@piagent/platform 1.6.0 → 1.7.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +104 -0
- package/README.md +24 -15
- package/README.vi.md +36 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +27 -3
- package/benchmarks/production-v1/suite.json +5 -1
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +16 -16
- package/docs/authority-manifest.md +1 -1
- package/docs/auto-delegation-policy.md +38 -73
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +30 -35
- package/docs/context-engine.md +5 -3
- package/docs/context-window-policy.md +37 -1
- package/docs/distribution-standard.md +9 -9
- package/docs/en/architecture.md +26 -13
- package/docs/languages.json +4 -4
- package/docs/operator-manual-vietnamese.md +26 -46
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +7 -0
- package/docs/quality-benchmark.md +299 -23
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +32 -9
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/runtime-harness-standard.md +3 -3
- package/docs/security-threat-model.md +1 -1
- package/docs/subagent-orchestration-capabilities.md +19 -51
- package/docs/subagents-and-multiagent.md +21 -47
- package/docs/task-implementation-contract.md +3 -3
- package/docs/task-lifecycle-vietnamese.md +1 -1
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs/vi/architecture.md +26 -13
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +89 -95
- package/docs-site/commands.html +1 -1
- package/docs-site/content/benchmark.html +88 -94
- package/docs-site/content/en/benchmark.html +134 -90
- package/docs-site/content/en/index.html +19 -19
- package/docs-site/content/en/runtime.html +5 -5
- package/docs-site/content/en/whats-new.html +23 -1
- package/docs-site/content/en/workflows.html +2 -2
- package/docs-site/content/index.html +19 -19
- package/docs-site/content/runtime.html +6 -6
- package/docs-site/content/whats-new.html +23 -1
- package/docs-site/content/workflows.html +2 -2
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +135 -91
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +20 -20
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +6 -6
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -2
- package/docs-site/en/workflows.html +3 -3
- package/docs-site/index.html +20 -20
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +7 -7
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -2
- package/docs-site/workflows.html +3 -3
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
- package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
- package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/helper-request.invalid.json +1 -1
- package/evals/fixtures/helper-request.valid.json +1 -1
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
- package/evals/fs4-readiness-matrix.v1.json +2 -2
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +10 -6
- package/evals/fs5-pilot-protocol.v1.json +3 -3
- package/evals/fs5-pilot-protocol.v2.json +8 -8
- package/evals/fs5-pilot-protocol.v3.json +8 -8
- package/evals/fs5-pilot-protocol.v4.json +7 -7
- package/evals/fs5-pilot-protocol.v5.json +10 -8
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/evals/runtime-conformance-v1/README.md +43 -0
- package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
- package/evals/runtime-conformance-v1/lane.json +57 -0
- package/evals/runtime-conformance-v1/paid-canary.json +31 -0
- package/evals/runtime-conformance-v1/runner.mjs +456 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +15 -6
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +12 -17
- package/packages/piagent-core/extensions/context-index-policy.js +4 -0
- package/packages/piagent-core/extensions/context-query-paths.js +57 -0
- package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
- package/packages/piagent-core/extensions/criterion-graph.js +4 -1
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
- package/packages/piagent-core/extensions/guard-types.ts +9 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +140 -43
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +43 -2
- package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
- package/packages/piagent-core/prompts/be-to-fe.md +2 -8
- package/packages/piagent-core/prompts/plan.md +1 -1
- package/packages/piagent-core/prompts/platform-improve.md +2 -6
- package/packages/piagent-core/prompts/review.md +2 -7
- package/packages/piagent-core/prompts/scout.md +1 -1
- package/packages/piagent-core/prompts/task.md +4 -3
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
- package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
- package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
- package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
- package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
- package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
- package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
- package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
- package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
- package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
- package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
- package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
- package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
- package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
- package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
- package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
- package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
- package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
- package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
- package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
- package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
- package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
- package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
- package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
- package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
- package/packages/piagent-core/subagents/piagent-planner.md +6 -4
- package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
- package/packages/piagent-core/subagents/piagent-scout.md +4 -2
- package/packages/piagent-core/subagents/piagent-worker.md +6 -4
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
- package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
- package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
- package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
- package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
- package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +150 -11
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/project-profile.schema.json +1 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +45 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +224 -387
- package/scripts/benchmark-runner-finalization.mjs +144 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-subagents.sh +42 -34
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/helper-evaluation.mjs +1 -1
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/team-doctor.sh +1 -1
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +41 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
|
@@ -0,0 +1,126 @@
|
|
|
1
|
+
import { parse } from "@babel/parser";
|
|
2
|
+
import { protocolShape as shape, validateValue, numberValue } from "./acceptance-executor/values.mjs";
|
|
3
|
+
|
|
4
|
+
export const DECLARED_API_VERSION = "declared-positional-api-v1";
|
|
5
|
+
const NAME = /^[a-zA-Z_$][\w$]{0,127}$/;
|
|
6
|
+
const fail = () => { throw new TypeError("declared-api-unsupported"); };
|
|
7
|
+
|
|
8
|
+
/** Explicit host expectation for declarations only. Parameter names and function
|
|
9
|
+
* bodies are not API proofs; behavior/default results require the other cases.
|
|
10
|
+
* This comparator never executes source or reads a model-supplied expectation.
