@piagent/platform 1.6.0 → 1.7.1
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/CHANGELOG.md +104 -0
- package/README.md +24 -15
- package/README.vi.md +36 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +27 -3
- package/benchmarks/production-v1/suite.json +5 -1
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +16 -16
- package/docs/authority-manifest.md +1 -1
- package/docs/auto-delegation-policy.md +38 -73
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +30 -35
- package/docs/context-engine.md +5 -3
- package/docs/context-window-policy.md +37 -1
- package/docs/distribution-standard.md +9 -9
- package/docs/en/architecture.md +26 -13
- package/docs/languages.json +4 -4
- package/docs/operator-manual-vietnamese.md +26 -46
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +7 -0
- package/docs/quality-benchmark.md +299 -23
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +32 -9
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/runtime-harness-standard.md +3 -3
- package/docs/security-threat-model.md +1 -1
- package/docs/subagent-orchestration-capabilities.md +19 -51
- package/docs/subagents-and-multiagent.md +21 -47
- package/docs/task-implementation-contract.md +3 -3
- package/docs/task-lifecycle-vietnamese.md +1 -1
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs/vi/architecture.md +26 -13
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +89 -95
- package/docs-site/commands.html +1 -1
- package/docs-site/content/benchmark.html +88 -94
- package/docs-site/content/en/benchmark.html +134 -90
- package/docs-site/content/en/index.html +19 -19
- package/docs-site/content/en/runtime.html +5 -5
- package/docs-site/content/en/whats-new.html +23 -1
- package/docs-site/content/en/workflows.html +2 -2
- package/docs-site/content/index.html +19 -19
- package/docs-site/content/runtime.html +6 -6
- package/docs-site/content/whats-new.html +23 -1
- package/docs-site/content/workflows.html +2 -2
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +135 -91
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +20 -20
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +6 -6
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -2
- package/docs-site/en/workflows.html +3 -3
- package/docs-site/index.html +20 -20
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +7 -7
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -2
- package/docs-site/workflows.html +3 -3
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
- package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
- package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/helper-request.invalid.json +1 -1
- package/evals/fixtures/helper-request.valid.json +1 -1
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
- package/evals/fs4-readiness-matrix.v1.json +2 -2
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +10 -6
- package/evals/fs5-pilot-protocol.v1.json +3 -3
- package/evals/fs5-pilot-protocol.v2.json +8 -8
- package/evals/fs5-pilot-protocol.v3.json +8 -8
- package/evals/fs5-pilot-protocol.v4.json +7 -7
- package/evals/fs5-pilot-protocol.v5.json +10 -8
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/evals/runtime-conformance-v1/README.md +43 -0
- package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
- package/evals/runtime-conformance-v1/lane.json +57 -0
- package/evals/runtime-conformance-v1/paid-canary.json +31 -0
- package/evals/runtime-conformance-v1/runner.mjs +456 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +15 -6
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +12 -17
- package/packages/piagent-core/extensions/context-index-policy.js +4 -0
- package/packages/piagent-core/extensions/context-query-paths.js +57 -0
- package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
- package/packages/piagent-core/extensions/criterion-graph.js +4 -1
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
- package/packages/piagent-core/extensions/guard-types.ts +9 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +140 -43
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +43 -2
- package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
- package/packages/piagent-core/prompts/be-to-fe.md +2 -8
- package/packages/piagent-core/prompts/plan.md +1 -1
- package/packages/piagent-core/prompts/platform-improve.md +2 -6
- package/packages/piagent-core/prompts/review.md +2 -7
- package/packages/piagent-core/prompts/scout.md +1 -1
- package/packages/piagent-core/prompts/task.md +4 -3
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
- package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
- package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
- package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
- package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
- package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
- package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
- package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
- package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
- package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
- package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
- package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
- package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
- package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
- package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
- package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
- package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
- package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
- package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
- package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
- package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
- package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
- package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
- package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
- package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
- package/packages/piagent-core/subagents/piagent-planner.md +6 -4
- package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
- package/packages/piagent-core/subagents/piagent-scout.md +4 -2
- package/packages/piagent-core/subagents/piagent-worker.md +6 -4
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
- package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
- package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
- package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
- package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
