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 +1,46 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
'use strict';
|
|
3
|
+
|
|
4
|
+
const { finalizeExecution } = require('./action-gate');
|
|
5
|
+
|
|
6
|
+
function inferOutcome(payload) {
|
|
7
|
+
const response = (payload && (payload.tool_response || payload.toolResponse)) || {};
|
|
8
|
+
const rawExitCode = response.exitCode ?? response.exit_code ?? payload?.exitCode ?? payload?.exit_code;
|
|
9
|
+
if (typeof rawExitCode === 'number') return rawExitCode === 0 ? 'success' : 'failure';
|
|
10
|
+
if (response.success === false || response.ok === false) return 'failure';
|
|
11
|
+
|
|
12
|
+
const stderr = response.stderr ?? payload?.stderr ?? '';
|
|
13
|
+
if (typeof stderr === 'string' && stderr.trim()) {
|
|
14
|
+
if (/\b(error|fail|failed|failure|exception|fatal|panic|traceback|denied|refused|cannot|unable)\b/i.test(stderr)) {
|
|
15
|
+
return 'failure';
|
|
16
|
+
}
|
|
17
|
+
}
|
|
18
|
+
return 'success';
|
|
19
|
+
}
|
|
20
|
+
|
|
21
|
+
function processPayload(payload) {
|
|
22
|
+
return finalizeExecution(payload, inferOutcome(payload));
|
|
23
|
+
}
|
|
24
|
+
|
|
25
|
+
function run() {
|
|
26
|
+
let raw = '';
|
|
27
|
+
process.stdin.setEncoding('utf8');
|
|
28
|
+
process.stdin.on('data', chunk => { raw += chunk; });
|
|
29
|
+
process.stdin.on('end', () => {
|
|
30
|
+
try {
|
|
31
|
+
processPayload(JSON.parse(raw || '{}'));
|
|
32
|
+
} catch (err) {
|
|
33
|
+
// PostToolUse runs after the side effect. Audit failures must never be
|
|
34
|
+
// misrepresented as a successful enforcement decision or block the host.
|
|
35
|
+
console.error(`[Harness actionGate] Codex post-tool audit failed: ${err.message}`);
|
|
36
|
+
}
|
|
37
|
+
process.exitCode = 0;
|
|
38
|
+
});
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
if (require.main === module) run();
|
|
42
|
+
|
|
43
|
+
module.exports = {
|
|
44
|
+
inferOutcome,
|
|
45
|
+
processPayload,
|
|
46
|
+
};
|
|
@@ -0,0 +1,140 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
'use strict';
|
|
3
|
+
|
|
4
|
+
const crypto = require('crypto');
|
|
5
|
+
const fs = require('fs');
|
|
6
|
+
const path = require('path');
|
|
7
|
+
const {
|
|
8
|
+
getWorkspaceRoot,
|
|
9
|
+
getSessionDir,
|
|
10
|
+
} = require('./lib/harness-state');
|
|
11
|
+
const {
|
|
12
|
+
RULES_PATH,
|
|
13
|
+
classify,
|
|
14
|
+
exactPayloadText,
|
|
15
|
+
loadRuleTable,
|
|
16
|
+
} = require('./action-gate');
|
|
17
|
+
|
|
18
|
+
function hookEventOf(payload) {
|
|
19
|
+
return String((payload && (payload.hook_event_name || payload.hookEventName)) || 'PermissionRequest');
|
|
20
|
+
}
|
|
21
|
+
|
|
22
|
+
function toolNameOf(payload) {
|
|
23
|
+
return String((payload && (payload.tool_name || payload.toolName)) || '');
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
function hashExact(value) {
|
|
27
|
+
return crypto.createHash('sha256').update(String(value || ''), 'utf8').digest('hex');
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
function safePart(value, fallback) {
|
|
31
|
+
const text = String(value || '').replace(/[^A-Za-z0-9._-]+/g, '_').slice(0, 120);
|
|
32
|
+
return text || fallback;
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
function sessionContext(payload, overrideSessionDir) {
|
|
36
|
+
if (overrideSessionDir) return { sessionDir: overrideSessionDir };
|
|
37
|
+
const root = getWorkspaceRoot(payload);
|
|
38
|
+
const sessionId = payload && (payload.session_id || payload.sessionId);
|
|
39
|
+
return { sessionDir: getSessionDir(root, sessionId, payload) };
|
|
40
|
+
}
|
|
41
|
+
|
|
42
|
+
function atomicWriteJson(filePath, value) {
|
|
43
|
+
fs.mkdirSync(path.dirname(filePath), { recursive: true });
|
|
44
|
+
const temp = `${filePath}.${process.pid}.${Date.now()}.tmp`;
|
|
45
|
+
fs.writeFileSync(temp, `${JSON.stringify(value, null, 2)}\n`, 'utf8');
