@piagent/platform 1.6.1 → 1.8.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/CHANGELOG.md +71 -0
- package/README.md +8 -2
- package/README.vi.md +11 -2
- package/adapters/common/assurance-producers.json +109 -0
- package/adapters/common/structured-assurance-facts.mjs +153 -0
- package/adapters/node-typescript/contract-families.json +2811 -0
- package/adapters/node-typescript/contract-families.md +90 -0
- package/benchmarks/production-v1/grade.mjs +20 -0
- package/benchmarks/production-v1/spend-control.v1.json +3 -3
- package/benchmarks/production-v2/grade.mjs +892 -0
- package/benchmarks/production-v2/project/README.md +7 -0
- package/benchmarks/production-v2/project/config/service.json +4 -0
- package/benchmarks/production-v2/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v2/project/docs/ops.md +3 -0
- package/benchmarks/production-v2/project/logs/incident.log +2 -0
- package/benchmarks/production-v2/project/package.json +11 -0
- package/benchmarks/production-v2/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v2/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v2/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v2/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v2/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v2/project/src/backend/revocation-cache.js +3 -0
- package/benchmarks/production-v2/project/src/data/csv.js +3 -0
- package/benchmarks/production-v2/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v2/project/src/data/migration.js +9 -0
- package/benchmarks/production-v2/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v2/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v2/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/request-lifecycle.js +15 -0
- package/benchmarks/production-v2/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v2/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v2/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v2/project/src/platform/args.js +10 -0
- package/benchmarks/production-v2/project/src/platform/config.js +7 -0
- package/benchmarks/production-v2/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v2/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v2/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/expiry.js +5 -0
- package/benchmarks/production-v2/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v2/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v2/project/vendor/README.md +1 -0
- package/benchmarks/production-v2/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v2/prompts/backend-frontend-contract-sync.md +12 -0
- package/benchmarks/production-v2/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v2/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v2/prompts/chunked-record-boundary.md +9 -0
- package/benchmarks/production-v2/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v2/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v2/prompts/destructive-history-refusal.md +3 -0
- package/benchmarks/production-v2/prompts/expiry-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/idempotent-replay-conflict.md +11 -0
- package/benchmarks/production-v2/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v2/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v2/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v2/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v2/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v2/prompts/pagination-boundary.md +7 -0
- package/benchmarks/production-v2/prompts/protected-env-refusal.md +2 -0
- package/benchmarks/production-v2/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v2/prompts/reconnect-chat-event-order.md +19 -0
- package/benchmarks/production-v2/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v2/prompts/resumable-checkpoint-partial-failure.md +11 -0
- package/benchmarks/production-v2/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v2/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v2/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v2/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v2/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v2/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v2/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v2/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v2/spend-control.v1.json +115 -0
- package/benchmarks/production-v2/suite.json +1009 -0
- package/benchmarks/production-v2/variant.mjs +169 -0
- package/benchmarks/production-v3/CONTRACT-REVISION-20260922.md +31 -0
- package/benchmarks/production-v3/calibration/corpus.v1.json +63 -0
- package/benchmarks/production-v3/calibration/human-review-rubric.v1.json +36 -0
- package/benchmarks/production-v3/grade.mjs +1017 -0
- package/benchmarks/production-v3/project/README.md +7 -0
- package/benchmarks/production-v3/project/config/service.json +4 -0
- package/benchmarks/production-v3/project/data/audit-ledger.jsonl +1 -0
- package/benchmarks/production-v3/project/docs/ops.md +3 -0
- package/benchmarks/production-v3/project/logs/incident.log +2 -0
- package/benchmarks/production-v3/project/package.json +11 -0
- package/benchmarks/production-v3/project/scripts/check.mjs +21 -0
- package/benchmarks/production-v3/project/src/backend/auth.js +3 -0
- package/benchmarks/production-v3/project/src/backend/billing-window.js +4 -0
- package/benchmarks/production-v3/project/src/backend/cache.js +15 -0
- package/benchmarks/production-v3/project/src/backend/invoice.js +4 -0
- package/benchmarks/production-v3/project/src/backend/revocation-cache.js +12 -0
- package/benchmarks/production-v3/project/src/data/csv.js +3 -0
- package/benchmarks/production-v3/project/src/data/dedup.js +5 -0
- package/benchmarks/production-v3/project/src/data/migration.js +9 -0
- package/benchmarks/production-v3/project/src/data/ndjson-stream.js +5 -0
- package/benchmarks/production-v3/project/src/data/versioned-replay.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/chat-events.js +7 -0
- package/benchmarks/production-v3/project/src/frontend/pagination.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/request-lifecycle.js +23 -0
- package/benchmarks/production-v3/project/src/frontend/search-state.js +8 -0
- package/benchmarks/production-v3/project/src/frontend/unicode-search.js +7 -0
- package/benchmarks/production-v3/project/src/fullstack/contract-sync.js +10 -0
- package/benchmarks/production-v3/project/src/platform/args.js +10 -0
- package/benchmarks/production-v3/project/src/platform/config.js +7 -0
- package/benchmarks/production-v3/project/src/platform/workflow-session.js +9 -0
- package/benchmarks/production-v3/project/src/platform/workspace.js +13 -0
- package/benchmarks/production-v3/project/src/reliability/checkpoint.js +5 -0
- package/benchmarks/production-v3/project/src/reliability/expiry.js +14 -0
- package/benchmarks/production-v3/project/src/reliability/retry.js +15 -0
- package/benchmarks/production-v3/project/test/smoke.test.js +21 -0
- package/benchmarks/production-v3/project/vendor/README.md +1 -0
- package/benchmarks/production-v3/prompts/abort-reconnect-supersession.md +11 -0
- package/benchmarks/production-v3/prompts/backend-frontend-contract-sync.md +20 -0
- package/benchmarks/production-v3/prompts/billing-cutoff-clock-skew.md +11 -0
- package/benchmarks/production-v3/prompts/bounded-retry.md +7 -0
- package/benchmarks/production-v3/prompts/chunked-record-boundary.md +11 -0
- package/benchmarks/production-v3/prompts/cli-double-dash.md +7 -0
- package/benchmarks/production-v3/prompts/config-precedence.md +6 -0
- package/benchmarks/production-v3/prompts/destructive-history-refusal.md +9 -0
- package/benchmarks/production-v3/prompts/expiry-boundary.md +14 -0
- package/benchmarks/production-v3/prompts/idempotent-replay-conflict.md +13 -0
- package/benchmarks/production-v3/prompts/incident-diagnosis.md +4 -0
- package/benchmarks/production-v3/prompts/invoice-rounding.md +9 -0
- package/benchmarks/production-v3/prompts/journeys/config-precedence-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/expiry-boundary-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/quoted-csv-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/recover-after-uncertain-send.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/repository-prompt-injection-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/resumable-checkpoint-partial-failure-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/stable-dedup-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/stale-search-response-scout.md +3 -0
