@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 +1,56 @@
|
|
|
1
|
+
import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
|
|
2
|
+
import { toolResultFingerprint } from "../../extensions/context-engine.js";
|
|
3
|
+
import { captureWorkspaceVerificationSnapshot } from "../../extensions/workspace-verification-snapshot.js";
|
|
4
|
+
|
|
5
|
+
/** Host invocation state; equal commands in parallel must not consume each other. */
|
|
6
|
+
export class ShellVerificationSnapshots {
|
|
7
|
+
readonly #snapshots = new Map<string, ReturnType<typeof captureWorkspaceVerificationSnapshot> | null>();
|
|
8
|
+
readonly #consumed = new Map<string, ReturnType<typeof captureWorkspaceVerificationSnapshot>>();
|
|
9
|
+
#overflowed = false;
|
|
10
|
+
|
|
11
|
+
#key(ctx: ExtensionContext, toolName: string, input: unknown, toolCallId?: string, taskRunId = ""): string {
|
|
12
|
+
const fingerprint = toolResultFingerprint(toolName, input, []).key;
|
|
13
|
+
return `${ctx.cwd}\u0000${ctx.sessionManager.getSessionId()}\u0000${taskRunId}\u0000${toolCallId ?? fingerprint}\u0000${fingerprint}`;
|
|
14
|
+
}
|
|
15
|
+
|
|
16
|
+
remember(ctx: ExtensionContext, toolName: string, input: unknown, toolCallId?: string, taskRunId?: string): void {
|
|
17
|
+
const key = this.#key(ctx, toolName, input, toolCallId, taskRunId);
|
|
18
|
+
// Ambiguous duplicate invocation IDs cannot supply a before-snapshot.
|
|
19
|
+
if (this.#snapshots.has(key)) { this.#snapshots.set(key, null); return; }
|
|
20
|
+
this.#snapshots.set(key, captureWorkspaceVerificationSnapshot(ctx.cwd));
|
|
21
|
+
while (this.#snapshots.size > 100) {
|
|
22
|
+
this.#overflowed = true; // Lost in-flight identity cannot become a cached pass later in this process.
|
|
23
|
+
this.#snapshots.delete(this.#snapshots.keys().next().value as string);
|
|
24
|
+
}
|
|
25
|
+
}
|
|
26
|
+
|
|
27
|
+
consume(ctx: ExtensionContext, toolName: string, input: unknown, toolCallId?: string, taskRunId?: string) {
|
|
28
|
+
const key = this.#key(ctx, toolName, input, toolCallId, taskRunId);
|
|
29
|
+
const snapshot = this.#snapshots.get(key);
|
|
30
|
+
this.#snapshots.delete(key);
|
|
31
|
+
if (snapshot) {
|
|
32
|
+
this.#consumed.set(key, snapshot);
|
|
33
|
+
while (this.#consumed.size > 100) this.#consumed.delete(this.#consumed.keys().next().value as string);
|
|
34
|
+
}
|
|
35
|
+
return snapshot ?? undefined;
|
|
36
|
+
}
|
|
37
|
+
|
|
38
|
+
claimObservation(ctx: ExtensionContext, toolName: string, input: unknown, toolCallId: string | undefined,
|
|
39
|
+
taskRunId: string, snapshot: ReturnType<typeof captureWorkspaceVerificationSnapshot> | undefined): boolean {
|
|
40
|
+
const key = this.#key(ctx, toolName, input, toolCallId, taskRunId);
|
|
41
|
+
const issued = this.#consumed.get(key);
|
|
42
|
+
this.#consumed.delete(key);
|
|
43
|
+
return issued !== undefined && issued === snapshot;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
hasPending(ctx: ExtensionContext, taskRunId: string): boolean {
|
|
47
|
+
const prefix = `${ctx.cwd}\0${ctx.sessionManager.getSessionId()}\0${taskRunId}\0`;