|
|
11
|
+
*/
|
|
12
|
+
export function validateDeclaredApi(value) {
|
|
13
|
+
shape(value, ["version", "exports"]);
|
|
14
|
+
if (value.version !== DECLARED_API_VERSION || !Array.isArray(value.exports) || value.exports.length < 1 || value.exports.length > 32) fail();
|
|
15
|
+
let previous = "";
|
|
16
|
+
for (const entry of value.exports) {
|
|
17
|
+
if (typeof entry.name !== "string" || !NAME.test(entry.name) || entry.name <= previous) fail(); previous = entry.name;
|
|
18
|
+
if (entry.kind === "function") {
|
|
19
|
+
shape(entry, ["name", "kind", "async", "generator", "parameters"]);
|
|
20
|
+
if (typeof entry.async !== "boolean" || typeof entry.generator !== "boolean" || !Array.isArray(entry.parameters) || entry.parameters.length > 16) fail();
|
|
21
|
+
for (const parameter of entry.parameters) {
|
|
22
|
+
if (parameter === "required") continue;
|
|
23
|
+
if (parameter?.kind === "default-binding") {
|
|
24
|
+
shape(parameter, ["kind", "name", "value", "frozen"]); if (typeof parameter.name !== "string" || !NAME.test(parameter.name) || typeof parameter.frozen !== "boolean") fail();
|
|
25
|
+
validateValue(parameter.value, false);
|
|
26
|
+
} else {
|
|
27
|
+
shape(parameter, ["kind", "value"]); if (parameter.kind !== "default") fail();
|
|
28
|
+
validateValue(parameter.value, false);
|
|
29
|
+
}
|
|
30
|
+
}
|
|
31
|
+
} else {
|
|
32
|
+
shape(entry, ["name", "kind", "value", "frozen"]);
|
|
33
|
+
if (entry.kind !== "const-value" || typeof entry.frozen !== "boolean") fail();
|
|
34
|
+
validateValue(entry.value, false);
|
|
35
|
+
}
|
|
36
|
+
}
|
|
37
|
+
if (JSON.stringify(value).length > 65536) fail();
|
|
38
|
+
return value;
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
/** Closed, bounded declaration inspection, not a return-type or equivalence
|
|
42
|
+
* prover. Dynamic/re-exported APIs, destructuring, rest parameters and arbitrary
|
|
43
|
+
* initializer execution remain unknown. Const values must be inert literals.
|
|
44
|
+
*/
|
|
45
|
+
export function observeDeclaredApi(source) {
|
|
46
|
+
try {
|
|
47
|
+
if (typeof source !== "string" || Buffer.byteLength(source) > 64000) fail();
|
|
48
|
+
const ast = parse(source, { sourceType: "module", errorRecovery: false, attachComment: false });
|
|
49
|
+
const bindings = new Map(), exported = new Map(); let nodes = 0;
|
|
50
|
+
function visit(node, depth = 0) {
|
|
51
|
+
if (!node || typeof node.type !== "string") return;
|
|
52
|
+
if (++nodes > 12000 || depth > 128) fail();
|
|
53
|
+
// Conservative refusal of rebinding a public name, including ambiguous
|
|
54
|
+
// shadowing. Do not certify a declaration that source later replaces.
|
|
55
|
+
const target = node.type === "AssignmentExpression" ? node.left : node.type === "UpdateExpression" ? node.argument : null;
|
|
56
|
+
if (target?.type === "Identifier" && [...exported.values()].includes(target.name)) fail();
|
|
57
|
+
for (const [key, value] of Object.entries(node)) {
|
|
58
|
+
if (["loc", "extra", "comments", "tokens", "errors"].includes(key)) continue;
|
|
59
|
+
if (Array.isArray(value)) value.forEach(item => visit(item, depth + 1));
|
|
60
|
+
else if (value && typeof value === "object") visit(value, depth + 1);
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
for (const statement of ast.program.body) {
|
|
64
|
+
const isExport = statement.type === "ExportNamedDeclaration";
|
|
65
|
+
if (isExport && statement.source) fail();
|
|
66
|
+
const node = isExport ? statement.declaration : statement;
|
|
67
|
+
if (isExport && !node) {
|
|
68
|
+
for (const item of statement.specifiers) {
|
|
69
|
+
if (item.type !== "ExportSpecifier" || item.local.type !== "Identifier" || item.exported.type !== "Identifier" || exported.has(item.exported.name)) fail();
|
|
70
|
+
exported.set(item.exported.name, item.local.name);
|
|
71
|
+
}
|
|
72
|
+
continue;
|
|
73
|
+
}
|
|
74
|
+
const declarations = node?.type === "FunctionDeclaration" ? [[node.id?.name, node]]
|
|
75
|
+
: node?.type === "VariableDeclaration" && node.kind === "const" ? node.declarations.map(item => [item.id.type === "Identifier" ? item.id.name : null, item]) : null;
|
|
76
|
+
if (!declarations) fail();
|
|
77
|
+
for (const [name, declaration] of declarations) {
|
|
78
|
+
if (!name || bindings.has(name)) fail(); bindings.set(name, declaration);
|
|
79
|
+
if (isExport) { if (exported.has(name)) fail(); exported.set(name, name); }
|
|
80
|
+
}
|
|
81
|
+
}
|
|
82
|
+
visit(ast);
|
|
83
|
+
const literal = node => {
|
|
84
|
+
if (node?.type === "NullLiteral") return { type: "null" };
|
|
85
|
+
if (node?.type === "Identifier" && node.name === "undefined" && !bindings.has("undefined")) return { type: "undefined" };
|
|
86
|
+
if (node?.type === "NumericLiteral") return { type: "number", value: numberValue(node.value) };
|
|
87
|
+
if (node?.type === "StringLiteral") return { type: "string", value: node.value };
