchestrator.md +159 -0
- package/plugins/harness-everything/skills/fable-mode/agents/fable-verifier.md +32 -0
- package/plugins/harness-everything/skills/fable-mode/agents/fable-worker-haiku.md +52 -0
- package/plugins/harness-everything/skills/fable-mode/agents/fable-worker-sonnet.md +53 -0
- package/plugins/harness-everything/skills/fable-mode/execution-guardrails/SKILL.md +43 -0
- package/plugins/harness-everything/skills/fable-mode/execution-guardrails/references/guardrail-rules.md +56 -0
- package/plugins/harness-everything/skills/fable-mode/fable-haiku/SKILL.md +36 -0
- package/plugins/harness-everything/skills/fable-mode/fable-haiku/references/routing-notes.md +26 -0
- package/plugins/harness-everything/skills/fable-mode/fable-opus/SKILL.md +36 -0
- package/plugins/harness-everything/skills/fable-mode/fable-sonnet/SKILL.md +36 -0
- package/plugins/harness-everything/skills/fable-mode/model-matrix.json +32 -0
- package/plugins/harness-everything/skills/fable-mode/references/execution-phases.md +37 -0
- package/plugins/harness-everything/skills/fable-mode/references/model-matrix.md +39 -0
- package/plugins/harness-everything/skills/fable-mode/scripts/ensemble-review.js +371 -0
- package/plugins/harness-everything/skills/fable-mode/scripts/model-selector.js +260 -0
- package/plugins/harness-everything/skills/fable-mode/scripts/workflow-plan-consumer.js +337 -0
- package/plugins/harness-everything/skills/find-skills/SKILL.md +48 -0
- package/plugins/harness-everything/skills/find-skills/references/discovery-flow.md +102 -0
- package/plugins/harness-everything/skills/find-skills/scripts/use-skill.js +112 -0
- package/plugins/harness-everything/skills/git-commit/SKILL.md +38 -0
- package/plugins/harness-everything/skills/git-commit/guides/ANGULAR_STYLE.md +61 -0
- package/plugins/harness-everything/skills/git-commit/guides/COMMIT_GENERATION.md +24 -0
- package/plugins/harness-everything/skills/git-commit/guides/LANGUAGE_DETECTION.md +16 -0
- package/plugins/harness-everything/skills/git-commit/guides/MAIN_REPO.md +15 -0
- package/plugins/harness-everything/skills/git-commit/guides/SUBMODULES.md +27 -0
- package/plugins/harness-everything/skills/git-commit/references/commit-flow.md +30 -0
- package/plugins/harness-everything/skills/grill-me/SKILL.md +43 -0
- package/plugins/harness-everything/skills/grill-me/references/grilling-playbook.md +38 -0
- package/plugins/harness-everything/skills/grill-with-docs/ADR-FORMAT.md +47 -0
- package/plugins/harness-everything/skills/grill-with-docs/CONTEXT-FORMAT.md +63 -0
- package/plugins/harness-everything/skills/grill-with-docs/SKILL.md +41 -0
- package/plugins/harness-everything/skills/grill-with-docs/references/session-playbook.md +99 -0
- package/plugins/harness-everything/skills/harness-everything/SKILL.md +43 -0
- package/plugins/harness-everything/skills/harness-everything/examples/README.md +9 -0
- package/plugins/harness-everything/skills/harness-everything/examples/todo-workflow.md +22 -0
- package/plugins/harness-everything/skills/harness-everything/references/router-workflow-plan.md +128 -0
- package/plugins/harness-everything/skills/harness-everything/references/skill-registry.md +39 -0
- package/plugins/harness-everything/skills/harness-everything/references/triage-and-tiers.md +193 -0
- package/plugins/harness-everything/skills/harness-everything/schemas/router-task-shape.schema.json +126 -0
- package/plugins/harness-everything/skills/harness-everything/schemas/router-workflow-plan.schema.json +149 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/bootstrap.js +212 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/ensemble-policy.js +102 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/kernel-router.js +212 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/router-contract.js +612 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/routing-keywords.json +320 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/self-heal.js +250 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/tier-router.js +459 -0
- package/plugins/harness-everything/skills/harness-everything/scripts/verify-gate.js +100 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/SKILL.md +38 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/guides/DEEPENING.md +37 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/guides/HTML-REPORT.md +123 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/guides/INTERFACE-DESIGN.md +44 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/guides/LANGUAGE.md +53 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/guides/refactoring-workflow.md +59 -0
- package/plugins/harness-everything/skills/improve-codebase-architecture/references/README.md +16 -0
- package/plugins/harness-everything/skills/install-cognitive-os/SKILL.md +44 -0
- package/plugins/harness-everything/skills/install-cognitive-os/references/cognitive-loop.md +27 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/NOTICE-agency-agents.md +27 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/SKILL.md +38 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/examples/csharp-monorepo.md +123 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/examples/typescript-cli.md +126 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-claude.md +77 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-codex.md +146 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-continue.md +38 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-copilot.md +74 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-cursor.md +40 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/platform-hermes.md +38 