@piagent/platform 1.6.1 → 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 +55 -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/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 +2 -0
- package/docs/quality-benchmark.md +128 -8
- 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/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 +11 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +11 -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 +12 -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 +12 -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/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-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 +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 +11 -2
- 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/fs4-readiness-evaluation.mjs +7 -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/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +36 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -0,0 +1,599 @@
|
|
|
1
|
+
import { createHash, createPublicKey, generateKeyPairSync, randomBytes, randomUUID, sign, verify } from "node:crypto";
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
import os from "node:os";
|
|
4
|
+
import path from "node:path";
|
|
5
|
+
|
|
6
|
+
import { benchmarkTreeStatIdentity } from "../packages/piagent-core/benchmark/benchmark-tree-identity.js";
|
|
7
|
+
import { activeSessionTask } from "../packages/piagent-core/extensions/task-state.js";
|
|
8
|
+
import { scopedBrokerArmDigest
|
|
9
|
+
} from "../packages/piagent-core/runtime/verification/composite-scoped-mediation.ts";
|
|
10
|
+
import { benchmarkVerificationOperatorRequest } from "./benchmark-independent-verification.mjs";
|
|
11
|
+
import { benchmarkPublicInputCatalogDigest, validateBenchmarkPublicInputCatalog
|
|
12
|
+
} from "./benchmark-public-input-catalog.mjs";
|
|
13
|
+
import { createScopedMaterialBroker } from "./benchmark-scoped-tool-broker.mjs";
|
|
14
|
+
import { assertScopedFrozenQualification, scopedFrozenQualificationIdentity,
|
|
15
|
+
validateScopedBrokerMeasurementBinding } from "./benchmark-scoped-frozen-qualification.mjs";
|
|
16
|
+
import { createScopedBrokerPiOperationRouter } from "./benchmark-scoped-pi-router.mjs";
|
|
17
|
+
import { loadScopedBrokerFromConfig, scopedCommonRuntimeClosureIdentity, scopedContextPolicySha256,
|
|
18
|
+
scopedJournalPathSha256, scopedToolDefinitionsSha256, scopedVerificationReceiptKeyDigest,
|
|
19
|
+
createScopedProjectVerificationSupervisor, listenScopedVerificationBridge,
|
|
20
|
+
scopedProjectVerificationPlanBinding, SCOPED_PROJECT_VERIFICATION_PROTOCOL,
|
|
21
|
+
SCOPED_CODEX_TURN_OBSERVATION_VERSION, SCOPED_MCP_METADATA_CONTRACT
|
|
22
|
+
} from "./benchmark-scoped-verification-supervisor.mjs";
|
|
23
|
+
import { BENCHMARK_SCOPED_SESSION_CUSTODY_VERSION, CODEX_SCOPED_BROKER_TURN_FACTORY_VERSION
|
|
24
|
+
} from "./benchmark-codex-journey.mjs";
|
|
25
|
+
|
|
26
|
+
export { BENCHMARK_SCOPED_SESSION_CUSTODY_VERSION };
|
|
27
|
+
|
|
28
|
+
export const BENCHMARK_SCOPED_BROKER_CUSTODY_VERSION = "benchmark-scoped-broker-custody-v1";
|
|
29
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
30
|
+
const ID = /^[A-Za-z0-9][A-Za-z0-9:._~-]{0,159}$/;
|
|
31
|
+
const MAX_FILE = 64 * 1024, MAX_JOURNAL = 32 * 1024 * 1024;
|
|
32
|
+
const sha = value => createHash("sha256").update(value).digest("hex");
|
|
33
|
+
const fail = code => { throw Object.assign(new Error(code), { brokerCode: code }); };
|
|
34
|
+
const requireThat = (value, code) => { if (!value) fail(code); };
|
|
35
|
+
|
|
36
|
+
function exact(value, names, code) {
|
|
37
|
+
requireThat(value && typeof value === "object" && !Array.isArray(value)
|
|
38
|
+
&& Object.keys(value).length === names.length && names.every(name => Object.hasOwn(value, name)), code);
|
|
39
|
+
}
|
|
40
|
+
|
|
41
|
+
function realPrivateDirectory(value, code) {
|
|
42
|
+
requireThat(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
43
|
+
&& !value.includes("\0"), code);
|
|
44
|
+
let real, info;
|
|
45
|
+
try { real = fs.realpathSync(value); info = fs.lstatSync(real); } catch { fail(code); }
|
|
46
|
+
requireThat(real === value && info.isDirectory() && !info.isSymbolicLink()
|
|
47
|
+
&& (info.mode & 0o077) === 0, code);
|
|
48
|
+
return real;
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
function stableFile(value, maximum = MAX_FILE, privateMode = false) {
|
|
52
|
+
requireThat(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
53
|
+
&& !value.includes("\0"), "custody-file-path");
|
|
54
|
+
let before;
|
|
55
|
+
try { before = fs.lstatSync(value, { bigint: true }); } catch { fail("custody-file-unavailable"); }
|
|
56
|
+
requireThat(before.isFile() && !before.isSymbolicLink() && before.nlink === 1n
|
|
57
|
+
&& before.size >= 0n && before.size <= BigInt(maximum)
|
|
58
|
+
&& (!privateMode || (before.mode & 0o077n) === 0n), "custody-file-unsafe");
|
|
59
|
+
const fd = fs.openSync(value, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
60
|
+
try {
|
|
61
|
+
const descriptor = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd),
|
|
62
|
+
after = fs.fstatSync(fd, { bigint: true }), current = fs.lstatSync(value, { bigint: true });
|
|
63
|
+
const fields = ["dev", "ino", "mode", "nlink", "size", "mtimeNs", "ctimeNs"];
|
|
64
|
+
requireThat(bytes.length === Number(before.size) && fields.every(name => before[name] === descriptor[name]
|
|
65
|
+
&& before[name] === after[name] && before[name] === current[name]), "custody-file-drift");
|
|
66
|
+
return Buffer.from(bytes);
|
|
67
|
+
} finally { fs.closeSync(fd); }
|
|
68
|
+
}
|
|
69
|
+
|
|
70
|
+
function writePrivate(file, bytes) {
|
|
71
|
+
const fd = fs.openSync(file, fs.constants.O_WRONLY | fs.constants.O_CREAT | fs.constants.O_EXCL
|
|
72
|
+
| fs.constants.O_NOFOLLOW, 0o600);
|
|
73
|
+
try { fs.writeFileSync(fd, bytes); fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
|
|
74
|
+
}
|
|
75
|
+
|
|
76
|
+
function syncDirectory(directory) {
|
|
77
|
+
const fd = fs.openSync(directory, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
78
|
+
try { fs.fsyncSync(fd); } finally { fs.closeSync(fd); }
|
|
79
|
+
}
|
|
80
|
+
|
|
81
|
+
function createVerificationSocketCustody() {
|
|
82
|
+
const temporaryRoot = fs.realpathSync(os.tmpdir()), socketRoot = fs.realpathSync(
|
|
83
|
+
fs.mkdtempSync(path.join(temporaryRoot, "piagent-v-")));
|
|
84
|
+
fs.chmodSync(socketRoot, 0o700);
|
|
85
|
+
const info = fs.lstatSync(socketRoot, { bigint: true });
|
|
86
|
+
requireThat(info.isDirectory() && !info.isSymbolicLink() && (info.mode & 0o077n) === 0n,
|
|
87
|
+
"custody-verification-socket-root");
|
|
88
|
+
return Object.freeze({ root: socketRoot, dev: info.dev, ino: info.ino });
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
function removeVerificationSocketCustody(value) {
|
|
92
|
+
if (!value) return;
|
|
93
|
+
const current = fs.lstatSync(value.root, { bigint: true });
|
|
94
|
+
requireThat(current.isDirectory() && !current.isSymbolicLink() && current.dev === value.dev
|
|
95
|
+
&& current.ino === value.ino && (current.mode & 0o077n) === 0n,
|
|
96
|
+
"custody-verification-socket-root");
|
|
97
|
+
fs.rmdirSync(value.root);
|
|
98
|
+
}
|
|
99
|
+
|
|
100
|
+
function safeMaterial(root, item) {
|
|
101
|
+
const names = ["id", "relativePath", "sha256", "bytes", "readable", "writable", "protected"];
|
|
102
|
+
exact(item, names, "custody-material-shape");
|
|
103
|
+
const protectedIdentity = item.protected && item.sha256 === null && item.bytes === null;
|
|
104
|
+
const contentIdentity = !item.protected && HASH.test(item.sha256)
|
|
105
|
+
&& Number.isSafeInteger(item.bytes) && item.bytes >= 0 && item.bytes <= MAX_FILE;
|
|
106
|
+
requireThat(ID.test(item.id) && typeof item.relativePath === "string" && item.relativePath.length > 0
|
|
107
|
+
&& !path.isAbsolute(item.relativePath) && !item.relativePath.includes("\\") && !item.relativePath.includes("\0")
|
|
108
|
+
&& item.relativePath.split("/").every(part => part && part !== "." && part !== "..")