|
|
88
|
+
if (node?.type === "BooleanLiteral") return { type: "boolean", value: node.value };
|
|
89
|
+
if (node?.type === "UnaryExpression" && node.operator === "-" && node.argument.type === "NumericLiteral") return { type: "number", value: numberValue(-node.argument.value) };
|
|
90
|
+
if (node?.type === "ArrayExpression") return { type: "array", value: node.elements.map(literal) };
|
|
91
|
+
if (node?.type === "ObjectExpression") {
|
|
92
|
+
const keys = new Set();
|
|
93
|
+
return { type: "record", value: node.properties.map(item => {
|
|
94
|
+
if (item.type !== "ObjectProperty" || item.computed || item.method || item.shorthand) fail();
|
|
95
|
+
const key = item.key.type === "Identifier" ? item.key.name : item.key.type === "StringLiteral" ? item.key.value : null;
|
|
96
|
+
if (key === null || key === "__proto__" || keys.has(key)) fail(); keys.add(key);
|
|
97
|
+
return { key, value: literal(item.value) };
|
|
98
|
+
}) };
|
|
99
|
+
}
|
|
100
|
+
return fail();
|
|
101
|
+
};
|
|
102
|
+
const constValue = node => {
|
|
103
|
+
if (node?.type !== "VariableDeclarator") fail();
|
|
104
|
+
let value = node.init, frozen = false;
|
|
105
|
+
if (value?.type === "CallExpression" && !value.optional && value.arguments.length === 1 && value.callee.type === "MemberExpression"
|
|
106
|
+
&& !value.callee.computed && value.callee.object.type === "Identifier" && value.callee.object.name === "Object"
|
|
107
|
+
&& value.callee.property.name === "freeze" && !bindings.has("Object")) { value = value.arguments[0]; frozen = true; }
|
|
108
|
+
return { value: literal(value), frozen };
|
|
109
|
+
};
|
|
110
|
+
const entries = [...exported].map(([name, local]) => {
|
|
111
|
+
const node = bindings.get(local); if (!node) fail();
|
|
112
|
+
if (node.type !== "FunctionDeclaration") return { name, kind: "const-value", ...constValue(node) };
|
|
113
|
+
const parameters = node.params.map(parameter => {
|
|
114
|
+
if (parameter.type === "Identifier") return "required";
|
|
115
|
+
if (parameter.type !== "AssignmentPattern" || parameter.left.type !== "Identifier") fail();
|
|
116
|
+
const right = parameter.right;
|
|
117
|
+
if (right.type === "Identifier" && bindings.has(right.name)) {
|
|
118
|
+
return { kind: "default-binding", name: right.name, ...constValue(bindings.get(right.name)) };
|
|
119
|
+
}
|
|
120
|
+
return { kind: "default", value: literal(right) };
|
|
121
|
+
});
|
|
122
|
+
return { name, kind: "function", async: node.async, generator: node.generator, parameters };
|
|
123
|
+
}).sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0);
|
|
124
|
+
return { status: "observed", value: validateDeclaredApi({ version: DECLARED_API_VERSION, exports: entries }) };
|
|
125
|
+
} catch { return { status: "unsupported", reason: "declared-api-unsupported" }; }
|
|
126
|
+
}
|
|
@@ -0,0 +1,14 @@
|
|
|
1
|
+
import type { BasePolicy, FinalGateConfig } from "./guard-types.ts";
|
|
2
|
+
|
|
3
|
+
export function finalGateConfig(policy: BasePolicy, fallback: BasePolicy): Required<FinalGateConfig> {
|
|
4
|
+
const diagnostic = policy.finalGate?.acceptanceProofMode === "diagnostic";
|
|
5
|
+
return {
|
|
6
|
+
acceptanceProofMode: diagnostic ? "diagnostic" : "enforce",
|
|
7
|
+
defaultMode: policy.finalGate?.defaultMode ?? fallback.finalGate?.defaultMode ?? "enforce",
|
|
8
|
+
requireTaskContract: diagnostic || (policy.finalGate?.requireTaskContract ?? true),
|
|
9
|
+
requireContextManifest: diagnostic || (policy.finalGate?.requireContextManifest ?? true),
|
|
10
|
+
requireVerifyEvidence: diagnostic || (policy.finalGate?.requireVerifyEvidence ?? true),
|
|
11
|
+
requireTrace: diagnostic || (policy.finalGate?.requireTrace ?? true),
|
|
12
|
+
requirePassingVerify: diagnostic || (policy.finalGate?.requirePassingVerify ?? true)
|
|
13
|
+
};
|
|
14
|
+
}
|
|
@@ -0,0 +1,73 @@
|
|
|
1
|
+
import crypto from 'node:crypto';
|
|
2
|
+
import { acceptanceCriterionBindingValid } from './acceptance-behavior-proof.js';
|
|
3
|
+
import { durableContextEvidenceEntries } from './context-evidence.js';
|
|
4
|
+
import { matchesAnyPath } from './policy-core.js';
|
|
5
|
+
import { captureWorkspaceVerificationSnapshot } from './workspace-verification-snapshot.js';
|
|
6
|
+
import { readWorkspaceFile } from '../security/workspace-file-reader.ts';
|
|
7
|
+
|
|
8
|
+
const sha = value => crypto.createHash('sha256').update(value).digest('hex');
|
|
9
|
+
const normalize = value => value.replace(/\s+/g, ' ').trim();
|
|
10
|
+
const safePath = value => /^[A-Za-z0-9_./-]{1,240}$/.test(value)
|
|
11
|
+
&& value.split('/').every(part => part && part !== '.' && part !== '..' && !part.startsWith('.'));
|
|
12
|
+
|
|
13
|
+
// Only operator-bound literal-copy requests qualify. Data values never supply
|
|
14
|
+
// paths, commands, permission or new obligations to this reader.
