@piagent/platform 1.6.1 → 1.8.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +71 -0
- package/README.md +8 -2
- package/README.vi.md +11 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +3 -3
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +2 -2
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/model-options.md +20 -15
- package/docs/operator-manual-vietnamese.md +12 -12
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +3 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +38 -11
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +23 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +23 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +2 -2
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +3 -3
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -1
- package/docs-site/workflows.html +1 -1
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +107 -107
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +63 -50
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +69 -8
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +162 -45
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +144 -128
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +395 -33
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +3 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +264 -70
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +139 -152
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +1 -0
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/model-route-policy.ts +33 -13
- package/packages/piagent-core/runtime/model/model-route-types.ts +10 -6
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +12 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +11 -11
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DMU7Xe_O.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-Cui9AAV_.js +176 -0
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +141 -19
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +14 -6
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +65 -12
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +32 -1
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +30 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-model-scope.sh +6 -2
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/install-global.sh +2 -2
- package/scripts/model-route-evaluation-core.ts +1 -0
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-model-route.mjs +1 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/setup.sh +2 -2
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +37 -56
- package/scripts/verify-release-identity.mjs +5 -3
- package/templates/global/settings.json +5 -1
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -0,0 +1,676 @@
|
|
|
1
|
+
import { execFileSync } from "node:child_process";
|
|
2
|
+
import crypto from "node:crypto";
|
|
3
|
+
import fs from "node:fs";
|
|
4
|
+
import path from "node:path";
|
|
5
|
+
import { fileURLToPath, pathToFileURL } from "node:url";
|
|
6
|
+
import { types } from "node:util";
|
|
7
|
+
|
|
8
|
+
import {
|
|
9
|
+
validBenchmarkCandidateProvenance,
|
|
10
|
+
verifyMaterializedBenchmarkCandidate
|
|
11
|
+
} from "../packages/piagent-core/benchmark/benchmark-candidate.js";
|
|
12
|
+
|
|
13
|
+
export const SCOPED_BROKER_FROZEN_IDENTITY_VERSION = 4;
|
|
14
|
+
export const SCOPED_BROKER_MEASUREMENT_BINDING_VERSION = "benchmark-turn-binding-v1";
|
|
15
|
+
const HASH = /^[a-f0-9]{64}$/;
|
|
16
|
+
const ID = /^[A-Za-z0-9][A-Za-z0-9:._~-]{0,159}$/;
|
|
17
|
+
const OPERATOR_REQUEST = /^operator-request-v1:[a-f0-9]{64}$/;
|
|
18
|
+
const hash = value => crypto.createHash("sha256").update(value).digest("hex");
|
|
19
|
+
|
|
20
|
+
function fail(code) {
|
|
21
|
+
throw Object.assign(new Error(code), { brokerCode: code });
|
|
22
|
+
}
|
|
23
|
+
function requireThat(value, code) { if (!value) fail(code); }
|
|
24
|
+
|
|
25
|
+
function realDirectory(value, code) {
|
|
26
|
+
requireThat(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
27
|
+
&& !value.includes("\0"), code);
|
|
28
|
+
let real, info;
|
|
29
|
+
try { real = fs.realpathSync(value); info = fs.statSync(real); }
|
|
30
|
+
catch { fail(code); }
|
|
31
|
+
requireThat(real === value && info.isDirectory(), code);
|
|
32
|
+
return real;
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
function stablePrivateIndex(file, expectedSha256) {
|
|
36
|
+
requireThat(typeof file === "string" && path.isAbsolute(file) && path.normalize(file) === file
|
|
37
|
+
&& !file.includes("\0") && HASH.test(expectedSha256), "frozen-index-binding");
|
|
38
|
+
const parent = realDirectory(path.dirname(file), "frozen-index-parent");
|
|
39
|
+
requireThat(parent === path.dirname(file), "frozen-index-parent");
|
|
40
|
+
let before;
|
|
41
|
+
try { before = fs.lstatSync(file, { bigint: true }); } catch { fail("frozen-index-file"); }
|
|
42
|
+
requireThat(before.isFile() && !before.isSymbolicLink() && before.nlink === 1n
|
|
43
|
+
&& before.size > 0n && before.size <= 16n * 1024n * 1024n && (before.mode & 0o077n) === 0n,
|
|
44
|
+
"frozen-index-file");
|
|
45
|
+
const fd = fs.openSync(file, fs.constants.O_RDONLY | fs.constants.O_NOFOLLOW);
|
|
46
|
+
try {
|
|
47
|
+
const descriptor = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd);
|
|
48
|
+
const after = fs.fstatSync(fd, { bigint: true }), current = fs.lstatSync(file, { bigint: true });
|
|
49
|
+
const fields = ["dev", "ino", "mode", "nlink", "size", "mtimeNs", "ctimeNs"];
|
|
50
|
+
requireThat(bytes.length === Number(before.size) && fields.every(key => before[key] === descriptor[key]
|
|
51
|
+
&& before[key] === after[key] && before[key] === current[key]) && hash(bytes) === expectedSha256,
|
|
52
|
+
"frozen-index-drift");
|
|
53
|
+
let index;
|
|
54
|
+
try { index = JSON.parse(new TextDecoder("utf-8", { fatal: true }).decode(bytes)); }
|
|
55
|
+
catch { fail("frozen-index-json"); }
|
|
56
|
+
return index;
|
|
57
|
+
} finally { fs.closeSync(fd); }
|
|
58
|
+
}
|
|
59
|
+
|
|
60
|
+
/** Same full-tree algorithm as scoped qualification v3. Links are bound as
|
|
61
|
+
* link text and never followed. */
|
|
62
|
+
export function scopedFrozenTreeIdentity(value) {
|
|
63
|
+
const root = realDirectory(value, "frozen-tree-root"), entries = [];
|
|
64
|
+
let files = 0, bytes = 0;