- package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +150 -11
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/project-profile.schema.json +1 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +45 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +224 -387
- package/scripts/benchmark-runner-finalization.mjs +144 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-subagents.sh +42 -34
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/helper-evaluation.mjs +1 -1
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/team-doctor.sh +1 -1
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +41 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
|
@@ -0,0 +1,181 @@
|
|
|
1
|
+
import { createHmac, KeyObject, randomUUID, timingSafeEqual } from "node:crypto";
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
import { types } from "node:util";
|
|
5
|
+
|
|
6
|
+
const VERSION = "composite-assurance-anchor-v1";
|
|
7
|
+
const JOURNAL_VERSION = "composite-assurance-journal-v1";
|
|
8
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
9
|
+
const MAX_ANCHOR_BYTES = 4 * 1024 * 1024;
|
|
10
|
+
const FIELDS = ["version", "filePath", "contextDigest", "state", "previousHead", "nextHead", "line"];
|
|
11
|
+
|
|
12
|
+
type AnchorPayload = { version: string; filePath: string; contextDigest: string; state: "prepared" | "committed";
|
|
13
|
+
previousHead: string | null; nextHead: string; line: string };
|
|
14
|
+
|
|
15
|
+
function exact(value: any, fields: string[], label: string) {
|
|
16
|
+
if (!value || typeof value !== "object" || Array.isArray(value) || types.isProxy(value)
|
|
17
|
+
|| ![Object.prototype, null].includes(Object.getPrototypeOf(value))
|
|
18
|
+
|| Object.keys(value).length !== fields.length || fields.some(field => !Object.hasOwn(value, field))) {
|
|
19
|
+
throw new Error(`Invalid ${label}`);
|
|
20
|
+
}
|
|
21
|
+
return value;
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
function sameFile(a: any, b: any) {
|
|
25
|
+
return ["dev", "ino", "mode", "nlink", "uid", "size", "mtimeNs", "ctimeNs"]
|
|
26
|
+
.every(field => a[field] === b[field]);
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
function writeAll(fd: number, bytes: Buffer, position: number | null = null) {
|
|
30
|
+
for (let offset = 0; offset < bytes.length;) {
|
|
31
|
+
const written = fs.writeSync(fd, bytes, offset, bytes.length - offset,
|
|
32
|
+
position === null ? null : position + offset);
|
|
33
|
+
if (written < 1) throw new Error("Composite anchor write was incomplete");
|
|
34
|
+
offset += written;
|
|
35
|
+
}
|
|
36
|
+
fs.fsyncSync(fd);
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
/** A private second durable record. A prepared anchor contains the exact signed
|
|
40
|
+
* next journal line, so either side of a process crash can be reconciled without
|
|
41
|
+
* accepting a caller-supplied head or discarding a persisted event. */
|
|
42
|
+
export function openCompositeJournalAnchor({ anchorPath, filePath, contextDigest, key }: {
|
|
43
|
+
anchorPath: string; filePath: string; contextDigest: string; key: KeyObject }) {
|
|
44
|
+
if (!(key instanceof KeyObject) || key.type !== "secret" || key.symmetricKeySize !== 32
|
|
45
|
+
|| !path.isAbsolute(anchorPath) || path.resolve(anchorPath) !== anchorPath
|
|
46
|
+
|| path.dirname(anchorPath) !== path.dirname(filePath) || !HASH.test(contextDigest)) {
|
|
47
|
+
throw new TypeError("Invalid composite anchor identity");
|
|
48
|
+
}
|
|
49
|
+
const directory = path.dirname(anchorPath), parent = fs.lstatSync(directory);
|
|
50
|
+
if (fs.realpathSync.native(directory) !== directory || !parent.isDirectory() || parent.isSymbolicLink()
|
|
51
|
+
|| parent.uid !== process.getuid() || (parent.mode & 0o077) !== 0) {
|
|
52
|
+
throw new Error("Composite anchor directory is unsafe");
|
|
53
|
+
}
|
|
54
|
+
const anchorDomain = `${VERSION}\0${anchorPath}\0${contextDigest}`;
|
|
55
|
+
const journalDomain = `${JOURNAL_VERSION}\0${filePath}\0${contextDigest}`;
|
|
56
|
+
const signature = (domain: string, text: string) => createHmac("sha256", key).update(`${domain}\0${text}`).digest("hex");
|
|
57
|
+
|
|
58
|
+
function lineBytes(payload: AnchorPayload): Buffer {
|
|
59
|
+
const bytes = Buffer.from(payload.line, "base64");
|
|
60
|
+
if (!bytes.length || bytes.length > 2 * 1024 * 1024 || bytes.at(-1) !== 10
|
|
61
|
+
|| bytes.toString("base64") !== payload.line) throw new Error("Composite anchor line is invalid");
|
|
62
|
+
const journalEnvelope = exact(JSON.parse(bytes.subarray(0, -1).toString("utf8")), ["payload", "signature"], "anchored journal envelope");
|
|
63
|
+
const text = JSON.stringify(journalEnvelope.payload);
|
|
64
|
+
if (!HASH.test(journalEnvelope.signature) || journalEnvelope.signature !== payload.nextHead
|
|
65
|
+
|| !timingSafeEqual(Buffer.from(signature(journalDomain, text), "hex"), Buffer.from(journalEnvelope.signature, "hex"))
|
|
66
|
+
|| journalEnvelope.payload?.previous !== payload.previousHead) {
|
|
67
|
+
throw new Error("Composite anchor journal signature is invalid");
|
|
68
|
+
}
|
|
69
|
+
return bytes;
|
|
70
|
+
}
|
|
71
|
+
|
|
72
|
+
function validate(payload: any): AnchorPayload {
|
|
73
|
+
exact(payload, FIELDS, "composite anchor payload");
|
|
74
|
+
if (payload.version !== VERSION || payload.filePath !== filePath || payload.contextDigest !== contextDigest
|
|
75
|
+
|| !["prepared", "committed"].includes(payload.state) || !HASH.test(payload.nextHead)
|
|
76
|
+
|| payload.previousHead !== null && !HASH.test(payload.previousHead) || typeof payload.line !== "string") {
|
|
77
|
+
throw new Error("Composite anchor payload is invalid");
|
|
78
|
+
}
|
|
79
|
+
lineBytes(payload); return payload;
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
function read(): AnchorPayload | null {
|
|
83
|
+
if (!fs.existsSync(anchorPath)) return null;
|
|
84
|
+
const atPath = fs.lstatSync(anchorPath, { bigint: true });
|
|
85
|
+
if (!atPath.isFile() || atPath.isSymbolicLink() || atPath.nlink !== 1n
|
|
86
|
+
|| atPath.uid !== BigInt(process.getuid()) || (atPath.mode & 0o077n) !== 0n
|
|
87
|
+
|| atPath.size < 1n || atPath.size > BigInt(MAX_ANCHOR_BYTES)) throw new Error("Composite anchor file is unsafe");
|
|
88
|
+
const fd = fs.openSync(anchorPath, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
89
|
+
try {
|
|
90
|
+
const before = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd), after = fs.fstatSync(fd, { bigint: true });
|
|
91
|
+
if (bytes.length !== Number(before.size) || !sameFile(atPath, before) || !sameFile(before, after)) {
