@piagent/platform 1.6.0 → 1.7.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +104 -0
- package/README.md +24 -15
- package/README.vi.md +36 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +27 -3
- package/benchmarks/production-v1/suite.json +5 -1
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +16 -16
- package/docs/authority-manifest.md +1 -1
- package/docs/auto-delegation-policy.md +38 -73
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +30 -35
- package/docs/context-engine.md +5 -3
- package/docs/context-window-policy.md +37 -1
- package/docs/distribution-standard.md +9 -9
- package/docs/en/architecture.md +26 -13
- package/docs/languages.json +4 -4
- package/docs/operator-manual-vietnamese.md +26 -46
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +7 -0
- package/docs/quality-benchmark.md +299 -23
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +32 -9
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/runtime-harness-standard.md +3 -3
- package/docs/security-threat-model.md +1 -1
- package/docs/subagent-orchestration-capabilities.md +19 -51
- package/docs/subagents-and-multiagent.md +21 -47
- package/docs/task-implementation-contract.md +3 -3
- package/docs/task-lifecycle-vietnamese.md +1 -1
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs/vi/architecture.md +26 -13
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +89 -95
- package/docs-site/commands.html +1 -1
- package/docs-site/content/benchmark.html +88 -94
- package/docs-site/content/en/benchmark.html +134 -90
- package/docs-site/content/en/index.html +19 -19
- package/docs-site/content/en/runtime.html +5 -5
- package/docs-site/content/en/whats-new.html +23 -1
- package/docs-site/content/en/workflows.html +2 -2
- package/docs-site/content/index.html +19 -19
- package/docs-site/content/runtime.html +6 -6
- package/docs-site/content/whats-new.html +23 -1
- package/docs-site/content/workflows.html +2 -2
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +135 -91
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +20 -20
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +6 -6
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -2
- package/docs-site/en/workflows.html +3 -3
- package/docs-site/index.html +20 -20
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +7 -7
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -2
- package/docs-site/workflows.html +3 -3
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-assurance-evidence.invalid.json +1 -1
- package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/codex-relative-efficiency-v1.invalid.json +4 -0
- package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/helper-request.invalid.json +1 -1
- package/evals/fixtures/helper-request.valid.json +1 -1
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fixtures/task-authority-snapshot.valid.json +4 -4
- package/evals/fs4-readiness-matrix.v1.json +2 -2
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +10 -6
- package/evals/fs5-pilot-protocol.v1.json +3 -3
- package/evals/fs5-pilot-protocol.v2.json +8 -8
- package/evals/fs5-pilot-protocol.v3.json +8 -8
- package/evals/fs5-pilot-protocol.v4.json +7 -7
- package/evals/fs5-pilot-protocol.v5.json +10 -8
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v1.json +1 -1
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/evals/runtime-conformance-v1/README.md +43 -0
- package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
- package/evals/runtime-conformance-v1/lane.json +57 -0
- package/evals/runtime-conformance-v1/paid-canary.json +31 -0
- package/evals/runtime-conformance-v1/runner.mjs +456 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +15 -6
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +699 -0
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +111 -110
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +67 -52
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +70 -7
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +462 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +343 -8
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-causal.js +244 -0
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +246 -33
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +232 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +410 -37
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +6 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +1000 -0
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +208 -180
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +12 -17
- package/packages/piagent-core/extensions/context-index-policy.js +4 -0
- package/packages/piagent-core/extensions/context-query-paths.js +57 -0
- package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
- package/packages/piagent-core/extensions/criterion-graph.js +4 -1