|
|
46
|
+
fs.renameSync(temp, filePath);
|
|
47
|
+
}
|
|
48
|
+
|
|
49
|
+
function auditPermissionRequest(payload, options = {}) {
|
|
50
|
+
const event = hookEventOf(payload);
|
|
51
|
+
if (event !== 'PermissionRequest') return { kind: 'noop', exitCode: 0, record: null, auditFile: null };
|
|
52
|
+
|
|
53
|
+
const context = sessionContext(payload, options.sessionDir);
|
|
54
|
+
const table = options.ruleTable || loadRuleTable(options.rulePath || RULES_PATH);
|
|
55
|
+
const rule = classify(payload, table.rules, context.sessionDir);
|
|
56
|
+
const exact = exactPayloadText(payload);
|
|
57
|
+
const payloadHash = hashExact(exact);
|
|
58
|
+
const turnId = (payload && (payload.turn_id || payload.turnId)) || null;
|
|
59
|
+
const now = new Date().toISOString();
|
|
60
|
+
const record = {
|
|
61
|
+
schemaVersion: 1,
|
|
62
|
+
event: 'PermissionRequest',
|
|
63
|
+
host: 'codex',
|
|
64
|
+
turnId,
|
|
65
|
+
toolName: toolNameOf(payload),
|
|
66
|
+
payloadHash,
|
|
67
|
+
matchedRule: rule ? rule.id : null,
|
|
68
|
+
ruleReason: rule ? rule.reason : null,
|
|
69
|
+
permissionMode: (payload && (payload.permission_mode || payload.permissionMode)) || null,
|
|
70
|
+
disposition: 'deferred-to-codex-approval',
|
|
71
|
+
correlation: 'turn-id-and-payload-hash-only',
|
|
72
|
+
exactToolCallCorrelation: false,
|
|
73
|
+
ruleTableSource: table.source,
|
|
74
|
+
degradedRuleTable: table.degraded,
|
|
75
|
+
degradationReason: table.degradationReason,
|
|
76
|
+
createdAt: now,
|
|
77
|
+
updatedAt: now,
|
|
78
|
+
};
|
|
79
|
+
|
|
80
|
+
const dir = path.join(context.sessionDir, 'action-gate', 'codex-permission-request');
|
|
81
|
+
const file = path.join(dir, `${safePart(turnId, 'turn')}-${payloadHash.slice(0, 24)}.json`);
|
|
82
|
+
atomicWriteJson(file, record);
|
|
83
|
+
|
|
84
|
+
// PermissionRequest means Codex is already entering its native approval flow.
|
|
85
|
+
// Returning no allow/deny decision preserves that flow. This hook must never
|
|
86
|
+
// be treated as a way to manufacture an approval prompt for an otherwise
|
|
87
|
+
// permitted action.
|
|
88
|
+
return { kind: 'defer', exitCode: 0, stdout: null, stderr: null, record, auditFile: file };
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
function internalErrorResult(err) {
|
|
92
|
+
return {
|
|
93
|
+
kind: 'audit-error',
|
|
94
|
+
exitCode: 0,
|
|
95
|
+
stdout: null,
|
|
96
|
+
stderr: `[Harness actionGate] Codex PermissionRequest audit failed: ${err.message}. Native Codex approval flow remains authoritative.`,
|
|
97
|
+
};
|
|
98
|
+
}
|
|
99
|
+
|
|
100
|
+
function processPayload(payload, options = {}) {
|
|
101
|
+
try {
|
|
102
|
+
return auditPermissionRequest(payload, options);
|
|
103
|
+
} catch (err) {
|
|
104
|
+
return internalErrorResult(err);
|
|
105
|
+
}
|
|
106
|
+
}
|
|
107
|
+
|
|
108
|
+
function emitResult(result) {
|
|
109
|
+
if (result.stderr) console.error(result.stderr);
|
|
110
|
+
if (result.stdout) process.stdout.write(`${JSON.stringify(result.stdout)}\n`);
|
|
111
|
+
return result.exitCode || 0;
|
|
112
|
+
}
|
|
113
|
+
|
|
114
|
+
function run() {
|
|
115
|
+
let raw = '';
|
|
116
|
+
process.stdin.setEncoding('utf8');
|
|
117
|
+
process.stdin.on('data', chunk => { raw += chunk; });
|
|
118
|
+
process.stdin.on('end', () => {
|
|
119
|
+
let payload = {};
|
|
120
|
+
try {
|
|
121
|
+
payload = JSON.parse(raw || '{}');
|
|
122
|
+
} catch (err) {
|
|
123
|
+
process.exitCode = emitResult(internalErrorResult(new Error(`invalid hook JSON: ${err.message}`)));
|
|
124
|
+
return;
|
|
125
|
+
}
|
|
126
|
+
process.exitCode = emitResult(processPayload(payload));
|
|
127
|
+
});
|
|
128
|
+
process.stdin.on('error', err => {
|
|
129
|
+
process.exitCode = emitResult(internalErrorResult(err));
|
|
130
|
+
});
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