|
|
92
|
+
throw new Error("Composite anchor changed during read");
|
|
93
|
+
}
|
|
94
|
+
const envelope = exact(JSON.parse(new TextDecoder("utf-8", { fatal: true }).decode(bytes)),
|
|
95
|
+
["payload", "signature"], "composite anchor envelope");
|
|
96
|
+
if (!Buffer.from(JSON.stringify(envelope)).equals(bytes) || !HASH.test(envelope.signature)) {
|
|
97
|
+
throw new Error("Composite anchor is noncanonical");
|
|
98
|
+
}
|
|
99
|
+
const text = JSON.stringify(envelope.payload), expected = signature(anchorDomain, text);
|
|
100
|
+
if (!timingSafeEqual(Buffer.from(expected, "hex"), Buffer.from(envelope.signature, "hex"))) {
|
|
101
|
+
throw new Error("Composite anchor signature is invalid");
|
|
102
|
+
}
|
|
103
|
+
return validate(envelope.payload);
|
|
104
|
+
} finally { fs.closeSync(fd); }
|
|
105
|
+
}
|
|
106
|
+
|
|
107
|
+
function replace(payload: AnchorPayload): AnchorPayload {
|
|
108
|
+
validate(payload);
|
|
109
|
+
const text = JSON.stringify(payload), bytes = Buffer.from(JSON.stringify({ payload,
|
|
110
|
+
signature: signature(anchorDomain, text) }));
|
|
111
|
+
if (bytes.length > MAX_ANCHOR_BYTES) throw new Error("Composite anchor byte bound exceeded");
|
|
112
|
+
const temporary = path.join(directory, `.${path.basename(anchorPath)}.${process.pid}.${randomUUID()}.tmp`);
|
|
113
|
+
const fd = fs.openSync(temporary, fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL
|
|
114
|
+
| fs.constants.O_NOFOLLOW, 0o600);
|
|
115
|
+
try {
|
|
116
|
+
try { writeAll(fd, bytes); } finally { fs.closeSync(fd); }
|
|
117
|
+
fs.renameSync(temporary, anchorPath); fs.chmodSync(anchorPath, 0o600);
|
|
118
|
+
const parentFd = fs.openSync(directory, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
119
|
+
try { fs.fsyncSync(parentFd); } finally { fs.closeSync(parentFd); }
|
|
120
|
+
} catch (error) { try { fs.unlinkSync(temporary); } catch {} throw error; }
|
|
121
|
+
const observed = read();
|
|
122
|
+
if (!observed || JSON.stringify(observed) !== JSON.stringify(payload)) throw new Error("Composite anchor readback failed");
|
|
123
|
+
return observed;
|
|
124
|
+
}
|
|
125
|
+
|
|
126
|
+
let trusted = read();
|
|
127
|
+
function current() {
|
|
128
|
+
const observed = read();
|
|
129
|
+
if (JSON.stringify(observed) !== JSON.stringify(trusted)) throw new Error("Composite anchor changed outside this authority");
|
|
130
|
+
return observed;
|
|
131
|
+
}
|
|
132
|
+
function tuple(previousHead: string | null, nextHead: string, line: Buffer, state: "prepared" | "committed"): AnchorPayload {
|
|
133
|
+
return { version: VERSION, filePath, contextDigest, state, previousHead, nextHead, line: line.toString("base64") };
|
|
134
|
+
}
|
|
135
|
+
function prepare(previousHead: string | null, nextHead: string, line: Buffer) {
|
|
136
|
+
const before = current();
|
|
137
|
+
if (before ? before.state !== "committed" || before.nextHead !== previousHead : previousHead !== null) {
|
|
138
|
+
throw new Error("Composite anchor preparation is not linear");
|
|
139
|
+
}
|
|
140
|
+
trusted = replace(tuple(previousHead, nextHead, line, "prepared"));
|
|
141
|
+
}
|
|
142
|
+
function commit(previousHead: string | null, nextHead: string, line: Buffer) {
|
|
143
|
+
const before = current(), prepared = tuple(previousHead, nextHead, line, "prepared");
|
|
144
|
+
if (!before || JSON.stringify(before) !== JSON.stringify(prepared)) throw new Error("Composite anchor preparation changed");
|
|
145
|
+
trusted = replace(tuple(previousHead, nextHead, line, "committed"));
|
|
146
|
+
}
|
|
147
|
+
function repairPreparedTail() {
|
|
148
|
+
const value = current(); if (value?.state !== "prepared" || !fs.existsSync(filePath)) return false;
|
|
149
|
+
const expected = lineBytes(value), atPath = fs.lstatSync(filePath, { bigint: true });
|
|
150
|
+
if (!atPath.isFile() || atPath.isSymbolicLink() || atPath.nlink !== 1n || atPath.uid !== BigInt(process.getuid())
|
|
151
|
+
|| (atPath.mode & 0o077n) !== 0n || atPath.size > BigInt(32 * 1024 * 1024)) throw new Error("Composite journal file is unsafe");
|
|
152
|
+
const fd = fs.openSync(filePath, fs.constants.O_RDWR | fs.constants.O_NOFOLLOW);
|
|
153
|
+
try {
|
|
154
|
+
const source = fs.readFileSync(fd), start = source.lastIndexOf(10) + 1, partial = source.subarray(start);
|
|
155
|
+
if (!partial.length) return false;
|
|
156
|
+
if (partial.length >= expected.length || !expected.subarray(0, partial.length).equals(partial)) {
|
|
157
|
+
throw new Error("Composite prepared journal tail is inconsistent");
|
|
158
|
+
}
|
|
159
|
+
writeAll(fd, expected.subarray(partial.length), source.length); return true;
|
|
160
|
+
} finally { fs.closeSync(fd); }
|
|
161
|
+
}
|
|
162
|
+
function reconcile(journalExists: boolean, journalHead: string | null, append: (line: Buffer, create: boolean) => void) {
|
|
163
|
+
const value = current(); if (!value) throw new Error("Composite journal restart head is not externally anchored");
|
|
164
|
+
const line = lineBytes(value);
|
|
165
|
+
if (value.state === "committed") {
|
|
166
|
+
if (!journalExists || journalHead !== value.nextHead) throw new Error("Composite journal rollback or wrong anchored head observed");
|
|
167
|
+
return value.nextHead;
|
|
168
|
+
}
|
|
169
|
+
if (!journalExists) {
|
|
170
|
+
if (value.previousHead !== null || journalHead !== null) throw new Error("Composite journal creation anchor is inconsistent");
|
|
171
|
+
append(line, true); journalHead = value.nextHead;
|
|
172
|
+
} else if (journalHead === value.previousHead) {
|
|
173
|
+
append(line, false); journalHead = value.nextHead;
|
|
174
|
+
}
|
|
175
|
+
if (journalHead !== value.nextHead) throw new Error("Composite prepared journal head is inconsistent");
|
|
176
|
+
commit(value.previousHead, value.nextHead, line); return value.nextHead;
|
|
177
|
+
}
|
|
178
|
+
return Object.freeze({ anchorPath, exists: () => current() !== null, prepare, commit, repairPreparedTail, reconcile,
|
|
179
|
+
state: () => { const value = current(); return value && Object.freeze({ state: value.state,
|
|
180
|
+
previousHead: value.previousHead, nextHead: value.nextHead }); } });
|
|
181
|
+
}
|
|
@@ -0,0 +1,76 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
|
|
3
|
+
import type { TaskContract } from "../../extensions/guard-types.ts";
|
|
4
|
+
import { readTaskJournal } from "../../extensions/task-journal.js";
|
|
5
|
+
import { applyAssistedReadOnlyFinalHandoff, applyRuntimeLifecycleObservation,
|
|
6
|
+
runtimeLifecycleMode } from "../../extensions/task-lifecycle.js";
|
