@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 +1,488 @@
|
|
|
1
|
+
import crypto from "node:crypto";
|
|
2
|
+
import { isApiPreservationCriterion } from "../../extensions/acceptance-api-baseline.js";
|
|
3
|
+
import fs from "node:fs";
|
|
4
|
+
import path from "node:path";
|
|
5
|
+
import { execFileSync } from "node:child_process";
|
|
6
|
+
import { parse } from "@babel/parser";
|
|
7
|
+
import { acceptanceCriterionBindingValid } from "../../extensions/acceptance-behavior-proof.js";
|
|
8
|
+
import { acceptanceLanguageAdapterForPath, isAcceptanceTestPath } from "../../extensions/acceptance-language-adapters.js";
|
|
9
|
+
import { matchesAnyPath } from "../../extensions/policy-core.js";
|
|
10
|
+
import { allConfiguredVerifierEvidenceCurrent, latestObservedVerificationEvidence } from "../../extensions/verification-intelligence.js";
|
|
11
|
+
import { workingTreeEvidenceDigest } from "../../extensions/working-tree-digest.js";
|
|
12
|
+
import { captureWorkspaceVerificationSnapshot } from "../../extensions/workspace-verification-snapshot.js";
|
|
13
|
+
import { decodeBaselineRepoPath, readTaskBaselineBlob, readTaskBaselineManifest,
|
|
14
|
+
taskBaselineRetentionState } from "../inspection/source-evidence-store.ts";
|
|
15
|
+
import { readWorkspaceFile } from "../inspection/workspace-file-reader.ts";
|
|
16
|
+
import { routeNamedSourceTargets } from "../../extensions/acceptance-source-origin.js";
|
|
17
|
+
|
|
18
|
+
const MAX_FILE_BYTES = 64_000, MAX_FILES = 8, MAX_NODES = 12_000;
|
|
19
|
+
const hash = value => `sha256:${crypto.createHash("sha256").update(value).digest("hex")}`;
|
|
20
|
+
const reject = reason => { throw Object.assign(new Error(reason), { apiProofReason: reason }); };
|
|
21
|
+
const mathMethods = new Set(["ceil", "floor", "round", "trunc", "abs", "min", "max"]);
|
|
22
|
+
|
|
23
|
+
function safePath(value) {
|
|
24
|
+
if (typeof value !== "string" || !value || value.includes("\\") || path.posix.isAbsolute(value)
|
|
25
|
+
|| value.split("/").some(part => !part || part === "..") || value.includes("\0")) reject("api-source-scope-unavailable");
|
|
26
|
+
return path.posix.normalize(value);
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
/** A syntax-only, closed API contract: static named function exports, plain
|
|
30
|
+
* positional parameters and a proven primitive or local-array return representation. Bodies
|
|
31
|
+
* are not equated; behavioral compatibility remains with the other criteria.
|
|
32
|
+
*/
|
|
33
|
+
function publicApi(text, includeImplementation = false) {
|
|
34
|
+
if (Buffer.byteLength(text, "utf8") > MAX_FILE_BYTES) reject("api-proof-limit");
|
|
35
|
+
const ast = parse(text, { sourceType: "module", errorRecovery: false, attachComment: false });
|
|
36
|
+
const functions = new Map(), exports = [];
|
|
37
|
+
const parameterName = node => node.type === "AssignmentPattern" ? node.left.name : node.name;
|
|
38
|
+
const parameterDefault = node => {
|
|
39
|
+
if (node.type === "Identifier") return undefined;
|
|
40
|
+
if (node.type !== "AssignmentPattern" || node.left.type !== "Identifier") reject("api-contract-unsupported");
|
|
41
|
+
const value = node.right;
|
|
42
|
+
if (value.type === "Identifier" && value.name === "undefined") return { intrinsic: "undefined" };
|
|
43
|
+
if (["NumericLiteral", "StringLiteral", "BooleanLiteral", "NullLiteral"].includes(value.type)) return { type: value.type, value: value.value ?? null };
|
|
44
|
+
if (value.type === "UnaryExpression" && ["+", "-"].includes(value.operator) && value.argument.type === "NumericLiteral") {
|
|
45
|
+
return { type: "signed-number", operator: value.operator, value: value.argument.value };
|
|
46
|
+
}
|
|
47
|
+
if (value.type === "MemberExpression" && !value.computed && value.object.name === "Number" && value.property.name === "MAX_SAFE_INTEGER") return { intrinsic: "Number.MAX_SAFE_INTEGER" };
|
|
48
|
+
if (value.type === "CallExpression" && value.arguments.length === 0 && value.callee.type === "MemberExpression"
|
|
49
|
+
&& !value.callee.computed && value.callee.object.name === "Date" && value.callee.property.name === "now") return { intrinsic: "Date.now()" };
|
|
50
|
+
reject("api-contract-unsupported");
|
|
51
|
+
};
|
|
52
|
+
let nodes = 0;
|
|
53
|
+
let booleanSurface = false, numericSurface = false, dateIntrinsicsUsed = false, arrayIntrinsicUsed = false;
|
|
54
|
+
const booleanExpression = node => node?.type === "BooleanLiteral"
|
|
55
|
+
|| node?.type === "UnaryExpression" && node.operator === "!"