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/guidelines/role-capabilities.md +108 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/references/agency-agents.md +31 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/references/architecture-guide.md +36 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/references/orchestration.md +65 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/agency-catalog.js +190 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/consume-workflow-plan.js +129 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/index_memory.js +104 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/path-boundary.js +59 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/project-docs-resolver.js +357 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/scripts/scaffold.js +408 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/AGENTS.md +20 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/README.md +17 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/agent-launcher.md +68 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/backend-developer-csharp.md +56 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/backend-developer-polyglot.md +62 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/backend-developer-ts.md +57 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/memory-keeper.md +37 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/requirement-analyzer.md +66 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/specialist-agent.md +34 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/task-verifier.md +56 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/templates/tool-maker.md +33 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/workflows/01-init.md +22 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/workflows/02-analysis.md +23 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/workflows/03-generation.md +25 -0
- package/plugins/harness-everything/skills/multi-agent-workspace/workflows/04-launcher.md +16 -0
- package/plugins/harness-everything/skills/repo-docs/SKILL.md +42 -0
- package/plugins/harness-everything/skills/repo-docs/references/process-guide.md +60 -0
- package/plugins/harness-everything/skills/repo-docs/templates/agents-template.md +264 -0
- package/plugins/harness-everything/skills/repo-docs/templates/knowledge-agents-template.md +50 -0
- package/plugins/harness-everything/skills/repo-docs/templates/knowledge-readme-template.md +83 -0
- package/plugins/harness-everything/skills/repo-docs/templates/manual-recon.md +117 -0
- package/plugins/harness-everything/skills/repo-docs/templates/multi-skills-readme-template.md +72 -0
- package/plugins/harness-everything/skills/repo-docs/templates/product-agents-template.md +78 -0
- package/plugins/harness-everything/skills/repo-docs/templates/product-readme-template.md +95 -0
- package/plugins/harness-everything/skills/repo-docs/templates/readme-template.md +136 -0
- package/plugins/harness-everything/skills/repo-docs/templates/single-skill-readme-template.md +70 -0
- package/plugins/harness-everything/skills/repo-docs/templates/skills-agents-template.md +60 -0
- package/plugins/harness-everything/skills/rewrite-commits/SKILL.md +37 -0
- package/plugins/harness-everything/skills/rewrite-commits/references/workflow.md +18 -0
- package/plugins/harness-everything/skills/security-review/SKILL.md +47 -0
- package/plugins/harness-everything/skills/security-review/guides/OWASP-PATTERNS.md +31 -0
- package/plugins/harness-everything/skills/security-review/guides/STRIDE-THREAT-MODEL.md +21 -0
- package/plugins/harness-everything/skills/security-review/references/security-checklist.md +478 -0
- package/plugins/harness-everything/skills/security-review/scripts/audit-secrets.js +85 -0
- package/plugins/harness-everything/skills/self-evolve/SKILL.md +45 -0
- package/plugins/harness-everything/skills/self-evolve/references/memory-resolution.md +61 -0
- package/plugins/harness-everything/skills/self-evolve/scripts/persist-memory.js +272 -0
- package/plugins/harness-everything/skills/self-evolve/scripts/register-dynamic-skill.js +253 -0
- package/plugins/harness-everything/skills/self-evolve/scripts/self-regression.js +125 -0
- package/plugins/harness-everything/skills/skill-creator/SKILL.md +47 -0
- package/plugins/harness-everything/skills/skill-creator/references/authoring-workflow.md +14 -0
- package/plugins/harness-everything/skills/skill-creator/references/dynamic-generation-contract.md +21 -0
- package/plugins/harness-everything/skills/skill-creator/references/quality-checklist.md +16 -0
- package/plugins/harness-everything/skills/skill-creator/references/quality-principles.md +81 -0
- package/plugins/harness-everything/skills/skill-creator/references/testing-workflow.md +27 -0
- package/plugins/harness-everything/skills/skill-style/SKILL.md +42 -0
- package/plugins/harness-everything/skills/skill-style/references/style-guide.md +62 -0
- package/plugins/harness-everything/skills/tdd/SKILL.md +38 -0
- package/plugins/harness-everything/skills/tdd/fixtures/pass-evidence.json +100 -0
- package/plugins/harness-everything/skills/tdd/fixtures/quality-scenarios.json +20 -0
- package/plugins/harness-everything/skills/tdd/guides/deep-modules.md +33 -0
- package/plugins/harness-everything/skills/tdd/guides/interface-design.md +31 -0
- package/plugins/harness-everything/skills/tdd/guides/mocking.md +59 -0
- package/plugins/harness-everything/skills/tdd/guides/refactoring.md +10 -0
- package/plugins/harness-everything/skills/tdd/guides/tests.md +61 -0
- package/plugins/harness-everything/skills/tdd/references/core-discipline.md +51 -0
- package/plugins/harness-everything/skills/tdd/references/integration-testing.md +13 -0