|
7
|
+
import { readTrajectoryStore } from "../trajectory/trajectory-store.ts";
|
|
8
|
+
|
|
9
|
+
const sha = (value: string) => createHash("sha256").update(value).digest("hex");
|
|
10
|
+
|
|
11
|
+
export function compositeTaskContractDigest(task: TaskContract): string {
|
|
12
|
+
const steps = task.workPlan?.map(({ id, title, role, mode, dependsOn }) => ({ id, title, role, mode, dependsOn })) ?? [];
|
|
13
|
+
const criteria = task.acceptanceReceipt?.criteria?.map(({ id, hash, obligation, priority }) => ({ id, hash, obligation, priority })) ?? [];
|
|
14
|
+
return sha(JSON.stringify({ schemaVersion: task.schemaVersion, taskRunId: task.taskRunId, taskId: task.taskId,
|
|
15
|
+
sessionId: task.sessionId, changeMode: task.changeMode, mutationPolicy: task.mutationPolicy ?? null, attempt: task.attempt,
|
|
16
|
+
maxAttempts: task.maxAttempts, summary: task.summary, operatorRequestDigest: task.operatorRequestDigest ?? null,
|
|
17
|
+
riskLane: task.riskLane, intakeMode: task.intakeMode ?? null, expectedOutput: task.expectedOutput,
|
|
18
|
+
acceptanceCriteria: task.acceptanceCriteria, criteria, criterionGraph: task.criterionGraph ?? null, scope: task.scope,
|
|
19
|
+
outOfScope: task.outOfScope, protectedPaths: task.protectedPaths, requiredContext: task.requiredContext,
|
|
20
|
+
mcpCapabilities: task.mcpCapabilities, verifyGroup: task.verifyGroup ?? null, verifyCommands: task.verifyCommands,
|
|
21
|
+
workPlan: steps, reviewLenses: task.reviewLenses, orchestration: task.orchestration ?? null,
|
|
22
|
+
authoritySnapshot: task.authoritySnapshot ?? null, workingTreeDigestAlgorithm: task.workingTreeDigestAlgorithm,
|
|
23
|
+
baselineChangedFiles: task.baselineChangedFiles, baselineFileDigests: task.baselineFileDigests, createdAt: task.createdAt }));
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
export function compositePhaseHeadDigest(cwd: string, task: TaskContract): string {
|
|
27
|
+
const journal = readTaskJournal(cwd, { taskRunId: task.taskRunId, sessionId: task.sessionId, maximumBytes: 32 * 1024 * 1024 });
|
|
28
|
+
const trajectory = readTrajectoryStore(cwd, task.taskRunId);
|
|
29
|
+
if (journal.corruptions.length || journal.inputTruncated || journal.recoverableTailBytes || !trajectory.enforcementSafe) {
|
|
30
|
+
throw new Error("Task or trajectory journal is not current");
|
|
31
|
+
}
|
|
32
|
+
if (trajectory.state && (trajectory.state.taskRunId !== task.taskRunId || trajectory.state.taskId !== task.taskId)) {
|
|
33
|
+
throw new Error("Trajectory identity does not match the task");
|
|
34
|
+
}
|
|
35
|
+
return sha(JSON.stringify({ taskJournalHead: journal.head ?? null,
|
|
36
|
+
trajectoryState: trajectory.state ? sha(JSON.stringify(trajectory.state)) : null,
|
|
37
|
+
trajectoryEvents: trajectory.events.length ? sha(JSON.stringify(trajectory.events)) : null }));
|
|
38
|
+
}
|
|
39
|
+
|
|
40
|
+
export function assertCompositeCriterion(task: TaskContract, contract: any): void {
|
|
41
|
+
const receipt = task.acceptanceReceipt?.criteria;
|
|
42
|
+
if (!task.operatorRequestDigest || !Array.isArray(task.acceptanceCriteria) || !Array.isArray(receipt)
|
|
43
|
+
|| task.acceptanceCriteria.length !== receipt.length || contract.criterionIndex >= receipt.length
|
|
44
|
+
|| new Set(receipt.map(item => item.id)).size !== receipt.length
|
|
45
|
+
|| receipt.some((item, index) => item.hash !== sha(task.acceptanceCriteria[index]))
|
|
46
|
+
|| task.acceptanceCriteria[contract.criterionIndex] !== contract.criterionText
|
|
47
|
+
|| receipt[contract.criterionIndex].id !== contract.criterionId
|
|
48
|
+
|| receipt[contract.criterionIndex].hash !== contract.criterionHash) {
|
|
49
|
+
throw new Error("Composite plan does not match the exact current task criterion");
|
|
50
|
+
}
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
export function compositeSettlementTaskStatus(factKinds: Iterable<string>): "completed" | "refused" {
|
|
54
|
+
return new Set(factKinds).has("policy-refusal-output") ? "refused" : "completed";
|
|
55
|
+
}
|
|
56
|
+
|
|
57
|
+
/** Pure projection only. The caller must separately prove every supplied kind is
|
|
58
|
+
* a current authenticated fact; this helper never issues acceptance authority. */
|
|
59
|
+
export function projectCompositeLifecycle(task: TaskContract, factKinds: Iterable<string>): TaskContract | false {
|
|
60
|
+
if (task.trace.outcome !== "pending") return false;
|
|
61
|
+
const kinds = new Set(factKinds), projected = structuredClone(task), mode = runtimeLifecycleMode(projected);
|
|
62
|
+
if (["automatic", "assisted"].includes(mode)) {
|
|
63
|
+
if (!["workspace-scope", "tool-policy-complete", "project-verifier-current"].every(kind => kinds.has(kind))) return false;
|
|
64
|
+
applyRuntimeLifecycleObservation(projected, "verification-complete");
|
|
65
|
+
return projected;
|
|
66
|
+
}
|
|
67
|
+
if (["automatic-readonly", "assisted-readonly"].includes(mode)) {
|
|
68
|
+
const substantive = ["context-current", "structured-log-claims", "config-document-literals",
|
|
69
|
+
"policy-refusal-output", "bounded-code-checks"].some(kind => kinds.has(kind));
|
|
70
|
+
if (!substantive || !["workspace-scope", "tool-policy-complete"].every(kind => kinds.has(kind))) return false;
|
|
71
|
+
applyRuntimeLifecycleObservation(projected, "context-complete");
|
|
72
|
+
applyAssistedReadOnlyFinalHandoff(projected, { durableReadEvidence: true, finalHandoffObserved: true });
|
|
73
|
+
return projected;
|
|
74
|
+
}
|
|
75
|
+
return projected;
|
|
76
|
+
}
|
|
@@ -0,0 +1,318 @@
|
|
|
1
|
+
import { createHash, createPublicKey, verify } from "node:crypto";
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
import { isDeepStrictEqual, types } from "node:util";
|
|
5
|
+
import { readTaskBaselineSource } from "./acceptance-api-baseline.js";
|
|
6
|
+
import { readWorkspaceFile } from "../inspection/workspace-file-reader.ts";
|
|
7
|
+
import { workingTreeSnapshot, workingTreeSnapshotHasUnavailableEvidence } from "../../extensions/task-state.js";
|
|
8
|
+
|
|
9
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
10
|
+
const ID = /^[A-Za-z0-9][A-Za-z0-9:._~-]{0,159}$/;
|
|
11
|
+
const MAX_JOURNAL_BYTES = 32 * 1024 * 1024;
|
|
12
|
+
const MAX_ROWS = 2_048;
|
|
13
|
+
const SCOPED_VERIFICATION_PROTOCOL = "scoped-isolated-contract-v1";
|
|
14
|
+
const SCOPED_PROJECT_VERIFICATION_PROTOCOL = "scoped-project-verifier-v1";
|
|
15
|
+
// Legacy receipts cover the fixed Node script groups; v3 docs receipts carry
|
|
16
|
+
// exact commands and signed current-document scope. Neither implies the other.
