@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
|
@@ -13,13 +13,13 @@ const MAX_BOOTSTRAP_BODY_BYTES = 4_096;
|
|
|
13
13
|
const MAX_CONTROL_BODY_BYTES = 70_000;
|
|
14
14
|
const MAX_ATTACHMENT_BODY_BYTES = 11_250_000;
|
|
15
15
|
const REQUESTS_PER_MINUTE = 120;
|
|
16
|
-
const
|
|
16
|
+
const BOOTSTRAP_FAILURES_PER_MINUTE = 8;
|
|
17
17
|
const CONTROLS_PER_MINUTE = 60;
|
|
18
18
|
const MAX_JSON_RESPONSE_BYTES = 16 * 1024 * 1024;
|
|
19
19
|
const CURSOR = /^[A-Za-z0-9][A-Za-z0-9._~-]{0,159}$/;
|
|
20
20
|
|
|
21
21
|
type ReadCapabilities = () => unknown | Promise<unknown>;
|
|
22
|
-
type RateState = { windowStart: number; requests: number;
|
|
22
|
+
type RateState = { windowStart: number; requests: number; bootstrapFailures: number };
|
|
23
23
|
type ControlRateState = { windowStart: number; controls: number };
|
|
24
24
|
|
|
25
25
|
export type LoopbackServer = {
|
|
@@ -109,8 +109,8 @@ export async function startLoopbackServer(options: {
|
|
|
109
109
|
const browserSession = auth.authenticate(request, now);
|
|
110
110
|
const rateKey = browserSession ? `${remote}:${browserSession.id}` : `remote:${remote}`;
|
|
111
111
|
for (const [key, value] of rates) if (now - value.windowStart >= 60_000) rates.delete(key);
|
|
112
|
-
const rate = rates.get(rateKey) ?? { windowStart: now, requests: 0,
|
|
113
|
-
if (now - rate.windowStart >= 60_000) Object.assign(rate, { windowStart: now, requests: 0,
|
|
112
|
+
const rate = rates.get(rateKey) ?? { windowStart: now, requests: 0, bootstrapFailures: 0 };
|
|
113
|
+
if (now - rate.windowStart >= 60_000) Object.assign(rate, { windowStart: now, requests: 0, bootstrapFailures: 0 });
|
|
114
114
|
rate.requests += 1; rates.set(rateKey, rate);
|
|
115
115
|
while (rates.size > 128) rates.delete(rates.keys().next().value as string);
|
|
116
116
|
if (rate.requests > REQUESTS_PER_MINUTE) return errorResponse(response, 429, "rate-limit");
|
|
@@ -120,8 +120,6 @@ export async function startLoopbackServer(options: {
|
|
|
120
120
|
if (url.hash || url.username || url.password || url.origin !== origin) return errorResponse(response, 400, "invalid-url");
|
|
121
121
|
|
|
122
122
|
if (request.method === "POST" && url.pathname === "/api/v1/bootstrap") {
|
|
123
|
-
rate.bootstraps += 1;
|
|
124
|
-
if (rate.bootstraps > BOOTSTRAPS_PER_MINUTE) return errorResponse(response, 429, "bootstrap-rate-limit");
|
|
125
123
|
if (requestOrigin !== origin) return errorResponse(response, 403, "origin-required");
|
|
126
124
|
if (!String(request.headers["content-type"] ?? "").toLowerCase().startsWith("application/json")) return errorResponse(response, 415, "content-type");
|
|
127
125
|
let value: unknown;
|
|
@@ -129,7 +127,11 @@ export async function startLoopbackServer(options: {
|
|
|
129
127
|
catch (error) { return errorResponse(response, (error as Error).message === "body-limit" ? 413 : 400, "invalid-bootstrap"); }
|
|
130
128
|
const capability = value && typeof value === "object" ? (value as Record<string, unknown>).capability : null;
|
|
131
129
|
const session = typeof capability === "string" ? auth.exchange(capability) : null;
|
|
132
|
-
if (!session)
|
|
130
|
+
if (!session) {
|
|
131
|
+
rate.bootstrapFailures += 1;
|
|
132
|
+
if (rate.bootstrapFailures > BOOTSTRAP_FAILURES_PER_MINUTE) return errorResponse(response, 429, "bootstrap-rate-limit");
|
|
133
|
+
return errorResponse(response, 403, "bootstrap-rejected");
|
|
134
|
+
}
|
|
133
135
|
response.setHeader("Set-Cookie", auth.cookieHeader(session));
|
|
134
136
|
return jsonResponse(response, 200, { authenticated: true, csrfToken: session.csrf, expiresAt: new Date(session.expiresAt).toISOString() });
|
|
135
137
|
}
|
|
@@ -0,0 +1,144 @@
|
|
|
1
|
+
import { isDeepStrictEqual } from "node:util";
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
|
|
4
|
+
import { webUiMessageCorrelationEntry, type WebUiMessageCorrelation } from "../shared/message-correlation.ts";
|
|
5
|
+
|
|
6
|
+
// As in CoreInspectionProvider, load core through the server boundary. Its
|
|
7
|
+
// internal source types are not part of the strict browser contract graph.
