@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
|
@@ -6,7 +6,7 @@ import { redactSensitiveText } from "../../extensions/redaction-core.js";
|
|
|
6
6
|
import { workingTreeSnapshot } from "../../extensions/task-state.js";
|
|
7
7
|
import { workingTreeEvidenceDigest, workingTreeSnapshotUsesCurrentAlgorithm } from "../../extensions/working-tree-digest.js";
|
|
8
8
|
import { inspectTaskContinuationBudget } from "../recovery/continuation-budget.ts";
|
|
9
|
-
import { readHandoffProjection } from "../recovery/handoff-projection.ts";
|
|
9
|
+
import { handoffIdentityMatchesTask, readHandoffProjection } from "../recovery/handoff-projection.ts";
|
|
10
10
|
import type { ActivityInspectorEvent, CurrentActivity } from "../product/activity-inspector.ts";
|
|
11
11
|
import { recoveredToolCalls } from "./activity-recovery.ts";
|
|
12
12
|
import { classifyToolFailure, handledToolFailure } from "./tool-failure-classification.ts";
|
|
@@ -129,7 +129,7 @@ function reconciledToolResults(events: ActivityInspectorEvent[], evidence: Retur
|
|
|
129
129
|
const raw = evidence.results.get(call.toolCallId);
|
|
130
130
|
if (!raw) continue;
|
|
131
131
|
const reasonCode = classifyToolFailure(raw.toolName || call.toolName, raw.isError, raw.content, evidence.inputs.get(call.toolCallId));
|
|
132
|
-
const effectiveError = raw.isError || Boolean(reasonCode && !handledToolFailure(reasonCode));
|
|
132
|
+
const effectiveError = raw.isError || Boolean(reasonCode && !handledToolFailure(reasonCode, raw.toolName || call.toolName));
|
|
133
133
|
reconciled.push({ activityId: `canonical-result:${call.toolCallId}`, event: "tool_result",
|
|
134
134
|
recordedAt: raw.recordedAt ?? call.recordedAt, sessionId: call.sessionId, taskRunId: call.taskRunId,
|
|
135
135
|
toolCallId: call.toolCallId, toolName: raw.toolName || call.toolName, isError: effectiveError,
|
|
@@ -162,29 +162,64 @@ function sourceSummary(document: SourceChangeDocument | null, view: "task" | "wo
|
|
|
162
162
|
};
|
|
163
163
|
}
|
|
164
164
|
|
|
165
|
-
function
|
|
165
|
+
function sourceProjectionRevision(views: CriteriaLinkProjection["sourceViews"]): string {
|
|
166
|
+
const content = (["task", "workingTree", "staged"] as const).map((key) => {
|
|
167
|
+
const document = views[key];
|
|
168
|
+
if (!document) return null;
|
|
169
|
+
const { generatedAt: _generatedAt, identity: _identity, snapshotBinding: _snapshotBinding, ...stable } = document;
|
|
170
|
+
const bases = Array.isArray(stable.bases) ? stable.bases.map((value) => {
|
|
171
|
+
if (!value || typeof value !== "object") return value;
|
|
172
|
+
const { taskRevision: _taskRevision, ...sourceBasis } = value as Record<string, unknown>;
|
|
173
|
+
return sourceBasis;
|
|
174
|
+
}) : stable.bases;
|
|
175
|
+
return { ...stable, bases };
|
|
176
|
+
});
|
|
177
|
+
return token("source-projection-rev", content);
|
|
178
|
+
}
|
|
179
|
+
|
|
180
|
+
function activityProjection(events: ActivityInspectorEvent[], current: CurrentActivity[], generatedAt: string,
|
|
181
|
+
operationLiveness?: "idle" | "running" | "unknown") {
|
|
166
182
|
const results = new Map(events.filter((event) => event.event === "tool_result").map((event) => [event.toolCallId, event]));
|
|
167
183
|
const decisions = new Map(events.filter((event) => event.event === "tool_decision").map((event) => [event.toolCallId, event]));
|
|
168
184
|
const recoveries = recoveredToolCalls(events);
|
|
185
|
+
// Tool-call telemetry is deliberately written before execution. A missing
|
|
186
|
+
// result therefore cannot remain authoritative forever: compaction, abrupt
|
|
187
|
+
// host replacement, or an older runtime may have lost the matching result.
|
|
188
|
+
// A later host lifecycle boundary proves only that the call is no longer
|
|
189
|
+
// running, not that it passed or failed, so project it as terminal/unknown.
