@piagent/platform 1.6.1 → 1.8.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +71 -0
- package/README.md +8 -2
- package/README.vi.md +11 -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 +3 -3
- 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 +2 -2
- 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 +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/model-options.md +20 -15
- package/docs/operator-manual-vietnamese.md +12 -12
- 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 +3 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +38 -11
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +23 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +23 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- 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 +2 -2
- 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 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- 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 +3 -3
- 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 -1
- package/docs-site/workflows.html +1 -1
- 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-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/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/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -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/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 +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- 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.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- 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 +2 -1
- 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 +107 -107
- 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 +63 -50
- 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 +69 -8
- 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 +162 -45
- 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 +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- 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-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 +144 -128
- 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 +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- 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 +395 -33
- 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 +3 -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 +264 -70
- 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 +139 -152
- 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 +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- 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-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- 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 +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- 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 +1 -0
- 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 +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -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 +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- 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/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- 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/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/model-route-policy.ts +33 -13
- package/packages/piagent-core/runtime/model/model-route-types.ts +10 -6
- 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/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 +12 -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/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 +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- 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/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- 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 +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- 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 +11 -11
- 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 +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- 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-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DMU7Xe_O.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-Cui9AAV_.js +176 -0
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- 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 +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- 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 +141 -19
- 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 +14 -6
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- 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 +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- 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 +65 -12
- 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/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 +32 -1
- 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/transcript-v1.schema.json +2 -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 +30 -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 +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -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-model-scope.sh +6 -2
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/install-global.sh +2 -2
- package/scripts/model-route-evaluation-core.ts +1 -0
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-model-route.mjs +1 -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/setup.sh +2 -2
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +37 -56
- package/scripts/verify-release-identity.mjs +5 -3
- package/templates/global/settings.json +5 -1
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -0,0 +1,461 @@
|
|
|
1
|
+
import { isScopedDocsPolicy, validateScopedDocsPolicy, scopedDocsPlanBinding, prepareScopedDocsAttempt,
|
|
2
|
+
runScopedDocsVerification, SCOPED_DOCS_RECEIPT, SCOPED_DOCS_WORKER, DOCS_COMMAND_IDS } from "./benchmark-scoped-docs-verifier.mjs";
|
|
3
|
+
import { validateProjectReceipt } from "./benchmark-scoped-project-verification-receipt.mjs";
|
|
4
|
+
import { spawn } from "node:child_process";
|
|
5
|
+
import { createHash, createPrivateKey, createPublicKey, randomUUID, sign } from "node:crypto";
|
|
6
|
+
import fs from "node:fs";
|
|
7
|
+
import path from "node:path";
|
|
8
|
+
import { fileURLToPath } from "node:url";
|
|
9
|
+
import { types } from "node:util";
|
|
10
|
+
|
|
11
|
+
export const SCOPED_PROJECT_VERIFICATION_PROTOCOL = "scoped-project-verifier-v1";
|
|
12
|
+
export const SCOPED_PROJECT_VERIFICATION_RECEIPT = "scoped-project-verification-receipt-v2";
|
|
13
|
+
export const SCOPED_PROJECT_VERIFIER_POLICY = "scoped-project-verifier-policy-v1";
|
|
14
|
+
export const SCOPED_PROJECT_VERIFIER_WORKER = "scoped-project-verifier-worker-v1";
|
|
15
|
+
|
|
16
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
17
|
+
const ID = /^[A-Za-z0-9_.:-]{1,160}$/;
|
|
18
|
+
const UUID_V4 = /^[a-f0-9]{8}-[a-f0-9]{4}-4[a-f0-9]{3}-[89ab][a-f0-9]{3}-[a-f0-9]{12}$/;
|
|
19
|
+
const SAFE_PATH = /^(?!\/)(?!.*\/\/)(?!.*\/$)(?!.*(?:^|\/)(?:\.{1,2}|\.git|\.pi|node_modules)(?:\/|$))[^\\\0-\x1f\x7f:%?#]+$/;
|
|
20
|
+
const POLICY_FIELDS = ["version", "configuredScripts", "configurationFiles", "syntaxFiles", "testFiles"];
|
|
21
|
+
const EXPECTED_SCRIPTS = Object.freeze({ "type-check": "node scripts/check.mjs", lint: "node scripts/check.mjs",
|
|
22
|
+
test: "node --test test/*.test.js", "test:e2e": "node --test test/*.test.js" });
|
|
23
|
+
// Policy v1 expands this reviewed dependency-free checker into fixed --check
|
|
24
|
+
// invocations. Other script bodies are not equivalent merely because the npm
|
|
25
|
+
// script name stayed unchanged. Keep them outside this fixed policy.
