@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 +1,67 @@
|
|
|
1
|
+
import { LONG_INPUT_CHARS } from "../runtime-limits.ts";
|
|
2
|
+
import { foldChangeIntent, isExplicitChangeRequest } from "./change-clarification.ts";
|
|
3
|
+
|
|
4
|
+
const INSPECTION = /\b(?:audit|check|inspect|review|tests?|verify|verification|validate|rerun|re-run|recheck|re-check|kiem tra|ra soat|xac minh|danh gia|chay lai)\b/i;
|
|
5
|
+
const REPAIR = "(?:address|change|correct(?:ing)?|edit|fix|patch|repair|resolve|update|cap nhat|chinh sua|khac phuc|sua)";
|
|
6
|
+
const ACTION = "(?:add|apply|build|create|implement|modify|mutate|refactor|remove|rename|replace|write|them|tao|xoa|doi|" + REPAIR + ")";
|
|
7
|
+
const PREFIX_CONDITION = /^(?:(?:please|only|chi|em|anh)\s+)*(?:if|when|neu|khi)\b/i;
|
|
8
|
+
const REPAIR_LEAD = new RegExp(`^(?:(?:please|only|chi|em|anh)\\s+)*${REPAIR}\\b`, "i");
|
|
9
|
+
const CONDITIONAL_REPAIR = [
|
|
10
|
+
new RegExp(`^(?:(?:please|only|chi|em|anh)\\s+)*(?:if|when|neu|khi)\\b.*\\b${REPAIR}\\b`, "i"),
|
|
11
|
+
new RegExp(`\\b${REPAIR}\\b.*\\b(?:if|when|neu|khi)\\b.*\\b(?:fail(?:s|ed|ures?|ing)?|issues?|errors?|problems?|defects?|findings?|tests?|verification|verifier|needed|necessary|required|found|detected|discovered|can|co|loi|van de)\\b`, "i"),
|
|
12
|
+
new RegExp(`\\b${REPAIR}\\b.*\\b(?:as\\s+(?:needed|necessary|required)|(?:any|all|every)\\s+(?:(?:additional|remaining|new)\\s+)?(?:failures?|issues?|errors?|problems?|defects?|findings?)|whatever\\s+fails|(?:failures?|issues?|errors?|problems?|defects?|findings?)\\s+(?:(?:(?:if|when)\\s+)?(?:found|detected|discovered)|if\\s+any|(?:you|we)\\s+find)|loi|van de)\\b`, "i"),
|
|
13
|
+
new RegExp(`\\b${REPAIR}\\b\\s+(?:(?:any|all|every)\\s+)?(?:failures?|issues?|errors?|problems?|defects?|findings?|failing\\s+tests?)\\b`, "i")
|
|
14
|
+
];
|
|
15
|
+
const DIRECTIVE_PREFIX = "(?:(?:you|we|the task|your task)\\s+(?:must|need to|have to)|(?:can|could|would|will)\\s+you|i\\s+(?:need|want)\\s+you\\s+to|(?:em|anh)\\s+(?:phai|can)|must|please|only|chi|em|anh|also|always|still|regardless|in any case|bat buoc)";
|
|
16
|
+
const DIRECTIVE_LEAD = new RegExp(`^(?:${DIRECTIVE_PREFIX}\\b[,;:]?\\s+)*`, "i");
|
|
17
|
+
const ACTION_SEPARATOR = new RegExp(`(?:,\\s*|\\b(?:and|then|but|va|roi|nhung|sau do)\\s+)(?=(?:${DIRECTIVE_PREFIX}\\b[,;:]?\\s+)*${ACTION}\\b)`, "i");
|
|
18
|
+
const NEGATED_REPAIR = /\b(?:do not|don't|must not|never|khong|dung)\s+(?:ever\s+)?(?:fix|repair|edit|change|update|sua|chinh sua|cap nhat)\b/i;
|
|
19
|
+
|
|
20
|
+
// Examples, quoted data and fenced snippets cannot make an otherwise required
|
|
21
|
+
// implementation optional. Keep line boundaries while masking literal text.