- package/benchmarks/production-v3/prompts/journeys/tenant-cache-isolation-scout.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/verify-current-contract.md +4 -0
- package/benchmarks/production-v3/prompts/journeys/workflow-switch-same-session-scout.md +4 -0
- package/benchmarks/production-v3/prompts/pagination-boundary.md +9 -0
- package/benchmarks/production-v3/prompts/protected-env-refusal.md +8 -0
- package/benchmarks/production-v3/prompts/quoted-csv.md +6 -0
- package/benchmarks/production-v3/prompts/reconnect-chat-event-order.md +20 -0
- package/benchmarks/production-v3/prompts/repository-prompt-injection.md +5 -0
- package/benchmarks/production-v3/prompts/resumable-checkpoint-partial-failure.md +15 -0
- package/benchmarks/production-v3/prompts/revoked-session-cache.md +10 -0
- package/benchmarks/production-v3/prompts/schema-migration.md +6 -0
- package/benchmarks/production-v3/prompts/stable-dedup.md +6 -0
- package/benchmarks/production-v3/prompts/stale-search-response.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-cache-isolation.md +6 -0
- package/benchmarks/production-v3/prompts/tenant-role-authorization.md +6 -0
- package/benchmarks/production-v3/prompts/unicode-search.md +6 -0
- package/benchmarks/production-v3/prompts/workflow-switch-same-session.md +30 -0
- package/benchmarks/production-v3/prompts/workspace-order.md +7 -0
- package/benchmarks/production-v3/public-checks/abort-reconnect-supersession.js +34 -0
- package/benchmarks/production-v3/public-checks/expiry-boundary.js +29 -0
- package/benchmarks/production-v3/public-checks/revoked-session-cache.js +20 -0
- package/benchmarks/production-v3/spend-control.v1.json +121 -0
- package/benchmarks/production-v3/suite.json +1010 -0
- package/benchmarks/production-v3/variant.mjs +178 -0
- package/catalog/capabilities.json +2 -2
- package/docs/benchmark-codex-mcp-metadata-contract-v1.md +159 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v2.md +182 -0
- package/docs/benchmark-codex-mcp-metadata-contract-v3.md +194 -0
- package/docs/benchmark-public-contract-addendum-v2.md +85 -0
- package/docs/benchmark-public-contract-addendum-v3.md +116 -0
- package/docs/command-reference-vietnamese.md +10 -7
- package/docs/context-window-policy.md +3 -3
- package/docs/distribution-standard.md +9 -9
- package/docs/languages.json +2 -2
- package/docs/model-options.md +20 -15
- package/docs/operator-manual-vietnamese.md +12 -12
- package/docs/piagent-codex-benchmark-recovery-plan.md +1817 -0
- package/docs/piagent-core-evaluation-specs-2026-09-06.json +706 -0
- package/docs/piagent-core-improvement-plan.md +528 -0
- package/docs/piagent-core-offline-implementation-2026-09-06.md +219 -0
- package/docs/piagent-core-plan-checkpoint-2026-09-06.json +69 -0
- package/docs/piagent-core-reasoning-research-2026-09-06.md +225 -0
- package/docs/piagent-core-research-sources-2026-09-06.json +416 -0
- package/docs/piagent-open-session-wait.md +19 -0
- package/docs/piagent-runtime-benchmark-execution-plan.md +685 -0
- package/docs/publishing-for-teams.md +3 -0
- package/docs/quality-benchmark.md +128 -8
- package/docs/quickstart-vietnamese.md +2 -2
- package/docs/release-install-policy.md +38 -11
- package/docs/research/harness-next-evidence-review.md +113 -0
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +1 -1
- package/docs-site/commands.html +1 -1
- package/docs-site/content/en/runtime.html +1 -1
- package/docs-site/content/en/whats-new.html +23 -0
- package/docs-site/content/runtime.html +2 -2
- package/docs-site/content/whats-new.html +23 -0
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +1 -1
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +1 -1
- package/docs-site/en/mcp-approval.html +1 -1
- package/docs-site/en/mcp-auth.html +1 -1
- package/docs-site/en/mcp-servers.html +1 -1
- package/docs-site/en/mcp.html +1 -1
- package/docs-site/en/permissions.html +1 -1
- package/docs-site/en/profiles.html +1 -1
- package/docs-site/en/quickstart.html +1 -1
- package/docs-site/en/runtime.html +2 -2
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -1
- package/docs-site/en/workflows.html +1 -1
- package/docs-site/index.html +1 -1
- package/docs-site/mcp-approval.html +1 -1
- package/docs-site/mcp-auth.html +1 -1
- package/docs-site/mcp-servers.html +1 -1
- package/docs-site/mcp.html +1 -1
- package/docs-site/permissions.html +1 -1
- package/docs-site/profiles.html +1 -1
- package/docs-site/quickstart.html +1 -1
- package/docs-site/runtime.html +3 -3
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -1
- package/docs-site/workflows.html +1 -1
- package/evals/architecture-conformance-v1/lane.json +18 -0
- package/evals/architecture-conformance-v1/runner.mjs +77 -0
- package/evals/fixtures/approved-host-contracts.invalid.json +43 -0
- package/evals/fixtures/approved-host-contracts.valid.json +43 -0
- package/evals/fixtures/benchmark-attempt-outcome.invalid.json +42 -0
- package/evals/fixtures/benchmark-attempt-outcome.valid.json +42 -0
- package/evals/fixtures/benchmark-independent-verification-plan.invalid.json +7 -0
- package/evals/fixtures/benchmark-independent-verification-plan.valid.json +21 -0
- package/evals/fixtures/composite-criterion.invalid.json +10 -0
- package/evals/fixtures/composite-criterion.valid.json +33 -0
- package/evals/fixtures/contract-family-library.invalid.json +42 -0
- package/evals/fixtures/contract-family-library.valid.json +41 -0
- package/evals/fixtures/contract-selection-recipe.invalid.json +29 -0
- package/evals/fixtures/contract-selection-recipe.valid.json +29 -0
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.invalid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v2.valid.json +42 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.invalid.json +57 -0
- package/evals/fixtures/scoped-project-verification-receipt-v3.valid.json +57 -0
- package/evals/fixtures/scoped-verification-receipt-v1.invalid.json +40 -0
- package/evals/fixtures/scoped-verification-receipt-v1.valid.json +40 -0
- package/evals/fs5-artifact-archive/README.md +32 -0
- package/evals/fs5-artifact-archive/manifest.json +148 -0
- package/evals/fs5-artifact-archive/sha256/1f073f6d42c093a3182685e0e096144c8b3befd87594aeea1af05ce17653860c.txt +380 -0
- package/evals/fs5-artifact-archive/sha256/1ff46984c5ff668647dfad94248cd9dcfb72384a8aa529279345582f90d7486a.txt +331 -0
- package/evals/fs5-artifact-archive/sha256/20a5e1ef1cab3234a888b876b3cec52e702a05bc5ac000b691f94c651beed9ca.txt +800 -0
- package/evals/fs5-artifact-archive/sha256/29bb79e96b047d79056626d747baa0b3a6ca9d4da7ad80b9668c0085e5581706.txt +116 -0
- package/evals/fs5-artifact-archive/sha256/2bcc0cf6c18eee0c080c30e727610b0f10a278f5862855d920ba5797f07a2f2d.txt +252 -0
- package/evals/fs5-artifact-archive/sha256/37ecc65febe2c5947ac9288ad958bc149e0f08c6beda137edb1c660908853832.txt +324 -0
- package/evals/fs5-artifact-archive/sha256/5d937632d2cfff34dbbb781dd367bb78feb64c36d571c36ca15f2cc40c576af8.txt +142 -0
- package/evals/fs5-artifact-archive/sha256/6057c7bf2f40f8e3db7e293f08d745133c05293b75e9f363fb4e2e53bd09a76a.txt +244 -0
- package/evals/fs5-artifact-archive/sha256/62ea354afb6816b148ae9bd95ffb1ee5e29580eb4a22473e1ae19dbb471ec36b.txt +930 -0
- package/evals/fs5-artifact-archive/sha256/8987775b742f3b3650d75cce84e6d58fcdd3dfb835d03dac4c2bfe6d9eea3a2e.txt +903 -0
- package/evals/fs5-artifact-archive/sha256/902e886d0920b66811f54d0d1d933d70f8fab8704813b46d578bee6af1f21eef.txt +942 -0
- package/evals/fs5-artifact-archive/sha256/976a9908adfd765ee7be0655616f39a44fff64824f5e8535d726082e550b02eb.txt +910 -0
- package/evals/fs5-artifact-archive/sha256/9c07bda5bfe7bb7b0959844ccea6790735cc9f154d8d5972e18785148cea6d9a.txt +101 -0
- package/evals/fs5-artifact-archive/sha256/b936a604db87a697bf69581c33d753e26d9e5e2ccce40bf9c9b2070948f49919.txt +931 -0
- package/evals/fs5-artifact-archive/sha256/bf91f69473f48fc3de7c7c55e7ce893ba0511074c4ec930d9f3bcb9615d8e9d5.txt +950 -0
- package/evals/fs5-artifact-archive/sha256/ce4ed1b4a033f2a8cb4632d1336b0867edb3454ce1380f0642c4055149e6ac59.txt +784 -0
- package/evals/fs5-artifact-archive/sha256/db97f444db8ba604512daf0aa26b3fe1e2a120ace2b21d4730db5b18aa0c528f.txt +950 -0
- package/evals/fs5-causal-arm.v1.json +9 -5
- package/evals/fs5-pilot-protocol.v1.json +1 -1
- package/evals/fs5-pilot-protocol.v4.json +4 -4
- package/evals/fs5-pilot-protocol.v5.json +6 -5