|
|
56
|
+
|| node?.type === "BinaryExpression" && ["===", "!==", "==", "!=", "<", "<=", ">", ">=", "in", "instanceof"].includes(node.operator);
|
|
57
|
+
const memberPath = node => node?.type === "Identifier" ? node.name
|
|
58
|
+
: node?.type === "MemberExpression" && !node.computed ? `${memberPath(node.object)}.${node.property.name}` : "";
|
|
59
|
+
const booleanHelperCall = node => {
|
|
60
|
+
if (!booleanSurface || node.callee?.type !== "MemberExpression" || node.callee.computed) return false;
|
|
61
|
+
const callee = node.callee, member = memberPath(callee);
|
|
62
|
+
if (["Date.now", "Date.parse", "Date.UTC", "Date.prototype.getTime.call", "Number.isNaN"].includes(member)) return true;
|
|
63
|
+
if (callee.object.type === "RegExpLiteral") return callee.property.name === "exec";
|
|
64
|
+
return ["slice", "padEnd", "getTime", "setUTCFullYear", "setUTCHours", "getUTCFullYear", "getUTCMonth", "getUTCDate"].includes(callee.property.name);
|
|
65
|
+
};
|
|
66
|
+
const numericExpression = node => node?.type === "NumericLiteral"
|
|
67
|
+
|| node?.type === "CallExpression" && (node.callee.type === "Identifier" && node.callee.name === "Number"
|
|
68
|
+
|| node.callee.type === "MemberExpression" && !node.callee.computed && node.callee.object.name === "Math" && mathMethods.has(node.callee.property.name));
|
|
69
|
+
const reduceCall = node => node?.type === "CallExpression" && node.callee.type === "MemberExpression"
|
|
70
|
+
&& !node.callee.computed && node.callee.property.name === "reduce" && node.arguments.length === 2
|
|
71
|
+
&& node.arguments[0].type === "ArrowFunctionExpression" && !node.arguments[0].async
|
|
72
|
+
&& node.arguments[0].params.every(parameter => parameter.type === "Identifier")
|
|
73
|
+
&& new Set(node.arguments[0].params.map(parameter => parameter.name)).size === node.arguments[0].params.length;
|
|
74
|
+
const collectionBindings = new Map(), collectionValidity = new Map();
|
|
75
|
+
const collectionMethods = { array: new Set(["push", "map", "sort"]), map: new Set(["get", "has", "set", "values"]) };
|
|
76
|
+
const bindingsFor = fn => {
|
|
77
|
+
const names = new Set(), collections = new Map();
|
|
78
|
+
let valid = true;
|
|
79
|
+
const bind = id => {
|
|
80
|
+
if (id?.type !== "Identifier" || names.has(id.name) || ["Map", "Math", "Number"].includes(id.name)) { valid = false; return; }
|
|
81
|
+
names.add(id.name);
|
|
82
|
+
};
|
|
83
|
+
fn.params.forEach(node => bind(node.type === "AssignmentPattern" ? node.left : node));
|
|
84
|
+
let bindingNodes = 0;
|
|
85
|
+
const visit = (node, depth = 0) => {
|
|
86
|
+
if (!node || typeof node.type !== "string") return;
|
|
87
|
+
if (++bindingNodes > MAX_NODES || depth > 128) reject("api-proof-limit");
|
|
88
|
+
if (["FunctionExpression", "FunctionDeclaration", "ObjectMethod", "ClassMethod", "ClassPrivateMethod"].includes(node.type)) valid = false;
|
|
89
|
+
if (node.type === "VariableDeclarator") {
|
|
90
|
+
bind(node.id);
|
|
91
|
+
if (node.init?.type === "ArrayExpression") collections.set(node.id.name, "array");
|
|
92
|
+
if (node.init?.type === "NewExpression" && node.init.callee.type === "Identifier"
|
|
93
|
+
&& node.init.callee.name === "Map" && !functions.has("Map") && node.init.arguments.length === 0) collections.set(node.id.name, "map");
|
|
94
|
+
}
|
|
95
|
+
if (node.type === "ArrowFunctionExpression") {
|
|
96
|
+
if (node.async) valid = false;
|
|
97
|
+
node.params.forEach(bind);
|
|
98
|
+
}
|
|
99
|
+
for (const [key, value] of Object.entries(node)) {
|
|
100
|
+
if (["loc", "extra", "comments", "tokens", "errors"].includes(key)) continue;
|
|
101
|
+
if (Array.isArray(value)) value.forEach(item => visit(item, depth + 1));
|
|
102
|
+
else if (value && typeof value === "object") visit(value, depth + 1);
|
|
103
|
+
}
|
|
104
|
+
};
|
|
105
|
+
visit(fn.body); collectionValidity.set(fn, valid); return collections;
|
|
106
|
+
};
|
|
107
|
+
const collectionCall = (node, owner) => {
|
|
108
|
+
const callee = node?.callee;
|
|
109
|
+
if (node?.type !== "CallExpression" || callee?.type !== "MemberExpression" || callee.computed) return false;
|
|
110
|
+
const kind = callee.object.type === "ArrayExpression" ? "array"
|
|
111
|
+
: callee.object.type === "Identifier" ? collectionBindings.get(owner)?.get(callee.object.name) : undefined;
|
|
112
|
+
return Boolean(kind && collectionMethods[kind].has(callee.property.name) && collectionValidity.get(owner));
|
|
113
|
+
};
|
|
114
|
+
const inspect = (node, parent, grandparent, depth = 0, owner = null) => {
|
|
115
|
+
if (!node || typeof node.type !== "string") return;
|
|
116
|
+
if (++nodes > MAX_NODES || depth > 128) reject("api-proof-limit");
|
|
117
|
+
if (node.type === "FunctionDeclaration") {
|
|
118
|
+
owner = node;
|
|
119
|
+
collectionBindings.set(owner, bindingsFor(node));
|
|
120
|
+
}
|
|
121
|
+
if (["AssignmentExpression", "UpdateExpression", "AwaitExpression", "YieldExpression"].includes(node.type)) reject("api-contract-unsupported");
|
|
122
|
+
if (node.type === "VariableDeclarator" && node.id.type === "Identifier" && node.id.name === "undefined") reject("api-contract-unsupported");
|
|
123
|
+
const numericDefault = node.type === "Identifier" && node.name === "Number" && parent?.type === "MemberExpression"
|
|
124
|
+
&& parent.object === node && !parent.computed && parent.property.name === "MAX_SAFE_INTEGER"
|
|
125
|
+
&& grandparent?.type === "AssignmentPattern" && grandparent.right === parent;
|
|
126
|
+
if (!numericDefault && node.type === "Identifier" && ["Math", "Number"].includes(node.name)
|
|
127
|
+
&& !(node.name === "Number" && parent?.type === "CallExpression" && parent.callee === node)
|
|
128
|
+
&& !(parent?.type === "MemberExpression" && !parent.computed && parent.object === node
|
|
129
|
+
&& (node.name === "Math" ? mathMethods.has(parent.property.name) : ["isInteger", "isSafeInteger", "isFinite", ...(booleanSurface ? ["isNaN"] : [])].includes(parent.property.name))
|
|
130
|
+
&& grandparent?.type === "CallExpression" && grandparent.callee === parent)) {
|
|
131
|
+
reject("api-contract-unsupported");
|
|
132
|
+
}
|
|
133
|
+
if (numericSurface && arrayIntrinsicUsed && node.type === "Identifier" && node.name === "Array"
|
|
134
|
+
&& !(parent?.type === "MemberExpression" && parent.object === node && !parent.computed && parent.property.name === "isArray"
|
|
135
|
+
&& grandparent?.type === "CallExpression" && grandparent.callee === parent)) reject("api-contract-unsupported");
|
|
136
|
+
if (dateIntrinsicsUsed && node.type === "Identifier" && node.name === "Date"
|
|
137
|
+
&& !(parent?.type === "NewExpression" && parent.callee === node)
|
|
138
|
+
&& !(parent?.type === "BinaryExpression" && parent.operator === "instanceof" && parent.right === node)
|
|
139
|
+
&& !(parent?.type === "MemberExpression" && !parent.computed && parent.object === node
|
|
140
|
+
&& ["now", "parse", "UTC", "prototype"].includes(parent.property.name))) reject("api-contract-unsupported");
|
|
141
|
+
if (node.type === "CallExpression") {
|
|
142
|
+
const callee = node.callee;
|
|
143
|
+
const intrinsic = callee.type === "MemberExpression" && !callee.computed && callee.object.type === "Identifier"
|
|
144
|
+
&& (callee.object.name === "Math" && mathMethods.has(callee.property.name)
|
|
145
|
+
|| callee.object.name === "Number" && ["isInteger", "isSafeInteger", "isFinite"].includes(callee.property.name));
|
|
146
|
+
if (!intrinsic && !collectionCall(node, owner) && !booleanHelperCall(node)
|
|
147
|
+
&& !(parent?.type === "AssignmentPattern" && parent.right === node && memberPath(callee) === "Date.now" && node.arguments.length === 0)
|
|
148
|
+
&& !(numericSurface && (reduceCall(node) || memberPath(callee) === "Array.isArray"))
|
|
149
|
+
&& !(callee.type === "Identifier" && callee.name === "Number" && !functions.has("Number"))
|
|
150
|
+
&& !(callee.type === "Identifier" && functions.has(callee.name))) reject("api-contract-unsupported");
|
|
151
|
+
}
|
|
152
|
+
if (node.type === "NewExpression" && !(node.callee.type === "Identifier"
|
|
153
|
+
&& (["Error", "TypeError", "RangeError"].includes(node.callee.name) || booleanSurface && node.callee.name === "Date" || node.callee.name === "Map" && node.arguments.length === 0 && collectionValidity.get(owner))
|
|
154
|
+
&& !functions.has(node.callee.name))) reject("api-contract-unsupported");
|
|
155
|
+
for (const [key, value] of Object.entries(node)) {
|
|
156
|
+
if (["loc", "extra", "comments", "tokens", "errors"].includes(key)) continue;
|
|
157
|
+
if (Array.isArray(value)) value.forEach(item => inspect(item, node, parent, depth + 1, owner));
|
|
158
|
+
else if (value && typeof value === "object") inspect(value, node, parent, depth + 1, owner);
|
|
159
|
+
}
|
|
160
|
+
};
|
|
161
|
+
for (const statement of ast.program.body) {
|
|
162
|
+
const node = statement.type === "ExportNamedDeclaration" && !statement.source ? statement.declaration : statement;
|
|
163
|
+
if (node?.type !== "FunctionDeclaration" || !node.id || node.id.name === "undefined" || functions.has(node.id.name)
|
|
164
|
+
|| node.async || node.generator || node.params.some(item => !["Identifier", "AssignmentPattern"].includes(item.type))
|
|
165
|
+
|| node.params.some(item => parameterName(item) === "undefined")
|
|
166
|
+
|| new Set(node.params.map(parameterName)).size !== node.params.length) reject("api-contract-unsupported");
|
|
167
|
+
node.params.forEach(parameterDefault);
|
|
168
|
+
functions.set(node.id.name, node);
|
|
169
|
+
if (statement.type === "ExportNamedDeclaration") exports.push(node);
|
|
170
|
+
}
|
|
171
|
+
if (exports.length === 0) reject("api-contract-unsupported");
|
|
172
|
+
// Only a closed module with syntactically boolean public returns admits these helper calls.