if (require.main === module) run();
|
|
134
|
+
|
|
135
|
+
module.exports = {
|
|
136
|
+
auditPermissionRequest,
|
|
137
|
+
hashExact,
|
|
138
|
+
internalErrorResult,
|
|
139
|
+
processPayload,
|
|
140
|
+
};
|
|
@@ -0,0 +1,115 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
const { execSync } = require('child_process');
|
|
3
|
+
const fs = require('fs');
|
|
4
|
+
const path = require('path');
|
|
5
|
+
const { getWorkspaceRoot, getSessionDir } = require('./lib/harness-state');
|
|
6
|
+
|
|
7
|
+
function main(payload) {
|
|
8
|
+
try {
|
|
9
|
+
// Let's run a quick Git check to estimate context size from modified files
|
|
10
|
+
const root = getWorkspaceRoot(payload);
|
|
11
|
+
const statusRaw = execSync('git status --porcelain', { cwd: root, encoding: 'utf8', stdio: ['ignore', 'pipe', 'ignore'] }).trim();
|
|
12
|
+
if (!statusRaw) {
|
|
13
|
+
process.exit(0);
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
const lines = statusRaw.split('\n').filter(Boolean);
|
|
17
|
+
const modifiedCount = lines.length;
|
|
18
|
+
|
|
19
|
+
let totalDiffLines = 0;
|
|
20
|
+
try {
|
|
21
|
+
const diffStat = execSync('git diff --numstat', { cwd: root, encoding: 'utf8', stdio: ['ignore', 'pipe', 'ignore'] }).trim();
|
|
22
|
+
if (diffStat) {
|
|
23
|
+
diffStat.split('\n').forEach(line => {
|
|
24
|
+
const [added, deleted] = line.split('\t');
|
|
25
|
+
if (added && added !== '-' && deleted && deleted !== '-') {
|
|
26
|
+
totalDiffLines += parseInt(added) + parseInt(deleted);
|
|
27
|
+
}
|
|
28
|
+
});
|
|
29
|
+
}
|
|
30
|
+
} catch (err) {
|
|
31
|
+
// Ignore git errors
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
// Thresholds for warning
|
|
35
|
+
const HIGH_FILE_LIMIT = 6;
|
|
36
|
+
const HIGH_LINE_LIMIT = 400;
|
|
37
|
+
|
|
38
|
+
const overThreshold = modifiedCount >= HIGH_FILE_LIMIT || totalDiffLines >= HIGH_LINE_LIMIT;
|
|
39
|
+
|
|
40
|
+
// Throttle: this hook fires on EVERY Bash/Edit/Write call, and a warning
|
|
41
|
+
// about context bloat that repeats on every call becomes context bloat
|
|
42
|
+
// itself (and trains the model to ignore hook output). Warn on the first
|
|
43
|
+
// crossing, then only every 10th call while still over - or when the diff
|
|
44
|
+
// has grown another 50% since the last warning.
|
|
45
|
+
const stateFile = path.join(getSessionDir(root, payload && (payload.session_id || payload.sessionId)), 'context-compact-state.json');
|
|
46
|
+
let state = { overCount: 0, lastWarnedLines: 0 };
|
|
47
|
+
try {
|
|
48
|
+
if (fs.existsSync(stateFile)) state = JSON.parse(fs.readFileSync(stateFile, 'utf8'));
|
|
49
|
+
} catch (err) { /* use default */ }
|
|
50
|
+
|
|
51
|
+
if (!overThreshold) {
|
|
52
|
+
if (state.overCount > 0) {
|
|
53
|
+
try {
|
|
54
|
+
fs.writeFileSync(stateFile, JSON.stringify({ overCount: 0, lastWarnedLines: 0 }), 'utf8');
|
|
55
|
+
} catch (err) { /* ignore */ }
|
|
56
|
+
}
|
|
57
|
+
process.exit(0);
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
state.overCount += 1;
|
|
61
|
+
const onBeat = state.overCount === 1 || state.overCount % 10 === 0;
|
|
62
|
+
const grewALot = totalDiffLines >= (state.lastWarnedLines || 0) * 1.5 && totalDiffLines > state.lastWarnedLines;
|
|
63
|
+
const shouldWarn = onBeat || grewALot;
|
|
64
|
+
if (shouldWarn) state.lastWarnedLines = totalDiffLines;
|
|
65
|
+
try {
|
|
66
|
+
fs.writeFileSync(stateFile, JSON.stringify(state), 'utf8');
|
|
67
|
+
} catch (err) { /* ignore */ }
|
|
68
|
+
|
|
69
|
+
if (shouldWarn) {
|
|
70
|
+
// Plain stdout on exit 0 is only surfaced to Claude for UserPromptSubmit/
|
|
71
|
+
// SessionStart - for PreToolUse it's written to a debug log Claude never
|
|
72
|
+
// sees. To actually get an advisory (non-blocking) message in front of
|
|
73
|
+
// Claude here, it has to go through hookSpecificOutput.additionalContext.