|
|
65
|
+
function walk(directory, prefix) {
|
|
66
|
+
for (const entry of fs.readdirSync(directory, { withFileTypes: true })
|
|
67
|
+
.sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0)) {
|
|
68
|
+
const relative = prefix ? `${prefix}/${entry.name}` : entry.name, file = path.join(directory, entry.name);
|
|
69
|
+
requireThat(entries.length < 100000, "frozen-tree-entry-limit");
|
|
70
|
+
if (entry.isDirectory()) { entries.push(["d", relative]); walk(file, relative); continue; }
|
|
71
|
+
if (entry.isSymbolicLink()) { entries.push(["l", relative, fs.readlinkSync(file)]); continue; }
|
|
72
|
+
requireThat(entry.isFile(), "frozen-tree-node");
|
|
73
|
+
const before = fs.lstatSync(file, { bigint: true }), content = fs.readFileSync(file);
|
|
74
|
+
const after = fs.lstatSync(file, { bigint: true });
|
|
75
|
+
requireThat(before.dev === after.dev && before.ino === after.ino && before.size === after.size
|
|
76
|
+
&& before.mtimeNs === after.mtimeNs && before.ctimeNs === after.ctimeNs, "frozen-tree-drift");
|
|
77
|
+
bytes += content.length; files++;
|
|
78
|
+
requireThat(bytes <= 1024 * 1024 * 1024, "frozen-tree-byte-limit");
|
|
79
|
+
entries.push(["f", relative, content.length, hash(content)]);
|
|
80
|
+
}
|
|
81
|
+
}
|
|
82
|
+
walk(root, "");
|
|
83
|
+
return Object.freeze({ version: 2, files, bytes,
|
|
84
|
+
sha256: hash(JSON.stringify({ version: 2, entries })) });
|
|
85
|
+
}
|
|
86
|
+
|
|
87
|
+
/** Recomputes the immutable bootstrap snapshot, separately frozen WebUI assets,
|
|
88
|
+
* SDK tree and caller-supplied recomputed common closure. */
|
|
89
|
+
export function scopedFrozenQualificationIdentity(value, brokerClosureSha256) {
|
|
90
|
+
const names = ["version", "candidateRoot", "assetsRoot", "sdkRoot", "candidateIndexPath",
|
|
91
|
+
"candidateIndexSha256"];
|
|
92
|
+
requireThat(value && typeof value === "object" && !Array.isArray(value)
|
|
93
|
+
&& Object.keys(value).length === names.length && names.every(name => Object.hasOwn(value, name))
|
|
94
|
+
&& value.version === SCOPED_BROKER_FROZEN_IDENTITY_VERSION && HASH.test(brokerClosureSha256),
|
|
95
|
+
"frozen-qualification-fields");
|
|
96
|
+
const candidateRoot = realDirectory(value.candidateRoot, "frozen-candidate-root"),
|
|
97
|
+
assetsRoot = realDirectory(value.assetsRoot, "frozen-assets-root"),
|
|
98
|
+
sdkRoot = realDirectory(value.sdkRoot, "frozen-sdk-root");
|
|
99
|
+
requireThat(!assetsRoot.startsWith(candidateRoot + path.sep) && assetsRoot !== candidateRoot,
|
|
100
|
+
"frozen-assets-separation");
|
|
101
|
+
const index = stablePrivateIndex(value.candidateIndexPath, value.candidateIndexSha256);
|
|
102
|
+
let source;
|
|
103
|
+
try { source = verifyMaterializedBenchmarkCandidate(candidateRoot, index); }
|
|
104
|
+
catch { fail("frozen-candidate-mismatch"); }
|
|
105
|
+
requireThat(validBenchmarkCandidateProvenance(source), "frozen-candidate-identity");
|
|
106
|
+
const assets = scopedFrozenTreeIdentity(assetsRoot), sdk = scopedFrozenTreeIdentity(sdkRoot);
|
|
107
|
+
return Object.freeze({ version: SCOPED_BROKER_FROZEN_IDENTITY_VERSION, candidateRoot, assetsRoot,
|
|
108
|
+
sdkRoot, candidateIndexPath: value.candidateIndexPath, candidateIndexSha256: value.candidateIndexSha256,
|
|
109
|
+
sourceSha256: source.contentDigest, assetTreeSha256: assets.sha256, sdkTreeSha256: sdk.sha256,
|
|
110
|
+
brokerClosureSha256 });
|
|
111
|
+
}
|
|
112
|
+
|
|
113
|
+
export function assertScopedFrozenQualification(value, identity, brokerClosureSha256) {
|
|
114
|
+
const names = ["version", "candidateRoot", "assetsRoot", "sdkRoot", "candidateIndexPath",
|
|
115
|
+
"candidateIndexSha256", "sourceSha256", "assetTreeSha256", "sdkTreeSha256", "brokerClosureSha256"];
|
|
116
|
+
requireThat(value && typeof value === "object" && !Array.isArray(value)
|
|
117
|
+
&& Object.keys(value).length === names.length && names.every(name => Object.hasOwn(value, name)),
|
|
118
|
+
"manifest-identity-v4");
|
|
119
|
+
const actual = scopedFrozenQualificationIdentity(Object.fromEntries(names.slice(0, 6)
|
|
120
|
+
.map(name => [name, value[name]])), brokerClosureSha256);
|
|
121
|
+
requireThat(Object.keys(actual).every(key => actual[key] === value[key])
|
|
122
|
+
&& ["sourceSha256", "assetTreeSha256", "sdkTreeSha256", "brokerClosureSha256"]
|
|
123
|
+
.every(key => identity?.[key] === actual[key]), "manifest-identity-v4");
|
|
124
|
+
return true;
|
|
125
|
+
}
|
|
126
|
+
|
|
127
|
+
const MEASUREMENT_BINDING_FIELDS = Object.freeze(["version", "authority", "runId", "armId", "suiteId",
|
|
128
|
+
"scenarioId", "surface", "repeat", "infrastructureAttempt", "turnIndex", "turnId", "catalogSha256",
|
|
129
|
+
"configurationSha256", "publicContractSha256", "planSha256", "turnBindingSha256", "workspaceSha256",
|
|
130
|
+
"operatorRequestDigest", "profile", "materialManifestSha256", "verificationManifestSha256",
|
|
131
|
+
"contextPolicySha256", "nodeSha256", "runtimeSha256", "modelSha256"]);
|
|
132
|
+
const MEASUREMENT_HASH_FIELDS = Object.freeze(MEASUREMENT_BINDING_FIELDS.filter(name => name.endsWith("Sha256")));
|
|
133
|
+
|
|
134
|
+
/** Data-only predispatch coordinate. It binds bytes through the config digest
|
|
135
|
+
* but cannot grant approval, criterion status, provider dispatch or admission. */
|
|
136
|
+
export function validateScopedBrokerMeasurementBinding(value) {
|
|
137
|
+
requireThat(value && typeof value === "object" && !Array.isArray(value)
|
|
138
|
+
&& Object.keys(value).length === MEASUREMENT_BINDING_FIELDS.length
|
|
139
|
+
&& MEASUREMENT_BINDING_FIELDS.every(name => Object.hasOwn(value, name))
|
|
140
|
+
&& value.version === SCOPED_BROKER_MEASUREMENT_BINDING_VERSION && value.authority === "none"
|
|
141
|
+
&& [value.runId, value.armId, value.suiteId, value.scenarioId, value.turnId].every(item =>
|
|
142
|
+
typeof item === "string" && ID.test(item))
|
|
143
|
+
&& ["piagent", "codex-cli"].includes(value.surface)
|
|
144
|
+
&& Number.isSafeInteger(value.repeat) && value.repeat >= 1 && value.repeat <= 10
|
|
145
|
+
&& Number.isSafeInteger(value.infrastructureAttempt) && value.infrastructureAttempt >= 1
|
|
146
|
+
&& value.infrastructureAttempt <= 3 && Number.isSafeInteger(value.turnIndex)
|
|
147
|
+
&& value.turnIndex >= 1 && value.turnIndex <= 3
|
|
148
|
+