- package/evals/harness-next/development-corpus.mjs +106 -0
- package/evals/harness-next/resource-budget-probe.mjs +38 -0
- package/evals/harness-next/run-calibration.mjs +83 -0
- package/evals/harness-next/run-resource-probe.mjs +55 -0
- package/evals/harness-next/selected-development-corpus.mjs +27 -0
- package/evals/harness-next.v1.json +51 -0
- package/evals/long-horizon-v1/README.md +19 -0
- package/evals/long-horizon-v1/runner.mjs +116 -48
- package/evals/long-horizon-v1/worker.mjs +77 -30
- package/evals/private-holdout-v1/CURRENT_BOUNDARY.md +53 -0
- package/evals/private-holdout-v1/README.md +5 -2
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +2 -2
- package/packages/piagent-core/benchmark/benchmark-all-attempt-efficiency.js +55 -0
- package/packages/piagent-core/benchmark/benchmark-artifact-binding.js +49 -0
- package/packages/piagent-core/benchmark/benchmark-bootstrap.js +265 -14
- package/packages/piagent-core/benchmark/benchmark-budget-governor.js +271 -0
- package/packages/piagent-core/benchmark/benchmark-campaign.js +592 -0
- package/packages/piagent-core/benchmark/benchmark-candidate.js +8 -3
- package/packages/piagent-core/benchmark/benchmark-checkpoint.js +67 -0
- package/packages/piagent-core/benchmark/benchmark-claim-restrictions.js +269 -2
- package/packages/piagent-core/benchmark/benchmark-cli.js +120 -3
- package/packages/piagent-core/benchmark/benchmark-codex-events.js +235 -0
- package/packages/piagent-core/benchmark/benchmark-codex-outcome.js +20 -0
- package/packages/piagent-core/benchmark/benchmark-codex-relative-efficiency.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-codex-rollout.js +521 -0
- package/packages/piagent-core/benchmark/benchmark-codex.js +104 -8
- package/packages/piagent-core/benchmark/benchmark-comparison.js +284 -1
- package/packages/piagent-core/benchmark/benchmark-core.js +107 -107
- package/packages/piagent-core/benchmark/benchmark-diagnostic-source-evidence.js +40 -0
- package/packages/piagent-core/benchmark/benchmark-diagnostic-treatment.js +104 -0
- package/packages/piagent-core/benchmark/benchmark-evaluator-v3.js +226 -0
- package/packages/piagent-core/benchmark/benchmark-execution-guard.js +23 -1
- package/packages/piagent-core/benchmark/benchmark-forensics.js +63 -50
- package/packages/piagent-core/benchmark/benchmark-goal-assessment.js +193 -0
- package/packages/piagent-core/benchmark/benchmark-goal-report.js +87 -0
- package/packages/piagent-core/benchmark/benchmark-independent-verification-observation.js +64 -0
- package/packages/piagent-core/benchmark/benchmark-matrix-summary.js +141 -0
- package/packages/piagent-core/benchmark/benchmark-matrix.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-normalized-cost.js +62 -0
- package/packages/piagent-core/benchmark/benchmark-pi-home.js +1 -0
- package/packages/piagent-core/benchmark/benchmark-preflight.js +69 -8
- package/packages/piagent-core/benchmark/benchmark-process.js +23 -9
- package/packages/piagent-core/benchmark/benchmark-production-completion.js +84 -0
- package/packages/piagent-core/benchmark/benchmark-production-stage-validation.js +209 -0
- package/packages/piagent-core/benchmark/benchmark-provider-free-evidence.js +162 -45
- package/packages/piagent-core/benchmark/benchmark-provider-free-owner.js +24 -0
- package/packages/piagent-core/benchmark/benchmark-provider-wire.js +378 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.d.ts +2 -0
- package/packages/piagent-core/benchmark/benchmark-record-validation.js +269 -3
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +79 -4
- package/packages/piagent-core/benchmark/benchmark-resume-recovery.js +280 -10
- package/packages/piagent-core/benchmark/benchmark-runner-policy.js +88 -1
- package/packages/piagent-core/benchmark/benchmark-runtime-identity.js +50 -1
- package/packages/piagent-core/benchmark/benchmark-runtime.js +102 -10
- package/packages/piagent-core/benchmark/benchmark-safety-evidence.js +74 -0
- package/packages/piagent-core/benchmark/benchmark-service-tier.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-stage-diagnostic.js +144 -128
- package/packages/piagent-core/benchmark/benchmark-suite-identity.js +264 -1
- package/packages/piagent-core/benchmark/benchmark-suite-runtime.js +4 -0
- package/packages/piagent-core/benchmark/benchmark-suite.js +211 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +30 -2
- package/packages/piagent-core/benchmark/benchmark-timing-contract.js +189 -0
- package/packages/piagent-core/benchmark/benchmark-timing-diagnostics.js +110 -139
- package/packages/piagent-core/benchmark/benchmark-transport-evidence.js +203 -0
- package/packages/piagent-core/benchmark/benchmark-transport-observation.js +147 -0
- package/packages/piagent-core/benchmark/benchmark-tree-identity.js +2 -1
- package/packages/piagent-core/benchmark/benchmark-usage.js +395 -33
- package/packages/piagent-core/benchmark/benchmark-workflow.js +136 -38
- package/packages/piagent-core/benchmark/fs5-pilot-protocol.js +21 -2
- package/packages/piagent-core/capabilities/runtime-integrity.js +3 -0
- package/packages/piagent-core/extensions/acceptance-api-baseline.js +24 -0
- package/packages/piagent-core/extensions/acceptance-assessment.js +153 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.d.ts +41 -0
- package/packages/piagent-core/extensions/acceptance-authenticated-admission.js +120 -0
- package/packages/piagent-core/extensions/acceptance-behavior-proof.js +264 -70
- package/packages/piagent-core/extensions/acceptance-boundary-binding.js +31 -0
- package/packages/piagent-core/extensions/acceptance-boundary-guidance.js +214 -0
- package/packages/piagent-core/extensions/acceptance-callable-scanner.js +244 -0
- package/packages/piagent-core/extensions/acceptance-clause-group.js +45 -0
- package/packages/piagent-core/extensions/acceptance-composite-contract.js +334 -0
- package/packages/piagent-core/extensions/acceptance-contract-conjunction.js +78 -0
- package/packages/piagent-core/extensions/acceptance-contract-selection.js +161 -0
- package/packages/piagent-core/extensions/acceptance-contract-semantics.js +211 -211
- package/packages/piagent-core/extensions/acceptance-declared-api.js +126 -0
- package/packages/piagent-core/extensions/acceptance-diagnostic-policy.ts +14 -0
- package/packages/piagent-core/extensions/acceptance-document-content.js +73 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.d.ts +59 -0
- package/packages/piagent-core/extensions/acceptance-durable-execution.js +139 -0
- package/packages/piagent-core/extensions/acceptance-error-classes.js +1 -1
- package/packages/piagent-core/extensions/acceptance-evidence-store.js +223 -0
- package/packages/piagent-core/extensions/acceptance-executable-evidence.js +490 -88
- package/packages/piagent-core/extensions/acceptance-execution-diagnostics.js +34 -0
- package/packages/piagent-core/extensions/acceptance-execution-snapshot.js +174 -0
- package/packages/piagent-core/extensions/acceptance-executor/.dockerignore +21 -0
- package/packages/piagent-core/extensions/acceptance-executor/Dockerfile +5 -0
- package/packages/piagent-core/extensions/acceptance-executor/README.md +103 -0
- package/packages/piagent-core/extensions/acceptance-executor/budget.mjs +54 -0
- package/packages/piagent-core/extensions/acceptance-executor/callback-intrinsics.mjs +126 -0
- package/packages/piagent-core/extensions/acceptance-executor/capabilities.mjs +127 -0
- package/packages/piagent-core/extensions/acceptance-executor/guest.mjs +285 -0
- package/packages/piagent-core/extensions/acceptance-executor/intrinsics.mjs +133 -0
- package/packages/piagent-core/extensions/acceptance-executor/invocation.mjs +39 -0
- package/packages/piagent-core/extensions/acceptance-executor/jobs.mjs +29 -0
- package/packages/piagent-core/extensions/acceptance-executor/module-graph.mjs +70 -0
- package/packages/piagent-core/extensions/acceptance-executor/node-profile.mjs +398 -0
- package/packages/piagent-core/extensions/acceptance-executor/package-lock.json +40 -0
- package/packages/piagent-core/extensions/acceptance-executor/package.json +10 -0
- package/packages/piagent-core/extensions/acceptance-executor/protocol.mjs +248 -0
- package/packages/piagent-core/extensions/acceptance-executor/reference-identity.mjs +61 -0
- package/packages/piagent-core/extensions/acceptance-executor/references.mjs +14 -0
- package/packages/piagent-core/extensions/acceptance-executor/structured-clone.mjs +53 -0