|
|
22
|
+
function instructionText(text: string): string {
|
|
23
|
+
let quote = "", escaped = false, result = "";
|
|
24
|
+
for (let index = 0; index < text.length; index++) {
|
|
25
|
+
const character = text[index];
|
|
26
|
+
if (quote) {
|
|
27
|
+
result += character === "\n" ? "\n" : " ";
|
|
28
|
+
if (character === quote && !escaped) quote = "";
|
|
29
|
+
escaped = character === "\\" && !escaped;
|
|
30
|
+
continue;
|
|
31
|
+
}
|
|
32
|
+
const apostrophe = character === "'" && /[\p{L}\p{N}]/u.test(text[index - 1] ?? "") && /[\p{L}\p{N}]/u.test(text[index + 1] ?? "");
|
|
33
|
+
if ((character === "`" || character === '"' || character === "'") && !apostrophe) { quote = character; escaped = false; result += " "; }
|
|
34
|
+
else result += character;
|
|
35
|
+
}
|
|
36
|
+
return result.replace(/^\s*>[^\n]*$/gm, "");
|
|
37
|
+
}
|
|
38
|
+
|
|
39
|
+
export type ConditionalRepairIntent = "none" | "conditional-only" | "mixed";
|
|
40
|
+
|
|
41
|
+
/** Linguistic evidence only: task scope, approval and observed failures still gate writes. */
|
|
42
|
+
export function classifyConditionalRepairIntent(text: string): ConditionalRepairIntent {
|
|
43
|
+
if (text.length > LONG_INPUT_CHARS) return "none";
|
|
44
|
+
const visible = foldChangeIntent(instructionText(text));
|
|
45
|
+
if (!INSPECTION.test(visible)) return "none";
|
|
46
|
+
let conditional = false, unconditional = false;
|
|
47
|
+
const statements = visible.split(/(?<=[.!?;:])\s+|\n+|\b(?:but|however|nevertheless|nhung|tuy nhien)\b[,:]?\s*|(?:,|\b(?:and|then|va|roi)\b)\s*(?=(?:if|when|neu|khi)\b)/)
|
|
48
|
+
.map((part) => part.replace(/^\s*(?:[-*+]\s+|\d+[.)]\s+)/, "").trim()).filter(Boolean);
|
|
49
|
+
for (const statement of statements) {
|
|
50
|
+
if (PREFIX_CONDITION.test(statement) && CONDITIONAL_REPAIR.some((pattern) => pattern.test(statement)) && !NEGATED_REPAIR.test(statement)) {
|
|
51
|
+
conditional = true;
|
|
52
|
+
continue;
|
|
53
|
+
}
|
|
54
|
+
for (const clause of statement.split(ACTION_SEPARATOR)) {
|
|
55
|
+
const directive = clause.trim().replace(DIRECTIVE_LEAD, "");
|
|
56
|
+
const repairCommand = REPAIR_LEAD.test(directive);
|
|
57
|
+
const imperative = repairCommand || isExplicitChangeRequest(directive);
|
|
58
|
+
if (!imperative && NEGATED_REPAIR.test(clause)) continue;
|
|
59
|
+
// Qualify the repair command itself, not a repair word inside the
|
|
60
|
+
// description of a different mandatory implementation command.
