@piagent/platform 1.6.1 → 1.7.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +55 -0
- package/README.md +8 -2
- package/README.vi.md +11 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +3 -3
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +2 -2
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/operator-manual-vietnamese.md +12 -12
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +2 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +32 -9
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +11 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +11 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +2 -2
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +12 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +3 -3
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +12 -1
- package/docs-site/workflows.html +1 -1
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +107 -107
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +63 -50
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +69 -8
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +162 -45
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +144 -128
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +395 -33
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +3 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +264 -70
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +139 -152
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +1 -0
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +12 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +11 -11
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DGk34zJT.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-DiKRmCEP.js +176 -0
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +141 -19
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +11 -2
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +65 -12
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +32 -1
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +30 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +36 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "http://json-schema.org/draft-07/schema#",
|
|
3
|
+
"$id": "https://piagent.dev/schemas/contract-family-library.schema.json",
|
|
4
|
+
"title": "Declarative reusable contract library",
|
|
5
|
+
"description": "Bounded data templates, not executable oracle code. A {$parameter: name} node is replaced with one typed binding, never evaluated or recursively interpreted. Runtime additionally checks unique identities, used references and full expanded contract semantics.",
|
|
6
|
+
"type": "object", "additionalProperties": false, "required": ["schemaVersion", "families"],
|
|
7
|
+
"properties": {
|
|
8
|
+
"schemaVersion": { "const": 1 },
|
|
9
|
+
"families": { "type": "array", "minItems": 1, "maxItems": 32, "items": {
|
|
10
|
+
"type": "object", "additionalProperties": false,
|
|
11
|
+
"required": ["id", "version", "domain", "description", "parameters", "template"],
|
|
12
|
+
"properties": {
|
|
13
|
+
"id": { "type": "string", "pattern": "^[a-zA-Z0-9][a-zA-Z0-9:._-]{0,79}$" },
|
|
14
|
+
"version": { "type": "integer", "minimum": 1, "maximum": 1000000 },
|
|
15
|
+
"domain": { "enum": ["pure-function", "temporal-input", "configuration-precedence", "stateful-recovery"] },
|
|
16
|
+
"description": { "type": "string", "minLength": 1, "maxLength": 2000 },
|
|
17
|
+
"parameters": { "type": "object", "maxProperties": 32, "propertyNames": { "pattern": "^[a-zA-Z][a-zA-Z0-9_]{0,63}$" },
|
|
18
|
+
"additionalProperties": { "enum": ["export", "string", "number", "value"] } },
|
|
19
|
+
"template": { "type": "object", "additionalProperties": false, "required": ["exportName", "checks"], "properties": {
|
|
20
|
+
"exportName": { "oneOf": [ { "type": "string", "pattern": "^[a-zA-Z_$][a-zA-Z0-9_$]{0,127}$" }, { "$ref": "#/$defs/reference" } ] },
|
|
21
|
+
"checks": { "type": "array", "minItems": 1, "maxItems": 256, "items": { "type": "object" } }
|
|
22
|
+
} }
|
|
23
|
+
}
|
|
24
|
+
} }
|
|
25
|
+
},
|
|
26
|
+
"$defs": { "reference": { "type": "object", "additionalProperties": false, "required": ["$parameter"],
|
|
27
|
+
"properties": { "$parameter": { "type": "string", "pattern": "^[a-zA-Z][a-zA-Z0-9_]{0,63}$" } } } }
|
|
28
|