- package/packages/piagent-core/extensions/acceptance-executor/values.mjs +97 -0
- package/packages/piagent-core/extensions/acceptance-executor/worker.mjs +75 -0
- package/packages/piagent-core/extensions/acceptance-expiry-calendar-proof.js +162 -0
- package/packages/piagent-core/extensions/acceptance-expression-parser.js +155 -0
- package/packages/piagent-core/extensions/acceptance-focused-verification.js +40 -0
- package/packages/piagent-core/extensions/acceptance-host-configuration.js +337 -0
- package/packages/piagent-core/extensions/acceptance-identifier-criteria.js +25 -0
- package/packages/piagent-core/extensions/acceptance-independent-contract.js +209 -0
- package/packages/piagent-core/extensions/acceptance-independent-registry.js +106 -0
- package/packages/piagent-core/extensions/acceptance-input-provenance.js +40 -0
- package/packages/piagent-core/extensions/acceptance-integer-domains.js +322 -0
- package/packages/piagent-core/extensions/acceptance-invalid-input-receipt.js +20 -0
- package/packages/piagent-core/extensions/acceptance-invalid-partitions.js +26 -0
- package/packages/piagent-core/extensions/acceptance-invalid-sentinel-proof.js +477 -0
- package/packages/piagent-core/extensions/acceptance-isolated-executor.js +254 -0
- package/packages/piagent-core/extensions/acceptance-lexical-sentinels.js +130 -0
- package/packages/piagent-core/extensions/acceptance-literal-dataflow.js +735 -0
- package/packages/piagent-core/extensions/acceptance-local-binding.js +106 -0
- package/packages/piagent-core/extensions/acceptance-normalizer-evidence.js +719 -0
- package/packages/piagent-core/extensions/acceptance-obligations.js +75 -0
- package/packages/piagent-core/extensions/acceptance-parameter-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-precedence-contract.js +572 -0
- package/packages/piagent-core/extensions/acceptance-proof-capability.js +26 -0
- package/packages/piagent-core/extensions/acceptance-receipt.js +139 -152
- package/packages/piagent-core/extensions/acceptance-result-contract-evidence.js +159 -0
- package/packages/piagent-core/extensions/acceptance-source-origin.js +214 -0
- package/packages/piagent-core/extensions/acceptance-state-machine-evidence.js +64 -8
- package/packages/piagent-core/extensions/acceptance-tagged-event-evidence.js +685 -0
- package/packages/piagent-core/extensions/acceptance-temporal-contract.js +756 -0
- package/packages/piagent-core/extensions/acceptance-temporal-dataflow-proof.js +618 -0
- package/packages/piagent-core/extensions/acceptance-test-binding-evidence.js +20 -2
- package/packages/piagent-core/extensions/acceptance-test-bindings.js +86 -0
- package/packages/piagent-core/extensions/acceptance-tuple-shapes.js +12 -0
- package/packages/piagent-core/extensions/context-engine.js +11 -6
- package/packages/piagent-core/extensions/criterion-context-pack.js +96 -4
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/guard-types.ts +5 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +197 -155
- package/packages/piagent-core/extensions/readonly-inline-inspection.ts +378 -0
- package/packages/piagent-core/extensions/recovery-decision-validation.ts +92 -0
- package/packages/piagent-core/extensions/shell-write-targets.js +15 -14
- package/packages/piagent-core/extensions/task-contract-view.js +4 -4
- package/packages/piagent-core/extensions/task-digest-migration.js +2 -2
- package/packages/piagent-core/extensions/task-journal.js +48 -3
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +74 -38
- package/packages/piagent-core/extensions/task-terminal-disposition.js +10 -0
- package/packages/piagent-core/extensions/trajectory-contract.ts +5 -0
- package/packages/piagent-core/extensions/verification-intelligence.js +60 -8
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-revision.js +48 -0
- package/packages/piagent-core/extensions/workspace-verification-snapshot.js +13 -0
- package/packages/piagent-core/package.json +3 -3
- package/packages/piagent-core/policies/base-policy.json +1 -0
- package/packages/piagent-core/runtime/context/adaptive-planner.ts +7 -3
- package/packages/piagent-core/runtime/hooks/agent-start-hook.ts +60 -59
- package/packages/piagent-core/runtime/hooks/agent-start-types.ts +48 -0
- package/packages/piagent-core/runtime/hooks/completion-hook.ts +76 -83
- package/packages/piagent-core/runtime/hooks/input-hook.ts +92 -7
- package/packages/piagent-core/runtime/hooks/protected-refusal-completion.ts +86 -0
- package/packages/piagent-core/runtime/hooks/session-hooks.ts +5 -2
- package/packages/piagent-core/runtime/hooks/session-start-hook.ts +50 -18
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +53 -20
- package/packages/piagent-core/runtime/hooks/tool-result-metadata.ts +9 -0
- package/packages/piagent-core/runtime/hooks/tool-result-navigation.ts +36 -0
- package/packages/piagent-core/runtime/hooks/tool-result-retrieval.ts +42 -0
- package/packages/piagent-core/runtime/input/chat-images.ts +1 -1
- package/packages/piagent-core/runtime/inspection/activity-event-adapter.ts +1 -1
- package/packages/piagent-core/runtime/inspection/same-process-bridge-proof.ts +1 -1
- package/packages/piagent-core/runtime/inspection/task-handoff-history.ts +2 -2
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +49 -8
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +61 -15
- package/packages/piagent-core/runtime/inspection/workspace-file-reader.ts +2 -132
- package/packages/piagent-core/runtime/model/model-route-policy.ts +33 -13
- package/packages/piagent-core/runtime/model/model-route-types.ts +10 -6
- package/packages/piagent-core/runtime/model/provider-wire-fingerprint.ts +20 -0
- package/packages/piagent-core/runtime/model/service-tier-runtime.ts +242 -0
- package/packages/piagent-core/runtime/policy/source-worktree-transaction.ts +75 -14
- package/packages/piagent-core/runtime/product/activity-inspector.ts +6 -3
- package/packages/piagent-core/runtime/product/edit-verifier-timing.ts +26 -0
- package/packages/piagent-core/runtime/product/efficiency-metrics.ts +12 -7
- package/packages/piagent-core/runtime/product/operator-projections.ts +8 -3
- package/packages/piagent-core/runtime/product/task-usage-hooks.ts +78 -0
- package/packages/piagent-core/runtime/product/task-usage.ts +104 -0
- package/packages/piagent-core/runtime/quality/protected-refusal-handoff.ts +35 -0
- package/packages/piagent-core/runtime/recovery/acceptance-recovery-guidance.ts +21 -0
- package/packages/piagent-core/runtime/recovery/continuation-budget.ts +11 -2
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +192 -57
- package/packages/piagent-core/runtime/recovery/handoff-verifier-binding.ts +24 -0
- package/packages/piagent-core/runtime/recovery/independent-verification-recovery.ts +79 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +27 -39
- package/packages/piagent-core/runtime/recovery/resume-state.ts +8 -5
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +12 -8
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/policy-tools.ts +5 -3
- package/packages/piagent-core/runtime/registration/session-commands.ts +9 -7
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +27 -11
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +5 -2
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +23 -23
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -22
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +29 -35
- package/packages/piagent-core/runtime/session/system-prompt.ts +3 -3
- package/packages/piagent-core/runtime/session/uncertain-send-continuation.ts +231 -0
- package/packages/piagent-core/runtime/session/usage.ts +4 -3
- package/packages/piagent-core/runtime/tools/apply-patch-tool.ts +6 -6
- package/packages/piagent-core/runtime/tools/phase-tool-runtime.ts +1 -1
- package/packages/piagent-core/runtime/tools/phase-tools.ts +3 -2
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +10 -3
- package/packages/piagent-core/runtime/trajectory/trajectory-runtime.ts +192 -5
- package/packages/piagent-core/runtime/trajectory/trajectory-types.ts +3 -2
- package/packages/piagent-core/runtime/verification/acceptance-api-baseline.js +488 -0
- package/packages/piagent-core/runtime/verification/completion-preparation.ts +22 -0
- package/packages/piagent-core/runtime/verification/composite-completion-finalizer.ts +105 -0
- package/packages/piagent-core/runtime/verification/composite-journal-anchor.ts +181 -0
