@piagent/platform 1.6.1 → 1.8.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +71 -0
- package/README.md +8 -2
- package/README.vi.md +11 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +3 -3
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +2 -2
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/model-options.md +20 -15
- package/docs/operator-manual-vietnamese.md +12 -12
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +3 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +38 -11
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +23 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +23 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +2 -2
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +3 -3
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -1
- package/docs-site/workflows.html +1 -1
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +107 -107
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +63 -50
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +69 -8
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +162 -45
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +144 -128
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +395 -33
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +3 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +264 -70
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +139 -152
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +1 -0
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/model-route-policy.ts +33 -13
- package/packages/piagent-core/runtime/model/model-route-types.ts +10 -6
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +12 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +11 -11
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DMU7Xe_O.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-Cui9AAV_.js +176 -0
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +141 -19
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +14 -6
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +65 -12
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +32 -1
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +30 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-model-scope.sh +6 -2
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/install-global.sh +2 -2
- package/scripts/model-route-evaluation-core.ts +1 -0
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-model-route.mjs +1 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/setup.sh +2 -2
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +37 -56
- package/scripts/verify-release-identity.mjs +5 -3
- package/templates/global/settings.json +5 -1
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -1,4 +1,5 @@
|
|
|
1
1
|
import crypto from "node:crypto";
|
|
2
|
+
import { currentWorkspaceRevisionDigest } from "../../extensions/workspace-revision.js";
|
|
2
3
|
|
|
3
4
|
import { hasDurableContextEvidence } from "../../extensions/context-evidence.js";
|
|
4
5
|
import type { TaskContract } from "../../extensions/guard-types.ts";
|
|
@@ -9,6 +10,7 @@ import { createTrajectoryState, createTrajectoryTransition, reduceTrajectory, tr
|
|
|
9
10
|
import { appendTrajectoryTransition, readTrajectoryStore, writeTrajectoryState } from "./trajectory-store.ts";
|
|
10
11
|
import type { TrajectoryRecommendationRef, TrajectorySourceHook, TrajectoryState, TrajectoryTransitionEvent } from "./trajectory-types.ts";
|
|
11
12
|
import type { SolverDecision } from "../solver/solver-types.ts";
|
|
13
|
+
import { buildOpenAiCodexWireFingerprint } from "../model/provider-wire-fingerprint.ts";