- package/plugins/harness-everything/skills/tdd/references/quality-model.md +68 -0
- package/plugins/harness-everything/skills/tdd/references/unit-testing.md +15 -0
- package/plugins/harness-everything/skills/tdd/schemas/quality-evidence.schema.json +211 -0
- package/plugins/harness-everything/skills/tdd/schemas/quality-report.schema.json +110 -0
- package/plugins/harness-everything/skills/tdd/scripts/quality-gate.js +475 -0
- package/plugins/harness-everything/skills/to-spec/SKILL.md +40 -0
- package/plugins/harness-everything/skills/to-spec/references/process.md +77 -0
- package/plugins/harness-everything/skills/to-spec/scripts/check-project-docs.js +239 -0
- package/plugins/harness-everything/skills/to-spec/scripts/project-docs-resolver.js +111 -0
- package/plugins/harness-everything/skills/to-spec/templates/cli-reference.md +31 -0
- package/plugins/harness-everything/skills/to-spec/templates/dev-doc.md +27 -0
- package/plugins/harness-everything/skills/to-spec/templates/feature-spec.md +59 -0
- package/plugins/harness-everything/skills/to-spec/templates/schema-doc.md +31 -0
- package/plugins/harness-everything/skills/to-tickets/SKILL.md +42 -0
- package/plugins/harness-everything/skills/to-tickets/references/publishing-and-templates.md +97 -0
- package/plugins/harness-everything/skills/todo-driven-workflow/SKILL.md +42 -0
- package/plugins/harness-everything/skills/todo-driven-workflow/references/execution-guide.md +27 -0
- package/plugins/harness-everything/skills/using-git-worktrees/SKILL.md +49 -0
- package/plugins/harness-everything/skills/using-git-worktrees/references/workflow-details.md +108 -0
- package/plugins/harness-everything/skills/verification-loop/SKILL.md +46 -0
- package/plugins/harness-everything/skills/verification-loop/references/verification-phases.md +116 -0
- package/plugins/harness-everything/skills/verification-loop/templates/verification-report.template.md +15 -0
- package/plugins/harness-everything/skills/verify-before-claim/SKILL.md +43 -0
- package/plugins/harness-everything/skills/verify-before-claim/references/verification-guide.md +73 -0
- package/plugins/harness-everything/skills/zoom-out/SKILL.md +42 -0
- package/plugins/harness-everything/skills/zoom-out/references/circuit-breaker.md +66 -0
- package/plugins/harness-everything/skills/zoom-out/templates/zoom-out-report.template.md +20 -0
- package/repo-docs/SKILL.md +3 -7
- package/rewrite-commits/SKILL.md +2 -2
- package/scripts/build-openai-submission.js +158 -0
- package/scripts/installer.js +86 -11
- package/scripts/lib/advisory-text.js +25 -18
- package/scripts/lib/execution-contract.js +70 -0
- package/scripts/lib/manifest.js +34 -38
- package/scripts/lib/path-boundary.js +90 -0
- package/scripts/lib/skills.js +124 -70
- package/scripts/lib/workspace.js +269 -7
- package/scripts/repository-contract.js +213 -0
- package/scripts/sync-current-docs.js +12 -0
- package/scripts/sync-openai-plugin.js +91 -0
- package/scripts/sync-release-version.js +212 -0
- package/security-review/SKILL.md +4 -7
- package/security-review/scripts/audit-secrets.js +9 -8
- package/self-evolve/SKILL.md +9 -11
- package/self-evolve/references/memory-resolution.md +4 -1
- package/self-evolve/scripts/persist-memory.js +221 -88
- package/self-evolve/scripts/register-dynamic-skill.js +12 -10
- package/self-evolve/scripts/self-regression.js +23 -8
- package/skill-creator/SKILL.md +12 -14
- package/skill-creator/references/quality-checklist.md +1 -0
- package/skill-style/SKILL.md +4 -4
- package/skill-style/references/style-guide.md +29 -3
- package/submission/openai/README.md +67 -0
- package/submission/openai/listing.json +31 -0
- package/submission/openai/test-cases.json +61 -0
- package/tdd/SKILL.md +20 -26
- package/tdd/fixtures/pass-evidence.json +100 -0
- package/tdd/fixtures/quality-scenarios.json +20 -0
- package/tdd/references/core-discipline.md +2 -0
- package/tdd/references/integration-testing.md +13 -0
- package/tdd/references/quality-model.md +68 -0
- package/tdd/references/unit-testing.md +15 -0
- package/tdd/schemas/quality-evidence.schema.json +211 -0
- package/tdd/schemas/quality-report.schema.json +110 -0
- package/tdd/scripts/quality-gate.js +475 -0
- package/to-spec/SKILL.md +3 -5
- package/to-spec/scripts/check-project-docs.js +42 -62
- package/to-spec/scripts/project-docs-resolver.js +111 -0
- package/to-tickets/SKILL.md +4 -4
- package/todo-driven-workflow/SKILL.md +6 -4
- package/using-git-worktrees/SKILL.md +3 -3
- package/verification-loop/SKILL.md +4 -4
- package/verify-before-claim/SKILL.md +2 -2
- package/zoom-out/SKILL.md +4 -4
- package/ab-test-deep-analysis-2026-08-27.md +0 -383
- package/ab-test-results-2026-08-27.md +0 -152
- package/ab-test-statistical-analysis-31cases.md +0 -193
- package/opencode-plugin/hooks/circuit-breaker.js +0 -121
- package/opencode-plugin/hooks/compliance.js +0 -128
- package/opencode-plugin/hooks/post-edit.js +0 -62
- package/opencode-plugin/hooks/pre-complete.js +0 -65
- package/opencode-plugin/test-hook-firing.js +0 -198
|
@@ -0,0 +1,371 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
'use strict';
|
|
3
|
+
|
|
4
|
+
const crypto = require('crypto');
|
|
5
|
+
const fs = require('fs');
|
|
6
|
+
const path = require('path');
|
|
7
|
+
const {
|
|
8
|
+
derivePlanId,
|
|
9
|
+
getWorkspaceStateRoot,
|
|
10
|
+
} = require('./workflow-plan-consumer');
|
|
11
|
+
|
|
12
|
+
const ENSEMBLE_EVIDENCE_VERSION = 1;
|
|
13
|
+
const SUPPORTED_DIVERSITY = new Set(['model', 'prompt', 'evidence']);
|
|
14
|
+
const VERIFICATION_STATUSES = new Set(['pass', 'fail', 'inconclusive']);
|
|
15
|
+