+
}
|
|
@@ -0,0 +1,31 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "http://json-schema.org/draft-07/schema#",
|
|
3
|
+
"$id": "https://piagent.dev/schemas/contract-selection-recipe.schema.json",
|
|
4
|
+
"title": "Reusable contract selection recipe (not authority or evidence)",
|
|
5
|
+
"description": "Exact criterion text/obligation matching against a task snapshot. No fuzzy source matching. Parameters substitute complete data nodes; the compiler validates the fully expanded independent plan. Unknown or ambiguous requested selections produce no plan.",
|
|
6
|
+
"type": "object", "additionalProperties": false,
|
|
7
|
+
"required": ["schemaVersion", "backend", "selections"],
|
|
8
|
+
"properties": {
|
|
9
|
+
"schemaVersion": { "const": 1 },
|
|
10
|
+
"backend": { "$ref": "approved-host-contracts.schema.json#/$defs/backend" },
|
|
11
|
+
"selections": { "type": "array", "minItems": 1, "maxItems": 12, "items": {
|
|
12
|
+
"type": "object", "additionalProperties": false,
|
|
13
|
+
"required": ["criterion", "family", "parameters", "sourcePath", "maxAttempts"],
|
|
14
|
+
"properties": {
|
|
15
|
+
"criterion": { "type": "object", "additionalProperties": false, "required": ["text", "obligation"], "properties": {
|
|
16
|
+
"text": { "type": "string", "minLength": 1, "maxLength": 4000 },
|
|
17
|
+
"obligation": { "type": "string", "pattern": "^[a-zA-Z0-9][a-zA-Z0-9:._-]{0,79}$" }
|
|
18
|
+
} },
|
|
19
|
+
"family": { "type": "object", "additionalProperties": false, "required": ["id", "version"], "properties": {
|
|
20
|
+
"id": { "type": "string", "pattern": "^[a-zA-Z0-9][a-zA-Z0-9:._-]{0,79}$" },
|
|
21
|
+
"version": { "type": "integer", "minimum": 1, "maximum": 1000000 },
|
|
22
|
+
"digest": { "$ref": "approved-host-contracts.schema.json#/$defs/hash" }
|
|
23
|
+
} },
|
|
24
|
+
"parameters": { "type": "object", "maxProperties": 32 },
|
|
25
|
+
"sourcePath": { "$ref": "approved-host-contracts.schema.json#/$defs/contracts/items/properties/sourcePath" },
|
|
26
|
+
"modulePaths": { "$ref": "approved-host-contracts.schema.json#/$defs/contracts/items/properties/modulePaths" },
|
|
27
|
+
"maxAttempts": { "type": "integer", "minimum": 1, "maximum": 8 }
|
|
28
|
+
}
|
|
29
|
+
} }
|
|
30
|
+
}
|
|
31
|
+
}
|
|
@@ -10,11 +10,11 @@
|
|
|
10
10
|
"parserVersion": { "const": "failure-parser-v1" },
|
|
11
11
|
"exitCode": { "type": "integer", "minimum": 0, "maximum": 255 },
|
|
12
12
|
"signals": { "type": "array", "maxItems": 16, "uniqueItems": true, "items": { "$ref": "#/$defs/reason" } },
|
|
13
|
-
"structuredEvents": { "type": "array", "maxItems": 16, "uniqueItems": true, "items": { "enum": ["scope-violation", "protected-path", "permission-denied", "policy-block", "provider-network", "environment", "flaky-infrastructure", "stale-verifier", "missing-verifier"] } },
|
|
13
|
+
"structuredEvents": { "type": "array", "maxItems": 16, "uniqueItems": true, "items": { "enum": ["scope-violation", "protected-path", "permission-denied", "policy-block", "provider-network", "environment", "flaky-infrastructure", "stale-verifier", "missing-verifier", "verification-gap"] } },
|
|
14
14
|
"outputRef": { "$ref": "#/$defs/outputRef" }
|
|
15
15
|
},
|
|
16
16
|
"$defs": {
|
|
17
|
-
"reason": { "enum": ["exit-zero", "structured-scope", "structured-protected-path", "structured-permission", "structured-policy", "structured-provider", "structured-environment", "structured-flaky-infrastructure", "structured-stale-verifier", "structured-missing-verifier", "typescript-diagnostic", "python-type-diagnostic", "go-compile-diagnostic", "rust-compile-diagnostic", "generic-compile-diagnostic", "test-assertion-diagnostic", "lint-format-diagnostic", "dependency-config-diagnostic", "environment-diagnostic", "provider-rate-limit", "provider-transport", "permission-diagnostic", "policy-diagnostic", "scope-diagnostic", "local-port-conflict", "transient-infrastructure", "unknown-diagnostic"] },
|
|
17
|
+