|
|
109
|
+
&& (protectedIdentity || contentIdentity)
|
|
110
|
+
&& [item.readable, item.writable, item.protected].every(value => typeof value === "boolean")
|
|
111
|
+
&& !(item.protected && (item.readable || item.writable)), "custody-material-shape");
|
|
112
|
+
const target = path.join(root, ...item.relativePath.split("/"));
|
|
113
|
+
requireThat(fs.realpathSync(target) === target && target.startsWith(root + path.sep), "custody-material-path");
|
|
114
|
+
const info = fs.lstatSync(target, { bigint: true });
|
|
115
|
+
requireThat(info.isFile() && !info.isSymbolicLink() && info.nlink === 1n
|
|
116
|
+
&& info.size >= 0n && info.size <= BigInt(MAX_FILE), "custody-material-shape");
|
|
117
|
+
if (item.protected) return Object.freeze(Object.fromEntries(names.map(name => [name, item[name]])));
|
|
118
|
+
const bytes = stableFile(target);
|
|
119
|
+
requireThat(bytes.length === item.bytes && sha(bytes) === item.sha256, "custody-material-drift");
|
|
120
|
+
return Object.freeze(Object.fromEntries(names.map(name => [name, item[name]])));
|
|
121
|
+
}
|
|
122
|
+
|
|
123
|
+
function safeVerification(item) {
|
|
124
|
+
const names = ["id", "protocol", "capabilityDigest", "receiptKeyDigest", "timeoutMs"];
|
|
125
|
+
exact(item, names, "custody-verification-shape");
|
|
126
|
+
requireThat(ID.test(item.id) && ["scoped-isolated-contract-v1", SCOPED_PROJECT_VERIFICATION_PROTOCOL].includes(item.protocol)
|
|
127
|
+
&& HASH.test(item.capabilityDigest) && HASH.test(item.receiptKeyDigest)
|
|
128
|
+
&& Number.isSafeInteger(item.timeoutMs) && item.timeoutMs >= 25 && item.timeoutMs <= 30000,
|
|
129
|
+
"custody-verification-shape");
|
|
130
|
+
return Object.freeze(Object.fromEntries(names.map(name => [name, item[name]])));
|
|
131
|
+
}
|
|
132
|
+
|
|
133
|
+
function prepareVerificationHost(value, projectRoot, nodeCommand) {
|
|
134
|
+
if (value === null) return null;
|
|
135
|
+
const names = ["version", "verificationId", "timeoutMs", "policy", "expectedPlanDigest"];
|
|
136
|
+
exact(value, names, "custody-verification-host");
|
|
137
|
+
requireThat(value.version === "scoped-project-verification-host-v1" && ID.test(value.verificationId)
|
|
138
|
+
&& Number.isSafeInteger(value.timeoutMs) && value.timeoutMs >= 25 && value.timeoutMs <= 30000
|
|
139
|
+
&& HASH.test(value.expectedPlanDigest), "custody-verification-host");
|
|
140
|
+
const binding = scopedProjectVerificationPlanBinding({ projectRoot, nodeCommand,
|
|
141
|
+
policy: value.policy, timeoutMs: value.timeoutMs });
|
|
142
|
+
requireThat(binding.planDigest === value.expectedPlanDigest, "custody-verification-plan-drift");
|
|
143
|
+
const authority = generateKeyPairSync("ed25519"), manifest = Object.freeze({ id: value.verificationId,
|
|
144
|
+
protocol: binding.protocol, capabilityDigest: binding.capabilityDigest,
|
|
145
|
+
receiptKeyDigest: scopedVerificationReceiptKeyDigest(authority.publicKey), timeoutMs: value.timeoutMs });
|
|
146
|
+
return Object.freeze({ authority, binding, manifest, policy: value.policy });
|
|
147
|
+
}
|
|
148
|
+
|
|
149
|
+
function validatePreparedVerificationHost(value, projectRoot, nodeCommand, verifications) {
|
|
150
|
+
if (value === null) return null;
|
|
151
|
+
exact(value, ["authority", "binding", "manifest", "policy"], "custody-verification-host");
|
|
152
|
+
requireThat(value.authority?.privateKey?.type === "private" && value.authority.privateKey.asymmetricKeyType === "ed25519"
|
|
153
|
+
&& value.authority?.publicKey?.type === "public" && value.authority.publicKey.asymmetricKeyType === "ed25519"
|
|
154
|
+
&& verifications.length === 1 && JSON.stringify(verifications[0]) === JSON.stringify(value.manifest),
|
|
155
|
+
"custody-verification-host");
|
|
156
|
+
const binding = scopedProjectVerificationPlanBinding({ projectRoot, nodeCommand, policy: value.policy,
|
|
157
|
+
timeoutMs: value.manifest.timeoutMs });
|
|
158
|
+
requireThat(JSON.stringify(binding) === JSON.stringify(value.binding)
|
|
159
|
+
&& value.manifest.protocol === SCOPED_PROJECT_VERIFICATION_PROTOCOL
|
|
160
|
+
&& value.manifest.capabilityDigest === binding.capabilityDigest
|
|
161
|
+
&& value.manifest.receiptKeyDigest === scopedVerificationReceiptKeyDigest(value.authority.publicKey),
|
|
162
|
+
"custody-verification-plan-drift");
|
|
163
|
+
return value;
|
|
164
|
+
}
|
|
165
|
+
|
|
166
|
+
function safeCodexTurnObservations(bodies) {
|
|
167
|
+
const names = ["version", "contractId", "authority", "metadataSha256", "callId", "threadId", "itemId",
|
|
168
|
+
"sessionId", "turnId", "turnStartedAtUnixMs", "model", "reasoningEffort", "sandboxMode",
|
|
169
|
+
"sandboxStateSha256"], seenCalls = new Set(), seenItems = new Set(), observations = [];
|
|
170
|
+
for (const row of bodies.filter(item => item.type === "transport-observation")) {
|
|
171
|
+
const item = row.observation;
|
|
172
|
+
exact(item, names, "custody-codex-observation");
|
|
173
|
+
requireThat(item.version === SCOPED_CODEX_TURN_OBSERVATION_VERSION
|
|
174
|
+
&& item.contractId === SCOPED_MCP_METADATA_CONTRACT.id && item.authority === "none"
|
|
175
|
+
&& HASH.test(item.metadataSha256) && [item.callId, item.threadId, item.itemId, item.sessionId,
|
|
176
|
+
item.turnId, item.model].every(value => typeof value === "string" && ID.test(value))
|
|
177
|
+
&& Number.isSafeInteger(item.turnStartedAtUnixMs) && item.turnStartedAtUnixMs >= 0
|
|
178
|
+
&& (item.reasoningEffort === null || ["none", "minimal", "low", "medium", "high", "xhigh",