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
- package/packages/piagent-core/extensions/guard-types.ts +9 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-lifecycle.js +32 -0
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +140 -43
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +109 -20
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +43 -2
- package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
- package/packages/piagent-core/prompts/be-to-fe.md +2 -8
- package/packages/piagent-core/prompts/plan.md +1 -1
- package/packages/piagent-core/prompts/platform-improve.md +2 -6
- package/packages/piagent-core/prompts/review.md +2 -7
- package/packages/piagent-core/prompts/scout.md +1 -1
- package/packages/piagent-core/prompts/task.md +4 -3
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +83 -89
- package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +150 -19
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/hooks/tool-result-value-helpers.ts +66 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/bounded-emission-guard.ts +286 -0
- package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/task-subagent-tree.ts +27 -4
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/orchestration/helper-lifecycle.ts +16 -12
- package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
- package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
- package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
- package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
- package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +15 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/edit-freshness-guard.ts +102 -0
- package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
- package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +12 -3
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/pristine-task-policy-refresh.ts +91 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
- package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
- package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
- package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
- package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
- package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
- package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
- package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
- package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
- package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
- package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
- package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
- package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
- package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
- package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
- package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +10 -9
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +101 -0
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +201 -0
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-core/subagents/piagent-oracle.md +5 -3
- package/packages/piagent-core/subagents/piagent-planner.md +6 -4
- package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
- package/packages/piagent-core/subagents/piagent-scout.md +4 -2
- package/packages/piagent-core/subagents/piagent-worker.md +6 -4
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.js → MarkdownMessage-DGk34zJT.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +13 -3
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
- package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +170 -32
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
- package/packages/piagent-webui/gateway/session-operation-lifecycle.ts +237 -0
- package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +118 -117
- package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
- package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +150 -11
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/codex-relative-efficiency-v1.schema.json +155 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +46 -5
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/subagent-tree-v1.schema.json +11 -1
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/project-profile.schema.json +1 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +45 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +224 -387
- package/scripts/benchmark-runner-finalization.mjs +144 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-subagents.sh +42 -34
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/helper-evaluation.mjs +1 -1
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/team-doctor.sh +1 -1