|
|
8
|
+
const CORE_ROOT = "../../piagent-core";
|
|
9
|
+
const [{ activeSessionTask }, { isUncertainSendContinuation, terminalUncertainSendReceipt }] = await Promise.all([
|
|
10
|
+
import(`${CORE_ROOT}/extensions/task-state.js`),
|
|
11
|
+
import(`${CORE_ROOT}/runtime/session/uncertain-send-continuation.ts`)
|
|
12
|
+
]);
|
|
13
|
+
|
|
14
|
+
export type TerminalDeliveryReceipt = { customType: string; content: string; details: {
|
|
15
|
+
taskId: string; taskRunId: string; outcome: string; completionApproved: boolean; gateDecision: string;
|
|
16
|
+
[key: string]: unknown;
|
|
17
|
+
} };
|
|
18
|
+
|
|
19
|
+
export function readTerminalDeliveryReceipt(cwd: string, task: unknown): TerminalDeliveryReceipt | undefined {
|
|
20
|
+
return terminalUncertainSendReceipt(cwd, task);
|
|
21
|
+
}
|
|
22
|
+
|
|
23
|
+
export const TERMINAL_DELIVERY_CONFIRMATION_TYPE = "piagent-webui-terminal-delivery-v1";
|
|
24
|
+
export type TerminalDeliveryPair = { requestEntry: any; receiptEntry: any; confirmationEntry?: any; request: string;
|
|
25
|
+
correlation: WebUiMessageCorrelation; receipt: TerminalDeliveryReceipt };
|
|
26
|
+
|
|
27
|
+
/** Strict bounded readback; an in-memory Pi branch is not a persistence acknowledgment. */
|
|
28
|
+
export function terminalDeliverySessionEntries(file: string | undefined, sessionId: string): unknown[] {
|
|
29
|
+
if (!file) return [];
|
|
30
|
+
let descriptor: number | undefined;
|
|
31
|
+
try {
|
|
32
|
+
descriptor = fs.openSync(file, "r");
|
|
33
|
+
const stat = fs.fstatSync(descriptor);
|
|
34
|
+
if (!stat.isFile() || stat.size > 64 * 1024 * 1024) return [];
|
|
35
|
+
const source = fs.readFileSync(descriptor, "utf8");
|
|
36
|
+
if (!source.endsWith("\n") || Buffer.byteLength(source) > 64 * 1024 * 1024) return [];
|
|
37
|
+
const lines = source.trimEnd().split("\n");
|
|
38
|
+
if (lines.length > 50_001) return [];
|
|
39
|
+
// Pi's generic parser tolerates corrupt lines. Delivery requires every
|
|
40
|
+
// persisted record, including the confirmation, to be complete and valid.