|
|
15
|
+
export function documentContentRule(request) {
|
|
16
|
+
if (typeof request !== 'string' || request.length > 16000) return null;
|
|
17
|
+
const text = normalize(request);
|
|
18
|
+
const update = [...text.matchAll(/(?:^|[.!?] )Update `([^`]+)` with the ([A-Za-z][A-Za-z ,_-]{1,140}) from `([^`]+)`\./g)];
|
|
19
|
+
const scope = [...text.matchAll(/(?:^|[.!?] )Change only `([^`]+)`, include (both|all) configuration values verbatim, and run the configured verification\./g)];
|
|
20
|
+
if (update.length !== 1 || scope.length !== 1) return null;
|
|
21
|
+
const [, documentPath, names, configPath] = update[0];
|
|
22
|
+
if (scope[0][1] !== documentPath || !safePath(documentPath) || !safePath(configPath)
|
|
23
|
+
|| !/\.(?:md|txt)$/.test(documentPath) || !/\.json$/.test(configPath)) return null;
|
|
24
|
+
const labels = names.split(/,\s*(?:and\s+)?|\s+and\s+/).map(value => value.trim());
|
|
25
|
+
if (labels.length < 2 || labels.length > 8 || labels.some(value => !/^[A-Za-z][A-Za-z _-]{0,50}$/.test(value))
|
|
26
|
+
|| new Set(labels).size !== labels.length || scope[0][2] === 'both' && labels.length !== 2) return null;
|
|
27
|
+
return { documentPath, configPath, labels, clauses: [update[0][0].replace(/^[.!?] /, ''), scope[0][0].replace(/^[.!?] /, '')] };
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
export function documentContentCriterionEvidence({task, criterion, corpus, cwd, passingVerifier, verifierEvidence, currentWorkingTreeDigest, workspaceRevisionDigest}) {
|
|
31
|
+
const rule = documentContentRule(task?.operatorRequest);
|
|
32
|
+
if (!rule) return {handled: false};
|
|
33
|
+
const index = task.acceptanceReceipt?.criteria.findIndex(item => item.id === criterion?.id) ?? -1;
|
|
34
|
+
const selected = task.acceptanceCriteria?.[index];
|
|
35
|
+
if (typeof selected !== 'string' || !rule.clauses.includes(normalize(selected))) return {handled: false};
|
|
36
|
+
const missing = {handled: true};
|
|
37
|
+
if (task.operatorRequestDigest !== `operator-request-v1:${sha(task.operatorRequest)}`
|
|
38
|
+
|| !acceptanceCriterionBindingValid(task, task.acceptanceReceipt, criterion, index)
|
|
39
|
+
|| task.changeMode !== 'source-change' || !passingVerifier || !verifierEvidence || !cwd
|
|
40
|
+
|| verifierEvidence.workingTreeDigest !== currentWorkingTreeDigest || verifierEvidence.exitCode !== 0
|
|
41
|
+
|| corpus.files.length !== 1 || corpus.files[0] !== rule.documentPath) return missing;
|
|
42
|
+
const forbidden = [...(task.protectedPaths ?? []), ...(task.outOfScope ?? [])];
|
|
43
|
+
if ([rule.documentPath, rule.configPath].some(file => matchesAnyPath(file, forbidden))) return missing;
|
|
44
|
+
if (!durableContextEvidenceEntries(task).some(entry => entry.path === rule.configPath)) return missing;
|
|
45
|
+
try {
|
|
46
|
+
const before = captureWorkspaceVerificationSnapshot(cwd);
|
|
47
|
+
if (!before.proofCapable || before.digest !== currentWorkingTreeDigest
|
|
48
|
+
|| before.workspaceRevisionDigest !== workspaceRevisionDigest) return missing;
|
|
49
|
+
const configBytes = readWorkspaceFile(cwd, rule.configPath, 65536);
|
|
50
|
+
const documentBytes = readWorkspaceFile(cwd, rule.documentPath, 262144);
|
|
51
|
+
const decode = bytes => new TextDecoder('utf-8', {fatal: true}).decode(bytes);
|
|
52
|
+
const config = JSON.parse(decode(configBytes)), document = decode(documentBytes);
|
|
53
|
+
if (!config || typeof config !== 'object' || Array.isArray(config)) return missing;
|
|
54
|
+
const keyName = value => value.replace(/[ _-]/g, '').toLowerCase();
|
|
55
|
+
const keys = rule.labels.map(label => {
|
|
56
|
+
const labels = new Set([keyName(label), keyName(label.replace(/ name$/i, ''))]);
|
|
57
|
+
const candidates = Object.keys(config).filter(key => labels.has(keyName(key)));
|
|
58
|
+
return candidates.length === 1 ? candidates[0] : null;
|
|
59
|
+
});
|
|
60
|
+
if (keys.some(key => key === null) || new Set(keys).size !== keys.length) return missing;
|
|
61
|
+
if (keys.some(key => typeof config[key] !== 'string' || !config[key].trim() || !document.includes(config[key]))) return missing;
|
|
62
|
+
// Recheck both reads before publishing their combined content binding.