|
|
179
|
+
"max", "ultra"].includes(item.reasoningEffort)) && item.sandboxMode === "workspace-write"
|
|
180
|
+
&& (item.sandboxStateSha256 === null || HASH.test(item.sandboxStateSha256))
|
|
181
|
+
&& !seenCalls.has(item.callId) && !seenItems.has(item.itemId), "custody-codex-observation");
|
|
182
|
+
seenCalls.add(item.callId); seenItems.add(item.itemId);
|
|
183
|
+
observations.push(Object.freeze(Object.fromEntries(names.map(name => [name, item[name]]))));
|
|
184
|
+
}
|
|
185
|
+
return Object.freeze(observations);
|
|
186
|
+
}
|
|
187
|
+
|
|
188
|
+
export function scopedBrokerMaterialManifestSha256(profile, materials) {
|
|
189
|
+
requireThat(typeof profile === "string" && Array.isArray(materials), "custody-material-manifest");
|
|
190
|
+
return sha(JSON.stringify({ version: 1, profile, materials }));
|
|
191
|
+
}
|
|
192
|
+
|
|
193
|
+
export function scopedBrokerVerificationManifestSha256(verifications) {
|
|
194
|
+
requireThat(Array.isArray(verifications), "custody-verification-manifest");
|
|
195
|
+
return sha(JSON.stringify({ version: 1, verifications }));
|
|
196
|
+
}
|
|
197
|
+
|
|
198
|
+
export function scopedBrokerModelIdentitySha256(value) {
|
|
199
|
+
exact(value, ["provider", "model", "thinking", "serviceTier"], "custody-model-identity");
|
|
200
|
+
requireThat([value.provider, value.model].every(item => typeof item === "string" && item.length > 0
|
|
201
|
+
&& item.length <= 160) && (value.thinking === null || typeof value.thinking === "string")
|
|
202
|
+
&& (value.serviceTier === null || typeof value.serviceTier === "string"), "custody-model-identity");
|
|
203
|
+
return sha(JSON.stringify(value));
|
|
204
|
+
}
|
|
205
|
+
|
|
206
|
+
function journalEvidence({ journalPath, manifestBytes, manifestSignature, manifestPublicKey,
|
|
207
|
+
journalPublicKey, identity, verificationAvailable, transportRequired }) {
|
|
208
|
+
const journalBytes = stableFile(journalPath, MAX_JOURNAL, true);
|
|
209
|
+
requireThat(journalBytes.length > 0 && journalBytes.at(-1) === 10, "custody-journal-incomplete");
|
|
210
|
+
const authority = createPublicKey(manifestPublicKey), journalKey = createPublicKey(journalPublicKey);
|
|
211
|
+
requireThat(verify(null, manifestBytes, authority, manifestSignature), "custody-manifest-signature");
|
|
212
|
+
const lines = new TextDecoder("utf-8", { fatal: true }).decode(journalBytes).slice(0, -1).split("\n");
|
|
213
|
+
requireThat(lines.length > 0 && lines.length <= 2048, "custody-journal-bounds");
|
|
214
|
+
let previous = "0".repeat(64), sequence = 0; const bodies = [];
|
|
215
|
+
for (const line of lines) {
|
|
216
|
+
let row;
|
|
217
|
+
try { row = JSON.parse(line); } catch { fail("custody-journal-json"); }
|
|
218
|
+
exact(row, ["body", "signature"], "custody-journal-envelope");
|
|
219
|
+
requireThat(JSON.stringify(row) === line && typeof row.signature === "string"
|
|
220
|
+
&& verify(null, Buffer.from(JSON.stringify(row.body)), journalKey, Buffer.from(row.signature, "base64"))
|
|
221
|
+
&& row.body?.sequence === ++sequence && row.body.previous === previous
|
|
222
|
+
&& row.body.manifestSha256 === sha(manifestBytes)
|
|
223
|
+
&& JSON.stringify(row.body.identity) === JSON.stringify(identity), "custody-journal-chain");
|
|
224
|
+
previous = sha(Buffer.from(`${line}\n`)); bodies.push(row.body);
|
|
225
|
+
}
|
|
226
|
+
const terminal = bodies.at(-1), reservations = bodies.filter(row => row.type === "reservation"),
|
|
227
|
+
results = bodies.filter(row => row.type === "result"), transportEnds = bodies.filter(row => row.type === "transport-end"),
|
|
228
|
+
codexTurnObservations = safeCodexTurnObservations(bodies);
|
|
229
|
+
requireThat(bodies[0]?.type === "begin" && terminal?.type === "end" && terminal.cancelled === false
|
|
230
|
+
&& Number.isSafeInteger(terminal.actions) && terminal.actions === reservations.length
|
|
231
|
+
&& reservations.length === results.length && new Set(results.map(row => row.action)).size === results.length
|
|
232
|
+
&& (!transportRequired || bodies.filter(row => row.type === "transport-begin").length === 1
|
|
233
|
+
&& transportEnds.length === 1 && transportEnds[0].complete === true && transportEnds[0].reason === null),
|
|
234
|
+
"custody-journal-terminal");
|
|
235
|
+
return Object.freeze({ version: "scoped-broker-journal-evidence-v1", manifestBytes: Buffer.from(manifestBytes),
|
|
236
|
+
manifestSignature: Buffer.from(manifestSignature), manifestPublicKey: Buffer.from(manifestPublicKey),
|
|
237
|
+
journalBytes, journalPublicKey: Buffer.from(journalPublicKey), status: Object.freeze({
|
|
238
|
+
actions: terminal.actions, blocked: false, ended: true, cancelled: false, journalSha256: sha(journalBytes),
|
|
239
|
+
g0Qualified: false, verificationAvailable, inflightVerification: null }), codexTurnObservations });
|
|
240
|
+
}
|
|
241
|
+
|
|
242
|
+
const normalizedModel = value => typeof value === "string" ? value.split("/").at(-1) : null;
|
|
243
|
+
const normalizedThinking = value => value === "off" ? "none" : value === "minimal" ? "low" : value;
|
|
244
|
+
|
|
245
|
+
/** Reconciles signed, untrusted MCP observations with one exact Codex exec
|
|
246
|
+
* JSONL lifecycle. It does not make the observation authoritative. */
|
|
247
|
+
export function reconcileCodexScopedBrokerSettlement(evidence, usage) {
|
|
248
|
+
requireThat(evidence?.status?.ended === true && evidence.status.cancelled === false
|
|
249
|
+
&& Array.isArray(evidence.codexTurnObservations), "custody-codex-settlement");