&& OPERATOR_REQUEST.test(value.operatorRequestDigest)
|
|
149
|
+
&& ["incident", "document", "protected-env-refusal", "destructive-history-refusal"].includes(value.profile)
|
|
150
|
+
&& MEASUREMENT_HASH_FIELDS.every(name => HASH.test(value[name])), "measurement-binding-invalid");
|
|
151
|
+
return Object.freeze(Object.fromEntries(MEASUREMENT_BINDING_FIELDS.map(name => [name, value[name]])));
|
|
152
|
+
}
|
|
153
|
+
|
|
154
|
+
export function scopedBrokerMeasurementBindingSha256(value) {
|
|
155
|
+
return hash(JSON.stringify(validateScopedBrokerMeasurementBinding(value)));
|
|
156
|
+
}
|
|
157
|
+
|
|
158
|
+
export const SCOPED_BROKER_IDENTITY_VERSION = 2;
|
|
159
|
+
export const SCOPED_BROKER_QUALIFICATION_IDENTITY_VERSION = 3;
|
|
160
|
+
export const SCOPED_CONTEXT_POLICY_VERSION = 2;
|
|
161
|
+
const validHash = value => typeof value === "string" && HASH.test(value);
|
|
162
|
+
const plain = value => Boolean(value && typeof value === "object" && !types.isProxy(value)
|
|
163
|
+
&& [null, Object.prototype].includes(Object.getPrototypeOf(value)));
|
|
164
|
+
function deepFreeze(value) {
|
|
165
|
+
if (!value || typeof value !== "object" || Object.isFrozen(value)) return value;
|
|
166
|
+
for (const child of Object.values(value)) deepFreeze(child);
|
|
167
|
+
return Object.freeze(value);
|
|
168
|
+
}
|
|
169
|
+
|
|
170
|
+
const BROKER_CONFIG_BYTES = 64 * 1024;
|
|
171
|
+
const BROKER_JOURNAL_BYTES = 32 * 1024 * 1024;
|
|
172
|
+
const BROKER_TRANSPORT_BYTES = 1024 * 1024;
|
|
173
|
+
const brokerFail = code => { throw Object.assign(new Error(code), { brokerCode: code }); };
|
|
174
|
+
const brokerRequire = (value, code) => { if (!value) brokerFail(code); };
|
|
175
|
+
const brokerWellFormed = value => typeof value === "string" && value.isWellFormed();
|
|
176
|
+
function brokerFields(value, names) {
|
|
177
|
+
brokerRequire(plain(value), "invalid-object");
|
|
178
|
+
const descriptors = Object.getOwnPropertyDescriptors(value);
|
|
179
|
+
brokerRequire(Reflect.ownKeys(descriptors).length === names.length && names.every(name =>
|
|
180
|
+
descriptors[name]?.enumerable && Object.hasOwn(descriptors[name], "value")), "invalid-fields");
|
|
181
|
+
}
|
|
182
|
+
const brokerSchema = properties => ({ type: "object", properties, required: Object.keys(properties),
|
|
183
|
+
additionalProperties: false });
|
|
184
|
+
|
|
185
|
+
export const SCOPED_TOOL_DEFINITIONS = deepFreeze([
|
|
186
|
+
{ name: "scoped_read", inputSchema: brokerSchema({ materialId: { type: "string" } }) },
|
|
187
|
+
{ name: "scoped_write_document", inputSchema: brokerSchema({ materialId: { type: "string" },
|
|
188
|
+
expectedSha256: { type: "string" }, utf8: { type: "string" } }) },
|
|
189
|
+
{ name: "scoped_verify", inputSchema: brokerSchema({ verificationId: { type: "string" } }) }
|
|
190
|
+
]);
|
|
191
|
+
|
|
192
|
+
export const SCOPED_BROKER_IDENTITY_V2_FIELDS = Object.freeze(["version", "armId", "taskId", "sessionId",
|
|
193
|
+
"requestId", "operationId", "nonce", "sourceSha256", "configSha256", "brokerClosureSha256",
|
|
194
|
+
"toolDefinitionsSha256", "manifestAuthoritySha256", "journalSignerSha256", "journalPathSha256",
|
|
195
|
+
"contextPolicySha256", "sdkTreeSha256"]);
|
|
196
|
+
export const SCOPED_BROKER_IDENTITY_V3_FIELDS = Object.freeze(["version", "armId", "taskId", "sessionId",
|
|
197
|
+
"requestId", "operationId", "nonce", "sourceSha256", "assetTreeSha256", "configSha256",
|
|
198
|
+
"brokerClosureSha256", "toolDefinitionsSha256", "manifestAuthoritySha256", "journalSignerSha256",
|
|
199
|
+
"journalPathSha256", "contextPolicySha256", "sdkTreeSha256"]);
|
|
200
|
+
export const SCOPED_BROKER_IDENTITY_V4_FIELDS = Object.freeze([
|
|
201
|
+
...SCOPED_BROKER_IDENTITY_V3_FIELDS.slice(0, 10),
|
|
202
|
+
"measurementConfigurationSha256",
|
|
203
|
+
...SCOPED_BROKER_IDENTITY_V3_FIELDS.slice(10)
|
|
204
|
+
]);
|
|
205
|
+
|
|
206
|
+
export function scopedToolDefinitionsSha256() {
|
|
207
|
+
return hash(JSON.stringify(SCOPED_TOOL_DEFINITIONS));
|
|
208
|
+
}
|
|
209
|
+
|
|
210
|
+
export function scopedBrokerSourceClosureSha256() {
|
|
211
|
+
const directory = path.dirname(fileURLToPath(import.meta.url));
|
|
212
|
+
const files = [
|
|
213
|
+
["broker", "benchmark-scoped-tool-broker.mjs"],
|
|
214
|
+
["supervisor", "benchmark-scoped-verification-supervisor.mjs"],
|
|
215
|
+
["project-verifier", "benchmark-scoped-project-verifier.mjs"],
|
|
216
|
+
["qualification", "benchmark-scoped-frozen-qualification.mjs"],
|
|
217
|
+
["transport", "benchmark-scoped-broker-custody.mjs"]
|
|
218
|
+
].map(([role, name]) => ({ role, sha256: hash(fs.readFileSync(path.join(directory, name))) }));
|
|
219
|
+
return hash(JSON.stringify({ version: 4, files,
|
|
220
|
+
toolDefinitionsSha256: scopedToolDefinitionsSha256() }));
|
|
221
|
+
}
|
|
222
|
+
|
|
223
|
+
export function scopedJournalPathSha256(value) {
|
|
224
|
+
brokerRequire(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
225
|
+
&& !value.includes("\0"), "journal-path-binding");
|
|
226
|
+
const parent = fs.realpathSync(path.dirname(value));
|
|
227
|
+
brokerRequire(parent === path.dirname(value), "journal-path-binding");
|
|
228
|
+
return hash(JSON.stringify({ version: 2, path: value }));
|
|
229
|
+
}
|
|
230
|
+
|
|
231
|
+
/** Deterministic full-tree pin. Symlinks are bound as links and never followed. */
|
|
232
|
+
export function scopedTreeIdentity(value) {
|
|
233
|
+
brokerRequire(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
234
|
+
&& fs.realpathSync(value) === value && fs.statSync(value).isDirectory(), "tree-root-invalid");
|
|
235
|
+
const entries = []; let files = 0, bytes = 0;
|
|
236
|
+
function walk(directory, prefix) {
|
|
237
|
+
for (const entry of fs.readdirSync(directory, { withFileTypes: true })
|
|
238
|
+
.sort((a, b) => a.name < b.name ? -1 : a.name > b.name ? 1 : 0)) {
|
|
239
|
+
const relative = prefix ? `${prefix}/${entry.name}` : entry.name, file = path.join(directory, entry.name);
|
|
240
|
+
brokerRequire(entries.length < 100000, "tree-entry-limit");
|
|
241
|
+
if (entry.isDirectory()) { entries.push(["d", relative]); walk(file, relative); continue; }
|
|
242
|
+
if (entry.isSymbolicLink()) { entries.push(["l", relative, fs.readlinkSync(file)]); continue; }
|
|
243
|
+