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +41 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
|
@@ -5,6 +5,7 @@ import os from "node:os";
|
|
|
5
5
|
import path from "node:path";
|
|
6
6
|
|
|
7
7
|
import { materializeBenchmarkCandidate } from "./benchmark-candidate.js";
|
|
8
|
+
import { defaultBenchmarkSurfaces } from "./benchmark-cli.js";
|
|
8
9
|
import {
|
|
9
10
|
benchmarkPiCredentialFileIdentity,
|
|
10
11
|
benchmarkPiHomeConfigIdentity,
|
|
@@ -17,11 +18,15 @@ import { inspectBenchmarkLedger } from "./benchmark-ledger.js";
|
|
|
17
18
|
import { completedBenchmarkRecord } from "./benchmark-record-validation.js";
|
|
18
19
|
import { benchmarkGitEnvironment, benchmarkHostEnvironment } from "./benchmark-runtime.js";
|
|
19
20
|
import {
|
|
21
|
+
assertRegisteredBenchmarkPublicAssets,
|
|
20
22
|
canonicalBuiltInBenchmarkSuiteId,
|
|
21
|
-
isReservedBenchmarkSuiteId
|
|
23
|
+
isReservedBenchmarkSuiteId,
|
|
24
|
+
loadRegisteredBenchmarkMeasurement
|
|
22
25
|
} from "./benchmark-suite-identity.js";
|
|
23
26
|
|
|
24
27
|
const metadataVariable = "PIAGENT_BENCHMARK_BOOTSTRAP_METADATA";
|
|
28
|
+
export const registeredApprovalFileVariable = "PIAGENT_BENCHMARK_REGISTERED_APPROVAL_FILE";
|
|
29
|
+
export const registeredApprovalKeyVariable = "PIAGENT_BENCHMARK_REGISTERED_APPROVAL_PUBLIC_KEY_FILE";
|
|
25
30
|
|
|
26
31
|
function fail(message) {
|
|
27
32
|
const error = new Error(message);
|
|
@@ -34,6 +39,49 @@ function optionValue(argv, name) {
|
|
|
34
39
|
return index >= 0 ? argv[index + 1] : undefined;
|
|
35
40
|
}
|
|
36
41
|
|
|
42
|
+
function stableHostFile(file, label, maximum = 2 * 1024 * 1024) {
|
|
43
|
+
if (typeof file !== "string" || !path.isAbsolute(file) || path.normalize(file) !== file || file.includes("\0")) {
|
|
44
|
+
fail(`${label} path must be canonical and absolute`);
|
|
45
|
+
}
|
|
46
|
+
let resolved, before;
|
|
47
|
+
try { resolved = fs.realpathSync(file); before = fs.lstatSync(resolved, { bigint: true }); }
|
|
48
|
+
catch (error) { fail(`Cannot read ${label}: ${error.message}`); }
|
|
49
|
+
if (resolved !== file || !before.isFile() || before.isSymbolicLink() || before.nlink !== 1n
|
|
50
|
+
|| before.size < 1n || before.size > BigInt(maximum)
|
|
51
|
+
|| typeof process.getuid === "function" && before.uid !== BigInt(process.getuid())
|
|
52
|
+
|| (before.mode & 0o022n) !== 0n) fail(`${label} must be one host-owned non-writable-linked regular file`);
|
|
53
|
+
const fd = fs.openSync(resolved, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0));
|
|
54
|
+
try {
|
|
55
|
+
const descriptor = fs.fstatSync(fd, { bigint: true }), bytes = fs.readFileSync(fd),
|
|
56
|
+
after = fs.fstatSync(fd, { bigint: true }), current = fs.lstatSync(resolved, { bigint: true }),
|
|
57
|
+
fields = ["dev", "ino", "mode", "nlink", "size", "mtimeNs", "ctimeNs"];
|
|
58
|
+
if (bytes.length !== Number(before.size) || fields.some(name => before[name] !== descriptor[name]
|
|
59
|
+
|| before[name] !== after[name] || before[name] !== current[name])) fail(`${label} changed while read`);
|
|
60
|
+
return { path: resolved, bytes };
|
|
61
|
+
} finally { fs.closeSync(fd); }
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
function registeredManifestRequest(argv, cwd) {
|
|
65
|
+
const direct = optionValue(argv, "--registered-measurement");
|
|
66
|
+
if (direct) return direct;
|
|
67
|
+
const resume = optionValue(argv, "--resume");
|
|
68
|
+
if (!resume) return null;
|
|
69
|
+
const manifestPath = path.join(runRootFromResume(resume, cwd), "run-manifest.json");
|
|
70
|
+
if (!fs.existsSync(manifestPath)) return null;
|
|
71
|
+
const prior = jsonFile(manifestPath, "benchmark resume manifest")?.registeredMeasurement?.manifestOrigin;
|
|
72
|
+
return typeof prior === "string" ? prior : null;
|
|
73
|
+
}
|
|
74
|
+
|
|
75
|
+
function registeredAssetRoot(manifestFile) {
|
|
76
|
+
if (path.basename(manifestFile) !== "registered-suite.json"
|
|
77
|
+
|| path.basename(path.dirname(manifestFile)) !== "measurement") {
|
|
78
|
+
fail("Registered measurement manifest must be measurement/registered-suite.json inside its exact asset root");
|
|
79
|
+
}
|
|
80
|
+
const root = path.dirname(path.dirname(manifestFile));
|
|
81
|
+
if (fs.realpathSync(root) !== root) fail("Registered measurement asset root must be canonical");
|
|
82
|
+
return root;
|
|
83
|
+
}
|
|
84
|
+
|
|
37
85
|
function runRootFromResume(input, cwd) {
|
|
38
86
|
const target = path.resolve(cwd, input);
|
|
39
87
|
let stat;
|
|
@@ -124,6 +172,7 @@ export function requestedSuite(argv, cwd, replaySnapshot) {
|
|
|
124
172
|
if (argv.includes("--production")) return "production-v1";
|
|
125
173
|
if (argv.includes("--deep")) return "deep-logic-v1";
|
|
126
174
|
if (argv.includes("--capability")) return "capability-v1";
|