|
|
190
|
+
let lastLifecycleBoundary = -1;
|
|
191
|
+
for (let index = events.length - 1; index >= 0; index -= 1) {
|
|
192
|
+
if (!["turn_end", "agent_settled", "session_shutdown"].includes(String(events[index]?.event))) continue;
|
|
193
|
+
lastLifecycleBoundary = index;
|
|
194
|
+
break;
|
|
195
|
+
}
|
|
169
196
|
const calls = events.filter((event) => event.event === "tool_call").slice(-200);
|
|
170
197
|
const projected = calls.map((call) => {
|
|
171
198
|
const result = results.get(call.toolCallId), decision = decisions.get(call.toolCallId);
|
|
172
199
|
const blocked = decision?.decision === "blocked";
|
|
173
|
-
const handled = Boolean(result && handledToolFailure(result.reasonCode));
|
|
200
|
+
const handled = Boolean(result && handledToolFailure(result.reasonCode, result.toolName ?? call.toolName));
|
|
174
201
|
const failed = !blocked && !handled && Boolean(result && (result.isError === true || (typeof result.exitCode === "number" && result.exitCode !== 0)));
|
|
175
202
|
const recovery = call.toolCallId ? recoveries.get(call.toolCallId) : undefined;
|
|
176
|
-
const
|
|
203
|
+
const lifecycleSettled = !result && !blocked && (operationLiveness === "idle"
|
|
204
|
+
|| lastLifecycleBoundary > events.indexOf(call));
|
|
205
|
+
const state = blocked ? "blocked" : lifecycleSettled ? "unknown" : !result ? "running" : failed && !recovery ? "failed" : "passed";
|
|
177
206
|
const toolName = display(call.toolName ?? "unknown", 80).replace(/[^A-Za-z0-9._:@~-]/g, "-") || "unknown";
|
|
178
207
|
const rawId = String(call.toolCallId ?? call.activityId ?? `${call.recordedAt}:${toolName}`);
|
|
179
208
|
const command = display(call.command ?? call.targetPath ?? toolName, 65_536);
|
|
209
|
+
const boundaryAt = lastLifecycleBoundary > events.indexOf(call)
|
|
210
|
+
? timestamp(events[lastLifecycleBoundary]?.recordedAt, timestamp(call.recordedAt, generatedAt))
|
|
211
|
+
: timestamp(events.at(-1)?.recordedAt, timestamp(call.recordedAt, generatedAt));
|
|
180
212
|
return {
|
|
181
213
|
activityRef: token("activity", [rawId, call.recordedAt]), kind: ["bash", "shell", "exec"].includes(toolName) ? "command" : "tool", state,
|
|
182
|
-
label: display(recovery ? `${toolName} recovered` : handled ? `${toolName} warning`
|
|
214
|
+
label: display(recovery ? `${toolName} recovered` : handled ? `${toolName} warning`
|
|
215
|
+
: state === "unknown" ? `${toolName} ended · result unavailable`
|
|
216
|
+
: state === "running" ? `${toolName} running` : `${toolName} ${state}`, 500), preview: command,
|
|
183
217
|
toolCallId: token("tool", rawId), toolName, commandDigest: call.command ? `sha256:${hash(call.command)}` : null,
|
|
184
218
|
logRef: null, exitCode: recovery?.exitCode ?? (typeof result?.exitCode === "number" ? result.exitCode : null),
|
|
185
219
|
exitCodeExact: recovery?.exitCodeExact ?? (result?.exitCodeExact === true),
|
|
186
220
|
startedAt: timestamp(call.recordedAt), finishedAt: recovery ? timestamp(recovery.recoveredAt, timestamp(call.recordedAt))
|
|
187
|
-
: result || blocked ? timestamp(result?.recordedAt ?? decision?.recordedAt, timestamp(call.recordedAt))
|
|
221
|
+
: result || blocked ? timestamp(result?.recordedAt ?? decision?.recordedAt, timestamp(call.recordedAt))
|
|
222
|
+
: lifecycleSettled ? boundaryAt : null
|
|
188
223
|
};
|
|
189
224
|
});
|
|
190
225
|
const callIds = new Set(calls.map((call) => String(call.toolCallId ?? "")));
|
|
@@ -213,7 +248,7 @@ function capabilities(identity: WebUiIdentity, generatedAt: string, resyncRequir
|
|
|
213
248
|
capabilities: {
|
|
214
249
|
inspect: { status: "available", version: 1, reason: null, sourceViews: ["task", "working-tree", "staged"] },
|
|
215
250
|
"control.chat": unavailableCapability("chat-control-not-enabled", "The same-session bridge exists, but transcript, queue, server control and browser UX are not enabled"),
|
|
216
|
-
"control.lifecycle": unavailableCapability("lifecycle-contract-incomplete", "Pi
|
|
251
|
+
"control.lifecycle": unavailableCapability("lifecycle-contract-incomplete", "Pi host lacks semantic Pause and an acknowledged phase-complete Stop contract"),
|
|
217
252
|
"control.resumeAndContinue": unavailableCapability("compound-control-unavailable", "Resume is unavailable until a durable acknowledged Pause contract exists"),
|
|
218
253
|
"control.sessionOptions": unavailableCapability("effect-scope-not-productionized", "Model and thinking APIs exist but lifecycle and persistence scope enforcement is not enabled"),
|
|
219
254
|
attachments: unavailableCapability("attachments-not-implemented", "Attachments are not implemented"),
|
|
@@ -299,7 +334,7 @@ function handoff(cwd: string, task: TaskContract | undefined, currentDigest: str
|
|
|
299
334
|
if (!task) return null;
|
|
300
335
|
let value;
|
|
301
336
|