const VERIFIER_VERDICTS = new Set(['accept', 'reject', 'inconclusive']);
|
|
16
|
+
const CORRECTNESS_STATES = new Set(['pass', 'fail', 'inconclusive']);
|
|
17
|
+
|
|
18
|
+
function sanitizeId(value, label) {
|
|
19
|
+
const text = String(value || '').trim();
|
|
20
|
+
if (!text) throw new Error(`${label} is required`);
|
|
21
|
+
if (!/^[A-Za-z0-9._-]+$/.test(text)) throw new Error(`${label} contains unsupported characters: ${text}`);
|
|
22
|
+
return text.slice(0, 128);
|
|
23
|
+
}
|
|
24
|
+
|
|
25
|
+
function boundedString(value, label, max = 512) {
|
|
26
|
+
const text = String(value || '').trim();
|
|
27
|
+
if (!text) throw new Error(`${label} is required`);
|
|
28
|
+
if (text.length > max) throw new Error(`${label} exceeds ${max} characters`);
|
|
29
|
+
return text;
|
|
30
|
+
}
|
|
31
|
+
|
|
32
|
+
function normalizeEvidenceRefs(values, label) {
|
|
33
|
+
if (!Array.isArray(values)) throw new Error(`${label} must be an array`);
|
|
34
|
+
return [...new Set(values.map((value, index) => boundedString(value, `${label}[${index}]`, 512)))].sort();
|
|
35
|
+
}
|
|
36
|
+
|
|
37
|
+
function normalizeMetricNumber(value, label, integer = true) {
|
|
38
|
+
if (value === undefined || value === null) return null;
|
|
39
|
+
const number = Number(value);
|
|
40
|
+
if (!Number.isFinite(number) || number < 0 || (integer && !Number.isInteger(number))) {
|
|
41
|
+
throw new Error(`${label} must be a non-negative ${integer ? 'integer' : 'number'} or null`);
|
|
42
|
+
}
|
|
43
|
+
return number;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
function normalizeMetrics(raw = {}) {
|
|
47
|
+
const correctness = raw.correctness === undefined || raw.correctness === null
|
|
48
|
+
? null
|
|
49
|
+
: String(raw.correctness);
|
|
50
|
+
if (correctness !== null && !CORRECTNESS_STATES.has(correctness)) {
|
|
51
|
+
throw new Error(`metrics.correctness is invalid: ${correctness}`);
|
|
52
|
+
}
|
|
53
|
+
const failureMode = raw.failureMode === undefined || raw.failureMode === null
|
|
54
|
+
? null
|
|
55
|
+
: boundedString(raw.failureMode, 'metrics.failureMode', 256);
|
|
56
|
+
return {
|
|
57
|
+
correctness,
|
|
58
|
+
toolCalls: normalizeMetricNumber(raw.toolCalls, 'metrics.toolCalls'),
|
|
59
|
+
tokens: normalizeMetricNumber(raw.tokens, 'metrics.tokens'),
|
|
60
|
+
elapsedMs: normalizeMetricNumber(raw.elapsedMs, 'metrics.elapsedMs', false),
|
|
61
|
+
retries: normalizeMetricNumber(raw.retries, 'metrics.retries'),
|
|
62
|
+
failureMode,
|
|
63
|
+
};
|
|
64
|
+
}
|
|
65
|
+
|
|
66
|
+
function normalizeCandidate(raw) {
|
|
67
|
+
if (!raw || typeof raw !== 'object' || Array.isArray(raw)) throw new Error('candidate must be an object');
|
|
68
|
+
const candidateId = sanitizeId(raw.candidateId, 'candidateId');
|
|
69
|
+
const position = boundedString(raw.position, `${candidateId}.position`, 512);
|
|
70
|
+
const diversity = raw.diversity;
|
|
71
|
+
if (!diversity || typeof diversity !== 'object' || Array.isArray(diversity)) {
|
|
72
|
+
throw new Error(`${candidateId}.diversity must be an object`);
|
|
73
|
+
}
|
|
74
|
+
const modelId = boundedString(diversity.modelId, `${candidateId}.diversity.modelId`, 128);
|
|
75
|
+
const promptVariantId = boundedString(diversity.promptVariantId, `${candidateId}.diversity.promptVariantId`, 128);
|
|
76
|
+
const evidenceSetId = boundedString(diversity.evidenceSetId, `${candidateId}.diversity.evidenceSetId`, 128);
|
|
77
|
+
|
|
78
|
+
const verification = raw.objectiveVerification;
|
|
79
|
+
if (!verification || typeof verification !== 'object' || Array.isArray(verification)) {
|
|
80
|
+
throw new Error(`${candidateId}.objectiveVerification must be an object`);
|
|
81
|
+
}
|
|
82
|
+
const status = String(verification.status || '');
|
|
83
|
+
if (!VERIFICATION_STATUSES.has(status)) {
|
|
84
|
+
throw new Error(`${candidateId}.objectiveVerification.status is invalid: ${status}`);
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
return {
|
|
88
|
+
candidateId,
|
|
89
|
+
position,
|
|
90
|
+
diversity: { modelId, promptVariantId, evidenceSetId },
|
|
91
|
+
objectiveVerification: {
|
|
92
|
+
status,
|
|
93
|
+
evidenceRefs: normalizeEvidenceRefs(verification.evidenceRefs || [], `${candidateId}.objectiveVerification.evidenceRefs`),
|
|
94
|
+
},
|
|
95
|
+
metrics: normalizeMetrics(raw.metrics || {}),
|
|
96
|
+
};
|
|
97
|
+
}
|
|
98
|
+
|
|
99
|
+
function diversityFingerprint(candidate) {
|
|
100
|
+
return [
|
|
101
|
+
candidate.diversity.modelId,
|
|
102
|
+
candidate.diversity.promptVariantId,
|
|
103
|
+
candidate.diversity.evidenceSetId,
|
|
104
|
+
].join('|');
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
function fingerprintHash(candidate) {
|
|
108
|
+
return crypto.createHash('sha256').update(diversityFingerprint(candidate)).digest('hex').slice(0, 16);
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
function normalizeVerifier(raw, candidateIds) {
|
|
112
|
+
if (!raw || typeof raw !== 'object' || Array.isArray(raw)) throw new Error('independent verifier evidence is required');
|
|
113
|
+
const verifierId = sanitizeId(raw.verifierId, 'verifierId');
|
|
114
|
+
if (candidateIds.has(verifierId)) throw new Error('verifierId must not reuse a candidateId');
|
|
115
|
+
if (raw.independent !== true) throw new Error('verifier must declare independent=true');
|
|
116
|
+
const verdict = String(raw.verdict || '');
|
|
117
|
+
if (!VERIFIER_VERDICTS.has(verdict)) throw new Error(`verifier verdict is invalid: ${verdict}`);
|
|
118
|
+
const position = raw.position === null || raw.position === undefined
|
|
119
|
+
? null
|
|
120
|
+
: boundedString(raw.position, 'verifier.position', 512);
|
|
121
|
+