brokerRequire(entry.isFile(), "tree-node-invalid");
|
|
244
|
+
const before = fs.lstatSync(file, { bigint: true }), content = fs.readFileSync(file);
|
|
245
|
+
const after = fs.lstatSync(file, { bigint: true });
|
|
246
|
+
brokerRequire(before.dev === after.dev && before.ino === after.ino && before.size === after.size
|
|
247
|
+
&& before.mtimeNs === after.mtimeNs && before.ctimeNs === after.ctimeNs, "tree-mutated");
|
|
248
|
+
bytes += content.length; files++; brokerRequire(bytes <= 1024 * 1024 * 1024, "tree-byte-limit");
|
|
249
|
+
entries.push(["f", relative, content.length, hash(content)]);
|
|
250
|
+
}
|
|
251
|
+
}
|
|
252
|
+
walk(value, "");
|
|
253
|
+
return deepFreeze({ version: 2, files, bytes, sha256: hash(JSON.stringify({ version: 2, entries })) });
|
|
254
|
+
}
|
|
255
|
+
|
|
256
|
+
const qualificationRoot = (value, code) => {
|
|
257
|
+
brokerRequire(typeof value === "string" && path.isAbsolute(value) && path.normalize(value) === value
|
|
258
|
+
&& fs.realpathSync(value) === value && fs.statSync(value).isDirectory(), code);
|
|
259
|
+
return value;
|
|
260
|
+
};
|
|
261
|
+
const git = (root, args) => {
|
|
262
|
+
try { return execFileSync("git", ["-C", root, ...args], { maxBuffer: 16 * 1024 * 1024,
|
|
263
|
+
env: { ...process.env, GIT_OPTIONAL_LOCKS: "0", LC_ALL: "C" } }); }
|
|
264
|
+
catch { brokerFail("tracked-source-unavailable"); }
|
|
265
|
+
};
|
|
266
|
+
|
|
267
|
+
/** Recomputes the clean tracked working tree, binding Git mode, index object,
|
|
268
|
+
* path and the bytes actually present on disk. Untracked build assets remain a
|
|
269
|
+
* separate qualification input. */
|
|
270
|
+
export function scopedTrackedSourceIdentity(value) {
|
|
271
|
+
const root = qualificationRoot(value, "tracked-source-root-invalid");
|
|
272
|
+
const decoder = new TextDecoder("utf-8", { fatal: true });
|
|
273
|
+
let top, head;
|
|
274
|
+
try {
|
|
275
|
+
top = decoder.decode(git(root, ["rev-parse", "--show-toplevel"])).trimEnd();
|
|
276
|
+
head = decoder.decode(git(root, ["rev-parse", "--verify", "HEAD"])).trimEnd();
|
|
277
|
+
} catch { brokerFail("tracked-source-unavailable"); }
|
|
278
|
+
brokerRequire(fs.realpathSync(top) === root && /^[a-f0-9]{40,64}$/.test(head), "tracked-source-root-invalid");
|
|
279
|
+
const clean = () => git(root, ["status", "--porcelain=v1", "-z", "--untracked-files=no",
|
|
280
|
+
"--ignore-submodules=none"]);
|
|
281
|
+
brokerRequire(clean().length === 0, "tracked-source-dirty");
|
|
282
|
+
const index = git(root, ["ls-files", "--stage", "-z"]), entries = [];
|
|
283
|
+
let start = 0, bytes = 0;
|
|
284
|
+
for (let cursor = 0; cursor <= index.length; cursor++) {
|
|
285
|
+
if (cursor < index.length && index[cursor] !== 0) continue;
|
|
286
|
+
if (cursor === start) { start = cursor + 1; continue; }
|
|
287
|
+
let row;
|
|
288
|
+
try { row = decoder.decode(index.subarray(start, cursor)); } catch { brokerFail("tracked-source-index-invalid"); }
|
|
289
|
+
start = cursor + 1;
|
|
290
|
+
const tab = row.indexOf("\t"), prefix = row.slice(0, tab), relative = row.slice(tab + 1);
|
|
291
|
+
const match = /^(100644|100755|120000) ([a-f0-9]{40,64}) 0$/.exec(prefix);
|
|
292
|
+
brokerRequire(tab > 0 && match && relative && !path.isAbsolute(relative) && !relative.includes("\\")
|
|
293
|
+
&& relative.split("/").every(part => part && part !== "." && part !== "..")
|
|
294
|
+
&& entries.length < 100000, "tracked-source-index-invalid");
|
|
295
|
+
const file = path.join(root, ...relative.split("/")), before = fs.lstatSync(file, { bigint: true });
|
|
296
|
+
let content;
|
|
297
|
+
if (match[1] === "120000") {
|
|
298
|
+
brokerRequire(before.isSymbolicLink(), "tracked-source-node-invalid");
|
|
299
|
+
content = Buffer.from(fs.readlinkSync(file));
|
|
300
|
+
} else {
|
|
301
|
+
brokerRequire(before.isFile() && !before.isSymbolicLink(), "tracked-source-node-invalid");
|
|
302
|
+
content = fs.readFileSync(file);
|
|
303
|
+
const after = fs.lstatSync(file, { bigint: true });
|
|
304
|
+
brokerRequire(before.dev === after.dev && before.ino === after.ino && before.size === after.size
|
|
305
|
+
&& before.mtimeNs === after.mtimeNs && before.ctimeNs === after.ctimeNs, "tracked-source-mutated");
|
|
306
|
+
}
|
|
307
|
+
bytes += content.length; brokerRequire(bytes <= 1024 * 1024 * 1024, "tracked-source-byte-limit");
|
|
308
|
+
entries.push([match[1], match[2], relative, content.length, hash(content)]);
|
|
309
|
+
}
|
|
310
|
+
brokerRequire(clean().length === 0 && decoder.decode(git(root, ["rev-parse", "--verify", "HEAD"])).trimEnd() === head,
|
|
311
|
+
"tracked-source-mutated");
|
|
312
|
+
return deepFreeze({ version: 3, head, files: entries.length, bytes,
|
|
313
|
+
sha256: hash(JSON.stringify({ version: 3, head, entries })) });
|
|
314
|
+
}
|
|
315
|
+
|
|
316
|
+
const COMMON_RUNTIME_SEEDS = Object.freeze(["scripts/benchmark-scoped-tool-broker.mjs",
|
|
317
|
+
"scripts/benchmark-scoped-verification-supervisor.mjs", "scripts/benchmark-arm-observer.mjs",
|
|
318
|
+
"scripts/benchmark-webui-journey.mjs"]);
|
|
319
|
+
const LOCAL_IMPORT_PATTERNS = [
|
|
320
|
+
/\bimport\s+["']([^"']+)["']/g,
|
|
321
|
+
/\bimport\s(?!\()[^"'`;]*?\bfrom\s*["']([^"']+)["']/g,
|
|
322
|
+
/\bexport\s+(?:\*|\{[^}]*\})\s+from\s+["']([^"']+)["']/g,
|
|
323
|
+
/\bimport\s*\(\s*["']([^"']+)["']\s*\)/g
|
|
324
|
+
];
|
|
325
|
+
function localImports(source) {
|
|
326
|
+
const result = new Set();
|
|
327
|
+
for (const pattern of LOCAL_IMPORT_PATTERNS) {
|
|
328
|
+
pattern.lastIndex = 0;
|
|
329
|
+
for (const match of source.matchAll(pattern)) if (match[1].startsWith(".")) result.add(match[1]);
|
|
330
|
+
}
|
|
331
|
+
const constants = new Map();
|
|
332
|
+
for (const match of source.matchAll(/\bconst\s+([A-Za-z_$][A-Za-z0-9_$]*)\s*=\s*(["'])([^"'\\\r\n]*)\2\s*;/g))
|
|
333
|
+
constants.set(match[1], match[3]);
|
|
334
|
+
for (const match of source.matchAll(/\bimport\s*\(\s*`([^`\\]*(?:\\.[^`\\]*)*)`\s*\)/g)) {
|
|
335
|
+
let complete = true;
|
|
336
|
+
const specifier = match[1].replace(/\$\{([A-Za-z_$][A-Za-z0-9_$]*)\}/g, (_token, name) => {
|
|
337
|
+
if (!constants.has(name)) { complete = false; return ""; }
|
|
338
|
+
return constants.get(name);
|
|
339
|
+
});
|
|
340
|
+
if (complete && !specifier.includes("${") && specifier.startsWith(".")) result.add(specifier);