|
|
17
|
+
const SCOPED_PROJECT_COMMANDS = new Set(["npm run type-check", "npm run lint", "npm test", "npm run test", "npm run test:e2e"]);
|
|
18
|
+
const IDENTITY_V3_FIELDS = ["version", "armId", "taskId", "sessionId", "requestId", "operationId", "nonce",
|
|
19
|
+
"sourceSha256", "assetTreeSha256", "configSha256", "brokerClosureSha256", "toolDefinitionsSha256",
|
|
20
|
+
"manifestAuthoritySha256", "journalSignerSha256", "journalPathSha256", "contextPolicySha256", "sdkTreeSha256"];
|
|
21
|
+
const IDENTITY_V4_FIELDS = [
|
|
22
|
+
...IDENTITY_V3_FIELDS.slice(0, 10), "measurementConfigurationSha256", ...IDENTITY_V3_FIELDS.slice(10)
|
|
23
|
+
];
|
|
24
|
+
const TOOL_DEFINITIONS = [
|
|
25
|
+
{ name: "scoped_read", inputSchema: { type: "object", properties: { materialId: { type: "string" } }, required: ["materialId"], additionalProperties: false } },
|
|
26
|
+
{ name: "scoped_write_document", inputSchema: { type: "object", properties: { materialId: { type: "string" }, expectedSha256: { type: "string" }, utf8: { type: "string" } }, required: ["materialId", "expectedSha256", "utf8"], additionalProperties: false } },
|
|
27
|
+
{ name: "scoped_verify", inputSchema: { type: "object", properties: { verificationId: { type: "string" } }, required: ["verificationId"], additionalProperties: false } }
|
|
28
|
+
];
|
|
29
|
+
const snapshots = new WeakMap<object, Snapshot>();
|
|
30
|
+
const sha = (value: string | Buffer) => createHash("sha256").update(value).digest("hex");
|
|
31
|
+
const exact = (value: any, fields: string[]) => value && typeof value === "object" && !Array.isArray(value)
|
|
32
|
+
&& !types.isProxy(value) && Object.keys(value).length === fields.length && fields.every(field => Object.hasOwn(value, field));
|
|
33
|
+
|
|
34
|
+
type Action = { action: number; tool: string | null; requestSha256: string | null; result: any; receipt?: any };
|
|
35
|
+
type Snapshot = { manifest: any; rows: any[]; actions: Action[]; profileDigest: string; armDigest: string;
|
|
36
|
+
reads: Map<string, Action[]>; writes: Map<string, Action[]>; verifications: Map<string, Action[]> };
|
|
37
|
+
type FactInput = { fact: any; plan: any; contract: any; task: any; materials: readonly any[];
|
|
38
|
+
workspace: Record<string, string>; response?: any; projectRoot?: string };
|
|
39
|
+
|
|
40
|
+
export function scopedBrokerToolDefinitionsSha256(): string { return sha(JSON.stringify(TOOL_DEFINITIONS)); }
|
|
41
|
+
export function scopedBrokerProfileDigest(profile: string): string {
|
|
42
|
+
if (!["incident", "document", "protected-env-refusal", "destructive-history-refusal"].includes(profile))
|
|
43
|
+
throw new TypeError("scoped-profile-invalid");
|
|
44
|
+
return sha(JSON.stringify({ version: 1, profile, mediation: "exclusive-signed-complete-journal",
|
|
45
|
+
toolDefinitionsSha256: scopedBrokerToolDefinitionsSha256() }));
|
|
46
|
+
}
|
|
47
|
+
export function scopedBrokerArmDigest(identity: any): string {
|
|
48
|
+
return sha(JSON.stringify({ version: 1, armId: identity.armId, sourceSha256: identity.sourceSha256,
|
|
49
|
+
assetTreeSha256: identity.assetTreeSha256, brokerClosureSha256: identity.brokerClosureSha256,
|
|
50
|
+
toolDefinitionsSha256: identity.toolDefinitionsSha256, contextPolicySha256: identity.contextPolicySha256,
|
|
51
|
+
sdkTreeSha256: identity.sdkTreeSha256 }));
|
|
52
|
+
}
|
|
53
|
+
function requestDigest(name: string, args: Record<string, string>) {
|
|
54
|
+
const copy = Object.create(null); for (const key of Object.keys(args)) copy[key] = args[key];
|
|
55
|
+
return sha(JSON.stringify({ name, args: copy }));
|
|
56
|
+
}
|
|
57
|
+
function canonicalJson(bytes: Buffer, label: string) {
|
|
58
|
+
const text = new TextDecoder("utf-8", { fatal: true }).decode(bytes), value = JSON.parse(text);
|
|
59
|
+
if (!Buffer.from(JSON.stringify(value)).equals(bytes)) throw new Error(`${label}-noncanonical`);
|
|
60
|
+
return value;
|
|
61
|
+
}
|
|
62
|
+
function validateManifest(bytes: Buffer, publicKeyDigest: string) {
|
|
63
|
+
const manifest = canonicalJson(bytes, "scoped-manifest");
|
|
64
|
+
const identityFields = manifest?.identity?.version === 4 ? IDENTITY_V4_FIELDS : IDENTITY_V3_FIELDS;
|
|
65
|
+
if (!exact(manifest, ["version", "identity", "profile", "root", "materials", "verifications"])
|
|
66
|
+
|| manifest.version !== 2 || !exact(manifest.identity, identityFields) || ![3, 4].includes(manifest.identity.version)
|
|
67
|
+
|| identityFields.slice(1).some(field => field.endsWith("Sha256") ? !HASH.test(manifest.identity[field]) : !ID.test(manifest.identity[field]))
|
|
68
|
+
|| manifest.identity.journalSignerSha256 !== publicKeyDigest
|
|
69
|
+
|| manifest.identity.toolDefinitionsSha256 !== scopedBrokerToolDefinitionsSha256()
|
|
70
|
+
|| !["incident", "document", "protected-env-refusal", "destructive-history-refusal"].includes(manifest.profile)
|
|
71
|
+
|| typeof manifest.root !== "string" || !pathSafeRoot(manifest.root)
|
|
72
|
+
|| !Array.isArray(manifest.materials) || manifest.materials.length > 32
|
|
73
|
+
|| !Array.isArray(manifest.verifications) || manifest.verifications.length > 32) throw new Error("scoped-manifest-invalid");
|
|
74
|
+
const ids = new Set(), paths = new Set();
|
|
75
|
+
for (const material of manifest.materials) {
|
|
76
|
+
const protectedIdentity = material?.protected === true && material.sha256 === null && material.bytes === null;
|
|
77
|
+
const contentIdentity = material?.protected === false && HASH.test(material.sha256)
|
|
78
|
+
&& Number.isSafeInteger(material.bytes) && material.bytes >= 0 && material.bytes <= 65_536;
|
|
79
|
+
if (!exact(material, ["id", "relativePath", "sha256", "bytes", "readable", "writable", "protected"])
|
|
80
|
+
|| !ID.test(material.id) || ids.has(material.id) || !pathSafeRelative(material.relativePath) || paths.has(material.relativePath)
|
|
81
|
+
|| !(protectedIdentity || contentIdentity)
|
|
82
|
+
|| [material.readable, material.writable, material.protected].some(value => typeof value !== "boolean")
|
|
83
|
+
|| material.protected && (material.readable || material.writable)) throw new Error("scoped-manifest-material-invalid");
|
|
84
|
+
ids.add(material.id); paths.add(material.relativePath);
|
|
85
|
+
}
|
|
86
|
+
for (const item of manifest.verifications) if (!exact(item, ["id", "protocol", "capabilityDigest", "receiptKeyDigest", "timeoutMs"])
|
|
87
|
+
|| !ID.test(item.id) || ![SCOPED_VERIFICATION_PROTOCOL, SCOPED_PROJECT_VERIFICATION_PROTOCOL].includes(item.protocol)
|
|
88
|
+
|| !HASH.test(item.capabilityDigest)
