@piagent/platform 1.6.0 → 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 +104 -0
- package/README.md +24 -15
- package/README.vi.md +36 -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 +27 -3
- package/benchmarks/production-v1/suite.json +5 -1
- 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 +16 -16
- package/docs/authority-manifest.md +1 -1
- package/docs/auto-delegation-policy.md +38 -73
- 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 +30 -35
- package/docs/context-engine.md +5 -3
- package/docs/context-window-policy.md +37 -1
- package/docs/distribution-standard.md +9 -9
- package/docs/en/architecture.md +26 -13
- package/docs/languages.json +4 -4
- package/docs/operator-manual-vietnamese.md +26 -46
- 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 +7 -0
- package/docs/quality-benchmark.md +299 -23
- 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/runtime-harness-standard.md +3 -3
- package/docs/security-threat-model.md +1 -1
- package/docs/subagent-orchestration-capabilities.md +19 -51
- package/docs/subagents-and-multiagent.md +21 -47
- package/docs/task-implementation-contract.md +3 -3
- package/docs/task-lifecycle-vietnamese.md +1 -1
- package/docs/team-onboarding.md +7 -7
- package/docs/third-party-neutrality.md +7 -0
- package/docs/usage-observability.md +1 -1
- package/docs/vi/architecture.md +26 -13
- package/docs-site/architecture.html +1 -1
- package/docs-site/benchmark.html +89 -95
- package/docs-site/commands.html +1 -1
- package/docs-site/content/benchmark.html +88 -94
- package/docs-site/content/en/benchmark.html +134 -90
- package/docs-site/content/en/index.html +19 -19
- package/docs-site/content/en/runtime.html +5 -5
- package/docs-site/content/en/whats-new.html +23 -1
- package/docs-site/content/en/workflows.html +2 -2
- package/docs-site/content/index.html +19 -19
- package/docs-site/content/runtime.html +6 -6
- package/docs-site/content/whats-new.html +23 -1
- package/docs-site/content/workflows.html +2 -2
- package/docs-site/context-engine.html +1 -1
- package/docs-site/en/architecture.html +1 -1
- package/docs-site/en/benchmark.html +135 -91
- package/docs-site/en/commands.html +1 -1
- package/docs-site/en/context-engine.html +1 -1
- package/docs-site/en/index.html +20 -20
- 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 +6 -6
- package/docs-site/en/scope.html +1 -1
- package/docs-site/en/security.html +1 -1
- package/docs-site/en/team.html +1 -1
- package/docs-site/en/whats-new.html +24 -2
- package/docs-site/en/workflows.html +3 -3
- package/docs-site/index.html +20 -20
- 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 +7 -7
- package/docs-site/scope.html +1 -1
- package/docs-site/security.html +1 -1
- package/docs-site/team.html +1 -1
- package/docs-site/whats-new.html +24 -2
- package/docs-site/workflows.html +3 -3
- 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-assurance-evidence.invalid.json +1 -1
- package/evals/fixtures/benchmark-assurance-evidence.valid.json +1 -1
- 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/codex-relative-efficiency-v1.invalid.json +4 -0
- package/evals/fixtures/codex-relative-efficiency-v1.valid.json +94 -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/helper-request.invalid.json +1 -1
- package/evals/fixtures/helper-request.valid.json +1 -1
- package/evals/fixtures/host-contract-plan.invalid.json +42 -0
- package/evals/fixtures/host-contract-plan.valid.json +41 -0
- package/evals/fixtures/piagent-webui/subagent-tree-v1.valid.json +7 -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/fixtures/task-authority-snapshot.valid.json +4 -4
- package/evals/fs4-readiness-matrix.v1.json +2 -2
- 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 +10 -6
- package/evals/fs5-pilot-protocol.v1.json +3 -3
- package/evals/fs5-pilot-protocol.v2.json +8 -8
- package/evals/fs5-pilot-protocol.v3.json +8 -8
- package/evals/fs5-pilot-protocol.v4.json +7 -7
- package/evals/fs5-pilot-protocol.v5.json +10 -8
- 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.v1.json +1 -1
- package/evals/private-holdout-v1/public-exposure.v2.json +273 -0
- package/evals/runtime-conformance-v1/README.md +43 -0
- package/evals/runtime-conformance-v1/evaluate-paid-canary.mjs +198 -0
- package/evals/runtime-conformance-v1/lane.json +57 -0
- package/evals/runtime-conformance-v1/paid-canary.json +31 -0
- package/evals/runtime-conformance-v1/runner.mjs +456 -0
- package/package.json +9 -4
- package/packages/piagent-core/README.md +15 -6
- 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 +699 -0
- 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 +111 -110
- 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 +67 -52
- 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 +70 -7
- 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 +462 -0
- 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 +343 -8
- package/packages/piagent-core/benchmark/benchmark-release-claim-controls.js +132 -0
- package/packages/piagent-core/benchmark/benchmark-report.js +94 -5
- 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-causal.js +244 -0
- 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 +246 -33
- 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 +232 -4
- package/packages/piagent-core/benchmark/benchmark-summary-support.js +33 -1
- 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 +410 -37