|
|
41
|
+
const [header, ...entries] = lines.map((line) => JSON.parse(line));
|
|
42
|
+
return header?.type === "session" && header.version === 3 && header.id === sessionId ? entries : [];
|
|
43
|
+
} catch { return []; }
|
|
44
|
+
finally { if (descriptor !== undefined) fs.closeSync(descriptor); }
|
|
45
|
+
}
|
|
46
|
+
|
|
47
|
+
/** Only this handled-input protocol needs a request anchor outside model history. */
|
|
48
|
+
export function terminalDeliveryRequest(message: unknown, images?: unknown[]): string | undefined {
|
|
49
|
+
return typeof message === "string" && message.length <= 65_536 && !images?.length
|
|
50
|
+
&& isUncertainSendContinuation(message) ? message : undefined;
|
|
51
|
+
}
|
|
52
|
+
|
|
53
|
+
function matchesReceipt(value: any, expected: TerminalDeliveryReceipt): boolean {
|
|
54
|
+
return value?.display === true && value.customType === expected.customType && value.content === expected.content
|
|
55
|
+
&& isDeepStrictEqual(value.details, expected.details);
|
|
56
|
+
}
|
|
57
|
+
|
|
58
|
+
function hasEntryIdentity(value: any): boolean {
|
|
59
|
+
return typeof value?.id === "string" && value.id.length > 0 && value.id.length <= 160
|
|
60
|
+
&& typeof value.timestamp === "string" && Number.isFinite(Date.parse(value.timestamp));
|
|
61
|
+
}
|
|
62
|
+
|
|
63
|
+
function confirmationData(pair: TerminalDeliveryPair): Record<string, unknown> {
|
|
64
|
+
return { schemaVersion: 1, ...pair.correlation, receiptEntryId: pair.receiptEntry.id,
|
|
65
|
+
taskId: pair.receipt.details.taskId, taskRunId: pair.receipt.details.taskRunId };
|
|
66
|
+
}
|
|
67
|
+
|
|
68
|
+
function receiptPairs(entries: unknown[], expected: TerminalDeliveryReceipt | undefined,
|
|
69
|
+
confirmed: boolean): TerminalDeliveryPair[] {
|
|
70
|
+
if (!expected || !Array.isArray(entries) || entries.length > 50_000) return [];
|
|
71
|
+
const pairs: TerminalDeliveryPair[] = [];
|
|
72
|
+
const operations = new Map<string, number>(), requests = new Map<string, number>(), entryIds = new Map<string, number>();
|
|
73
|
+
for (const entry of entries) {
|
|
74
|
+
const id = (entry as any)?.id;
|
|
75
|
+
if (typeof id === "string") entryIds.set(id, (entryIds.get(id) ?? 0) + 1);
|
|
76
|
+
const correlation = webUiMessageCorrelationEntry(entry);
|
|
77
|
+
if (correlation) {
|
|
78
|
+
operations.set(correlation.operationRef, (operations.get(correlation.operationRef) ?? 0) + 1);
|
|
79
|
+
requests.set(correlation.messageRequestId, (requests.get(correlation.messageRequestId) ?? 0) + 1);
|
|
80
|
+
}
|
|
81
|
+
}
|
|
82
|
+
for (let index = 1; index < entries.length; index += 1) {
|
|
83
|
+
const requestEntry = entries[index - 1] as any, receiptEntry = entries[index] as any;
|
|
84
|
+
const correlation = webUiMessageCorrelationEntry(requestEntry);
|
|
85
|
+
const request = terminalDeliveryRequest(requestEntry?.data?.terminalDeliveryRequest);
|
|
86
|
+
if (!correlation || request === undefined || receiptEntry?.type !== "custom_message"
|
|
87
|
+
|| !hasEntryIdentity(requestEntry) || !hasEntryIdentity(receiptEntry)
|
|
88
|
+
|| entryIds.get(requestEntry.id) !== 1 || entryIds.get(receiptEntry.id) !== 1
|
|
89
|
+
|| !matchesReceipt(receiptEntry, expected)) continue;
|
|
90
|
+
const pair: TerminalDeliveryPair = { requestEntry, receiptEntry, request, correlation, receipt: expected };