try { value = readHandoffProjection(cwd, task.taskRunId); } catch { return null; }
|
|
302
|
-
if (!value) return null;
|
|
337
|
+
if (!value || !handoffIdentityMatchesTask(value.identity, task)) return null;
|
|
303
338
|
const stale = Boolean(currentDigest && value.tree.currentDigest && currentDigest !== value.tree.currentDigest);
|
|
304
339
|
return { handoffRef: token("handoff", task.taskRunId), state: stale ? "stale" : "ready", summary: display(value.goal.summary, 500),
|
|
305
340
|
blocker: value.failure.warnings[0] ? display(value.failure.warnings[0], 500) : null,
|
|
@@ -324,6 +359,7 @@ export async function buildWebUiInspectionProjection(input: {
|
|
|
324
359
|
eventReplay?: { eventRetentionCount: number; eventRetentionSeconds: number };
|
|
325
360
|
model?: WebUiRuntimeModel;
|
|
326
361
|
thinkingLevel?: unknown;
|
|
362
|
+
operationLiveness?: "idle" | "running" | "unknown";
|
|
327
363
|
}): Promise<WebUiInspectionProjection> {
|
|
328
364
|
const generatedAt = timestamp(input.generatedAt);
|
|
329
365
|
const identity: WebUiIdentity = { projectRef: webUiProjectRef(input.cwd), runtimeInstanceId: input.runtimeInstanceId ?? WEBUI_RUNTIME_INSTANCE_REF,
|
|
@@ -339,14 +375,18 @@ export async function buildWebUiInspectionProjection(input: {
|
|
|
339
375
|
const events = reconciledToolResults(scopedEvents(input.events ?? [], input.sessionId, input.task), canonicalEvidence);
|
|
340
376
|
const linked = projectCriteriaFileVerifier({ cwd: input.cwd, task: input.task, sourceViews: views, currentSnapshot,
|
|
341
377
|
protectedPaths: input.protectedPaths, events, at: new Date(generatedAt) });
|
|
342
|
-
const sourceChanges = {
|
|
378
|
+
const sourceChanges = { projectionRevision: sourceProjectionRevision(linked.sourceViews),
|
|
379
|
+
task: sourceSummary(linked.sourceViews.task, "task"), workingTree: sourceSummary(linked.sourceViews.workingTree, "working-tree"),
|
|
380
|
+
staged: sourceSummary(linked.sourceViews.staged, "staged") };
|
|
343
381
|
const currentDigest = workingTreeSnapshotUsesCurrentAlgorithm(currentSnapshot) ? workingTreeEvidenceDigest(currentSnapshot) : null;
|
|
344
|
-
const current = (input.current ?? []).filter((item) => (item.status ?? "running") !== "running" ||
|
|
382
|
+
const current = (input.current ?? []).filter((item) => ((item.status ?? "running") !== "running" || input.operationLiveness !== "idle")
|
|
383
|
+
&& ((item.status ?? "running") !== "running" || !canonicalEvidence.results.has(item.toolCallId)));
|
|
345
384
|
const runningCurrent = current.find((item) => (item.status ?? "running") === "running");
|
|
346
385
|
const controlState = input.task?.trace.outcome === "pending" ? "active" as const : "terminal" as const;
|
|
347
386
|
const context = contextProjection(input.contextUsage, generatedAt), usageFacts = usage(input.sessionEntries ?? []);
|
|
348
387
|
const eventCursor = input.eventCursor ?? token("event-cursor", events.map((event) => event.activityId ?? [event.event, event.toolCallId, event.recordedAt]));
|
|
349
|
-
const runtimeRevision = token("runtime-rev", [taskRevision, sourceChanges.workingTree.revision, sourceChanges.staged.revision,
|
|
388
|
+
const runtimeRevision = token("runtime-rev", [taskRevision, sourceChanges.workingTree.revision, sourceChanges.staged.revision,
|
|
389
|
+
eventCursor, current, input.operationLiveness ?? null]);
|
|
350
390
|
const approvals = { state: "unknown", pending: [], recent: [], health: { state: "unavailable", reasonCode: "approval-projection-unavailable", message: "Current host approval state is not exposed to the read-only projector" } };
|
|
351
391
|
const task = input.task ? taskProjection(input.task, linked.criteria, controlState) : null;
|
|
352
392
|
const issues = [
|
|
@@ -364,8 +404,8 @@ export async function buildWebUiInspectionProjection(input: {
|
|
|
364
404
|
capabilities: capabilities(identity, generatedAt, input.resyncRequired === true, input.eventReplay),
|
|
365
405
|
session: {
|
|
366
406
|
connectionState: input.resyncRequired ? "resync-required" : "connected", connectionReason: input.resyncRequired ? "event-replay-gap" : null, displayName: input.task?.sessionName ?? null,
|
|
367
|
-
operation: runningCurrent
|
|
368
|
-
? { liveness: "running", operationRef: null, hostPhase: { state: "known", value: "tool", evidence: "derived", reasonCode: null }, startedAt: timestamp(runningCurrent.startedAt, generatedAt), settledAt: null, reasonCode: null }
|
|
407
|
+
operation: input.operationLiveness === "running" || runningCurrent
|
|
408
|
+
? { liveness: "running", operationRef: null, hostPhase: { state: "known", value: runningCurrent ? "tool" : "other", evidence: "derived", reasonCode: null }, startedAt: runningCurrent ? timestamp(runningCurrent.startedAt, generatedAt) : null, settledAt: null, reasonCode: null }
|
|
369
409
|