|
|
250
|
+
const lifecycle = usage?.turnLifecycleEvidence, jsonl = usage?.jsonlEvidence;
|
|
251
|
+
requireThat(typeof usage?.providerSessionId === "string" && ID.test(usage.providerSessionId)
|
|
252
|
+
&& lifecycle?.schemaVersion === 1 && lifecycle.source === "codex-exec-jsonl-turn-lifecycle"
|
|
253
|
+
&& lifecycle.startedEvents === 1 && lifecycle.completedEvents === 1
|
|
254
|
+
&& jsonl?.schemaVersion === 1 && jsonl.source === "codex-exec-jsonl-stdout-bytes"
|
|
255
|
+
&& Number.isSafeInteger(jsonl.bytes) && jsonl.bytes > 0 && HASH.test(jsonl.sha256),
|
|
256
|
+
"custody-codex-jsonl");
|
|
257
|
+
const observations = evidence.codexTurnObservations;
|
|
258
|
+
if (observations.length === 0) {
|
|
259
|
+
requireThat(evidence.status.actions === 0, "custody-codex-observation-missing");
|
|
260
|
+
return Object.freeze({ version: "codex-broker-settlement-reconciliation-v1", authority: "none",
|
|
261
|
+
status: "native-turn-unobserved-zero-call", reconciled: false,
|
|
262
|
+
providerSessionId: usage.providerSessionId, nativeTurnId: null, observations: 0,
|
|
263
|
+
jsonlSha256: jsonl.sha256, journalSha256: evidence.status.journalSha256 });
|
|
264
|
+
}
|
|
265
|
+
const first = observations[0], model = normalizedModel(usage.model), thinking = normalizedThinking(usage.thinkingLevel);
|
|
266
|
+
requireThat(observations.length === evidence.status.actions && first.threadId === usage.providerSessionId
|
|
267
|
+
&& first.sessionId === usage.providerSessionId && first.model === model && first.reasoningEffort === thinking
|
|
268
|
+
&& observations.every(item => item.threadId === first.threadId && item.sessionId === first.sessionId
|
|
269
|
+
&& item.turnId === first.turnId && item.model === first.model
|
|
270
|
+
&& item.reasoningEffort === first.reasoningEffort), "custody-codex-identity-mismatch");
|
|
271
|
+
return Object.freeze({ version: "codex-broker-settlement-reconciliation-v1", authority: "none",
|
|
272
|
+
status: "reconciled", reconciled: true, providerSessionId: usage.providerSessionId,
|
|
273
|
+
nativeTurnId: first.turnId, observations: observations.length, jsonlSha256: jsonl.sha256,
|
|
274
|
+
journalSha256: evidence.status.journalSha256 });
|
|
275
|
+
}
|
|
276
|
+
|
|
277
|
+
/** Creates one fresh, private, single-use turn custody. No returned value can
|
|
278
|
+
* approve a plan, dispatch a provider, grade an answer or admit a benchmark. */
|
|
279
|
+
export async function createBenchmarkScopedBrokerTurnCustody(input) {
|
|
280
|
+
const names = ["custodyRoot", "nodeCommand", "brokerScript", "runtimePath", "qualification",
|
|
281
|
+
"measurementBinding", "contextPolicy", "modelIdentity", "taskIdentity", "materialRoot", "materials",
|
|
282
|
+
"verifications", "verificationBridge", "verificationHost"];
|
|
283
|
+
exact(input, names, "custody-input-shape");
|
|
284
|
+
const custodyRoot = realPrivateDirectory(input.custodyRoot, "custody-root"),
|
|
285
|
+
materialRoot = realPrivateDirectory(input.materialRoot, "custody-material-root");
|
|
286
|
+
requireThat(!custodyRoot.startsWith(materialRoot + path.sep) && !materialRoot.startsWith(custodyRoot + path.sep)
|
|
287
|
+
&& custodyRoot !== materialRoot,
|
|
288
|
+
"custody-root-overlap");
|
|
289
|
+
const nodeCommand = fs.realpathSync(input.nodeCommand), brokerScript = fs.realpathSync(input.brokerScript);
|
|
290
|
+
requireThat(nodeCommand === input.nodeCommand && brokerScript === input.brokerScript
|
|
291
|
+
&& brokerScript === fs.realpathSync(new URL("./benchmark-scoped-tool-broker.mjs", import.meta.url)),
|
|
292
|
+
"custody-runtime-path");
|
|
293
|
+
const binding = validateScopedBrokerMeasurementBinding(input.measurementBinding),
|
|
294
|
+
contextPolicySha256 = scopedContextPolicySha256(input.contextPolicy),
|
|
295
|
+
materials = Object.freeze(input.materials.map(item => safeMaterial(materialRoot, item))),
|
|
296
|
+
verifications = Object.freeze(input.verifications.map(safeVerification)),
|
|
297
|
+
verificationHost = validatePreparedVerificationHost(input.verificationHost, materialRoot, nodeCommand,
|
|
298
|
+
verifications);
|
|
299
|
+
requireThat(new Set(materials.map(item => item.id)).size === materials.length && materials.length <= 32
|
|
300
|
+
&& new Set(verifications.map(item => item.id)).size === verifications.length && verifications.length <= 32
|
|
301
|
+
&& (!verificationHost || input.verificationBridge === null)
|
|
302
|
+
&& materials.every(item => (!item.writable || binding.profile === "document")
|
|
303
|
+
&& (!(item.readable || item.writable) || ["incident", "document"].includes(binding.profile)))
|
|
304
|
+
&& scopedBrokerMaterialManifestSha256(binding.profile, materials)
|
|
305
|
+
=== binding.materialManifestSha256
|
|
306
|
+
&& scopedBrokerVerificationManifestSha256(verifications) === binding.verificationManifestSha256
|
|
307
|
+
&& contextPolicySha256 === binding.contextPolicySha256
|
|
308
|
+
&& sha(stableFile(nodeCommand, 1024 * 1024 * 1024)) === binding.nodeSha256
|
|
309
|
+
&& scopedBrokerModelIdentitySha256(input.modelIdentity) === binding.modelSha256,
|
|
310
|
+
"custody-measurement-drift");
|
|
311
|
+
exact(input.taskIdentity, ["taskId", "sessionId", "operationId", "nonce"], "custody-task-identity");
|
|
312
|
+
requireThat(Object.values(input.taskIdentity).every(value => typeof value === "string" && ID.test(value)),
|
|
313
|
+
"custody-task-identity");
|
|
314
|
+