|
|
91
|
+
if (confirmed) {
|
|
92
|
+
const confirmation = entries[index + 1] as any;
|
|
93
|
+
if (confirmation?.type !== "custom" || confirmation.customType !== TERMINAL_DELIVERY_CONFIRMATION_TYPE
|
|
94
|
+
|| !hasEntryIdentity(confirmation) || entryIds.get(confirmation.id) !== 1
|
|
95
|
+
|| !isDeepStrictEqual(confirmation.data, confirmationData(pair))) continue;
|
|
96
|
+
pair.confirmationEntry = confirmation;
|
|
97
|
+
}
|
|
98
|
+
pairs.push(pair);
|
|
99
|
+
}
|
|
100
|
+
return pairs.filter((pair) => operations.get(pair.correlation.operationRef) === 1
|
|
101
|
+
&& requests.get(pair.correlation.messageRequestId) === 1);
|
|
102
|
+
}
|
|
103
|
+
|
|
104
|
+
/** A projection may show only a receipt whose operation reached durable delivery settlement. */
|
|
105
|
+
export function terminalDeliveryPairs(entries: unknown[], expected: TerminalDeliveryReceipt | undefined,
|
|
106
|
+
persistedEntries: unknown[] = []): TerminalDeliveryPair[] {
|
|
107
|
+
const persisted = new Map(receiptPairs(persistedEntries, expected, true).map((pair) => [pair.receiptEntry.id, pair]));
|
|
108
|
+
return receiptPairs(entries, expected, true).filter((pair) => {
|
|
109
|
+
const disk = persisted.get(pair.receiptEntry.id);
|
|
110
|
+
return disk && isDeepStrictEqual(pair.requestEntry, disk.requestEntry) && isDeepStrictEqual(pair.receiptEntry, disk.receiptEntry)
|
|
111
|
+
&& isDeepStrictEqual(pair.confirmationEntry, disk.confirmationEntry);
|
|
112
|
+
});
|
|
113
|
+
}
|
|
114
|
+
|
|
115
|
+
/**
|
|
116
|
+
* Compare the observed custom message with current core state and this exact
|
|
117
|
+
* operation's latest durable receipt. Persist a delivery observation before
|
|
118
|
+
* the gateway acknowledges settlement. This marker grants no task authority;
|
|
119
|
+
* transcript reads still require the independently reread core receipt.
|
|
120
|
+
*/
|
|
121
|
+
export function terminalDeliveryReceiptCommitter(input: { cwd: string; sessionId: string; operationRef: string;
|
|
122
|
+
messageRequestId: string | null; request?: string; entries(): unknown[]; persistedEntries(): unknown[];
|
|
123
|
+
appendCustomEntry(type: string, data: Record<string, unknown>): unknown }):
|
|
124
|
+
(message: unknown) => TerminalDeliveryReceipt | undefined {
|
|
125
|
+
return (message) => {
|
|
126
|
+
if (input.request === undefined || !input.messageRequestId || (message as any)?.role !== "custom") return undefined;
|
|
127
|
+
try {
|
|
128
|
+
const expected = readTerminalDeliveryReceipt(input.cwd, activeSessionTask(input.cwd, input.sessionId));
|
|
129
|
+
const entries = input.entries(), pair = receiptPairs(entries, expected, false).find((candidate) =>
|
|
130
|
+
candidate.correlation.operationRef === input.operationRef
|
|
131
|
+
&& candidate.correlation.messageRequestId === input.messageRequestId && candidate.request === input.request);
|
|
132
|
+
if (!pair || pair.receiptEntry !== entries.at(-1) || !matchesReceipt(message, pair.receipt)) return undefined;
|
|
133
|
+
const before = input.persistedEntries();
|
|
134
|
+
const disk = receiptPairs(before, expected, false).find((candidate) => candidate.receiptEntry.id === pair.receiptEntry.id);
|
|
135
|
+
if (!disk || disk.receiptEntry !== before.at(-1) || !isDeepStrictEqual(pair.requestEntry, disk.requestEntry)
|
|
136