|
|
12
14
|
|
|
13
15
|
export type TrajectorySyncOptions = {
|
|
14
16
|
sourceHook: TrajectorySourceHook;
|
|
@@ -57,7 +59,8 @@ function phaseObservation(
|
|
|
57
59
|
task: TaskContract,
|
|
58
60
|
phase: string,
|
|
59
61
|
options: TrajectorySyncOptions,
|
|
60
|
-
currentTreeDigest?: string | null
|
|
62
|
+
currentTreeDigest?: string | null,
|
|
63
|
+
workspaceRevisionDigest?: string | null
|
|
61
64
|
): TrajectoryTransitionEvent["cause"] | undefined {
|
|
62
65
|
const latest = latestVerification(task);
|
|
63
66
|
if (phase === "scout" && (options.contextObserved === true || hasDurableContextEvidence(task) || stepObserved(task, ["scope", "challenge", "scout"]))) return "context-observed";
|
|
@@ -68,7 +71,7 @@ function phaseObservation(
|
|
|
68
71
|
}
|
|
69
72
|
if (phase === "verify") {
|
|
70
73
|
if (options.verificationStarted === true) return "verification-started";
|
|
71
|
-
if (verificationEvidenceProvesStableTree(latest, currentTreeDigest)) return "verification-passed";
|
|
74
|
+
if (verificationEvidenceProvesStableTree(latest, currentTreeDigest, workspaceRevisionDigest)) return "verification-passed";
|
|
72
75
|
}
|
|
73
76
|
if (phase === "review" && stepObserved(task, ["review"])) return "review-observed";
|
|
74
77
|
if (phase === "handoff" && options.handoffObserved === true) return "handoff-observed";
|
|
@@ -98,8 +101,19 @@ export class TrajectoryRuntime {
|
|
|
98
101
|
}
|
|
99
102
|
|
|
100
103
|
sync(cwd: string, sessionId: string, task: TaskContract, options: TrajectorySyncOptions): TrajectorySyncResult {
|
|
104
|
+
try { return this.syncDurable(cwd, sessionId, task, options); }
|
|
105
|
+
catch (error) {
|
|
106
|
+
wireSessions.get(wireKey(cwd, sessionId))?.invalidate("trajectory-sync-failed");
|
|
107
|
+
throw error;
|
|
108
|
+
}
|
|
109
|
+
}
|
|
110
|
+
|
|
111
|
+
private syncDurable(cwd: string, sessionId: string, task: TaskContract, options: TrajectorySyncOptions): TrajectorySyncResult {
|
|
101
112
|
const stored = readTrajectoryStore(cwd, task.taskRunId);
|
|
102
|
-
if (!stored.enforcementSafe)
|
|
113
|
+
if (!stored.enforcementSafe) {
|
|
114
|
+
wireSessions.get(wireKey(cwd, sessionId))?.invalidate("trajectory-store-corrupt");
|
|
115
|
+
return { status: "corrupt", initialized: false, enforcementSafe: false, transitions: [], warnings: stored.warnings };
|
|
116
|
+
}
|
|
103
117
|
let initialized = false;
|
|
104
118
|
let state = stored.state;
|
|
105
119
|
if (!state) {
|
|
@@ -123,6 +137,7 @@ export class TrajectoryRuntime {
|
|
|
123
137
|
const transitions: TrajectoryTransitionEvent[] = [];
|
|
124
138
|
const observedAt = options.observedAt ?? new Date().toISOString();
|
|
125
139
|
const taskDigest = digest(task), currentTreeDigest = treeDigest(task);
|
|
140
|
+
const workspaceRevisionDigest = currentWorkspaceRevisionDigest(cwd);
|
|
126
141
|
const latest = latestVerification(task);
|
|
127
142
|
if (state.currentPhase === "intake" && sourcePlanningAuthorized(task)) {
|
|
128
143
|
const path = trajectoryPath(state.changeMode, state.riskLane);
|
|
@@ -184,7 +199,7 @@ export class TrajectoryRuntime {
|
|
|
184
199
|
state = recordTransition(cwd, state, event);
|
|
185
200
|
transitions.push(event);
|
|
186
201
|
}
|
|
187
|
-
const passingVerifierAfterRepair = verificationEvidenceProvesStableTree(latest, currentTreeDigest)
|
|
202
|
+
const passingVerifierAfterRepair = verificationEvidenceProvesStableTree(latest, currentTreeDigest, workspaceRevisionDigest)
|
|
188
203
|
&& Date.parse(String(latest?.observedAt ?? latest?.recordedAt)) >= Date.parse(state.updatedAt);