"reason": { "enum": ["exit-zero", "structured-scope", "structured-protected-path", "structured-permission", "structured-policy", "structured-provider", "structured-environment", "structured-flaky-infrastructure", "structured-stale-verifier", "structured-missing-verifier", "structured-verification-gap", "typescript-diagnostic", "python-type-diagnostic", "go-compile-diagnostic", "rust-compile-diagnostic", "generic-compile-diagnostic", "test-assertion-diagnostic", "lint-format-diagnostic", "dependency-config-diagnostic", "environment-diagnostic", "provider-rate-limit", "provider-transport", "permission-diagnostic", "policy-diagnostic", "scope-diagnostic", "local-port-conflict", "transient-infrastructure", "unknown-diagnostic"] },
|
|
18
18
|
"outputRef": { "type": "object", "additionalProperties": false, "required": ["sha256", "chars", "truncated", "captureRef"], "properties": { "sha256": { "type": "string", "pattern": "^[a-f0-9]{64}$" }, "chars": { "type": "integer", "minimum": 0, "maximum": 1000000000 }, "truncated": { "type": "boolean" }, "captureRef": { "type": ["string", "null"], "pattern": "^[A-Za-z0-9][A-Za-z0-9:._-]{0,255}$" } } }
|
|
19
19
|
}
|
|
20
20
|
}
|
|
@@ -1,11 +1,11 @@
|
|
|
1
1
|
{
|
|
2
2
|
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
3
|
"$id": "https://piagent.io.vn/schemas/helper-request.schema.json",
|
|
4
|
-
"title": "Piagent HelperRequest
|
|
4
|
+
"title": "Piagent HelperRequest v2",
|
|
5
5
|
"type": "object",
|
|
6
|
-
"required": ["schemaVersion", "policyVersion", "role", "objectiveHash", "objectiveText", "taskId", "taskRunId", "sessionHash", "authority", "readScope", "writeScope", "allowedTools", "model", "contextBudget", "ceilings", "outputSchema", "stoppingRule", "approvalRestrictions", "parentAuthorityDigest", "singleWriterOwnership", "deduplicationKey"],
|
|
6
|
+
"required": ["schemaVersion", "policyVersion", "role", "objectiveHash", "objectiveText", "taskId", "taskRunId", "sessionHash", "authority", "readScope", "writeScope", "allowedTools", "model", "contextBudget", "ceilings", "outputSchema", "stoppingRule", "approvalRestrictions", "parentAuthorityDigest", "singleWriterOwnership", "deduplicationKey", "contextTransfer"],
|
|
7
7
|
"properties": {
|
|
8
|
-
"schemaVersion": { "const":
|
|
8
|
+
"schemaVersion": { "const": 2 }, "policyVersion": { "const": "role-policy-v1" },
|
|
9
9
|
"role": { "enum": ["retriever", "scout", "planner", "worker", "reviewer", "oracle", "researcher"] },
|
|
10
10
|
"objectiveHash": { "$ref": "#/$defs/hash" }, "objectiveText": { "type": "string", "minLength": 3, "maxLength": 1000 },
|
|
11
11
|
"taskId": { "$ref": "#/$defs/ref" }, "taskRunId": { "$ref": "#/$defs/ref" }, "sessionHash": { "$ref": "#/$defs/hash" },
|
|
@@ -13,7 +13,19 @@
|
|
|
13
13
|
"model": { "anyOf": [{ "type": "null" }, { "type": "object", "required": ["provider", "modelId", "effort", "source"], "properties": { "provider": { "$ref": "#/$defs/text" }, "modelId": { "$ref": "#/$defs/text" }, "effort": { "$ref": "#/$defs/text" }, "source": { "$ref": "#/$defs/text" } }, "additionalProperties": false }] },
|
|
14
14
|
"contextBudget": { "type": "integer", "minimum": 256, "maximum": 100000 }, "ceilings": { "$ref": "#/$defs/ceilings" },
|
|
15
15
|
"outputSchema": { "type": "string", "minLength": 3, "maxLength": 200 }, "stoppingRule": { "type": "string", "minLength": 3, "maxLength": 500 }, "approvalRestrictions": { "$ref": "#/$defs/names" },
|
|
16
|
-
"parentAuthorityDigest": { "$ref": "#/$defs/hash" }, "singleWriterOwnership": { "type": ["string", "null"], "pattern": "^[a-zA-Z0-9][a-zA-Z0-9:._/-]{0,255}$" }, "deduplicationKey": { "$ref": "#/$defs/hash" }
|
|
16
|
+
"parentAuthorityDigest": { "$ref": "#/$defs/hash" }, "singleWriterOwnership": { "type": ["string", "null"], "pattern": "^[a-zA-Z0-9][a-zA-Z0-9:._/-]{0,255}$" }, "deduplicationKey": { "$ref": "#/$defs/hash" },
|
|
17
|
+
"contextTransfer": {
|
|
18
|
+
"description": "Only a bounded isolated seed is transferred. Runtime validation also recomputes estimatedSeedTokens and requires it not to exceed maxSeedTokens.",
|
|
19
|
+
"type": "object",
|
|
20
|
+
"required": ["mode", "inheritParentHistory", "maxSeedTokens", "estimatedSeedTokens"],
|
|
21
|
+
"properties": {
|
|
22
|
+
"mode": { "const": "isolated-minimal" },
|
|
23
|
+
"inheritParentHistory": { "const": false },
|
|
24
|
+
"maxSeedTokens": { "type": "integer", "minimum": 256, "maximum": 2048 },
|
|
25
|
+
"estimatedSeedTokens": { "type": "integer", "minimum": 0, "maximum": 2048 }
|
|
26
|
+
},
|
|
27
|
+
"additionalProperties": false
|
|
28
|
+
}
|
|
17
29
|
},
|
|
18
30
|
"additionalProperties": false,
|
|