|
|
341
|
+
}
|
|
342
|
+
return [...result];
|
|
343
|
+
}
|
|
344
|
+
function resolveLocalImport(root, importer, specifier) {
|
|
345
|
+
const requested = path.resolve(path.dirname(importer), specifier), extensions = ["", ".js", ".mjs", ".cjs",
|
|
346
|
+
".ts", ".tsx", ".json"];
|
|
347
|
+
const stem = /\.(?:[cm]?js)$/.test(requested) ? requested.replace(/\.(?:[cm]?js)$/, "") : requested;
|
|
348
|
+
const candidates = extensions.map(extension => requested + extension)
|
|
349
|
+
.concat(stem === requested ? [] : [stem + ".ts", stem + ".tsx", stem + ".mjs"])
|
|
350
|
+
.concat(extensions.slice(1).map(extension => path.join(requested, `index${extension}`)));
|
|
351
|
+
const file = candidates.find(candidate => { try { return fs.lstatSync(candidate).isFile(); } catch { return false; } });
|
|
352
|
+
brokerRequire(file, "runtime-closure-import-unresolved");
|
|
353
|
+
const info = fs.lstatSync(file);
|
|
354
|
+
brokerRequire(info.isFile() && !info.isSymbolicLink(), "runtime-closure-node-invalid");
|
|
355
|
+
const real = fs.realpathSync(file);
|
|
356
|
+
brokerRequire(real.startsWith(root + path.sep), "runtime-closure-import-outside-root");
|
|
357
|
+
return real;
|
|
358
|
+
}
|
|
359
|
+
|
|
360
|
+
/** Signed source closure for the common broker/supervisor/observer/journey
|
|
361
|
+
* bridge and every statically discoverable local import reachable from it. */
|
|
362
|
+
export function scopedCommonRuntimeClosureIdentity(value = path.resolve(path.dirname(fileURLToPath(import.meta.url)), "..")) {
|
|
363
|
+
const root = qualificationRoot(fs.realpathSync(value), "runtime-closure-root-invalid"), pending = [];
|
|
364
|
+
for (const relative of COMMON_RUNTIME_SEEDS) {
|
|
365
|
+
const file = fs.realpathSync(path.join(root, relative));
|
|
366
|
+
brokerRequire(file.startsWith(root + path.sep), "runtime-closure-seed-invalid"); pending.push(file);
|
|
367
|
+
}
|
|
368
|
+
const visited = new Set(), rows = []; let bytes = 0;
|
|
369
|
+
while (pending.length) {
|
|
370
|
+
const file = pending.pop(); if (visited.has(file)) continue; visited.add(file);
|
|
371
|
+
brokerRequire(visited.size <= 4096, "runtime-closure-file-limit");
|
|
372
|
+
const before = fs.lstatSync(file, { bigint: true }), content = fs.readFileSync(file), after = fs.lstatSync(file, { bigint: true });
|
|
373
|
+
brokerRequire(before.isFile() && !before.isSymbolicLink() && before.dev === after.dev && before.ino === after.ino
|
|
374
|
+
&& before.size === after.size && before.mtimeNs === after.mtimeNs && before.ctimeNs === after.ctimeNs,
|
|
375
|
+
"runtime-closure-mutated");
|
|
376
|
+
bytes += content.length; brokerRequire(bytes <= 256 * 1024 * 1024, "runtime-closure-byte-limit");
|
|
377
|
+
const relative = path.relative(root, file).split(path.sep).join("/");
|
|
378
|
+
rows.push({ path: relative, bytes: content.length, sha256: hash(content) });
|
|
379
|
+
if (!/\.json$/i.test(file)) for (const specifier of localImports(content.toString("utf8")))
|
|
380
|
+
pending.push(resolveLocalImport(root, file, specifier));
|
|
381
|
+
}
|
|
382
|
+
rows.sort((a, b) => a.path < b.path ? -1 : a.path > b.path ? 1 : 0);
|
|
383
|
+
return deepFreeze({ version: 3, files: rows.length, bytes,
|
|
384
|
+
sha256: hash(JSON.stringify({ version: 3, seeds: COMMON_RUNTIME_SEEDS, files: rows })), entries: rows });
|
|
385
|
+
}
|
|
386
|
+
|
|
387
|
+
export function scopedQualificationIdentity(value) {
|
|
388
|
+
brokerFields(value, ["candidateRoot", "assetsRoot", "sdkRoot"]);
|
|
389
|
+
const candidateRoot = qualificationRoot(value.candidateRoot, "qualification-candidate-root-invalid");
|
|
390
|
+
const assetsRoot = qualificationRoot(value.assetsRoot, "qualification-assets-root-invalid");
|
|
391
|
+
const sdkRoot = qualificationRoot(value.sdkRoot, "qualification-sdk-root-invalid");
|
|
392
|
+
const expectedAssets = fs.realpathSync(path.join(candidateRoot, "packages/piagent-webui/dist/client"));
|
|
393
|
+
brokerRequire(assetsRoot === expectedAssets, "qualification-assets-binding");
|
|
394
|
+
const source = scopedTrackedSourceIdentity(candidateRoot), assets = scopedTreeIdentity(assetsRoot);
|
|
395
|
+
const sdk = scopedTreeIdentity(sdkRoot), closure = scopedCommonRuntimeClosureIdentity();
|
|
396
|
+
return deepFreeze({ version: 3, candidateRoot, assetsRoot, sdkRoot, sourceSha256: source.sha256,
|
|
397
|
+
assetTreeSha256: assets.sha256, sdkTreeSha256: sdk.sha256, brokerClosureSha256: closure.sha256 });
|
|
398
|
+
}
|
|
399
|
+
|
|
400
|
+
function strictJsonCopy(value, code, state = { nodes: 0 }) {
|
|
401
|
+
brokerRequire(++state.nodes <= 100000, code);
|
|
402
|
+
if (value === undefined) return null;
|
|
403
|
+
if (value === null || typeof value === "string" || typeof value === "boolean") return value;
|
|
404
|
+
if (typeof value === "number" && Number.isFinite(value)) return value;
|
|
405
|
+
brokerRequire(value && typeof value === "object" && !types.isProxy(value), code);
|
|
406
|
+
const array = Array.isArray(value), prototype = Object.getPrototypeOf(value);
|
|
407
|
+
brokerRequire(array || prototype === Object.prototype || prototype === null, code);
|
|
408
|
+
const descriptors = Object.getOwnPropertyDescriptors(value), result = array ? [] : Object.create(null);
|
|
409
|
+
for (const key of Reflect.ownKeys(descriptors)) {
|
|
410
|
+
if (array && key === "length") continue;
|
|
411
|
+
const descriptor = descriptors[key];
|
|
412
|
+
brokerRequire(typeof key === "string" && descriptor.enumerable && Object.hasOwn(descriptor, "value")
|
|
413
|
+
&& (!array || /^(0|[1-9][0-9]*)$/.test(key)), code);
|
|
414
|
+
Object.defineProperty(result, key, { value: strictJsonCopy(descriptor.value, code, state), enumerable: true,
|
|
415
|
+
writable: true, configurable: true });
|
|
416
|
+
}
|
|
417
|
+
brokerRequire(!array || result.length === value.length && Object.keys(result).length === value.length, code);
|
|
418
|
+
return result;
|
|
419
|
+
}
|
|
420
|
+
|
|
421
|
+
function observableJsonCopy(value, code, state = { nodes: 0 }) {
|
|
422
|
+
brokerRequire(++state.nodes <= 100000, code);
|
|
423
|
+
if (value === undefined) return null;
|
|
424
|
+
if (value === null || typeof value === "string" || typeof value === "boolean") return value;
|
|
425
|
+
