@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 +1,48 @@
|
|
|
1
|
+
import { execFileSync } from "node:child_process";
|
|
2
|
+
import { createHash } from "node:crypto";
|
|
3
|
+
import fs from "node:fs";
|
|
4
|
+
import { gitEvidenceRootDetails } from "./workspace-evidence-roots.js";
|
|
5
|
+
|
|
6
|
+
export const WORKSPACE_REVISION_VERSION = "workspace-revision-v1";
|
|
7
|
+
const hash = (value) => createHash("sha256").update(value).digest("hex");
|
|
8
|
+
const PREFIX = `${WORKSPACE_REVISION_VERSION}:`;
|
|
9
|
+
export const isWorkspaceRevisionDigest = (value) => typeof value === "string" && /^workspace-revision-v1:[a-f0-9]{64}$/.test(value);
|
|
10
|
+
|
|
11
|
+
function git(cwd, args) {
|
|
12
|
+
return execFileSync("git", ["-c", "core.fsmonitor=false", "--no-optional-locks", "-C", cwd, ...args], {
|
|
13
|
+
encoding: "utf8", timeout: 3000, maxBuffer: 8192, stdio: ["ignore", "pipe", "pipe"]
|
|
14
|
+
}).trim();
|
|
15
|
+
}
|
|
16
|
+
|
|
17
|
+
export function repositoryHeadIdentity(cwd) {
|
|
18
|
+
try {
|
|
19
|
+
const head = git(cwd, ["rev-parse", "--verify", "--quiet", "HEAD"]);
|
|
20
|
+
if (!/^(?:[a-f0-9]{40}|[a-f0-9]{64})$/.test(head)) throw new Error("Invalid repository revision");
|
|
21
|
+
return head;
|
|
22
|
+
} catch (error) {
|
|
23
|
+
if (error.status !== 1) throw error;
|
|
24
|
+
const reference = git(cwd, ["symbolic-ref", "--quiet", "HEAD"]);
|
|
25
|
+
if (!reference.startsWith("refs/heads/")) throw new Error("Invalid unborn repository");
|
|
26
|
+
try { git(cwd, ["show-ref", "--verify", "--quiet", reference]); }
|
|
27
|
+
catch (missing) { if (missing.status === 1) return `unborn:${reference}`; throw missing; }
|
|
28
|
+
throw new Error("Repository revision changed during capture");
|
|
29
|
+
}
|
|
30
|
+
}
|
|
31
|
+
|
|
32
|
+
/** Baseline identity, not a content digest or authenticated verifier receipt. */
|
|
33
|
+
export function captureWorkspaceRevision(cwd) {
|
|
34
|
+
const root = fs.realpathSync.native(cwd);
|
|
35
|
+
const details = gitEvidenceRootDetails(root);
|
|
36
|
+
if (!details.complete || details.roots.length === 0 || details.roots.length > 128) throw new Error("Workspace baseline inventory is unavailable");
|
|
37
|
+
const repositories = details.roots.map((item) => Object.freeze({
|
|
38
|
+
prefix: item.prefix,
|
|
39
|
+
rootId: hash(fs.realpathSync.native(git(item.cwd, ["rev-parse", "--show-toplevel"]))),
|
|
40
|
+
head: repositoryHeadIdentity(item.cwd)
|
|
41
|
+
}));
|
|
42
|
+
const binding = Object.freeze({ version: WORKSPACE_REVISION_VERSION, projectId: hash(root), repositories: Object.freeze(repositories) });
|
|
43
|
+
return Object.freeze({ ...binding, digest: `${PREFIX}${hash(JSON.stringify(binding))}` });
|
|
44
|
+
}
|
|
45
|
+
|
|
46
|
+
export function currentWorkspaceRevisionDigest(cwd) {
|
|
47
|
+
try { return captureWorkspaceRevision(cwd).digest; } catch { return null; }
|
|
48
|
+
}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import { workingTreeSnapshot } from "./task-state.js";
|
|
2
|
+
import { workingTreeObservation } from "./working-tree-digest.js";
|
|
3
|
+
import { currentWorkspaceRevisionDigest } from "./workspace-revision.js";
|
|
4
|
+
|
|
5
|
+
/** Bracket content capture with baseline checks; neither alone proves current code. */
|
|
6
|
+
export function captureWorkspaceVerificationSnapshot(cwd, options = {}) {
|
|
7
|
+
const before = currentWorkspaceRevisionDigest(cwd);
|
|
8
|
+
const observation = workingTreeObservation(workingTreeSnapshot(cwd, options));
|
|
9
|
+
const after = currentWorkspaceRevisionDigest(cwd);
|
|
10
|
+
const workspaceRevisionDigest = before !== null && before === after ? before : null;
|
|
11
|
+
return Object.freeze({ ...observation, workspaceRevisionDigest,
|
|
12
|
+
proofCapable: observation.proofCapable && workspaceRevisionDigest !== null });
|
|
13
|
+
}
|
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@piagent/core",
|
|
3
|
-
"version": "1.