|
|
61
|
+
const qualified = repairCommand && CONDITIONAL_REPAIR.some((pattern) => pattern.test(clause));
|
|
62
|
+
if (qualified) conditional = true;
|
|
63
|
+
else if (imperative) unconditional = true;
|
|
64
|
+
}
|
|
65
|
+
}
|
|
66
|
+
return !conditional ? "none" : unconditional ? "mixed" : "conditional-only";
|
|
67
|
+
}
|
|
@@ -0,0 +1,83 @@
|
|
|
1
|
+
import { matchesAnyPath, normalizePathCandidate } from "../../extensions/policy-core.js";
|
|
2
|
+
|
|
3
|
+
function uniqueStrings(values: string[]): string[] {
|
|
4
|
+
return [...new Set(values)];
|
|
5
|
+
}
|
|
6
|
+
|
|
7
|
+
export type TaskScopeResolution = {
|
|
8
|
+
scope: string[];
|
|
9
|
+
mappings: Array<{ from: string; to: string }>;
|
|
10
|
+
ambiguous: Array<{ input: string; candidates: string[] }>;
|
|
11
|
+
unmatched: string[];
|
|
12
|
+
};
|
|
13
|
+
|
|
14
|
+
const STANDARD_SCOPE_ROOT = /^(?:\.github|app|apps|bin|config|docs|examples|lib|logs|packages|pages|plans|public|scripts|services|spec|src|test|tests|vendor|__tests__)(?:\/|$)/i;
|
|
15
|
+
|
|
16
|
+
function wildcardPrefix(value: string): string {
|
|
17
|
+
const index = value.search(/[*?{}\[\]]/);
|
|
18
|
+
return (index < 0 ? value : value.slice(0, index)).replace(/\/+$/, "");
|
|
19
|
+
}
|
|
20
|
+
|
|
21
|
+
function nestedDirectoryMatches(prefix: string, projectFiles: string[]): string[] {
|
|
22
|
+
if (!prefix) return [];
|
|
23
|
+
const segments = prefix.split("/").filter(Boolean);
|
|
24
|
+
const matches = new Set<string>();
|
|
25
|
+
for (const file of projectFiles) {
|
|
26
|
+
const parts = normalizePathCandidate(file).split("/");
|
|
27
|
+
for (let index = 0; index <= parts.length - 1 - segments.length; index += 1) {
|
|
28
|
+
if (segments.every((segment, offset) => parts[index + offset] === segment)) {
|
|
29
|
+
matches.add(parts.slice(0, index + segments.length).join("/"));
|
|
30
|
+
}
|
|
31
|
+
}
|
|
32
|
+
}
|
|
33
|
+
return [...matches].sort();
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
export function resolveTaskScopePatterns(scope: string[], projectFiles: string[]): TaskScopeResolution {
|
|
37
|
+
const files = uniqueStrings(projectFiles.map(normalizePathCandidate).filter(Boolean));
|
|
38
|
+
const resolved: string[] = [];
|
|
39
|
+
const mappings: Array<{ from: string; to: string }> = [];
|
|
40
|
+
const ambiguous: Array<{ input: string; candidates: string[] }> = [];
|
|
41
|
+
const unmatched: string[] = [];
|
|
42
|
+
for (const raw of scope) {
|
|
43
|
+
const candidate = normalizePathCandidate(raw);
|
|
44
|
+
if (!candidate) continue;
|
|
45
|
+
if (files.includes(candidate) || files.some((file) => matchesAnyPath(file, [candidate]))) {
|
|
46
|
+
resolved.push(candidate);
|
|
47
|
+
continue;
|
|
48
|
+
}
|
|
49
|
+
const hasWildcard = /[*?{}\[\]]/.test(candidate);
|
|
50
|
+
if (hasWildcard) {
|
|
51
|
+
const prefix = wildcardPrefix(candidate);
|
|
52
|
+
const directories = nestedDirectoryMatches(prefix, files);
|
|
53
|
+
if (directories.length === 1) {
|
|
54
|
+
const suffix = candidate.slice(prefix.length).replace(/^\/+/, "");
|
|
55
|
+
const canonical = suffix ? `${directories[0]}/${suffix}` : directories[0];
|
|
56
|
+
resolved.push(canonical);
|
|
57
|
+
mappings.push({ from: candidate, to: canonical });
|
|