|
|
63
|
+
if (!configBytes.equals(readWorkspaceFile(cwd, rule.configPath, 65536))
|
|
64
|
+
|| !documentBytes.equals(readWorkspaceFile(cwd, rule.documentPath, 262144))) return missing;
|
|
65
|
+
const after = captureWorkspaceVerificationSnapshot(cwd);
|
|
66
|
+
if (!after.proofCapable || before.digest !== after.digest || before.workspaceRevisionDigest !== after.workspaceRevisionDigest) return missing;
|
|
67
|
+
const binding = sha(JSON.stringify({version: 1, request: task.operatorRequestDigest,
|
|
68
|
+
criterion: criterion.hash, config: sha(configBytes), document: sha(documentBytes), keys}));
|
|
69
|
+
return {handled: true, evidence: {...verifierEvidence, kind: 'document-content-proof',
|
|
70
|
+
summary: `Operator-requested configuration values occur verbatim in the sole changed document; content binding sha256:${binding}.`,
|
|
71
|
+
paths: [rule.documentPath, rule.configPath]}};
|
|
72
|
+
} catch { return missing; }
|
|
73
|
+
}
|
|
@@ -0,0 +1,59 @@
|
|
|
1
|
+
import type { openAcceptanceEvidenceStore } from "./acceptance-evidence-store.js";
|
|
2
|
+
import type { captureExecutionSnapshot } from "./acceptance-execution-snapshot.js";
|
|
3
|
+
import type { AuthenticatedAssessment } from "./acceptance-authenticated-admission.js";
|
|
4
|
+
|
|
5
|
+
export const DURABLE_EXECUTION_VERSION: "durable-module-contract-v2";
|
|
6
|
+
export type LegacyDurableContractScope = { taskRunId: string; criterionId: string };
|
|
7
|
+
export type FactDurableContractScope = { version: "authenticated-contract-fact-scope-v2";
|
|
8
|
+
taskRunId: string; criterionId: string; factId: string };
|
|
9
|
+
export type DurableContractScope = LegacyDurableContractScope | FactDurableContractScope;
|
|
10
|
+
export type ProjectVerificationRequest = DurableContractScope & {
|
|
11
|
+
criterionHash: string;
|
|
12
|
+
snapshot: ReturnType<typeof captureExecutionSnapshot>["binding"];
|
|
13
|
+
snapshotDigest: string;
|
|
14
|
+
};
|
|
15
|
+
export type DurableContractConfiguration = {
|
|
16
|
+
store: ReturnType<typeof openAcceptanceEvidenceStore>;
|
|
17
|
+
projectRoot: string;
|
|
18
|
+
sourcePath: string;
|
|
19
|
+
modulePaths?: readonly string[];
|
|
20
|
+
authorizeSourceRead: (input: { projectRoot: string; sourcePath: string }) => boolean;
|
|
21
|
+
exportName: string;
|
|
22
|
+
checks: unknown[];
|
|
23
|
+
profile?: Readonly<{ id: string; digest: string; workerVersion: string }>;
|
|
24
|
+
imageId: string;
|
|
25
|
+
dockerSocket: string;
|
|
26
|
+
dockerCommand?: Readonly<{ path: string; sha256: string }>;
|
|
27
|
+
verifierDigest: string;
|
|
28
|
+
getProjectVerificationDigest: (input: ProjectVerificationRequest) => string | null | Promise<string | null>;
|
|
29
|
+
timeoutMs?: number;
|
|
30
|
+
/** Extra attached-command wait; guest watchdog and CPU/wall budgets are unchanged. */
|
|
31
|
+
startupAllowanceMs?: number;
|
|
32
|
+
};
|
|
33
|
+
export type DurableContractResult = Readonly<{
|
|
34
|
+
version: typeof DURABLE_EXECUTION_VERSION;
|
|
35
|
+
verdict: "pass" | "fail" | "unknown" | "error";
|
|
36
|
+
completionAllowed: false;
|
|
37
|
+
reason?: string;
|
|
38
|
+
attemptId?: string;
|
|
39
|
+
reused?: boolean;
|
|
40
|
+
evidence?: Readonly<{
|
|
41
|
+
version: typeof DURABLE_EXECUTION_VERSION;
|
|
42
|
+
verdict: "pass" | "fail" | "unknown" | "error";
|
|
43
|
+
snapshotDigest: string;
|
|
44
|
+
planDigest: string;
|
|
45
|
+
reason?: string;
|
|
46
|
+
observed: unknown;
|
|
47
|
+
}>;
|
|
48
|
+
}>;
|
|
49
|
+
export function createDurableContractRunner(configuration: DurableContractConfiguration): Readonly<{
|
|
50
|
+
version: typeof DURABLE_EXECUTION_VERSION;
|
|
51
|
+
run(input: {
|
|
52
|
+
scope: DurableContractScope;
|
|
53
|
+
criterionHash: string;
|
|
54
|
+
maxAttempts: number;
|
|
55
|
+
retry?: boolean;
|
|
56
|
+
signal?: AbortSignal;
|
|
57
|
+
}): Promise<DurableContractResult>;
|
|
58
|
+
assess(result: unknown, options?: { policy?: string }): Promise<AuthenticatedAssessment>;