: { liveness: "idle", operationRef: null, hostPhase: { state: "known", value: "idle", evidence: "derived", reasonCode: null }, startedAt: null, settledAt: null, reasonCode: null },
|
|
370
410
|
controlState: input.task ? controlState : "active", taskOutcome: input.task?.trace.outcome ?? null, approvalState: approvals.state,
|
|
371
411
|
verificationState: linked.verification.state, permissionProfile: { state: "unavailable", value: null, evidence: null, reasonCode: "permission-profile-not-projected" },
|
|
@@ -373,12 +413,18 @@ export async function buildWebUiInspectionProjection(input: {
|
|
|
373
413
|
thinking: thinkingFact(input.thinkingLevel),
|
|
374
414
|
queue: { state: "unavailable", hasPending: null, heldCount: null, revision: null, reasonCode: "host-queue-api-unavailable" }, context
|
|
375
415
|
},
|
|
376
|
-
task, sourceChanges, activity: activityProjection(events, current), approvals, verification: linked.verification,
|
|
416
|
+
task, sourceChanges, activity: activityProjection(events, current, generatedAt, input.operationLiveness), approvals, verification: linked.verification,
|
|
377
417
|
usage: { context, latestTurn: usageCounter(usageFacts.latest, "no-assistant-turn"), sessionTotal: usageCounter(usageFacts.observed ? usageFacts.totals : null, "host-total-usage-unavailable"),
|
|
378
418
|
taskTotal: usageCounter(null, input.task ? "task-usage-boundary-unavailable" : "no-active-task"), capturedAt: generatedAt,
|
|
379
419
|
health: { state: "degraded", reasonCode: "partial-usage-only", message: "Task-attributed usage is not available" } },
|
|
380
420
|
continuation: continuation(input.cwd, input.task), handoff: handoff(input.cwd, input.task, currentDigest),
|
|
381
421
|
health: { state: issues.length ? "degraded" : "ok", issues, resyncRequired: input.resyncRequired === true, generatedFromRevision: runtimeRevision }
|
|
382
422
|
};
|
|
423
|
+
const snapshotBinding = { sourceProjectionRevision: snapshot.sourceChanges.projectionRevision,
|
|
424
|
+
taskViewRevision: snapshot.sourceChanges.task.revision, workspaceRevision: snapshot.sourceChanges.workingTree.revision,
|
|
425
|
+
indexRevision: snapshot.sourceChanges.staged.revision };
|
|
426
|
+
for (const document of Object.values(linked.sourceViews)) {
|
|
427
|
+
if (document) (document as SourceChangeDocument & { snapshotBinding: typeof snapshotBinding }).snapshotBinding = snapshotBinding;
|
|
428
|
+
}
|
|
383
429
|
return { snapshot, sourceViews: linked.sourceViews, relations: linked.relations, verifierAttempts: linked.verifierAttempts, scopedEvents: events };
|
|
384
430
|
}
|
|
@@ -1,132 +1,2 @@
|
|
|
1
|
-
|
|
2
|
-
|
|
3
|
-
import path from "node:path";
|
|
4
|
-
|
|
5
|
-
export class WorkspaceReadError extends Error {
|
|
6
|
-
readonly code: "outside-root" | "symlink-ancestor" | "not-file" | "oversized" | "changed-during-read";
|
|
7
|
-
|
|
8
|
-
constructor(code: WorkspaceReadError["code"], message: string) {
|
|
9
|
-
super(message);
|
|
10
|
-
this.name = "WorkspaceReadError";
|
|
11
|
-
this.code = code;
|
|
12
|
-
}
|
|
13
|
-
}
|
|
14
|
-
|
|
15
|
-
type SafeCandidate = { root: string; absolute: string; parent: string };
|
|
16
|
-
|
|
17
|
-
function inside(root: string, candidate: string): boolean {
|
|
18
|
-
const relative = path.relative(root, candidate);
|
|
19
|
-
return relative === "" || (relative !== ".." && !relative.startsWith(`..${path.sep}`) && !path.isAbsolute(relative));
|
|
20
|
-
}
|
|
21
|
-
|
|
22
|
-
function safeCandidate(repoRoot: string, repoPath: string): SafeCandidate {
|
|
23
|
-
const root = fs.realpathSync.native(repoRoot);
|
|
24
|
-
const segments = repoPath.split("/");
|
|
25
|
-
if (!repoPath || path.posix.isAbsolute(repoPath) || segments.includes("..") || segments.includes("")) {
|
|
26
|
-
throw new WorkspaceReadError("outside-root", "Workspace path is not a safe repository-relative path");
|
|
27
|
-
}
|
|
28
|
-
let parent = root;
|
|
29
|
-
for (const segment of segments.slice(0, -1)) {
|
|
30
|
-
parent = path.join(parent, segment);
|
|
31
|
-
const stat = fs.lstatSync(parent);
|
|
32
|
-
if (stat.isSymbolicLink()) throw new WorkspaceReadError("symlink-ancestor", "Workspace path traverses a symbolic link");
|
|
33
|
-
if (!stat.isDirectory()) throw new WorkspaceReadError("not-file", "Workspace path has a non-directory ancestor");
|
|
34
|
-
}
|
|
35
|
-
const canonicalParent = fs.realpathSync.native(parent);
|
|
36
|
-
if (!inside(root, canonicalParent)) throw new WorkspaceReadError("outside-root", "Workspace path resolves outside its repository root");
|
|
37
|
-
const absolute = path.join(canonicalParent, segments.at(-1) as string);
|
|
38
|
-
if (!inside(root, absolute)) throw new WorkspaceReadError("outside-root", "Workspace path escapes its repository root");
|
|
39
|
-
return { root, absolute, parent: canonicalParent };
|
|
40
|
-
}
|
|
41
|
-
|
|
42
|
-
function sameFile(left: fs.Stats, right: fs.Stats): boolean {