|
175
|
+
if (optionValue(argv, "--registered-measurement")) return "production-v2";
|
|
127
176
|
return optionValue(argv, "--suite") ?? "core-v1";
|
|
128
177
|
}
|
|
129
178
|
|
|
@@ -131,6 +180,9 @@ function snapshotReplay(argv, cwd, temporaryRoot) {
|
|
|
131
180
|
const input = optionValue(argv, "--replay-failures");
|
|
132
181
|
if (!input) return null;
|
|
133
182
|
const origin = path.resolve(cwd, input);
|
|
183
|
+
if (argv.includes("--failed-attempts-only") && fs.existsSync(path.join(path.dirname(origin), "aborted.json"))) {
|
|
184
|
+
fail("Failed-only replay cannot snapshot an aborted source run");
|
|
185
|
+
}
|
|
134
186
|
let bytes;
|
|
135
187
|
let report;
|
|
136
188
|
try {
|
|
@@ -219,6 +271,16 @@ function validateOutputIsolation(argv, cwd, repositoryRoot, suiteManifest) {
|
|
|
219
271
|
}
|
|
220
272
|
}
|
|
221
273
|
|
|
274
|
+
function validateRegisteredOutputIsolation(argv, cwd, assetRoot) {
|
|
275
|
+
const resume = optionValue(argv, "--resume");
|
|
276
|
+
const requested = resume
|
|
277
|
+
? runRootFromResume(resume, cwd)
|
|
278
|
+
: optionValue(argv, "--output") ? path.resolve(cwd, optionValue(argv, "--output")) : null;
|
|
279
|
+
if (requested && inside(assetRoot, canonicalProspective(requested))) {
|
|
280
|
+
fail("Benchmark output must not be nested inside the registered measurement asset source");
|
|
281
|
+
}
|
|
282
|
+
}
|
|
283
|
+
|
|
222
284
|
function copyTree(source, target) {
|
|
223
285
|
const pending = [[source, target]];
|
|
224
286
|
while (pending.length > 0) {
|
|
@@ -291,6 +353,89 @@ function runtimeDependencies(candidateRoot, resolutionRoot) {
|
|
|
291
353
|
};
|
|
292
354
|
}
|
|
293
355
|
|
|
356
|
+
function fileDigest(file) {
|
|
357
|
+
return crypto.createHash("sha256").update(fs.readFileSync(file)).digest("hex");
|
|
358
|
+
}
|
|
359
|
+
|
|
360
|
+
function suiteNeedsWebUiAssets(manifest) {
|
|
361
|
+
const suite = jsonFile(manifest, "benchmark suite manifest");
|
|
362
|
+
return Array.isArray(suite?.scenarios) && suite.scenarios.some((scenario) => scenario?.userJourney);
|
|
363
|
+
}
|
|
364
|
+
|
|
365
|
+
function snapshotWebUiAssets(candidateRoot, liveRoot, temporaryRoot, candidateProvenance) {
|
|
366
|
+
const sourcePackage = path.join(candidateRoot, "packages", "piagent-webui");
|
|
367
|
+
const sourceManifest = jsonFile(path.join(sourcePackage, "package.json"), "candidate WebUI package manifest");
|
|
368
|
+
const nodeModules = fs.realpathSync(path.join(liveRoot, "node_modules"));
|
|
369
|
+
const installed = (name) => jsonFile(path.join(nodeModules, ...name.split("/"), "package.json"), `WebUI build dependency ${name}`);
|
|
370
|
+
const vite = installed("vite"), reactPlugin = installed("@vitejs/plugin-react");
|
|
371
|
+
if (sourceManifest.devDependencies?.vite !== vite.version
|
|
372
|
+
|| sourceManifest.devDependencies?.["@vitejs/plugin-react"] !== reactPlugin.version) {
|
|
373
|
+
fail("Cannot build frozen WebUI assets: installed Vite toolchain does not match the candidate manifest");
|
|
374
|
+
}
|
|
375
|
+
const buildRoot = path.join(temporaryRoot, "webui-build-source");
|
|
376
|
+
const buildPackage = path.join(buildRoot, "packages", "piagent-webui");
|
|
377
|
+
const assetRoot = path.join(temporaryRoot, "webui-assets");
|
|
378
|
+
fs.mkdirSync(path.dirname(buildPackage), { recursive: true, mode: 0o700 });
|
|
379
|
+
copyTree(sourcePackage, buildPackage);
|
|
380
|
+
chmodTree(buildRoot, true);
|
|
381
|
+
fs.symlinkSync(nodeModules, path.join(buildRoot, "node_modules"), "dir");
|
|
382
|
+
try {
|
|
383
|
+
const viteCli = path.join(nodeModules, "vite", String(vite.bin?.vite ?? "bin/vite.js"));
|
|
384
|
+
execFileSync(process.execPath, [viteCli, "build", "--config", path.join(buildPackage, "vite.config.ts"),
|
|
385
|
+
"--outDir", assetRoot, "--emptyOutDir"], {
|
|
386
|
+
cwd: buildRoot,
|
|
387
|
+
env: benchmarkHostEnvironment({ ...process.env, NO_COLOR: "1" }),
|
|
388
|
+
stdio: ["ignore", "pipe", "pipe"],
|
|
389
|
+
maxBuffer: 32 * 1024 * 1024
|
|
390
|
+
});
|
|
391
|
+
} catch (error) {
|
|
392
|
+
fail(`Cannot build frozen WebUI assets: ${String(error?.stderr ?? error?.message ?? error).trim()}`);
|
|
393
|
+
} finally {
|
|
394
|
+
chmodTree(buildRoot, true);
|
|
395
|
+
fs.rmSync(buildRoot, { recursive: true, force: true });
|
|
396
|
+
}
|
|
397
|
+
if (!fs.existsSync(path.join(assetRoot, "index.html"))) fail("Frozen WebUI build did not produce index.html");
|
|
398
|
+
chmodTree(assetRoot, false);
|
|
399
|
+
const identity = {
|
|
400
|
+
schemaVersion: 1,
|
|
401
|
+
sourceCandidateDigest: candidateProvenance.contentDigest,
|
|
402
|
+
lockfileDigest: fileDigest(path.join(candidateRoot, "package-lock.json")),
|
|
403
|
+
pipeline: "vite-production-build-v1",
|
|
404
|
+
node: process.version,