|
|
3
|
+
"version": "1.7.1",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"private": true,
|
|
6
6
|
"license": "MIT",
|
|
@@ -33,8 +33,8 @@
|
|
|
33
33
|
}
|
|
34
34
|
},
|
|
35
35
|
"peerDependencies": {
|
|
36
|
-
"@earendil-works/pi-coding-agent": "0.
|
|
37
|
-
"typebox": "1.3.
|
|
36
|
+
"@earendil-works/pi-coding-agent": "0.86.1",
|
|
37
|
+
"typebox": "1.3.27"
|
|
38
38
|
},
|
|
39
39
|
"peerDependenciesMeta": {
|
|
40
40
|
"@earendil-works/pi-coding-agent": {
|
|
@@ -13,7 +13,7 @@ type ContextUsage = {
|
|
|
13
13
|
|
|
14
14
|
export type AdaptiveContextPlan = {
|
|
15
15
|
schemaVersion: 1;
|
|
16
|
-
phase: "utility" | "protected" | "intake" | "scout" | "execute" | "review" | "release";
|
|
16
|
+
phase: "utility" | "protected" | "intake" | "scout" | "discuss" | "plan" | "execute" | "review" | "release";
|
|
17
17
|
lane: "tiny" | "normal" | "high-risk";
|
|
18
18
|
shouldInject: boolean;
|
|
19
19
|
budgetTokens: number;
|
|
@@ -64,6 +64,8 @@ function phaseFor(signal: ReturnType<typeof classifyContextTask>, input: Planner
|
|
|
64
64
|
if (signal.workflow === "usage") return "utility";
|
|
65
65
|
if (signal.workflow === "review") return "review";
|
|
66
66
|
if (signal.workflow === "scout" || signal.workflow === "onboard") return "scout";
|
|
67
|
+
if (signal.workflow === "discuss") return "discuss";
|
|
68
|
+
if (signal.workflow === "plan") return "plan";
|
|
67
69
|
if (signal.workflow === "release") return "release";
|
|
68
70
|
if (input.runtimeIntake) return "intake";
|
|
69
71
|
return "execute";
|
|
@@ -74,6 +76,8 @@ function baseBudget(phase: AdaptiveContextPlan["phase"], lane: AdaptiveContextPl
|
|
|
74
76
|
if (phase === "review") return lane === "high-risk" ? 720 : 520;
|
|
75
77
|
if (phase === "release") return 620;
|
|
76
78
|
if (phase === "scout") return lane === "high-risk" ? 820 : 640;
|
|
79
|
+
if (phase === "plan") return lane === "high-risk" ? 680 : 520;
|
|
80
|
+
if (phase === "discuss") return lane === "high-risk" ? 480 : 360;
|
|
77
81
|
if (phase === "intake") {
|
|
78
82
|
if (explicitPaths > 0) return lane === "high-risk" ? 560 : 420;
|
|
79
83
|
return lane === "high-risk" ? 780 : lane === "tiny" ? 360 : 540;
|
|
@@ -130,13 +134,13 @@ export function planAdaptiveContext(input: PlannerInput): AdaptiveContextPlan {
|
|
|
130
134
|
const limit = explicitPathCount > 0
|
|
131
135
|
? 4
|
|
132
136
|
: lane === "high-risk" ? 8 : lane === "tiny" ? 4 : 6;
|
|
133
|
-
const includeCode = phase === "intake" || phase === "review" || explicitPathCount > 0;
|
|
137
|
+
const includeCode = phase === "intake" || phase === "plan" || phase === "review" || explicitPathCount > 0;
|
|
134
138
|
const includePatterns = phase === "intake" ? DEFAULT_INTAKE_PATTERNS : undefined;
|
|
135
139
|
const reranker = "off" as const;
|
|
136
140
|
if (process.env.PIAGENT_LOCAL_RERANKER === "on") reasons.push("local-reranker:unavailable");
|
|
137
141
|