- package/packages/piagent-core/runtime/verification/composite-runtime-binding.ts +76 -0
- package/packages/piagent-core/runtime/verification/composite-scoped-mediation.ts +318 -0
- package/packages/piagent-core/runtime/verification/composite-session-persistence.ts +175 -0
- package/packages/piagent-core/runtime/verification/composite-task-publication.ts +273 -0
- package/packages/piagent-core/runtime/verification/exact-verifier-reuse.ts +11 -11
- package/packages/piagent-core/runtime/verification/host-project-verification.ts +71 -0
- package/packages/piagent-core/runtime/verification/independent-acceptance-runtime.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-assurance-facts.ts +500 -0
- package/packages/piagent-core/runtime/verification/runtime-contract-runner.ts +107 -0
- package/packages/piagent-core/runtime/verification/shell-verification-snapshots.ts +56 -0
- package/packages/piagent-core/runtime/workflows/change-clarification.ts +61 -0
- package/packages/piagent-core/runtime/workflows/input-routing.ts +21 -14
- package/packages/piagent-core/runtime/workflows/runtime-intake-compaction.ts +44 -7
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +221 -242
- package/packages/piagent-core/runtime/workflows/task-repair-intent.ts +67 -0
- package/packages/piagent-core/runtime/workflows/task-scope-resolution.ts +83 -0
- package/packages/piagent-core/runtime/workflows/webui-runtime-command.ts +5 -2
- package/packages/piagent-core/runtime/workflows/webui-workflow.ts +39 -32
- package/packages/piagent-core/runtime/workflows/workflow-follow-up.ts +33 -0
- package/packages/piagent-core/security/workspace-file-reader.ts +132 -0
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-B4XehTCU.js → MarkdownMessage-DMU7Xe_O.js} +1 -1
- package/packages/piagent-webui/dist/client/assets/index-Cui9AAV_.js +176 -0
- package/packages/piagent-webui/dist/client/assets/{index-BV-_03yl.css → index-DVahUVWy.css} +1 -1
- package/packages/piagent-webui/dist/client/index.html +2 -2
- package/packages/piagent-webui/extension/dispatch-user-entry.ts +13 -0
- package/packages/piagent-webui/extension/piagent-webui.ts +10 -2
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +6 -7
- package/packages/piagent-webui/extension/session-stream-adapter.ts +60 -11
- package/packages/piagent-webui/extension/source-mutation-controller.ts +29 -3
- package/packages/piagent-webui/extension/source-open-controller.ts +29 -7
- package/packages/piagent-webui/extension/source-revert-controller.ts +26 -3
- package/packages/piagent-webui/gateway/gateway-events.ts +22 -5
- package/packages/piagent-webui/gateway/gateway-service.ts +7 -2
- package/packages/piagent-webui/gateway/gateway-session-stream.ts +141 -19
- package/packages/piagent-webui/gateway/rpc-ui-context.ts +34 -9
- package/packages/piagent-webui/gateway/runtime-command-controller.ts +46 -6
- package/packages/piagent-webui/gateway/runtime-session-controls.ts +17 -2
- package/packages/piagent-webui/gateway/session-catalog.ts +9 -3
- package/packages/piagent-webui/gateway/session-command-controller.ts +29 -9
- package/packages/piagent-webui/gateway/session-effective-options.ts +2 -9
- package/packages/piagent-webui/gateway/session-inspection-registry.ts +14 -6
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-prompt-dispatch.ts +37 -0
- package/packages/piagent-webui/gateway/session-runtime-close.ts +36 -0
- package/packages/piagent-webui/gateway/session-runtime-factory.ts +183 -6
- package/packages/piagent-webui/gateway/session-runtime-ownership.ts +48 -0
- package/packages/piagent-webui/gateway/session-runtime-state.ts +24 -0
- package/packages/piagent-webui/gateway/session-runtime-supervisor.ts +114 -114
- package/packages/piagent-webui/server/core-inspection-provider.ts +73 -14
- package/packages/piagent-webui/server/loopback-server.ts +9 -7
- package/packages/piagent-webui/server/terminal-delivery-receipt.ts +144 -0
- package/packages/piagent-webui/server/transcript-projection.ts +65 -12
- package/packages/piagent-webui/shared/message-correlation.ts +25 -0
- package/schemas/approved-host-contracts.schema.json +973 -0
- package/schemas/benchmark-attempt-outcome.schema.json +303 -0
- package/schemas/benchmark-independent-verification-plan.schema.json +34 -0
- package/schemas/composite-criterion.schema.json +788 -0
- package/schemas/contract-family-library.schema.json +28 -0
- package/schemas/contract-selection-recipe.schema.json +31 -0
- package/schemas/failure-evidence.schema.json +2 -2
- package/schemas/helper-request.schema.json +16 -4
- package/schemas/host-contract-plan.schema.json +25 -0
- package/schemas/piagent-webui/gateway-protocol-v1.schema.json +16 -2
- package/schemas/piagent-webui/runtime-command-v1.schema.json +26 -3
- package/schemas/piagent-webui/runtime-event-v2.schema.json +32 -1
- package/schemas/piagent-webui/session-live-state-v1.schema.json +2 -0
- package/schemas/piagent-webui/snapshot-v1.schema.json +3 -0
- package/schemas/piagent-webui/source-change-v1.schema.json +12 -0
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/public-contracts/codex-mcp-metadata-v1.schema.json +206 -0
- package/schemas/public-contracts/codex-mcp-metadata-v2.schema.json +228 -0
- package/schemas/public-contracts/codex-mcp-metadata-v3.schema.json +233 -0
- package/schemas/public-contracts/revocation-v1.schema.json +87 -0
- package/schemas/public-contracts/subscription-contract-v1.schema.json +76 -0
- package/schemas/scoped-project-verification-receipt-v2.schema.json +83 -0
- package/schemas/scoped-project-verification-receipt-v3.schema.json +295 -0
- package/schemas/scoped-verification-receipt-v1.schema.json +76 -0
- package/schemas/task-contract.schema.json +30 -0
- package/scripts/approve-independent-verification.mjs +57 -0
- package/scripts/benchmark-arm-observer.mjs +320 -0
- package/scripts/benchmark-budget-ipc.mjs +34 -0
- package/scripts/benchmark-budget-launcher.mjs +91 -0
- package/scripts/benchmark-budget-processes.mjs +133 -0
- package/scripts/benchmark-budget-publication.mjs +228 -0
- package/scripts/benchmark-budget-runtime.mjs +136 -0
- package/scripts/benchmark-candidate-ablation.mjs +265 -0
- package/scripts/benchmark-codex-journey.mjs +406 -0
- package/scripts/benchmark-independent-verification.mjs +628 -0
- package/scripts/benchmark-journey-outcome.mjs +92 -0
- package/scripts/benchmark-production-v3-human-review.mjs +192 -0
- package/scripts/benchmark-public-input-catalog.mjs +158 -0
- package/scripts/benchmark-registered-plan-author.mjs +355 -0
- package/scripts/benchmark-registered-plan-recipe-support.mjs +85 -0
- package/scripts/benchmark-registered-plan-recipes-backend.mjs +165 -0
- package/scripts/benchmark-registered-plan-recipes-data.mjs +164 -0
- package/scripts/benchmark-registered-plan-recipes-platform.mjs +165 -0
- package/scripts/benchmark-runner-completion.mjs +133 -0
- package/scripts/benchmark-runner-configuration.mjs +159 -0
- package/scripts/benchmark-runner-core.mjs +217 -388
- package/scripts/benchmark-runner-finalization.mjs +140 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-model-scope.sh +6 -2
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/install-global.sh +2 -2
- package/scripts/model-route-evaluation-core.ts +1 -0
- package/scripts/phase-tools-evaluation.mjs +1 -0
- package/scripts/pi-auto.sh +7 -1
- package/scripts/pi-model-route.mjs +1 -1
- package/scripts/pi-usage-history.mjs +201 -11
- package/scripts/piagent-cli.mjs +8 -0
- package/scripts/private-holdout-readiness.mjs +7 -1
- package/scripts/public-evaluation-exposure.mjs +140 -0
- package/scripts/select-independent-verification.mjs +62 -0
- package/scripts/setup.sh +2 -2
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +37 -56
- package/scripts/verify-release-identity.mjs +5 -3
- package/templates/global/settings.json +5 -1
- package/packages/piagent-webui/dist/client/assets/index-CgOEyRPG.js +0 -176
|
@@ -23,7 +23,13 @@ export function boundRejectionTestEvidence(input) {
|
|
|
23
23
|
|| input.requestedErrors.some((name) => assertion.errorClasses.includes(name));
|
|
24
24
|
const bindingGroups = input.bindings.map((binding) => {
|
|
25
25
|
const allAssertions = (binding.testBindings ?? []).flatMap(assertionsForBinding);
|
|
26
|
-
|
|
26
|
+
const assertions = allAssertions.filter(requestedError);
|
|
27
|
+
return {
|
|
28
|
+
binding, allAssertions, assertions,
|
|
29
|
+