if (verdict === 'accept' && !position) throw new Error('accept verdict requires verifier.position');
|
|
122
|
+
const evidenceRefs = normalizeEvidenceRefs(raw.evidenceRefs || [], 'verifier.evidenceRefs');
|
|
123
|
+
if (evidenceRefs.length === 0) throw new Error('independent verifier must provide at least one evidence reference');
|
|
124
|
+
return { verifierId, independent: true, verdict, position, evidenceRefs };
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
function validateEnsembleConfig(plan) {
|
|
128
|
+
if (!plan || typeof plan !== 'object' || Array.isArray(plan)) throw new Error('workflow plan must be an object');
|
|
129
|
+
const config = plan.ensemble;
|
|
130
|
+
if (!config || typeof config !== 'object' || Array.isArray(config)) throw new Error('workflow plan does not enable ensemble review');
|
|
131
|
+
if (!Number.isInteger(config.maxCandidates) || config.maxCandidates < 2 || config.maxCandidates > 4) {
|
|
132
|
+
throw new Error('ensemble.maxCandidates must be an integer from 2 through 4');
|
|
133
|
+
}
|
|
134
|
+
if (!Array.isArray(config.diversity) || config.diversity.length === 0) {
|
|
135
|
+
throw new Error('ensemble.diversity must be a non-empty array');
|
|
136
|
+
}
|
|
137
|
+
for (const dimension of config.diversity) {
|
|
138
|
+
if (!SUPPORTED_DIVERSITY.has(dimension)) throw new Error(`unsupported ensemble diversity dimension: ${dimension}`);
|
|
139
|
+
}
|
|
140
|
+
if (config.synthesis !== 'preserve-disagreement') throw new Error('ensemble.synthesis must be preserve-disagreement');
|
|
141
|
+
if (config.verifier !== 'independent') throw new Error('ensemble.verifier must be independent');
|
|
142
|
+
return {
|
|
143
|
+
maxCandidates: config.maxCandidates,
|
|
144
|
+
diversity: [...new Set(config.diversity)],
|
|
145
|
+
synthesis: config.synthesis,
|
|
146
|
+
verifier: config.verifier,
|
|
147
|
+
reasonCodes: Array.isArray(config.reasonCodes) ? [...new Set(config.reasonCodes.filter(Boolean))] : [],
|
|
148
|
+
};
|
|
149
|
+
}
|
|
150
|
+
|
|
151
|
+
function pairedEvidenceAllowsImprovementClaim(raw) {
|
|
152
|
+
if (!raw || typeof raw !== 'object' || Array.isArray(raw)) return false;
|
|
153
|
+
return raw.paired === true &&
|
|
154
|
+
raw.sameEngineModelFixtureRubricConfig === true &&
|
|
155
|
+
Number.isInteger(raw.sampleCount) && raw.sampleCount > 0 &&
|
|
156
|
+
raw.effectEstimate !== undefined && raw.effectEstimate !== null &&
|
|
157
|
+
raw.uncertainty !== undefined && raw.uncertainty !== null;
|
|
158
|
+
}
|
|
159
|
+
|
|
160
|
+
function synthesizeEnsemble({ plan, candidates, verifier, pairedEvidence = null }) {
|
|
161
|
+
const config = validateEnsembleConfig(plan);
|
|
162
|
+
if (!Array.isArray(candidates)) throw new Error('candidates must be an array');
|
|
163
|
+
if (candidates.length < 2) throw new Error('ensemble review requires at least two candidates');
|
|
164
|
+
if (candidates.length > config.maxCandidates) {
|
|
165
|
+
throw new Error(`candidate count ${candidates.length} exceeds ensemble.maxCandidates=${config.maxCandidates}`);
|
|
166
|
+
}
|
|
167
|
+
|
|
168
|
+
const normalized = candidates.map(normalizeCandidate);
|
|
169
|
+
const candidateIds = new Set();
|
|
170
|
+
const fingerprints = new Set();
|
|
171
|
+
for (const candidate of normalized) {
|
|
172
|
+
if (candidateIds.has(candidate.candidateId)) throw new Error(`duplicate candidateId: ${candidate.candidateId}`);
|
|
173
|
+
candidateIds.add(candidate.candidateId);
|
|
174
|
+
const fingerprint = diversityFingerprint(candidate);
|
|
175
|
+
if (fingerprints.has(fingerprint)) {
|
|
176
|
+
throw new Error(`candidates are not independent: duplicate model/prompt/evidence fingerprint for ${candidate.candidateId}`);
|
|
177
|
+
}
|
|
178
|
+
fingerprints.add(fingerprint);
|
|
179
|
+
}
|
|
180
|
+
|
|
181
|
+
const normalizedVerifier = normalizeVerifier(verifier, candidateIds);
|
|
182
|
+
const byPosition = new Map();
|
|
183
|
+
for (const candidate of normalized) {
|
|
184
|
+
if (!byPosition.has(candidate.position)) byPosition.set(candidate.position, []);
|
|
185
|
+
byPosition.get(candidate.position).push(candidate);
|
|
186
|
+
}
|
|
187
|
+
|
|
188
|
+
const positions = [...byPosition.entries()]
|
|
189
|
+
.map(([position, members]) => ({
|
|
190
|
+
position,
|
|
191
|
+
count: members.length,
|
|
192
|
+
candidateIds: members.map(member => member.candidateId).sort(),
|
|
193
|
+
verificationStatuses: members.map(member => member.objectiveVerification.status).sort(),
|
|
194
|
+
evidenceRefs: [...new Set(members.flatMap(member => member.objectiveVerification.evidenceRefs))].sort(),
|
|
195
|
+
}))
|
|
196
|
+
.sort((left, right) => left.position.localeCompare(right.position));
|
|
197
|
+
|
|
198
|
+
const maxCount = Math.max(...positions.map(entry => entry.count));
|
|
199
|
+
const leaders = positions.filter(entry => entry.count === maxCount);
|
|
200
|
+
const strictMajority = leaders.length === 1 && maxCount > normalized.length / 2
|
|
201
|
+
? leaders[0].position
|
|
202
|
+
: null;
|
|
203
|
+
const unanimous = positions.length === 1 ? positions[0].position : null;
|
|
204
|
+
const proposedPosition = unanimous || strictMajority;
|
|
205
|
+
const proposedMembers = proposedPosition ? byPosition.get(proposedPosition) : [];
|
|
206
|
+
const objectiveFailureBlocks = proposedMembers.some(candidate => candidate.objectiveVerification.status === 'fail');
|
|
207
|
+
|
|
208
|
+
let disposition = 'disagreement-preserved';
|
|
209
|
+
let selectedPosition = null;
|
|
210
|
+
const reasonCodes = [...config.reasonCodes];
|
|
211
|
+
if (objectiveFailureBlocks) {
|
|
212
|
+
disposition = 'blocked-objective-failure';
|
|
213
|
+