|
|
74
|
+
const message = [
|
|
75
|
+
`Harness OS - Context Bloat Warning`,
|
|
76
|
+
`- Modified Files: ${modifiedCount}`,
|
|
77
|
+
`- Accumulated Diff: ${totalDiffLines} lines`,
|
|
78
|
+
``,
|
|
79
|
+
`Caution: Large number of modified files or big diffs will bloat the Agent context window, leading to:`,
|
|
80
|
+
` 1. "Lost in the Middle" reasoning degradation`,
|
|
81
|
+
` 2. Rapidly inflating token costs`,
|
|
82
|
+
``,
|
|
83
|
+
`Recommended Action:`,
|
|
84
|
+
` - Run atomic commits or stage completed files.`,
|
|
85
|
+
` - Use targeted sub-folders rather than broad searches.`,
|
|
86
|
+
` - Ask user to run '/compact' or '/clear' if supported.`
|
|
87
|
+
].join('\n');
|
|
88
|
+
|
|
89
|
+
console.log(JSON.stringify({
|
|
90
|
+
hookSpecificOutput: {
|
|
91
|
+
hookEventName: "PreToolUse",
|
|
92
|
+
additionalContext: message
|
|
93
|
+
}
|
|
94
|
+
}));
|
|
95
|
+
}
|
|
96
|
+
} catch (err) {
|
|
97
|
+
// Fail open - never block user execution on compaction checker
|
|
98
|
+
}
|
|
99
|
+
|
|
100
|
+
process.exit(0);
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
let inputData = '';
|
|
104
|
+
const timeout = setTimeout(() => { main(null); }, 200);
|
|
105
|
+
process.stdin.on('data', chunk => { inputData += chunk; });
|
|
106
|
+
process.stdin.on('end', () => {
|
|
107
|
+
clearTimeout(timeout);
|
|
108
|
+
let payload = null;
|
|
109
|
+
try { payload = JSON.parse(inputData.trim()); } catch (err) { /* no payload */ }
|
|
110
|
+
main(payload);
|
|
111
|
+
});
|
|
112
|
+
process.stdin.on('error', () => {
|
|
113
|
+
clearTimeout(timeout);
|
|
114
|
+
main(null);
|
|
115
|
+
});
|
|
@@ -0,0 +1,146 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
/**
|
|
3
|
+
* Contract Test (PostToolUse: Bash|PowerShell)
|
|
4
|
+
*
|
|
5
|
+
* Phase 3 resolves Fable stage checks against run-scoped contracts under the
|
|
6
|
+
* workspace-keyed global state root:
|
|
7
|
+
* state/fable-runs/<runId>/contracts/<stageId>.json
|
|
8
|
+
*
|
|
9
|
+
* A check is updated only when the command can be correlated to exactly one
|
|
10
|
+
* run/stage (session/worker metadata narrows the match when the host exposes
|
|
11
|
+
* it). Ambiguous matches are surfaced instead of mutating multiple runs.
|
|
12
|
+
* Verification evidence is written beside the run so planId/runId/stageId
|
|
13
|
+
* remain auditable without moving Fable's audit/model-selection ownership.