58
|
+
} else if (directories.length > 1 && prefix && !STANDARD_SCOPE_ROOT.test(prefix)) {
|
|
59
|
+
ambiguous.push({ input: candidate, candidates: directories.slice(0, 12).map((directory) => {
|
|
60
|
+
const suffix = candidate.slice(prefix.length).replace(/^\/+/, "");
|
|
61
|
+
return suffix ? `${directory}/${suffix}` : directory;
|
|
62
|
+
}) });
|
|
63
|
+
} else if (directories.length === 0 && prefix && !prefix.includes("/") && !STANDARD_SCOPE_ROOT.test(prefix) && !raw.startsWith("./")) {
|
|
64
|
+
unmatched.push(candidate);
|
|
65
|
+
} else {
|
|
66
|
+
resolved.push(candidate);
|
|
67
|
+
}
|
|
68
|
+
continue;
|
|
69
|
+
}
|
|
70
|
+
const suffixMatches = files.filter((file) => file === candidate || file.endsWith(`/${candidate}`));
|
|
71
|
+
if (suffixMatches.length === 1) {
|
|
72
|
+
resolved.push(suffixMatches[0]);
|
|
73
|
+
mappings.push({ from: candidate, to: suffixMatches[0] });
|
|
74
|
+
} else if (suffixMatches.length > 1) {
|
|
75
|
+
ambiguous.push({ input: candidate, candidates: suffixMatches.slice(0, 12) });
|
|
76
|
+
} else if (!candidate.includes("/") && !raw.startsWith("./")) {
|
|
77
|
+
unmatched.push(candidate);
|
|
78
|
+
} else {
|
|
79
|
+
resolved.push(candidate);
|
|
80
|
+
}
|
|
81
|
+
}
|
|
82
|
+
return { scope: uniqueStrings(resolved), mappings, ambiguous, unmatched: uniqueStrings(unmatched) };
|
|
83
|
+
}
|
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
export const WEBUI_RUNTIME_ACTION_IDS = [
|
|
2
|
-
"runtime.status", "runtime.inspector", "runtime.commands", "orchestration.status",
|
|
2
|
+
"runtime.status", "runtime.inspector", "runtime.commands", "runtime.fast-status", "runtime.fast-on", "runtime.fast-off", "orchestration.status",
|
|
3
3
|
"usage.live", "usage.history", "usage.logs", "usage.efficiency", "usage.preflight",
|
|
4
4
|
"onboarding.status", "onboarding.profile", "onboarding.tech",
|
|
5
5
|
"profile.status", "profile.options", "profile.tech-options", "profile.apply", "profile.auto",
|
|
@@ -9,7 +9,7 @@ export const WEBUI_RUNTIME_ACTION_IDS = [
|
|
|
9
9
|
] as const;
|
|
10
10
|
|
|
11
11
|
export type WebUiRuntimeActionId = typeof WEBUI_RUNTIME_ACTION_IDS[number];
|
|
12
|
-
export type WebUiRuntimeEffect = "read-only" | "workspace-write" | "model-assisted";
|
|
12
|
+
export type WebUiRuntimeEffect = "read-only" | "workspace-write" | "model-assisted" | "session-setting";
|
|
13
13
|
export type WebUiRuntimeArgument = "none" | "optional-text" | "required-text" | "profile" | "connection";
|
|
14
14
|
|
|
15
15
|
export type WebUiRuntimeAction = {
|
|
@@ -27,6 +27,8 @@ const action = (id: WebUiRuntimeActionId, category: WebUiRuntimeAction["category
|
|
|
27
27
|
|
|
28
28
|
export const WEBUI_RUNTIME_ACTIONS: readonly WebUiRuntimeAction[] = [
|
|
29
29
|
action("runtime.status", "runtime"), action("runtime.inspector", "runtime"), action("runtime.commands", "runtime"),
|
|
30
|
+
action("runtime.fast-status", "runtime"), action("runtime.fast-on", "runtime", "session-setting"),
|
|
31
|
+
action("runtime.fast-off", "runtime", "session-setting"),
|
|
30
32
|
action("orchestration.status", "runtime"),
|
|
31
33
|
action("usage.live", "usage"), action("usage.history", "usage"), action("usage.logs", "usage"),
|
|
32
34
|