reasonCodes.push('objective-verification-failure');
|
|
214
|
+
} else if (!proposedPosition) {
|
|
215
|
+
reasonCodes.push('no-strict-majority');
|
|
216
|
+
} else if (normalizedVerifier.verdict === 'reject') {
|
|
217
|
+
disposition = 'verifier-rejected';
|
|
218
|
+
reasonCodes.push('independent-verifier-rejected');
|
|
219
|
+
} else if (normalizedVerifier.verdict === 'inconclusive') {
|
|
220
|
+
disposition = 'verifier-inconclusive';
|
|
221
|
+
reasonCodes.push('independent-verifier-inconclusive');
|
|
222
|
+
} else if (normalizedVerifier.position !== proposedPosition) {
|
|
223
|
+
disposition = 'verifier-position-mismatch';
|
|
224
|
+
reasonCodes.push('independent-verifier-position-mismatch');
|
|
225
|
+
} else {
|
|
226
|
+
disposition = 'accepted';
|
|
227
|
+
selectedPosition = proposedPosition;
|
|
228
|
+
reasonCodes.push('independent-verifier-accepted');
|
|
229
|
+
}
|
|
230
|
+
|
|
231
|
+
const improvementAllowed = pairedEvidenceAllowsImprovementClaim(pairedEvidence);
|
|
232
|
+
const minorityPositions = positions
|
|
233
|
+
.filter(entry => entry.position !== selectedPosition)
|
|
234
|
+
.map(entry => entry.position);
|
|
235
|
+
|
|
236
|
+
return {
|
|
237
|
+
schemaVersion: ENSEMBLE_EVIDENCE_VERSION,
|
|
238
|
+
ensemble: config,
|
|
239
|
+
candidateCount: normalized.length,
|
|
240
|
+
candidateSummaries: normalized.map(candidate => ({
|
|
241
|
+
candidateId: candidate.candidateId,
|
|
242
|
+
position: candidate.position,
|
|
243
|
+
diversityFingerprint: fingerprintHash(candidate),
|
|
244
|
+
objectiveVerification: candidate.objectiveVerification,
|
|
245
|
+
})),
|
|
246
|
+
positions,
|
|
247
|
+
disagreement: positions.length > 1,
|
|
248
|
+
minorityPositions,
|
|
249
|
+
proposedPosition,
|
|
250
|
+
selectedPosition,
|
|
251
|
+
disposition,
|
|
252
|
+
verifier: normalizedVerifier,
|
|
253
|
+
reasonCodes: [...new Set(reasonCodes)],
|
|
254
|
+
measurement: {
|
|
255
|
+
sourceIssues: ['#71', '#83'],
|
|
256
|
+
correctness: normalized.map(candidate => ({
|
|
257
|
+
candidateId: candidate.candidateId,
|
|
258
|
+
value: candidate.metrics.correctness,
|
|
259
|
+
})),
|
|
260
|
+
executionCost: normalized.map(candidate => ({
|
|
261
|
+
candidateId: candidate.candidateId,
|
|
262
|
+
toolCalls: candidate.metrics.toolCalls,
|
|
263
|
+
tokens: candidate.metrics.tokens,
|
|
264
|
+
elapsedMs: candidate.metrics.elapsedMs,
|
|
265
|
+
retries: candidate.metrics.retries,
|
|
266
|
+
})),
|
|
267
|
+
failureModes: normalized.map(candidate => ({
|
|
268
|
+
candidateId: candidate.candidateId,
|
|
269
|
+
failureMode: candidate.metrics.failureMode,
|
|
270
|
+
})),
|
|
271
|
+
telemetryOptional: true,
|
|
272
|
+
},
|
|
273
|
+
improvementClaim: {
|
|
274
|
+
allowed: improvementAllowed,
|
|
275
|
+
reasonCode: improvementAllowed ? 'paired-evidence-present' : 'paired-evidence-required',
|
|
276
|
+
},
|
|
277
|
+
};
|
|
278
|
+
}
|
|
279
|
+
|
|
280
|
+
function atomicWriteJson(filePath, value) {
|
|
281
|
+
fs.mkdirSync(path.dirname(filePath), { recursive: true });
|
|
282
|
+
const tempPath = `${filePath}.${process.pid}.${Date.now()}.tmp`;
|
|
283
|
+
fs.writeFileSync(tempPath, `${JSON.stringify(value, null, 2)}\n`, 'utf8');
|
|
284
|
+
fs.renameSync(tempPath, filePath);
|
|
285
|
+
}
|
|
286
|
+
|
|
287
|
+
function writeCorrelatedEvidence({ routerContract, workspaceRoot, runId, synthesis }) {
|
|
288
|
+
const root = path.resolve(workspaceRoot || process.cwd());
|
|
289
|
+
const planId = derivePlanId(routerContract);
|
|
290
|
+
const cleanRunId = sanitizeId(runId, 'runId');
|
|
291
|
+
const runRoot = path.join(getWorkspaceStateRoot(root), 'fable-runs', cleanRunId);
|
|
292
|
+
const manifestPath = path.join(runRoot, 'run.json');
|
|
293
|
+
if (!fs.existsSync(manifestPath)) throw new Error(`run manifest not found: ${manifestPath}`);
|
|
294
|
+
const manifest = JSON.parse(fs.readFileSync(manifestPath, 'utf8'));
|
|
295
|
+
if (manifest.planId !== planId) throw new Error('run manifest planId does not match router contract');
|
|
296
|
+
if (manifest.runId !== cleanRunId) throw new Error('run manifest runId mismatch');
|
|
297
|
+
const artifact = {
|
|
298
|
+
...synthesis,
|
|
299
|
+
planId,
|
|
300
|
+
runId: cleanRunId,
|
|
301
|
+
generatedAt: new Date().toISOString(),
|
|
302
|
+
};
|
|
303
|
+
const evidencePath = path.join(runRoot, 'ensemble.json');
|
|
304
|
+
atomicWriteJson(evidencePath, artifact);
|
|
305
|
+
return { artifact, evidencePath };
|
|
306
|
+
}
|
|
307
|
+
|
|
308
|
+
function parseArgs(argv) {
|
|
309
|
+
const args = {};
|
|
310
|
+
for (let index = 0; index < argv.length; index++) {
|
|
311
|
+
const token = argv[index];
|
|
312
|
+
if (token === '--plan-file') args.planFile = argv[++index];
|
|
313
|
+
else if (token === '--candidates-file') args.candidatesFile = argv[++index];
|
|
314
|
+
else if (token === '--verifier-file') args.verifierFile = argv[++index];
|
|
315
|
+
else if (token === '--paired-evidence-file') args.pairedEvidenceFile = argv[++index];
|
|
316
|
+
else if (token === '--root') args.root = argv[++index];
|
|
317
|
+
else if (token === '--run-id') args.runId = argv[++index];
|
|
318
|
+
else if (token === '--help' || token === '-h') args.help = true;
|
|
319
|
+
else throw new Error(`unknown argument: ${token}`);
|
|
320
|
+
}
|
|
321
|
+
return args;
|
|
322
|
+
}
|
|
323
|
+
|
|
324
|
+
function usage() {
|
|
325
|
+
return 'Usage: node fable-mode/scripts/ensemble-review.js --plan-file <router-contract.json> --candidates-file <candidates.json> --verifier-file <verifier.json> --root <workspace> --run-id <runId> [--paired-evidence-file <paired.json>]';
|
|
326
|
+
}
|
|
327
|
+
|
|
328
|
+