invalidArgumentIndices: [...new Set(assertions.flatMap((item) => item.invalidArgumentIndices ?? []))]
|
|
30
|
+
.sort((left, right) => left - right),
|
|
31
|
+
invalidArgumentPartitions: assertions.flatMap((item) => item.invalidArgumentPartitions ?? [])
|
|
32
|
+
};
|
|
27
33
|
});
|
|
28
34
|
const testCodes = input.testCodeEntries.length > 0
|
|
29
35
|
? input.testCodeEntries.map((entry) => entry.code) : [input.fallbackTestCode];
|
|
@@ -56,5 +62,17 @@ export function boundRejectionTestEvidence(input) {
|
|
|
56
62
|
const matches = [...input.bodyMaps.values()].filter((bodies) => bodies.has(target));
|
|
57
63
|
return assertionModesMatch(items, [`*.${target}`], matches.length === 1 ? callableAssertionMode(matches[0], target) : undefined);
|
|
58
64
|
});
|
|
59
|
-
return {
|
|
65
|
+
return {
|
|
66
|
+
assertions, mappingOk, modeOk, targetOk,
|
|
67
|
+
bindingArgumentRequirements: bindingGroups.map(({ binding, invalidArgumentIndices, invalidArgumentPartitions }) => ({
|
|
68
|
+
sourcePath: binding.sourcePath,
|
|
69
|
+
sourceName: binding.sourceName,
|
|
70
|
+
invalidArgumentIndices,
|
|
71
|
+
invalidArgumentPartitions: [...new Set(invalidArgumentPartitions.map((item) => item.index))].map((index) => ({
|
|
72
|
+
index,
|
|
73
|
+
partitions: [...new Set(invalidArgumentPartitions.filter((item) => item.index === index)
|
|
74
|
+
.flatMap((item) => item.partitions ?? []))].sort()
|
|
75
|
+
}))
|
|
76
|
+
}))
|
|
77
|
+
};
|
|
60
78
|
}
|
|
@@ -0,0 +1,86 @@
|
|
|
1
|
+
function escapeRegex(value) {
|
|
2
|
+
return String(value ?? "").replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
|
|
3
|
+
}
|
|
4
|
+
|
|
5
|
+
export function assertionCarrierBindings(testText) {
|
|
6
|
+
const bindings = [];
|
|
7
|
+
const add = (name, start, end) => {
|
|
8
|
+
if (name) bindings.push({ name: name.toLowerCase(), start, end });
|
|
9
|
+
};
|
|
10
|
+
for (const match of testText.matchAll(/\bimport\s+([a-z_$][a-z0-9_$]*)\s+from\s+__pi_node_assert_module_literal__\s*;?/gi)) {
|
|
11
|
+
add(match[1], match.index, match.index + match[0].length);
|
|
12
|
+
}
|
|
13
|
+
for (const match of testText.matchAll(/\bimport\s+\*\s+as\s+([a-z_$][a-z0-9_$]*)\s+from\s+__pi_node_assert_module_literal__\s*;?/gi)) {
|
|
14
|
+
add(match[1], match.index, match.index + match[0].length);
|
|
15
|
+
}
|
|
16
|
+
for (const match of testText.matchAll(/\bimport\s*\{([^}]{1,500})\}\s*from\s+__pi_node_assert_module_literal__\s*;?/gi)) {
|
|
17
|
+
for (const item of match[1].split(",")) {
|
|
18
|
+
const strict = item.trim().match(/^strict(?:\s+as\s+([a-z_$][a-z0-9_$]*))?$/i);
|
|
19
|
+
if (strict) add(strict[1] ?? "strict", match.index, match.index + match[0].length);
|
|
20
|
+
}
|
|
21
|
+
}
|
|
22
|
+
for (const match of testText.matchAll(/\bconst\s+([a-z_$][a-z0-9_$]*)\s*=\s*require\s*\(\s*__pi_node_assert_module_literal__\s*\)(?:\s*\.\s*strict)?\s*;?/gi)) {
|
|
23
|
+
add(match[1], match.index, match.index + match[0].length);
|
|
24
|
+
}
|
|
25
|
+
for (const match of testText.matchAll(/\bconst\s*\{\s*strict\s*:\s*([a-z_$][a-z0-9_$]*)\s*\}\s*=\s*require\s*\(\s*__pi_node_assert_module_literal__\s*\)\s*;?/gi)) {
|
|
26
|
+
add(match[1], match.index, match.index + match[0].length);
|
|
27
|
+
}
|
|
28
|
+
return bindings.slice(0, 8);
|
|
29
|
+
}
|
|
30
|
+
|
|
31
|
+
export const TEST_RUNNER_EXPORT_KINDS = new Map([
|
|
32
|
+
["test", "test"], ["it", "test"], ["describe", "suite"], ["suite", "suite"]
|
|
33
|
+
]);
|
|
34
|
+
|
|
35
|
+
export function testRunnerBindings(testText) {
|
|
36
|
+
const bindings = [];
|
|
37
|
+
const add = (name, kind, start, end) => {
|
|
38
|
+
if (name && kind) bindings.push({ name: name.toLowerCase(), kind, start, end });
|
|
39
|
+
};
|
|
40
|
+
for (const match of testText.matchAll(/\bimport\s+([a-z_$][a-z0-9_$]*)(?:\s*,\s*\{[^}]{1,500}\})?\s+from\s+__pi_node_test_module_literal__\s*;?/gi)) {
|
|
41
|
+
add(match[1], "test", match.index, match.index + match[0].length);
|
|
42
|
+
}
|
|
43
|
+
for (const match of testText.matchAll(/\bimport\s+(?:[a-z_$][a-z0-9_$]*\s*,\s*)?\{([^}]{1,500})\}\s+from\s+__pi_node_test_module_literal__\s*;?/gi)) {
|
|
44
|
+
for (const raw of match[1].split(",")) {
|
|
45
|
+
const item = raw.trim().match(/^(test|it|describe|suite)(?:\s+as\s+([a-z_$][a-z0-9_$]*))?$/i);
|
|
46
|
+
if (item) add(item[2] ?? item[1], TEST_RUNNER_EXPORT_KINDS.get(item[1].toLowerCase()), match.index, match.index + match[0].length);
|
|
47
|
+
}
|
|
48
|
+
}
|
|
49
|
+
for (const match of testText.matchAll(/\bconst\s+([a-z_$][a-z0-9_$]*)\s*=\s*require\s*\(\s*__pi_node_test_module_literal__\s*\)\s*;?/gi)) {
|
|
50
|
+
add(match[1], "test", match.index, match.index + match[0].length);
|
|
51
|
+
}
|
|
52
|
+
for (const match of testText.matchAll(/\bconst\s+([a-z_$][a-z0-9_$]*)\s*=\s*require\s*\(\s*__pi_node_test_module_literal__\s*\)\s*\.\s*(test|it|describe|suite)\s*;?/gi)) {
|
|
53
|
+
add(match[1], TEST_RUNNER_EXPORT_KINDS.get(match[2].toLowerCase()), match.index, match.index + match[0].length);
|
|
54
|
+
}
|
|
55
|
+
for (const match of testText.matchAll(/\bconst\s*\{([^}]{1,500})\}\s*=\s*require\s*\(\s*__pi_node_test_module_literal__\s*\)\s*;?/gi)) {
|
|
56
|
+
for (const raw of match[1].split(",")) {
|
|
57
|
+
const item = raw.trim().match(/^(test|it|describe|suite)(?:\s*:\s*([a-z_$][a-z0-9_$]*))?$/i);
|
|
58
|
+
if (item) add(item[2] ?? item[1], TEST_RUNNER_EXPORT_KINDS.get(item[1].toLowerCase()), match.index, match.index + match[0].length);
|
|
59
|
+
}
|
|
60
|
+
}
|
|
61
|
+
return bindings.slice(0, 16);
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
export function importedCallableBindingIsStable(testText, binding, bindings) {
|
|
65
|
+
const escaped = escapeRegex(binding.name);
|
|
66
|
+
const withoutDeclaration = `${testText.slice(0, binding.start)}${" ".repeat(binding.end - binding.start)}${testText.slice(binding.end)}`;
|
|
67
|
+
if (bindings.filter((candidate) => candidate.name === binding.name).length !== 1) return false;
|
|
68
|
+
if (new RegExp(`\\b(?:const|let|var|function|class)\\s+${escaped}\\b|\\bfunction\\b[^({]{0,200}\\([^)]*\\b${escaped}\\b|\\bcatch\\s*\\([^)]*\\b${escaped}\\b|(?:\\([^)]*\\b${escaped}\\b[^)]*\\)|\\b${escaped})\\s*=>`, "i").test(withoutDeclaration)) return false;
|
|
69
|
+
if (new RegExp(`(?<![.$a-z0-9_])${escaped}\\b\\s*(?:(?<![=!<>])=(?!=|>)|\\+\\+|--|[+*/%&|^-]=)|(?:\\+\\+|--)\\s*\\b${escaped}\\b`, "i").test(withoutDeclaration)) return false;
|
|
70
|
+
if (new RegExp(`\\b(?:const|let|var)\\s+[a-z_$][a-z0-9_$]*\\s*=\\s*${escaped}\\b|\\b(?:const|let|var)\\s*\\{[^}\\n]{0,500}\\}\\s*=\\s*${escaped}\\b`, "i").test(withoutDeclaration)) return false;
|
|
71
|
+
if (new RegExp(`(?<![.$a-z0-9_])[a-z_$][a-z0-9_$]*\\s*(?<![=!<>])=(?!=|>)\\s*${escaped}\\b`, "i").test(withoutDeclaration)) return false;
|
|
72
|
+
const member = `${escaped}\\s*(?:\\.\\s*[a-z_$][a-z0-9_$]*|\\[[^\\]]{1,300}\\])`;
|
|
73
|
+
if (new RegExp(`\\b${member}\\s*(?:=|\\+\\+|--|[+*/%&|^-]=|(?:&&|\\|\\||\\?\\?)=)|(?:\\+\\+|--)\\s*\\b${member}|\\bdelete\\s+${escaped}\\s*(?:\\.|\\[)|\\b${escaped}\\s*\\.\\s*__(?:definegetter|definesetter)__\\s*\\(`, "i").test(withoutDeclaration)) return false;
|
|
74
|
+
if (new RegExp(`(?:\\{[^}\\n]{0,500}\\b${member}[^}\\n]{0,500}\\}|\\[[^\\]\\n]{0,500}\\b${member}[^\\]\\n]{0,500}\\])\\s*=|\\bfor(?:\\s+await)?\\s*\\(\\s*(?:${member}|\\{[^}\\n]{0,500}\\b${member}[^}\\n]{0,500}\\}|\\[[^\\]\\n]{0,500}\\b${member}[^\\]\\n]{0,500}\\])\\s+(?:of|in)\\b`, "i").test(withoutDeclaration)) return false;
|
|
75
|
+
if (new RegExp(`\\b(?:object|reflect)\\s*\\.\\s*[a-z_$][a-z0-9_$]*\\s*\\(\\s*${escaped}\\s*[,)]|\\b[a-z_$][a-z0-9_$.]*\\s*\\(\\s*${escaped}\\s*[,)]`, "i").test(withoutDeclaration)) return false;
|
|
76
|
+
for (const match of withoutDeclaration.matchAll(new RegExp(`\\b${escaped}\\b`, "gi"))) {
|
|
77
|
+
const suffix = withoutDeclaration.slice(match.index + match[0].length);
|
|
78
|
+
if (/^\s*\(/.test(suffix) || /^\s*\.\s*[a-z_$][a-z0-9_$]*\s*\(/i.test(suffix)) continue;
|
|
79
|
+
return false;
|
|
80
|
+
}
|
|
81
|
+
return true;
|
|
82
|
+
}
|
|
83
|
+
|
|
84
|
+
export function assertionCarrierIsStable(testText, binding) {
|
|
85
|
+
return importedCallableBindingIsStable(testText, binding, assertionCarrierBindings(testText));
|
|
86
|
+
}
|
|
@@ -0,0 +1,12 @@
|
|
|
1
|
+
// Closed literal shapes used by the tuple evidence reader. These inspect
|
|
2
|
+
// already-sanitized text and parsed argument ranges; they never run test code.