const minConfidence = explicitPathCount > 0 || lane === "tiny"
|
|
138
142
|
? "medium"
|
|
139
|
-
: phase === "review" || phase === "release" || lane === "high-risk"
|
|
143
|
+
: phase === "plan" || phase === "review" || phase === "release" || lane === "high-risk"
|
|
140
144
|
? "high"
|
|
141
145
|
: "medium";
|
|
142
146
|
return {
|
|
@@ -1,92 +1,67 @@
|
|
|
1
1
|
import crypto from "node:crypto";
|
|
2
|
-
import type { ExtensionAPI
|
|
2
|
+
import type { ExtensionAPI } from "@earendil-works/pi-coding-agent";
|
|
3
3
|
|
|
4
4
|
import {
|
|
5
5
|
buildContextPack,
|
|
6
6
|
classifyContextTask,
|
|
7
7
|
ensureContextIndexV2,
|
|
8
|
-
estimateContextTokens
|
|
8
|
+
estimateContextTokens,
|
|
9
|
+
shouldIndexPath
|
|
9
10
|
} from "../../extensions/context-engine.js";
|
|
10
11
|
import {
|
|
11
12
|
buildSelectedContextPack,
|
|
12
13
|
composeCriterionContextEntries
|
|
13
14
|
} from "../../extensions/criterion-context-pack.js";
|
|
14
15
|
import { matchesProtectedPath } from "../../extensions/policy-core.js";
|
|
16
|
+
import { hasDurableContextEvidence } from "../../extensions/context-evidence.js";
|
|
15
17
|
import { selectRepositoryMemoryFacts } from "../../extensions/repository-memory.js";
|
|
16
|
-
import type { TaskContract } from "../../extensions/guard-types.js";
|
|
17
18
|
import {
|
|
18
19
|
contextPlanAcceptsConfidence,
|
|
19
20
|
planAdaptiveContext
|
|
20
21
|
} from "../context/adaptive-planner.ts";
|
|
21
22
|
import { stageContextDelivery } from "../context/context-delivery.ts";
|
|
22
|
-
import { measureContextDeltaShadow
|
|
23
|
+
import { measureContextDeltaShadow } from "../context/context-delta-shadow.ts";
|
|
23
24
|
import { buildPrefixTelemetry } from "../context/prefix-telemetry.ts";
|
|
24
25
|
import { formatRepositoryMemoryHints } from "../context/repository-memory-hints.ts";
|
|
25
26
|
import { modelCapabilityFromContext } from "../model/capabilities.ts";
|
|
26
|
-
import type { RuntimeModelSnapshot } from "../model/runtime-snapshot.ts";
|
|
27
27
|
import type { ModelRouteEvaluation } from "../model/model-route-runtime.ts";
|
|
28
|
+
import type { RuntimeModelSnapshot } from "../model/runtime-snapshot.ts";
|
|
28
29
|
import { buildTaskResumeContext } from "../recovery/resume-state.ts";
|
|
29
30
|
import { planRetrievalRoute } from "../context/retrieval-route-policy.ts";
|
|
30
|
-
import type { SolverShadowEvaluation } from "../solver/solver-shadow.ts";
|
|
31
31
|
import { observeTrajectorySync } from "../trajectory/trajectory-observability.ts";
|
|
32
|
-
import type {
|
|
32
|
+
import type { SolverShadowEvaluation } from "../solver/solver-shadow.ts";
|
|
33
33
|
import { trajectoryRecommendationRef } from "../trajectory/trajectory-runtime.ts";
|
|
34
|
-
import {
|
|
34
|
+
import type { ContextInjectionItem } from "../session/runtime-state.ts";
|
|
35
|
+
import { isUncertainSendContinuation } from "../session/uncertain-send-continuation.ts";