|
+
|| !isDeepStrictEqual(pair.receiptEntry, disk.receiptEntry)) return undefined;
|
|
137
|
+
input.appendCustomEntry(TERMINAL_DELIVERY_CONFIRMATION_TYPE, confirmationData(pair));
|
|
138
|
+
const current = input.entries(), persisted = input.persistedEntries();
|
|
139
|
+
return terminalDeliveryPairs(current, expected, persisted).some((candidate) => candidate.receiptEntry.id === pair.receiptEntry.id
|
|
140
|
+
&& candidate.confirmationEntry === current.at(-1) && candidate.confirmationEntry.id === (persisted.at(-1) as any)?.id)
|
|
141
|
+
? pair.receipt : undefined;
|
|
142
|
+
} catch { return undefined; }
|
|
143
|
+
};
|
|
144
|
+
}
|
|
@@ -1,13 +1,22 @@
|
|
|
1
1
|
import { createHash } from "node:crypto";
|
|
2
2
|
|
|
3
3
|
import { redactSensitiveText } from "../../piagent-core/extensions/redaction-core.js";
|
|
4
|
+
import { looksLikeCompletionClaim } from "../../piagent-core/runtime/session/completion-signals.ts";
|
|
5
|
+
import { isLightweightNonAuthorizingChangeLanguage } from "../../piagent-core/runtime/workflows/change-clarification.ts";
|
|
6
|
+
import { workflowCommandPattern } from "../../piagent-core/runtime/workflows/webui-workflow.ts";
|
|
4
7
|
import { hasVisibleText } from "../shared/text-visibility.ts";
|
|
8
|
+
import { WEBUI_MESSAGE_CORRELATION_ENTRY_TYPE, webUiMessageCorrelationEntry,
|
|
9
|
+
type WebUiMessageCorrelation } from "../shared/message-correlation.ts";
|
|
10
|
+
import { terminalDeliveryPairs, type TerminalDeliveryPair, type TerminalDeliveryReceipt } from "./terminal-delivery-receipt.ts";
|
|
5
11
|
|
|
6
12
|
const MAX_ENTRIES = 50_000;
|
|
7
13
|
const MAX_ITEMS = 200;
|
|
8
14
|
const MAX_TEXT = 16_384;
|
|
9
15
|
const ANSI = /\u001b(?:\[[0-?]*[ -/]*[@-~]|[@-_])/g;
|
|
10
16
|
const COMPLETION_GATE = /^\[Piagent completion gate: (CONTINUING|NOT APPROVED)\]/i;
|
|
17
|
+
const INTERNAL_FRESH_TRANSITION = new RegExp(
|
|
18
|
+
`^\\/fresh\\s+(?:${workflowCommandPattern({ aliases: false })})\\s+(?:--session-title\\s+"[^"\\r\\n]{1,64}"\\s+)?Read task intake from \\.pi\\/task-inbox\\/[A-Za-z0-9._-]+\\.md\\.\\s+Current session is near context limits; use a fresh governed session\\.$`, "u"
|
|
19
|
+
);
|
|
11
20
|
|
|
12
21
|
type TranscriptIdentity = { projectRef: string; runtimeInstanceId: string; sessionRef: string; taskId: string | null; taskRunId: string | null;
|
|
13
22
|
agentOperationId: null; toolCallId: null };
|
|
@@ -16,8 +25,8 @@ type TranscriptRevision = { runtimeRevision: string; taskRevision: string | null
|
|
|
16
25
|
type TranscriptContent = { state: "available" | "redacted" | "unavailable"; text: string | null; textChars: number | null; digest: string | null;
|
|
17
26
|
truncated: boolean; redacted: boolean; imageCount: number; reasonCode: string | null };
|
|
18
27
|
type TranscriptAttachment = { displayName: string; kind: "file" | "image" | "document"; mimeType: string; truncated: boolean };
|
|
19
|
-
type TranscriptItem = { messageRef: string; parentMessageRef: string | null; role: "user" | "assistant" | "tool-result"; recordedAt: string;
|
|
20
|
-
agentOperationId: string | null; turnIndex: number | null; content: TranscriptContent;
|
|
28
|
+
type TranscriptItem = { messageRef: string; parentMessageRef: string | null; role: "user" | "assistant" | "tool-result" | "custom"; recordedAt: string;
|
|
29
|