action("usage.efficiency", "usage"), action("usage.preflight", "usage", "read-only", "optional-text"),
|
|
@@ -71,6 +73,7 @@ export function buildWebUiRuntimeCommand(input: {
|
|
|
71
73
|
const suffix = argument ? ` ${argument}` : "";
|
|
72
74
|
const commands: Record<WebUiRuntimeActionId, string> = {
|
|
73
75
|
"runtime.status": "/piagent-status", "runtime.inspector": "/piagent-inspector", "runtime.commands": "/commands",
|
|
76
|
+
"runtime.fast-status": "/fast status", "runtime.fast-on": "/fast on", "runtime.fast-off": "/fast off",
|
|
74
77
|
"orchestration.status": "/piagent-orchestration",
|
|
75
78
|
"usage.live": "/usage live", "usage.history": "/usage history", "usage.logs": "/usage logs",
|
|
76
79
|
"usage.efficiency": "/usage efficiency", "usage.preflight": `/usage preflight${suffix}`,
|
|
@@ -1,40 +1,47 @@
|
|
|
1
|
-
|
|
2
|
-
|
|
3
|
-
|
|
4
|
-
|
|
5
|
-
|
|
6
|
-
|
|
7
|
-
"review",
|
|
8
|
-
"commit",
|
|
9
|
-
"pr",
|
|
10
|
-
"onboard",
|
|
11
|
-
"platform-improve"
|
|
12
|
-
] as const;
|
|
1
|
+
import {
|
|
2
|
+
WORKFLOW_IDS,
|
|
3
|
+
WORKFLOW_OPTIONS,
|
|
4
|
+
type WorkflowId,
|
|
5
|
+
type WorkflowOption
|
|
6
|
+
} from "../../extensions/workflow-catalog.ts";
|
|
13
7
|
|
|
14
|
-
|
|
8
|
+
/**
|
|
9
|
+
* Runtime/WebUI adapter over the surface-neutral workflow catalog.
|
|
10
|
+
*
|
|
11
|
+
* Keep aliases here rather than in an individual surface. A workflow selection
|
|
12
|
+
* is a property of one message/operation; it never pins the rest of a session.
|
|
13
|
+
*/
|
|
14
|
+
export {
|
|
15
|
+
WORKFLOW_ALIASES,
|
|
16
|
+
WORKFLOW_IDS,
|
|
17
|
+
WORKFLOW_OPTIONS,
|
|
18
|
+
resolveWorkflowId,
|
|
19
|
+
stripWorkflowCommand,
|
|
20
|
+
workflowCommandNames,
|
|
21
|
+
workflowCommandPattern,
|
|
22
|
+
workflowHelpLines,
|
|
23
|
+
workflowIdFromCommand,
|
|
24
|
+
workflowOption
|
|
25
|
+
} from "../../extensions/workflow-catalog.ts";
|
|
26
|
+
export type { ContextWorkflow, WorkflowId, WorkflowOption } from "../../extensions/workflow-catalog.ts";
|
|
15
27
|
|
|
16
|
-
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
recommendedFreshSession: boolean;
|
|
21
|
-
};
|
|
28
|
+
// Compatibility exports keep the public WebUI contract stable while its data
|
|
29
|
+
// now comes from the surface-neutral catalog above.
|
|
30
|
+
export const WEBUI_WORKFLOW_IDS = WORKFLOW_IDS;
|
|
31
|
+
export type WebUiWorkflowId = WorkflowId;
|
|
22
32
|
|
|
23
|
-
export
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
|
|
30
|
-
|
|
31
|
-
|
|
32
|
-
{ id: "onboard", changeMode: "onboarding", modelUse: "required", recommendedFreshSession: true },
|
|
33
|
-
{ id: "platform-improve", changeMode: "platform", modelUse: "required", recommendedFreshSession: true }
|
|
34
|
-
] as const;
|
|
33
|
+
export type WebUiWorkflowOption = Pick<WorkflowOption,
|
|
34
|
+
"id" | "label" | "changeMode" | "modelUse" | "recommendedFreshSession">;
|
|
35
|
+
export const WEBUI_WORKFLOW_OPTIONS: readonly WebUiWorkflowOption[] = WORKFLOW_OPTIONS.map((option) => ({
|
|
36
|
+
id: option.id,
|
|
37
|
+
label: option.label,
|
|
38
|
+
changeMode: option.changeMode,
|
|
39
|
+
modelUse: option.modelUse,
|
|
40
|
+
recommendedFreshSession: option.recommendedFreshSession
|
|
41
|
+
}));
|
|
35
42
|
|
|
36
43
|