if (typeof value === "number" && Number.isFinite(value)) return value;
|
|
426
|
+
if (["function", "symbol", "bigint"].includes(typeof value)) return `[non-json:${typeof value}]`;
|
|
427
|
+
brokerRequire(value && typeof value === "object" && !types.isProxy(value), code);
|
|
428
|
+
const array = Array.isArray(value), prototype = Object.getPrototypeOf(value);
|
|
429
|
+
brokerRequire(array || prototype === Object.prototype || prototype === null, code);
|
|
430
|
+
const descriptors = Object.getOwnPropertyDescriptors(value), result = array ? [] : Object.create(null);
|
|
431
|
+
for (const key of Reflect.ownKeys(descriptors)) {
|
|
432
|
+
if (array && key === "length") continue;
|
|
433
|
+
const descriptor = descriptors[key];
|
|
434
|
+
brokerRequire(typeof key === "string" && descriptor.enumerable && Object.hasOwn(descriptor, "value")
|
|
435
|
+
&& (!array || /^(0|[1-9][0-9]*)$/.test(key)), code);
|
|
436
|
+
Object.defineProperty(result, key, { value: observableJsonCopy(descriptor.value, code, state), enumerable: true });
|
|
437
|
+
}
|
|
438
|
+
brokerRequire(!array || result.length === value.length && Object.keys(result).length === value.length, code);
|
|
439
|
+
return result;
|
|
440
|
+
}
|
|
441
|
+
|
|
442
|
+
function scopedContextPolicy(policy) {
|
|
443
|
+
brokerFields(policy, ["version", "systemPrompt", "allowedUserMessages", "removableUserMessages"]);
|
|
444
|
+
brokerRequire(policy.version === SCOPED_CONTEXT_POLICY_VERSION && typeof policy.systemPrompt === "string"
|
|
445
|
+
&& policy.systemPrompt.length > 0 && Buffer.byteLength(policy.systemPrompt) <= 4096
|
|
446
|
+
&& Array.isArray(policy.allowedUserMessages) && policy.allowedUserMessages.length > 0
|
|
447
|
+
&& policy.allowedUserMessages.length <= 32 && Array.isArray(policy.removableUserMessages)
|
|
448
|
+
&& policy.removableUserMessages.length <= 32, "context-policy-invalid");
|
|
449
|
+
const all = [...policy.allowedUserMessages, ...policy.removableUserMessages];
|
|
450
|
+
brokerRequire(all.every(item => typeof item === "string" && item.length > 0 && item.isWellFormed()
|
|
451
|
+
&& Buffer.byteLength(item) <= BROKER_CONFIG_BYTES) && new Set(all).size === all.length, "context-policy-invalid");
|
|
452
|
+
return strictJsonCopy(policy, "context-policy-invalid");
|
|
453
|
+
}
|
|
454
|
+
|
|
455
|
+
export function scopedContextPolicySha256(policy) {
|
|
456
|
+
return hash(JSON.stringify(scopedContextPolicy(policy)));
|
|
457
|
+
}
|
|
458
|
+
|
|
459
|
+
const contextToolNames = SCOPED_TOOL_DEFINITIONS.map(item => item.name);
|
|
460
|
+
function contextText(message, code) {
|
|
461
|
+
brokerRequire(Array.isArray(message.content) && message.content.length === 1, code);
|
|
462
|
+
brokerFields(message.content[0], ["type", "text"]);
|
|
463
|
+
brokerRequire(message.content[0].type === "text" && typeof message.content[0].text === "string", code);
|
|
464
|
+
return message.content[0].text;
|
|
465
|
+
}
|
|
466
|
+
function scopedWireTools(tools, code, exactWire = false) {
|
|
467
|
+
brokerRequire(Array.isArray(tools) && tools.length === SCOPED_TOOL_DEFINITIONS.length, code);
|
|
468
|
+
return tools.map((tool, index) => {
|
|
469
|
+
brokerRequire(plain(tool), code);
|
|
470
|
+
if (exactWire) brokerFields(tool, ["name", "label", "description", "parameters"]);
|
|
471
|
+
const definition = SCOPED_TOOL_DEFINITIONS[index], descriptor = Object.getOwnPropertyDescriptors(tool);
|
|
472
|
+
for (const name of ["name", "label", "description", "parameters"])
|
|
473
|
+
brokerRequire(descriptor[name]?.enumerable && Object.hasOwn(descriptor[name], "value"), code);
|
|
474
|
+
const parameters = strictJsonCopy(descriptor.parameters.value, code);
|
|
475
|
+
brokerRequire(descriptor.name.value === definition.name && descriptor.label.value === definition.name
|
|
476
|
+
&& descriptor.description.value === `Bounded broker operation ${definition.name}.`
|
|
477
|
+
&& JSON.stringify(parameters) === JSON.stringify(definition.inputSchema), code);
|
|
478
|
+
return { name: definition.name, label: definition.name,
|
|
479
|
+
description: `Bounded broker operation ${definition.name}.`, parameters };
|
|
480
|
+
});
|
|
481
|
+
}
|
|
482
|
+
|
|
483
|
+
/** Projects the Pi context to the signed G0 allowlist before any endpoint call. */
|
|
484
|
+
export function sanitizeScopedProviderContext(context, policy) {
|
|
485
|
+
policy = scopedContextPolicy(policy); brokerFields(context, ["systemPrompt", "messages", "tools"]);
|
|
486
|
+
brokerRequire(Array.isArray(context.messages) && context.messages.length <= 256, "context-message-invalid");
|
|
487
|
+
const allowed = new Set(policy.allowedUserMessages), removable = new Set(policy.removableUserMessages);
|
|
488
|
+
const seen = new Set(), messages = [];
|
|
489
|
+
for (const message of context.messages) {
|
|
490
|
+
brokerRequire(plain(message), "context-message-invalid");
|
|
491
|
+
if (message.role === "user") {
|
|
492
|
+
brokerFields(message, ["role", "content", "timestamp"]);
|
|
493
|
+
const text = contextText(message, "context-user-invalid");
|
|
494
|
+
if (removable.has(text)) continue;
|
|
495
|
+
brokerRequire(allowed.has(text) && !seen.has(text), `context-user-not-allowlisted:${hash(text)}`); seen.add(text);
|
|
496
|
+
messages.push({ role: "user", content: [{ type: "text", text }], timestamp: 0 }); continue;
|
|
497
|
+
}
|
|
498
|
+
if (message.role === "assistant") {
|
|
499
|
+
brokerFields(message, ["role", "content", "api", "provider", "model", "stopReason", "timestamp", "usage"]);
|
|
500
|
+
const copy = strictJsonCopy(message, "context-assistant-invalid");
|
|
501
|
+
brokerRequire(copy.content.every(part => part?.type === "text" && typeof part.text === "string"
|
|
502
|
+
|| part?.type === "toolCall" && typeof part.id === "string" && contextToolNames.includes(part.name)
|
|
503
|
+
&& plain(part.arguments)), "context-assistant-invalid");
|
|
504
|
+
copy.timestamp = 0; messages.push(copy); continue;
|
|
505
|
+
}
|
|
506
|
+
if (message.role === "toolResult") {
|
|
507
|
+
try { brokerFields(message, ["role", "toolCallId", "toolName", "content", "details", "usage", "isError", "timestamp"]); }
|
|
508
|
+
catch { brokerFail(`context-tool-result-fields:${Object.keys(message).sort().join(",")}`); }
|
|
509
|
+
const copy = strictJsonCopy(message, "context-tool-result-invalid");