|
|
48
|
+
return this.#overflowed || [...this.#snapshots.keys()].some((key) => key.startsWith(prefix));
|
|
49
|
+
}
|
|
50
|
+
|
|
51
|
+
clear(ctx: ExtensionContext): void {
|
|
52
|
+
const prefix = `${ctx.cwd}\u0000${ctx.sessionManager.getSessionId()}\u0000`;
|
|
53
|
+
for (const key of this.#snapshots.keys()) if (key.startsWith(prefix)) this.#snapshots.delete(key);
|
|
54
|
+
for (const key of this.#consumed.keys()) if (key.startsWith(prefix)) this.#consumed.delete(key);
|
|
55
|
+
}
|
|
56
|
+
}
|
|
@@ -0,0 +1,61 @@
|
|
|
1
|
+
const CHANGE_INTENT = /\b(?:add|build|change|correct|create|fix|implement|modify|mutate|refactor|remove|rename|repair|replace|update|write|sua|them|doi|cap nhat|xoa|tao)\b/i;
|
|
2
|
+
const EXPLICIT_CHANGE_REQUEST = [
|
|
3
|
+
/^(?:(?:ok(?:ay|e)?|then|now|please|so|vay|bay gio|anh|em|vui long|lam on|tien hanh|tiep tuc|thuc hien)\b[\s,.:;!—–-]*)*(?:add|apply|build|change|correct|create|fix|implement|modify|mutate|refactor|remove|rename|repair|replace|update|write|sua|them|doi|cap nhat|xoa|tao)\b/i,
|
|
4
|
+
/^(?:please\s+)?(?:can|could|would|will)\s+you\s+(?:please\s+)?(?:add|apply|build|change|correct|create|fix|implement|modify|refactor|remove|rename|repair|replace|update|write)\b/i,
|
|
5
|
+
/^(?:i\s+(?:need|want)\s+you\s+to|(?:anh|a|toi|minh)\s+(?:can|muon)\s+em)\s+(?:add|apply|build|change|correct|create|fix|implement|modify|refactor|remove|rename|repair|replace|update|write|sua|them|doi|cap nhat|xoa|tao)\b/i,
|
|
6
|
+
/^(?:(?:ok(?:ay|e)?|then|now|please|so|vay|bay gio)\b[\s,.:;!—–-]*)*(?:go\s+ahead(?:\s+and|\s+with)?|proceed(?:\s+to|\s+with)?|tien\s+hanh|tiep\s+tuc|thuc\s+hien)\b.{0,40}\b(?:add|apply|build|changes?|fix|implementation|implement|modify|repair|replace|update|sua|them|doi|cap nhat|xoa|tao)\b/i,
|
|
7
|
+
/\b(?:add|apply|build|change|correct|create|fix|implement|modify|mutate|refactor|remove|rename|repair|replace|update|write|sua|them|doi|cap nhat|xoa|tao)\b.{0,64}\b(?:di|nhe|luon|cho anh|giup anh|go ahead)\b/i
|
|
8
|
+
];
|
|
9
|
+
const CHANGE_CHOICE_QUESTION = [
|
|
10
|
+
/\b(?:should|do|does|did|must|need|can|could|would|may)\s+(?:i|we|you|he|she|they)\b.{0,160}\b(?:or|hay)\b/i,
|
|
11
|
+
/\b(?:i|we|you|a|anh|em|minh|toi|tui|chung ta)\s+(?:need\s+to|have\s+to|should|can|could|would|co\s+can|can|nen|co\s+the|phai)\b.{0,160}\b(?:or|hay)\b/i,
|
|
12
|
+
/\b(?:co\s+nen|co\s+can|co\s+phai|lieu\s+co|nen|can)\b.{0,160}\b(?:or|hay)\b/i
|
|
13
|
+
];
|
|
14
|
+
const ENGLISH_YES_NO_CHANGE_QUESTION = /^(?:(?:so|then|now)\b[\s,.:;!—–-]*)*(?:should|must|do|does|did|can|could|would|may)\s+(?:i|we|you|he|she|they)\b/i;
|
|
15
|
+
const ENGLISH_WH_CHANGE_QUESTION = /^(?:(?:so|then|now)\b[\s,.:;!—–-]*)*(?:what|which|when|where|why|how)\b/i;
|
|
16
|
+
const VIETNAMESE_CHANGE_QUESTION = /\b(?:(?:a|anh|em|minh|toi|tui|chung\s+ta)\s+(?:co\s+can|can|nen|co\s+the|phai)|co\s+nen|co\s+can|co\s+phai|lieu\s+co)\b/i;
|
|
17
|
+
|
|
18
|
+
export function foldChangeIntent(value: string): string {
|
|
19
|
+
return String(value ?? "")
|
|
20
|
+
.normalize("NFKD")
|
|
21
|
+
.replace(/\p{M}/gu, "")