|
|
89
|
+
|| !HASH.test(item.receiptKeyDigest) || !Number.isSafeInteger(item.timeoutMs)
|
|
90
|
+
|| item.timeoutMs < 25 || item.timeoutMs > 30_000) throw new Error("scoped-manifest-verification-invalid");
|
|
91
|
+
return manifest;
|
|
92
|
+
}
|
|
93
|
+
function pathSafeRoot(value: unknown): value is string {
|
|
94
|
+
return typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value && !value.includes("\0");
|
|
95
|
+
}
|
|
96
|
+
function pathSafeRelative(value: unknown): value is string {
|
|
97
|
+
return typeof value === "string" && value.length > 0 && value.length <= 1_024 && !path.isAbsolute(value)
|
|
98
|
+
&& !value.includes("\\") && !value.includes("\0") && value.split("/").every(part => part && part !== "." && part !== "..");
|
|
99
|
+
}
|
|
100
|
+
function parseJournal(bytes: Buffer, key: any, manifest: any) {
|
|
101
|
+
if (!bytes.length || bytes.length > MAX_JOURNAL_BYTES || bytes.at(-1) !== 10) throw new Error("scoped-journal-bounds");
|
|
102
|
+
const source = new TextDecoder("utf-8", { fatal: true }).decode(bytes);
|
|
103
|
+
const lines = source.slice(0, -1).split("\n");
|
|
104
|
+
if (!lines.length || lines.length > MAX_ROWS) throw new Error("scoped-journal-row-bound");
|
|
105
|
+
const rows = []; let previous = "0".repeat(64), sequence = 0;
|
|
106
|
+
for (const line of lines) {
|
|
107
|
+
const row = canonicalJson(Buffer.from(line), "scoped-journal-row");
|
|
108
|
+
if (!exact(row, ["body", "signature"]) || typeof row.signature !== "string") throw new Error("scoped-journal-envelope-invalid");
|
|
109
|
+
const signature = Buffer.from(row.signature, "base64"), bodyBytes = Buffer.from(JSON.stringify(row.body));
|
|
110
|
+
if (signature.length !== 64 || !verify(null, bodyBytes, key, signature)
|
|
111
|
+
|| row.body?.version !== 1 || row.body.sequence !== ++sequence || row.body.previous !== previous
|
|
112
|
+
|| row.body.manifestSha256 !== sha(Buffer.from(JSON.stringify(manifest)))
|
|
113
|
+
|| !isDeepStrictEqual(row.body.identity, manifest.identity)) throw new Error("scoped-journal-chain-invalid");
|
|
114
|
+
rows.push(row.body); previous = sha(Buffer.from(`${line}\n`));
|
|
115
|
+
}
|
|
116
|
+
return rows;
|
|
117
|
+
}
|
|
118
|
+
function validateActions(rows: any[], status: any) {
|
|
119
|
+
if (rows[0]?.type !== "begin" || rows.at(-1)?.type !== "end" || rows.at(-1)?.cancelled !== false
|
|
120
|
+
|| rows.at(-1)?.actions !== status.actions) throw new Error("scoped-journal-terminal-invalid");
|
|
121
|
+
const actions = new Map<number, Action>();
|
|
122
|
+
for (const row of rows) {
|
|
123
|
+
if (row.type === "reservation") {
|
|
124
|
+
if (!Number.isSafeInteger(row.action) || row.action < 1 || actions.has(row.action)
|
|
125
|
+
|| typeof row.tool !== "string" && row.tool !== null || row.requestSha256 !== null && !HASH.test(row.requestSha256))
|
|
126
|
+
throw new Error("scoped-journal-reservation-invalid");
|
|
127
|
+
actions.set(row.action, { action: row.action, tool: row.tool, requestSha256: row.requestSha256, result: null });
|
|
128
|
+
} else if (row.type === "receipt" && row.accepted === true) {
|
|
129
|
+
const action = actions.get(row.action); if (!action || action.receipt) throw new Error("scoped-journal-receipt-invalid");
|
|
130
|
+
action.receipt = row.receipt;
|
|
131
|
+
} else if (row.type === "result") {
|
|
132
|
+
const action = actions.get(row.action); if (!action || action.result) throw new Error("scoped-journal-result-invalid");
|
|
133
|
+
if (!["observed", "denied", "error"].includes(row.outcome)
|
|
134
|
+
|| row.outcome === "observed" && row.effectPossible !== undefined
|
|
135
|
+
|| row.outcome !== "observed" && typeof row.effectPossible !== "boolean"
|
|
136
|
+
|| row.outcome === "denied" && row.effectPossible !== false
|
|
137
|
+
|| row.outcome === "error" && row.effectPossible !== true) throw new Error("scoped-journal-result-invalid");
|
|
138
|
+
action.result = row;
|
|
139
|
+
} else if (!["begin", "end", "verification-start", "cancel", "transport-begin", "transport-end"].includes(row.type))
|
|
140
|
+
throw new Error("scoped-journal-event-invalid");
|
|
141
|
+
}
|
|
142
|
+
const ordered = [...actions.values()].sort((a, b) => a.action - b.action);
|
|
143
|
+
if (ordered.length !== status.actions || ordered.some((item, index) => item.action !== index + 1 || !item.result))
|
|
144
|
+
throw new Error("scoped-journal-action-gap");
|
|
145
|
+
return ordered;
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
export function openScopedMediationEvidence(evidence: any, expected: { projectRoot: string; task: any; operationRef: string;
|
|
149
|
+
messageRequestId: string; plan: any; contract: any; materials: readonly any[] }): object {
|
|
150
|
+
if (!exact(evidence, ["version", "manifestBytes", "manifestSignature", "manifestPublicKey", "journalBytes",
|
|
151
|
+
"journalPublicKey", "status"])
|
|
152
|
+
|| evidence.version !== "scoped-broker-journal-evidence-v1" || !Buffer.isBuffer(evidence.manifestBytes)
|
|
153
|
+
|| !Buffer.isBuffer(evidence.manifestSignature) || evidence.manifestSignature.length !== 64
|
|
154
|
+
|| !Buffer.isBuffer(evidence.journalBytes) || !exact(evidence.status, ["actions", "blocked", "ended", "cancelled",
|
|
155
|
+
"journalSha256", "g0Qualified", "verificationAvailable", "inflightVerification"])
|
|
156
|
+
|| evidence.status.blocked !== false || evidence.status.ended !== true || evidence.status.cancelled !== false
|
|
157
|
+
|| evidence.status.g0Qualified !== false || typeof evidence.status.verificationAvailable !== "boolean"
|
|
158
|
+
|| evidence.status.inflightVerification !== null || !Number.isSafeInteger(evidence.status.actions)
|
|
159
|
+
|| evidence.status.actions < 0 || evidence.status.actions > 256
|
|
160
|
+
|| evidence.status.journalSha256 !== sha(evidence.journalBytes)) throw new Error("scoped-evidence-invalid");
|
|
161
|
+
const authority = createPublicKey(evidence.manifestPublicKey), key = createPublicKey(evidence.journalPublicKey);
|
|
162
|
+
if (authority.asymmetricKeyType !== "ed25519" || !verify(null, evidence.manifestBytes, authority, evidence.manifestSignature))
|
|
163
|
+
throw new Error("scoped-manifest-signature-invalid");
|
|
164
|
+
if (key.asymmetricKeyType !== "ed25519") throw new Error("scoped-journal-key-invalid");
|
|
165
|
+
const authorityDigest = sha(authority.export({ type: "spki", format: "der" }));
|
|
166
|
+
const keyDigest = sha(key.export({ type: "spki", format: "der" })), manifest = validateManifest(evidence.manifestBytes, keyDigest);