|
|
26
|
+
const FIXED_SYNTAX_CHECKER_SHA256 = "9af8698a75a04abce876f881181d6d9688220b1d0d71f08be98919a033cacdb5";
|
|
27
|
+
const MAX_FILE_BYTES = 64 * 1024, MAX_FOOTPRINT_BYTES = 2 * 1024 * 1024, MAX_OUTPUT_BYTES = 256 * 1024;
|
|
28
|
+
const REAP_GRACE_MS = 250;
|
|
29
|
+
const sha = value => createHash("sha256").update(value).digest("hex");
|
|
30
|
+
const fail = code => { throw Object.assign(new Error(code), { supervisorCode: code }); };
|
|
31
|
+
const requireThat = (value, code) => { if (!value) fail(code); };
|
|
32
|
+
const plain = value => Boolean(value && typeof value === "object" && !types.isProxy(value)
|
|
33
|
+
&& [null, Object.prototype].includes(Object.getPrototypeOf(value)));
|
|
34
|
+
|
|
35
|
+
function exact(value, names, code) {
|
|
36
|
+
requireThat(plain(value), code);
|
|
37
|
+
const descriptors = Object.getOwnPropertyDescriptors(value);
|
|
38
|
+
requireThat(Reflect.ownKeys(descriptors).length === names.length && names.every(name =>
|
|
39
|
+
descriptors[name]?.enumerable && Object.hasOwn(descriptors[name], "value")), code);
|
|
40
|
+
}
|
|
41
|
+
function deepFreeze(value) {
|
|
42
|
+
if (!value || typeof value !== "object" || Object.isFrozen(value)) return value;
|
|
43
|
+
for (const child of Object.values(value)) deepFreeze(child);
|
|
44
|
+
return Object.freeze(value);
|
|
45
|
+
}
|
|
46
|
+
const validHash = value => typeof value === "string" && HASH.test(value);
|
|
47
|
+
const validTime = value => Number.isSafeInteger(value) && value >= 0;
|
|
48
|
+
function canonicalDirectory(value, code) {
|
|
49
|
+
requireThat(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
50
|
+
&& !value.includes("\0"), code);
|
|
51
|
+
let real, info;
|
|
52
|
+
try { real = fs.realpathSync(value); info = fs.lstatSync(real); } catch { fail(code); }
|
|
53
|
+
requireThat(real === value && info.isDirectory() && !info.isSymbolicLink(), code);
|
|
54
|
+
return real;
|
|
55
|
+
}
|
|
56
|
+
function outside(left, right) {
|
|
57
|
+
const relative = path.relative(left, right);
|
|
58
|
+
return relative === ".." || relative.startsWith(`..${path.sep}`) || path.isAbsolute(relative);
|
|
59
|
+
}
|
|
60
|
+
function stableFile(file, maximum, code) {
|
|
61
|
+
let before;
|
|
62
|
+
try { before = fs.lstatSync(file, { bigint: true }); } catch { fail(code); }
|
|
63
|
+
requireThat(before.isFile() && !before.isSymbolicLink() && before.nlink === 1n
|
|
64
|
+
&& before.size >= 0n && before.size <= BigInt(maximum), code);
|
|
65
|
+
const fd = fs.openSync(file, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0));
|
|
66
|
+
try {
|
|
67
|
+
const descriptor = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd),
|
|
68
|
+
after = fs.fstatSync(fd, { bigint: true }), current = fs.lstatSync(file, { bigint: true }),
|
|
69
|
+
fields = ["dev", "ino", "mode", "nlink", "size", "mtimeNs", "ctimeNs"];
|
|
70
|
+
requireThat(bytes.length === Number(before.size) && fields.every(name => before[name] === descriptor[name]
|
|
71
|
+
&& before[name] === after[name] && before[name] === current[name]), code);
|
|
72
|
+
return Buffer.from(bytes);
|
|
73
|
+
} finally { fs.closeSync(fd); }
|
|
74
|
+
}
|
|
75
|
+
function validatePaths(value, minimum, maximum, code) {
|
|
76
|
+
requireThat(Array.isArray(value) && value.length >= minimum && value.length <= maximum
|
|
77
|
+
&& new Set(value).size === value.length && value.every(item => typeof item === "string"
|
|
78
|
+
&& item.length <= 1024 && SAFE_PATH.test(item))
|
|
79
|
+
&& value.every((item, index) => index === 0 || value[index - 1] < item), code);
|
|
80
|
+
return value;
|
|
81
|
+
}
|
|
82
|
+
function validatePolicy(value) {
|
|
83
|
+
if (isScopedDocsPolicy(value)) return validateScopedDocsPolicy(value, docsHelpers());
|
|
84
|
+
exact(value, POLICY_FIELDS, "invalid-project-verifier-policy");
|
|
85
|
+
exact(value.configuredScripts, ["type-check", "lint", "test", "test:e2e"], "invalid-project-verifier-policy");
|
|
86
|
+
requireThat(value.version === SCOPED_PROJECT_VERIFIER_POLICY
|
|
87
|
+
&& JSON.stringify(value.configuredScripts) === JSON.stringify(EXPECTED_SCRIPTS),
|
|
88
|
+
"invalid-project-verifier-policy");
|
|
89
|
+
validatePaths(value.configurationFiles, 2, 8, "invalid-project-verifier-policy");
|
|
90
|
+
validatePaths(value.syntaxFiles, 1, 64, "invalid-project-verifier-policy");
|
|
91
|
+
validatePaths(value.testFiles, 1, 16, "invalid-project-verifier-policy");
|
|
92
|
+
requireThat(JSON.stringify(value.configurationFiles) === JSON.stringify(["package.json", "scripts/check.mjs"])
|
|
93
|
+
&& value.syntaxFiles.every(item => item.endsWith(".js"))
|
|
94
|
+
&& value.testFiles.every(item => item.startsWith("test/") && item.endsWith(".test.js")