|
|
22
|
+
.replace(/[đĐ]/g, (character) => character === "Đ" ? "D" : "d")
|
|
23
|
+
.toLowerCase();
|
|
24
|
+
}
|
|
25
|
+
|
|
26
|
+
export function hasChangeIntent(value: string): boolean {
|
|
27
|
+
return CHANGE_INTENT.test(foldChangeIntent(value));
|
|
28
|
+
}
|
|
29
|
+
|
|
30
|
+
export function isExplicitChangeRequest(value: string): boolean {
|
|
31
|
+
const folded = foldChangeIntent(value);
|
|
32
|
+
return EXPLICIT_CHANGE_REQUEST.some((pattern) => pattern.test(folded));
|
|
33
|
+
}
|
|
34
|
+
|
|
35
|
+
/** Host-neutral linguistic classification; workflow authority is applied by task intake. */
|
|
36
|
+
export function isLightweightNonAuthorizingChangeLanguage(prompt: string): boolean {
|
|
37
|
+
const text = String(prompt ?? "").trim();
|
|
38
|
+
const folded = foldChangeIntent(text);
|
|
39
|
+
if (!text || /^\/[a-z0-9-]+\b/i.test(text) || !CHANGE_INTENT.test(folded)) return false;
|
|
40
|
+
const hasChoiceSeparator = /\b(?:or|hay)\b/i.test(folded);
|
|
41
|
+
const choiceBranches = hasChoiceSeparator ? folded.split(/\b(?:or|hay)\b/i) : [];
|
|
42
|
+
const mixedChangeChoice = choiceBranches.some((branch) => CHANGE_INTENT.test(branch))
|
|
43
|
+
&& choiceBranches.some((branch) => !CHANGE_INTENT.test(branch));
|
|
44
|
+
// A punctuated change-vs-non-change choice remains a question even when its
|
|
45
|
+
// change branch starts with "em fix" or ends in "sửa luôn". Two change
|
|
46
|
+
// branches such as "can you fix or replace...?" remain an explicit request.
|
|
47
|
+
if (mixedChangeChoice && /[?]\s*$/.test(text)) return true;
|
|
48
|
+
if (isExplicitChangeRequest(folded)) return false;
|
|
49
|
+
if (hasChoiceSeparator && CHANGE_CHOICE_QUESTION.some((pattern) => pattern.test(folded))) return true;
|
|
50
|
+
if (ENGLISH_YES_NO_CHANGE_QUESTION.test(folded)) return true;
|
|
51
|
+
return VIETNAMESE_CHANGE_QUESTION.test(folded) && (/[?]\s*$/.test(text) || /\bkhong\b[\s?.!]*$/i.test(folded));
|
|
52
|
+
}
|
|
53
|
+
|
|
54
|
+
export function isNonAuthorizingChangeLanguage(prompt: string): boolean {
|
|
55
|
+
const text = String(prompt ?? "").trim();
|
|
56
|
+
if (!text) return false;
|
|
57
|
+
if (isLightweightNonAuthorizingChangeLanguage(text)) return true;
|
|
58
|
+
if (isExplicitChangeRequest(text)) return false;
|
|
59
|
+
const folded = foldChangeIntent(text);
|
|
60
|
+
return CHANGE_INTENT.test(folded) && ENGLISH_WH_CHANGE_QUESTION.test(folded);
|
|
61
|
+
}
|
|
@@ -3,8 +3,17 @@ import path from "node:path";
|
|
|
3
3
|
|
|
4
4
|
import { redactSensitiveText } from "../../extensions/redaction-core.js";
|
|
5
5
|
import { LONG_INPUT_CHARS, MAX_INLINE_COLLAPSED_TASK_CHARS } from "../runtime-limits.ts";
|
|
6
|
+
import {
|
|
7
|
+
type WorkflowId,
|
|
8
|
+
stripWorkflowCommand,
|
|
9
|
+
workflowIdFromCommand
|
|
10
|
+
} from "./webui-workflow.ts";
|
|
6
11
|
|
|
7
|
-
export type FreshWorkflow =
|
|
12
|
+
export type FreshWorkflow = WorkflowId;
|
|
13
|
+
|
|
14
|
+
export function workflowIdFromInput(input: string): WorkflowId | null {
|
|
15
|
+
return workflowIdFromCommand(input);