19
31
|
"$defs": {
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "http://json-schema.org/draft-07/schema#",
|
|
3
|
+
"$id": "https://piagent.dev/schemas/host-contract-plan.schema.json",
|
|
4
|
+
"title": "Operator-reviewed independent verification plan",
|
|
5
|
+
"description": "Input to piagent approve-verification. Expected results must be reviewed independently of the candidate; this file is not itself authority or evidence.",
|
|
6
|
+
"oneOf": [
|
|
7
|
+
{"$ref": "approved-host-contracts.schema.json#/$defs/plan"},
|
|
8
|
+
{
|
|
9
|
+
"type": "object",
|
|
10
|
+
"additionalProperties": false,
|
|
11
|
+
"required": ["schemaVersion","plans"],
|
|
12
|
+
"properties": {
|
|
13
|
+
"schemaVersion": {"const": 2},
|
|
14
|
+
"plans": {"type": "array","minItems": 1,"maxItems": 32,"items": {"$ref": "approved-host-contracts.schema.json#/$defs/plan"}}
|
|
15
|
+
}
|
|
16
|
+
},
|
|
17
|
+
{"allOf": [{"$ref": "approved-host-contracts.schema.json#/$defs/planV2"}, {"not": {"required": ["nativeOnly"]}}]},
|
|
18
|
+
{
|
|
19
|
+
"type": "object",
|
|
20
|
+
"additionalProperties": false,
|
|
21
|
+
"required": ["schemaVersion","plans"],
|
|
22
|
+
"properties": {"schemaVersion": {"const": 4},"plans": {"type": "array","minItems": 1,"maxItems": 32,"items": {"$ref": "approved-host-contracts.schema.json#/$defs/planV2"}}}
|
|
23
|
+
}
|
|
24
|
+
]
|
|
25
|
+
}
|
|
@@ -326,6 +326,7 @@
|
|
|
326
326
|
"sessionRevision": { "$ref": "common-v1.schema.json#/$defs/revisionToken" },
|
|
327
327
|
"liveState": { "enum": ["offline", "idle", "running", "paused", "waiting-approval", "uncertain"] },
|
|
328
328
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/nullableOpaqueRef" },
|
|
329
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
329
330
|
"reasonCode": { "$ref": "common-v1.schema.json#/$defs/nullableReasonCode" }
|
|
330
331
|
},
|
|
331
332
|
"allOf": [
|
|
@@ -379,6 +380,7 @@
|
|
|
379
380
|
"properties": {
|
|
380
381
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
381
382
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
383
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
382
384
|
"messageRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
383
385
|
"messageSequence": { "$ref": "#/$defs/stateVersion" },
|
|
384
386
|
"delta": { "type": "string", "maxLength": 16384, "pattern": "^[^\\u0000]*$" }
|
|
@@ -402,6 +404,7 @@
|
|
|
402
404
|
"properties": {
|
|
403
405
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
404
406
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
407
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
405
408
|
"messageRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
406
409
|
"sessionRevision": { "$ref": "common-v1.schema.json#/$defs/revisionToken" },
|
|
407
410
|
"truncated": { "type": "boolean" }
|
|
@@ -414,23 +417,26 @@
|
|
|
414
417
|
"operationCompletedPayload": {
|
|
415
418
|
"type": "object",
|
|
416
419
|
"additionalProperties": false,
|
|
417
|
-
"required": ["sessionRef", "operationRef", "messageRef", "sessionRevision", "settlement", "reasonCode"],
|
|
420
|
+
"required": ["sessionRef", "operationRef", "messageRef", "sessionRevision", "settlement", "taskStatus", "reasonCode"],
|
|
418
421
|
"properties": {
|
|
419
422
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
420
423
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
424
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
421
425
|
"messageRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
422
426
|
"sessionRevision": { "$ref": "common-v1.schema.json#/$defs/revisionToken" },
|
|
423
427
|
"settlement": { "const": "completed" },
|
|
428
|
+
"taskStatus": { "enum": ["pending", "completed", "refused", "failed", "unknown"] },
|
|
424
429
|
"reasonCode": { "type": "null" }
|
|
425
430
|
}
|
|
426
431
|
},
|
|
427
432
|
"operationIncompletePayload": {
|
|
428
433
|
"type": "object",
|
|
429
434
|
"additionalProperties": false,
|
|
430
|
-
"required": ["sessionRef", "operationRef", "messageRef", "sessionRevision", "settlement", "reasonCode"],
|
|
435
|
+
"required": ["sessionRef", "operationRef", "messageRef", "sessionRevision", "settlement", "taskStatus", "reasonCode"],
|
|
431
436
|
"properties": {
|
|
432
437
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
433
438