|
|
95
|
+
&& value.syntaxFiles.includes(item)), "invalid-project-verifier-policy");
|
|
96
|
+
return deepFreeze(structuredClone(value));
|
|
97
|
+
}
|
|
98
|
+
function declaredRootFiles(root, directory, suffix) {
|
|
99
|
+
const base = path.join(root, directory), result = [];
|
|
100
|
+
function walk(current) {
|
|
101
|
+
let entries;
|
|
102
|
+
try { entries = fs.readdirSync(current, { withFileTypes: true }); } catch { fail("project-footprint-unavailable"); }
|
|
103
|
+
for (const entry of entries.sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0)) {
|
|
104
|
+
const file = path.join(current, entry.name);
|
|
105
|
+
requireThat(!entry.isSymbolicLink(), "project-footprint-node");
|
|
106
|
+
if (entry.isDirectory()) walk(file);
|
|
107
|
+
else {
|
|
108
|
+
requireThat(entry.isFile(), "project-footprint-node");
|
|
109
|
+
if (file.endsWith(suffix)) result.push(path.relative(root, file).split(path.sep).join("/"));
|
|
110
|
+
else fail("project-footprint-undeclared");
|
|
111
|
+
}
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
walk(base); return result;
|
|
115
|
+
}
|
|
116
|
+
function captureFootprint(projectRoot, policy) {
|
|
117
|
+
const declared = [...new Set([...policy.configurationFiles, ...policy.syntaxFiles])].sort(), actual = [
|
|
118
|
+
...declaredRootFiles(projectRoot, "src", ".js"),
|
|
119
|
+
...declaredRootFiles(projectRoot, "scripts", ".mjs"),
|
|
120
|
+
...declaredRootFiles(projectRoot, "test", ".js"), "package.json"
|
|
121
|
+
].sort();
|
|
122
|
+
requireThat(JSON.stringify(actual) === JSON.stringify(declared), "project-footprint-mismatch");
|
|
123
|
+
let total = 0;
|
|
124
|
+
const files = declared.map(relativePath => {
|
|
125
|
+
const file = path.join(projectRoot, ...relativePath.split("/"));
|
|
126
|
+
requireThat(fs.realpathSync(file) === file && file.startsWith(projectRoot + path.sep), "project-footprint-path");
|
|
127
|
+
const bytes = stableFile(file, MAX_FILE_BYTES, "project-footprint-drift"); total += bytes.length;
|
|
128
|
+
requireThat(total <= MAX_FOOTPRINT_BYTES, "project-footprint-size");
|
|
129
|
+
return { relativePath, bytes, byteLength: bytes.length, sha256: sha(bytes) };
|
|
130
|
+
});
|
|
131
|
+
let packageValue;
|
|
132
|
+
try { packageValue = JSON.parse(files.find(item => item.relativePath === "package.json").bytes); }
|
|
133
|
+
catch { fail("project-package-json"); }
|
|
134
|
+
requireThat(plain(packageValue?.scripts)
|
|
135
|
+
&& JSON.stringify(packageValue.scripts) === JSON.stringify(policy.configuredScripts), "project-configured-scripts");
|
|
136
|
+
requireThat(files.find(item => item.relativePath === "scripts/check.mjs")?.sha256 === FIXED_SYNTAX_CHECKER_SHA256,
|
|
137
|
+
"project-configured-checker-unsupported");
|
|
138
|
+
const identity = files.map(({ relativePath, byteLength, sha256 }) => ({ relativePath, byteLength, sha256 }));
|
|
139
|
+
return { files, sourceDigest: sha(JSON.stringify({ version: 1, files: identity })) };
|
|
140
|
+
}
|
|
141
|
+
function stableNodeIdentity(nodeCommand) {
|
|
142
|
+
requireThat(typeof nodeCommand === "string" && path.isAbsolute(nodeCommand) && path.normalize(nodeCommand) === nodeCommand
|
|
143
|
+
&& fs.realpathSync(nodeCommand) === nodeCommand, "invalid-project-verifier-node");
|
|
144
|
+
const nodeSha256 = sha(stableFile(nodeCommand, 1024 * 1024 * 1024, "project-verifier-node-drift"));
|
|
145
|
+
return { nodeSha256, environmentDigest: sha(JSON.stringify({ version: 1, nodeSha256 })) };
|
|
146
|
+
}
|
|
147
|
+
|
|
148
|
+
function docsHelpers() { return { exact, validatePaths, validatePolicy, deepFreeze, stableFile, captureFootprint, scopedProjectVerifierDigest }; }
|
|
149
|
+
export function scopedProjectVerifierDigest() {
|
|
150
|
+
return sha(JSON.stringify([import.meta.url, new URL("./benchmark-scoped-docs-verifier.mjs", import.meta.url).href,
|
|
151
|
+
new URL("./benchmark-scoped-project-verification-receipt.mjs", import.meta.url).href]
|
|
152
|
+
.map(url => ({ name: path.basename(fileURLToPath(url)), sha256: sha(stableFile(fileURLToPath(url),
|
|
153
|
+
1024 * 1024, "project-verifier-source-drift")) }))));
|
|
154
|
+
}
|
|
155
|
+
|
|
156
|
+
export function scopedProjectVerificationPlanBinding({ projectRoot, nodeCommand, policy, timeoutMs } = {}) {
|
|
157
|
+
const root = canonicalDirectory(projectRoot, "invalid-project-verifier-root"), approved = validatePolicy(policy);
|
|
158
|
+
requireThat(Number.isSafeInteger(timeoutMs) && timeoutMs >= 25 && timeoutMs <= 30000,
|
|
159
|
+
"invalid-project-verifier-plan");
|
|
160
|
+
if (isScopedDocsPolicy(approved)) return scopedDocsPlanBinding({ projectRoot: root, nodeCommand, policy: approved,