|
|
405
|
+
tools: { vite: vite.version, reactPlugin: reactPlugin.version },
|
|
406
|
+
tree: benchmarkTreeIdentity(assetRoot, { rejectSymlinks: true })
|
|
407
|
+
};
|
|
408
|
+
return {
|
|
409
|
+
...identity,
|
|
410
|
+
digest: crypto.createHash("sha256").update(JSON.stringify(identity)).digest("hex"),
|
|
411
|
+
root: assetRoot
|
|
412
|
+
};
|
|
413
|
+
}
|
|
414
|
+
|
|
415
|
+
function validWebUiAssets(value, candidateProvenance) {
|
|
416
|
+
if (value === null) return true;
|
|
417
|
+
if (!value || typeof value !== "object" || Array.isArray(value)
|
|
418
|
+
|| value.schemaVersion !== 1 || typeof value.root !== "string" || !path.isAbsolute(value.root)
|
|
419
|
+
|| value.sourceCandidateDigest !== candidateProvenance?.contentDigest
|
|
420
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.lockfileDigest ?? ""))
|
|
421
|
+
|| value.pipeline !== "vite-production-build-v1" || value.node !== process.version
|
|
422
|
+
|| typeof value.tools?.vite !== "string" || typeof value.tools?.reactPlugin !== "string"
|
|
423
|
+
|| value.tree?.schemaVersion !== 1 || typeof value.tree?.algorithm !== "string"
|
|
424
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.tree?.contentDigest ?? ""))
|
|
425
|
+
|| !Number.isInteger(value.tree?.entryCount) || value.tree.entryCount < 1
|
|
426
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.digest ?? ""))) return false;
|
|
427
|
+
const identity = {
|
|
428
|
+
schemaVersion: value.schemaVersion,
|
|
429
|
+
sourceCandidateDigest: value.sourceCandidateDigest,
|
|
430
|
+
lockfileDigest: value.lockfileDigest,
|
|
431
|
+
pipeline: value.pipeline,
|
|
432
|
+
node: value.node,
|
|
433
|
+
tools: value.tools,
|
|
434
|
+
tree: value.tree
|
|
435
|
+
};
|
|
436
|
+
return crypto.createHash("sha256").update(JSON.stringify(identity)).digest("hex") === value.digest;
|
|
437
|
+
}
|
|
438
|
+
|
|
294
439
|
function writeBoundJson(file, value) {
|
|
295
440
|
const bytes = Buffer.from(`${JSON.stringify(value)}\n`, "utf8");
|
|
296
441
|
fs.writeFileSync(file, bytes, { mode: 0o400 });
|
|
@@ -329,9 +474,11 @@ function piAuthInvocation(argv, cwd, replay) {
|
|
|
329
474
|
if (!/^[a-f0-9]{32}$/.test(String(manifest.piCredentialVaultId ?? ""))) fail("Cannot resume benchmark: private Pi credential vault identity is missing or unsupported");
|
|
330
475
|
return { authorized: manifest.allowPiAuthWriteback === true, model: manifest.model ?? null, vaultId: manifest.piCredentialVaultId };
|
|
331
476
|
}
|
|
477
|
+
const registered = registeredManifestRequest(argv, cwd);
|
|
332
478
|
return {
|
|
333
479
|
authorized: argv.includes("--allow-pi-auth-writeback"),
|
|
334
|
-
model: optionValue(argv, "--model")
|
|
480
|
+
model: (registered ? "openai-codex/gpt-5.6-luna" : optionValue(argv, "--model"))
|
|
481
|
+
?? replay?.manifest?.model ?? replay?.report?.environment?.requestedModel ?? null,
|
|
335
482
|
vaultId: crypto.randomBytes(16).toString("hex")
|
|
336
483
|
};
|
|
337
484
|
}
|
|
@@ -358,7 +505,8 @@ function snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay, {
|
|
|
358
505
|
catch (error) { if (error?.code === "ENOENT") continue; throw error; }
|
|
359
506
|
copied.push(name);
|
|
360
507
|
}
|
|
361
|
-
|
|
508
|
+
// Cache refreshes are provider requests; keep them outside the staged measurement.
|
|
509
|
+
writeBoundJson(path.join(configRoot, "settings.json"), { cacheWarming: "off" });
|
|
362
510
|
fs.chmodSync(configRoot, 0o500);
|
|
363
511
|
fs.chmodSync(runtimeParent, 0o700);
|
|
364
512
|
const seedIdentity = benchmarkPiHomeConfigIdentity(configRoot, { requiredFileMode: "400" });
|
|
@@ -372,7 +520,7 @@ function snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay, {
|
|
|
372
520
|
} : null;
|
|
373
521
|
const identity = {
|
|
374
522
|
...benchmarkPiHomePublicIdentity(configRoot, credentialReadiness.map(({ providerId, type }) => ({ providerId, type }))),
|
|
375
|
-
settingsPolicy: "deterministic-
|
|
523
|
+
settingsPolicy: "deterministic-cache-warming-off-v1",
|
|
376
524
|
operatorPackagesAndResources: "excluded"
|
|
377
525
|
};
|
|
378
526
|
return {
|
|
@@ -396,37 +544,42 @@ function snapshotPiAgentHome(temporaryRoot, runtimeParent, argv, cwd, replay, {
|
|
|
396
544
|
function snapshotCodexCredential(temporaryRoot) {
|
|
397
545
|
const source = path.join(path.resolve(process.env.CODEX_HOME || path.join(os.homedir(), ".codex")), "auth.json");
|
|
398
546
|
if (!fs.existsSync(source)) return null;
|
|
399
|
-
|
|
547
|
+
const stable = stableHostFile(source, "Codex credential", 2 * 1024 * 1024);
|
|
400
548
|
const targetRoot = path.join(temporaryRoot, "codex-credential");
|
|
401
549
|
fs.mkdirSync(targetRoot, { recursive: true, mode: 0o700 });
|
|
402
550
|
const target = path.join(targetRoot, "auth.json");
|
|
403
|
-
fs.