|
|
189
204
|
if (state.currentPhase === "repair" && (options.verificationStarted || passingVerifierAfterRepair)) {
|
|
190
205
|
const event = createTrajectoryTransition(state, { to: "verify", cause: "verification-started", sourceHook: options.sourceHook, taskDigest, treeDigest: currentTreeDigest, observedAt });
|
|
@@ -195,7 +210,7 @@ export class TrajectoryRuntime {
|
|
|
195
210
|
while (state.currentPhase !== "terminal") {
|
|
196
211
|
const currentIndex = path.indexOf(state.currentPhase);
|
|
197
212
|
const next = currentIndex >= 0 ? path[currentIndex + 1] : undefined;
|
|
198
|
-
const cause = next && next !== "terminal" ? phaseObservation(task, next, options, currentTreeDigest) : undefined;
|
|
213
|
+
const cause = next && next !== "terminal" ? phaseObservation(task, next, options, currentTreeDigest, workspaceRevisionDigest) : undefined;
|
|
199
214
|
if (!next || next === "terminal" || !cause) break;
|
|
200
215
|
const event = createTrajectoryTransition(state, { to: next, cause, sourceHook: options.sourceHook, taskDigest, treeDigest: currentTreeDigest, observedAt });
|
|
201
216
|
state = recordTransition(cwd, state, event);
|
|
@@ -219,6 +234,7 @@ export class TrajectoryRuntime {
|
|
|
219
234
|
state = recordTransition(cwd, state, event);
|
|
220
235
|
transitions.push(event);
|
|
221
236
|
}
|
|
237
|
+
observeHostWireTrajectory(cwd, sessionId, task, state, transitions, initialized);
|
|
222
238
|
return { status: "ok", initialized, enforcementSafe: true, state, transitions, warnings: stored.warnings };
|
|
223
239
|
}
|
|
224
240
|
|
|
@@ -234,3 +250,174 @@ export class TrajectoryRuntime {
|
|
|
234
250
|
return task ? this.syncToolCall(cwd, sessionId, task, event, observedAt) : undefined;
|
|
235
251
|
}
|
|
236
252
|
}
|
|
253
|
+
|
|
254
|
+
export const HOST_WIRE_PHASE_EDGES: Record<string, string[]> = {
|
|
255
|
+
intake: ["scout", "plan", "execute", "review", "handoff", "terminal"],
|
|
256
|
+
scout: ["plan", "review", "handoff", "terminal"], plan: ["execute", "review", "handoff", "terminal"],
|
|
257
|
+
execute: ["verify", "repair", "handoff", "terminal"], verify: ["repair", "review", "handoff", "terminal"],
|
|
258
|
+
repair: ["verify", "handoff", "terminal"], review: ["repair", "handoff", "terminal"], handoff: ["terminal"], terminal: []
|
|
259
|
+
};
|
|
260
|
+
for (const targets of Object.values(HOST_WIRE_PHASE_EDGES)) Object.freeze(targets);
|
|
261
|
+
Object.freeze(HOST_WIRE_PHASE_EDGES);
|
|
262
|
+
type WireRecord = Record<string, any>;
|
|
263
|
+
export type HostWireSessionOptions = {
|
|
264
|
+
cwd: string; sessionId: string; runtimeInstanceRef: string; manifest: WireRecord;
|
|
265
|
+
signReceipt: (material: string) => string;
|
|
266
|
+
record: (envelope: { material: string; signature: string }) => void | Promise<void>;
|
|
267
|
+
};
|
|
268
|
+
export type HostWireOperation = { operationRef: string; messageRequestId: string; inputText: string };
|
|
269
|
+
export type HostWireSession = {
|
|
270
|
+
beginOperation(input: HostWireOperation): (reason: string) => void;
|
|
271
|
+
validatePayload(input: { payload: unknown; model: { provider?: string; id?: string }; hookErrors: number }): Promise<void>;
|
|
272
|
+
dispose(): void;
|
|
273
|
+
};
|
|
274
|
+
type WireInput = { turnId: string; promptHash: string; text: string; source: string; hasImages: boolean };
|
|
275
|
+
type WireState = {
|
|
276
|
+
invalidate(reason: string): void;
|
|
277
|
+
observeInput(input: WireInput): void;
|
|
278
|
+
observeTrajectory(task: TaskContract, state: TrajectoryState, events: TrajectoryTransitionEvent[], initialized: boolean): void;