+
agentOperationId: string | null; messageRequestId?: string; turnIndex: number | null; content: TranscriptContent;
|
|
21
30
|
attachments?: TranscriptAttachment[];
|
|
22
31
|
toolCalls: Array<{ toolCallRef: string; toolName: string; state: "requested" | "completed" | "failed" | "unknown" }> };
|
|
23
32
|
export type TranscriptDocument = { schemaVersion: 1; version: "piagent-webui-transcript-v1"; generatedAt: string; identity: TranscriptIdentity;
|
|
@@ -33,6 +42,8 @@ export type TranscriptProjectionInput = {
|
|
|
33
42
|
limit?: number;
|
|
34
43
|
generatedAt?: string;
|
|
35
44
|
taskOutcome?: string | null;
|
|
45
|
+
terminalDeliveryReceipt?: TerminalDeliveryReceipt;
|
|
46
|
+
terminalDeliveryEntries?: unknown[];
|
|
36
47
|
};
|
|
37
48
|
|
|
38
49
|
function hash(value: string): string { return createHash("sha256").update(value).digest("hex"); }
|
|
@@ -135,16 +146,15 @@ function role(message: any): TranscriptItem["role"] | null {
|
|
|
135
146
|
}
|
|
136
147
|
function isInternalFreshTransition(message: any): boolean {
|
|
137
148
|
if (message?.role !== "user") return false;
|
|
138
|
-
return
|
|
139
|
-
.test(messageText(message).trim());
|
|
149
|
+
return INTERNAL_FRESH_TRANSITION.test(messageText(message).trim());
|
|
140
150
|
}
|
|
141
151
|
function unavailableContent(reasonCode: string): TranscriptContent {
|
|
142
152
|
return { state: "unavailable", text: null, textChars: null, digest: null, truncated: false, redacted: false, imageCount: 0, reasonCode };
|
|
143
153
|
}
|
|
144
|
-
function item(entry: any, identity: TranscriptIdentity): TranscriptItem | null {
|
|
154
|
+
function item(entry: any, identity: TranscriptIdentity, customReceipt = false): TranscriptItem | null {
|
|
145
155
|
if (!entry || entry.type !== "message" || !entry.message) return null;
|
|
146
156
|
if (isInternalFreshTransition(entry.message)) return null;
|
|
147
|
-
const itemRole = role(entry.message), recordedAt = timestamp(entry.timestamp ?? entry.message.timestamp);
|
|
157
|
+
const itemRole = customReceipt ? "custom" : role(entry.message), recordedAt = timestamp(entry.timestamp ?? entry.message.timestamp);
|
|
148
158
|
if (!itemRole || !recordedAt || typeof entry.id !== "string" || entry.id.length === 0) return null;
|
|
149
159
|
const userProjection = userMessageProjection(entry.message);
|
|
150
160
|
const projectedToolCalls = toolCalls(entry.message, identity.sessionRef, itemRole);
|
|
@@ -163,6 +173,41 @@ function item(entry: any, identity: TranscriptIdentity): TranscriptItem | null {
|
|
|
163
173
|
|
|
164
174
|
type ProjectedTranscriptItem = { entry: any; item: TranscriptItem; cursor: string };
|
|
165
175
|
|
|
176
|
+
function correlatedTranscriptItems(entries: unknown[], identity: TranscriptIdentity,
|
|
177
|
+
receiptPairs: Map<unknown, TerminalDeliveryPair>): ProjectedTranscriptItem[] {
|
|
178
|
+
let pending: WebUiMessageCorrelation | null = null;
|
|
179
|
+
let turn: WebUiMessageCorrelation | null = null;
|
|
180
|
+
const values: ProjectedTranscriptItem[] = [];
|
|
181
|
+
for (const entry of entries as any[]) {
|
|
182
|
+
const pair = receiptPairs.get(entry);
|
|
183
|
+
if (pair) {
|
|
184
|
+
// Project the admitted request and real custom receipt without writing
|
|
185
|
+
// fabricated user/assistant messages into Pi's durable model history.