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
439
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
434
440
|
"messageRef": { "$ref": "common-v1.schema.json#/$defs/nullableOpaqueRef" },
|
|
435
441
|
"sessionRevision": {
|
|
436
442
|
"anyOf": [
|
|
@@ -439,6 +445,7 @@
|
|
|
439
445
|
]
|
|
440
446
|
},
|
|
441
447
|
"settlement": { "enum": ["blocked", "aborted", "error", "unknown"] },
|
|
448
|
+
"taskStatus": { "enum": ["pending", "completed", "refused", "failed", "unknown"] },
|
|
442
449
|
"reasonCode": { "$ref": "common-v1.schema.json#/$defs/reasonCode" }
|
|
443
450
|
}
|
|
444
451
|
},
|
|
@@ -473,8 +480,15 @@
|
|
|
473
480
|
"properties": {
|
|
474
481
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
475
482
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
483
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
476
484
|
"toolCallRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
477
485
|
"toolLabel": { "type": "string", "minLength": 1, "maxLength": 160, "pattern": "^[^\\u0000-\\u001F\\u007F]+$" },
|
|
486
|
+
"fileLabel": {
|
|
487
|
+
"type": ["string", "null"],
|
|
488
|
+
"minLength": 1,
|
|
489
|
+
"maxLength": 160,
|
|
490
|
+
"pattern": "^[^/\\\\\\u0000-\\u001F\\u007F]+$"
|
|
491
|
+
},
|
|
478
492
|
"isError": { "type": ["boolean", "null"] },
|
|
479
493
|
"reasonCode": { "$ref": "common-v1.schema.json#/$defs/nullableReasonCode" }
|
|
480
494
|
}
|
|
@@ -9,7 +9,7 @@
|
|
|
9
9
|
"$defs": {
|
|
10
10
|
"action": {
|
|
11
11
|
"enum": [
|
|
12
|
-
"runtime.status", "runtime.inspector", "runtime.commands", "orchestration.status",
|
|
12
|
+
"runtime.status", "runtime.inspector", "runtime.commands", "runtime.fast-status", "runtime.fast-on", "runtime.fast-off", "orchestration.status",
|
|
13
13
|
"usage.live", "usage.history", "usage.logs", "usage.efficiency", "usage.preflight",
|
|
14
14
|
"onboarding.status", "onboarding.profile", "onboarding.tech",
|
|
15
15
|
"profile.status", "profile.options", "profile.tech-options", "profile.apply", "profile.auto",
|
|
@@ -18,7 +18,29 @@
|
|
|
18
18
|
"mcp.status", "mcp.doctor", "mcp.detail", "mcp.approve", "mcp.reject", "mcp.reset"
|
|
19
19
|
]
|
|
20
20
|
},
|
|
21
|
-
"effect": { "enum": ["read-only", "workspace-write", "model-assisted"] },
|
|
21
|
+
"effect": { "enum": ["read-only", "workspace-write", "model-assisted", "session-setting"] },
|
|
22
|
+
"serviceTierReceipt": {
|
|
23
|
+
"type": "object",
|
|
24
|
+
"additionalProperties": false,
|
|
25
|
+
"required": [
|
|
26
|
+
"schemaVersion", "enabled", "mode", "source", "requestedServiceTier", "observedRequestServiceTier",
|
|
27
|
+
"providerResponseServiceTier", "providerResponseEvidence", "provider", "modelId", "applied", "reasonCode"
|
|
28
|
+
],
|
|
29
|
+
"properties": {
|
|
30
|
+
"schemaVersion": { "const": 1 },
|
|
31
|
+
"enabled": { "type": "boolean" },
|
|
32
|
+
"mode": { "enum": ["standard", "fast"] },
|
|
33
|
+
"source": { "enum": ["default", "environment", "session-command", "session-state"] },
|
|
34
|
+
"requestedServiceTier": { "enum": ["fast", null] },
|
|
35
|
+
"observedRequestServiceTier": { "enum": ["priority", null] },
|
|
36
|
+
"providerResponseServiceTier": { "type": "null" },
|
|
37
|
+
"providerResponseEvidence": { "const": "unavailable-host-api" },
|
|
38
|
+
"provider": { "type": ["string", "null"], "maxLength": 512, "pattern": "^[^\\u0000]*$" },
|
|
39
|
+
"modelId": { "type": ["string", "null"], "maxLength": 512, "pattern": "^[^\\u0000]*$" },
|
|
40
|
+
"applied": { "type": "boolean" },
|
|
41
|
+
"reasonCode": { "enum": ["fast-mode-disabled", "provider-not-supported", "fast-request-pending", "fast-request-observed"] }
|
|
42
|
+
}
|
|
43
|
+
},
|
|
22
44
|
"output": {
|
|
23
45
|
"type": "object",
|
|
24
46
|
"additionalProperties": false,
|
|
@@ -27,7 +49,8 @@
|
|
|
27
49
|
"customType": { "type": "string", "minLength": 1, "maxLength": 120, "pattern": "^[A-Za-z0-9._-]+$" },
|
|
28
50
|
"content": { "type": "string", "maxLength": 12000, "pattern": "^[^\\u0000]*$" },
|
|
29
51
|
"truncated": { "type": "boolean" },
|
|
30
|
-
"redacted": { "type": "boolean" }
|
|
52
|
+
"redacted": { "type": "boolean" },
|
|
53
|
+
"details": { "$ref": "#/$defs/serviceTierReceipt" }
|
|
31
54
|
}
|
|
32
55
|
},
|
|
33
56
|
"command": {
|
|
@@ -2051,6 +2051,25 @@
|
|
|