|
|
43
|
-
return left.dev === right.dev && left.ino === right.ino && left.mode === right.mode;
|
|
44
|
-
}
|
|
45
|
-
|
|
46
|
-
function openRegularFile(repoRoot: string, repoPath: string): { descriptor: number; stat: fs.Stats; candidate: SafeCandidate } {
|
|
47
|
-
const candidate = safeCandidate(repoRoot, repoPath);
|
|
48
|
-
const descriptor = fs.openSync(candidate.absolute, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0));
|
|
49
|
-
try {
|
|
50
|
-
const stat = fs.fstatSync(descriptor);
|
|
51
|
-
if (!stat.isFile()) throw new WorkspaceReadError("not-file", "Workspace entry is not a regular file");
|
|
52
|
-
const canonical = fs.realpathSync.native(candidate.absolute);
|
|
53
|
-
const current = fs.lstatSync(candidate.absolute);
|
|
54
|
-
if (!inside(candidate.root, canonical) || !sameFile(stat, current)) {
|
|
55
|
-
throw new WorkspaceReadError("changed-during-read", "Workspace path changed while it was opened");
|
|
56
|
-
}
|
|
57
|
-
return { descriptor, stat, candidate };
|
|
58
|
-
} catch (error) {
|
|
59
|
-
fs.closeSync(descriptor);
|
|
60
|
-
throw error;
|
|
61
|
-
}
|
|
62
|
-
}
|
|
63
|
-
|
|
64
|
-
function verifyAfterRead(descriptor: number, before: fs.Stats, candidate: SafeCandidate): void {
|
|
65
|
-
const after = fs.fstatSync(descriptor);
|
|
66
|
-
const current = fs.lstatSync(candidate.absolute);
|
|
67
|
-
const canonical = fs.realpathSync.native(candidate.absolute);
|
|
68
|
-
if (
|
|
69
|
-
!sameFile(before, after)
|
|
70
|
-
|| !sameFile(after, current)
|
|
71
|
-
|| before.size !== after.size
|
|
72
|
-
|| before.mtimeMs !== after.mtimeMs
|
|
73
|
-
|| before.ctimeMs !== after.ctimeMs
|
|
74
|
-
|| !inside(candidate.root, canonical)
|
|
75
|
-
) {
|
|
76
|
-
throw new WorkspaceReadError("changed-during-read", "Workspace file changed during inspection");
|
|
77
|
-
}
|
|
78
|
-
}
|
|
79
|
-
|
|
80
|
-
export function inspectWorkspaceEntry(repoRoot: string, repoPath: string): fs.Stats {
|
|
81
|
-
const candidate = safeCandidate(repoRoot, repoPath);
|
|
82
|
-
return fs.lstatSync(candidate.absolute);
|
|
83
|
-
}
|
|
84
|
-
|
|
85
|
-
export function readWorkspaceLink(repoRoot: string, repoPath: string): string {
|
|
86
|
-
const candidate = safeCandidate(repoRoot, repoPath);
|
|
87
|
-
const before = fs.lstatSync(candidate.absolute);
|
|
88
|
-
if (!before.isSymbolicLink()) throw new WorkspaceReadError("not-file", "Workspace entry is not a symbolic link");
|
|
89
|
-
const target = fs.readlinkSync(candidate.absolute);
|
|
90
|
-
const after = fs.lstatSync(candidate.absolute);
|
|
91
|
-
if (!sameFile(before, after) || before.mtimeMs !== after.mtimeMs || before.ctimeMs !== after.ctimeMs) {
|
|
92
|
-
throw new WorkspaceReadError("changed-during-read", "Workspace link changed during inspection");
|
|
93
|
-
}
|
|
94
|
-
return target;
|
|
95
|
-
}
|
|
96
|
-
|
|
97
|
-
export function readWorkspaceFile(repoRoot: string, repoPath: string, maxBytes: number): Buffer {
|
|
98
|
-
const opened = openRegularFile(repoRoot, repoPath);
|
|
99
|
-
try {
|
|
100
|
-
if (opened.stat.size > maxBytes) throw new WorkspaceReadError("oversized", "Workspace file exceeds its read cap");
|
|
101
|
-
const content = Buffer.allocUnsafe(opened.stat.size);
|
|
102
|
-
let position = 0;
|
|
103
|
-
while (position < content.length) {
|
|
104
|
-
const count = fs.readSync(opened.descriptor, content, position, content.length - position, position);
|
|
105
|
-
if (count <= 0) throw new WorkspaceReadError("changed-during-read", "Workspace file ended during inspection");
|
|
106
|
-
position += count;
|
|
107
|
-
}
|
|
108
|
-
verifyAfterRead(opened.descriptor, opened.stat, opened.candidate);
|
|
109
|
-
return content;
|
|
110
|
-
} finally {
|
|
111
|
-
fs.closeSync(opened.descriptor);
|
|
112
|
-
}
|
|
113
|
-
}
|
|
114
|
-
|
|
115
|
-
export function hashWorkspaceFile(repoRoot: string, repoPath: string): string {
|
|
116
|
-
const opened = openRegularFile(repoRoot, repoPath);
|
|
117
|
-
try {
|
|
118
|
-
const state = createHash("sha256");
|
|
119
|
-
const buffer = Buffer.allocUnsafe(256 * 1024);
|
|
120
|
-
let position = 0;
|
|
121
|
-
while (position < opened.stat.size) {
|
|
122
|
-
const count = fs.readSync(opened.descriptor, buffer, 0, Math.min(buffer.length, opened.stat.size - position), position);
|
|
123
|
-
if (count <= 0) throw new WorkspaceReadError("changed-during-read", "Workspace file ended during hashing");
|
|
124
|
-
state.update(buffer.subarray(0, count));
|
|
125
|
-
position += count;
|
|
126
|
-
}
|
|
127
|
-
verifyAfterRead(opened.descriptor, opened.stat, opened.candidate);
|
|
128
|
-
return state.digest("hex");
|
|
129
|
-
} finally {
|
|
130
|
-
fs.closeSync(opened.descriptor);
|
|
131
|
-
}
|
|
132
|
-
}
|
|
1
|
+
// Compatibility entrypoint; core and runtime share the same bounded file reader.