|
|
279
|
+
};
|
|
280
|
+
const WIRE_REGISTRY_KEY = Symbol.for("piagent.host-wire-session.v1");
|
|
281
|
+
const wireRoot = globalThis as typeof globalThis & { [WIRE_REGISTRY_KEY]?: Map<string, WireState> };
|
|
282
|
+
const wireSessions = wireRoot[WIRE_REGISTRY_KEY] ??= new Map<string, WireState>();
|
|
283
|
+
const wireKey = (cwd: string, sessionId: string) => `${cwd}\0${sessionId}`;
|
|
284
|
+
const wireHash = (value: string) => crypto.createHash("sha256").update(value).digest("hex");
|
|
285
|
+
const wireId = (value: unknown) => typeof value === "string" && value.length > 0 && value.length <= 2048 && !value.includes("\0");
|
|
286
|
+
const wireAssert = (condition: unknown, reason: string): void => { if (!condition) throw new Error(`host-wire-${reason}`); };
|
|
287
|
+
|
|
288
|
+
export function observeHostWireInput(cwd: string, sessionId: string, input: WireInput): void {
|
|
289
|
+
wireSessions.get(wireKey(cwd, sessionId))?.observeInput(input);
|
|
290
|
+
}
|
|
291
|
+
|
|
292
|
+
function observeHostWireTrajectory(cwd: string, sessionId: string, task: TaskContract,
|
|
293
|
+
state: TrajectoryState, events: TrajectoryTransitionEvent[], initialized: boolean): void {
|
|
294
|
+
wireSessions.get(wireKey(cwd, sessionId))?.observeTrajectory(task, state, events, initialized);
|
|
295
|
+
}
|
|
296
|
+
|
|
297
|
+
/** Host callbacks are capabilities; neither a transcript entry nor a serialized task can invoke them. */
|
|
298
|
+
export function createHostWireSession(options: HostWireSessionOptions): HostWireSession {
|
|
299
|
+
const manifest = structuredClone(options.manifest), key = wireKey(options.cwd, options.sessionId);
|
|
300
|
+
wireAssert(!wireSessions.has(key), "session-already-bound");
|
|
301
|
+
wireAssert(wireId(options.sessionId) && wireId(options.runtimeInstanceRef), "runtime-identity-missing");
|
|
302
|
+
wireAssert(manifest.schemaVersion === 1 && manifest.protocol === "phase-valid-configuration-v1"
|
|
303
|
+
&& manifest.source === "pinned-host-definitions" && manifest.workingDirectory === options.cwd, "manifest-invalid");
|
|
304
|
+
wireAssert([manifest.candidateDigest, manifest.configurationDigest, manifest.definitionDigest]
|
|
305
|
+
.every((value) => /^[a-f0-9]{64}$/.test(value ?? "")), "manifest-identity-missing");
|
|
306
|
+
wireAssert(manifest.requestedTier === "fast" && manifest.requestTier === "priority", "requested-tier-mismatch");
|
|
307
|
+
wireAssert(JSON.stringify(manifest.allowedEdges) === JSON.stringify(HOST_WIRE_PHASE_EDGES), "phase-policy-mismatch");
|
|
308
|
+
wireAssert(Array.isArray(manifest.states) && manifest.states.length > 0 && manifest.states.length <= 4096, "states-invalid");
|
|
309
|
+
const selectors = new Set<string>(), stateIds = new Set<string>();
|
|
310
|
+
for (const state of manifest.states) {
|
|
311
|
+
wireAssert(state && wireId(state.id) && !stateIds.has(state.id), "state-id-invalid"); stateIds.add(state.id);
|
|
312
|
+
wireAssert([state.operatorInputHash, state.inputHash].every((value) => /^[a-f0-9]{64}$/.test(value ?? "")), "input-hash-invalid");
|
|
313
|
+
wireAssert((state.taskPresence === "none" && state.phase === null)
|
|
314
|
+
|| (state.taskPresence === "current" && Object.hasOwn(HOST_WIRE_PHASE_EDGES, state.phase)), "state-phase-invalid");
|
|
315
|
+
const selector = JSON.stringify([state.operatorInputHash, state.inputHash, state.taskPresence, state.phase]);
|
|
316
|
+
wireAssert(!selectors.has(selector), "state-selector-ambiguous"); selectors.add(selector);
|
|
317
|
+