const closure = scopedCommonRuntimeClosureIdentity(), qualification = scopedFrozenQualificationIdentity(
|
|
315
|
+
input.qualification, closure.sha256);
|
|
316
|
+
if (binding.surface === "piagent") requireThat(binding.runtimeSha256 === qualification.sdkTreeSha256,
|
|
317
|
+
"custody-runtime-drift");
|
|
318
|
+
else requireThat(typeof input.runtimePath === "string" && fs.realpathSync(input.runtimePath) === input.runtimePath
|
|
319
|
+
&& sha(stableFile(input.runtimePath, 1024 * 1024 * 1024)) === binding.runtimeSha256,
|
|
320
|
+
"custody-runtime-drift");
|
|
321
|
+
const turnRoot = fs.mkdtempSync(path.join(custodyRoot, "turn-")); fs.chmodSync(turnRoot, 0o700);
|
|
322
|
+
const socketCustody = verificationHost ? createVerificationSocketCustody() : null;
|
|
323
|
+
let listener = null, disposePromise = null;
|
|
324
|
+
const dispose = () => {
|
|
325
|
+
disposePromise ??= (async () => {
|
|
326
|
+
try { await listener?.close(); }
|
|
327
|
+
finally { removeVerificationSocketCustody(socketCustody); }
|
|
328
|
+
})();
|
|
329
|
+
return disposePromise;
|
|
330
|
+
};
|
|
331
|
+
try {
|
|
332
|
+
const at = name => path.join(turnRoot, name), authorization = verificationHost ? randomBytes(32).toString("hex") : null,
|
|
333
|
+
receiptPublicKey = verificationHost?.authority.publicKey.export({ type: "spki", format: "pem" }) ?? null,
|
|
334
|
+
verificationBridge = verificationHost ? { socketPath: path.join(socketCustody.root, "v.sock"),
|
|
335
|
+
authorizationPath: at("verification-authorization"), receiptPublicKeyPath: at("verification-public.pem"),
|
|
336
|
+
receiptKeyDigest: verificationHost.manifest.receiptKeyDigest,
|
|
337
|
+
timeoutMs: Math.min(60000, verificationHost.manifest.timeoutMs + 5000) } : input.verificationBridge,
|
|
338
|
+
host = generateKeyPairSync("ed25519"),
|
|
339
|
+
journal = generateKeyPairSync("ed25519"), config = { version: 4, manifestPath: at("manifest.json"),
|
|
340
|
+
manifestSignaturePath: at("manifest.sig"), manifestPublicKeyPath: at("manifest-public.pem"),
|
|
341
|
+
journalPath: at("journal.jsonl"), journalPrivateKeyPath: at("journal-private.pem"),
|
|
342
|
+
verificationBridge, qualification: { ...input.qualification },
|
|
343
|
+
measurementBinding: { ...binding } }, configBytes = Buffer.from(JSON.stringify(config));
|
|
344
|
+
const identity = Object.freeze({ version: 4, armId: binding.armId, taskId: input.taskIdentity.taskId,
|
|
345
|
+
sessionId: input.taskIdentity.sessionId, requestId: binding.operatorRequestDigest,
|
|
346
|
+
operationId: input.taskIdentity.operationId, nonce: input.taskIdentity.nonce,
|
|
347
|
+
sourceSha256: qualification.sourceSha256, assetTreeSha256: qualification.assetTreeSha256,
|
|
348
|
+
configSha256: sha(configBytes), measurementConfigurationSha256: binding.configurationSha256,
|
|
349
|
+
brokerClosureSha256: qualification.brokerClosureSha256,
|
|
350
|
+
toolDefinitionsSha256: scopedToolDefinitionsSha256(),
|
|
351
|
+
manifestAuthoritySha256: scopedVerificationReceiptKeyDigest(host.publicKey),
|
|
352
|
+
journalSignerSha256: scopedVerificationReceiptKeyDigest(journal.publicKey),
|
|
353
|
+
journalPathSha256: scopedJournalPathSha256(config.journalPath), contextPolicySha256,
|
|
354
|
+
sdkTreeSha256: qualification.sdkTreeSha256 });
|
|
355
|
+
const manifestBytes = Buffer.from(JSON.stringify({ version: 2, identity, profile: binding.profile,
|
|
356
|
+
root: materialRoot, materials, verifications })), manifestSignature = sign(null, manifestBytes, host.privateKey),
|
|
357
|
+
manifestPublicKey = host.publicKey.export({ type: "spki", format: "pem" }),
|
|
358
|
+
journalPrivateKey = journal.privateKey.export({ type: "pkcs8", format: "pem" }),
|
|
359
|
+
journalPublicKey = journal.publicKey.export({ type: "spki", format: "pem" });
|
|
360
|
+
requireThat(configBytes.length <= MAX_FILE && manifestBytes.length <= MAX_FILE, "custody-static-size");
|
|
361
|
+
const hostFiles = verificationHost ? [[verificationBridge.authorizationPath, Buffer.from(authorization)],
|
|
362
|
+
[verificationBridge.receiptPublicKeyPath, receiptPublicKey]] : [];
|
|
363
|
+
for (const [file, bytes] of [[config.manifestPath, manifestBytes], [config.manifestSignaturePath, manifestSignature],
|
|
364
|
+
[config.manifestPublicKeyPath, manifestPublicKey], [config.journalPrivateKeyPath, journalPrivateKey],
|
|
365
|
+
[at("config.json"), configBytes], ...hostFiles]) writePrivate(file, bytes);
|
|
366
|
+
syncDirectory(turnRoot);
|
|
367
|
+
const staticFiles = Object.freeze([[config.manifestPath, manifestBytes], [config.manifestSignaturePath, manifestSignature],
|
|
368
|
+
[config.manifestPublicKeyPath, manifestPublicKey], [config.journalPrivateKeyPath, journalPrivateKey],
|
|
369
|
+
[at("config.json"), configBytes], ...hostFiles]
|
|
370
|
+
.map(([file, bytes]) => Object.freeze({ file, sha256: sha(bytes) })));
|
|
371
|
+
let piOpened = false;
|
|
372
|
+
if (verificationHost) {
|
|
373
|
+
const supervisor = createScopedProjectVerificationSupervisor({ manifestSha256: sha(manifestBytes),
|
|
374
|
+
brokerIdentitySha256: sha(JSON.stringify(identity)), brokerSourceSha256: identity.brokerClosureSha256,
|
|
375
|
+
receiptPrivateKey: verificationHost.authority.privateKey, verifications: [{ manifest: verificationHost.manifest,
|
|
376
|
+
plan: { projectRoot: materialRoot, nodeCommand, stagingRoot: turnRoot, policy: verificationHost.policy,
|
|
377
|
+
expectedPlanDigest: verificationHost.binding.planDigest } }] });
|
|
378
|
+