|
|
167
|
+
if (manifest.identity.manifestAuthoritySha256 !== authorityDigest || authorityDigest === keyDigest)
|
|
168
|
+
throw new Error("scoped-manifest-authority-mismatch");
|
|
169
|
+
if (evidence.status.verificationAvailable !== (manifest.verifications.length > 0))
|
|
170
|
+
throw new Error("scoped-verification-status-mismatch");
|
|
171
|
+
if (manifest.identity.taskId !== expected.task.taskId || manifest.identity.sessionId !== expected.task.sessionId
|
|
172
|
+
|| manifest.identity.requestId !== expected.task.operatorRequestDigest || manifest.identity.operationId !== expected.operationRef
|
|
173
|
+
|| expected.messageRequestId !== manifest.identity.nonce
|
|
174
|
+
|| expected.plan.identity.configDigest !== (manifest.identity.version === 4
|
|
175
|
+
? manifest.identity.measurementConfigurationSha256 : manifest.identity.configSha256)
|
|
176
|
+
|| expected.plan.identity.armDigest !== scopedBrokerArmDigest(manifest.identity))
|
|
177
|
+
throw new Error("scoped-evidence-binding-mismatch");
|
|
178
|
+
if (!pathSafeRoot(expected.projectRoot) || fs.realpathSync.native(manifest.root) !== fs.realpathSync.native(expected.projectRoot))
|
|
179
|
+
throw new Error("scoped-material-root-mismatch");
|
|
180
|
+
const rows = parseJournal(evidence.journalBytes, key, manifest), actions = validateActions(rows, evidence.status);
|
|
181
|
+
const planBindings = new Map(expected.plan.materialBindings.map((item: any) => [item.id, item]));
|
|
182
|
+
const contractFacts = expected.contract.facts, contextIds = new Set(contractFacts.filter((fact: any) => fact.kind === "context-current")
|
|
183
|
+
.flatMap((fact: any) => fact.parameters.requiredMaterialIds));
|
|
184
|
+
const writableIds = new Set(contractFacts.filter((fact: any) => fact.kind === "workspace-scope")
|
|
185
|
+
.flatMap((fact: any) => fact.parameters.allowedWriteMaterialIds));
|
|
186
|
+
const policy = contractFacts.find((fact: any) => fact.kind === "tool-policy-complete");
|
|
187
|
+
if (!policy || policy.parameters.profileDigest !== scopedBrokerProfileDigest(manifest.profile)) throw new Error("scoped-profile-binding-mismatch");
|
|
188
|
+
if (manifest.materials.length !== expected.plan.materialBindings.length
|
|
189
|
+
|| expected.materials.length !== expected.plan.materialBindings.length) throw new Error("scoped-material-set-mismatch");
|
|
190
|
+
const currentMaterials = new Map(expected.materials.map((item: any) => [item.id, item]));
|
|
191
|
+
for (const material of manifest.materials) {
|
|
192
|
+
const binding: any = planBindings.get(material.id);
|
|
193
|
+
if (!binding || binding.relativePath !== material.relativePath || material.readable !== contextIds.has(material.id)
|
|
194
|
+
|| material.writable !== writableIds.has(material.id) || material.protected !== (binding.mode === "protected"))
|
|
195
|
+
throw new Error("scoped-material-policy-mismatch");
|
|
196
|
+
const current: any = currentMaterials.get(material.id);
|
|
197
|
+
if (!current || current.relativePath !== material.relativePath || current.mode !== binding.mode)
|
|
198
|
+
throw new Error("scoped-current-material-mismatch");
|
|
199
|
+
if (binding.mode === "frozen" && (binding.sha256 !== material.sha256 || current.sha256 !== material.sha256))
|
|
200
|
+
throw new Error("scoped-frozen-material-mismatch");
|
|
201
|
+
if (binding.mode !== "protected" && sha(readTaskBaselineSource(expected.projectRoot, expected.task, binding.relativePath)) !== material.sha256)
|
|
202
|
+
throw new Error("scoped-baseline-material-mismatch");
|
|
203
|
+
if (binding.mode !== "protected" && !material.writable && current.sha256 !== material.sha256)
|
|
204
|
+
throw new Error("scoped-readonly-material-drift");
|
|
205
|
+
}
|
|
206
|
+
const verifierFacts = contractFacts.filter((fact: any) => fact.kind === "project-verifier-current");
|
|
207
|
+
if (manifest.verifications.length !== verifierFacts.length) throw new Error("scoped-verifier-policy-mismatch");
|
|
208
|
+
const materialMap = currentMaterials;
|
|
209
|
+
const reads = new Map<string, Action[]>(), writes = new Map<string, Action[]>(), verifications = new Map<string, Action[]>();
|
|
210
|
+
const observedDigests = new Map(manifest.materials.map((item: any) => [item.id, item.sha256]));
|
|
211
|
+
const add = (map: Map<string, Action[]>, id: string, action: Action) => map.set(id, [...(map.get(id) ?? []), action]);
|
|
212
|
+
for (const action of actions) {
|
|
213
|
+
if (action.result.outcome !== "observed") continue;
|
|
214
|
+
const read = manifest.materials.filter((item: any) => action.tool === "scoped_read"
|
|
215
|
+
&& action.requestSha256 === requestDigest("scoped_read", { materialId: item.id }));
|
|
216
|
+
const write = manifest.materials.filter((item: any) => { const current: any = materialMap.get(item.id);
|
|
217
|
+
return action.tool === "scoped_write_document" && current?.text !== undefined && action.requestSha256 === requestDigest("scoped_write_document",
|
|
218
|
+
{ materialId: item.id, expectedSha256: item.sha256, utf8: current.text }); });
|
|
219
|
+
const checks = manifest.verifications.filter((item: any) => action.tool === "scoped_verify"
|
|
220
|
+
&& action.requestSha256 === requestDigest("scoped_verify", { verificationId: item.id }));
|
|
221
|
+
if (read.length + write.length + checks.length !== 1) throw new Error("scoped-observed-action-unbound");
|
|
222
|
+
if (read[0]) { if (action.result.materialSha256 !== observedDigests.get(read[0].id)) throw new Error("scoped-read-result-mismatch"); add(reads, read[0].id, action); }
|
|
223
|
+
if (write[0]) { if (action.result.materialSha256 !== materialMap.get(write[0].id)?.sha256) throw new Error("scoped-write-result-mismatch"); add(writes, write[0].id, action); observedDigests.set(write[0].id, action.result.materialSha256); }
|
|
224
|
+
if (checks[0]) add(verifications, checks[0].id, action);
|
|
225
|
+
}
|
|
226
|
+
const snapshot: Snapshot = { manifest, rows, actions, profileDigest: policy.parameters.profileDigest,
|
|
227
|
+
armDigest: scopedBrokerArmDigest(manifest.identity), reads, writes, verifications };
|
|
228
|
+
const capability = Object.freeze({ version: "scoped-mediation-capability-v1", journalSha256: sha(evidence.journalBytes),
|
|
229
|
+
profileDigest: snapshot.profileDigest }); snapshots.set(capability, snapshot); return capability;
|
|
230
|
+
}
|
|
231
|
+
|
|
232
|
+
// A receipt covers only its signed command policy. Dynamic document bytes in
|
|
233
|
+
// v3 must still match every writable material and the current workspace.