|
|
404
|
-
fs.chmodSync(target, 0o400);
|
|
551
|
+
fs.writeFileSync(target, stable.bytes, { flag: "wx", mode: 0o400 });
|
|
405
552
|
fs.chmodSync(targetRoot, 0o500);
|
|
406
553
|
return {
|
|
407
554
|
path: target,
|
|
408
555
|
privateIdentity: benchmarkTreeIdentity(targetRoot, { rejectSymlinks: true }),
|
|
409
|
-
identity: { schemaVersion: 1, credentialPresent: true, contentBinding: "private-runtime-only"
|
|
556
|
+
identity: { schemaVersion: 1, credentialPresent: true, contentBinding: "private-runtime-only",
|
|
557
|
+
environmentCredentialPolicy: "excluded-from-frozen-controlled-runtime" }
|
|
410
558
|
};
|
|
411
559
|
}
|
|
412
560
|
|
|
413
561
|
function requestsCodexSurface(argv, cwd, replay) {
|
|
414
|
-
|
|
415
|
-
if (explicit) return explicit.split(",").map((value) => value.trim()).includes("codex-cli");
|
|
562
|
+
if (registeredManifestRequest(argv, cwd)) return true;
|
|
416
563
|
const resume = optionValue(argv, "--resume");
|
|
417
564
|
if (resume) {
|
|
418
565
|
const manifestPath = path.join(runRootFromResume(resume, cwd), "run-manifest.json");
|
|
419
566
|
return fs.existsSync(manifestPath) && jsonFile(manifestPath, "benchmark resume manifest")?.surfaces?.includes("codex-cli") === true;
|
|
420
567
|
}
|
|
421
|
-
|
|
568
|
+
if (replay) {
|
|
569
|
+
const surfaces = replay.manifest?.surfaces ?? replay.report?.environment?.surfaces;
|
|
570
|
+
return Array.isArray(surfaces) && surfaces.includes("codex-cli");
|
|
571
|
+
}
|
|
572
|
+
const explicit = optionValue(argv, "--surfaces");
|
|
573
|
+
if (explicit) return explicit.split(",").map((value) => value.trim()).includes("codex-cli");
|
|
574
|
+
return defaultBenchmarkSurfaces.includes("codex-cli");
|
|
422
575
|
}
|
|
423
576
|
|
|
424
577
|
function isolatedSnapshotParent() {
|
|
425
|
-
return fs.mkdtempSync(path.join(os.tmpdir(), "piagent-benchmark-snapshot-"));
|
|
578
|
+
return fs.realpathSync.native(fs.mkdtempSync(path.join(os.tmpdir(), "piagent-benchmark-snapshot-")));
|
|
426
579
|
}
|
|
427
580
|
|
|
428
581
|
function isolatedPiRuntimeParent() {
|
|
429
|
-
return fs.mkdtempSync(path.join(os.tmpdir(), "piagent-benchmark-pi-runtime-"));
|
|
582
|
+
return fs.realpathSync.native(fs.mkdtempSync(path.join(os.tmpdir(), "piagent-benchmark-pi-runtime-")));
|
|
430
583
|
}
|
|
431
584
|
|
|
432
585
|
function gitSourceIdentity(root) {
|
|
@@ -447,6 +600,60 @@ function gitSourceIdentity(root) {
|
|
|
447
600
|
};
|
|
448
601
|
}
|
|
449
602
|
|
|
603
|
+
export function snapshotRegisteredBenchmarkMeasurement({ argv, cwd, temporaryRoot, scenarioIds }) {
|
|
604
|
+
const requested = registeredManifestRequest(argv, cwd);
|
|
605
|
+
if (!requested) return null;
|
|
606
|
+
if (!path.isAbsolute(requested) || path.normalize(requested) !== requested || requested.includes("\0")) {
|
|
607
|
+
fail("Registered measurement manifest path must be canonical and absolute");
|
|
608
|
+
}
|
|
609
|
+
let manifestFile;
|
|
610
|
+
try { manifestFile = fs.realpathSync(requested); }
|
|
611
|
+
catch (error) { fail(`Cannot resolve registered measurement manifest: ${error.message}`); }
|
|
612
|
+
if (manifestFile !== requested) fail("Registered measurement manifest path must be canonical and absolute");
|
|
613
|
+
const approval = stableHostFile(process.env[registeredApprovalFileVariable],
|
|
614
|
+
"registered measurement detached approval");
|
|
615
|
+
const publicKeyFile = stableHostFile(process.env[registeredApprovalKeyVariable],
|
|
616
|
+
"registered measurement trusted public key", 64 * 1024);
|
|
617
|
+
let publicKey;
|
|
618
|
+
try { publicKey = crypto.createPublicKey(publicKeyFile.bytes); }
|
|
619
|
+
catch { fail("Registered measurement trusted public key is invalid"); }
|
|
620
|
+
const registration = loadRegisteredBenchmarkMeasurement(manifestFile, {
|
|
621
|
+
approvalFile: approval.path,
|
|
622
|
+
approvalPublicKey: publicKey,
|
|
623
|
+
scenarioIds
|
|
624
|
+
});
|
|
625
|
+
const sourceRoot = registeredAssetRoot(manifestFile);
|
|
626
|
+
validateRegisteredOutputIsolation(argv, cwd, sourceRoot);
|
|
627
|
+
const sourceInventory = assertRegisteredBenchmarkPublicAssets(registration, sourceRoot, scenarioIds);