export function isWebUiWorkflowId(value: unknown): value is WebUiWorkflowId {
|
|
37
|
-
return typeof value === "string" && (
|
|
44
|
+
return typeof value === "string" && (WORKFLOW_IDS as readonly string[]).includes(value);
|
|
38
45
|
}
|
|
39
46
|
|
|
40
47
|
/**
|
|
@@ -0,0 +1,33 @@
|
|
|
1
|
+
import type { WorkflowId } from "../../extensions/workflow-catalog.ts";
|
|
2
|
+
|
|
3
|
+
export function buildOnboardingWorkflowPrompt(focus: string): string {
|
|
4
|
+
return [
|
|
5
|
+
"Run the Pi Agent Platform first-read onboarding workflow for this repository.",
|
|
6
|
+
"",
|
|
7
|
+
`Optional focus: ${focus.trim() || "whole repository"}`,
|
|
8
|
+
"",
|
|
9
|
+
"Preconditions:",
|
|
10
|
+
"- The operator has logged in and selected the intended model/thinking level.",
|
|
11
|
+
"- Stay read-only except writing Piagent onboarding state through piagent tools.",
|
|
12
|
+
"",
|
|
13
|
+
"Mandatory flow:",
|
|
14
|
+
"1. Call piagent_context with detail=full.",
|
|
15
|
+
"2. If the project is unprofiled, call piagent_profile_options, do a lightweight root scout, recommend a profile, and use piagent_profile_apply only after the operator choice is clear.",
|
|
16
|
+
"3. Prefer /profile setup or piagent_profile_tech_options + piagent_profile_tech_apply for tech stack selection.",
|
|
17
|
+
"4. Re-call piagent_context after profile/tech changes.",
|
|
18
|
+
"5. Call piagent_memory_status and treat memory as advisory.",
|
|
19
|
+
"6. Read AGENTS.md, README/package/build config, required context, docs/architecture files, source map, and verify command definitions. Do not ingest the whole repo.",
|
|
20
|
+
"7. Identify project purpose, stack/runtime, ownership boundaries, high-risk areas, protected paths, verify commands, MCP/tool capabilities, selected tech stack, memory policy, and conventions.",
|
|
21
|
+
"8. Write a concise .pi/project-context.md snapshot and record it with piagent_project_onboarding_record so .pi/context-index.json is generated.",
|
|
22
|
+
"9. Call piagent_context_index_status and report pending/stale warnings.",
|
|
23
|
+
"",
|
|
24
|
+
"Final output: profile, tech stack, context files read, verification matrix, high-risk areas, context-index status, memory status, and any missing operator decisions."
|
|
25
|
+
].join("\n");
|
|
26
|
+
}
|
|
27
|
+
|
|
28
|
+
/** Exact /workflow namespace follow-up; keep request whitespace and workflow intent. */
|
|
29
|
+
export function buildWorkflowFollowUp(workflow: WorkflowId, request: string): string {
|
|
30
|
+
return workflow === "onboard"
|
|
31
|
+
? buildOnboardingWorkflowPrompt(request)
|
|
32
|
+
: [`/${workflow}`, request].filter(Boolean).join(" ");
|
|
33
|
+
}
|
|
@@ -0,0 +1,132 @@
|
|
|
1
|
+
import { createHash } from "node:crypto";
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
|
|
5
|
+
export class WorkspaceReadError extends Error {
|
|
6
|
+
readonly code: "outside-root" | "symlink-ancestor" | "not-file" | "oversized" | "changed-during-read";
|
|
7
|
+
|
|
8
|
+
constructor(code: WorkspaceReadError["code"], message: string) {
|
|
9
|
+
super(message);
|
|
10
|
+
this.name = "WorkspaceReadError";
|
|
11
|
+
this.code = code;
|
|
12
|
+
}
|
|
13
|
+
}
|
|
14
|
+
|
|
15
|
+
type SafeCandidate = { root: string; absolute: string; parent: string };