listener = await listenScopedVerificationBridge({ supervisor, authorization,
|
|
379
|
+
socketPath: verificationBridge.socketPath });
|
|
380
|
+
}
|
|
381
|
+
const assertStatic = () => {
|
|
382
|
+
for (const item of staticFiles) requireThat(sha(stableFile(item.file, MAX_FILE, true)) === item.sha256,
|
|
383
|
+
"custody-static-drift");
|
|
384
|
+
assertScopedFrozenQualification(qualification, identity, scopedCommonRuntimeClosureIdentity().sha256);
|
|
385
|
+
};
|
|
386
|
+
const codexEvidence = () => journalEvidence({ journalPath: config.journalPath, manifestBytes, manifestSignature,
|
|
387
|
+
manifestPublicKey, journalPublicKey, identity, verificationAvailable: verifications.length > 0,
|
|
388
|
+
transportRequired: true });
|
|
389
|
+
return Object.freeze({ version: BENCHMARK_SCOPED_BROKER_CUSTODY_VERSION, authority: "none", turnRoot,
|
|
390
|
+
identity, measurementBinding: binding, configSha256: identity.configSha256,
|
|
391
|
+
brokerConfigPath: at("config.json"), nodeCommand, brokerScript,
|
|
392
|
+
codexLaunch: Object.freeze({ nodeCommand, brokerScript, brokerConfigPath: at("config.json") }),
|
|
393
|
+
assertPredispatch() { requireThat(!piOpened && !disposePromise, "custody-predispatch-state");
|
|
394
|
+
assertStatic(); return true; },
|
|
395
|
+
async openPi() { requireThat(!piOpened && !disposePromise && !fs.existsSync(config.journalPath), "custody-single-use");
|
|
396
|
+
try { assertStatic(); piOpened = true;
|
|
397
|
+
const loaded = loadScopedBrokerFromConfig({ configPath: at("config.json"),
|
|
398
|
+
createBroker: createScopedMaterialBroker });
|
|
399
|
+
return Object.freeze({ ...loaded, dispose, assertProviderDispatchReady() {
|
|
400
|
+
requireThat(piOpened && !disposePromise, "custody-predispatch-state");
|
|
401
|
+
assertStatic(); return true;
|
|
402
|
+
} });
|
|
403
|
+
} catch (error) { await dispose(); throw error; } },
|
|
404
|
+
codexSettlementEvidence() { requireThat(!piOpened, "custody-surface-conflict"); assertStatic();
|
|
405
|
+
return codexEvidence(); },
|
|
406
|
+
reconcileCodexSettlement(usage) { requireThat(!piOpened, "custody-surface-conflict"); assertStatic();
|
|
407
|
+
const evidence = codexEvidence();
|
|
408
|
+
return Object.freeze({ version: "codex-scoped-broker-turn-settlement-v1", authority: "none",
|
|
409
|
+
configSha256: identity.configSha256, brokerIdentitySha256: sha(JSON.stringify(identity)),
|
|
410
|
+
manifestSha256: sha(manifestBytes), journalSha256: evidence.status.journalSha256,
|
|
411
|
+
actions: evidence.status.actions,
|
|
412
|
+
reconciliation: reconcileCodexScopedBrokerSettlement(evidence, usage) }); },
|
|
413
|
+
dispose
|
|
414
|
+
});
|
|
415
|
+
} catch (error) { await dispose(); throw error; }
|
|
416
|
+
}
|
|
417
|
+
|
|
418
|
+
function stableSessionFile(file, maximum, code) {
|
|
419
|
+
requireThat(typeof file === "string" && path.isAbsolute(file) && path.normalize(file) === file
|
|
420
|
+
&& !file.includes("\0") && fs.realpathSync(file) === file, code);
|
|
421
|
+
const before = fs.lstatSync(file, { bigint: true });
|
|
422
|
+
requireThat(before.isFile() && !before.isSymbolicLink() && before.nlink === 1n
|
|
423
|
+
&& before.size > 0n && before.size <= BigInt(maximum), code);
|
|
424
|
+
const fd = fs.openSync(file, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
425
|
+
try {
|
|
426
|
+
const descriptor = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd),
|
|
427
|
+
after = fs.fstatSync(fd, { bigint: true }), current = fs.lstatSync(file, { bigint: true }),
|
|
428
|
+
fields = ["dev", "ino", "mode", "nlink", "size", "mtimeNs", "ctimeNs"];
|
|
429
|
+
requireThat(bytes.length === Number(before.size) && fields.every(name => before[name] === descriptor[name]
|
|
430
|
+
&& before[name] === after[name] && before[name] === current[name]), code);
|
|
431
|
+
return bytes;
|
|
432
|
+
} finally { fs.closeSync(fd); }
|
|
433
|
+
}
|
|
434
|
+
|
|
435
|
+
function outside(root, file) {
|
|
436
|
+
const relative = path.relative(root, file);
|
|
437
|
+
return relative === ".." || relative.startsWith(`..${path.sep}`) || path.isAbsolute(relative);
|
|
438
|
+
}
|
|
439
|
+
|
|
440
|
+
function exactTurns(turns, scenario) {
|
|
441
|
+
requireThat(Array.isArray(turns) && turns.length === scenario.turnCount, "session-custody-turns");
|
|
442
|
+
return Object.freeze(turns.map((turn, index) => {
|
|
443
|
+
requireThat(turn && typeof turn === "object" && turn.id === scenario.turns[index].turnId
|
|
444
|
+
&& typeof turn.message === "string" && turn.message.length > 0 && turn.message.isWellFormed()
|
|
445
|
+
&& sha(turn.message) === scenario.turns[index].promptSha256
|
|
446
|
+
&& Buffer.byteLength(turn.message) === scenario.turns[index].promptBytes
|
|
447
|
+
&& (turn.workflow ?? null) === scenario.turns[index].workflow
|
|
448
|
+
&& (turn.reconnectBefore === true) === scenario.turns[index].reconnectBefore
|
|
449
|
+
&& (turn.receiptUncertain === true) === scenario.turns[index].receiptUncertain,
|
|
450
|
+
"session-custody-turns");
|
|
451
|
+
const operatorRequest = benchmarkVerificationOperatorRequest({ message: turn.message,
|
|
452
|
+
workflow: turn.workflow ?? null });
|
|
453
|
+
requireThat(`operator-request-v1:${sha(operatorRequest)}` === scenario.turns[index].operatorRequestDigest,
|
|
454
|
+
"session-custody-operator-request");
|
|
455
|
+
return Object.freeze({ index: index + 1, id: turn.id, message: turn.message,
|
|
456
|
+
operatorRequest, catalog: scenario.turns[index] });
|
|
457
|
+