wireAssert(state.fingerprint?.state === "known" && /^[a-f0-9]{64}$/.test(state.fingerprint.requestPrefixFingerprint), "fingerprint-invalid");
|
|
318
|
+
}
|
|
319
|
+
const publicKey = crypto.createPublicKey({ key: Buffer.from(manifest.receiptPublicKey, "base64"), format: "der", type: "spki" });
|
|
320
|
+
wireAssert(publicKey.asymmetricKeyType === "ed25519" && typeof options.signReceipt === "function"
|
|
321
|
+
&& typeof options.record === "function", "receipt-authority-missing");
|
|
322
|
+
const manifestDigest = wireHash(JSON.stringify(manifest));
|
|
323
|
+
let operation: (HostWireOperation & { operatorInputHash: string; token: string }) | undefined;
|
|
324
|
+
let accepted: (WireInput & { inputHash: string }) | undefined;
|
|
325
|
+
let task: { taskId: string; taskRunId: string; phase: string; sequence: number; outcome: string; stateDigest: string } | undefined;
|
|
326
|
+
let disposed = false, pending = false, fault: string | undefined, sequence = 0, previousReceiptHash: string | null = null;
|
|
327
|
+
const events: WireRecord[] = [];
|
|
328
|
+
const fail = (reason: string): never => { fault = reason; throw new Error(`host-wire-${reason}`); };
|
|
329
|
+
const live = () => { if (disposed || wireSessions.get(key) !== state) fail("runtime-retired"); if (fault) fail(fault); };
|
|
330
|
+
const enqueue = (event: WireRecord) => { if (events.length >= 1024) fail("event-budget-exhausted"); events.push(structuredClone(event)); };
|
|
331
|
+
const assertCurrentTrajectory = () => {
|
|
332
|
+
if (!task) return;
|
|
333
|
+
const stored = readTrajectoryStore(options.cwd, task.taskRunId);
|
|
334
|
+
if (!stored.enforcementSafe || !stored.state || digest(stored.state) !== task.stateDigest) fail("trajectory-custody-changed");
|
|
335
|
+
};
|
|
336
|
+
const state: WireState = {
|
|
337
|
+
invalidate(reason) { fault = reason; },
|
|
338
|
+
observeInput(input) {
|
|
339
|
+
live(); if (!operation || accepted) fail("input-not-admitted");
|
|
340
|
+
if (input.hasImages !== false) fail("input-images-unprojected");
|
|
341
|
+
if (!wireId(input.turnId) || !["rpc", "interactive", "extension"].includes(input.source)) fail("input-identity-invalid");
|
|
342
|
+
const inputHash = wireHash(input.text.trim());
|
|
343
|
+
if (!manifest.states.some((row: WireRecord) => row.operatorInputHash === operation!.operatorInputHash && row.inputHash === inputHash)) fail("input-not-in-manifest");
|
|
344
|
+
accepted = { ...input, inputHash };
|
|
345
|
+
enqueue({ type: "input-accepted", operationRef: operation.operationRef, messageRequestId: operation.messageRequestId,
|
|
346
|
+
operatorInputHash: operation.operatorInputHash, inputHash, turnId: input.turnId, source: input.source });
|
|
347
|
+
},
|
|
348
|
+
observeTrajectory(observedTask, observedState, transitions, initialized) {
|
|
349
|
+
if (!operation || !accepted) return;
|
|
350
|
+
live();
|
|
351
|
+
const stored = readTrajectoryStore(options.cwd, observedTask.taskRunId);
|
|
352
|
+
if (!stored.enforcementSafe || JSON.stringify(stored.state) !== JSON.stringify(observedState)) fail("trajectory-readback-mismatch");
|
|
353
|
+
if (observedTask.sessionId !== options.sessionId || observedState.taskId !== observedTask.taskId || observedState.taskRunId !== observedTask.taskRunId
|
|
354
|
+
|| observedState.sessionHash !== wireHash(options.sessionId)) fail("task-identity-mismatch");
|
|
355
|
+
if (!task || task.taskRunId !== observedTask.taskRunId) {
|
|
356
|
+
if (task && task.outcome === "pending") fail("pending-task-replaced");
|
|
357
|
+
if (!initialized || transitions.length === 0 && observedState.sequence !== 0) fail("task-attachment-unproven");