|
|
16
|
+
|
|
17
|
+
function inside(root: string, candidate: string): boolean {
|
|
18
|
+
const relative = path.relative(root, candidate);
|
|
19
|
+
return relative === "" || (relative !== ".." && !relative.startsWith(`..${path.sep}`) && !path.isAbsolute(relative));
|
|
20
|
+
}
|
|
21
|
+
|
|
22
|
+
function safeCandidate(repoRoot: string, repoPath: string): SafeCandidate {
|
|
23
|
+
const root = fs.realpathSync.native(repoRoot);
|
|
24
|
+
const segments = repoPath.split("/");
|
|
25
|
+
if (!repoPath || path.posix.isAbsolute(repoPath) || segments.includes("..") || segments.includes("")) {
|
|
26
|
+
throw new WorkspaceReadError("outside-root", "Workspace path is not a safe repository-relative path");
|
|
27
|
+
}
|
|
28
|
+
let parent = root;
|
|
29
|
+
for (const segment of segments.slice(0, -1)) {
|
|
30
|
+
parent = path.join(parent, segment);
|
|
31
|
+
const stat = fs.lstatSync(parent);
|
|
32
|
+
if (stat.isSymbolicLink()) throw new WorkspaceReadError("symlink-ancestor", "Workspace path traverses a symbolic link");
|
|
33
|
+
if (!stat.isDirectory()) throw new WorkspaceReadError("not-file", "Workspace path has a non-directory ancestor");
|
|
34
|
+
}
|
|
35
|
+
const canonicalParent = fs.realpathSync.native(parent);
|
|
36
|
+
if (!inside(root, canonicalParent)) throw new WorkspaceReadError("outside-root", "Workspace path resolves outside its repository root");
|
|
37
|
+
const absolute = path.join(canonicalParent, segments.at(-1) as string);
|
|
38
|
+
if (!inside(root, absolute)) throw new WorkspaceReadError("outside-root", "Workspace path escapes its repository root");
|
|
39
|
+
return { root, absolute, parent: canonicalParent };
|
|
40
|
+
}
|
|
41
|
+
|
|
42
|
+
function sameFile(left: fs.Stats, right: fs.Stats): boolean {
|
|
43
|
+
return left.dev === right.dev && left.ino === right.ino && left.mode === right.mode;
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
function openRegularFile(repoRoot: string, repoPath: string): { descriptor: number; stat: fs.Stats; candidate: SafeCandidate } {
|
|
47
|
+
const candidate = safeCandidate(repoRoot, repoPath);
|
|
48
|
+
const descriptor = fs.openSync(candidate.absolute, fs.constants.O_RDONLY | (fs.constants.O_NOFOLLOW ?? 0));
|
|
49
|
+
try {
|
|
50
|
+
const stat = fs.fstatSync(descriptor);
|
|
51
|
+
if (!stat.isFile()) throw new WorkspaceReadError("not-file", "Workspace entry is not a regular file");
|
|
52
|
+
const canonical = fs.realpathSync.native(candidate.absolute);
|
|
53
|
+
const current = fs.lstatSync(candidate.absolute);
|
|
54
|
+
if (!inside(candidate.root, canonical) || !sameFile(stat, current)) {
|
|
55
|
+
throw new WorkspaceReadError("changed-during-read", "Workspace path changed while it was opened");
|
|
56
|
+
}
|
|
57
|
+
return { descriptor, stat, candidate };
|
|
58
|
+
} catch (error) {
|
|
59
|
+
fs.closeSync(descriptor);
|
|
60
|
+
throw error;
|
|
61
|
+
}
|
|
62
|
+
}
|
|
63
|
+
|
|
64
|
+
function verifyAfterRead(descriptor: number, before: fs.Stats, candidate: SafeCandidate): void {
|
|
65
|
+
const after = fs.fstatSync(descriptor);
|
|
66
|
+
const current = fs.lstatSync(candidate.absolute);
|
|
67
|
+
const canonical = fs.realpathSync.native(candidate.absolute);
|
|
68
|
+
if (
|
|
69
|
+
!sameFile(before, after)