|
|
35
36
|
import {
|
|
36
37
|
compactManagedProjectInstructions,
|
|
37
38
|
rewriteLegacyProjectInstructions
|
|
38
39
|
} from "../session/system-prompt.ts";
|
|
39
40
|
import { PIAGENT_TOOL_NAMES } from "../tools/tool-groups.ts";
|
|
40
|
-
import {
|
|
41
|
+
import { agentStartTaskRequest } from "../workflows/input-routing.ts";
|
|
41
42
|
import {
|
|
42
43
|
AUTO_INTAKE_SNAPSHOT_PATTERNS,
|
|
43
|
-
automaticTaskIntakeMode
|
|
44
|
+
automaticTaskIntakeMode,
|
|
45
|
+
isLightweightNonAuthorizingChangeContinuation,
|
|
46
|
+
isNonAuthorizingChangeClarification
|
|
44
47
|
} from "../workflows/task-intake.ts";
|
|
45
48
|
import { CONTEXT_PACK_MAX_TOKENS } from "../runtime-limits.ts";
|
|
46
49
|
import { modelRouteTelemetry, runtimeSnapshotTelemetry, solverShadowTelemetry } from "./agent-prompt-telemetry.ts";
|
|
47
|
-
|
|
48
|
-
type RuntimeIntakeResult = {
|
|
49
|
-
started: boolean;
|
|
50
|
-
text: string;
|
|
51
|
-
task?: TaskContract;
|
|
52
|
-
plannedContext?: Array<{ path: string; reason: string }>;
|
|
53
|
-
plannedContextComplete?: boolean;
|
|
54
|
-
};
|
|
55
|
-
|
|
56
|
-
type AgentStartHookDependencies = {
|
|
57
|
-
state: RuntimeSessionState;
|
|
58
|
-
autoContextEnabled: boolean;
|
|
59
|
-
contextDeltaShadowMode: ContextDeltaShadowMode;
|
|
60
|
-
activeTask: (ctx: ExtensionContext) => TaskContract | undefined;
|
|
61
|
-
readProtectedPaths: (ctx: ExtensionContext) => string[];
|
|
62
|
-
contextExcludePatterns: (ctx: ExtensionContext) => string[];
|
|
63
|
-
promptPackKey: (ctx: ExtensionContext, promptHash: string) => string;
|
|
64
|
-
retrievalKey: (ctx: ExtensionContext, query: string) => string;
|
|
65
|
-
startAutomaticTask: (query: string, ctx: ExtensionContext) => Promise<RuntimeIntakeResult | undefined>;
|
|
66
|
-
runtimeSnapshot?: (ctx: ExtensionContext) => RuntimeModelSnapshot | undefined;
|
|
67
|
-
persistRuntimeSnapshot?: (ctx: ExtensionContext, snapshot: RuntimeModelSnapshot) => unknown;
|
|
68
|
-
shadowSolver?: (input: {
|
|
69
|
-
request: string;
|
|
70
|
-
ctx: ExtensionContext;
|
|
71
|
-
activeTask?: TaskContract;
|
|
72
|
-
runtimeSnapshot?: RuntimeModelSnapshot;
|
|
73
|
-
protectedTarget: boolean;
|
|
74
|
-
}) => SolverShadowEvaluation;
|
|
75
|
-
modelRoute?: (input: {
|
|
76
|
-
ctx: ExtensionContext;
|
|
77
|
-
features: NonNullable<Extract<SolverShadowEvaluation, { status: "ok" }>["features"]>;
|
|
78
|
-
runtimeSnapshot?: RuntimeModelSnapshot;
|
|
79
|
-
}) => Promise<ModelRouteEvaluation>;
|
|
80
|
-
syncTrajectory?: (ctx: ExtensionContext, task: TaskContract, options: TrajectorySyncOptions) => TrajectorySyncResult;
|
|
81
|
-
telemetry: (ctx: ExtensionContext, payload: Record<string, unknown>) => void;
|
|
82
|
-
};
|
|
50
|
+
import type { AgentStartHookDependencies } from "./agent-start-types.ts";
|
|
83
51
|
|
|
84
52
|
export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStartHookDependencies): void {