|
|
173
|
+
// Function bindings cannot escape or be shadowed; the normal return-path walk still applies.
|
|
174
|
+
let surfaceNodes = 0, closed = true, numericClosed = true;
|
|
175
|
+
const inspectSurface = (node, parent, returns, depth = 0) => {
|
|
176
|
+
if (!node || typeof node.type !== "string") return;
|
|
177
|
+
if (++surfaceNodes > MAX_NODES || depth > 128) reject("api-proof-limit");
|
|
178
|
+
if (["FunctionExpression", "ArrowFunctionExpression", "FunctionDeclaration", "ObjectMethod", "ClassMethod"].includes(node.type)) {
|
|
179
|
+
closed = false;
|
|
180
|
+
if (node.type !== "ArrowFunctionExpression" || !reduceCall(parent) || parent.arguments[0] !== node) numericClosed = false;
|
|
181
|
+
}
|
|
182
|
+
if (node.type === "Identifier" && functions.has(node.name)
|
|
183
|
+
&& !(parent?.type === "CallExpression" && parent.callee === node)) closed = numericClosed = false;
|
|
184
|
+
if (node.type === "NewExpression" && node.callee?.name === "Date"
|
|
185
|
+
|| node.type === "CallExpression" && memberPath(node.callee).startsWith("Date.")) dateIntrinsicsUsed = true;
|
|
186
|
+
if (node.type === "CallExpression" && memberPath(node.callee) === "Array.isArray") arrayIntrinsicUsed = true;
|
|
187
|
+
if (node.type === "ReturnStatement") returns.push(node.argument);
|
|
188
|
+
for (const [key, value] of Object.entries(node)) {
|
|
189
|
+
if (["loc", "extra", "comments", "tokens", "errors"].includes(key)) continue;
|
|
190
|
+
if (Array.isArray(value)) value.forEach(item => inspectSurface(item, node, node.type === "ArrowFunctionExpression" ? [] : returns, depth + 1));
|
|
191
|
+
else if (value && typeof value === "object") inspectSurface(value, node, node.type === "ArrowFunctionExpression" ? [] : returns, depth + 1);
|
|
192
|
+
}
|
|
193
|
+
};
|
|
194
|
+
const publicReturns = [], numericReturns = [];
|
|
195
|
+
for (const fn of functions.values()) {
|
|
196
|
+
const returns = [];
|
|
197
|
+
if (fn.params.some(parameter => functions.has(parameterName(parameter)))) closed = numericClosed = false;
|
|
198
|
+
fn.params.forEach(parameter => inspectSurface(parameter, fn, []));
|
|
199
|
+
inspectSurface(fn.body, fn, returns);
|
|
200
|
+
if (exports.includes(fn)) {
|
|
201
|
+
publicReturns.push(returns.length > 0 && returns.every(booleanExpression));
|
|
202
|
+
numericReturns.push(returns.length > 0 && returns.every(numericExpression));
|
|
203
|
+
}
|
|
204
|
+
}
|
|
205
|
+
booleanSurface = closed && publicReturns.every(Boolean);
|
|
206
|
+
numericSurface = numericClosed && numericReturns.every(Boolean);
|
|
207
|
+
inspect(ast);
|
|
208
|
+
const kinds = new Map(), active = new Set();
|
|
209
|
+
const resultKind = name => {
|
|
210
|
+
if (kinds.has(name)) return kinds.get(name);
|
|
211
|
+
if (active.has(name) || !functions.has(name)) return undefined;
|
|
212
|
+
active.add(name);
|
|
213
|
+
const fn = functions.get(name), locals = new Map(fn.params.map(item => [parameterName(item), undefined]));
|
|
214
|
+
const expressionKind = expression => {
|
|
215
|
+
if (!expression) return undefined;
|
|
216
|
+
const literal = { NumericLiteral: "number", StringLiteral: "string", BooleanLiteral: "boolean", NullLiteral: "null" }[expression.type];
|
|
217
|
+
if (literal) return literal;
|
|
218
|
+
if (booleanExpression(expression)) return "boolean";
|
|
219
|
+
if (expression.type === "ArrayExpression") return "array";
|
|
220
|
+
if (expression.type === "NewExpression" && expression.callee.type === "Identifier"
|
|
221
|
+
&& expression.callee.name === "Map") return "local-map";
|
|
222
|
+
if (expression.type === "Identifier") return locals.get(expression.name);
|
|
223
|
+
if (expression.type === "UnaryExpression" && ["+", "-"].includes(expression.operator)
|
|
224
|
+
&& expressionKind(expression.argument) === "number") return "number";
|
|
225
|
+
if (expression.type === "BinaryExpression" && ["+", "-", "*", "/", "%", "**"].includes(expression.operator)
|
|
226
|
+
&& expressionKind(expression.left) === "number" && expressionKind(expression.right) === "number") return "number";
|
|
227
|
+
if (expression.type === "ConditionalExpression") {
|
|
228
|
+
const yes = expressionKind(expression.consequent), no = expressionKind(expression.alternate);
|
|
229
|
+
return yes && yes === no ? yes : undefined;
|
|
230
|
+
}
|
|
231
|
+
if (expression.type === "CallExpression") {
|
|
232
|
+
const callee = expression.callee;
|
|
233
|
+
if (callee.type === "Identifier" && callee.name === "Number" && !locals.has("Number") && !functions.has("Number")) return "number";
|
|
234
|
+
if (callee.type === "MemberExpression" && !callee.computed && ["map", "sort"].includes(callee.property.name)
|
|
235
|
+
&& expressionKind(callee.object) === "array" && collectionCall(expression, fn)) return "array";
|
|
236
|
+
if (callee.type === "MemberExpression" && !callee.computed && callee.object.type === "Identifier"
|
|
237
|
+
&& callee.object.name === "Math" && mathMethods.has(callee.property.name)) return "number";
|
|
238
|
+
if (callee.type === "Identifier" && !locals.has(callee.name)) return resultKind(callee.name);
|
|
239
|
+
}
|
|
240
|
+
return undefined;
|
|
241
|
+
};
|
|
242
|
+
const walk = statements => {
|
|
243
|
+
const returns = new Set();
|
|
244
|
+
for (const statement of statements) {
|
|
245
|
+
if (statement.type === "ReturnStatement") {
|
|
246
|
+