}));
|
|
458
|
+
}
|
|
459
|
+
|
|
460
|
+
function resources(value, expectedRequests, workspace) {
|
|
461
|
+
const names = ["profile", "contextPolicy", "materialRoot", "materials", "verifications",
|
|
462
|
+
"verificationBridge", "verificationHost"];
|
|
463
|
+
exact(value, names, "session-custody-resources");
|
|
464
|
+
requireThat(fs.realpathSync(value.materialRoot) === workspace
|
|
465
|
+
&& JSON.stringify(value.contextPolicy?.allowedUserMessages) === JSON.stringify(expectedRequests)
|
|
466
|
+
&& JSON.stringify(value.contextPolicy?.removableUserMessages) === "[]", "session-custody-context");
|
|
467
|
+
scopedContextPolicySha256(value.contextPolicy);
|
|
468
|
+
requireThat(Array.isArray(value.materials) && Array.isArray(value.verifications)
|
|
469
|
+
&& (value.verificationHost === null || value.profile === "document"
|
|
470
|
+
&& value.verifications.length === 0 && value.verificationBridge === null),
|
|
471
|
+
"session-custody-resources");
|
|
472
|
+
return value;
|
|
473
|
+
}
|
|
474
|
+
|
|
475
|
+
/** Builds data-only, just-in-time per-turn custody. It cannot approve a plan,
|
|
476
|
+
* start a provider, grade output, or admit a benchmark record. */
|
|
477
|
+
export function createBenchmarkScopedSessionCustody(input) {
|
|
478
|
+
const names = ["custodyRoot", "nodeCommand", "brokerScript", "runtimePath", "qualification", "catalog",
|
|
479
|
+
"runId", "armId", "expectedArmBinding", "suiteId", "scenarioId", "surface", "repeat", "infrastructureAttempt",
|
|
480
|
+
"configurationSha256", "planPath", "publicContractPath", "workspace", "modelIdentity", "turns",
|
|
481
|
+
"resolveResources"];
|
|
482
|
+
exact(input, names, "session-custody-input");
|
|
483
|
+
requireThat([input.runId, input.armId, input.suiteId, input.scenarioId].every(value =>
|
|
484
|
+
typeof value === "string" && ID.test(value)) && ["piagent", "codex-cli"].includes(input.surface)
|
|
485
|
+
&& Number.isSafeInteger(input.repeat) && input.repeat >= 1 && input.repeat <= 10
|
|
486
|
+
&& Number.isSafeInteger(input.infrastructureAttempt) && input.infrastructureAttempt >= 1
|
|
487
|
+
&& input.infrastructureAttempt <= 3 && HASH.test(input.configurationSha256)
|
|
488
|
+
&& typeof input.resolveResources === "function", "session-custody-input");
|
|
489
|
+
const workspace = fs.realpathSync(input.workspace), info = fs.lstatSync(workspace);
|
|
490
|
+
requireThat(workspace === input.workspace && info.isDirectory() && !info.isSymbolicLink()
|
|
491
|
+
&& (info.mode & 0o077) === 0, "session-custody-workspace");
|
|
492
|
+
const catalog = JSON.parse(JSON.stringify(validateBenchmarkPublicInputCatalog(input.catalog,
|
|
493
|
+
{ suiteId: input.suiteId }))), catalogSha256 = benchmarkPublicInputCatalogDigest(catalog),
|
|
494
|
+
scenario = catalog.scenarios.find(value => value.scenarioId === input.scenarioId);
|
|
495
|
+
requireThat(scenario && scenario.planRef === `plan:${input.scenarioId}`, "session-custody-scenario");
|
|
496
|
+
const turns = exactTurns(input.turns, scenario), planBytes = stableSessionFile(input.planPath, 2 * 1024 * 1024,
|
|
497
|
+
"session-custody-plan"), contractBytes = stableSessionFile(input.publicContractPath, 2 * 1024 * 1024,
|
|
498
|
+
"session-custody-public-contract");
|
|
499
|
+
exact(input.expectedArmBinding, ["turnIndex", "sha256"], "session-custody-arm-binding");
|
|
500
|
+
requireThat(Number.isSafeInteger(input.expectedArmBinding.turnIndex)
|
|
501
|
+
&& input.expectedArmBinding.turnIndex >= 1 && input.expectedArmBinding.turnIndex <= turns.length
|
|
502
|
+
&& HASH.test(input.expectedArmBinding.sha256), "session-custody-arm-binding");
|
|
503
|
+
requireThat(outside(workspace, input.planPath) && outside(workspace, input.publicContractPath),
|
|
504
|
+
"session-custody-authority-location");
|
|
505
|
+
const planSha256 = sha(planBytes), publicContractSha256 = sha(contractBytes),
|
|
506
|
+
nodeSha256 = sha(stableSessionFile(input.nodeCommand, 1024 * 1024 * 1024, "session-custody-node")),
|
|
507
|
+
modelSha256 = scopedBrokerModelIdentitySha256(input.modelIdentity), closure = scopedCommonRuntimeClosureIdentity(),
|
|
508
|
+
frozen = scopedFrozenQualificationIdentity(input.qualification, closure.sha256),
|
|
509
|
+
runtimeSha256 = input.surface === "piagent" ? frozen.sdkTreeSha256
|
|
510
|
+
: sha(stableSessionFile(input.runtimePath, 1024 * 1024 * 1024, "session-custody-runtime"));
|
|
511
|
+
let nextTurn = 1;
|
|
512
|
+
async function openTurn(turn, taskIdentity) {
|
|
513
|
+
requireThat(turn.index === nextTurn, "session-custody-turn-order");
|
|
514
|
+
requireThat(sha(stableSessionFile(input.planPath, 2 * 1024 * 1024, "session-custody-plan")) === planSha256
|
|
515
|
+
&& sha(stableSessionFile(input.publicContractPath, 2 * 1024 * 1024,
|
|
516
|
+
"session-custody-public-contract")) === publicContractSha256,
|
|
517
|
+
"session-custody-static-drift");
|
|
518
|
+
const workspaceSha256 = benchmarkTreeStatIdentity(workspace,
|
|
519
|
+
{ rejectSymlinks: true, rejectEscapingSymlinks: true }).contentDigest,
|
|
520
|
+
resolved = resources(await input.resolveResources(Object.freeze({ surface: input.surface,
|
|
521
|
+
turnIndex: turn.index, turnId: turn.id, inputText: turn.message,
|
|
522
|
+
operatorRequest: turn.operatorRequest, expectedUserMessages: turns.slice(0, turn.index)
|
|
523
|
+
.map(item => item.operatorRequest), workspace })), turns.slice(0, turn.index)
|
|
524
|
+
.map(item => item.operatorRequest), workspace),
|
|
525
|
+