|
|
2
|
+
export * from "../../security/workspace-file-reader.ts";
|
|
@@ -4,8 +4,11 @@ import {
|
|
|
4
4
|
authenticatedCatalogDigest,
|
|
5
5
|
CAPABILITY_BANDS,
|
|
6
6
|
createModelRouteDecision,
|
|
7
|
+
LEGACY_OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION,
|
|
8
|
+
OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION,
|
|
7
9
|
type CapabilityBand,
|
|
8
10
|
type ModelRouteDecision,
|
|
11
|
+
type ModelRouteMappingVersion,
|
|
9
12
|
type ModelSelectionSource,
|
|
10
13
|
type ParentRoutingMode,
|
|
11
14
|
type RoutingObjective
|
|
@@ -22,33 +25,47 @@ export type ModelRoutePolicyInput = {
|
|
|
22
25
|
current: { provider: string | null; modelId: string | null; effort: string | null };
|
|
23
26
|
freshTaskBoundary: boolean;
|
|
24
27
|
hostBoundary: ModelRouteHostBoundary;
|
|
28
|
+
mappingVersion?: ModelRouteMappingVersion;
|
|
25
29
|
};
|
|
26
30
|
|
|
27
31
|
type Candidate = { provider: "openai-codex"; modelId: string; effort: string };
|
|
28
32
|
|
|
29
|
-
const
|
|
33
|
+
const LEGACY_BAND_CANDIDATES: Record<CapabilityBand, Candidate[]> = {
|
|
30
34
|
low: [{ provider: "openai-codex", modelId: "gpt-5.6-luna", effort: "medium" }],
|
|
31
35
|
medium: [{ provider: "openai-codex", modelId: "gpt-5.6-terra", effort: "medium" }],
|
|
32
36
|
high: [{ provider: "openai-codex", modelId: "gpt-5.6-sol", effort: "high" }],
|
|
33
37
|
ultra: [{ provider: "openai-codex", modelId: "gpt-5.6-sol", effort: "xhigh" }]
|
|
34
38
|
};
|
|
39
|
+
const BAND_CANDIDATES: Record<CapabilityBand, Candidate[]> = {
|
|
40
|
+
low: [{ provider: "openai-codex", modelId: "gpt-6-luna", effort: "medium" }, ...LEGACY_BAND_CANDIDATES.low],
|
|
41
|
+
medium: [{ provider: "openai-codex", modelId: "gpt-6-sol", effort: "medium" }, ...LEGACY_BAND_CANDIDATES.medium],
|
|
42
|
+
high: [{ provider: "openai-codex", modelId: "gpt-6-sol", effort: "high" }, ...LEGACY_BAND_CANDIDATES.high],
|
|
43
|
+
ultra: [{ provider: "openai-codex", modelId: "gpt-6-astra", effort: "xhigh" }, ...LEGACY_BAND_CANDIDATES.ultra]
|
|
44
|
+
};
|
|
35
45
|
const BAND_INDEX = new Map(CAPABILITY_BANDS.map((band, index) => [band, index]));
|
|
36
46
|
|
|
37
|
-
function exactCandidate(catalog: AuthenticatedModelCatalog, band: CapabilityBand): { model: AuthenticatedModelCatalogEntry; effort: string } | undefined {
|
|
38
|
-
|
|
47
|
+
function exactCandidate(catalog: AuthenticatedModelCatalog, band: CapabilityBand, mappingVersion: ModelRouteMappingVersion): { model: AuthenticatedModelCatalogEntry; effort: string; legacyFallback: boolean } | undefined {
|
|
48
|
+
const candidates = mappingVersion === LEGACY_OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION ? LEGACY_BAND_CANDIDATES[band] : BAND_CANDIDATES[band];
|
|
49
|
+
for (const [index, candidate] of candidates.entries()) {
|
|
39
50
|
const model = catalog.models.find((entry) => entry.provider === candidate.provider
|
|
40
51
|
&& entry.modelId === candidate.modelId
|
|
41
52
|
&& (entry.supportedThinkingLevels === null || entry.supportedThinkingLevels.includes(candidate.effort)));
|
|
42
|
-
if (model) return { model, effort: candidate.effort };
|
|
53
|
+
if (model) return { model, effort: candidate.effort, legacyFallback: mappingVersion === OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION && index > 0 };
|
|
43
54
|
}
|
|
44
55
|
return undefined;
|
|
45
56
|
}
|
|
46
57
|
|
|
47
|
-
function currentBand(modelId: string | null): CapabilityBand | undefined {
|
|
48
|
-
|
|
49
|
-
|
|
50
|
-
|
|
51
|
-
|
|
58
|
+
function currentBand(modelId: string | null, effort: string | null): CapabilityBand | undefined {
|
|
59
|
+
if (modelId === "gpt-6-luna" || modelId === "gpt-5.6-luna") return "low";
|
|
60
|
+
if (modelId === "gpt-6-astra") return "ultra";
|
|
61
|
+
if (modelId === "gpt-6-sol") {
|
|
62
|
+
if (effort === "medium") return "medium";
|
|
63
|
+
if (effort === "xhigh" || effort === "max") return "ultra";
|
|
64
|
+
return "high";
|
|
65
|
+
}
|
|
66
|
+