|
|
161
|
+
timeoutMs, protocol: SCOPED_PROJECT_VERIFICATION_PROTOCOL, verifierDigest: scopedProjectVerifierDigest() }, docsHelpers());
|
|
162
|
+
const footprint = captureFootprint(root, approved), node = stableNodeIdentity(nodeCommand),
|
|
163
|
+
requestDigest = sha(JSON.stringify(approved)), verifierDigest = scopedProjectVerifierDigest();
|
|
164
|
+
const planDigest = sha(JSON.stringify({ version: 1, protocol: SCOPED_PROJECT_VERIFICATION_PROTOCOL,
|
|
165
|
+
requestDigest, sourceDigest: footprint.sourceDigest, environmentDigest: node.environmentDigest,
|
|
166
|
+
verifierDigest, timeoutMs }));
|
|
167
|
+
const capabilityDigest = sha(JSON.stringify({ version: 2, protocol: SCOPED_PROJECT_VERIFICATION_PROTOCOL,
|
|
168
|
+
planDigest, requestDigest, sourceDigest: footprint.sourceDigest, environmentDigest: node.environmentDigest,
|
|
169
|
+
verifierDigest, timeoutMs }));
|
|
170
|
+
return deepFreeze({ protocol: SCOPED_PROJECT_VERIFICATION_PROTOCOL, planDigest, requestDigest,
|
|
171
|
+
sourceDigest: footprint.sourceDigest, environmentDigest: node.environmentDigest, verifierDigest, timeoutMs,
|
|
172
|
+
capabilityDigest });
|
|
173
|
+
}
|
|
174
|
+
|
|
175
|
+
function killProcess(child, signal) {
|
|
176
|
+
if (!child?.pid) return;
|
|
177
|
+
try { process.kill(process.platform === "win32" ? child.pid : -child.pid, signal); } catch {}
|
|
178
|
+
}
|
|
179
|
+
function runNode(nodeCommand, args, cwd, signal) {
|
|
180
|
+
return new Promise(resolve => {
|
|
181
|
+
let child;
|
|
182
|
+
try { child = spawn(nodeCommand, args, { cwd, detached: process.platform !== "win32",
|
|
183
|
+
stdio: ["ignore", "pipe", "pipe"], env: { LANG: "C", LC_ALL: "C", TZ: "UTC", NODE_NO_WARNINGS: "1" } }); }
|
|
184
|
+
catch { resolve({ status: "error", reason: "process-spawn-failed" }); return; }
|
|
185
|
+
let outputBytes = 0, overflow = false, spawnFailed = false;
|
|
186
|
+
const chunks = [], onData = chunk => {
|
|
187
|
+
outputBytes += chunk.length;
|
|
188
|
+
if (outputBytes > MAX_OUTPUT_BYTES) { overflow = true; killProcess(child, "SIGKILL"); return; }
|
|
189
|
+
chunks.push(Buffer.from(chunk));
|
|
190
|
+
};
|
|
191
|
+
child.stdout.on("data", onData); child.stderr.on("data", onData);
|
|
192
|
+
let aborted = signal.aborted, reapTimer;
|
|
193
|
+
const abort = () => { aborted = true; killProcess(child, "SIGTERM");
|
|
194
|
+
reapTimer = setTimeout(() => killProcess(child, "SIGKILL"), REAP_GRACE_MS); };
|
|
195
|
+
signal.addEventListener("abort", abort, { once: true }); if (aborted) abort();
|
|
196
|
+
child.once("error", () => { spawnFailed = true; });
|
|
197
|
+
child.once("close", (code, childSignal) => {
|
|
198
|
+
clearTimeout(reapTimer); signal.removeEventListener("abort", abort);
|
|
199
|
+
const output = Buffer.concat(chunks);
|
|
200
|
+
resolve(aborted ? { status: "cancelled", reason: "project-verifier-cancelled" }
|
|
201
|
+
: spawnFailed ? { status: "error", reason: "process-spawn-failed" }
|
|
202
|
+
: overflow ? { status: "error", reason: "project-verifier-output-limit" }
|
|
203
|
+
: { status: "completed", code, signal: childSignal, outputSha256: sha(output), outputBytes: output.length });
|
|
204
|
+
});
|
|
205
|
+
});
|
|
206
|
+
}
|
|
207
|
+
async function runGroup(id, invocations, nodeCommand, stage, signal) {
|
|
208
|
+
const results = []; let outputBytes = 0;
|
|
209
|
+
for (const args of invocations) {
|
|
210
|
+
const result = await runNode(nodeCommand, args, stage, signal);
|
|
211
|
+
if (result.status !== "completed") return result;
|
|
212
|
+
results.push(result); outputBytes += result.outputBytes;
|
|
213
|
+
if (outputBytes > MAX_OUTPUT_BYTES) return { status: "error", reason: "project-verifier-output-limit" };
|
|
214
|
+
}
|
|
215
|
+
const passed = results.every(item => item.code === 0 && item.signal === null);
|
|
216
|
+
return { status: "completed", summary: { id, invocationCount: results.length,
|
|
217
|
+
outcome: passed ? "passed" : "failed", outputSha256: sha(JSON.stringify(results)),
|
|
218
|
+
outputBytes } };
|
|
219
|
+
}
|
|
220
|
+
function materializeStage(stage, footprint) {
|
|
221
|
+
for (const item of footprint.files) {
|
|
222
|
+
const file = path.join(stage, ...item.relativePath.split("/"));
|
|
223
|
+
fs.mkdirSync(path.dirname(file), { recursive: true, mode: 0o700 });
|
|
224
|
+
const fd = fs.openSync(file, fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL
|
|
225
|
+
| (fs.constants.O_NOFOLLOW ?? 0), 0o400);
|
|
226
|
+
try { fs.writeFileSync(fd, item.bytes); fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
|
|
227
|
+
}
|
|
228
|
+
}
|
|
229
|
+