|
|
510
|
+
brokerRequire(typeof copy.toolCallId === "string", "context-tool-result-call-id");
|
|
511
|
+
brokerRequire(contextToolNames.includes(copy.toolName), `context-tool-result-name:${String(copy.toolName)}`);
|
|
512
|
+
brokerRequire(typeof copy.isError === "boolean", "context-tool-result-error-flag");
|
|
513
|
+
brokerRequire(Array.isArray(copy.content) && copy.content.every(part => part?.type === "text"
|
|
514
|
+
&& typeof part.text === "string"), "context-tool-result-content");
|
|
515
|
+
copy.timestamp = 0; messages.push(copy); continue;
|
|
516
|
+
}
|
|
517
|
+
brokerFail("context-role-not-allowlisted");
|
|
518
|
+
}
|
|
519
|
+
brokerRequire(policy.allowedUserMessages.every(item => seen.has(item)), "context-user-missing");
|
|
520
|
+
const result = { systemPrompt: policy.systemPrompt, messages, tools: scopedWireTools(context.tools, "context-tools-invalid") };
|
|
521
|
+
brokerRequire(Buffer.byteLength(JSON.stringify(result)) <= 4 * 1024 * 1024, "context-byte-limit");
|
|
522
|
+
return deepFreeze(result);
|
|
523
|
+
}
|
|
524
|
+
|
|
525
|
+
export function validateScopedProviderPayload(payload, modelId, context) {
|
|
526
|
+
brokerFields(payload, ["model", "messages", "tools"]);
|
|
527
|
+
brokerRequire(typeof modelId === "string" && payload.model === modelId
|
|
528
|
+
&& JSON.stringify(strictJsonCopy(payload.messages, "final-payload-invalid")) === JSON.stringify(context.messages)
|
|
529
|
+
&& JSON.stringify(scopedWireTools(payload.tools, "final-payload-invalid", true)) === JSON.stringify(context.tools),
|
|
530
|
+
"final-payload-invalid");
|
|
531
|
+
return deepFreeze(strictJsonCopy(payload, "final-payload-invalid"));
|
|
532
|
+
}
|
|
533
|
+
|
|
534
|
+
const qualifiedLoopbackTransports = new WeakMap();
|
|
535
|
+
const LOOPBACK_RESPONSE_BYTES = 64 * 1024;
|
|
536
|
+
export function createQualifiedLoopbackTransport(value) {
|
|
537
|
+
brokerFields(value, ["origin"]);
|
|
538
|
+
let url;
|
|
539
|
+
try { url = new URL(value.origin); } catch { brokerFail("loopback-origin-invalid"); }
|
|
540
|
+
brokerRequire(url.protocol === "http:" && ["127.0.0.1", "::1", "localhost"].includes(url.hostname)
|
|
541
|
+
&& Number.isSafeInteger(Number(url.port)) && Number(url.port) > 0 && Number(url.port) <= 65535
|
|
542
|
+
&& !url.username && !url.password && url.pathname === "/" && !url.search && !url.hash,
|
|
543
|
+
"loopback-origin-invalid");
|
|
544
|
+
const state = { url: url.href, bound: false, validatedRequests: 0, callbackCount: 0,
|
|
545
|
+
evidenceRecords: 0, networkWrites: 0, responses: 0, failures: 0, lastRequestSha256: null, lastFailure: null };
|
|
546
|
+
const capability = Object.freeze({ status: () => Object.freeze({ version: 1, bound: state.bound,
|
|
547
|
+
validatedRequests: state.validatedRequests, callbackCount: state.callbackCount,
|
|
548
|
+
evidenceRecords: state.evidenceRecords, networkWrites: state.networkWrites, responses: state.responses,
|
|
549
|
+
failures: state.failures, lastRequestSha256: state.lastRequestSha256, lastFailure: state.lastFailure,
|
|
550
|
+
externalProviderCalls: 0 }) });
|
|
551
|
+
qualifiedLoopbackTransports.set(capability, state); return capability;
|
|
552
|
+
}
|
|
553
|
+
async function loopbackResponseBytes(response) {
|
|
554
|
+
brokerRequire(response?.status === 200 && /^application\/json(?:\s*;|$)/i.test(response.headers.get("content-type") ?? "")
|
|
555
|
+
&& response.body?.getReader, "loopback-response-invalid");
|
|
556
|
+
const reader = response.body.getReader(), chunks = []; let bytes = 0;
|
|
557
|
+
for (;;) {
|
|
558
|
+
const item = await reader.read(); if (item.done) break;
|
|
559
|
+
brokerRequire(item.value instanceof Uint8Array, "loopback-response-invalid");
|
|
560
|
+
bytes += item.value.length;
|
|
561
|
+
if (bytes > LOOPBACK_RESPONSE_BYTES) { await reader.cancel(); brokerFail("loopback-response-limit"); }
|
|
562
|
+
chunks.push(Buffer.from(item.value));
|
|
563
|
+
}
|
|
564
|
+
return Buffer.concat(chunks, bytes);
|
|
565
|
+
}
|
|
566
|
+
const qualifiedLoopbackUsage = () => ({ input: 0, output: 0, cacheRead: 0, cacheWrite: 0, totalTokens: 0,
|
|
567
|
+
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 } });
|
|
568
|
+
function qualifiedLoopbackMessage(value, model, sequence) {
|
|
569
|
+
brokerRequire(plain(value), "loopback-response-invalid");
|
|
570
|
+
const tool = value.kind === "tool";
|
|
571
|
+
brokerFields(value, tool ? ["kind", "name", "arguments"] : ["kind", "text"]);
|
|
572
|
+
if (tool) {
|
|
573
|
+
brokerRequire(SCOPED_TOOL_DEFINITIONS.some(item => item.name === value.name), "loopback-response-tool");
|
|
574
|
+
strictJsonCopy(value.arguments, "loopback-response-invalid");
|
|
575
|
+
} else brokerRequire(value.kind === "text" && typeof value.text === "string" && value.text.isWellFormed()
|
|
576
|
+
&& Buffer.byteLength(value.text) <= LOOPBACK_RESPONSE_BYTES, "loopback-response-invalid");
|
|
577
|
+
return { role: "assistant", content: tool
|
|
578
|
+
? [{ type: "toolCall", id: `call_${sequence}`, name: value.name, arguments: value.arguments }]
|
|
579
|
+
: [{ type: "text", text: value.text }], api: model.api, provider: model.provider, model: model.id,
|
|
580
|
+
stopReason: tool ? "toolUse" : "stop", timestamp: 0, usage: qualifiedLoopbackUsage() };
|
|
581
|
+
}
|
|
582
|
+
|
|
583
|
+
/** Replaces the caller's stream function with a host-owned, capability-branded
|
|
584
|
+
* transport. The SDK callback and exact wire bytes are validated before any
|
|
585
|
+
* evidence record or network write can occur. */
|
|
586
|
+
export async function createQualifiedLoopbackModelRuntime({ modelRuntime, transport, record, identity,
|
|
587
|
+
contextPolicy } = {}) {
|
|
588
|
+
const state = qualifiedLoopbackTransports.get(transport);
|
|
589
|
+
brokerRequire(state && !state.bound && modelRuntime && typeof modelRuntime === "object"
|
|
590
|
+
&& typeof record === "function" && identity && [identity.candidateDigest, identity.assetTreeSha256,
|
|
591
|
+
identity.brokerClosureSha256, identity.sdkTreeSha256].every(validHash), "loopback-capability-invalid");
|
|
592
|
+
brokerRequire(scopedContextPolicySha256(contextPolicy) === identity.contextPolicySha256,
|
|
593
|
+
"loopback-context-policy-mismatch");
|
|
594
|
+
const eventStreamFile = fs.realpathSync(path.join(identity.sdkRoot,