if (modelId === "gpt-5.6-terra") return "medium";
|
|
67
|
+
if (modelId === "gpt-5.6-sol") return "high";
|
|
68
|
+
return undefined;
|
|
52
69
|
}
|
|
53
70
|
|
|
54
71
|
function maxBand(left: CapabilityBand, right: CapabilityBand): CapabilityBand {
|
|
@@ -102,6 +119,7 @@ function downgradeSteps(current: CapabilityBand | undefined, target: CapabilityB
|
|
|
102
119
|
}
|
|
103
120
|
|
|
104
121
|
export function routeParentModel(input: ModelRoutePolicyInput): ModelRouteDecision {
|
|
122
|
+
const mappingVersion = input.mappingVersion ?? OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION;
|
|
105
123
|
const floor = safetyFloor(input.features);
|
|
106
124
|
const desired = objectiveBand(floor.band, input.objective);
|
|
107
125
|
const catalogDigest = authenticatedCatalogDigest(input.catalog);
|
|
@@ -114,7 +132,8 @@ export function routeParentModel(input: ModelRoutePolicyInput): ModelRouteDecisi
|
|
|
114
132
|
currentModelId: input.current.modelId,
|
|
115
133
|
currentEffort: input.current.effort,
|
|
116
134
|
selectionSource: input.selectionSource,
|
|
117
|
-
catalogDigest
|
|
135
|
+
catalogDigest,
|
|
136
|
+
mappingVersion
|
|
118
137
|
} as const;
|
|
119
138
|
if (input.mode === "off") return createModelRouteDecision({
|
|
120
139
|
...common, capabilityBand: "abstain", fallbackBand: desired, provider: input.current.provider,
|
|
@@ -136,14 +155,15 @@ export function routeParentModel(input: ModelRoutePolicyInput): ModelRouteDecisi
|
|
|
136
155
|
confidence: "low", disposition: "unavailable", reasonCodes: ["authenticated-catalog-unavailable", ...floor.reasons],
|
|
137
156
|
downgradeSteps: 0, enforced: false
|
|
138
157
|
});
|
|
139
|
-
const candidate = exactCandidate(input.catalog, desired);
|
|
158
|
+
const candidate = exactCandidate(input.catalog, desired, mappingVersion);
|
|
140
159
|
if (!candidate) return createModelRouteDecision({
|
|
141
160
|
...common, capabilityBand: "abstain", fallbackBand: desired, provider: null, modelId: null, effort: null,
|
|
142
161
|
confidence: "low", disposition: "unavailable", reasonCodes: ["preferred-model-or-effort-unavailable", "no-silent-substitution", ...floor.reasons],
|
|
143
162
|
downgradeSteps: 0, enforced: false
|
|
144
163
|
});
|
|
145
|
-
const reasons = [...floor.reasons, `objective-${input.objective}`, `candidate-${desired}`, "runtime-catalog-match"
|
|
146
|
-
|
|
164
|
+
const reasons = [...floor.reasons, `objective-${input.objective}`, `candidate-${desired}`, "runtime-catalog-match",
|
|
165
|
+
...(candidate.legacyFallback ? ["legacy-model-fallback"] : [])];
|
|
166
|
+
const steps = downgradeSteps(currentBand(input.current.modelId, input.current.effort), desired);
|
|
147
167
|
if (input.mode === "shadow") return createModelRouteDecision({
|
|
148
168
|
...common, capabilityBand: desired, fallbackBand: null, provider: candidate.model.provider, modelId: candidate.model.modelId,
|
|
149
169
|
effort: candidate.effort, confidence: floor.band === "low" || floor.band === "ultra" ? "high" : "medium",
|
|
@@ -4,7 +4,9 @@ import type { AuthenticatedModelCatalog } from "./authenticated-catalog.ts";
|
|
|
4
4
|
|
|
5
5
|
export const MODEL_ROUTE_DECISION_SCHEMA_VERSION = 1 as const;
|
|
6
6
|
export const MODEL_ROUTE_POLICY_VERSION = "model-route-v1" as const;
|
|
7
|
-
export const
|
|
7
|
+
export const LEGACY_OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION = "openai-codex-model-route-map-v1" as const;
|
|
8
|
+
export const OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION = "openai-codex-model-route-map-v2" as const;
|
|
9
|
+
export type ModelRouteMappingVersion = typeof LEGACY_OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION | typeof OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION;
|
|
8
10
|
|
|
9
11
|
export const PARENT_ROUTING_MODES = Object.freeze(["off", "shadow", "recommend", "auto"] as const);
|
|
10
12
|
export const ROUTING_OBJECTIVES = Object.freeze(["intelligence", "balance", "cost"] as const);
|
|