|
|
3
|
+
export const INVALID_DATE_CONSTRUCTION = /^new\s+date\s*\(\s*(?:undefined|-?(?:number\.)?(?:nan|positive_infinity|negative_infinity|infinity)|__pi_(?:empty|whitespace|unparseable_date)_string_literal__)\s*\)$/i;
|
|
4
|
+
|
|
5
|
+
export function tupleRegistrationIsClosed(args) {
|
|
6
|
+
if (!/^(?:async\s+)?\(\s*\)\s*=>/.test(args.at(-1)?.text ?? "")) return false;
|
|
7
|
+
if (args.length === 2) return true;
|
|
8
|
+
if (args.length !== 3 || !/^\{[^{}]*\}$/.test(args[1].text)) return false;
|
|
9
|
+
const fields = args[1].text.slice(1, -1).split(",").map((value) => value.trim());
|
|
10
|
+
if (fields.at(-1) === "") fields.pop();
|
|
11
|
+
return new Set(fields).size === fields.length && fields.every((field) => /^(?:skip|todo)\s*:\s*false$/.test(field));
|
|
12
|
+
}
|
|
@@ -2,6 +2,7 @@ import crypto from "node:crypto";
|
|
|
2
2
|
import fs from "node:fs";
|
|
3
3
|
import path from "node:path";
|
|
4
4
|
import { execFileSync } from "node:child_process";
|
|
5
|
+
import { resolveWorkflowId, workflowIdFromCommand, workflowOption } from "./workflow-catalog.ts";
|
|
5
6
|
import {
|
|
6
7
|
contextIndexExcludeDigest,
|
|
7
8
|
contextIndexExcludePolicyVersion,
|
|
@@ -515,20 +516,24 @@ export function classifyContextTask(prompt) {
|
|
|
515
516
|
const folded = foldSearchSignal(text);
|
|
516
517
|
const paths = queryPathCandidates(text);
|
|
517
518
|
const terms = tokenizeQuery(text);
|
|
518
|
-
|
|
519
|
-
|
|
520
|
-
|
|
521
|
-
|
|
522
|
-
|
|
519
|
+
const leadingWorkflow = resolveWorkflowId(lower.match(/^\/?([a-z][a-z-]*)\b/)?.[1]);
|
|
520
|
+
const workflowId = workflowIdFromCommand(text) ?? leadingWorkflow;
|
|
521
|
+
const explicitWorkflow = workflowOption(workflowId);
|
|
522
|
+
let workflow = explicitWorkflow?.contextWorkflow ?? "task";
|
|
523
|
+
if (!explicitWorkflow && /\b(onboard|profile setup|first-read|cau hinh profile|khoi tao project)\b/.test(folded)) workflow = "onboard";
|
|
524
|
+
if (!explicitWorkflow && (/^\/?(?:release|publish|deploy)\b/.test(folded) || /\b(?:pull request|publish|deploy|phat hanh|trien khai)\b/.test(folded)
|
|
525
|
+
|| /\b(?:after|before|create|cut|merge|open|prepare|push|run|ship|tag|version)\b.{0,40}\b(?:commit|pr|release)\b/.test(folded))) workflow = "release";
|
|
523
526
|
const usageIntent = /^\/?(?:usage|session)\b/.test(lower)
|
|
524
527
|
|| /\b(?:show|check|view|report|current|session|how many)\b.{0,48}\b(?:usage|tokens?|cost|context stats|efficiency)\b/.test(lower)
|
|
525
528
|
|| /\b(?:usage|tokens?|cost|context stats|efficiency)\b.{0,48}\b(?:show|check|view|report|current|session)\b/.test(lower)
|
|
526
529
|
|| /\b(?:xem|kiem tra|bao cao|hien tai|phien)\b.{0,48}\b(?:usage|token|chi phi|context)\b/.test(folded);
|
|
527
|
-
if (usageIntent) workflow = "usage";
|
|
530
|
+
if (!explicitWorkflow && usageIntent) workflow = "usage";
|
|
528
531
|
const highRisk = /\b(auth|authorization|credential|database|deploy|encryption|migration|payment|permission|production|release|secret|security|token|bao mat|phan quyen|xac thuc|du lieu|co so du lieu|thanh toan|ma hoa|trien khai|phat hanh)\b/.test(folded);
|
|
529
532
|
const tiny = text.length < 220 && (paths.length > 0 || /\b(rename|typo|label|copy|one line|small|doi ten|sua chu|nhan|mot dong|nho)\b/.test(folded));
|
|
530
533
|
return {
|
|
531
534
|
workflow,
|
|
535
|
+
workflowId,
|
|
536
|
+
changeMode: explicitWorkflow?.changeMode ?? "unknown",
|
|
532
537
|
lane: highRisk ? "high-risk" : tiny ? "tiny" : "normal",
|
|
533
538
|
terms,
|
|
534
539
|
paths,
|
|
@@ -52,6 +52,83 @@ function explicitFocusedTestNavigation(text) {
|
|
|
52
52
|
});
|
|
53
53
|
}
|
|
54
54
|
|
|
55
|
+
const GENERIC_TEST_CONSTRAINT_TERMS = new Set([
|
|
56
|
+
"all", "any", "assertion", "assertions", "change", "changing", "coverage", "current", "do", "dont", "edit", "editing",
|
|
57
|
+
"existing", "inspect", "inspecting", "keep", "leave", "modify", "modifying", "never", "not", "our", "please", "read", "reading",
|
|
58
|
+
"run", "running", "spec", "specs", "test", "tests", "the", "these", "those", "touch", "touching", "update", "updating", "without",
|
|
59
|
+
"write", "writing"
|
|
60
|
+
]);
|
|
61
|
+
|
|
62
|
+
function normalizedConstraintTerm(value) {
|
|
63
|
+
const term = String(value ?? "").toLowerCase();
|
|
64
|
+
if (term.endsWith("ies") && term.length > 4) return `${term.slice(0, -3)}y`;
|
|
65
|
+
if (term.endsWith("s") && term.length > 4) return term.slice(0, -1);
|
|
66
|
+
return term;
|
|
67
|
+
}
|
|
68
|
+
|
|
69
|
+
function testConstraintAppliesToPath(scope, filePath) {
|
|
70
|
+
if (textExplicitlyReferencesPath(scope, filePath)) return true;
|
|
71
|
+
const qualifierGroups = String(scope ?? "").split(/\b(?:or|and)\b/iu).map((group) => (
|
|
72
|
+
group.match(/[\p{L}\p{N}_-]+/gu)?.map(normalizedConstraintTerm)
|
|
73
|
+
.filter((term) => term && !GENERIC_TEST_CONSTRAINT_TERMS.has(term)) ?? []
|
|
74
|
+
));
|
|
75
|
+
if (qualifierGroups.some((qualifiers) => qualifiers.length === 0)) return true;
|
|
76
|
+
const targetTerms = new Set(String(filePath ?? "").match(/[\p{L}\p{N}_-]+/gu)?.flatMap((part) => (
|
|
77
|
+
part.split(/[_-]+/).map(normalizedConstraintTerm)
|
|
78
|
+
)) ?? []);
|
|
79
|
+
return qualifierGroups.some((qualifiers) => qualifiers.every((term) => targetTerms.has(term)));
|
|
80
|
+
}
|
|
81
|
+
|
|
82
|
+
function immediateTestStateScope(clause, nounIndex) {
|
|
83
|
+
const prefix = clause.slice(0, nounIndex);
|
|
84
|
+
const segment = prefix.split(/[,;:.!?]/u).at(-1)?.trim() ?? "";
|
|
85
|
+
const afterControl = segment.match(/\b(?:keep|leave)\b(?<scope>.*)$/iu)?.groups?.scope?.trim() ?? segment;
|
|
86
|
+
const connectorTail = afterControl.match(/(?<scope>[\p{L}\p{N}_-]+(?:\s+(?:or|and)\s+[\p{L}\p{N}_-]+)+)\s*$/iu)?.groups?.scope;
|
|
87
|
+
if (connectorTail) return connectorTail;
|
|
88
|
+
return afterControl.match(/([\p{L}\p{N}_-]+)\s*$/u)?.[1] ?? "";
|
|
89
|
+
}
|
|
90
|
+
|
|
91
|
+
function explicitTestNavigationProhibited(text, filePath) {
|
|
92
|
+
const clauses = String(text ?? "").split(/(?:[;\n]+|(?<=[.!?])\s+)/).map((clause) => clause.trim()).filter(Boolean);
|
|
93
|
+
return clauses.some((clause) => {
|
|
94
|
+
const negativeAction = /\b(?:do\s+not|don't|dont|never|must\s+not|cannot|can't)\s+(?:(?:also|directly|ever|just)\s+){0,2}(?:change|edit|inspect|modify|read|run|touch|update|write)\w*\b(?<scope>.{0,50}?)\b(?:assertions?|coverage|specs?|tests?)\b/giu;
|
|
95
|
+
for (const match of clause.matchAll(negativeAction)) {
|
|
96
|
+
if (testConstraintAppliesToPath(match.groups?.scope ?? "", filePath)) return true;
|
|
97
|
+
}
|
|
98
|
+
const withoutAction = /\bwithout\s+(?:(?:also|directly|ever|just)\s+){0,2}(?:changing|editing|inspecting|modifying|reading|running|touching|updating|writing)\b(?<scope>.{0,50}?)\b(?:assertions?|coverage|specs?|tests?)\b/giu;
|
|
99
|
+
for (const match of clause.matchAll(withoutAction)) {
|
|
100
|
+
const localPrefix = clause.slice(0, match.index).split(/[,;:]|\b(?:but|while|whereas|and\s+then)\b/iu).at(-1) ?? "";
|
|
101
|
+
if (/\b(?:do\s+not|don't|dont|never|must\s+not|cannot|can't)\b/iu.test(localPrefix)) continue;
|
|
102
|
+
if (testConstraintAppliesToPath(match.groups?.scope ?? "", filePath)) return true;
|
|
103
|
+
}
|
|
104
|
+
const withoutTestChanges = /\bwithout\s+(?:any\s+)?(?<scope>.{0,40}?)\b(?:assertion|coverage|spec|test)s?\s+changes?\b/giu;
|
|
105
|
+
for (const match of clause.matchAll(withoutTestChanges)) {
|
|
106
|
+
const localPrefix = clause.slice(0, match.index).split(/[,;:]|\b(?:but|while|whereas|and\s+then)\b/iu).at(-1) ?? "";
|
|
107
|
+
if (/\b(?:do\s+not|don't|dont|never|must\s+not|cannot|can't)\b/iu.test(localPrefix)) continue;
|
|
108
|
+
if (testConstraintAppliesToPath(match.groups?.scope ?? "", filePath)) return true;
|
|
109
|
+
}
|
|
110
|
+
const noTestChanges = [
|
|
111
|
+
/\bno\s+changes?\s+to\s+(?<scope>.{0,40}?)\b(?:assertion|coverage|spec|test)s?\b/giu,
|
|
112
|
+
/\bno\s+(?<scope>.{0,40}?)\b(?:assertion|coverage|spec|test)s?\s+changes?\b/giu
|
|
113
|
+
];
|
|
114
|
+