if (require.main === module) {
|
|
329
|
+
try {
|
|
330
|
+
const args = parseArgs(process.argv.slice(2));
|
|
331
|
+
if (args.help) {
|
|
332
|
+
console.log(usage());
|
|
333
|
+
process.exit(0);
|
|
334
|
+
}
|
|
335
|
+
if (!args.planFile || !args.candidatesFile || !args.verifierFile || !args.root || !args.runId) {
|
|
336
|
+
throw new Error(usage());
|
|
337
|
+
}
|
|
338
|
+
const routerContract = JSON.parse(fs.readFileSync(path.resolve(args.planFile), 'utf8'));
|
|
339
|
+
const candidatesPayload = JSON.parse(fs.readFileSync(path.resolve(args.candidatesFile), 'utf8'));
|
|
340
|
+
const verifier = JSON.parse(fs.readFileSync(path.resolve(args.verifierFile), 'utf8'));
|
|
341
|
+
const pairedEvidence = args.pairedEvidenceFile
|
|
342
|
+
? JSON.parse(fs.readFileSync(path.resolve(args.pairedEvidenceFile), 'utf8'))
|
|
343
|
+
: null;
|
|
344
|
+
const synthesis = synthesizeEnsemble({
|
|
345
|
+
plan: routerContract.workflowPlan,
|
|
346
|
+
candidates: Array.isArray(candidatesPayload) ? candidatesPayload : candidatesPayload.candidates,
|
|
347
|
+
verifier,
|
|
348
|
+
pairedEvidence,
|
|
349
|
+
});
|
|
350
|
+
const result = writeCorrelatedEvidence({
|
|
351
|
+
routerContract,
|
|
352
|
+
workspaceRoot: args.root,
|
|
353
|
+
runId: args.runId,
|
|
354
|
+
synthesis,
|
|
355
|
+
});
|
|
356
|
+
process.stdout.write(`${JSON.stringify({ ...result.artifact, evidencePath: result.evidencePath })}\n`);
|
|
357
|
+
} catch (err) {
|
|
358
|
+
console.error(`[FABLE ENSEMBLE REVIEW] ${err.message}`);
|
|
359
|
+
process.exit(2);
|
|
360
|
+
}
|
|
361
|
+
}
|
|
362
|
+
|
|
363
|
+
module.exports = {
|
|
364
|
+
ENSEMBLE_EVIDENCE_VERSION,
|
|
365
|
+
diversityFingerprint,
|
|
366
|
+
normalizeCandidate,
|
|
367
|
+
pairedEvidenceAllowsImprovementClaim,
|
|
368
|
+
synthesizeEnsemble,
|
|
369
|
+
validateEnsembleConfig,
|
|
370
|
+
writeCorrelatedEvidence,
|
|
371
|
+
};
|
|
@@ -1,9 +1,139 @@
|
|
|
1
1
|
#!/usr/bin/env node
|
|
2
2
|
|
|
3
3
|
const fs = require('fs');
|
|
4
|
+
const os = require('os');
|
|
4
5
|
const path = require('path');
|
|
6
|
+
const crypto = require('crypto');
|
|
5
7
|
|
|
6
8
|
const MATRIX_PATH = path.join(__dirname, '..', 'model-matrix.json');
|
|
9
|
+
const SESSION_REGISTRY_DIR = 'session-workspaces';
|
|
10
|
+
const SESSION_REGISTRY_VERSION = 1;
|
|
11
|
+
const UNBOUND_WORKSPACE_KEY = 'unbound-workspace';
|
|
12
|
+
const HOST_CONTEXT_KEYS = ['host_id', 'hostId', 'agent_id', 'agentId', 'client_id', 'clientId', 'machine_id', 'machineId', 'runtime_id', 'runtimeId'];
|
|
13
|
+
|
|
14
|
+
// This script ships standalone (copied whole into every install target,
|
|
15
|
+
// e.g. .claude/skills/fable-mode/scripts/), so it can't require the source
|
|
16
|
+
// repo's scripts/lib/workspace.js. The audit log is genuine runtime state
|
|
17
|
+
// though - not a project artifact - so it needs to land in the same global,
|
|
18
|
+
// workspace-keyed root that hooks/scripts/lib/harness-state.js resolves to,
|
|
19
|
+
// not scattered under cwd (issue #42). Duplicated here, algorithm-for-
|
|
20
|
+
// algorithm, same as opencode-plugin/index.mjs's own inlined copy.
|
|
21
|
+
function findWorkspaceRoot(startPath, allowNonGit = false) {
|
|
22
|
+
if (typeof startPath !== 'string' || !startPath.trim()) return null;
|
|
23
|
+
let dir = path.resolve(startPath);
|
|
24
|
+
while (true) {
|
|
25
|
+
if (fs.existsSync(path.join(dir, '.git'))) return canonicalPath(dir);
|
|
26
|
+
const parent = path.dirname(dir);
|
|
27
|
+
if (parent === dir) break;
|
|
28
|
+
dir = parent;
|
|
29
|
+
}
|
|
30
|
+
return allowNonGit ? canonicalPath(startPath) : null;
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
function canonicalPath(value) {
|
|
34
|
+
if (typeof value !== 'string' || !value.trim()) return null;
|
|
35
|
+
const resolved = path.resolve(value);
|
|
36
|
+
try { return fs.realpathSync(resolved); } catch (err) { return resolved; }
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
function getSessionId(context) {
|
|
40
|
+
if (!context || typeof context !== 'object') return null;
|
|
41
|
+
return context.session_id || context.sessionId || null;
|
|
42
|
+
}
|
|
43
|
+
|
|
44
|
+
function getHostId(context) {
|
|
45
|
+
if (!context || typeof context !== 'object') return process.env.FABLE_HOST_ID || process.env.HARNESS_HOST_ID || null;
|
|
46
|
+
for (const key of HOST_CONTEXT_KEYS) {
|
|
47
|
+
if (typeof context[key] === 'string' && context[key].trim()) return context[key].trim();
|
|
48
|
+
}
|
|
49
|
+
return process.env.FABLE_HOST_ID || process.env.HARNESS_HOST_ID || null;
|
|
50
|
+
}
|
|
51
|
+
|
|
52
|
+
function getSessionRegistryPath(sessionId, hostId) {
|
|
53
|
+
const namespace = hostId ? `${hostId}\0${sessionId}` : String(sessionId);
|
|
54
|
+
const hash = crypto.createHash('sha256').update(namespace).digest('hex');
|
|
55
|
+
const home = process.env.HARNESS_STATE_HOME || path.join(os.homedir(), '.agents', 'harness-everything');
|
|
56
|
+
return path.join(home, SESSION_REGISTRY_DIR, `${hash}.json`);
|
|
57
|
+
}
|
|
58
|
+
|
|
59
|
+
function readRegistryRecord(filePath) {
|
|
60
|
+
try {
|
|
61
|
+
const record = JSON.parse(fs.readFileSync(filePath, 'utf8'));
|
|
62
|
+
return record.version === SESSION_REGISTRY_VERSION && typeof record.workspaceRoot === 'string' ? record : null;
|
|
63
|
+
} catch (err) {
|
|
64
|
+
return null;
|
|
65
|
+
}
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
function getBoundWorkspace(sessionId, context) {
|
|
69
|
+
if (!sessionId) return null;
|
|
70
|
+
const hostId = getHostId(context);
|
|
71
|
+
const records = [];
|
|
72
|
+
const preferred = readRegistryRecord(getSessionRegistryPath(sessionId, hostId));