const kind = expressionKind(statement.argument); if (!kind) reject("api-contract-unsupported");
|
|
247
|
+
returns.add(kind); return { terminal: true, returns };
|
|
248
|
+
}
|
|
249
|
+
if (statement.type === "ThrowStatement") return { terminal: true, returns };
|
|
250
|
+
if (statement.type === "IfStatement") {
|
|
251
|
+
const before = new Map(locals);
|
|
252
|
+
const left = walk(statement.consequent.type === "BlockStatement" ? statement.consequent.body : [statement.consequent]);
|
|
253
|
+
locals.clear(); before.forEach((value, key) => locals.set(key, value));
|
|
254
|
+
const right = statement.alternate ? walk(statement.alternate.type === "BlockStatement" ? statement.alternate.body : [statement.alternate])
|
|
255
|
+
: { terminal: false, returns: new Set() };
|
|
256
|
+
locals.clear(); before.forEach((value, key) => locals.set(key, value));
|
|
257
|
+
for (const kind of [...left.returns, ...right.returns]) returns.add(kind);
|
|
258
|
+
if (left.terminal && right.terminal) return { terminal: true, returns };
|
|
259
|
+
} else if (statement.type === "VariableDeclaration" && statement.kind === "const") {
|
|
260
|
+
for (const item of statement.declarations) {
|
|
261
|
+
if (item.id.type !== "Identifier" || locals.has(item.id.name)) reject("api-contract-unsupported");
|
|
262
|
+
locals.set(item.id.name, (booleanSurface || numericSurface) && exports.includes(fn) ? undefined : expressionKind(item.init));
|
|
263
|
+
}
|
|
264
|
+
} else if (statement.type === "ForOfStatement" && !statement.await
|
|
265
|
+
&& statement.left.type === "VariableDeclaration" && statement.left.kind === "const"
|
|
266
|
+
&& statement.left.declarations.length === 1 && statement.left.declarations[0].id.type === "Identifier") {
|
|
267
|
+
const before = new Map(locals), binding = statement.left.declarations[0].id.name;
|
|
268
|
+
if (locals.has(binding)) reject("api-contract-unsupported");
|
|
269
|
+
locals.set(binding, undefined);
|
|
270
|
+
const body = walk(statement.body.type === "BlockStatement" ? statement.body.body : [statement.body]);
|
|
271
|
+
if (body.returns.size || body.terminal) reject("api-contract-unsupported");
|
|
272
|
+
locals.clear(); before.forEach((value, key) => locals.set(key, value));
|
|
273
|
+
} else if (statement.type === "ExpressionStatement" && statement.expression.type === "CallExpression"
|
|
274
|
+
&& (collectionCall(statement.expression, fn)
|
|
275
|
+
|| statement.expression.callee.type === "Identifier" && functions.has(statement.expression.callee.name)
|
|
276
|
+
&& !locals.has(statement.expression.callee.name))) {
|
|
277
|
+
// Ignoring a local helper's result cannot establish a return type;
|
|
278
|
+
// the caller's own subsequent return must still prove it.
|
|
279
|
+
} else reject("api-contract-unsupported");
|
|
280
|
+
}
|
|
281
|
+
return { terminal: false, returns };
|
|
282
|
+
};
|
|
283
|
+
const result = walk(fn.body.body);
|
|
284
|
+
active.delete(name);
|
|
285
|
+
const kind = result.terminal && result.returns.size === 1 ? [...result.returns][0] : undefined;
|
|
286
|
+
kinds.set(name, kind); return kind;
|
|
287
|
+
};
|
|
288
|
+
const contracts = exports.map(fn => {
|
|
289
|
+
const result = resultKind(fn.id.name); if (!result || result === "local-map" || result === "array" && !collectionValidity.get(fn)) reject("api-contract-unsupported");
|
|
290
|
+
const defaults = fn.params.map(parameterDefault);
|
|
291
|
+
return { ...(defaults.some(value => value !== undefined) ? { defaults: defaults.map(value => value ?? null), functionLength: fn.params.findIndex(item => item.type === "AssignmentPattern") } : {}), exportName: fn.id.name, callable: "sync-function", parameters: fn.params.map(parameterName), result: result === "array" ? "collection:array" : `primitive:${result}` };
|
|
292
|
+
}).sort((a, b) => a.exportName.localeCompare(b.exportName, "en"));
|
|
293
|
+
if (!includeImplementation) return contracts;
|
|
294
|
+
const metadata = new Set(["start", "end", "loc", "extra", "leadingComments", "trailingComments", "innerComments", "comments"]);
|
|
295
|
+
const implementationClosure = name => {
|
|
296
|
+
const closure = new Map(), pending = [name];
|
|
297
|
+
while (pending.length) {
|
|
298
|
+
const next = pending.pop();
|
|
299
|
+
if (closure.has(next)) continue;
|
|
300
|
+
const fn = functions.get(next);
|
|
301
|
+
closure.set(next, JSON.stringify(fn, (key, value) => metadata.has(key) ? undefined : value));
|
|
302
|
+
const visit = node => {
|
|
303
|
+
if (!node || typeof node.type !== "string") return;
|
|
304
|
+
if (node.type === "Identifier" && functions.has(node.name) && !closure.has(node.name)) pending.push(node.name);
|
|
305
|
+
for (const [key, value] of Object.entries(node)) {
|
|
306
|
+
if (metadata.has(key)) continue;
|
|
307
|
+
if (Array.isArray(value)) value.forEach(visit);
|
|
308
|
+
else if (value && typeof value === "object") visit(value);
|
|
309
|
+
}
|
|
310
|
+
};
|
|
311
|
+
visit(fn);
|
|
312
|
+
}
|
|
313
|
+
return JSON.stringify([...closure].sort(([a], [b]) => a.localeCompare(b, "en")));
|
|
314
|
+
};
|
|
315
|
+
return { contracts, implementations: new Map(exports.map(fn => [fn.id.name, implementationClosure(fn.id.name)])) };
|
|
316
|
+
}
|
|
317
|
+
|
|
318
|
+
// Parameter spelling is local implementation detail. Keep the positions of any
|
|
319
|
+
// retained names as a conservative reorder check; defaults and arity stay exact.