|
|
85
53
|
pi.on("before_agent_start", async (event, ctx) => {
|
|
86
54
|
const projectInstructions = rewriteLegacyProjectInstructions(event.systemPrompt);
|
|
87
|
-
const query =
|
|
55
|
+
const query = agentStartTaskRequest(event.prompt);
|
|
88
56
|
const signal = classifyContextTask(query);
|
|
89
|
-
const
|
|
57
|
+
const uncertainSendContinuation = isUncertainSendContinuation(query);
|
|
58
|
+
// A mutation clarification still goes to the provider unchanged and never
|
|
59
|
+
// creates a task. Only a lightweight decision also skips repository retrieval.
|
|
60
|
+
const nonAuthorizingClarification = isNonAuthorizingChangeClarification(query);
|
|
61
|
+
const lightweightChangeContinuation = isLightweightNonAuthorizingChangeContinuation(query);
|
|
62
|
+
const turn = dependencies.state.classifyTurn(ctx, signal.promptHash, {
|
|
63
|
+
lightweightNonAuthorizingChange: lightweightChangeContinuation
|
|
64
|
+
});
|
|
90
65
|
const readProtectedPaths = dependencies.readProtectedPaths(ctx);
|
|
91
66
|
const protectedTarget = signal.paths.some((candidate) => matchesProtectedPath(candidate, readProtectedPaths));
|
|
92
67
|
const protectedOnlyTarget = signal.paths.length > 0
|
|
@@ -119,7 +94,9 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
119
94
|
const prefix = buildPrefixTelemetry(effectiveSystemPrompt, toolMetadata);
|
|
120
95
|
const toolSchemaTokens = estimateContextTokens(prefix.canonicalToolSurface);
|
|
121
96
|
const systemPromptTokens = estimateContextTokens(effectiveSystemPrompt);
|
|
122
|
-
const autoPackUseful =
|
|
97
|
+
const autoPackUseful = !lightweightChangeContinuation
|
|
98
|
+
&& !uncertainSendContinuation
|
|
99
|
+
&& activeTask?.trace.outcome !== "pending"
|
|
123
100
|
&& (runtimeIntake || signal.paths.length === 0);
|
|
124
101
|
let runtimeSnapshot: RuntimeModelSnapshot | undefined;
|
|
125
102
|
try {
|
|
@@ -172,10 +149,12 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
172
149
|
content: string;
|
|
173
150
|
details: Record<string, unknown>;
|
|
174
151
|
}) => {
|
|
175
|
-
const resumed = activeTask
|
|
152
|
+
const resumed = activeTask && !lightweightChangeContinuation
|
|
153
|
+
? dependencies.state.takeResumeContextState(ctx, activeTask.taskRunId)
|
|
154
|
+
: undefined;
|
|
176
155
|
const durableResume = activeTask && resumed ? buildTaskResumeContext(activeTask, resumed) : undefined;
|
|
177
156
|
const selectedContext = contextMessage ?? durableResume;
|
|
178
|
-
const intake = await dependencies.startAutomaticTask(query, ctx);
|
|
157
|
+
const intake = nonAuthorizingClarification ? undefined : await dependencies.startAutomaticTask(query, ctx);
|
|
179
158
|
if (intake?.task) observeTrajectorySync(ctx, dependencies.syncTrajectory?.(ctx, intake.task, { sourceHook: "agent-start", recommendationRef }), dependencies.telemetry);
|
|
180
159
|
if (!selectedContext && !intake) return systemPromptUpdate;
|
|
181
160