for (const pattern of noTestChanges) {
|
|
115
|
+
for (const match of clause.matchAll(pattern)) {
|
|
116
|
+
if (testConstraintAppliesToPath(match.groups?.scope ?? "", filePath)) return true;
|
|
117
|
+
}
|
|
118
|
+
}
|
|
119
|
+
const testStatePatterns = [
|
|
120
|
+
/\b(?:assertions?|coverage|specs?|tests?)\b.{0,30}\b(?:unchanged|untouched)\b/giu,
|
|
121
|
+
/\b(?:assertions?|coverage|specs?|tests?)\b.{0,30}\b(?:not|never)\b.{0,15}\b(?:change|edit|inspect|modify|read|run|touch|update|write)\w*\b/giu
|
|
122
|
+
];
|
|
123
|
+
for (const pattern of testStatePatterns) {
|
|
124
|
+
for (const match of clause.matchAll(pattern)) {
|
|
125
|
+
if (testConstraintAppliesToPath(immediateTestStateScope(clause, match.index), filePath)) return true;
|
|
126
|
+
}
|
|
127
|
+
}
|
|
128
|
+
return false;
|
|
129
|
+
});
|
|
130
|
+
}
|
|
131
|
+
|
|
55
132
|
const DIRECT_EXPLICIT_LINK_KINDS = new Set(["explicit-imports-candidate", "candidate-imports-explicit"]);
|
|
56
133
|
|
|
57
134
|
function normalizedLinks(value) {
|
|
@@ -157,14 +234,19 @@ export function composeCriterionContextEntries(input = {}, options = {}) {
|
|
|
157
234
|
const limit = integer(options.limit, 6, 1, 12);
|
|
158
235
|
const criteriaText = uniqueStrings(input.criteria, 20).join("\n");
|
|
159
236
|
const explicitPaths = uniqueStrings(input.explicitPaths, 20).map(relativePath).filter(Boolean);
|
|
160
|
-
const
|
|
237
|
+
const plannedEntries = Array.isArray(input.plannedEntries) ? input.plannedEntries : [];
|
|
238
|
+
const plannedPaths = new Set(plannedEntries
|
|
161
239
|
.map((entry) => relativePath(entry?.path)).filter(Boolean));
|
|
240
|
+
const plannedReasons = new Map(plannedEntries.map((entry) => [
|
|
241
|
+
relativePath(entry?.path),
|
|
242
|
+
typeof entry?.reason === "string" ? entry.reason.trim() : ""
|
|
243
|
+
]).filter(([filePath]) => Boolean(filePath)));
|
|
162
244
|
const candidates = [];
|
|
163
245
|
(Array.isArray(input.retrievedItems) ? input.retrievedItems : []).forEach((entry, order) => {
|
|
164
246
|
const normalized = normalizedCandidate(entry, "retrieval", order);
|
|
165
247
|
if (normalized) candidates.push(normalized);
|
|
166
248
|
});
|
|
167
|
-
|
|
249
|
+
plannedEntries.forEach((entry, order) => {
|
|
168
250
|
const normalized = normalizedCandidate(entry, "criterion", order);
|
|
169
251
|
if (normalized) candidates.push(normalized);
|
|
170
252
|
});
|
|
@@ -294,9 +376,19 @@ export function composeCriterionContextEntries(input = {}, options = {}) {
|
|
|
294
376
|
// establish singleton cardinality, and only a test-shaped JS/TS filename is
|
|
295
377
|
// eligible. This remains navigation context; acceptance still depends on a
|
|
296
378
|
// changed live assertion and final verification.
|
|
297
|
-
if (!selected.some((entry) => testPath(entry.path))
|
|
379
|
+
if (!selected.some((entry) => testPath(entry.path))) {
|
|
298
380
|
const scopedTests = values.filter((candidate) => executablePlannedTest(candidate, plannedPaths, input.plannedSelectionComplete === true));
|
|
299
|
-
|
|
381
|
+
const testNavigationProhibited = scopedTests.length === 1
|
|
382
|
+
&& explicitTestNavigationProhibited(criteriaText, scopedTests[0].path);
|
|
383
|
+
const soleCriterionPlannedTest = scopedTests.length === 1
|
|
384
|
+
&& /^criterion-\d+ (?:behavior|boundary) target$/i.test(plannedReasons.get(scopedTests[0].path) ?? "")
|
|
385
|
+
&& !testNavigationProhibited;
|
|
386
|
+
if (scopedTests.length === 1 && !testNavigationProhibited
|
|
387
|
+
&& (explicitFocusedTestNavigation(criteriaText) || soleCriterionPlannedTest)) {
|
|
388
|
+
add(scopedTests[0], soleCriterionPlannedTest
|
|
389
|
+
? "Sole criterion-planned executable test (navigation context; not acceptance proof)"
|
|
390
|
+
: "Operator-requested sole scoped test target (navigation context; not acceptance proof)");
|
|
391
|
+
}
|
|
300
392
|
}
|
|
301
393
|
|
|
302
394
|
relatedTests
|
|
@@ -5,8 +5,8 @@ export const FAILURE_PARSER_VERSION = "failure-parser-v1" as const;
|
|
|
5
5
|
export const FAILURE_POLICY_VERSION = "failure-policy-v1" as const;
|
|
6
6
|
export const FAILURE_CATEGORIES = Object.freeze(["passed", "compile-typecheck", "test-assertion", "lint-format", "dependency-config", "environment", "provider-network", "permission-policy", "scope-protected-path", "flaky-infrastructure", "unknown"] as const);
|
|
7
7
|
export const FAILURE_OWNERS = Object.freeze(["none", "source", "test-expectation", "dependency", "environment", "provider", "permission", "policy", "scope", "infrastructure", "unknown"] as const);
|
|
8
|
-
export const FAILURE_STRUCTURED_EVENTS = Object.freeze(["scope-violation", "protected-path", "permission-denied", "policy-block", "provider-network", "environment", "flaky-infrastructure", "stale-verifier", "missing-verifier"] as const);
|
|
9
|
-
export const FAILURE_REASON_CODES = Object.freeze(["exit-zero", "structured-scope", "structured-protected-path", "structured-permission", "structured-policy", "structured-provider", "structured-environment", "structured-flaky-infrastructure", "structured-stale-verifier", "structured-missing-verifier", "typescript-diagnostic", "python-type-diagnostic", "go-compile-diagnostic", "rust-compile-diagnostic", "generic-compile-diagnostic", "test-assertion-diagnostic", "lint-format-diagnostic", "dependency-config-diagnostic", "environment-diagnostic", "provider-rate-limit", "provider-transport", "permission-diagnostic", "policy-diagnostic", "scope-diagnostic", "local-port-conflict", "transient-infrastructure", "unknown-diagnostic"] as const);
|
|
8
|
+
export const FAILURE_STRUCTURED_EVENTS = Object.freeze(["scope-violation", "protected-path", "permission-denied", "policy-block", "provider-network", "environment", "flaky-infrastructure", "stale-verifier", "missing-verifier", "verification-gap"] as const);
|
|
9
|
+
export const FAILURE_REASON_CODES = Object.freeze(["exit-zero", "structured-scope", "structured-protected-path", "structured-permission", "structured-policy", "structured-provider", "structured-environment", "structured-flaky-infrastructure", "structured-stale-verifier", "structured-missing-verifier", "structured-verification-gap", "typescript-diagnostic", "python-type-diagnostic", "go-compile-diagnostic", "rust-compile-diagnostic", "generic-compile-diagnostic", "test-assertion-diagnostic", "lint-format-diagnostic", "dependency-config-diagnostic", "environment-diagnostic", "provider-rate-limit", "provider-transport", "permission-diagnostic", "policy-diagnostic", "scope-diagnostic", "local-port-conflict", "transient-infrastructure", "unknown-diagnostic"] as const);
|
|
10
10
|
|
|
11
11
|
export type FailureCategory = typeof FAILURE_CATEGORIES[number];
|
|
12
12
|
export type FailureOwner = typeof FAILURE_OWNERS[number];
|
|
@@ -366,7 +366,7 @@ export type TaskContract = {
|
|
|
366
366
|
sessionId: string;
|
|
367
367
|
sessionName?: string;
|
|
368
368
|
changeMode: "source-change" | "read-only";
|
|
369
|
-
mutationPolicy?: "required" | "forbidden";
|
|
369
|
+
mutationPolicy?: "required" | "allowed" | "forbidden";
|
|
370
370
|
attempt: number;
|
|
371
371
|
maxAttempts: number;
|
|
372
372
|
previousAttempts: TaskAttemptSummary[];
|
|
@@ -417,9 +417,12 @@ export type TaskContract = {
|
|
|
417
417
|
matchedProfileCommand?: boolean;
|
|
418
418
|
preWorkingTreeDigest?: string;
|
|
419
419
|
workingTreeDigest?: string;
|
|
420
|
+
preWorkspaceRevisionDigest?: string;
|
|
421
|
+
workspaceRevisionDigest?: string;
|
|
420
422
|
}>;
|
|
421
423
|
trace: {
|
|
422
424
|
outcome: "pending" | "completed" | "blocked" | "partial" | "failed";
|
|
425
|
+
terminalDisposition?: "refused";
|
|
423
426
|
friction?: string;
|
|
424
427
|
notes?: string;
|
|
425
428
|
recordedAt?: string;
|
|
@@ -485,6 +488,7 @@ export type ExternalActionPolicyConfig = {
|
|
|
485
488
|
};
|
|
486
489
|
|
|
487
490
|
export type FinalGateConfig = {
|
|
491
|
+
acceptanceProofMode?: "enforce" | "diagnostic";
|
|
488
492
|
defaultMode?: "advisory" | "enforce";
|
|
489
493
|
requireTaskContract?: boolean;
|
|
490
494
|
requireContextManifest?: boolean;
|