|
|
320
|
+
function samePublicContracts(before, after, permittedDefaults = new Set()) {
|
|
321
|
+
return before.length === after.length && before.every((previous, index) => {
|
|
322
|
+
const next = after[index], { parameters: oldNames, ...oldShape } = previous;
|
|
323
|
+
const { parameters: newNames, ...newShape } = next;
|
|
324
|
+
if (permittedDefaults.has(previous.exportName) && next.exportName === previous.exportName
|
|
325
|
+
&& oldShape.defaults && newShape.defaults && oldShape.defaults.length === newShape.defaults.length) {
|
|
326
|
+
const changes = oldShape.defaults.flatMap((value, position) => JSON.stringify(value) === JSON.stringify(newShape.defaults[position]) ? [] : [position]);
|
|
327
|
+
// Permission to change an initializer does not add/remove a parameter or
|
|
328
|
+
// a default, change function.length, or waive any behavioral criterion.
|
|
329
|
+
if (changes.length === 1 && oldShape.defaults[changes[0]] !== null && newShape.defaults[changes[0]] !== null) {
|
|
330
|
+
delete oldShape.defaults;
|
|
331
|
+
delete newShape.defaults;
|
|
332
|
+
}
|
|
333
|
+
}
|
|
334
|
+
return oldNames.length === newNames.length && JSON.stringify(oldShape) === JSON.stringify(newShape)
|
|
335
|
+
&& oldNames.every((name, position) => !newNames.includes(name) || newNames[position] === name);
|
|
336
|
+
});
|
|
337
|
+
}
|
|
338
|
+
|
|
339
|
+
function permittedDefaultInitializerChanges(task, targeted) {
|
|
340
|
+
const request = task.operatorRequest;
|
|
341
|
+
if (typeof request !== "string" || task.operatorRequestDigest !== hash(request).replace("sha256:", "operator-request-v1:")
|
|
342
|
+
|| targeted.size !== 1) return new Set();
|
|
343
|
+
const normalize = text => text.replace(/\s+/g, " ").trim();
|
|
344
|
+
const requestClauses = request.split(/\r?\n|(?<=[.!?;])\s+/).map(normalize);
|
|
345
|
+
const permitted = task.acceptanceCriteria.some((text, index) =>
|
|
346
|
+
acceptanceCriterionBindingValid(task, task.acceptanceReceipt, task.acceptanceReceipt.criteria[index], index)
|
|
347
|
+
&& /^Changing the formal default initializer(?: to [^.;\n]{1,200})? is permitted[.;]?$/i.test(text.trim())
|
|
348
|
+
&& requestClauses.includes(normalize(text)));
|
|
349
|
+
return permitted ? new Set(targeted) : new Set();
|
|
350
|
+
}
|
|
351
|
+
|
|
352
|
+
export function compareSupportedPublicApi(before, after) {
|
|
353
|
+
try {
|
|
354
|
+
const baseline = publicApi(before), current = publicApi(after);
|
|
355
|
+
return { compatible: samePublicContracts(baseline, current), baseline, current };
|
|
356
|
+
} catch (error) { return { compatible: null, reason: error.apiProofReason ?? "api-contract-unsupported" }; }
|
|
357
|
+
}
|
|
358
|
+
|
|
359
|
+
function targetedExports(task, apiCriterion, sourceEntries, inventories) {
|
|
360
|
+
const targets = new Set();
|
|
361
|
+
const exports = inventories.flatMap(({ file, current }) => current.contracts.map(({ exportName }) => ({
|
|
362
|
+
contractName: exportName.toLowerCase(), exportName: exportName.toLowerCase(), sourceName: exportName.toLowerCase(), sourcePath: file
|
|
363
|
+
})));
|
|
364
|
+
task.acceptanceReceipt.criteria.forEach((criterion, index) => {
|
|
365
|
+
if (criterion.id === apiCriterion.id || !["requested-behavior", "boundary-case", "invalid-input-rejection"].includes(criterion.obligation)
|
|
366
|
+
|| !acceptanceCriterionBindingValid(task, task.acceptanceReceipt, criterion, index)) return;
|
|
367
|
+
const text = task.acceptanceCriteria[index];
|
|
368
|
+
// Only an explicit callable subject can authorize its implementation change.
|
|
369
|
+
// A name in an example, comparison, dependency or test title is not a target.
|
|
370
|
+
const subject = /^\s*(?:`([a-z_$][a-z0-9_$]*)(?:\([^`\n]*\))?`|([a-z_$][a-z0-9_$]*)\([^\n)]*\))\s+(?:must\s+|shall\s+)?(?:use|returns?|rejects?|throws?|clamps?)\b/i.exec(text);
|
|
371
|
+
const explicitChange = /^\s*(?:Fix|Repair)\s+`([a-z_$][a-z0-9_$]*)\([^`\n]*\)`\s+in\s+`([^`\n]+)`\.\s*$/i.exec(text);
|
|
372
|
+
if (!subject && !explicitChange) return;
|
|
373
|
+
if (explicitChange && !sourceEntries.some(entry => entry.path === explicitChange[2])) return;
|
|
374
|
+
const name = (explicitChange?.[1] ?? subject[1] ?? subject[2]).toLowerCase();
|
|
375
|
+
const route = routeNamedSourceTargets({ namedTargets: [name], provenanceTargets: new Set(), sourceEntries, exports, taskText: text }).routes[0];
|
|
376
|
+
if (route?.kind === "export") targets.add(`${route.origin.path}:${route.origin.name}`);
|
|
377
|
+
});
|
|
378
|
+
return targets;
|
|
379
|
+
}
|
|
380
|
+
|
|
381
|
+
// Protected entries deliberately have no bytes. They do not invalidate a complete
|
|
382
|
+
// inventory's unrelated source entries, but truncation or failed reads still do.
|
|
383
|
+
function readableBaselineRoot(root) {
|
|
384
|
+
return root.state === "current" || root.state === "unavailable" && root.reasonCode === "protected-path"
|
|
385
|
+
&& root.entries.some(entry => entry.state === "protected")
|
|
386
|
+
&& root.entries.every(entry => !["oversized", "unavailable"].includes(entry.state));
|
|
387
|
+
}
|
|
388
|
+
function readableBaseline(manifest) {
|
|
389
|
+
return manifest && (manifest.captureState === "current"
|
|
390
|
+
|| manifest.captureState === "unavailable" && manifest.reasonCode === "protected-path")
|
|
391
|
+
&& manifest.roots.every(readableBaselineRoot);
|
|
392
|
+
}
|
|
393
|
+
|
|
394
|
+
// Reuse the captured task-start head and dirty blobs for both API and scoped
|
|
395
|
+
// material checks. The sparse working-tree map is not a raw content inventory.