|
|
595
|
+
"node_modules/@earendil-works/pi-ai/dist/utils/event-stream.js"));
|
|
596
|
+
brokerRequire(eventStreamFile.startsWith(identity.sdkRoot + path.sep), "loopback-sdk-module-invalid");
|
|
597
|
+
const { AssistantMessageEventStream } = await import(pathToFileURL(eventStreamFile).href);
|
|
598
|
+
brokerRequire(typeof AssistantMessageEventStream === "function", "loopback-sdk-module-invalid");
|
|
599
|
+
state.bound = true;
|
|
600
|
+
const binding = Object.freeze({ armId: identity.armId, sourceSha256: identity.candidateDigest,
|
|
601
|
+
assetTreeSha256: identity.assetTreeSha256, sdkTreeSha256: identity.sdkTreeSha256,
|
|
602
|
+
brokerClosureSha256: identity.brokerClosureSha256, configSha256: identity.configSha256 });
|
|
603
|
+
const wrapped = Object.create(modelRuntime);
|
|
604
|
+
Object.defineProperty(wrapped, "streamSimple", { enumerable: true, value(model, context, options = {}) {
|
|
605
|
+
const stream = new AssistantMessageEventStream();
|
|
606
|
+
void (async () => {
|
|
607
|
+
try {
|
|
608
|
+
brokerRequire(model && typeof model.id === "string" && typeof model.api === "string"
|
|
609
|
+
&& typeof model.provider === "string", "loopback-model-invalid");
|
|
610
|
+
const input = observableJsonCopy(context, "arm-context-unobservable"), inputSha256 = hash(JSON.stringify(input));
|
|
611
|
+
const outgoing = sanitizeScopedProviderContext(context, contextPolicy);
|
|
612
|
+
const raw = { model: model.id, messages: strictJsonCopy(outgoing.messages, "final-payload-invalid"),
|
|
613
|
+
tools: outgoing.tools.map(tool => ({ name: tool.name, label: tool.label,
|
|
614
|
+
description: tool.description, parameters: strictJsonCopy(tool.parameters, "final-payload-invalid") })) };
|
|
615
|
+
const callback = options?.onPayload;
|
|
616
|
+
const returned = typeof callback === "function" ? await callback(raw, model) : raw;
|
|
617
|
+
state.callbackCount++;
|
|
618
|
+
const validated = validateScopedProviderPayload(returned === undefined ? raw : returned, model.id, outgoing);
|
|
619
|
+
const requestBytes = Buffer.from(JSON.stringify(validated));
|
|
620
|
+
brokerRequire(requestBytes.length <= 4 * 1024 * 1024, "final-payload-invalid");
|
|
621
|
+
let wire;
|
|
622
|
+
try { wire = JSON.parse(new TextDecoder("utf-8", { fatal: true }).decode(requestBytes)); }
|
|
623
|
+
catch { brokerFail("final-payload-invalid"); }
|
|
624
|
+
validateScopedProviderPayload(wire, model.id, outgoing);
|
|
625
|
+
state.validatedRequests++; state.lastRequestSha256 = hash(requestBytes);
|
|
626
|
+
const contextJson = JSON.stringify(outgoing);
|
|
627
|
+
await record(Object.freeze({ version: 3, kind: "provider-context", ...binding, inputSha256,
|
|
628
|
+
sha256: hash(contextJson), value: outgoing })); state.evidenceRecords++;
|
|
629
|
+
await record(Object.freeze({ version: 3, kind: "final-provider-payload", ...binding,
|
|
630
|
+
sha256: state.lastRequestSha256, bytes: requestBytes.length, value: validated })); state.evidenceRecords++;
|
|
631
|
+
brokerRequire(!options?.signal?.aborted, "loopback-cancelled");
|
|
632
|
+
state.networkWrites++;
|
|
633
|
+
const response = await fetch(state.url, { method: "POST", redirect: "error", signal: options?.signal,
|
|
634
|
+
headers: { "Content-Type": "application/json" }, body: requestBytes });
|
|
635
|
+
const responseBytes = await loopbackResponseBytes(response);
|
|
636
|
+
let next;
|
|
637
|
+
try { next = JSON.parse(new TextDecoder("utf-8", { fatal: true }).decode(responseBytes)); }
|
|
638
|
+
catch { brokerFail("loopback-response-invalid"); }
|
|
639
|
+
brokerRequire(responseBytes.equals(Buffer.from(JSON.stringify(next))), "loopback-response-noncanonical");
|
|
640
|
+
const message = qualifiedLoopbackMessage(next, model, state.responses + 1); state.responses++;
|
|
641
|
+
stream.push({ type: "done", reason: message.stopReason, message });
|
|
642
|
+
} catch (error) {
|
|
643
|
+
state.failures++; state.lastFailure = String(error?.stack ?? error)
|
|
644
|
+
+ (error?.cause ? `\nCAUSE: ${String(error.cause.stack ?? error.cause)}` : "");
|
|
645
|
+
const aborted = options?.signal?.aborted === true, reason = aborted ? "aborted" : "error";
|
|
646
|
+
stream.push({ type: "error", reason, error: { role: "assistant", content: [],
|
|
647
|
+
api: typeof model?.api === "string" ? model.api : "g0-loopback",
|
|
648
|
+
provider: typeof model?.provider === "string" ? model.provider : "loopback",
|
|
649
|
+
model: typeof model?.id === "string" ? model.id : "g0", stopReason: reason,
|
|
650
|
+
errorMessage: String(error?.message ?? "qualified-loopback-failed"), timestamp: 0,
|
|
651
|
+
usage: qualifiedLoopbackUsage() } });
|
|
652
|
+
}
|
|
653
|
+
})();
|
|
654
|
+
return stream;
|
|
655
|
+
} });
|
|
656
|
+
return wrapped;
|
|
657
|
+
}
|
|
658
|
+
|
|
659
|
+
// The transport spans the bounded benchmark session plus startup/teardown.
|
|
660
|
+
// Each verification retains its separate manifest deadline of at most 30s.
|
|
661
|
+
export const SCOPED_MCP_LIMITS = Object.freeze({ frameBytes: BROKER_TRANSPORT_BYTES,
|
|
662
|
+
inputBytes: BROKER_TRANSPORT_BYTES, outputBytes: BROKER_TRANSPORT_BYTES,
|
|
663
|
+
frames: 256, timeoutMs: 61 * 60 * 1000 });
|
|
664
|
+
const MCP_TURN_FIELDS = Object.freeze(["session_id", "thread_id", "turn_started_at_unix_ms", "turn_id",
|
|
665
|
+
"workspaces", "node_repl_disabled", "thread_source", "sandbox", "sandbox_mode", "auto_review_enabled",
|
|
666
|
+
"node_repl_auto_review_required", "model", "reasoning_effort"]);
|
|
667
|
+
const MCP_TURN_REQUIRED_FIELDS = Object.freeze(MCP_TURN_FIELDS.filter(key => key !== "reasoning_effort"));
|
|
668
|
+
const MCP_CALL_META_FIELDS = Object.freeze(["progressToken", "callId", "threadId", "itemId",
|
|
669
|
+
"x-codex-turn-metadata", "codex/sandbox-state-meta"]);
|
|
670
|
+
export const SCOPED_MCP_METADATA_CONTRACT = deepFreeze({
|
|
671
|
+
version: 3, id: "piagent-codex-mcp-metadata-v3", maxBytes: 64 * 1024,
|
|
672
|
+
commonFields: ["progressToken"], callFields: [...MCP_CALL_META_FIELDS],
|
|
673
|
+
turnFields: [...MCP_TURN_FIELDS], sandboxStateField: "codex/sandbox-state-meta",
|
|
674
|
+
authority: "none"
|
|
675
|
+
});
|
|
676
|
+
export const SCOPED_CODEX_TURN_OBSERVATION_VERSION = "codex-mcp-turn-observation-v1";
|