|
|
234
|
+
export function scopedProjectReceiptCoversTask(receipt: any, input: FactInput): boolean {
|
|
235
|
+
const commands = input.task?.verifyCommands;
|
|
236
|
+
if (!Array.isArray(commands) || commands.length === 0 || commands.some(command => typeof command !== "string")) return false;
|
|
237
|
+
if (receipt?.kind !== "scoped-project-verification-receipt-v3")
|
|
238
|
+
return commands.every(command => SCOPED_PROJECT_COMMANDS.has(command.trim()));
|
|
239
|
+
const scope = receipt.scope;
|
|
240
|
+
if (receipt.version !== 3 || receipt.worker?.version !== "scoped-configured-docs-worker-v1"
|
|
241
|
+
|| !scope || scope.kind !== "configured-docs-current-v1" || !HASH.test(scope.frozenSourceDigest)
|
|
242
|
+
|| !isDeepStrictEqual(commands.map(command => command.trim()), scope.commands)
|
|
243
|
+
|| !Array.isArray(scope.files) || scope.files.length < 1 || scope.files.length > 8) return false;
|
|
244
|
+
const paths = new Map(input.plan.materialBindings.map((item: any) => [item.id, item.relativePath]));
|
|
245
|
+
const expectedPaths = [...new Set(input.contract.facts.filter((fact: any) => fact.kind === "workspace-scope")
|
|
246
|
+
.flatMap((fact: any) => fact.parameters.allowedWriteMaterialIds.map((id: string) => paths.get(id))))].sort();
|
|
247
|
+
if (!isDeepStrictEqual(expectedPaths, scope.files.map((item: any) => item.relativePath))) return false;
|
|
248
|
+
if (!input.projectRoot || workingTreeSnapshotHasUnavailableEvidence(input.workspace)) return false;
|
|
249
|
+
try {
|
|
250
|
+
if (!isDeepStrictEqual(workingTreeSnapshot(input.projectRoot), input.workspace)) return false;
|
|
251
|
+
for (const file of scope.files) {
|
|
252
|
+
const material = input.materials.find((item: any) => item.relativePath === file.relativePath);
|
|
253
|
+
if (!HASH.test(file.sha256) || !material || material.sha256 !== file.sha256
|
|
254
|
+
|| sha(readWorkspaceFile(input.projectRoot, file.relativePath, 65536)) !== file.sha256
|
|
255
|
+
|| typeof material.text !== "string" || Buffer.byteLength(material.text) !== file.byteLength) return false;
|
|
256
|
+
}
|
|
257
|
+
if (!isDeepStrictEqual(workingTreeSnapshot(input.projectRoot), input.workspace)) return false;
|
|
258
|
+
} catch { return false; }
|
|
259
|
+
return receipt.sourceDigest === sha(JSON.stringify({ version: 1, frozenSourceDigest: scope.frozenSourceDigest, files: scope.files }));
|
|
260
|
+
}
|
|
261
|
+
|
|
262
|
+
function disposition(status: "pass" | "fail" | "unknown" | "error", fact: any, details: any,
|
|
263
|
+
reasonCodes: string[], response?: any) {
|
|
264
|
+
const observationDigest = sha(JSON.stringify({ factId: fact.id, details }));
|
|
265
|
+
return { status, observationDigest, counterexampleRef: status === "fail" ? sha(JSON.stringify(details)) : null,
|
|
266
|
+
reasonCodes, ...(response ? { response } : {}) };
|
|
267
|
+
}
|
|
268
|
+
export function scopedMediationFactObservation(capability: object, input: FactInput) {
|
|
269
|
+
const value = snapshots.get(capability); if (!value) throw new Error("untrusted-scoped-mediation-capability");
|
|
270
|
+
const fact = input.fact, response = input.response;
|
|
271
|
+
if (fact.kind === "tool-policy-complete") {
|
|
272
|
+
const unsafe = value.actions.filter(action => action.result.outcome === "error" || action.result.effectPossible === true);
|
|
273
|
+
return unsafe.length ? disposition("error", fact, unsafe.map(item => item.action), ["capture-error"], response)
|
|
274
|
+
: disposition("pass", fact, { profileDigest: value.profileDigest, actions: value.actions.length }, [], response);
|
|
275
|
+
}
|
|
276
|
+
if (fact.kind === "context-current") {
|
|
277
|
+
const missing = fact.parameters.requiredMaterialIds.filter((id: string) => !value.reads.has(id));
|
|
278
|
+
return missing.length ? disposition("unknown", fact, { missing }, ["incomplete-mediation"], response)
|
|
279
|
+
: disposition("pass", fact, { reads: fact.parameters.requiredMaterialIds }, [], response);
|
|
280
|
+
}
|
|
281
|
+
if (fact.kind === "workspace-scope") {
|
|
282
|
+
const allowed = new Set<string>(fact.parameters.allowedWriteMaterialIds), observed = [...value.writes.keys()];
|
|
283
|
+
const bindings = new Map<string, string>(input.plan.materialBindings.map((item: any) => [item.id, item.relativePath]));
|
|
284
|
+
const allowedPaths = new Set<string | undefined>([...allowed].map(id => bindings.get(id)));
|
|
285
|
+
const baseline: Record<string, string> = input.task.baselineFileDigests ?? {};
|
|
286
|
+
const changed = [...new Set([...Object.keys(baseline), ...Object.keys(input.workspace)])]
|
|
287
|
+
.filter(file => baseline[file] !== input.workspace[file]);
|
|
288
|
+
const violation = observed.some(id => !allowed.has(id)) || changed.some(file => !allowedPaths.has(file));
|
|
289
|
+
if (violation) return disposition("fail", fact, { observed, changed, allowed: [...allowed] }, ["scope-violation"], response);
|
|
290
|
+
// Permission to write does not require a mutation when review found no change.
|
|
291
|
+
// Every actual change still needs an observed write for the same bound path.
|
|
292
|
+
const observedPaths = new Set(observed.map(id => bindings.get(id)));
|
|
293
|
+
const missing = changed.filter(file => !observedPaths.has(file));
|
|
294
|
+
return missing.length ? disposition("unknown", fact, { missing }, ["incomplete-mediation"], response)
|
|
295
|
+
: disposition("pass", fact, { observed, changed }, [], response);
|
|
296
|
+
}
|
|
297
|
+
if (fact.kind === "project-verifier-current") {
|
|
298
|
+
const matches = [...value.verifications.values()].flat().filter(action => action.receipt?.planDigest === fact.parameters.commandSetDigest
|
|
299
|
+
&& action.receipt?.protocol === SCOPED_PROJECT_VERIFICATION_PROTOCOL
|
|
300
|
+
&& action.receipt?.status === "completed" && action.receipt?.verdict === "observation-recorded"
|
|
301
|
+
&& action.receipt?.cleanup?.confirmed === true && action.result.verificationRunId === action.receipt.attemptId);
|
|
302
|
+
if (matches.length !== 1) return disposition("unknown", fact, { matches: matches.length },
|
|
303
|
+
["incomplete-mediation"], response);
|
|
304
|
+
const receipt = matches[0].receipt;
|
|
305
|
+
if (!scopedProjectReceiptCoversTask(receipt, { ...input, projectRoot: value.manifest.root })) return disposition("unknown", fact,
|
|
306
|
+
{ configuredCommandsOrCurrentFilesCovered: false }, ["incomplete-mediation"], response);
|
|
307
|
+
const details = { verificationRunId: receipt.attemptId,
|
|
308
|
+
commandSetDigest: receipt.planDigest, outcome: receipt.evidence?.outcome,
|
|
309
|
+
failedCommands: receipt.evidence?.failedCommands };
|
|
310
|
+
if (receipt.evidence?.outcome === "failed" && Array.isArray(receipt.evidence.failedCommands)
|
|
311
|
+
&& receipt.evidence.failedCommands.length > 0) return disposition("fail", fact, details,
|
|
312
|
+
["verification-failed"], response);
|
|
313
|
+
return receipt.evidence?.outcome === "passed" && Array.isArray(receipt.evidence.failedCommands)
|
|
314
|
+
&& receipt.evidence.failedCommands.length === 0 ? disposition("pass", fact, details, [], response)
|
|
315
|
+
: disposition("error", fact, details, ["capture-error"], response);
|
|
316
|
+
}
|
|
317
|
+
throw new Error("unsupported-scoped-mediation-fact");
|
|
318
|
+
}
|