2051
2051
|
}
|
|
2052
2052
|
]
|
|
2053
2053
|
},
|
|
2054
|
+
"streamActivity": {
|
|
2055
|
+
"type": "object",
|
|
2056
|
+
"additionalProperties": false,
|
|
2057
|
+
"description": "Content-free observations at the Pi message-update hook, not provider usage or network liveness. Only retained at message settlement; absent for legacy events.",
|
|
2058
|
+
"required": ["schemaVersion", "boundary", "observedUpdates", "textUpdates", "thinkingUpdates", "toolCallUpdates", "otherUpdates", "lastUpdateKind", "lastObservedAt", "countersCapped"],
|
|
2059
|
+
"properties": {
|
|
2060
|
+
"schemaVersion": { "const": 1 },
|
|
2061
|
+
"boundary": { "const": "pi-message-update-hook" },
|
|
2062
|
+
"observedUpdates": { "$ref": "#/$defs/streamActivityCount" },
|
|
2063
|
+
"textUpdates": { "$ref": "#/$defs/streamActivityCount" },
|
|
2064
|
+
"thinkingUpdates": { "$ref": "#/$defs/streamActivityCount" },
|
|
2065
|
+
"toolCallUpdates": { "$ref": "#/$defs/streamActivityCount" },
|
|
2066
|
+
"otherUpdates": { "$ref": "#/$defs/streamActivityCount" },
|
|
2067
|
+
"lastUpdateKind": { "enum": [null, "start", "text_start", "text_delta", "text_end", "thinking_start", "thinking_delta", "thinking_end", "toolcall_start", "toolcall_delta", "toolcall_end", "done", "error", "other"] },
|
|
2068
|
+
"lastObservedAt": { "anyOf": [{ "type": "null" }, { "type": "string", "format": "date-time", "maxLength": 40 }] },
|
|
2069
|
+
"countersCapped": { "type": "boolean" }
|
|
2070
|
+
}
|
|
2071
|
+
},
|
|
2072
|
+
"streamActivityCount": { "type": "integer", "minimum": 0, "maximum": 100000 },
|
|
2054
2073
|
"messageCompletedPayload": {
|
|
2055
2074
|
"type": "object",
|
|
2056
2075
|
"additionalProperties": false,
|
|
@@ -2090,6 +2109,9 @@
|
|
|
2090
2109
|
},
|
|
2091
2110
|
"usage": {
|
|
2092
2111
|
"$ref": "#/$defs/nullableTokenUsage"
|
|
2112
|
+
},
|
|
2113
|
+
"streamActivity": {
|
|
2114
|
+
"$ref": "#/$defs/streamActivity"
|
|
2093
2115
|
}
|
|
2094
2116
|
}
|
|
2095
2117
|
},
|
|
@@ -2121,6 +2143,9 @@
|
|
|
2121
2143
|
},
|
|
2122
2144
|
"contentDigest": {
|
|
2123
2145
|
"$ref": "#/$defs/nullableDigest"
|
|
2146
|
+
},
|
|
2147
|
+
"streamActivity": {
|
|
2148
|
+
"$ref": "#/$defs/streamActivity"
|
|
2124
2149
|
}
|
|
2125
2150
|
}
|
|
2126
2151
|
},
|
|
@@ -2973,7 +2998,13 @@
|
|
|
2973
2998
|
{
|
|
2974
2999
|
"properties": {
|
|
2975
3000
|
"isError": { "const": false },
|
|
2976
|
-
"reasonCode": {
|
|
3001
|
+
"reasonCode": {
|
|
3002
|
+
"enum": [
|
|
3003
|
+
"target-not-found",
|
|
3004
|
+
"target-is-directory",
|
|
3005
|
+
"search-target-missing"
|
|
3006
|
+
]
|
|
3007
|
+
}
|
|
2977
3008
|
}
|
|
2978
3009
|
}
|
|
2979
3010
|
]
|
|
@@ -48,6 +48,7 @@
|
|
|
48
48
|
"properties": {
|
|
49
49
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
50
50
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
51
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
51
52
|
"state": { "enum": ["running", "waiting-approval", "settling"] },
|
|
52
53
|
"abortable": { "type": "boolean" }
|
|
53
54
|
},
|
|
@@ -66,6 +67,7 @@
|
|
|
66
67
|
"properties": {
|
|
67
68
|
"sessionRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
68
69
|
"operationRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
70
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
69
71
|
"settlement": { "enum": ["blocked", "aborted", "error", "unknown"] },
|
|
70
72
|
"reasonCode": { "$ref": "common-v1.schema.json#/$defs/reasonCode" },
|
|
71
73
|
"settledAt": { "$ref": "common-v1.schema.json#/$defs/timestamp" },
|
|
@@ -27,6 +27,7 @@
|
|
|
27
27
|
"identity": { "$ref": "common-v1.schema.json#/$defs/identity" },
|
|
28
28
|
"view": { "$ref": "#/$defs/view" },
|
|
29
29
|
"viewRevision": { "$ref": "common-v1.schema.json#/$defs/revisionToken" },
|
|
30
|
+
"snapshotBinding": { "$ref": "#/$defs/snapshotBinding" },
|
|
30
31
|
"bases": {
|
|
31
32
|
"type": "array",
|
|
32
33
|
"minItems": 1,
|
|
@@ -174,6 +175,17 @@
|
|
|
174
175
|
"view": {
|
|
175
176
|
"enum": ["task", "working-tree", "staged"]
|
|
176
177
|
},
|
|
178
|
+
"snapshotBinding": {
|
|
179
|
+
"type": "object",
|
|
180
|
+
"additionalProperties": false,
|
|
181
|
+
"required": ["sourceProjectionRevision", "taskViewRevision", "workspaceRevision", "indexRevision"],
|
|
182
|
+
"properties": {
|
|
183
|