async function runPinnedProjectVerification(input) {
|
|
230
|
+
if (isScopedDocsPolicy(input.policy)) return runScopedDocsVerification(input, docsHelpers());
|
|
231
|
+
let stage = null, cleanupConfirmed = true, result;
|
|
232
|
+
try {
|
|
233
|
+
const footprint = captureFootprint(input.projectRoot, input.policy), node = stableNodeIdentity(input.nodeCommand);
|
|
234
|
+
requireThat(footprint.sourceDigest === input.sourceDigest && node.environmentDigest === input.environmentDigest
|
|
235
|
+
&& scopedProjectVerifierDigest() === input.verifierDigest, "project-verifier-binding-drift");
|
|
236
|
+
if (input.signal.aborted) return { runId: input.executionRunId, requestDigest: input.requestDigest,
|
|
237
|
+
sourceDigest: input.sourceDigest, environmentDigest: input.environmentDigest, planDigest: input.planDigest,
|
|
238
|
+
status: "cancelled", reason: "cancelled-before-execute", cleanupConfirmed: true };
|
|
239
|
+
stage = fs.mkdtempSync(path.join(input.stagingRoot, "project-verifier-")); fs.chmodSync(stage, 0o700);
|
|
240
|
+
materializeStage(stage, footprint);
|
|
241
|
+
requireThat(captureFootprint(stage, input.policy).sourceDigest === input.sourceDigest,
|
|
242
|
+
"project-verifier-stage-drift");
|
|
243
|
+
requireThat(captureFootprint(input.projectRoot, input.policy).sourceDigest === input.sourceDigest,
|
|
244
|
+
"project-verifier-binding-drift");
|
|
245
|
+
const syntax = input.policy.syntaxFiles.map(file => ["--check", file]);
|
|
246
|
+
const tests = [["--test", ...input.policy.testFiles]];
|
|
247
|
+
const commands = [];
|
|
248
|
+
for (const [id, invocations] of [["type-check", syntax], ["lint", syntax], ["test", tests], ["test:e2e", tests]]) {
|
|
249
|
+
requireThat(captureFootprint(stage, input.policy).sourceDigest === input.sourceDigest,
|
|
250
|
+
"project-verifier-stage-drift");
|
|
251
|
+
const group = await runGroup(id, invocations, input.nodeCommand, stage, input.signal);
|
|
252
|
+
if (group.status !== "completed") { result = { status: group.status, reason: group.reason }; break; }
|
|
253
|
+
commands.push(group.summary);
|
|
254
|
+
requireThat(captureFootprint(stage, input.policy).sourceDigest === input.sourceDigest,
|
|
255
|
+
"project-verifier-stage-drift");
|
|
256
|
+
}
|
|
257
|
+
if (!result) {
|
|
258
|
+
requireThat(captureFootprint(input.projectRoot, input.policy).sourceDigest === input.sourceDigest,
|
|
259
|
+
"project-verifier-binding-drift");
|
|
260
|
+
result = { status: "completed", observation: { schemaVersion: 1,
|
|
261
|
+
workerVersion: SCOPED_PROJECT_VERIFIER_WORKER, verificationRunId: input.executionRunId,
|
|
262
|
+
requestDigest: input.requestDigest, sourceDigest: input.sourceDigest, commandSetDigest: input.planDigest,
|
|
263
|
+
outcome: commands.every(item => item.outcome === "passed") ? "passed" : "failed", commands } };
|
|
264
|
+
}
|
|
265
|
+
} catch (error) {
|
|
266
|
+
result = { status: "error", reason: typeof error?.supervisorCode === "string"
|
|
267
|
+
? error.supervisorCode : "project-verifier-error" };
|
|
268
|
+
} finally {
|
|
269
|
+
if (stage) {
|
|
270
|
+
try { fs.rmSync(stage, { recursive: true, force: false }); cleanupConfirmed = !fs.existsSync(stage); }
|
|
271
|
+
catch { cleanupConfirmed = false; }
|
|
272
|
+
}
|
|
273
|
+
}
|
|
274
|
+
const common = { runId: input.executionRunId, requestDigest: input.requestDigest, sourceDigest: input.sourceDigest,
|
|
275
|
+
environmentDigest: input.environmentDigest, planDigest: input.planDigest, status: result.status, cleanupConfirmed };
|
|
276
|
+
return result.status === "completed" ? { ...common, observation: result.observation }
|
|
277
|
+
: { ...common, reason: result.reason };
|
|
278
|
+
}
|
|
279
|
+
|
|
280
|
+
function normalizeExecution(execution, state) {
|
|
281
|
+
const docs = isScopedDocsPolicy(state.policy), worker = docs ? SCOPED_DOCS_WORKER : SCOPED_PROJECT_VERIFIER_WORKER,
|
|
282
|
+
ids = docs ? DOCS_COMMAND_IDS : ["type-check", "lint", "test", "test:e2e"];
|
|
283
|
+
try {
|
|
284
|
+
requireThat(plain(execution) && execution.runId === state.attemptId
|
|
285
|
+
&& execution.requestDigest === state.requestDigest && execution.sourceDigest === state.sourceDigest
|
|
286
|
+
&& execution.environmentDigest === state.environmentDigest && execution.planDigest === state.planDigest
|
|
287
|
+
&& ["completed", "timeout", "cancelled", "error"].includes(execution.status)
|
|
288
|
+
&& typeof execution.cleanupConfirmed === "boolean", "project-execution-binding-invalid");
|
|
289
|
+
let observation = null;
|
|
290
|
+
if (execution.status === "completed") {
|
|
291
|
+
exact(execution, ["runId", "requestDigest", "sourceDigest", "environmentDigest", "planDigest", "status",