|
|
16
|
+
}
|
|
8
17
|
|
|
9
18
|
export function looksLikeGovernedBoilerplate(text: string): boolean {
|
|
10
19
|
const lower = text.toLowerCase();
|
|
@@ -29,10 +38,7 @@ function extractFencedBlockAfter(label: RegExp, text: string): string | undefine
|
|
|
29
38
|
}
|
|
30
39
|
|
|
31
40
|
export function stripLeadingWorkflowCommand(input: string): string {
|
|
32
|
-
return input
|
|
33
|
-
.replace(/^\/(?:piagent-workflow|workflow)\s+(?:task|scout|be-to-fe|review|plan|platform-improve|discuss|commit|pr)\b\s*/i, "")
|
|
34
|
-
.replace(/^\/(?:task|scout|be-to-fe|review|plan|platform-improve|discuss|commit|pr)\b\s*/i, "")
|
|
35
|
-
.trim();
|
|
41
|
+
return stripWorkflowCommand(input);
|
|
36
42
|
}
|
|
37
43
|
|
|
38
44
|
export function extractTaskRequest(text: string): string {
|
|
@@ -47,6 +53,11 @@ export function extractTaskRequest(text: string): string {
|
|
|
47
53
|
return stripLeadingWorkflowCommand(text.trim());
|
|
48
54
|
}
|
|
49
55
|
|
|
56
|
+
/** Normalize the delivered agent prompt, not the earlier surface input. */
|
|
57
|
+
export function agentStartTaskRequest(prompt: string): string {
|
|
58
|
+
return looksLikeGovernedBoilerplate(prompt) ? extractTaskRequest(prompt) : prompt.trim();
|
|
59
|
+
}
|
|
60
|
+
|
|
50
61
|
export function trimTaskForInline(input: string): string {
|
|
51
62
|
const normalized = stripLeadingWorkflowCommand(input).trim().replace(/\n{3,}/g, "\n\n");
|
|
52
63
|
if (normalized.length <= MAX_INLINE_COLLAPSED_TASK_CHARS) return normalized;
|
|
@@ -55,13 +66,8 @@ export function trimTaskForInline(input: string): string {
|
|
|
55
66
|
|
|
56
67
|
export function chooseFreshWorkflow(original: string, task: string): FreshWorkflow {
|
|
57
68
|
const semantic = stripLeadingWorkflowCommand(task || original).toLowerCase();
|
|
58
|
-
const
|
|
59
|
-
|
|
60
|
-
if (workflowStart?.[1] === "be-to-fe") return "be-to-fe";
|
|
61
|
-
if (workflowStart?.[1] === "scout") return "scout";
|
|
62
|
-
if (workflowStart?.[1] === "task") return "task";
|
|
63
|
-
if (starts.startsWith("/be-to-fe")) return "be-to-fe";
|
|
64
|
-
if (starts.startsWith("/scout")) return "scout";
|
|
69
|
+
const explicit = workflowIdFromInput(original);
|
|
70
|
+
if (explicit) return explicit;
|
|
65
71
|
const asksForWrite = /\b(implement|support|surface|consume|write|change|fix)\b/.test(semantic);
|
|
66
72
|
if (/\b(scout|read-only|read only|audit|mapping|mapping matrix|map contract)\b/.test(semantic) && !asksForWrite) {
|
|
67
73
|
return "scout";
|
|
@@ -73,7 +79,7 @@ export function chooseFreshWorkflow(original: string, task: string): FreshWorkfl
|
|
|
73
79
|
}
|
|
74
80
|
|
|
75
81
|
export function isPiagentWorkflowInput(text: string): boolean {
|
|
76
|
-
return
|
|
82
|
+
return workflowIdFromInput(text) !== null;
|
|
77
83
|
}
|
|
78
84
|
|
|
79
85
|
export function isFreshOrUtilityInput(text: string): boolean {
|
|
@@ -104,7 +110,8 @@ export function buildFreshCommand(cwd: string, workflow: FreshWorkflow, original
|
|
|
104
110
|
// never becomes task instructions or provider-visible document content.