|
|
396
|
+
export function readTaskBaselineSource(cwd, task, file) {
|
|
397
|
+
file = safePath(file);
|
|
398
|
+
const manifest = readTaskBaselineManifest(cwd, task.taskRunId);
|
|
399
|
+
if (!readableBaseline(manifest)) reject("api-baseline-unavailable");
|
|
400
|
+
if (manifest.taskId !== task.taskId || manifest.taskRunId !== task.taskRunId
|
|
401
|
+
|| manifest.sessionIdentityHash !== hash(task.sessionId) || manifest.capturedAt !== task.createdAt
|
|
402
|
+
|| manifest.baselineTreeDigest !== workingTreeEvidenceDigest(task.baselineFileDigests)
|
|
403
|
+
|| Date.parse(manifest.capturedAt) > Date.now()) reject("api-baseline-binding-mismatch");
|
|
404
|
+
if (taskBaselineRetentionState(manifest) !== "active") reject("api-baseline-expired");
|
|
405
|
+
const roots = manifest.roots.filter(root => root.projectPath === "." || file.startsWith(`${safePath(root.projectPath)}/`));
|
|
406
|
+
if (roots.length !== 1 || !readableBaselineRoot(roots[0])) reject("api-baseline-scope-unavailable");
|
|
407
|
+
const root = roots[0], relative = root.projectPath === "." ? file : file.slice(root.projectPath.length + 1);
|
|
408
|
+
const entries = root.entries.filter(entry => decodeBaselineRepoPath(entry) === relative);
|
|
409
|
+
if (entries.length > 1) reject("api-baseline-scope-unavailable");
|
|
410
|
+
if (entries.length === 1) {
|
|
411
|
+
const entry = entries[0];
|
|
412
|
+
if (entry.state !== "blob" || !["100644", "100755"].includes(entry.mode)) reject("api-baseline-unavailable");
|
|
413
|
+
return readTaskBaselineBlob(cwd, task.taskRunId, entry.contentRef, MAX_FILE_BYTES).toString("utf8");
|
|
414
|
+
}
|
|
415
|
+
if (root.headState !== "head" || !root.headOid) reject("api-baseline-unavailable");
|
|
416
|
+
const rootPath = root.projectPath === "." ? cwd : path.join(cwd, safePath(root.projectPath));
|
|
417
|
+
if (fs.realpathSync(rootPath) !== rootPath) reject("api-baseline-scope-unavailable");
|
|
418
|
+
const git = args => execFileSync("git", ["--no-replace-objects", "--literal-pathspecs", "-c", "core.fsmonitor=false",
|
|
419
|
+
"--no-optional-locks", "-C", rootPath, ...args], { timeout: 1500, maxBuffer: MAX_FILE_BYTES + 1, stdio: ["ignore", "pipe", "ignore"] });
|
|
420
|
+
const entriesAtHead = git(["ls-tree", "-z", root.headOid, "--", relative]).toString("utf8").split("\0").filter(Boolean);
|
|
421
|
+
if (entriesAtHead.length !== 1) reject("api-baseline-unavailable");
|
|
422
|
+
const tree = /^(100644|100755) blob ([a-f0-9]{40,64})\t([\s\S]+)$/.exec(entriesAtHead[0]);
|
|
423
|
+
if (!tree || tree[3] !== relative) reject("api-baseline-unavailable");
|
|
424
|
+
return git(["cat-file", "blob", tree[2]]).toString("utf8");
|
|
425
|
+
}
|
|
426
|
+
|
|
427
|
+
/** Receipt integration boundary. This never executes the target module, uses
|
|
428
|
+
* current HEAD as task-start authority, or replaces behavioral criteria.
|
|
429
|
+
* Unknown and incompatible outcomes carry explicit reasons and no evidence.
|
|
430
|
+
*/
|
|
431
|
+
export function apiBaselineCriterionEvidence(input) {
|
|
432
|
+
const { task, criterion, corpus, currentWorkingTreeDigest } = input;
|
|
433
|
+
const index = task?.acceptanceReceipt?.criteria?.findIndex(item => item.id === criterion?.id) ?? -1;
|
|
434
|
+
const text = task?.acceptanceCriteria?.[index];
|
|
435
|
+
if (typeof text !== "string" || !isApiPreservationCriterion(text)) return { handled: false };
|
|
436
|
+
try {
|
|
437
|
+
if (task.changeMode !== "source-change"
|
|
438
|
+
|| !acceptanceCriterionBindingValid(task, task.acceptanceReceipt, criterion, index)) reject("api-criterion-binding-mismatch");
|
|
439
|
+
const cwd = fs.realpathSync(input.cwd);
|
|
440
|
+
const before = captureWorkspaceVerificationSnapshot(cwd);
|
|
441
|
+
if (!before.proofCapable || before.digest !== currentWorkingTreeDigest
|
|
442
|
+
|| (input.workspaceRevisionDigest !== undefined && input.workspaceRevisionDigest !== before.workspaceRevisionDigest)) reject("api-current-tree-mismatch");
|
|
443
|
+
if (!allConfiguredVerifierEvidenceCurrent(task, before.digest, before.workspaceRevisionDigest)) reject("api-current-verifier-missing");
|
|
444
|
+
const manifest = readTaskBaselineManifest(cwd, task.taskRunId);
|
|
445
|
+
if (!readableBaseline(manifest)) reject("api-baseline-unavailable");
|
|
446
|
+
if (manifest.taskId !== task.taskId || manifest.taskRunId !== task.taskRunId
|
|
447
|
+
|| manifest.sessionIdentityHash !== hash(task.sessionId) || manifest.capturedAt !== task.createdAt
|
|
448
|
+
|| manifest.baselineTreeDigest !== workingTreeEvidenceDigest(task.baselineFileDigests)
|
|
449
|
+
|| Date.parse(manifest.capturedAt) > Date.now()) reject("api-baseline-binding-mismatch");
|
|
450
|
+
if (taskBaselineRetentionState(manifest) !== "active") reject("api-baseline-expired");
|
|
451
|
+
const files = (corpus?.files ?? []).filter(file => !isAcceptanceTestPath(file)).map(safePath);
|
|
452
|
+
if (!files.length || files.length > MAX_FILES || new Set(files).size !== files.length) reject("api-source-scope-unavailable");
|
|
453
|
+
if (files.some(file => acceptanceLanguageAdapterForPath(file).disposition !== "supported"
|
|
454
|
+
|| !matchesAnyPath(file, task.scope ?? []) || matchesAnyPath(file, [...(task.protectedPaths ?? []), ...(task.outOfScope ?? [])]))) {
|
|
455
|
+
reject("api-source-scope-unavailable");
|
|
456
|
+
}
|
|
457
|
+
const inventories = [];
|
|
458
|
+
for (const file of files) {
|
|
459
|
+
const entries = (corpus.sourceEntries ?? []).filter(entry => safePath(entry.path) === file);
|
|
460
|
+
const current = readWorkspaceFile(cwd, file, MAX_FILE_BYTES).toString("utf8");
|
|
461
|
+
if (entries.length !== 1 || entries[0].text !== current) reject("api-source-corpus-mismatch");
|
|
462
|
+
let previous, next;
|
|
463
|
+
try {
|
|
464
|
+
previous = publicApi(readTaskBaselineSource(cwd, task, file), true);
|
|
465
|
+
next = publicApi(current, true);
|
|
466
|
+
} catch (error) { reject(error.apiProofReason ?? "api-contract-unsupported"); }
|
|
467
|
+
inventories.push({ file, baseline: previous, current: next });
|
|
468
|
+
}
|
|
469
|
+
const targeted = targetedExports(task, criterion, corpus.sourceEntries.filter(entry => files.includes(entry.path)), inventories);
|
|
470
|
+
const permittedDefaults = permittedDefaultInitializerChanges(task, targeted);
|
|
471
|
+
for (const item of inventories) {
|
|
472
|
+