|
|
292
|
+
"cleanupConfirmed", "observation"], "project-execution-binding-invalid");
|
|
293
|
+
observation = execution.observation;
|
|
294
|
+
exact(observation, ["schemaVersion", "workerVersion", "verificationRunId", "requestDigest", "sourceDigest",
|
|
295
|
+
"commandSetDigest", "outcome", "commands"], "project-execution-binding-invalid");
|
|
296
|
+
requireThat(observation.schemaVersion === 1 && observation.workerVersion === worker
|
|
297
|
+
&& observation.verificationRunId === state.attemptId && observation.requestDigest === state.requestDigest
|
|
298
|
+
&& observation.sourceDigest === state.sourceDigest && observation.commandSetDigest === state.planDigest
|
|
299
|
+
&& ["passed", "failed"].includes(observation.outcome) && Array.isArray(observation.commands)
|
|
300
|
+
&& observation.commands.length === ids.length, "project-execution-binding-invalid");
|
|
301
|
+
for (const [index, command] of observation.commands.entries()) {
|
|
302
|
+
exact(command, ["id", "invocationCount", "outcome", "outputSha256", "outputBytes"],
|
|
303
|
+
"project-execution-binding-invalid");
|
|
304
|
+
requireThat(command.id === ids[index] && Number.isSafeInteger(command.invocationCount)
|
|
305
|
+
&& command.invocationCount >= 1 && command.invocationCount <= 64 && ["passed", "failed"].includes(command.outcome)
|
|
306
|
+
&& validHash(command.outputSha256) && Number.isSafeInteger(command.outputBytes)
|
|
307
|
+
&& command.outputBytes >= 0 && command.outputBytes <= MAX_OUTPUT_BYTES
|
|
308
|
+
&& (observation.outcome === "passed") === observation.commands.every(item => item.outcome === "passed"),
|
|
309
|
+
"project-execution-binding-invalid");
|
|
310
|
+
}
|
|
311
|
+
} else {
|
|
312
|
+
exact(execution, ["runId", "requestDigest", "sourceDigest", "environmentDigest", "planDigest", "status",
|
|
313
|
+
"cleanupConfirmed", "reason"], "project-execution-binding-invalid");
|
|
314
|
+
requireThat(typeof execution.reason === "string" && /^[a-z0-9-]{1,80}$/.test(execution.reason),
|
|
315
|
+
"project-execution-binding-invalid");
|
|
316
|
+
}
|
|
317
|
+
return { observation, executionSha256: sha(JSON.stringify(execution)),
|
|
318
|
+
observationSha256: observation ? sha(JSON.stringify(observation)) : null, status: execution.status,
|
|
319
|
+
reason: execution.reason ?? null, cleanupConfirmed: execution.cleanupConfirmed, integrityFailure: false };
|
|
320
|
+
} catch {
|
|
321
|
+
return { observation: null, executionSha256: sha("invalid-project-execution-result"), observationSha256: null,
|
|
322
|
+
status: "error", reason: "project-execution-binding-invalid", cleanupConfirmed: false, integrityFailure: true };
|
|
323
|
+
}
|
|
324
|
+
}
|
|
325
|
+
|
|
326
|
+
export function validateScopedProjectVerificationReceipt(receipt) {
|
|
327
|
+
return validateProjectReceipt(receipt, { exact, plain, requireThat, validHash, validTime, validatePaths,
|
|
328
|
+
HASH, ID, UUID_V4, SCOPED_PROJECT_VERIFICATION_PROTOCOL, SCOPED_PROJECT_VERIFICATION_RECEIPT, SCOPED_PROJECT_VERIFIER_WORKER });
|
|
329
|
+
}
|
|
330
|
+
|
|
331
|
+
export function createScopedProjectVerificationSupervisor({ manifestSha256, brokerIdentitySha256,
|
|
332
|
+
brokerSourceSha256, receiptPrivateKey, verifications, execute = runPinnedProjectVerification,
|
|
333
|
+
clock = Date.now, createAttemptId = randomUUID } = {}) {
|
|
334
|
+
requireThat([manifestSha256, brokerIdentitySha256, brokerSourceSha256].every(validHash)
|
|
335
|
+
&& Array.isArray(verifications) && verifications.length > 0 && verifications.length <= 32
|
|
336
|
+
&& typeof execute === "function" && typeof clock === "function" && typeof createAttemptId === "function",
|
|
337
|
+
"invalid-supervisor-configuration");
|
|
338
|
+
const privateKey = receiptPrivateKey?.type === "private" ? receiptPrivateKey : createPrivateKey(receiptPrivateKey);
|
|
339
|
+
requireThat(privateKey.asymmetricKeyType === "ed25519", "invalid-receipt-key");
|
|
340
|
+
const receiptPublicKey = createPublicKey(privateKey), receiptPublicKeyPem = receiptPublicKey.export({ type: "spki", format: "pem" }),
|
|
341
|
+
receiptKeyDigest = sha(receiptPublicKey.export({ type: "spki", format: "der" })), plans = new Map();
|
|
342
|
+
for (const item of verifications) {
|
|
343
|
+
exact(item, ["manifest", "plan"], "invalid-supervisor-configuration");
|
|
344
|
+
exact(item.manifest, ["id", "protocol", "capabilityDigest", "receiptKeyDigest", "timeoutMs"],
|
|
345
|
+
"invalid-supervisor-configuration");
|
|
346
|
+
exact(item.plan, ["projectRoot", "nodeCommand", "stagingRoot", "policy", "expectedPlanDigest"],
|
|
347
|
+
"invalid-supervisor-configuration");
|
|
348
|
+