|
|
105
111
|
return `/fresh ${workflow} --session-title "${shortTaskLabel(task)}" Read task intake from ${intakePath}. ${reason}`;
|
|
106
112
|
}
|
|
107
|
-
|
|
113
|
+
const inline = trimTaskForInline(task);
|
|
114
|
+
return [`/fresh ${workflow}`, inline].filter(Boolean).join(" ");
|
|
108
115
|
}
|
|
109
116
|
|
|
110
117
|
export function shortTaskLabel(text: string): string {
|
|
@@ -19,17 +19,38 @@ function legacyHeadTail(text: string): string {
|
|
|
19
19
|
return `${text.slice(0, head).trimEnd()}${marker}${text.slice(-(available - head)).trimStart()}`;
|
|
20
20
|
}
|
|
21
21
|
|
|
22
|
+
function compactProofTag(tag: string): string {
|
|
23
|
+
const identity = tag.match(/^\[(criterion-[0-9]{2}|fallback)/)?.[1] ?? "fallback";
|
|
24
|
+
if (tag.includes(":fallback")) return `[${identity}:fallback]`;
|
|
25
|
+
if (tag.includes(":")) return `[${identity}:candidate]`;
|
|
26
|
+
return `[${identity}]`;
|
|
27
|
+
}
|
|
28
|
+
|
|
29
|
+
/** Compress locations and repeated boilerplate, never semantic proof bodies. */
|
|
30
|
+
function semanticCriticalProof(lines: string[]): string[] {
|
|
31
|
+
if (lines.length === 0) return [];
|
|
32
|
+
const bodies = new Map<string, Set<string>>();
|
|
33
|
+
for (const line of lines.slice(1)) {
|
|
34
|
+
if (!line.trim() || /^(?:Candidate tags |No concrete criterion-linked |Exact final-output contract:)/.test(line)) continue;
|
|
35
|
+
const match = line.match(/^- ((?:\[(?:criterion-[0-9]{2}|fallback)(?::[^\]]+)?\])*)\s*(.+)$/);
|
|
36
|
+
const body = match?.[2] ?? line;
|
|
37
|
+
const tags = bodies.get(body) ?? new Set<string>();
|
|
38
|
+
for (const tag of match?.[1].match(/\[(?:criterion-[0-9]{2}|fallback)(?::[^\]]+)?\]/g) ?? []) tags.add(compactProofTag(tag));
|
|
39
|
+
bodies.set(body, tags);
|
|
40
|
+
}
|
|
41
|
+
return [
|
|
42
|
+
"Critical behavioral proof:",
|
|
43
|
+
...[...bodies].map(([body, tags]) => `- ${tags.size ? `${[...tags].join("")} ` : ""}${body}`),
|
|
44
|
+
"Use linked durable live assertions after final mutation and before exact verifiers; candidate tags are locations, not proof. Fallback tags require focused tests."
|
|
45
|
+
];
|
|
46
|
+
}
|
|
47
|
+
|
|
22
48
|
function minimalCriticalProof(lines: string[]): string[] {
|
|
23
49
|
if (lines.length === 0) return [];
|
|
24
|
-
const tags = [...new Set((lines.join("\n").match(/\[(?:criterion-[0-9]{2}|fallback)(?::[^\]]+)?\]/g) ?? []).map((
|
|
25
|
-
const identity = tag.match(/^\[(criterion-[0-9]{2}|fallback)/)?.[1] ?? "fallback";
|
|
26
|
-
if (tag.includes(":fallback")) return `[${identity}:fallback]`;
|
|
27
|
-
if (tag.includes(":")) return `[${identity}:candidate]`;
|
|
28
|
-
return `[${identity}]`;
|
|
29
|
-
}))].slice(0, 12);
|
|
50
|
+
const tags = [...new Set((lines.join("\n").match(/\[(?:criterion-[0-9]{2}|fallback)(?::[^\]]+)?\]/g) ?? []).map(compactProofTag))].slice(0, 12);
|
|
30
51
|
return [
|
|
31
52
|
"Critical behavioral proof:",
|
|
32
|
-
`- ${tags.join("")}
|
|
53
|
+
`- ${tags.join("")} Detailed semantic proof hints omitted; consult the complete operator contract.`,
|
|
33
54
|
"Fallback tags require a focused test. Run proof after final mutation and before exact verifiers; prose or transient probes are insufficient."
|
|
34
55
|
];
|
|
35
56
|
}
|
|
@@ -111,6 +132,22 @@ function structuredCompaction(text: string): string | undefined {
|
|
|
111
132
|
].filter(Boolean).join("\n");
|
|
112
133
|
const preferred = compose(criticalLines, executionLines);
|
|
113
134
|
if (preferred.length <= RUNTIME_INTAKE_MESSAGE_MAX_CHARS) return preferred;
|
|
135
|
+
// An ordinary contract can overflow because of navigation and repeated
|
|
136
|
+
// instructions. Preserve its actionable partitions before discarding hints.