|
|
628
|
+
const targetRoot = path.join(temporaryRoot, "registered-assets");
|
|
629
|
+
copyTree(sourceRoot, targetRoot);
|
|
630
|
+
chmodTree(targetRoot, false);
|
|
631
|
+
const snapshotRoot = fs.realpathSync(targetRoot), snapshotManifest = path.join(snapshotRoot,
|
|
632
|
+
"measurement", "registered-suite.json"), snapshotInventory = assertRegisteredBenchmarkPublicAssets(
|
|
633
|
+
registration, snapshotRoot, scenarioIds);
|
|
634
|
+
if (JSON.stringify(sourceInventory) !== JSON.stringify(snapshotInventory)
|
|
635
|
+
|| fileDigest(snapshotManifest) !== registration.identity.manifestSha256) {
|
|
636
|
+
fail("Registered measurement snapshot identity changed while it was frozen");
|
|
637
|
+
}
|
|
638
|
+
return {
|
|
639
|
+
schemaVersion: 1,
|
|
640
|
+
manifestOrigin: manifestFile,
|
|
641
|
+
manifestSnapshot: snapshotManifest,
|
|
642
|
+
assetRoot: snapshotRoot,
|
|
643
|
+
payload: registration.payload,
|
|
644
|
+
identity: registration.identity,
|
|
645
|
+
inventory: snapshotInventory,
|
|
646
|
+
tree: benchmarkTreeIdentity(snapshotRoot, { rejectSymlinks: true })
|
|
647
|
+
};
|
|
648
|
+
}
|
|
649
|
+
|
|
650
|
+
function requestedCandidateTreatment(argv, cwd) {
|
|
651
|
+
const resume = optionValue(argv, "--resume");
|
|
652
|
+
const manifest = resume ? jsonFile(path.join(runRootFromResume(resume, cwd), "run-manifest.json"), "resume treatment") : null;
|
|
653
|
+
return { piagentTreatment: manifest?.piagentTreatment ?? optionValue(argv, "--piagent-treatment"),
|
|
654
|
+
measurementOnly: manifest ? manifest.measurementOnly === true : argv.includes("--measurement-only") };
|
|
655
|
+
}
|
|
656
|
+
|
|
450
657
|
export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
|
|
451
658
|
const root = fs.realpathSync(liveRoot);
|
|
452
659
|
let temporaryRoot;
|
|
@@ -461,7 +668,7 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
|
|
|
461
668
|
try {
|
|
462
669
|
const sourceIdentityBefore = gitSourceIdentity(root);
|
|
463
670
|
const candidateRoot = path.join(temporaryRoot, "candidate");
|
|
464
|
-
const candidate = materializeBenchmarkCandidate(root, candidateRoot);
|
|
671
|
+
const candidate = materializeBenchmarkCandidate(root, candidateRoot, requestedCandidateTreatment(argv, cwd));
|
|
465
672
|
const sourceIdentityAfter = gitSourceIdentity(root);
|
|
466
673
|
if (JSON.stringify(sourceIdentityBefore) !== JSON.stringify(sourceIdentityAfter)) {
|
|
467
674
|
fail("Benchmark Git source identity changed while the candidate snapshot was being created");
|
|
@@ -487,6 +694,19 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
|
|
|
487
694
|
} else {
|
|
488
695
|
suiteRoot = path.join(candidateRoot, "benchmarks", suite.origin);
|
|
489
696
|
}
|
|
697
|
+
const frozenSuiteManifest = suiteSnapshot ?? path.join(suiteRoot, "suite.json");
|
|
698
|
+
const frozenSuite = jsonFile(frozenSuiteManifest, "frozen benchmark suite manifest");
|
|
699
|
+
const registeredMeasurement = snapshotRegisteredBenchmarkMeasurement({ argv, cwd, temporaryRoot,
|
|
700
|
+
scenarioIds: frozenSuite.scenarios?.map(value => value.id) });
|
|
701
|
+
if (registeredMeasurement && suite.origin !== registeredMeasurement.payload.baseSuiteId) {
|
|
702
|
+
fail("Registered measurement base suite does not match its frozen benchmark suite");
|
|
703
|
+
}
|
|
704
|
+
if (registeredMeasurement && !providerFreeFinalization && !codexCredential) {
|
|
705
|
+
fail("Registered measurement requires one frozen Codex auth.json credential");
|
|
706
|
+
}
|
|
707
|
+
const webUiAssets = suiteNeedsWebUiAssets(frozenSuiteManifest)
|
|
708
|
+
? snapshotWebUiAssets(candidateRoot, root, temporaryRoot, candidate.provenance)
|
|
709
|
+
: null;
|
|
490
710
|
const metadata = {
|
|
491
711
|
schemaVersion: 1,
|
|
492
712
|
liveRoot: root,
|
|
@@ -495,11 +715,14 @@ export function createBenchmarkExecutionSnapshot({ liveRoot, argv, cwd }) {
|
|
|
495
715
|
sourceIdentity: sourceIdentityAfter,
|
|
496
716
|
defaultOutputRoot: piAgentHome.defaultOutputRoot,
|
|
497
717
|
candidateProvenance: candidate.provenance,
|
|
718
|
+
...(candidate.treatmentDerivation ? { treatmentDerivation: candidate.treatmentDerivation } : {}),