|
|
358
|
+
task = { taskId: observedTask.taskId, taskRunId: observedTask.taskRunId, phase: "intake", sequence: 0, outcome: "pending", stateDigest: "" };
|
|
359
|
+
enqueue({ type: "task-attached", initialState: createTrajectoryState({ taskId: task.taskId, taskRunId: task.taskRunId,
|
|
360
|
+
sessionId: options.sessionId, changeMode: observedState.changeMode, riskLane: observedState.riskLane,
|
|
361
|
+
recommendationRef: observedState.recommendationRef, createdAt: observedState.createdAt }) });
|
|
362
|
+
}
|
|
363
|
+
for (const event of transitions) {
|
|
364
|
+
if (event.taskRunId !== task.taskRunId || event.taskId !== task.taskId || event.from !== task.phase
|
|
365
|
+
|| event.sequence !== task.sequence + 1 || !HOST_WIRE_PHASE_EDGES[event.from]?.includes(event.to)) fail("trajectory-edge-invalid");
|
|
366
|
+
enqueue({ type: "phase-transition", event }); task.phase = event.to; task.sequence = event.sequence;
|
|
367
|
+
}
|
|
368
|
+
if (task.phase !== observedState.currentPhase || task.sequence !== observedState.sequence) fail("trajectory-observation-gap");
|
|
369
|
+
task.outcome = observedTask.trace.outcome;
|
|
370
|
+
task.stateDigest = digest(observedState);
|
|
371
|
+
}
|
|
372
|
+
};
|
|
373
|
+
wireSessions.set(key, state);
|
|
374
|
+
return {
|
|
375
|
+
beginOperation(input) {
|
|
376
|
+
live(); if (operation || pending) fail("operation-already-active");
|
|
377
|
+
if (!wireId(input.operationRef) || !wireId(input.messageRequestId) || !input.inputText.trim()) fail("operation-identity-invalid");
|
|
378
|
+
const current = { ...input, operatorInputHash: wireHash(input.inputText.trim()), token: crypto.randomUUID() };
|
|
379
|
+
if (!manifest.states.some((row: WireRecord) => row.operatorInputHash === current.operatorInputHash)) fail("operation-not-in-manifest");
|
|
380
|
+
operation = current; accepted = undefined;
|
|
381
|
+
if (task && task.outcome !== "pending" && task.phase === "terminal") {
|
|
382
|
+
enqueue({ type: "task-detached", taskId: task.taskId, taskRunId: task.taskRunId, outcome: task.outcome }); task = undefined;
|
|
383
|
+
}
|
|
384
|
+
enqueue({ type: "operation-accepted", operationRef: current.operationRef, messageRequestId: current.messageRequestId,
|
|
385
|
+
operatorInputHash: current.operatorInputHash, predecessorReceiptHash: previousReceiptHash });
|
|
386
|
+
return (reason) => {
|
|
387
|
+
if (operation !== current) return;
|
|
388
|
+
enqueue({ type: "operation-closed", operationRef: current.operationRef, messageRequestId: current.messageRequestId, reason });
|
|
389
|
+
operation = undefined; accepted = undefined;
|
|
390
|
+
};
|
|
391
|
+
},
|
|
392
|
+
async validatePayload(input) {
|
|
393
|
+
live(); if (!operation || !accepted || pending) fail("payload-without-current-input");
|
|
394
|
+
assertCurrentTrajectory();
|
|
395
|
+
if (input.hookErrors !== 0) fail("extension-hook-error");
|
|
396
|
+
const current = operation, currentInput = accepted, phase = task?.phase ?? null, taskSequence = task?.sequence ?? 0;
|
|
397
|
+
const selected = manifest.states.find((row: WireRecord) => row.operatorInputHash === current.operatorInputHash
|
|
398
|
+
&& row.inputHash === currentInput.inputHash && row.taskPresence === (task ? "current" : "none") && row.phase === phase);
|
|
399
|
+
if (!selected) fail("state-not-in-manifest");
|
|
400
|
+
const fingerprint = buildOpenAiCodexWireFingerprint({ payload: input.payload, provider: input.model.provider,
|
|
401
|
+
modelId: input.model.id, workingDirectory: options.cwd, platformRoot: manifest.platformRoot });
|
|
402
|
+