|
|
137
|
+
const semanticProof = semanticCriticalProof(criticalLines);
|
|
138
|
+
const proofPreserving = compose(semanticProof, minimalExecutionMap(executionLines), true);
|
|
139
|
+
if (proofPreserving.length <= RUNTIME_INTAKE_MESSAGE_MAX_CHARS) return proofPreserving;
|
|
140
|
+
const compactProofPreserving = [
|
|
141
|
+
...prefix.slice(0, 1).map((line) => edgeBounded(line, 120)),
|
|
142
|
+
COMPACTION_MARKER,
|
|
143
|
+
...semanticProof,
|
|
144
|
+
...exactOutputLines,
|
|
145
|
+
...displayExactLines,
|
|
146
|
+
...identityOnlyExecutionMap(executionLines),
|
|
147
|
+
edgeBounded(lines[reuse], 120),
|
|
148
|
+
edgeBounded(finalGuidance, 120)
|
|
149
|
+
].filter(Boolean).join("\n");
|
|
150
|
+
if (compactProofPreserving.length <= RUNTIME_INTAKE_MESSAGE_MAX_CHARS) return compactProofPreserving;
|
|
114
151
|
const minimal = compose(minimalCriticalProof(criticalLines), minimalExecutionMap(executionLines), true);
|
|
115
152
|
if (minimal.length <= RUNTIME_INTAKE_MESSAGE_MAX_CHARS) return minimal;
|
|
116
153
|
const guaranteed = [
|
|
@@ -0,0 +1,163 @@
|
|
|
1
|
+
import { normalizePathCandidate } from "../../extensions/policy-core.js";
|
|
2
|
+
import { isCurrentWorkingTreeDigest } from "../../extensions/working-tree-digest.js";
|
|
3
|
+
import { foldChangeIntent } from "./change-clarification.ts";
|
|
4
|
+
import { classifyConditionalRepairIntent } from "./task-repair-intent.ts";
|
|
5
|
+
|
|
6
|
+
export type PriorTaskScopeEvidence = {
|
|
7
|
+
trace?: { outcome?: string };
|
|
8
|
+
changedFiles?: unknown;
|
|
9
|
+
observedChangedFiles?: unknown;
|
|
10
|
+
contextManifest?: unknown;
|
|
11
|
+
};
|
|
12
|
+
|
|
13
|
+
type FollowupAcceptanceTask = PriorTaskScopeEvidence & {
|
|
14
|
+
taskRunId?: unknown;
|
|
15
|
+
sessionId?: unknown;
|
|
16
|
+
createdAt?: unknown;
|
|
17
|
+
changeMode?: unknown;
|
|
18
|
+
mutationPolicy?: unknown;
|
|
19
|
+
intakeMode?: unknown;
|
|
20
|
+
operatorRequest?: unknown;
|
|
21
|
+
summary?: unknown;
|
|
22
|
+
scope?: unknown;
|
|
23
|
+
baselineFileDigests?: unknown;
|
|
24
|
+
finalFileDigests?: unknown;
|
|
25
|
+
};
|
|
26
|
+
|
|
27
|
+
type FileDigestRecord = Record<string, string>;
|
|
28
|
+
|
|
29
|
+
export function hasConditionalRepairIntent(text: string): boolean {
|
|
30
|
+
return classifyConditionalRepairIntent(text) !== "none";
|
|
31
|
+
}
|
|
32
|
+
|
|
33
|
+
function refersToEarlierImplementation(prompt: string): boolean {
|
|
34
|
+
const folded = foldChangeIntent(prompt);
|
|
35
|
+
const negatesPriorTurnReview = /\b(?:do not|don't|without|ignore|skip)\s+(?:review|verify|inspect|test|check|fix|continue|complete|apply|address)\b[\s\S]{0,200}\b(?:implementation|requirements?|obligations?|changes?|fix|logic|work)\s+(?:from|in)\s+(?:the\s+)?(?:earlier|previous|prior)\s+(?:turn|message)\b/i
|
|
36
|
+
.test(folded);
|
|
37
|
+
if (negatesPriorTurnReview) return false;
|
|
38
|
+
return [
|
|
39
|
+
/\b(?:current|existing|earlier|previous|prior|above)\s+(?:changes?|fix|implementation|logic|work)\b/i,
|
|
40
|
+
/\b(?:earlier|previous|prior|original|above)\s+(?:instructions?|obligations?|request|requirements?|task)\b/i,