|
const discoveryItems = selectedContext?.customType === "piagent-context-pack-v2" && Array.isArray(selectedContext.details.selectedItems)
|
|
@@ -192,14 +171,23 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
192
171
|
const criterionLimit = Number.isFinite(Number(discoveryPlan?.limit))
|
|
193
172
|
? Math.max(1, Math.min(6, Math.trunc(Number(discoveryPlan?.limit))))
|
|
194
173
|
: signal.paths.length > 0 ? 3 : 4;
|
|
195
|
-
const
|
|
196
|
-
const
|
|
174
|
+
const terminalUncertainSend = intake?.continuation === "terminal-uncertain-send";
|
|
175
|
+
const criterionMode = dependencies.autoContextEnabled
|
|
176
|
+
&& !terminalUncertainSend
|
|
177
|
+
&& intake?.task?.criterionGraph?.mode === "criterion-graph";
|
|
178
|
+
const contextExcludePatterns = dependencies.contextExcludePatterns(ctx);
|
|
179
|
+
// Apply the pack's path eligibility before ranking a singleton fallback.
|
|
180
|
+
// Otherwise an unsupported/protected candidate can displace a usable
|
|
181
|
+
// planned source and then be discarded by the pack builder.
|
|
182
|
+
const packEligible = (entry: { path?: unknown }) => typeof entry.path === "string"
|
|
183
|
+
&& shouldIndexPath(entry.path, { excludePatterns: contextExcludePatterns });
|
|
184
|
+
const criterionEntries = !terminalUncertainSend && intake?.task?.criterionGraph?.mode === "criterion-graph"
|
|
197
185
|
? composeCriterionContextEntries({
|
|
198
186
|
explicitPaths: signal.paths,
|
|
199
187
|
criteria: intake.task.acceptanceCriteria,
|
|
200
|
-
plannedEntries: intake.plannedContext ?? [],
|
|
188
|
+
plannedEntries: (intake.plannedContext ?? []).filter(packEligible),
|
|
201
189
|
plannedSelectionComplete: intake.plannedContextComplete === true,
|
|
202
|
-
retrievedItems: discoveryItems
|
|
190
|
+
retrievedItems: discoveryItems.filter(packEligible)
|
|
203
191
|
}, { limit: criterionLimit })
|
|
204
192
|
: [];
|
|
205
193
|
const criterionContext = criterionMode
|
|
@@ -207,7 +195,7 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
207
195
|
budgetTokens: criterionBudget,
|
|
208
196
|
limit: criterionLimit,
|
|
209
197
|
focusText: [query, ...intake.task.acceptanceCriteria].join("\n"),
|
|
210
|
-
excludePatterns:
|
|
198
|
+
excludePatterns: contextExcludePatterns
|
|
211
199
|
})
|
|
212
200
|
: undefined;
|
|
213
201
|
const criterionReasonCode = !dependencies.autoContextEnabled
|
|
@@ -227,7 +215,13 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
227
215
|
});
|
|
228
216
|
const composedContext = criterionContext?.selected.length ? criterionContext : undefined;
|
|
229
217
|
const deliveredSelectedContext = criterionMode ? undefined : selectedContext;
|
|
230
|
-
|
|
218
|
+
// An oversized file without evidenced ranges can legitimately produce an
|
|
219
|
+
// empty pack. Make the existing read-evidence requirement actionable;
|
|
220
|
+
// planned paths and earlier turns are not evidence of current delivery.