|
|
186
|
+
for (const [source, itemRole, content] of [[pair.requestEntry, "user", pair.request],
|
|
187
|
+
[pair.receiptEntry, "custom", pair.receipt.content]] as const) {
|
|
188
|
+
const projected = item({ ...source, type: "message", message: { role: itemRole, content } }, identity, itemRole === "custom");
|
|
189
|
+
if (projected) values.push({ entry: source, item: { ...projected, agentOperationId: pair.correlation.operationRef,
|
|
190
|
+
messageRequestId: pair.correlation.messageRequestId }, cursor: opaque("transcript", [identity.sessionRef, source.id]) });
|
|
191
|
+
}
|
|
192
|
+
pending = null; turn = null; continue;
|
|
193
|
+
}
|
|
194
|
+
if (entry?.type === "custom" && entry.customType === WEBUI_MESSAGE_CORRELATION_ENTRY_TYPE) {
|
|
195
|
+
// A newer marker supersedes an older marker that never reached a user
|
|
196
|
+
// message. Invalid markers fail closed instead of leaking correlation
|
|
197
|
+
// from an earlier admitted-but-undispatched operation into a later turn.
|
|
198
|
+
pending = webUiMessageCorrelationEntry(entry);
|
|
199
|
+
continue;
|
|
200
|
+
}
|
|
201
|
+
if (role(entry?.message) === "user") { turn = pending; pending = null; }
|
|
202
|
+
const projected = item(entry, identity);
|
|
203
|
+
if (!projected) continue;
|
|
204
|
+
values.push({ entry, item: turn ? { ...projected, agentOperationId: turn.operationRef,
|
|
205
|
+
messageRequestId: turn.messageRequestId } : projected,
|
|
206
|
+
cursor: opaque("transcript", [identity.sessionRef, entry?.id]) });
|
|
207
|
+
}
|
|
208
|
+
return values;
|
|
209
|
+
}
|
|
210
|
+
|
|
166
211
|
function linkTranscriptTurns(values: ProjectedTranscriptItem[]): ProjectedTranscriptItem[] {
|
|
167
212
|
let userMessageRef: string | null = null;
|
|
168
213
|
const toolCallOwners = new Map<string, string>();
|
|
@@ -187,15 +232,18 @@ function suppressOpenTaskHandoff(values: ProjectedTranscriptItem[], taskOutcome:
|
|
|
187
232
|
if (values[index]?.item.role === "user") { latestUser = index; break; }
|
|
188
233
|
}
|
|
189
234
|
if (latestUser < 0) return values;
|
|
235
|
+
const latestPrompt = values[latestUser]?.item.content.text ?? "";
|
|
236
|
+
if (isLightweightNonAuthorizingChangeLanguage(latestPrompt)) return values;
|
|
190
237
|
return values.map((value, index) => {
|
|
191
238
|
if (index <= latestUser || value.item.role !== "assistant" || value.item.toolCalls.length > 0
|
|
192
|
-
|| !["available", "redacted"].includes(value.item.content.state)
|
|
239
|
+
|| !["available", "redacted"].includes(value.item.content.state)
|
|
240
|
+
|| !looksLikeCompletionClaim(value.item.content.text ?? "")) return value;
|
|
193
241
|
return { ...value, item: { ...value.item, content: unavailableContent("assistant-task-pending") } };
|
|
194
242
|
});
|
|
195
243
|
}
|
|
196
244
|
|
|
197
245
|
function durableAssistant(item: TranscriptItem): boolean {
|
|
198
|
-
return item.role === "assistant" && item.toolCalls.length === 0
|
|
246
|
+
return (item.role === "assistant" || item.role === "custom") && item.toolCalls.length === 0
|
|
199
247
|
&& (item.content.state === "available" || item.content.state === "redacted")
|
|
200
248
|
&& hasVisibleText(item.content.text ?? "");
|
|
201
249
|
}
|
|
@@ -252,9 +300,13 @@ function unavailable(input: TranscriptProjectionInput, reasonCode: string, limit
|
|
|
252
300
|
export function projectTranscript(input: TranscriptProjectionInput): TranscriptDocument {
|
|
253
301
|
const limit = Math.max(1, Math.min(MAX_ITEMS, Number.isInteger(input.limit) ? Number(input.limit) : 50));
|
|
254
302
|
if (!Array.isArray(input.entries) || input.entries.length > MAX_ENTRIES) return unavailable(input, "transcript-history-unavailable", limit);
|
|
255
|
-
const
|
|
256
|
-
|
|
257
|
-
.