const entry = item.manifest, projectRoot = canonicalDirectory(item.plan.projectRoot, "invalid-project-verifier-root"),
|
|
349
|
+
stagingRoot = canonicalDirectory(item.plan.stagingRoot, "invalid-project-verifier-stage"),
|
|
350
|
+
binding = scopedProjectVerificationPlanBinding({ projectRoot, nodeCommand: item.plan.nodeCommand,
|
|
351
|
+
policy: item.plan.policy, timeoutMs: entry.timeoutMs });
|
|
352
|
+
requireThat(typeof entry.id === "string" && ID.test(entry.id) && !plans.has(entry.id)
|
|
353
|
+
&& entry.protocol === SCOPED_PROJECT_VERIFICATION_PROTOCOL && entry.capabilityDigest === binding.capabilityDigest
|
|
354
|
+
&& entry.receiptKeyDigest === receiptKeyDigest && item.plan.expectedPlanDigest === binding.planDigest
|
|
355
|
+
&& outside(projectRoot, stagingRoot) && outside(stagingRoot, projectRoot), "invalid-supervisor-configuration");
|
|
356
|
+
plans.set(entry.id, deepFreeze({ ...binding, id: entry.id, projectRoot, nodeCommand: item.plan.nodeCommand,
|
|
357
|
+
stagingRoot, policy: validatePolicy(item.plan.policy) }));
|
|
358
|
+
}
|
|
359
|
+
let active = null; const completed = new Map();
|
|
360
|
+
function stateFor(attemptId) {
|
|
361
|
+
requireThat(UUID_V4.test(attemptId) && active?.attemptId === attemptId, "verification-attempt-unavailable");
|
|
362
|
+
return active;
|
|
363
|
+
}
|
|
364
|
+
function envelope(state, normalized, settledAtMs) {
|
|
365
|
+
const docs = isScopedDocsPolicy(state.policy);
|
|
366
|
+
const receipt = { version: docs ? 3 : 2, kind: docs ? SCOPED_DOCS_RECEIPT : SCOPED_PROJECT_VERIFICATION_RECEIPT,
|
|
367
|
+
protocol: SCOPED_PROJECT_VERIFICATION_PROTOCOL, verificationId: state.verificationId, action: state.action,
|
|
368
|
+
attemptId: state.attemptId, broker: { identitySha256: brokerIdentitySha256, sourceSha256: brokerSourceSha256 },
|
|
369
|
+
manifestSha256, capabilityDigest: state.capabilityDigest, planDigest: state.planDigest,
|
|
370
|
+
requestDigest: state.requestDigest, sourceDigest: state.sourceDigest, environmentDigest: state.environmentDigest,
|
|
371
|
+
verifierDigest: state.verifierDigest, worker: { runId: state.attemptId,
|
|
372
|
+
version: normalized.observation?.workerVersion ?? null }, status: normalized.status,
|
|
373
|
+
verdict: normalized.status === "completed" && normalized.cleanupConfirmed && normalized.observation
|
|
374
|
+
? "observation-recorded" : "observation-unavailable", cleanup: { confirmed: normalized.cleanupConfirmed },
|
|
375
|
+
evidence: { executionSha256: normalized.executionSha256,
|
|
376
|
+
observationSha256: normalized.observationSha256, reason: normalized.reason,
|
|
377
|
+
outcome: normalized.observation?.outcome ?? null,
|
|
378
|
+
failedCommands: normalized.observation?.commands.filter(item => item.outcome === "failed").map(item => item.id) ?? [] },
|
|
379
|
+
startedAtMs: state.startedAtMs,
|
|
380
|
+
deadlineAtMs: state.deadlineAtMs, settledAtMs, completionAllowed: false, ...(docs ? { scope: state.docsScope } : {}) };
|
|
381
|
+
validateScopedProjectVerificationReceipt(receipt);
|
|
382
|
+
return deepFreeze({ receipt, signature: sign(null, Buffer.from(JSON.stringify(receipt)), privateKey).toString("base64") });
|
|
383
|
+
}
|
|
384
|
+
async function settle(state) {
|
|
385
|
+
if (state.promise) return state.promise;
|
|
386
|
+
state.promise = (async () => {
|
|
387
|
+
let execution;
|
|
388
|
+
try {
|
|
389
|
+
const invocation = state.controller.signal.aborted ? Promise.resolve({ runId: state.attemptId,
|
|
390
|
+
requestDigest: state.requestDigest, sourceDigest: state.sourceDigest,
|
|
391
|
+
environmentDigest: state.environmentDigest, planDigest: state.planDigest, status: "cancelled",
|
|
392
|
+
reason: "cancelled-before-execute", cleanupConfirmed: true }) : Promise.resolve(execute({ ...state,
|
|
393
|
+
signal: state.controller.signal, executionRunId: state.attemptId }));
|
|
394
|
+
invocation.catch(() => undefined); execution = await Promise.race([invocation, state.deadlineGate]);
|
|
395
|
+
} catch {
|
|
396
|
+
execution = { runId: state.attemptId, requestDigest: state.requestDigest, sourceDigest: state.sourceDigest,
|
|
397
|
+
environmentDigest: state.environmentDigest, planDigest: state.planDigest, status: "error",
|
|
398
|
+
reason: "executor-threw", cleanupConfirmed: false };
|
|
399
|
+
}
|
|
400
|
+
const normalized = normalizeExecution(execution, state), settledAtMs = clock();
|
|
401
|
+
requireThat(validTime(settledAtMs) && settledAtMs >= state.startedAtMs, "invalid-supervisor-clock");
|
|
402
|
+
if (state.deadlineExpired || settledAtMs >= state.deadlineAtMs) {
|
|
403
|
+
normalized.observation = null; normalized.observationSha256 = null;
|
|
404
|
+