|
|
14
|
+
*/
|
|
15
|
+
const fs = require('fs');
|
|
16
|
+
const path = require('path');
|
|
17
|
+
const { getSessionId, getWorkspaceRoot, getStateRoot } = require('./lib/harness-state');
|
|
18
|
+
const {
|
|
19
|
+
atomicWriteJson,
|
|
20
|
+
chooseCorrelatedContract,
|
|
21
|
+
commandMatches,
|
|
22
|
+
listRunContracts,
|
|
23
|
+
} = require('./lib/fable-contracts');
|
|
24
|
+
|
|
25
|
+
function resultState(payload) {
|
|
26
|
+
const toolResponse = payload.tool_response || payload.tool_result || {};
|
|
27
|
+
const stdout = toolResponse.stdout ?? payload.stdout ?? '';
|
|
28
|
+
const stderr = toolResponse.stderr ?? payload.stderr ?? '';
|
|
29
|
+
const rawExitCode = toolResponse.exitCode ?? toolResponse.exit_code ?? toolResponse.code ?? payload.exitCode;
|
|
30
|
+
const exitCode = typeof rawExitCode === 'number' ? rawExitCode : undefined;
|
|
31
|
+
const stderrSignal = typeof stderr === 'string' && stderr.trim().length > 0;
|
|
32
|
+
const failed = toolResponse.is_error === true || Boolean(toolResponse.error) ||
|
|
33
|
+
(exitCode !== undefined && exitCode !== 0) ||
|
|
34
|
+
(exitCode === undefined && stderrSignal);
|
|
35
|
+
const evidence = ((stderrSignal ? stderr : stdout) || '').slice(-1000).trim();
|
|
36
|
+
return { failed, exitCode, evidence };
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
function updateRunContract(entry, payload, command, sessionId) {
|
|
40
|
+
const { failed, exitCode, evidence } = resultState(payload);
|
|
41
|
+
const observedAt = new Date().toISOString();
|
|
42
|
+
const contract = {
|
|
43
|
+
...entry.contract,
|
|
44
|
+
status: failed ? 'fail' : 'pass',
|
|
45
|
+
lastObservedCommand: command,
|
|
46
|
+
lastObservedExitCode: exitCode ?? null,
|
|
47
|
+
evidence,
|
|
48
|
+
verifiedAt: observedAt,
|
|
49
|
+
updatedAt: observedAt,
|
|
50
|
+
};
|
|
51
|
+
const evidenceRelative = path.posix.join('evidence', `${contract.stageId}.json`);
|
|
52
|
+
contract.verificationEvidence = evidenceRelative;
|
|
53
|
+
|
|
54
|
+
atomicWriteJson(path.join(entry.runRoot, evidenceRelative), {
|
|
55
|
+
schemaVersion: 1,
|
|
56
|
+
planId: contract.planId,
|
|
57
|
+
runId: contract.runId,
|
|
58
|
+
stageId: contract.stageId,
|
|
59
|
+
sessionId: sessionId || contract.sessionId || null,
|
|
60
|
+
workerId: contract.workerId || null,
|
|
61
|
+
checkCommand: command,
|
|
62
|
+
status: failed ? 'fail' : 'pass',
|
|
63
|
+
exitCode: exitCode ?? null,
|
|
64
|
+
evidence,
|
|
65
|
+
observedAt,
|
|
66
|
+
});
|
|
67
|
+
atomicWriteJson(entry.filePath, contract);
|
|
68
|
+
return contract;
|
|
69
|
+
}
|
|
70
|
+
|
|
71
|
+
function resolveLegacyContract(stateRoot, command, payload) {
|
|
72
|
+
const contractsDir = path.join(stateRoot, 'contracts');
|
|
73
|
+
let files = [];
|
|
74
|
+
try { files = fs.readdirSync(contractsDir).filter(file => file.endsWith('.json')); } catch (_) { return null; }
|
|
75
|
+
const matches = [];
|
|
76
|
+
for (const file of files) {
|
|
77
|
+
const filePath = path.join(contractsDir, file);
|
|
78
|
+
let manifest;
|
|
79
|
+
try { manifest = JSON.parse(fs.readFileSync(filePath, 'utf8')); } catch (_) { continue; }
|
|
80
|
+
if (manifest.status !== 'pending' || !manifest.checkCommand || manifest.checkCommand.trim() !== command) continue;
|
|
81
|
+
matches.push({ manifest, filePath });
|
|
82
|
+
}
|
|
83
|
+
if (matches.length !== 1) return matches.length > 1 ? { ambiguous: true } : null;
|
|
84
|
+
const { failed, evidence } = resultState(payload);
|
|
85
|
+
const match = matches[0];
|
|
86
|
+
match.manifest.status = failed ? 'fail' : 'pass';
|
|
87
|
+
match.manifest.evidence = evidence;
|
|
88
|
+
match.manifest.verifiedAt = new Date().toISOString();
|
|
89
|
+
atomicWriteJson(match.filePath, match.manifest);
|
|
90
|
+
return { legacy: true, failed, contract: match.manifest };
|
|
91
|
+
}
|
|
92
|
+
|
|
93
|
+
let inputData = '';
|
|
94
|
+
process.stdin.on('data', chunk => { inputData += chunk; });
|
|
95
|
+
process.stdin.on('end', () => {
|
|
96
|
+
try {
|
|
97
|
+
const payload = JSON.parse(inputData);
|
|
98
|
+