|
|
73
|
+
if (preferred && (!hostId || !preferred.hostId || preferred.hostId === hostId)) records.push(preferred);
|
|
74
|
+
if (hostId) {
|
|
75
|
+
const unscoped = readRegistryRecord(getSessionRegistryPath(sessionId, null));
|
|
76
|
+
if (unscoped && !unscoped.hostId) records.push(unscoped);
|
|
77
|
+
}
|
|
78
|
+
try {
|
|
79
|
+
const home = process.env.HARNESS_STATE_HOME || path.join(os.homedir(), '.agents', 'harness-everything');
|
|
80
|
+
const dir = path.join(home, SESSION_REGISTRY_DIR);
|
|
81
|
+
for (const entry of fs.readdirSync(dir, { withFileTypes: true })) {
|
|
82
|
+
if (!entry.isFile() || !entry.name.endsWith('.json')) continue;
|
|
83
|
+
const record = readRegistryRecord(path.join(dir, entry.name));
|
|
84
|
+
if (record && record.sessionId === sessionId && (!hostId || !record.hostId || record.hostId === hostId)) records.push(record);
|
|
85
|
+
}
|
|
86
|
+
} catch (err) { /* no registry yet */ }
|
|
87
|
+
const unique = [...new Map(records.map(record => [`${record.hostId || ''}\0${record.workspaceRoot}`, record])).values()];
|
|
88
|
+
if (unique.length === 1) return unique[0].workspaceRoot;
|
|
89
|
+
if (unique.length > 1 && new Set(unique.map(record => canonicalPath(record.workspaceRoot))).size === 1) return unique[0].workspaceRoot;
|
|
90
|
+
return null;
|
|
91
|
+
}
|
|
92
|
+
|
|
93
|
+
function getRegistryRecord(sessionId, context) {
|
|
94
|
+
const root = getBoundWorkspace(sessionId, context);
|
|
95
|
+
return root ? { workspaceRoot: root } : null;
|
|
96
|
+
}
|
|
97
|
+
|
|
98
|
+
function bindWorkspace(sessionId, root, context) {
|
|
99
|
+
if (!sessionId) return;
|
|
100
|
+
if (!root) return;
|
|
101
|
+
const hostId = getHostId(context);
|
|
102
|
+
const target = getSessionRegistryPath(sessionId, hostId);
|
|
103
|
+
const record = { version: SESSION_REGISTRY_VERSION, sessionId, hostId: hostId || undefined, workspaceRoot: root, updatedAt: new Date().toISOString() };
|
|
104
|
+
let temporary;
|
|
105
|
+
try {
|
|
106
|
+
fs.mkdirSync(path.dirname(target), { recursive: true });
|
|
107
|
+
temporary = `${target}.${process.pid}.${Date.now()}.tmp`;
|
|
108
|
+
fs.writeFileSync(temporary, JSON.stringify(record, null, 2), 'utf8');
|
|
109
|
+
if (fs.existsSync(target)) fs.unlinkSync(temporary);
|
|
110
|
+
else fs.renameSync(temporary, target);
|
|
111
|
+
} catch (err) {
|
|
112
|
+
try { if (temporary && fs.existsSync(temporary)) fs.unlinkSync(temporary); } catch (cleanupErr) { /* ignore */ }
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
|
|
116
|
+
function getWorkspaceRoot(context) {
|
|
117
|
+
const sessionId = getSessionId(context);
|
|
118
|
+
const bound = getRegistryRecord(sessionId, context);
|
|
119
|
+
if (bound) return bound.workspaceRoot;
|
|
120
|
+
const explicit = context && (context.workspace_root || context.workspaceRoot || context.cwd);
|
|
121
|
+
const hostRoot = explicit || process.env.HARNESS_WORKSPACE_ROOT || process.env.FABLE_WORKSPACE_ROOT;
|
|
122
|
+
const root = findWorkspaceRoot(hostRoot || process.cwd(), Boolean(hostRoot));
|
|
123
|
+
if (sessionId) bindWorkspace(sessionId, root, context);
|
|
124
|
+
return root;
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
function getWorkspaceStateDir(root) {
|
|
128
|
+
const home = process.env.HARNESS_STATE_HOME || path.join(os.homedir(), '.agents', 'harness-everything');
|
|
129
|
+
if (!root) return path.join(home, 'workspaces', UNBOUND_WORKSPACE_KEY);
|
|
130
|
+
let real = path.resolve(root);
|
|
131
|
+
try { real = fs.realpathSync(real); } catch (err) { /* path may not exist yet */ }
|
|
132
|
+
const slug = path.basename(real).toLowerCase().replace(/[^a-z0-9._-]+/g, '-').replace(/^-+|-+$/g, '') || 'workspace';
|
|
133
|
+
const hashInput = process.platform === 'win32' ? real.toLowerCase() : real;
|
|
134
|
+
const hash = crypto.createHash('sha1').update(hashInput).digest('hex').slice(0, 12);
|
|
135
|
+
return path.join(home, 'workspaces', `${slug}-${hash}`);
|
|
136
|
+
}
|
|
7
137
|
const REQUIRED = ['stageBrief', 'passCondition', 'verificationCommand', 'verifierResult'];
|
|
8
138
|
const VALID_VERIFIER_RESULTS = new Set(['pending', 'pass', 'fail', 'not-run', 'blocked']);
|
|
9
139
|
|
|
@@ -101,8 +231,9 @@ function printHelp() {
|
|
|
101
231
|
console.log('Usage: node fable-mode/scripts/model-selector.js --requested <haiku|sonnet|sonnect|opus> --available <models> --available-agents <agent names> --stage-brief <text> --pass-condition <text> --verification-command <command> --verifier-result <pending|pass|fail|not-run|blocked> [--fallback <inline|stop>] [--host-model <model>] [--audit-file <path>]');
|
|
102
232
|
}
|
|
103
233
|
|
|
104
|
-
function appendAuditRecord(record, auditFile) {
|
|
105
|
-
const target = auditFile || process.env.FABLE_AUDIT_FILE ||
|
|
234
|
+
function appendAuditRecord(record, auditFile, context) {
|
|
235
|
+
const target = auditFile || process.env.FABLE_AUDIT_FILE ||
|
|
236
|
+
path.join(getWorkspaceStateDir(getWorkspaceRoot(context)), 'state', 'fable-mode', 'audit.jsonl');
|
|
106
237
|
fs.mkdirSync(path.dirname(path.resolve(target)), { recursive: true });
|
|
107
238
|
fs.appendFileSync(target, `${JSON.stringify(record)}\n`, 'utf8');
|
|
108
239
|
}
|
|
@@ -115,7 +246,7 @@ function main(argv = process.argv.slice(2)) {
|
|
|
115
246
|
return 0;
|
|
116
247
|
}
|
|
117
248
|
const record = resolveMode(input);
|
|
118
|
-
appendAuditRecord(record, input.auditFile);
|
|
249
|
+
appendAuditRecord(record, input.auditFile, input);
|
|
119
250
|
console.log(JSON.stringify(record, null, 2));
|
|
120
251
|
return record.status === 'blocked' ? 2 : 0;
|
|
121
252
|
} catch (error) {
|