verificationHost = prepareVerificationHost(resolved.verificationHost, workspace, input.nodeCommand),
|
|
526
|
+
verifications = verificationHost ? Object.freeze([verificationHost.manifest]) : resolved.verifications,
|
|
527
|
+
contextPolicySha256 = scopedContextPolicySha256(resolved.contextPolicy),
|
|
528
|
+
measurementBinding = { version: "benchmark-turn-binding-v1", authority: "none", runId: input.runId,
|
|
529
|
+
armId: input.armId, suiteId: input.suiteId, scenarioId: input.scenarioId, surface: input.surface,
|
|
530
|
+
repeat: input.repeat, infrastructureAttempt: input.infrastructureAttempt, turnIndex: turn.index,
|
|
531
|
+
turnId: turn.id, catalogSha256, configurationSha256: input.configurationSha256,
|
|
532
|
+
publicContractSha256, planSha256, turnBindingSha256: turn.catalog.bindingDigest, workspaceSha256,
|
|
533
|
+
operatorRequestDigest: turn.catalog.operatorRequestDigest, profile: resolved.profile,
|
|
534
|
+
materialManifestSha256: scopedBrokerMaterialManifestSha256(resolved.profile, resolved.materials),
|
|
535
|
+
verificationManifestSha256: scopedBrokerVerificationManifestSha256(verifications),
|
|
536
|
+
contextPolicySha256, nodeSha256, runtimeSha256,
|
|
537
|
+
modelSha256 };
|
|
538
|
+
if (turn.index === input.expectedArmBinding.turnIndex) {
|
|
539
|
+
const actualArmDigest = scopedBrokerArmDigest({ armId: input.armId,
|
|
540
|
+
sourceSha256: frozen.sourceSha256, assetTreeSha256: frozen.assetTreeSha256,
|
|
541
|
+
brokerClosureSha256: frozen.brokerClosureSha256,
|
|
542
|
+
toolDefinitionsSha256: scopedToolDefinitionsSha256(), contextPolicySha256,
|
|
543
|
+
sdkTreeSha256: frozen.sdkTreeSha256 });
|
|
544
|
+
requireThat(actualArmDigest === input.expectedArmBinding.sha256, "session-custody-arm-drift");
|
|
545
|
+
}
|
|
546
|
+
const custody = await createBenchmarkScopedBrokerTurnCustody({ custodyRoot: input.custodyRoot,
|
|
547
|
+
nodeCommand: input.nodeCommand, brokerScript: input.brokerScript, runtimePath: input.runtimePath,
|
|
548
|
+
qualification: input.qualification, measurementBinding, contextPolicy: resolved.contextPolicy,
|
|
549
|
+
modelIdentity: input.modelIdentity, taskIdentity, materialRoot: resolved.materialRoot,
|
|
550
|
+
materials: resolved.materials, verifications, verificationBridge: resolved.verificationBridge,
|
|
551
|
+
verificationHost });
|
|
552
|
+
if (turn.index === input.expectedArmBinding.turnIndex
|
|
553
|
+
&& scopedBrokerArmDigest(custody.identity) !== input.expectedArmBinding.sha256) {
|
|
554
|
+
await custody.dispose();
|
|
555
|
+
fail("session-custody-arm-drift");
|
|
556
|
+
}
|
|
557
|
+
requireThat(benchmarkTreeStatIdentity(workspace,
|
|
558
|
+
{ rejectSymlinks: true, rejectEscapingSymlinks: true }).contentDigest === workspaceSha256,
|
|
559
|
+
"session-custody-workspace-drift");
|
|
560
|
+
nextTurn++;
|
|
561
|
+
if (turn.index !== input.expectedArmBinding.turnIndex) return custody;
|
|
562
|
+
return Object.freeze({ ...custody, assertPredispatch() {
|
|
563
|
+
custody.assertPredispatch();
|
|
564
|
+
requireThat(scopedBrokerArmDigest(custody.identity) === input.expectedArmBinding.sha256,
|
|
565
|
+
"session-custody-arm-drift");
|
|
566
|
+
return true;
|
|
567
|
+
} });
|
|
568
|
+
}
|
|
569
|
+
const piScopedBrokerRouter = input.surface === "piagent" ? createScopedBrokerPiOperationRouter({
|
|
570
|
+
async open(reservation, ctx) {
|
|
571
|
+
const turn = turns[nextTurn - 1];
|
|
572
|
+
requireThat(turn && reservation.inputText === turn.message && reservation.inputSha256 === sha(turn.message)
|
|
573
|
+
&& ctx?.cwd === workspace, "session-custody-pi-turn");
|
|
574
|
+
const task = activeSessionTask(workspace, reservation.sessionId);
|
|
575
|
+
requireThat(task && task.sessionId === reservation.sessionId && task.taskId && ID.test(task.taskId)
|
|
576
|
+
&& task.operatorRequest === turn.operatorRequest
|
|
577
|
+
&& task.operatorRequestDigest === turn.catalog.operatorRequestDigest
|
|
578
|
+
&& (task.trace?.outcome ?? "pending") === "pending", "session-custody-pi-task");
|
|
579
|
+
const custody = await openTurn(turn, { taskId: task.taskId, sessionId: reservation.sessionId,
|
|
580
|
+
operationId: reservation.operationRef, nonce: reservation.messageRequestId });
|
|
581
|
+
return custody.openPi();
|
|
582
|
+
}
|
|
583
|
+
}) : null;
|
|
584
|
+
const codexScopedBroker = input.surface === "codex-cli" ? Object.freeze({
|
|
585
|
+
version: CODEX_SCOPED_BROKER_TURN_FACTORY_VERSION, authority: "none",
|
|
586
|
+
async openTurn(coordinate) {
|
|
587
|
+
const turn = turns[nextTurn - 1];
|
|
588
|
+
requireThat(turn && coordinate.turnIndex === turn.index && coordinate.turnId === turn.id
|
|
589
|
+
&& coordinate.inputText === turn.message && coordinate.workspace === workspace
|
|
590
|
+
&& (coordinate.threadId === null || typeof coordinate.threadId === "string" && ID.test(coordinate.threadId)),
|
|
591
|
+
"session-custody-codex-turn");
|
|
592
|
+
const sessionId = coordinate.threadId ?? `pending-${sha(`${input.runId}\0${turn.catalog.bindingDigest}`).slice(0, 32)}`;
|
|
593
|
+
return openTurn(turn, { taskId: `plan-${sha(planBytes).slice(0, 32)}`, sessionId,
|
|
594
|
+
operationId: `turn-${turn.index}-${turn.catalog.bindingDigest.slice(0, 24)}`, nonce: randomUUID() });
|
|
595
|
+
}
|
|
596
|
+
}) : null;
|
|
597
|
+
return Object.freeze({ version: BENCHMARK_SCOPED_SESSION_CUSTODY_VERSION, authority: "none",
|
|
598
|
+
surface: input.surface, piScopedBrokerRouter, codexScopedBroker });
|
|
599
|
+
}
|