|
|
59
|
+
}>;
|
|
@@ -0,0 +1,139 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
import { isAbsolute, normalize } from "node:path";
|
|
3
|
+
import { compileIndependentContract } from "./acceptance-independent-contract.js";
|
|
4
|
+
import { captureExecutionSnapshot, EXECUTION_SNAPSHOT_VERSION, runSnapshotBoundContract, snapshotPlanSource } from "./acceptance-execution-snapshot.js";
|
|
5
|
+
import { validateModulePaths } from "./acceptance-executor/module-graph.mjs";
|
|
6
|
+
import { createAuthenticatedAdmission, unavailableAuthenticatedAssessment } from "./acceptance-authenticated-admission.js";
|
|
7
|
+
import { FACT_EVIDENCE_SCOPE_VERSION, ROOT_AGGREGATE_FACT_ID } from "./acceptance-evidence-store.js";
|
|
8
|
+
|
|
9
|
+
export const DURABLE_EXECUTION_VERSION = "durable-module-contract-v2";
|
|
10
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
11
|
+
const hash = (value) => createHash("sha256").update(value).digest("hex");
|
|
12
|
+
const freeze = (value) => {
|
|
13
|
+
if (value && typeof value === "object") { for (const child of Object.values(value)) freeze(child); Object.freeze(value); }
|
|
14
|
+
return value;
|
|
15
|
+
};
|
|
16
|
+
const unavailable = (reason, attemptId) => Object.freeze({ version: DURABLE_EXECUTION_VERSION,
|
|
17
|
+
verdict: "unknown", reason, completionAllowed: false, ...(attemptId ? { attemptId } : {}) });
|
|
18
|
+
|
|
19
|
+
/**
|
|
20
|
+
* HOST-ONLY durable diagnostic runner, not a completion gate or model tool.
|
|
21
|
+
* All configuration/checks and the evidence-store handle must be host-approved.
|
|
22
|
+
* verifierDigest identifies the installed host verifier code; the project
|
|
23
|
+
* verification callback must return the digest of a current, actually observed
|
|
24
|
+
* project verifier, bound to this snapshot and task (or null if unavailable).
|
|
25
|
+
* Neither model-provided expected results nor project-authored receipt JSON may
|
|
26
|
+
* supply these capabilities. Runtime approval/observation wiring is separate.
|
|
27
|
+
*
|
|
28
|
+
* Only the current authenticated store event can be reused. A pending run never
|
|
29
|
+
* starts another worker automatically. Explicit retries consume the same finite
|
|
30
|
+
* attempt budget. Interruption reconciliation must establish worker termination
|
|
31
|
+
* before the store's recordStoppedAttempt capability can be used.
|
|
32
|
+
*/
|
|
33
|
+
export function createDurableContractRunner({ store, projectRoot, sourcePath, modulePaths, authorizeSourceRead,
|
|
34
|
+
exportName, checks, profile, imageId, dockerSocket, dockerCommand, verifierDigest,
|
|
35
|
+
getProjectVerificationDigest, timeoutMs = 10000, startupAllowanceMs = 0 } = {}) {
|
|
36
|
+
if (!store || typeof store.reserve !== "function" || typeof store.settle !== "function"
|
|
37
|
+
|| typeof verifierDigest !== "string" || !HASH.test(verifierDigest) || typeof getProjectVerificationDigest !== "function"
|
|
38
|
+
|| typeof imageId !== "string" || !/^sha256:[a-f0-9]{64}$/.test(imageId)
|
|
39
|
+
|| typeof dockerSocket !== "string" || !dockerSocket.startsWith("/") || dockerSocket.includes("\0")
|
|
40
|
+
|| dockerCommand !== undefined && (!dockerCommand || typeof dockerCommand !== "object"
|
|
41
|
+
|| Array.isArray(dockerCommand)
|
|
42
|
+
|| JSON.stringify(Object.keys(dockerCommand)) !== JSON.stringify(["path", "sha256"])
|
|
43
|
+
|| typeof dockerCommand.path !== "string" || !isAbsolute(dockerCommand.path)
|
|
44
|
+
|| normalize(dockerCommand.path) !== dockerCommand.path || dockerCommand.path.includes("\0")
|
|
45
|
+
|| !HASH.test(String(dockerCommand.sha256 ?? "")))
|
|
46
|
+
|| !Number.isSafeInteger(startupAllowanceMs) || startupAllowanceMs < 0 || startupAllowanceMs > 60000
|
|
47
|
+
|| !Number.isSafeInteger(timeoutMs) || timeoutMs < 25 || timeoutMs > 30000) throw new TypeError("Invalid approved durable verifier configuration");
|
|
48
|
+
// Compile before retaining the plan to detach it from later caller mutations.