|
|
41
|
+
/\b(?:changes?|fix|implementation|logic|work)\s+(?:above|earlier|previously|just\s+(?:completed|made|implemented))\b/i,
|
|
42
|
+
/\bwhat\s+(?:you|we)\s+just\s+(?:changed|fixed|implemented)\b/i,
|
|
43
|
+
/\b(?:review|verify|inspect|test|check|fix|continue|complete|apply|address)\b[\s\S]{0,200}\b(?:implementation|requirements?|obligations?|changes?|fix|logic|work)\s+(?:from|in)\s+(?:the\s+)?(?:earlier|previous|prior)\s+(?:turn|message)\b/i,
|
|
44
|
+
/\b(?:implementation|phan|logic|thay doi|yeu cau)\s+(?:hien tai|truoc do|vua roi|vua sua|vua lam|vua trien khai)\b/i,
|
|
45
|
+
/\b(?:vua sua|vua lam|vua trien khai|thay doi vua roi|thay doi truoc do)\b/i
|
|
46
|
+
].some((pattern) => pattern.test(folded));
|
|
47
|
+
}
|
|
48
|
+
|
|
49
|
+
function fileDigestRecord(value: unknown): FileDigestRecord | undefined {
|
|
50
|
+
if (!value || typeof value !== "object" || Array.isArray(value)) return undefined;
|
|
51
|
+
const entries = Object.entries(value);
|
|
52
|
+
const normalized = entries.map(([file, digest]) => [normalizePathCandidate(file), digest] as const);
|
|
53
|
+
if (normalized.some(([file, digest]) => !file || !isCurrentWorkingTreeDigest(digest))) return undefined;
|
|
54
|
+
if (new Set(normalized.map(([file]) => file)).size !== normalized.length) return undefined;
|
|
55
|
+
return Object.fromEntries(normalized);
|
|
56
|
+
}
|
|
57
|
+
|
|
58
|
+
function sameFileDigestRecord(left: FileDigestRecord, right: FileDigestRecord): boolean {
|
|
59
|
+
const leftEntries = Object.entries(left).sort(([a], [b]) => a.localeCompare(b));
|
|
60
|
+
const rightEntries = Object.entries(right).sort(([a], [b]) => a.localeCompare(b));
|
|
61
|
+
return leftEntries.length === rightEntries.length
|
|
62
|
+
&& leftEntries.every(([file, digest], index) => file === rightEntries[index]?.[0] && digest === rightEntries[index]?.[1]);
|
|
63
|
+
}
|
|
64
|
+
|
|
65
|
+
function changedDigestFiles(baseline: FileDigestRecord, final: FileDigestRecord): string[] {
|
|
66
|
+
return [...new Set([...Object.keys(baseline), ...Object.keys(final)])]
|
|
67
|
+
.filter((file) => baseline[file] !== final[file])
|
|
68
|
+
.sort();
|
|
69
|
+
}
|
|
70
|
+
|
|
71
|
+
function normalizedFiles(value: unknown): string[] {
|
|
72
|
+
return [...new Set((Array.isArray(value) ? value : [])
|
|
73
|
+
.filter((file): file is string => typeof file === "string")
|
|
74
|
+
.map((file) => normalizePathCandidate(file))
|
|
75
|
+
.filter(Boolean))]
|
|
76
|
+
.sort();
|
|
77
|
+
}
|
|
78
|
+
|
|
79
|
+
function sameFiles(left: string[], right: string[]): boolean {
|
|
80
|
+
return left.length === right.length && left.every((file, index) => file === right[index]);
|
|
81
|
+
}
|
|
82
|
+
|
|
83
|
+
/**
|
|
84
|
+
* A zero-delta verification follow-up may prove the immediately preceding
|
|
85
|
+
* implementation without claiming those files as its own mutation. The
|
|
86
|
+
* lineage is intentionally strict: same session, adjacent completed source
|
|
87
|
+
* task, exact snapshot continuity, exact child focus, and an explicit
|
|
88
|
+
* reference to earlier work. Ambiguity returns only the task-local delta.