|
|
498
719
|
candidateIndex,
|
|
499
720
|
runtimeDependencies: runtimeDependencies(candidateRoot, root),
|
|
721
|
+
webUiAssets,
|
|
500
722
|
providerFreeFinalization,
|
|
501
723
|
piAgentHome,
|
|
502
724
|
codexCredential,
|
|
725
|
+
registeredMeasurement,
|
|
503
726
|
replay: replay ? {
|
|
504
727
|
origin: replay.origin,
|
|
505
728
|
snapshot: replay.snapshot,
|
|
@@ -537,6 +760,12 @@ export function cleanupBenchmarkExecutionSnapshot(temporaryRoot, runtimeParent,
|
|
|
537
760
|
export function benchmarkBootstrapEnvironment(metadata, base = process.env) {
|
|
538
761
|
const environment = benchmarkHostEnvironment(base);
|
|
539
762
|
delete environment.PI_CODING_AGENT_DIR;
|
|
763
|
+
delete environment[registeredApprovalFileVariable];
|
|
764
|
+
delete environment[registeredApprovalKeyVariable];
|
|
765
|
+
if (metadata.registeredMeasurement) {
|
|
766
|
+
delete environment.OPENAI_API_KEY;
|
|
767
|
+
delete environment.CODEX_ACCESS_TOKEN;
|
|
768
|
+
}
|
|
540
769
|
return {
|
|
541
770
|
...environment,
|
|
542
771
|
...(metadata.codexCredential ? { PIAGENT_BENCHMARK_CODEX_AUTH_SNAPSHOT: metadata.codexCredential.path } : {}),
|
|
@@ -563,6 +792,7 @@ export function benchmarkBootstrapMetadata(env = process.env) {
|
|
|
563
792
|
|| typeof value.candidateIndex?.path !== "string"
|
|
564
793
|
|| !/^[a-f0-9]{64}$/.test(String(value.candidateIndex?.digest ?? ""))
|
|
565
794
|
|| typeof value.runtimeDependencies?.digest !== "string"
|
|
795
|
+
|| !validWebUiAssets(value.webUiAssets, value.candidateProvenance)
|
|
566
796
|
|| typeof value.providerFreeFinalization !== "boolean"
|
|
567
797
|
|| typeof value.piAgentHome?.configRoot !== "string"
|
|
568
798
|
|| typeof value.piAgentHome?.runtimeParent !== "string"
|
|
@@ -581,6 +811,27 @@ export function benchmarkBootstrapMetadata(env = process.env) {
|
|
|
581
811
|
typeof value.codexCredential?.path !== "string"
|
|
582
812
|
|| typeof value.codexCredential?.privateIdentity?.contentDigest !== "string"
|
|
583
813
|
))
|
|
814
|
+
|| (value.registeredMeasurement !== undefined && value.registeredMeasurement !== null
|
|
815
|
+
&& value.providerFreeFinalization === false && value.codexCredential === null)
|
|
816
|
+
|| (value.registeredMeasurement !== undefined && value.registeredMeasurement !== null && (
|
|
817
|
+
value.registeredMeasurement?.schemaVersion !== 1
|
|
818
|
+
|| typeof value.registeredMeasurement?.manifestOrigin !== "string"
|
|
819
|
+
|| typeof value.registeredMeasurement?.manifestSnapshot !== "string"
|
|
820
|
+
|| typeof value.registeredMeasurement?.assetRoot !== "string"
|
|
821
|
+
|| value.registeredMeasurement?.payload?.suiteId !== "production-v2-da2"
|
|
822
|
+
|| value.registeredMeasurement?.payload?.baseSuiteId !== "production-v2"
|
|
823
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.identity?.manifestSha256 ?? ""))
|
|
824
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.identity?.payloadSha256 ?? ""))
|
|
825
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.identity?.approvalRecordSha256 ?? ""))
|
|
826
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.identity?.authorityKeyId ?? ""))
|
|
827
|
+
|| value.registeredMeasurement?.inventory?.version !== "registered-public-asset-inventory-v1"
|
|
828
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.inventory?.publicAssetsDigest ?? ""))
|
|
829
|
+
|| value.registeredMeasurement?.inventory?.publicAssetsDigest !== value.registeredMeasurement?.payload?.publicAssetsDigest
|
|
830
|
+
|| value.registeredMeasurement?.inventory?.publicContractDigest !== value.registeredMeasurement?.payload?.publicContractDigest
|
|
831
|
+
|| !Array.isArray(value.registeredMeasurement?.inventory?.entries)
|
|
832
|
+
|| value.registeredMeasurement.inventory.entries.length !== 48
|
|
833
|
+
|| !/^[a-f0-9]{64}$/.test(String(value.registeredMeasurement?.tree?.contentDigest ?? ""))
|
|
834
|
+
))
|
|
584
835
|
|| typeof value.suite?.origin !== "string"
|
|
585
836
|
|| !(value.suite?.builtInId === null || isReservedBenchmarkSuiteId(value.suite?.builtInId))
|
|
586
837
|
|| (value.suite?.builtInId === null && typeof value.suite?.snapshot !== "string")
|