|
|
221
|
+
const contextReadGuidance = criterionMode && !composedContext && !hasDurableContextEvidence(intake?.task)
|
|
222
|
+
? "No file content was delivered in the bounded context pack. Before completion, use the native read tool on a relevant source or test file in the task scope; use offset/limit for a large file. Shell inspection and passing tests do not record this read evidence. Do not reread the whole repository."
|
|
223
|
+
: undefined;
|
|
224
|
+
const content = [deliveredSelectedContext?.content, intake?.text, composedContext?.text, contextReadGuidance].filter(Boolean).join("\n\n");
|
|
231
225
|
const deliveryTask = intake?.task ?? activeTask;
|
|
232
226
|
const selectedPackPaths = deliveredSelectedContext?.customType === "piagent-context-pack-v2" && Array.isArray(deliveredSelectedContext.details.paths)
|
|
233
227
|
? deliveredSelectedContext.details.paths.filter((value): value is string => typeof value === "string")
|
|
@@ -303,6 +297,13 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
303
297
|
intakeMode: intake.task.intakeMode
|
|
304
298
|
}
|
|
305
299
|
: undefined,
|
|
300
|
+
uncertainSendContinuation: terminalUncertainSend
|
|
301
|
+
? {
|
|
302
|
+
taskId: intake?.task?.taskId ?? null,
|
|
303
|
+
taskRunId: intake?.task?.taskRunId ?? null,
|
|
304
|
+
replacementTaskStarted: false
|
|
305
|
+
}
|
|
306
|
+
: undefined,
|
|
306
307
|
criterionContext: composedContext ? {
|
|
307
308
|
paths: composedContext.selected.map((entry) => entry.path), estimatedTokens: composedContext.estimatedTokens
|
|
308
309
|
} : undefined,
|
|
@@ -313,7 +314,7 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
313
314
|
};
|
|
314
315
|
|
|
315
316
|
const packKey = dependencies.promptPackKey(ctx, signal.promptHash);
|
|
316
|
-
if (!autoPackUseful && activeTask?.trace.outcome === "pending" && query.length >= 20 && signal.workflow !== "usage" && dependencies.autoContextEnabled) {
|
|
317
|
+
if (!lightweightChangeContinuation && !autoPackUseful && activeTask?.trace.outcome === "pending" && query.length >= 20 && signal.workflow !== "usage" && dependencies.autoContextEnabled) {
|
|
317
318
|
await measureContextDeltaShadow({ ctx, query, turnId: turn.turnId, task: activeTask, mode: dependencies.contextDeltaShadowMode, protectedTarget, excludePatterns: dependencies.contextExcludePatterns(ctx), telemetry: dependencies.telemetry });
|
|
318
319
|
}
|
|
319
320
|
if (
|
|
@@ -329,7 +330,7 @@ export function registerAgentStartHook(pi: ExtensionAPI, dependencies: AgentStar
|
|
|
329
330
|
dependencies.state.rememberAutoPackedPrompt(packKey);
|
|
330
331
|
try {
|
|
331
332
|
const excludePatterns = dependencies.contextExcludePatterns(ctx);
|
|
332
|
-
const ensured = await ensureContextIndexV2(ctx.cwd, {
|
|
333
|
+
const ensured = await (dependencies.ensureContextIndex ?? ensureContextIndexV2)(ctx.cwd, {
|
|
333
334
|
excludePatterns,
|
|
334