if (payload.tool_name !== 'Bash' && payload.tool_name !== 'PowerShell') process.exit(0);
|
|
99
|
+
|
|
100
|
+
const command = ((payload.tool_input && payload.tool_input.command) || '').trim();
|
|
101
|
+
if (!command) process.exit(0);
|
|
102
|
+
|
|
103
|
+
const root = getWorkspaceRoot(payload);
|
|
104
|
+
const stateRoot = getStateRoot(root, payload);
|
|
105
|
+
const sessionId = getSessionId(payload);
|
|
106
|
+
const runMatches = listRunContracts(stateRoot).filter(({ contract }) =>
|
|
107
|
+
['pending', 'planned', 'running'].includes(contract.status) && commandMatches(contract, command)
|
|
108
|
+
);
|
|
109
|
+
|
|
110
|
+
if (runMatches.length > 0) {
|
|
111
|
+
const resolution = chooseCorrelatedContract(runMatches, payload, sessionId);
|
|
112
|
+
if (resolution.ambiguous || !resolution.match) {
|
|
113
|
+
console.error(`[Contract Test] Ambiguous Fable check command matched ${runMatches.length} active run contract(s): ${command}`);
|
|
114
|
+
console.error('No contract was mutated. Correlate the worker/session to one run before accepting this check.');
|
|
115
|
+
process.exit(2);
|
|
116
|
+
}
|
|
117
|
+
|
|
118
|
+
const contract = updateRunContract(resolution.match, payload, command, sessionId);
|
|
119
|
+
console.log(`[Contract Test] ${contract.planId}/${contract.runId}/${contract.stageId}: ${contract.status.toUpperCase()} ${contract.checkCommand}`);
|
|
120
|
+
if (contract.status === 'fail') {
|
|
121
|
+
console.error(`Evidence: ${contract.evidence || '(no output captured)'}`);
|
|
122
|
+
console.error(`Do not build on this stage's output until it passes. Artifact: ${contract.outputPath || '(no path recorded)'}`);
|
|
123
|
+
process.exit(2);
|
|
124
|
+
}
|
|
125
|
+
process.exit(0);
|
|
126
|
+
}
|
|
127
|
+
|
|
128
|
+
// Temporary compatibility with pre-Phase-3 manifests. New orchestration
|
|
129
|
+
// must use run-scoped contracts; legacy manifests have no plan/run link.
|
|
130
|
+
const legacy = resolveLegacyContract(stateRoot, command, payload);
|
|
131
|
+
if (legacy && legacy.ambiguous) {
|
|
132
|
+
console.error(`[Contract Test] Ambiguous legacy stage contracts matched: ${command}`);
|
|
133
|
+
process.exit(2);
|
|
134
|
+
}
|
|
135
|
+
if (legacy && legacy.failed) {
|
|
136
|
+
console.error(`[Contract Test] Legacy stage "${legacy.contract.stageId}" FAILED its named check: ${legacy.contract.checkCommand}`);
|
|
137
|
+
console.error(`Evidence: ${legacy.contract.evidence || '(no output captured)'}`);
|
|
138
|
+
process.exit(2);
|
|
139
|
+
}
|
|
140
|
+
process.exit(0);
|
|
141
|
+
} catch (_) {
|
|
142
|
+
// Hook lookup/parsing remains best-effort. A resolved failed check above
|
|
143
|
+
// never passes through this path because it exits explicitly.
|
|
144
|
+
process.exit(0);
|
|
145
|
+
}
|
|
146
|
+
});
|
|
@@ -0,0 +1,65 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
/**
|
|
3
|
+
* Depth Guard (PreToolUse: Edit|Write)
|
|
4
|
+
* Claude Code already refuses to Edit a file that wasn't Read first in the
|
|
5
|
+
* same session - that's native behavior, not something this hook needs to
|
|
6
|
+
* re-check. The gap is Write: it happily overwrites an existing file with
|
|
7
|
+
* no prior Read at all. This hook closes that gap: blocks Write on a file
|
|
8
|
+
* that already exists on disk unless a Read of that exact path shows up
|
|
9
|
+
* earlier in this session's transcript.
|
|
10
|
+
* Fails open on any parse/lookup error - never blocks due to a bug here.
|
|
11
|
+
*/
|
|
12
|
+
const fs = require('fs');
|
|
13
|
+
const path = require('path');
|
|
14
|
+
|
|
15
|
+
let inputData = '';
|
|
16
|
+
process.stdin.on('data', chunk => { inputData += chunk; });
|
|
17
|
+
process.stdin.on('end', () => {
|
|
18
|
+
try {
|
|
19
|
+
const payload = JSON.parse(inputData);
|
|
20
|
+
if (payload.tool_name !== 'Write') {
|
|
21
|
+
process.exit(0);
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
const filePath = payload.tool_input && payload.tool_input.file_path;
|
|
25
|
+
if (!filePath || !fs.existsSync(filePath)) {
|
|
26
|
+
// New file - nothing to have read first.