@@ -26,7 +28,7 @@ export type ModelRouteDisposition = "preserved" | "shadowed" | "recommended" | "
|
|
|
26
28
|
export type ModelRouteDecision = {
|
|
27
29
|
schemaVersion: typeof MODEL_ROUTE_DECISION_SCHEMA_VERSION;
|
|
28
30
|
policyVersion: typeof MODEL_ROUTE_POLICY_VERSION;
|
|
29
|
-
mappingVersion:
|
|
31
|
+
mappingVersion: ModelRouteMappingVersion;
|
|
30
32
|
featureHash: string;
|
|
31
33
|
mode: ParentRoutingMode;
|
|
32
34
|
objective: RoutingObjective;
|
|
@@ -88,13 +90,14 @@ export function authenticatedCatalogDigest(catalog: AuthenticatedModelCatalog):
|
|
|
88
90
|
}
|
|
89
91
|
|
|
90
92
|
export function createModelRouteDecision(
|
|
91
|
-
input: Omit<ModelRouteDecision, "schemaVersion" | "policyVersion" | "mappingVersion" | "decisionDigest">
|
|
93
|
+
input: Omit<ModelRouteDecision, "schemaVersion" | "policyVersion" | "mappingVersion" | "decisionDigest"> & { mappingVersion?: ModelRouteMappingVersion }
|
|
92
94
|
): ModelRouteDecision {
|
|
95
|
+
const { mappingVersion = OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION, ...decisionInput } = input;
|
|
93
96
|
const payload = {
|
|
94
97
|
schemaVersion: MODEL_ROUTE_DECISION_SCHEMA_VERSION,
|
|
95
98
|
policyVersion: MODEL_ROUTE_POLICY_VERSION,
|
|
96
|
-
mappingVersion
|
|
97
|
-
...structuredClone(
|
|
99
|
+
mappingVersion,
|
|
100
|
+
...structuredClone(decisionInput)
|
|
98
101
|
} satisfies Omit<ModelRouteDecision, "decisionDigest">;
|
|
99
102
|
return validateModelRouteDecision({ ...payload, decisionDigest: digestPayload(payload) });
|
|
100
103
|
}
|
|
@@ -108,7 +111,8 @@ export function modelRouteDecisionValidationErrors(input: unknown): string[] {
|
|
|
108
111
|
];
|
|
109
112
|
if (value.schemaVersion !== MODEL_ROUTE_DECISION_SCHEMA_VERSION
|
|
110
113
|
|| value.policyVersion !== MODEL_ROUTE_POLICY_VERSION
|
|
111
|
-
|| value.mappingVersion !==
|
|
114
|
+
|| (value.mappingVersion !== LEGACY_OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION
|
|
115
|
+
&& value.mappingVersion !== OPENAI_CODEX_MODEL_ROUTE_MAPPING_VERSION)) errors.push("model route schema/policy/mapping version is invalid");
|
|
112
116
|
if (typeof value.featureHash !== "string" || !HASH.test(value.featureHash)) errors.push("featureHash must be sha256 hex");
|
|
113
117
|
if (!PARENT_ROUTING_MODES.includes(value.mode as ParentRoutingMode)) errors.push("mode is invalid");
|
|
114
118
|
if (!ROUTING_OBJECTIVES.includes(value.objective as RoutingObjective)) errors.push("objective is invalid");
|
|
@@ -236,3 +236,23 @@ export function buildOpenAiCodexWireFingerprint(input: {
|
|
|
236
236
|
return unavailable(modelId, error instanceof SurfaceBoundError ? error.reasonCode : "surface-not-json");
|
|
237
237
|
}
|
|
238
238
|
}
|
|
239
|
+
|
|
240
|
+
// Measurement remains outside the signed fingerprint contract. A new telemetry
|
|
241
|
+
// field must not silently change an existing signed-receipt schema.
|
|
242
|
+
export function measureProviderPayload(payload: unknown) {
|
|
243
|
+
if (!record(payload)) return null;
|
|
244
|
+
try {
|
|
245
|
+
// Validate the entire payload within the existing bounds before measuring.
|
|
246
|
+
// This is the JSON at this hook, not HTTP bytes or proof of a cache hit.
|
|
247
|
+
canonicalJson(payload);
|
|
248
|
+
const serialized = JSON.stringify(payload);
|
|
249
|
+
const measured = JSON.parse(serialized);
|
|
250
|
+
const bytes = (value: unknown) => Buffer.byteLength(JSON.stringify(value) ?? "", "utf8");
|
|
251
|
+
const instructionsBytes = bytes(measured.instructions);
|
|
252
|
+
const toolsBytes = bytes(measured.tools);
|
|
253
|
+
const inputBytes = bytes(measured.input);
|
|
254
|
+
const totalBytes = Buffer.byteLength(serialized, "utf8");
|
|
255
|
+
return { basis: "hook-json-utf8" as const, instructionsBytes, toolsBytes, inputBytes,
|
|
256
|
+
remainingBytes: totalBytes - instructionsBytes - toolsBytes - inputBytes, totalBytes, cacheHit: null };
|
|
257
|
+
} catch { return null; }
|
|
258
|
+
}
|