|
|
49
|
+
const nodeProfile = profile !== undefined;
|
|
50
|
+
const approvedDockerCommand = dockerCommand === undefined ? undefined
|
|
51
|
+
: Object.freeze({ path: dockerCommand.path, sha256: dockerCommand.sha256 });
|
|
52
|
+
const template = compileIndependentContract(JSON.stringify({ schemaVersion: nodeProfile ? 2 : 1,
|
|
53
|
+
...(nodeProfile ? { profile } : {}), source: "export const placeholder = 0;", exportName, checks }));
|
|
54
|
+
const plan = source => ({ schemaVersion: nodeProfile ? 2 : 1,
|
|
55
|
+
...(nodeProfile ? { profile: template.plan.profile } : {}), ...source, exportName, checks });
|
|
56
|
+
const approvedChecks = template.plan.checks;
|
|
57
|
+
const snapshotRequest = { projectRoot, sourcePath, authorizeSourceRead,
|
|
58
|
+
...(modulePaths === undefined ? {} : { modulePaths: Object.freeze(validateModulePaths(sourcePath, modulePaths)) }) };
|
|
59
|
+
const admission = createAuthenticatedAdmission({ store, snapshotRequest, verifierDigest, imageId, exportName, checks: approvedChecks,
|
|
60
|
+
...(nodeProfile ? { profile: template.plan.profile } : {}) });
|
|
61
|
+
const completed = new WeakMap();
|
|
62
|
+
const backendDigest = hash(JSON.stringify([DURABLE_EXECUTION_VERSION, template.version,
|
|
63
|
+
EXECUTION_SNAPSHOT_VERSION, imageId, dockerSocket, approvedDockerCommand ?? null,
|
|
64
|
+
timeoutMs, template.plan.profile ?? null, ...(startupAllowanceMs ? [startupAllowanceMs] : [])]));
|
|
65
|
+
|
|
66
|
+
async function run({ scope, criterionHash, maxAttempts, retry = false, signal } = {}) {
|
|
67
|
+
if (typeof criterionHash !== "string" || !HASH.test(criterionHash)) throw new TypeError("Invalid current criterion hash");
|
|
68
|
+
const factScoped = scope?.version === FACT_EVIDENCE_SCOPE_VERSION;
|
|
69
|
+
if (factScoped && scope.factId === ROOT_AGGREGATE_FACT_ID) throw new TypeError("The root aggregate is not a code child fact");
|
|
70
|
+
// Freeze identities before awaiting a host callback; the caller cannot swap
|
|
71
|
+
// tasks or criteria midway through a reserved execution.
|
|
72
|
+
const identity = Object.freeze({ ...(factScoped ? { version: FACT_EVIDENCE_SCOPE_VERSION, factId: scope.factId } : {}),
|
|
73
|
+
taskRunId: scope?.taskRunId, criterionId: scope?.criterionId, criterionHash });
|
|
74
|
+
const exactScope = Object.freeze({ ...(factScoped ? { version: FACT_EVIDENCE_SCOPE_VERSION } : {}),
|
|
75
|
+
taskRunId: identity.taskRunId, criterionId: identity.criterionId, ...(factScoped ? { factId: identity.factId } : {}) });
|
|
76
|
+
const before = captureExecutionSnapshot(snapshotRequest);
|
|
77
|
+
if (before.binding.projectId !== store.projectId) throw new Error("Evidence store belongs to another project");
|
|
78
|
+
const verificationDigest = await getProjectVerificationDigest({ ...identity, snapshot: before.binding, snapshotDigest: before.snapshotDigest });
|
|
79
|
+
if (typeof verificationDigest !== "string" || !HASH.test(verificationDigest)) return unavailable("current-project-verifier-missing");
|
|
80
|
+
const compiled = compileIndependentContract(JSON.stringify({ ...plan(snapshotPlanSource(before)), checks: approvedChecks }));
|
|
81
|
+
const binding = { criterionHash, snapshotDigest: before.snapshotDigest, verifierDigest,
|
|
82
|
+
projectVerificationDigest: verificationDigest, planDigest: compiled.planDigest, backendDigest };
|
|
83
|
+
if (signal?.aborted) return unavailable("cancelled-before-reservation");
|
|
84
|
+
const reserved = store.reserve({ scope: exactScope, binding, maxAttempts, retry });
|
|
85
|
+
if (!["current", "reserved"].includes(reserved.status)) return unavailable(`evidence-${reserved.status}`, reserved.event?.attemptId);
|
|
86
|
+
|
|
87
|
+
async function stillCurrent() {
|
|
88
|
+
try {
|
|
89
|
+
const after = captureExecutionSnapshot(snapshotRequest);
|
|
90
|
+
return after.snapshotDigest === before.snapshotDigest
|
|
91
|
+
&& await getProjectVerificationDigest({ ...identity, snapshot: after.binding, snapshotDigest: after.snapshotDigest }) === verificationDigest;
|
|
92
|
+
} catch { return false; }
|
|
93
|
+
}
|
|
94
|
+
function diagnostic(evidence, event, reused) {
|
|
95
|
+
const result = freeze({ version: DURABLE_EXECUTION_VERSION, verdict: evidence.verdict, completionAllowed: false,