|
|
27
|
+
process.exit(0);
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
const transcriptPath = payload.transcript_path;
|
|
31
|
+
if (!transcriptPath || !fs.existsSync(transcriptPath)) {
|
|
32
|
+
// Can't verify either way - fail open.
|
|
33
|
+
process.exit(0);
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
const target = path.resolve(filePath).toLowerCase();
|
|
37
|
+
const lines = fs.readFileSync(transcriptPath, 'utf8').split('\n').filter(Boolean);
|
|
38
|
+
|
|
39
|
+
const wasRead = lines.some(line => {
|
|
40
|
+
try {
|
|
41
|
+
const entry = JSON.parse(line);
|
|
42
|
+
const content = entry && entry.message && entry.message.content;
|
|
43
|
+
if (!Array.isArray(content)) return false;
|
|
44
|
+
return content.some(block => {
|
|
45
|
+
if (!block || block.type !== 'tool_use' || block.name !== 'Read') return false;
|
|
46
|
+
const readPath = block.input && block.input.file_path;
|
|
47
|
+
return readPath && path.resolve(readPath).toLowerCase() === target;
|
|
48
|
+
});
|
|
49
|
+
} catch (e) {
|
|
50
|
+
return false;
|
|
51
|
+
}
|
|
52
|
+
});
|
|
53
|
+
|
|
54
|
+
if (!wasRead) {
|
|
55
|
+
console.error(`[Depth Guard] BLOCKED: "${filePath}" already exists but hasn't been Read in this session.`);
|
|
56
|
+
console.error(`Write() overwrites the whole file - doing that blind risks destroying content you never looked at.`);
|
|
57
|
+
console.error(`Read the file first (even a quick pass), then retry the Write, or use Edit for a targeted change.`);
|
|
58
|
+
process.exit(2);
|
|
59
|
+
}
|
|
60
|
+
|
|
61
|
+
process.exit(0);
|
|
62
|
+
} catch (err) {
|
|
63
|
+
process.exit(0);
|
|
64
|
+
}
|
|
65
|
+
});
|
|
@@ -0,0 +1,121 @@
|
|
|
1
|
+
'use strict';
|
|
2
|
+
|
|
3
|
+
const fs = require('fs');
|
|
4
|
+
const path = require('path');
|
|
5
|
+
|
|
6
|
+
function readJson(filePath) {
|
|
7
|
+
try { return JSON.parse(fs.readFileSync(filePath, 'utf8')); } catch (_) { return null; }
|
|
8
|
+
}
|
|
9
|
+
|
|
10
|
+
function atomicWriteJson(filePath, value) {
|
|
11
|
+
fs.mkdirSync(path.dirname(filePath), { recursive: true });
|
|
12
|
+
const tempPath = `${filePath}.${process.pid}.${Date.now()}.tmp`;
|
|
13
|
+
fs.writeFileSync(tempPath, `${JSON.stringify(value, null, 2)}\n`, 'utf8');
|
|
14
|
+
fs.renameSync(tempPath, filePath);
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
function listRunContracts(stateRoot) {
|
|
18
|
+
const runsRoot = path.join(stateRoot, 'fable-runs');
|
|
19
|
+
let runDirs = [];
|
|
20
|
+
try {
|
|
21
|
+
runDirs = fs.readdirSync(runsRoot, { withFileTypes: true })
|
|
22
|
+
.filter(entry => entry.isDirectory())
|
|
23
|
+
.map(entry => path.join(runsRoot, entry.name));
|
|
24
|
+
} catch (_) {
|
|
25
|
+
return [];
|
|
26
|
+
}
|
|
27
|
+
|
|
28
|
+
const contracts = [];
|
|
29
|
+
for (const runRoot of runDirs) {
|
|
30
|
+
const contractsDir = path.join(runRoot, 'contracts');
|
|
31
|
+
let files = [];
|
|
32
|
+
try { files = fs.readdirSync(contractsDir).filter(file => file.endsWith('.json')); } catch (_) { continue; }
|
|
33
|
+
for (const file of files) {
|
|
34
|
+
const filePath = path.join(contractsDir, file);
|
|
35
|
+
const contract = readJson(filePath);
|
|
36
|
+
if (!contract || contract.schemaVersion !== 2 || !contract.stageId || !contract.runId || !contract.planId) continue;
|
|
37
|
+
contracts.push({ contract, filePath, runRoot });
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
return contracts;
|
|
41
|
+
}
|
|
42
|
+
|
|
43
|
+
function getWorkerId(payload) {
|
|
44
|
+
if (!payload || typeof payload !== 'object') return null;
|
|
45
|
+
const direct = payload.worker_id || payload.workerId || payload.subagent_id || payload.subagentId || payload.agent_id || payload.agentId;
|
|
46
|
+