+
"sourceProjectionRevision": { "$ref": "common-v1.schema.json#/$defs/revisionToken" },
|
|
184
|
+
"taskViewRevision": { "$ref": "common-v1.schema.json#/$defs/nullableRevisionToken" },
|
|
185
|
+
"workspaceRevision": { "$ref": "common-v1.schema.json#/$defs/nullableRevisionToken" },
|
|
186
|
+
"indexRevision": { "$ref": "common-v1.schema.json#/$defs/nullableRevisionToken" }
|
|
187
|
+
}
|
|
188
|
+
},
|
|
177
189
|
"availability": {
|
|
178
190
|
"type": "object",
|
|
179
191
|
"additionalProperties": false,
|
|
@@ -183,9 +183,10 @@
|
|
|
183
183
|
"properties": {
|
|
184
184
|
"messageRef": { "$ref": "common-v1.schema.json#/$defs/opaqueRef" },
|
|
185
185
|
"parentMessageRef": { "$ref": "common-v1.schema.json#/$defs/nullableOpaqueRef" },
|
|
186
|
-
"role": { "enum": ["user", "assistant", "tool-result"] },
|
|
186
|
+
"role": { "enum": ["user", "assistant", "tool-result", "custom"] },
|
|
187
187
|
"recordedAt": { "$ref": "common-v1.schema.json#/$defs/timestamp" },
|
|
188
188
|
"agentOperationId": { "$ref": "common-v1.schema.json#/$defs/nullableOpaqueRef" },
|
|
189
|
+
"messageRequestId": { "$ref": "common-v1.schema.json#/$defs/nullableOpaqueRef" },
|
|
189
190
|
"turnIndex": { "$ref": "#/$defs/nullableCount" },
|
|
190
191
|
"content": { "$ref": "#/$defs/content" },
|
|
191
192
|
"attachments": {
|
|
@@ -0,0 +1,206 @@
|
|
|
1
|
+
{
|
|
2
|
+
"$schema": "https://json-schema.org/draft/2020-12/schema",
|
|
3
|
+
"$id": "https://piagent.local/schemas/public-contracts/codex-mcp-metadata-v1.schema.json",
|
|
4
|
+
"title": "PiAgent Codex MCP Metadata Contract v1",
|
|
5
|
+
"description": "Untrusted, zero-authority params._meta accepted by the scoped benchmark broker.",
|
|
6
|
+
"oneOf": [
|
|
7
|
+
{
|
|
8
|
+
"type": "object",
|
|
9
|
+
"additionalProperties": false,
|
|
10
|
+
"properties": {
|
|
11
|
+
"progressToken": {"$ref": "#/$defs/progressToken"}
|
|
12
|
+
}
|
|
13
|
+
},
|
|
14
|
+
{
|
|
15
|
+
"type": "object",
|
|
16
|
+
"additionalProperties": false,
|
|
17
|
+
"required": ["callId", "threadId", "itemId", "x-codex-turn-metadata"],
|
|
18
|
+
"properties": {
|
|
19
|
+
"progressToken": {"$ref": "#/$defs/progressToken"},
|
|
20
|
+
"callId": {"$ref": "#/$defs/id160"},
|
|
21
|
+
"threadId": {"$ref": "#/$defs/id160"},
|
|
22
|
+
"itemId": {"$ref": "#/$defs/id160"},
|
|
23
|
+
"x-codex-turn-metadata": {"$ref": "#/$defs/turnMetadata"},
|
|
24
|
+
"codex/sandbox-state-meta": {"$ref": "#/$defs/sandboxState"}
|
|
25
|
+
}
|
|
26
|
+
}
|
|
27
|
+
],
|
|
28
|
+
"$defs": {
|
|
29
|
+
"id160": {
|
|
30
|
+
"type": "string",
|
|
31
|
+
"minLength": 1,
|
|
32
|
+
"maxLength": 160,
|
|
33
|
+
"pattern": "^[A-Za-z0-9_.:-]+$"
|
|
34
|
+
},
|
|
35
|
+
"progressToken": {
|
|
36
|
+
"oneOf": [
|
|
37
|
+
{"type": "integer", "minimum": -9007199254740991, "maximum": 9007199254740991},
|
|
38
|
+
{"type": "string", "maxLength": 160}
|
|
39
|
+
]
|
|
40
|
+
},
|
|
41
|
+
"turnMetadata": {
|
|
42
|
+
"type": "object",
|
|
43
|
+
"additionalProperties": false,
|
|
44
|
+
"required": [
|
|
45
|
+
"session_id",
|
|
46
|
+
"thread_id",
|
|
47
|
+
"turn_started_at_unix_ms",
|
|
48
|
+
"turn_id",
|
|
49
|
+
"node_repl_disabled",
|
|
50
|
+
"thread_source",
|
|
51
|
+
"sandbox",
|
|
52
|
+
"sandbox_mode",
|
|
53
|
+
"auto_review_enabled",
|
|
54
|
+
"node_repl_auto_review_required",
|
|
55
|
+
"model"
|
|
56
|
+
],
|
|
57
|
+
"properties": {
|
|
58
|
+
"session_id": {"$ref": "#/$defs/id160"},
|
|
59
|
+
"thread_id": {"$ref": "#/$defs/id160"},
|
|
60
|
+
"turn_started_at_unix_ms": {"type": "integer", "minimum": 0, "maximum": 9007199254740991},
|
|
61
|
+
"turn_id": {"$ref": "#/$defs/id160"},
|
|
62
|
+
"node_repl_disabled": {"type": "boolean"},
|
|
63
|
+
"thread_source": {"const": "user"},
|
|
64
|
+
"sandbox": {"type": "string", "minLength": 1, "maxLength": 80},
|
|
65
|
+
"sandbox_mode": {"const": "workspace-write"},
|
|
66
|
+
"auto_review_enabled": {"type": "boolean"},
|
|
67
|
+
"node_repl_auto_review_required": {"type": "boolean"},
|
|
68
|
+
"model": {"$ref": "#/$defs/id160"},
|
|
69
|
+
"reasoning_effort": {
|
|
70
|
+
"enum": ["none", "minimal", "low", "medium", "high", "xhigh", "max", "ultra"]
|
|
71
|
+
}
|
|
72
|
+
}
|
|
73
|
+
},
|
|
74
|
+
"sandboxState": {
|
|
75
|
+
"type": "object",
|
|
76
|
+
"additionalProperties": false,