normalized.status = "timeout"; normalized.reason = "supervisor-deadline";
|
|
405
|
+
} else if (state.cancelRequested) {
|
|
406
|
+
normalized.observation = null; normalized.observationSha256 = null;
|
|
407
|
+
normalized.status = "cancelled"; normalized.reason = "supervisor-cancelled";
|
|
408
|
+
}
|
|
409
|
+
const result = deepFreeze({ envelope: envelope(state, normalized, settledAtMs),
|
|
410
|
+
observation: normalized.observation, integrityFailure: normalized.integrityFailure });
|
|
411
|
+
completed.set(state.attemptId, result); if (completed.size > 32) completed.delete(completed.keys().next().value);
|
|
412
|
+
return result;
|
|
413
|
+
})().finally(() => { clearTimeout(state.deadlineTimer); clearTimeout(state.reapTimer);
|
|
414
|
+
if (active === state) active = null; });
|
|
415
|
+
return state.promise;
|
|
416
|
+
}
|
|
417
|
+
const bridge = { version: "scoped-verification-supervisor-v1", receiptPublicKey: receiptPublicKeyPem,
|
|
418
|
+
receiptKeyDigest,
|
|
419
|
+
begin(request) {
|
|
420
|
+
exact(request, ["verificationId", "action", "brokerIdentitySha256", "brokerSourceSha256", "manifestSha256",
|
|
421
|
+
"capabilityDigest"], "invalid-verification-request");
|
|
422
|
+
requireThat(!active && typeof request.verificationId === "string" && ID.test(request.verificationId)
|
|
423
|
+
&& Number.isSafeInteger(request.action) && request.action > 0
|
|
424
|
+
&& request.brokerIdentitySha256 === brokerIdentitySha256 && request.brokerSourceSha256 === brokerSourceSha256
|
|
425
|
+
&& request.manifestSha256 === manifestSha256, "verification-inflight-or-binding-invalid");
|
|
426
|
+
const plan = plans.get(request.verificationId);
|
|
427
|
+
requireThat(plan && request.capabilityDigest === plan.capabilityDigest, "verification-denied");
|
|
428
|
+
const startedAtMs = clock(), attemptId = createAttemptId();
|
|
429
|
+
requireThat(validTime(startedAtMs) && Number.isSafeInteger(startedAtMs + plan.timeoutMs)
|
|
430
|
+
&& UUID_V4.test(attemptId), "invalid-supervisor-runtime");
|
|
431
|
+
let expire; const deadlineGate = new Promise(resolve => { expire = resolve; });
|
|
432
|
+
const current = isScopedDocsPolicy(plan.policy) ? prepareScopedDocsAttempt(plan, docsHelpers()) : {};
|
|
433
|
+
const state = { ...plan, ...current, verificationId: request.verificationId, action: request.action, attemptId,
|
|
434
|
+
startedAtMs, deadlineAtMs: startedAtMs + plan.timeoutMs, controller: new AbortController(), promise: null,
|
|
435
|
+
cancelRequested: false, deadlineExpired: false, deadlineTimer: null, reapTimer: null, deadlineGate };
|
|
436
|
+
active = state; state.deadlineTimer = setTimeout(() => {
|
|
437
|
+
state.deadlineExpired = true; state.controller.abort(); state.reapTimer = setTimeout(() => expire({
|
|
438
|
+
runId: state.attemptId, requestDigest: state.requestDigest, sourceDigest: state.sourceDigest,
|
|
439
|
+
environmentDigest: state.environmentDigest, planDigest: state.planDigest, status: "timeout",
|
|
440
|
+
reason: "executor-reap-unconfirmed", cleanupConfirmed: false }), REAP_GRACE_MS);
|
|
441
|
+
void settle(state).catch(() => undefined);
|
|
442
|
+
}, plan.timeoutMs);
|
|
443
|
+
return deepFreeze({ attemptId, verificationId: state.verificationId, capabilityDigest: state.capabilityDigest,
|
|
444
|
+
planDigest: state.planDigest, requestDigest: state.requestDigest, sourceDigest: state.sourceDigest,
|
|
445
|
+
environmentDigest: state.environmentDigest, verifierDigest: state.verifierDigest,
|
|
446
|
+
startedAtMs, deadlineAtMs: state.deadlineAtMs });
|
|
447
|
+
},
|
|
448
|
+
execute(attemptId) { return completed.has(attemptId) ? Promise.resolve(completed.get(attemptId))
|
|
449
|
+
: settle(stateFor(attemptId)); },
|
|
450
|
+
cancel(attemptId) { if (!UUID_V4.test(attemptId) || active?.attemptId !== attemptId) return false;
|
|
451
|
+
if (!active.deadlineExpired) active.cancelRequested = true; active.controller.abort(); return true; },
|
|
452
|
+
reconcile(attemptId) { return completed.has(attemptId) ? Promise.resolve(completed.get(attemptId))
|
|
453
|
+
: settle(stateFor(attemptId)); },
|
|
454
|
+
status() { return deepFreeze({ active: active ? { attemptId: active.attemptId,
|
|
455
|
+
verificationId: active.verificationId, action: active.action, capabilityDigest: active.capabilityDigest,
|
|
456
|
+
planDigest: active.planDigest, requestDigest: active.requestDigest, sourceDigest: active.sourceDigest,
|
|
457
|
+
environmentDigest: active.environmentDigest, verifierDigest: active.verifierDigest,
|
|
458
|
+
startedAtMs: active.startedAtMs, deadlineAtMs: active.deadlineAtMs } : null, completed: completed.size }); }
|
|
459
|
+
};
|
|
460
|
+
return deepFreeze(bridge);
|
|
461
|
+
}
|