335
|
rebuildMissing: false
|
|
335
336
|
});
|
|
@@ -0,0 +1,48 @@
|
|
|
1
|
+
import type { ExtensionContext } from "@earendil-works/pi-coding-agent";
|
|
2
|
+
|
|
3
|
+
import type { TaskContract } from "../../extensions/guard-types.js";
|
|
4
|
+
import type { ContextDeltaShadowMode } from "../context/context-delta-shadow.ts";
|
|
5
|
+
import type { ModelRouteEvaluation } from "../model/model-route-runtime.ts";
|
|
6
|
+
import type { RuntimeModelSnapshot } from "../model/runtime-snapshot.ts";
|
|
7
|
+
import type { RuntimeSessionState } from "../session/runtime-state.ts";
|
|
8
|
+
import type { SolverShadowEvaluation } from "../solver/solver-shadow.ts";
|
|
9
|
+
import type { TrajectorySyncOptions, TrajectorySyncResult } from "../trajectory/trajectory-runtime.ts";
|
|
10
|
+
import type { ensureContextIndexV2 } from "../../extensions/context-engine.js";
|
|
11
|
+
|
|
12
|
+
export type RuntimeIntakeResult = {
|
|
13
|
+
started: boolean;
|
|
14
|
+
text: string;
|
|
15
|
+
task?: TaskContract;
|
|
16
|
+
plannedContext?: Array<{ path: string; reason: string }>;
|
|
17
|
+
plannedContextComplete?: boolean;
|
|
18
|
+
continuation?: "terminal-uncertain-send";
|
|
19
|
+
};
|
|
20
|
+
|
|
21
|
+
export type AgentStartHookDependencies = {
|
|
22
|
+
state: RuntimeSessionState;
|
|
23
|
+
autoContextEnabled: boolean;
|
|
24
|
+
contextDeltaShadowMode: ContextDeltaShadowMode;
|
|
25
|
+
activeTask: (ctx: ExtensionContext) => TaskContract | undefined;
|
|
26
|
+
readProtectedPaths: (ctx: ExtensionContext) => string[];
|
|
27
|
+
contextExcludePatterns: (ctx: ExtensionContext) => string[];
|
|
28
|
+
ensureContextIndex?: typeof ensureContextIndexV2;
|
|
29
|
+
promptPackKey: (ctx: ExtensionContext, promptHash: string) => string;
|
|
30
|
+
retrievalKey: (ctx: ExtensionContext, query: string) => string;
|
|
31
|
+
startAutomaticTask: (query: string, ctx: ExtensionContext) => Promise<RuntimeIntakeResult | undefined>;
|
|
32
|
+
runtimeSnapshot?: (ctx: ExtensionContext) => RuntimeModelSnapshot | undefined;
|
|
33
|
+
persistRuntimeSnapshot?: (ctx: ExtensionContext, snapshot: RuntimeModelSnapshot) => unknown;
|
|
34
|
+
shadowSolver?: (input: {
|
|
35
|
+
request: string;
|
|
36
|
+
ctx: ExtensionContext;
|
|
37
|
+
activeTask?: TaskContract;
|
|
38
|
+
runtimeSnapshot?: RuntimeModelSnapshot;
|
|
39
|
+
protectedTarget: boolean;
|
|
40
|
+
}) => SolverShadowEvaluation;
|
|
41
|
+
modelRoute?: (input: {
|
|
42
|
+
ctx: ExtensionContext;
|
|
43
|
+
features: NonNullable<Extract<SolverShadowEvaluation, { status: "ok" }>["features"]>;
|
|
44
|
+
runtimeSnapshot?: RuntimeModelSnapshot;
|
|
45
|
+
}) => Promise<ModelRouteEvaluation>;
|
|
46
|
+
syncTrajectory?: (ctx: ExtensionContext, task: TaskContract, options: TrajectorySyncOptions) => TrajectorySyncResult;
|
|
47
|
+
telemetry: (ctx: ExtensionContext, payload: Record<string, unknown>) => void;
|
|
48
|
+
};
|