|
|
77
|
+
"required": ["permissionProfile", "codexLinuxSandboxExe", "sandboxCwd", "useLegacyLandlock"],
|
|
78
|
+
"properties": {
|
|
79
|
+
"permissionProfile": {"$ref": "#/$defs/permissionProfile"},
|
|
80
|
+
"codexLinuxSandboxExe": {
|
|
81
|
+
"oneOf": [
|
|
82
|
+
{"type": "null"},
|
|
83
|
+
{"type": "string", "minLength": 1, "maxLength": 2048, "pattern": "^/"}
|
|
84
|
+
]
|
|
85
|
+
},
|
|
86
|
+
"sandboxCwd": {"type": "string", "minLength": 6, "maxLength": 2048, "pattern": "^file:"},
|
|
87
|
+
"useLegacyLandlock": {"type": "boolean"}
|
|
88
|
+
}
|
|
89
|
+
},
|
|
90
|
+
"permissionProfile": {
|
|
91
|
+
"oneOf": [
|
|
92
|
+
{
|
|
93
|
+
"type": "object",
|
|
94
|
+
"additionalProperties": false,
|
|
95
|
+
"required": ["type"],
|
|
96
|
+
"properties": {"type": {"const": "disabled"}}
|
|
97
|
+
},
|
|
98
|
+
{
|
|
99
|
+
"type": "object",
|
|
100
|
+
"additionalProperties": false,
|
|
101
|
+
"required": ["type", "network"],
|
|
102
|
+
"properties": {
|
|
103
|
+
"type": {"const": "external"},
|
|
104
|
+
"network": {"enum": ["restricted", "enabled"]}
|
|
105
|
+
}
|
|
106
|
+
},
|
|
107
|
+
{
|
|
108
|
+
"type": "object",
|
|
109
|
+
"additionalProperties": false,
|
|
110
|
+
"required": ["type", "file_system", "network"],
|
|
111
|
+
"properties": {
|
|
112
|
+
"type": {"const": "managed"},
|
|
113
|
+
"file_system": {"$ref": "#/$defs/fileSystemPermissions"},
|
|
114
|
+
"network": {"enum": ["restricted", "enabled"]}
|
|
115
|
+
}
|
|
116
|
+
}
|
|
117
|
+
]
|
|
118
|
+
},
|
|
119
|
+
"fileSystemPermissions": {
|
|
120
|
+
"oneOf": [
|
|
121
|
+
{
|
|
122
|
+
"type": "object",
|
|
123
|
+
"additionalProperties": false,
|
|
124
|
+
"required": ["type"],
|
|
125
|
+
"properties": {"type": {"const": "unrestricted"}}
|
|
126
|
+
},
|
|
127
|
+
{
|
|
128
|
+
"type": "object",
|
|
129
|
+
"additionalProperties": false,
|
|
130
|
+
"required": ["type", "entries"],
|
|
131
|
+
"properties": {
|
|
132
|
+
"type": {"const": "restricted"},
|
|
133
|
+
"entries": {
|
|
134
|
+
"type": "array",
|
|
135
|
+
"maxItems": 64,
|
|
136
|
+
"items": {"$ref": "#/$defs/fileSystemEntry"}
|
|
137
|
+
},
|
|
138
|
+
"glob_scan_max_depth": {"type": "integer", "minimum": 1, "maximum": 64}
|
|
139
|
+
}
|
|
140
|
+
}
|
|
141
|
+
]
|
|
142
|
+
},
|
|
143
|
+
"fileSystemEntry": {
|
|
144
|
+
"type": "object",
|
|
145
|
+
"additionalProperties": false,
|
|
146
|
+
"required": ["path", "access"],
|
|
147
|
+
"properties": {
|
|
148
|
+
"path": {"$ref": "#/$defs/fileSystemPath"},
|
|
149
|
+
"access": {"enum": ["read", "write", "deny"]},
|
|
150
|
+
"missing_path_behavior": {"const": "skip"}
|
|
151
|
+
}
|
|
152
|
+
},
|
|
153
|
+
"fileSystemPath": {
|
|
154
|
+
"oneOf": [
|
|
155
|
+
{
|
|
156
|
+
"type": "object",
|
|
157
|
+
"additionalProperties": false,
|
|
158
|
+
"required": ["type", "path"],
|
|
159
|
+
"properties": {
|
|
160
|
+
"type": {"const": "path"},
|
|
161
|
+
"path": {"type": "string", "minLength": 1, "maxLength": 2048}
|
|
162
|
+
}
|
|
163
|
+
},
|
|
164
|
+
{
|
|
165
|
+
"type": "object",
|
|
166
|
+
"additionalProperties": false,
|
|
167
|
+
"required": ["type", "pattern"],
|
|
168
|
+
"properties": {
|
|
169
|
+
"type": {"const": "glob_pattern"},
|
|
170
|
+
"pattern": {"type": "string", "minLength": 1, "maxLength": 2048}
|
|
171
|
+
}
|
|
172
|
+
},
|
|
173
|
+
{
|
|
174
|
+
"type": "object",
|
|
175
|
+
"additionalProperties": false,
|
|
176
|
+
"required": ["type", "value"],
|
|
177
|
+
"properties": {
|
|
178
|
+
"type": {"const": "special"},
|
|
179
|
+
"value": {"$ref": "#/$defs/specialPath"}
|
|
180
|
+
}
|
|
181
|
+
}
|
|
182
|
+
]
|
|
183
|
+
},
|
|
184
|
+
"specialPath": {
|
|
185
|
+
"oneOf": [
|
|
186
|
+
{
|
|
187
|
+
"type": "object",
|
|
188
|
+
"additionalProperties": false,
|
|
189
|
+
"required": ["kind"],
|
|
190
|
+
"properties": {
|
|
191
|
+
"kind": {"enum": ["root", "minimal", "tmpdir", "slash_tmp"]}
|
|
192
|
+
}
|
|
193
|
+
},
|
|
194
|
+
{
|
|
195
|
+
"type": "object",
|
|
196
|
+
"additionalProperties": false,
|
|
197
|
+
"required": ["kind"],
|
|
198
|
+
"properties": {
|
|
199
|
+
"kind": {"const": "project_roots"},
|
|
200
|
+
"subpath": {"type": "string", "minLength": 1, "maxLength": 1024}
|
|
201
|
+
}
|
|
202
|
+
}
|
|
203
|
+
]
|
|
204
|
+
}
|
|
205
|
+
}
|
|
206
|
+
}
|