const permittedNames = new Set([...permittedDefaults].filter(target => target.slice(0, target.lastIndexOf(":")) === item.file)
|
|
473
|
+
.map(target => target.slice(target.lastIndexOf(":") + 1)));
|
|
474
|
+
if (!samePublicContracts(item.baseline.contracts, item.current.contracts, permittedNames)) reject("api-contract-changed");
|
|
475
|
+
for (const [name, implementation] of item.baseline.implementations) {
|
|
476
|
+
if (!targeted.has(`${item.file}:${name}`) && implementation !== item.current.implementations.get(name)) {
|
|
477
|
+
reject("api-unrelated-implementation-changed");
|
|
478
|
+
}
|
|
479
|
+
}
|
|
480
|
+
}
|
|
481
|
+
const after = captureWorkspaceVerificationSnapshot(cwd);
|
|
482
|
+
if (!after.proofCapable || after.digest !== before.digest || after.workspaceRevisionDigest !== before.workspaceRevisionDigest) reject("api-current-tree-mismatch");
|
|
483
|
+
const command = task.verifyCommands[0], observed = latestObservedVerificationEvidence(task.verifyEvidence).get(command.trim());
|
|
484
|
+
return { handled: true, evidence: { kind: "task-baseline-public-api", summary:
|
|
485
|
+
"Task-start baseline preserves the supported public export, positional calling and primitive/local-array result contract, allowing only an operator-authorized initializer change when bound; all configured verifiers passed the current tree. Behavioral equivalence is not inferred.",
|
|
486
|
+
paths: files, command: observed.command, exitCode: 0, workingTreeDigest: before.digest } };
|
|
487
|
+
} catch (error) { return { handled: true, reason: error.apiProofReason ?? "api-baseline-unavailable" }; }
|
|
488
|
+
}
|
|
@@ -0,0 +1,22 @@
|
|
|
1
|
+
/** A lifecycle fence, not acceptance authority. The completion gate still runs. */
|
|
2
|
+
export type CompletionPreparation = Readonly<{
|
|
3
|
+
isCurrent: () => boolean;
|
|
4
|
+
completion?: "ready" | "deferred";
|
|
5
|
+
reason?: string;
|
|
6
|
+
}>;
|
|
7
|
+
|
|
8
|
+
export function completionPreparationDeferred(value: CompletionPreparation | boolean | void): boolean {
|
|
9
|
+
try { return value !== null && typeof value === "object" && value.completion === "deferred" && value.isCurrent() === true; }
|
|
10
|
+
catch { return false; }
|
|
11
|
+
}
|
|
12
|
+
|
|
13
|
+
export function completionPreparationCurrent(value: CompletionPreparation | boolean | void): boolean {
|
|
14
|
+
if (value === undefined || value === true) return true;
|
|
15
|
+
try {
|
|
16
|
+
return value !== null && typeof value === "object" && value.completion !== "deferred"
|
|
17
|
+
&& typeof value.isCurrent === "function" && value.isCurrent() === true;
|
|
18
|
+
} catch {
|
|
19
|
+
// A disposed host context may throw instead of returning a stale identity.
|
|
20
|
+
return false;
|
|
21
|
+
}
|
|
22
|
+
}
|
|
@@ -0,0 +1,105 @@
|
|
|
1
|
+
import type { ExtensionAPI, ExtensionContext } from "@earendil-works/pi-coding-agent";
|
|
2
|
+
import type { TaskContract } from "../../extensions/guard-types.js";
|
|
3
|
+
import { hasDurableContextEvidence } from "../../extensions/context-evidence.js";
|
|
4
|
+
import { applyAssistedReadOnlyFinalHandoff, runtimeLifecycleMode,
|
|
5
|
+
workingTreeEvidenceDigest } from "../../extensions/task-lifecycle.js";
|
|
6
|
+
import { recordCompletionAudit } from "../../extensions/task-runtime-audit.js";
|
|
7
|
+
import { workingTreeSnapshot } from "../../extensions/task-state.js";
|
|
8
|
+
import { taskDeltaFilesFromSnapshot } from "../../extensions/task-contract-view.js";
|
|
9
|
+
import { evaluateExactFinalOutputContract } from "../quality/exact-output-contract.ts";
|
|
10
|
+
import { taskPerformanceAssurance } from "../quality/performance-assurance.ts";
|
|
11
|
+
import type { RuntimeSessionState } from "../session/runtime-state.ts";
|
|
12
|
+
import { observeTrajectorySync } from "../trajectory/trajectory-observability.ts";
|
|
13
|
+
import type { TrajectorySyncOptions, TrajectorySyncResult } from "../trajectory/trajectory-runtime.ts";
|
|
14
|
+
import { compositeTaskPublicationDigest } from "./composite-task-publication.ts";
|
|
15
|
+
|
|
16
|
+
type Gate = { decision: "pass" | "fail"; missing: string[]; missingVerifyCommands: string[] };
|
|
17
|
+
type Projection = { task: TaskContract; projected: TaskContract; gate: Gate;
|
|
18
|
+
currentDigests: Record<string, string>; currentDigest: string };
|
|
19
|
+
type Options = { ctx: ExtensionContext; pi: ExtensionAPI; state: RuntimeSessionState; task: TaskContract; responseText: string;
|
|
20
|
+
activeTask: (ctx: ExtensionContext) => TaskContract | undefined;
|
|
21
|
+
completionProjection: (cwd: string, task: TaskContract, digests: Record<string, string>) => TaskContract;
|
|
22
|
+
evaluateGate: (cwd: string, task: TaskContract, digests: Record<string, string>, digest: string) => Gate;
|
|
23
|
+
semanticReviewAllowed: (task: TaskContract) => boolean;
|
|
24
|
+
syncTrajectory?: (ctx: ExtensionContext, task: TaskContract, options: TrajectorySyncOptions) => TrajectorySyncResult;
|
|
25
|
+
withRecoveryProvenance: (task: TaskContract, gate: Gate, digests: Record<string, string>) => TaskContract;
|
|
26
|
+
projectCompositeLifecycle?: (task: TaskContract) => TaskContract | false;
|
|
27
|
+
writeTask: (cwd: string, task: TaskContract) => TaskContract; activateBaseTools: (ctx: ExtensionContext) => unknown;
|
|
28
|
+
appendTrace: (cwd: string, payload: Record<string, unknown>) => void;
|
|
29
|
+
appendSessionTrace: (pi: ExtensionAPI, payload: Record<string, unknown>) => void;
|
|
30
|
+
telemetry: (ctx: ExtensionContext, payload: Record<string, unknown>) => void;
|
|
31
|
+
persistHandoff: (task: TaskContract, gate: Gate, digests: Record<string, string>) => void };
|
|
32
|
+
|
|
33
|
+
function exactPathCoverage(expectedPaths: string[], reviewedPaths: string[] | undefined): boolean {
|
|
34
|
+
const expected = [...new Set(expectedPaths)].sort(), reviewed = [...new Set(reviewedPaths ?? [])].sort();
|
|
35
|
+
return expected.length > 0 && expected.length === reviewed.length && expected.every((file, index) => file === reviewed[index]);
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
/** Rechecks the ordinary gate without emitting a turn or synthesizing output. */
|
|
39
|
+
export function createCompositeCompletionFinalizer(options: Options) {
|
|
40
|
+
const issued = new WeakMap<object, Projection>();
|
|
41
|
+
function reviewReady(task: TaskContract, digest: string, digests: Record<string, string>) {