if (`${input.model.provider}/${input.model.id}` !== manifest.model || fingerprint.state !== "known"
|
|
403
|
+
|| fingerprint.reasoningEffort !== manifest.thinking || (input.payload as WireRecord)?.service_tier !== manifest.requestTier
|
|
404
|
+
|| JSON.stringify(fingerprint) !== JSON.stringify(selected.fingerprint)) fail("final-payload-drift");
|
|
405
|
+
const material = JSON.stringify({ schemaVersion: 1, protocol: manifest.protocol, manifestDigest, sequence: sequence + 1,
|
|
406
|
+
previousReceiptHash, sessionId: options.sessionId, workingDirectory: options.cwd, runtimeInstanceRef: options.runtimeInstanceRef,
|
|
407
|
+
operationRef: current.operationRef, messageRequestId: current.messageRequestId, operatorInputHash: current.operatorInputHash,
|
|
408
|
+
inputHash: currentInput.inputHash, turnId: currentInput.turnId, taskId: task?.taskId ?? null, taskRunId: task?.taskRunId ?? null,
|
|
409
|
+
phase, trajectorySequence: taskSequence, stateId: selected.id, fingerprint, events: structuredClone(events) });
|
|
410
|
+
const envelope = { material, signature: options.signReceipt(material) };
|
|
411
|
+
if (!crypto.verify(null, Buffer.from(material), publicKey, Buffer.from(envelope.signature, "base64"))) fail("receipt-signature-invalid");
|
|
412
|
+
pending = true;
|
|
413
|
+
try {
|
|
414
|
+
await options.record(structuredClone(envelope)); live(); assertCurrentTrajectory();
|
|
415
|
+
if (operation !== current || accepted !== currentInput || task?.phase !== (phase ?? undefined)
|
|
416
|
+
|| (task?.sequence ?? 0) !== taskSequence) fail("payload-binding-superseded");
|
|
417
|
+
previousReceiptHash = wireHash(JSON.stringify(envelope)); sequence += 1; events.length = 0;
|
|
418
|
+
} catch (error) { fault ??= "receipt-not-durable"; throw error; }
|
|
419
|
+
finally { pending = false; }
|
|
420
|
+
},
|
|
421
|
+
dispose() { disposed = true; operation = undefined; accepted = undefined; if (wireSessions.get(key) === state) wireSessions.delete(key); }
|
|
422
|
+
};
|
|
423
|
+
}
|
|
@@ -1,13 +1,14 @@
|
|
|
1
1
|
import crypto from "node:crypto";
|
|
2
2
|
import { isCurrentWorkingTreeDigest } from "../../extensions/working-tree-digest.js";
|
|
3
|
+
import { TRAJECTORY_PHASES, type TrajectoryPhase } from "../../extensions/trajectory-contract.ts";
|
|
4
|
+
|
|
5
|
+
export { TRAJECTORY_PHASES, type TrajectoryPhase } from "../../extensions/trajectory-contract.ts";
|
|
3
6
|
|
|
4
7
|
export const TRAJECTORY_SCHEMA_VERSION = 1 as const;
|
|
5
8
|
export const TRAJECTORY_POLICY_VERSION = "trajectory-v1" as const;
|
|
6
|
-
export const TRAJECTORY_PHASES = Object.freeze(["intake", "scout", "plan", "execute", "verify", "repair", "review", "handoff", "terminal"] as const);
|
|
7
9
|
export const TRAJECTORY_CAUSES = Object.freeze(["task-started", "context-observed", "plan-observed", "execution-authorized", "mutation-observed", "verification-started", "verification-passed", "verification-failed", "review-observed", "handoff-observed", "task-terminal", "recovery-requested", "explicit-skip"] as const);
|
|
8
10
|
export const TRAJECTORY_SOURCE_HOOKS = Object.freeze(["input", "agent-start", "tool-call", "tool-result", "completion", "session-start", "task-state", "operator"] as const);
|
|
9
11
|
|
|
10
|
-
export type TrajectoryPhase = typeof TRAJECTORY_PHASES[number];
|
|
11
12
|
export type TrajectoryCause = typeof TRAJECTORY_CAUSES[number];
|
|
12
13
|
export type TrajectorySourceHook = typeof TRAJECTORY_SOURCE_HOOKS[number];
|
|
13
14
|
export type TerminalTaskOutcomeRef = { taskRunId: string; taskUpdatedAt: string; taskDigest: string };
|