|
|
96
|
+
reused, attemptId: event.attemptId, evidence });
|
|
97
|
+
completed.set(result, { scope: exactScope, binding, event, stillCurrent,
|
|
98
|
+
expectedChecks: compiled.plan.checks.map((check) => ({ id: check.id, caseCount: check.cases.length })) });
|
|
99
|
+
return result;
|
|
100
|
+
}
|
|
101
|
+
if (reserved.status === "current") {
|
|
102
|
+
if (!await stillCurrent()) return unavailable("cached-evidence-binding-drift", reserved.event.attemptId);
|
|
103
|
+
const evidence = JSON.parse(reserved.event.evidenceText);
|
|
104
|
+
if (evidence.version !== DURABLE_EXECUTION_VERSION || evidence.snapshotDigest !== before.snapshotDigest
|
|
105
|
+
|| evidence.planDigest !== compiled.planDigest || !["pass", "fail", "unknown", "error"].includes(evidence.verdict)) {
|
|
106
|
+
throw new Error("Authenticated evidence is incompatible with this verifier");
|
|
107
|
+
}
|
|
108
|
+
return diagnostic(evidence, reserved.event, true);
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
let observed;
|
|
112
|
+
try {
|
|
113
|
+
observed = await runSnapshotBoundContract({ ...snapshotRequest, exportName, checks: approvedChecks,
|
|
114
|
+
...(nodeProfile ? { profile: template.plan.profile } : {}), imageId, dockerSocket,
|
|
115
|
+
...(approvedDockerCommand === undefined ? {} : { dockerCommand: approvedDockerCommand }),
|
|
116
|
+
timeoutMs, startupAllowanceMs, signal, executionRunId: reserved.event.attemptId });
|
|
117
|
+
} catch {
|
|
118
|
+
observed = { verdict: "error", reason: "independent-execution-threw" };
|
|
119
|
+
}
|
|
120
|
+
const bound = observed.snapshotDigest === before.snapshotDigest && observed.result?.planDigest === compiled.planDigest;
|
|
121
|
+
const current = await stillCurrent();
|
|
122
|
+
const verdict = !current ? "unknown" : observed.verdict === "error" ? "error" : bound ? observed.verdict : "unknown";
|
|
123
|
+
const evidence = { version: DURABLE_EXECUTION_VERSION, snapshotDigest: before.snapshotDigest, planDigest: compiled.planDigest,
|
|
124
|
+
verdict, ...(!current || (!bound && observed.verdict !== "error") ? { reason: "execution-or-project-verifier-binding-drift" } : {}), observed };
|
|
125
|
+
// A failed settlement remains pending: there is no fallback to unsigned
|
|
126
|
+
// receipt files or a second execution. The host must reconcile explicitly.
|
|
127
|
+
const settled = store.settle(reserved.reservation, JSON.stringify(evidence));
|
|
128
|
+
return diagnostic(evidence, settled, false);
|
|
129
|
+
}
|
|
130
|
+
async function assess(result, { policy = "unknown" } = {}) {
|
|
131
|
+
if (policy !== "allow") return unavailableAuthenticatedAssessment("policy-unestablished-or-denied");
|
|
132
|
+
const observed = result && typeof result === "object" ? completed.get(result) : undefined;
|
|
133
|
+
if (!observed) return unavailableAuthenticatedAssessment("untrusted-durable-result");
|
|
134
|
+
if (!await observed.stillCurrent()) return unavailableAuthenticatedAssessment("current-project-or-source-binding-changed");
|
|
135
|
+
try { return admission.issue(observed); }
|
|
136
|
+
catch { return unavailableAuthenticatedAssessment("authenticated-admission-unavailable"); }
|
|
137
|
+
}
|
|
138
|
+
return Object.freeze({ version: DURABLE_EXECUTION_VERSION, run, assess });
|
|
139
|
+
}
|
|
@@ -147,7 +147,7 @@ export function rejectionStatementErrorClass(text, requestedErrors) {
|
|
|
147
147
|
}
|
|
148
148
|
if (close === -1) return null;
|
|
149
149
|
const argument = text.slice(open + 1, close - 1).trim();
|
|
150
|
-
if (argument && !/^__pi_(?:(?:empty|double_quote|whitespace)_string|string|node_assert_module|code_generation_module|module_loader_module|error_name_[a-z]+)_literal__$/i.test(argument)) return null;
|
|
150
|
+
if (argument && !/^__pi_(?:(?:empty|double_quote|whitespace)_string|invalid_(?:calendar_)?date_string|unparseable_date_string|string|node_assert_module|code_generation_module|module_loader_module|error_name_[a-z]+)_literal__$/i.test(argument)) return null;
|
|
151
151
|
if (!/^(?:[ \t]*;|[ \t]*(?:\}|$))/u.test(text.slice(close))) return null;
|
|
152
152
|
return requestedErrors.length === 0 || requestedErrors.includes(errorClass) ? errorClass : null;
|
|
153
153
|
}
|