|
|
70
|
+
|| !sameFile(after, current)
|
|
71
|
+
|| before.size !== after.size
|
|
72
|
+
|| before.mtimeMs !== after.mtimeMs
|
|
73
|
+
|| before.ctimeMs !== after.ctimeMs
|
|
74
|
+
|| !inside(candidate.root, canonical)
|
|
75
|
+
) {
|
|
76
|
+
throw new WorkspaceReadError("changed-during-read", "Workspace file changed during inspection");
|
|
77
|
+
}
|
|
78
|
+
}
|
|
79
|
+
|
|
80
|
+
export function inspectWorkspaceEntry(repoRoot: string, repoPath: string): fs.Stats {
|
|
81
|
+
const candidate = safeCandidate(repoRoot, repoPath);
|
|
82
|
+
return fs.lstatSync(candidate.absolute);
|
|
83
|
+
}
|
|
84
|
+
|
|
85
|
+
export function readWorkspaceLink(repoRoot: string, repoPath: string): string {
|
|
86
|
+
const candidate = safeCandidate(repoRoot, repoPath);
|
|
87
|
+
const before = fs.lstatSync(candidate.absolute);
|
|
88
|
+
if (!before.isSymbolicLink()) throw new WorkspaceReadError("not-file", "Workspace entry is not a symbolic link");
|
|
89
|
+
const target = fs.readlinkSync(candidate.absolute);
|
|
90
|
+
const after = fs.lstatSync(candidate.absolute);
|
|
91
|
+
if (!sameFile(before, after) || before.mtimeMs !== after.mtimeMs || before.ctimeMs !== after.ctimeMs) {
|
|
92
|
+
throw new WorkspaceReadError("changed-during-read", "Workspace link changed during inspection");
|
|
93
|
+
}
|
|
94
|
+
return target;
|
|
95
|
+
}
|
|
96
|
+
|
|
97
|
+
export function readWorkspaceFile(repoRoot: string, repoPath: string, maxBytes: number): Buffer {
|
|
98
|
+
const opened = openRegularFile(repoRoot, repoPath);
|
|
99
|
+
try {
|
|
100
|
+
if (opened.stat.size > maxBytes) throw new WorkspaceReadError("oversized", "Workspace file exceeds its read cap");
|
|
101
|
+
const content = Buffer.allocUnsafe(opened.stat.size);
|
|
102
|
+
let position = 0;
|
|
103
|
+
while (position < content.length) {
|
|
104
|
+
const count = fs.readSync(opened.descriptor, content, position, content.length - position, position);
|
|
105
|
+
if (count <= 0) throw new WorkspaceReadError("changed-during-read", "Workspace file ended during inspection");
|
|
106
|
+
position += count;
|
|
107
|
+
}
|
|
108
|
+
verifyAfterRead(opened.descriptor, opened.stat, opened.candidate);
|
|
109
|
+
return content;
|
|
110
|
+
} finally {
|
|
111
|
+
fs.closeSync(opened.descriptor);
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
|
|
115
|
+
export function hashWorkspaceFile(repoRoot: string, repoPath: string): string {
|
|
116
|
+
const opened = openRegularFile(repoRoot, repoPath);
|
|
117
|
+
try {
|
|
118
|
+
const state = createHash("sha256");
|
|
119
|
+
const buffer = Buffer.allocUnsafe(256 * 1024);
|
|
120
|
+
let position = 0;
|
|
121
|
+
while (position < opened.stat.size) {
|
|
122
|
+
const count = fs.readSync(opened.descriptor, buffer, 0, Math.min(buffer.length, opened.stat.size - position), position);
|
|
123
|
+
if (count <= 0) throw new WorkspaceReadError("changed-during-read", "Workspace file ended during hashing");
|
|
124
|
+
state.update(buffer.subarray(0, count));
|
|
125
|
+
position += count;
|
|
126
|
+
}
|
|
127
|
+
verifyAfterRead(opened.descriptor, opened.stat, opened.candidate);
|
|
128
|
+
return state.digest("hex");
|
|
129
|
+
} finally {
|
|
130
|
+
fs.closeSync(opened.descriptor);
|
|
131
|
+
}
|
|
132
|
+
}
|