|
|
42
|
+
const assurance = taskPerformanceAssurance(task); if (!options.semanticReviewAllowed(task) || !assurance.requiresReview) return true;
|
|
43
|
+
const expected = taskDeltaFilesFromSnapshot(task, digests), review = options.state.performanceReviewCheckpoint(task.taskRunId);
|
|
44
|
+
const checkpoint = review?.workingTreeDigest === digest && review.reviewSatisfied && !review.invalidated
|
|
45
|
+
&& exactPathCoverage(expected, review.expectedPaths) && exactPathCoverage(expected, review.reviewedPaths);
|
|
46
|
+
const credit = options.state.performanceReviewCredit(task.taskRunId, digest);
|
|
47
|
+
return Boolean(checkpoint || credit && exactPathCoverage(expected, credit.reviewedPaths));
|
|
48
|
+
}
|
|
49
|
+
function project(): Projection | false {
|
|
50
|
+
const task = options.activeTask(options.ctx);
|
|
51
|
+
if (!task || task.taskRunId !== options.task.taskRunId || task.sessionId !== options.task.sessionId
|
|
52
|
+
|| task.trace.outcome !== "pending") return false;
|
|
53
|
+
const currentDigests = workingTreeSnapshot(options.ctx.cwd) as Record<string, string>;
|
|
54
|
+
const currentDigest = workingTreeEvidenceDigest(currentDigests); let candidate = structuredClone(task);
|
|
55
|
+
if (options.projectCompositeLifecycle) {
|
|
56
|
+
const projectedLifecycle = options.projectCompositeLifecycle(candidate); if (!projectedLifecycle) return false;
|
|
57
|
+
candidate = projectedLifecycle;
|
|
58
|
+
}
|
|
59
|
+
const readOnly = (candidate.changeMode === "read-only" || candidate.mutationPolicy === "forbidden")
|
|
60
|
+
&& hasDurableContextEvidence(candidate);
|
|
61
|
+
applyAssistedReadOnlyFinalHandoff(candidate, { durableReadEvidence: readOnly, finalHandoffObserved: true });
|
|
62
|
+
const projected = options.completionProjection(options.ctx.cwd, candidate, currentDigests);
|
|
63
|
+
const base = options.evaluateGate(options.ctx.cwd, projected, currentDigests, currentDigest);
|
|
64
|
+
const exact = evaluateExactFinalOutputContract(projected, options.responseText, options.ctx.cwd);
|
|
65
|
+
const gate = exact.applicable && !exact.passed ? { ...base, decision: "fail" as const,
|
|
66
|
+
missing: [...base.missing, `exact final output contract (${exact.key}=<value> must copy the complete observed value verbatim as the last non-empty line)`] } : base;
|
|
67
|
+
return gate.decision === "pass" && reviewReady(projected, currentDigest, currentDigests)
|
|
68
|
+
? { task, projected: options.withRecoveryProvenance(projected, gate, currentDigests), gate, currentDigests, currentDigest } : false;
|
|
69
|
+
}
|
|
70
|
+
function publish(value: Projection): boolean {
|
|
71
|
+
let { task, projected } = value; const { gate, currentDigests, currentDigest } = value;
|
|
72
|
+
const expected = taskDeltaFilesFromSnapshot(projected, currentDigests), review = options.state.performanceReviewCheckpoint(task.taskRunId);
|
|
73
|
+
const credit = options.state.performanceReviewCredit(task.taskRunId, currentDigest);
|
|
74
|
+
if (options.semanticReviewAllowed(task) && taskPerformanceAssurance(projected).requiresReview && credit
|
|
75
|
+
&& exactPathCoverage(expected, credit.reviewedPaths) && !(review?.workingTreeDigest === currentDigest && review.reviewSatisfied
|
|
76
|
+
&& !review.invalidated && exactPathCoverage(expected, review.expectedPaths) && exactPathCoverage(expected, review.reviewedPaths))) {
|
|
77
|
+
const reused = { event: "performance_review_credit_reused", taskId: task.taskId, taskRunId: task.taskRunId,
|
|
78
|
+
sessionId: task.sessionId, workingTreeDigest: credit.workingTreeDigest, commandHash: credit.commandHash,
|
|
79
|
+
reviewedPaths: credit.reviewedPaths, reviewedAt: credit.recordedAt, reasonCodes: taskPerformanceAssurance(projected).reasonCodes };
|
|
80
|
+
options.appendTrace(options.ctx.cwd, reused); options.appendSessionTrace(options.pi, reused); options.telemetry(options.ctx, reused);
|
|
81
|
+
}
|
|
82
|
+
observeTrajectorySync(options.ctx, options.syncTrajectory?.(options.ctx, task, { sourceHook: "completion", handoffObserved: true }), options.telemetry);
|
|
83
|
+
task = options.writeTask(options.ctx.cwd, projected);
|
|
84
|
+
observeTrajectorySync(options.ctx, options.syncTrajectory?.(options.ctx, task, { sourceHook: "completion" }), options.telemetry);
|
|
85
|
+
options.state.clearPerformanceReview(task.taskRunId); options.state.clearObservedContext(options.ctx); options.activateBaseTools(options.ctx);
|
|
86
|
+
recordCompletionAudit(options.ctx, task, { outcome: "completed", evidence: { changedFiles: task.changedFiles,
|
|
87
|
+
lifecycleMode: runtimeLifecycleMode(task) } }); options.persistHandoff(task, gate, currentDigests);
|
|
88
|
+
const trace = { event: "task_auto_completed", taskId: task.taskId, taskRunId: task.taskRunId, sessionId: task.sessionId,
|
|
89
|
+
changedFiles: task.changedFiles, lifecycleMode: runtimeLifecycleMode(task) };
|
|
90
|
+
options.appendTrace(options.ctx.cwd, trace); options.appendSessionTrace(options.pi, trace); options.telemetry(options.ctx, trace); return true;
|
|
91
|
+
}
|
|
92
|
+
function current(value: Projection) {
|
|
93
|
+
const observed = project(); return observed && observed.currentDigest === value.currentDigest
|
|
94
|
+
&& compositeTaskPublicationDigest(observed.task) === compositeTaskPublicationDigest(value.task);
|
|
95
|
+
}
|
|
96
|
+
return Object.freeze({ preflight(): object | false { const value = project(); if (!value) return false;
|
|
97
|
+
const capability = Object.freeze({ version: "composite-completion-preflight-v1", taskRunId: value.task.taskRunId,
|
|
98
|
+
workingTreeDigest: value.currentDigest }); issued.set(capability, value); return capability; },
|
|
99
|
+
publication(capability: object): object | false { const value = issued.get(capability); if (!value || !current(value)) return false;
|
|
100
|
+
return Object.freeze({ version: "composite-terminal-task-target-v1", pendingTask: structuredClone(value.task),
|
|
101
|
+
terminalTask: structuredClone(value.projected), pendingTaskDigest: compositeTaskPublicationDigest(value.task),
|
|
102
|
+
terminalTaskDigest: compositeTaskPublicationDigest(value.projected), workingTreeDigest: value.currentDigest }); },
|
|
103
|
+
finalize(capability: object): boolean { const value = issued.get(capability); if (!value || !issued.delete(capability) || !current(value)) return false;
|
|
104
|
+
return publish(value); } });
|
|
105
|
+
}
|