|
|
303
|
+
const receipt = input.terminalDeliveryReceipt;
|
|
304
|
+
const boundReceipt = receipt?.details.taskId === input.identity.taskId && receipt?.details.taskRunId === input.identity.taskRunId
|
|
305
|
+
&& receipt?.details.outcome === input.taskOutcome ? receipt : undefined;
|
|
306
|
+
const receiptPairs = new Map(terminalDeliveryPairs(input.entries, boundReceipt, input.terminalDeliveryEntries)
|
|
307
|
+
.map((pair) => [pair.receiptEntry, pair]));
|
|
308
|
+
const projected = suppressOpenTaskHandoff(linkTranscriptTurns(correlatedTranscriptItems(input.entries, input.identity,
|
|
309
|
+
receiptPairs)), input.taskOutcome);
|
|
258
310
|
const cursors = projected.map((value) => value.cursor);
|
|
259
311
|
let end = projected.length;
|
|
260
312
|
if (input.beforeCursor) {
|
|
@@ -266,5 +318,6 @@ export function projectTranscript(input: TranscriptProjectionInput): TranscriptD
|
|
|
266
318
|
return { schemaVersion: 1, version: "piagent-webui-transcript-v1", generatedAt: input.generatedAt ?? new Date().toISOString(),
|
|
267
319
|
identity: structuredClone(input.identity), revision: structuredClone(input.revision), eventCursor: input.eventCursor,
|
|
268
320
|
state: "ready", items: selected, page: { beforeCursor: input.beforeCursor ?? null, nextBeforeCursor: hasOlder ? cursors[page.start] : null,
|
|
269
|
-
hasOlder, limit, truncated: page.compacted || projected.length !== input.entries.filter((entry: any) => entry?.type === "message").length
|
|
321
|
+
hasOlder, limit, truncated: page.compacted || projected.length !== input.entries.filter((entry: any) => entry?.type === "message").length
|
|
322
|
+
+ receiptPairs.size * 2 }, reasonCode: null };
|
|
270
323
|
}
|
|
@@ -0,0 +1,25 @@
|
|
|
1
|
+
export const WEBUI_MESSAGE_CORRELATION_ENTRY_TYPE = "piagent-webui-message-correlation";
|
|
2
|
+
|
|
3
|
+
const OPAQUE_REF = /^[A-Za-z0-9][A-Za-z0-9._~-]{0,159}$/;
|
|
4
|
+
|
|
5
|
+
export type WebUiMessageCorrelation = {
|
|
6
|
+
messageRequestId: string;
|
|
7
|
+
operationRef: string;
|
|
8
|
+
};
|
|
9
|
+
|
|
10
|
+
/**
|
|
11
|
+
* A correlation marker is persisted immediately before the host receives the
|
|
12
|
+
* user prompt. Plain custom entries do not enter model context, but they let a
|
|
13
|
+
* later transcript projection recover the exact browser request and operation
|
|
14
|
+
* without trusting browser/server clock alignment or repeated message text.
|
|
15
|
+
*/
|
|
16
|
+
export function webUiMessageCorrelationEntry(entry: unknown): WebUiMessageCorrelation | null {
|
|
17
|
+
if (!entry || typeof entry !== "object" || Array.isArray(entry)) return null;
|
|
18
|
+
const candidate = entry as Record<string, any>;
|
|
19
|
+
if (candidate.type !== "custom" || candidate.customType !== WEBUI_MESSAGE_CORRELATION_ENTRY_TYPE
|
|
20
|
+
|| !candidate.data || typeof candidate.data !== "object" || Array.isArray(candidate.data)) return null;
|
|
21
|
+
const data = candidate.data as Record<string, unknown>;
|
|
22
|
+
if (data.schemaVersion !== 1 || typeof data.messageRequestId !== "string" || !OPAQUE_REF.test(data.messageRequestId)
|
|
23
|
+
|| typeof data.operationRef !== "string" || !OPAQUE_REF.test(data.operationRef)) return null;
|
|
24
|
+
return { messageRequestId: data.messageRequestId, operationRef: data.operationRef };
|
|
25
|
+
}
|