if (typeof direct === 'string' && direct.trim()) return direct.trim();
|
|
47
|
+
for (const container of [payload.tool_input, payload.tool, payload.metadata, payload.context]) {
|
|
48
|
+
if (!container || typeof container !== 'object') continue;
|
|
49
|
+
const nested = getWorkerId(container);
|
|
50
|
+
if (nested) return nested;
|
|
51
|
+
}
|
|
52
|
+
return null;
|
|
53
|
+
}
|
|
54
|
+
|
|
55
|
+
function commandMatches(contract, command) {
|
|
56
|
+
return typeof contract.checkCommand === 'string' && contract.checkCommand.trim() === String(command || '').trim();
|
|
57
|
+
}
|
|
58
|
+
|
|
59
|
+
function chooseCorrelatedContract(matches, payload, sessionId) {
|
|
60
|
+
if (!Array.isArray(matches) || matches.length === 0) return { match: null, ambiguous: false };
|
|
61
|
+
let candidates = matches;
|
|
62
|
+
if (sessionId) {
|
|
63
|
+
candidates = candidates.filter(({ contract }) => !contract.sessionId || contract.sessionId === sessionId);
|
|
64
|
+
if (candidates.length === 0) return { match: null, ambiguous: false };
|
|
65
|
+
}
|
|
66
|
+
|
|
67
|
+
const workerId = getWorkerId(payload);
|
|
68
|
+
if (workerId) {
|
|
69
|
+
const hasBoundWorker = candidates.some(({ contract }) => Boolean(contract.workerId));
|
|
70
|
+
if (hasBoundWorker) {
|
|
71
|
+
candidates = candidates.filter(({ contract }) => !contract.workerId || contract.workerId === workerId);
|
|
72
|
+
if (candidates.length === 0) return { match: null, ambiguous: false };
|
|
73
|
+
}
|
|
74
|
+
const workerMatches = candidates.filter(({ contract }) => contract.workerId === workerId);
|
|
75
|
+
if (workerMatches.length === 1) return { match: workerMatches[0], ambiguous: false };
|
|
76
|
+
if (workerMatches.length > 1) return { match: null, ambiguous: true };
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
if (candidates.length === 1) return { match: candidates[0], ambiguous: false };
|
|
80
|
+
return { match: null, ambiguous: candidates.length > 1 };
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
function normalizeChangedPath(statusLine) {
|
|
84
|
+
const text = String(statusLine || '');
|
|
85
|
+
const raw = /^..\s/.test(text) ? text.slice(3).trim() : text.trim();
|
|
86
|
+
const renameTarget = raw.includes(' -> ') ? raw.split(' -> ').pop() : raw;
|
|
87
|
+
return renameTarget.replace(/\\/g, '/').replace(/^\.\//, '');
|
|
88
|
+
}
|
|
89
|
+
|
|
90
|
+
function pathWithinScope(filePath, scope) {
|
|
91
|
+
const file = normalizeChangedPath(filePath).replace(/\/+$/g, '');
|
|
92
|
+
const declared = String(scope || '').replace(/\\/g, '/').replace(/^\.\//, '').replace(/\/+$/g, '');
|
|
93
|
+
if (!file || !declared) return false;
|
|
94
|
+
return file === declared || file.startsWith(`${declared}/`);
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
function stagesForChangedPath(filePath, contractEntries) {
|
|
98
|
+
return (contractEntries || []).filter(({ contract }) =>
|
|
99
|
+
Array.isArray(contract.writeSet) && contract.writeSet.some(scope => pathWithinScope(filePath, scope))
|
|
100
|
+
);
|
|
101
|
+
}
|
|
102
|
+
|
|
103
|
+
function getActiveRunContracts(stateRoot, sessionId) {
|
|
104
|
+
return listRunContracts(stateRoot).filter(({ contract }) => {
|
|
105
|
+
if (!['pending', 'planned', 'running'].includes(contract.status)) return false;
|
|
106
|
+
return !sessionId || !contract.sessionId || contract.sessionId === sessionId;
|
|
107
|
+
});
|
|
108
|
+
}
|
|
109
|
+
|
|
110
|
+
module.exports = {
|
|
111
|
+
atomicWriteJson,
|
|
112
|
+
chooseCorrelatedContract,
|
|
113
|
+
commandMatches,
|
|
114
|
+
getActiveRunContracts,
|
|
115
|
+
getWorkerId,
|
|
116
|
+
listRunContracts,
|
|
117
|
+
normalizeChangedPath,
|
|
118
|
+
pathWithinScope,
|
|
119
|
+
readJson,
|
|
120
|
+
stagesForChangedPath,
|
|
121
|
+
};
|