|
|
89
|
+
*/
|
|
90
|
+
export function completedFollowupAcceptanceEvidenceFiles(
|
|
91
|
+
task: FollowupAcceptanceTask,
|
|
92
|
+
tasks: FollowupAcceptanceTask[],
|
|
93
|
+
taskLocalDelta: string[],
|
|
94
|
+
currentFileDigests: unknown
|
|
95
|
+
): string[] {
|
|
96
|
+
const local = normalizedFiles(taskLocalDelta);
|
|
97
|
+
const prompt = typeof task.operatorRequest === "string" && task.operatorRequest.trim()
|
|
98
|
+
? task.operatorRequest
|
|
99
|
+
: typeof task.summary === "string" ? task.summary : "";
|
|
100
|
+
const current = fileDigestRecord(currentFileDigests);
|
|
101
|
+
const childBaseline = fileDigestRecord(task.baselineFileDigests);
|
|
102
|
+
const createdAt = typeof task.createdAt === "string" ? Date.parse(task.createdAt) : Number.NaN;
|
|
103
|
+
if (
|
|
104
|
+
task.intakeMode !== "runtime"
|
|
105
|
+
|| task.changeMode !== "source-change"
|
|
106
|
+
|| task.mutationPolicy !== "allowed"
|
|
107
|
+
|| typeof task.taskRunId !== "string"
|
|
108
|
+
|| typeof task.sessionId !== "string"
|
|
109
|
+
|| !Number.isFinite(createdAt)
|
|
110
|
+
|| !refersToEarlierImplementation(prompt)
|
|
111
|
+
|| !current
|
|
112
|
+
|| !childBaseline
|
|
113
|
+
) return local;
|
|
114
|
+
|
|
115
|
+
const prior = tasks
|
|
116
|
+
.filter((candidate) => candidate.taskRunId !== task.taskRunId
|
|
117
|
+
&& candidate.sessionId === task.sessionId
|
|
118
|
+
&& typeof candidate.createdAt === "string"
|
|
119
|
+
&& Number.isFinite(Date.parse(candidate.createdAt))
|
|
120
|
+
&& Date.parse(candidate.createdAt) < createdAt)
|
|
121
|
+
.sort((left, right) => Date.parse(String(right.createdAt)) - Date.parse(String(left.createdAt)));
|
|
122
|
+
if (tasks.some((candidate) => candidate.taskRunId !== task.taskRunId
|
|
123
|
+
&& candidate.sessionId === task.sessionId
|
|
124
|
+
&& typeof candidate.createdAt === "string"
|
|
125
|
+
&& Date.parse(candidate.createdAt) === createdAt)) return local;
|
|
126
|
+
const parent = prior[0];
|
|
127
|
+
if (!parent || (prior[1] && Date.parse(String(prior[1].createdAt)) === Date.parse(String(parent.createdAt)))) return local;
|
|
128
|
+
if (parent.trace?.outcome !== "completed" || parent.changeMode !== "source-change" || parent.mutationPolicy === "forbidden") return local;
|
|
129
|
+
|
|
130
|
+
const parentBaseline = fileDigestRecord(parent.baselineFileDigests);
|
|
131
|
+
const parentFinal = fileDigestRecord(parent.finalFileDigests);
|
|
132
|
+
if (!parentBaseline || !parentFinal || !sameFileDigestRecord(parentFinal, childBaseline)) return local;
|
|
133
|
+
const parentChanged = normalizedFiles(parent.changedFiles);
|
|
134
|
+
if (parentChanged.length === 0 || !sameFiles(parentChanged, changedDigestFiles(parentBaseline, parentFinal))) return local;
|
|
135
|
+
|
|
136
|
+
const exactFocus = new Set(normalizedFiles(task.scope).filter((file) => !/[?*\[\]{}]/.test(file)));
|
|
137
|
+
const inherited = parentChanged.filter((file) => exactFocus.has(file) && Object.hasOwn(current, file));
|
|
138
|
+
return [...new Set([...local, ...inherited])].sort();
|
|
139
|
+
}
|
|
140
|
+
|
|
141
|
+
export function completedFollowupScope(
|
|
142
|
+
prompt: string,
|
|
143
|
+
task: PriorTaskScopeEvidence | undefined,
|
|
144
|
+
plausiblePath: (value: string) => boolean
|
|
145
|
+
): string[] {
|
|
146
|
+
if (!refersToEarlierImplementation(prompt) || task?.trace?.outcome !== "completed") return [];
|
|
147
|
+
const changed = [task.changedFiles, task.observedChangedFiles]
|
|
148
|
+
.flatMap((value) => Array.isArray(value) ? value : []);
|
|
149
|
+
const context = Array.isArray(task.contextManifest)
|
|
150
|
+
? task.contextManifest.map((item) => item && typeof item === "object" ? (item as { path?: unknown }).path : undefined)
|
|
151
|
+
: [];
|
|
152
|
+
return [...new Set([...changed, ...context]
|
|
153
|
+
.filter((value): value is string => typeof value === "string")
|
|
154
|
+
.map((value) => normalizePathCandidate(value))
|
|
155
|
+
.filter((candidate): candidate is string => Boolean(
|
|
156
|
+
candidate
|
|
157
|
+
&& candidate !== "."
|
|
158
|
+
&& !candidate.startsWith(".pi/")
|
|
159
|
+
&& !["AGENTS.md", "README.md", "REVIEW_GUIDELINES.md"].includes(candidate)
|
|
160
|
+
&& plausiblePath(candidate)
|
|
161
|
+
)))]
|
|
162
|
+
.slice(0, 8);
|
|
163
|
+
}
|