- 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 +6 -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 +1000 -0
- 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 +208 -180
- 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 +12 -17
- package/packages/piagent-core/extensions/context-index-policy.js +4 -0
- package/packages/piagent-core/extensions/context-query-paths.js +57 -0
- package/packages/piagent-core/extensions/criterion-context-pack.js +109 -4
- package/packages/piagent-core/extensions/criterion-graph.js +4 -1
- package/packages/piagent-core/extensions/failure-types.ts +2 -2
- package/packages/piagent-core/extensions/fallback-base-policy.ts +101 -0
- package/packages/piagent-core/extensions/guard-types.ts +9 -1
- package/packages/piagent-core/extensions/piagent-guard.ts +374 -364
- 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-lifecycle.js +32 -0
- package/packages/piagent-core/extensions/task-runtime-audit.js +8 -2
- package/packages/piagent-core/extensions/task-state.js +140 -43
- 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 +109 -20
- package/packages/piagent-core/extensions/workflow-catalog.ts +98 -0
- package/packages/piagent-core/extensions/workspace-evidence-roots.js +52 -16
- 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 +43 -2
- package/packages/piagent-core/policy/authority-manifest.v1.json +3 -3
- package/packages/piagent-core/prompts/be-to-fe.md +2 -8
- package/packages/piagent-core/prompts/plan.md +1 -1
- package/packages/piagent-core/prompts/platform-improve.md +2 -6
- package/packages/piagent-core/prompts/review.md +2 -7
- package/packages/piagent-core/prompts/scout.md +1 -1
- package/packages/piagent-core/prompts/task.md +4 -3
- 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 +83 -89
- package/packages/piagent-core/runtime/hooks/input-hook.ts +98 -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 +150 -19
- package/packages/piagent-core/runtime/hooks/tool-result-hook.ts +111 -92
- 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/hooks/tool-result-value-helpers.ts +66 -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/bounded-emission-guard.ts +286 -0
- package/packages/piagent-core/runtime/inspection/runtime-event-store.ts +37 -7
- 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/task-subagent-tree.ts +27 -4
- package/packages/piagent-core/runtime/inspection/tool-failure-classification.ts +69 -11
- package/packages/piagent-core/runtime/inspection/webui-snapshot.ts +104 -22
- 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/orchestration/helper-lifecycle.ts +16 -12
- package/packages/piagent-core/runtime/orchestration/minimal-delegation-policy.ts +71 -0
- package/packages/piagent-core/runtime/orchestration/owned-work-budget.ts +28 -20
- package/packages/piagent-core/runtime/orchestration/role-policy.ts +27 -9
- package/packages/piagent-core/runtime/orchestration/subagent-tool-policy.ts +329 -0
- package/packages/piagent-core/runtime/policy/block-remedy.ts +0 -4
- 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 +15 -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/edit-freshness-guard.ts +102 -0
- package/packages/piagent-core/runtime/quality/exact-output-contract.ts +20 -10
- package/packages/piagent-core/runtime/quality/performance-assurance.ts +3 -3
- 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 +12 -3
- package/packages/piagent-core/runtime/recovery/diagnostic-delivery.ts +61 -0
- package/packages/piagent-core/runtime/recovery/handoff-projection.ts +196 -58
- 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/pristine-task-policy-refresh.ts +91 -0
- package/packages/piagent-core/runtime/recovery/recovery-policy.ts +31 -40
- package/packages/piagent-core/runtime/recovery/resume-state.ts +120 -19
- package/packages/piagent-core/runtime/recovery/semantic-repair-handshake.ts +7 -13
- package/packages/piagent-core/runtime/recovery/task-evidence-coverage-refresh.ts +217 -0
- package/packages/piagent-core/runtime/registration/activity-inspector-command.ts +2 -1
- package/packages/piagent-core/runtime/registration/context-commands.ts +28 -10
- package/packages/piagent-core/runtime/registration/knowledge-tools.ts +9 -5
- package/packages/piagent-core/runtime/registration/operator-catalogs.ts +4 -9
- package/packages/piagent-core/runtime/registration/operator-request-intake.ts +49 -0
- package/packages/piagent-core/runtime/registration/policy-tools.ts +8 -6
- package/packages/piagent-core/runtime/registration/session-commands.ts +10 -8
- package/packages/piagent-core/runtime/registration/session-wait-tool.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-completion-tools.ts +39 -20
- package/packages/piagent-core/runtime/registration/task-evidence-tools.ts +21 -12
- package/packages/piagent-core/runtime/registration/task-start-conditional-mutation.ts +45 -0
- package/packages/piagent-core/runtime/registration/task-start-guidance.ts +24 -15
- package/packages/piagent-core/runtime/registration/task-start-manifest.ts +54 -0
- package/packages/piagent-core/runtime/registration/task-start-tool.ts +79 -79
- package/packages/piagent-core/runtime/registration/workflow-commands.ts +64 -93
- package/packages/piagent-core/runtime/runtime-limits.ts +16 -2
- package/packages/piagent-core/runtime/session/adaptive-context-analysis.ts +461 -0
- package/packages/piagent-core/runtime/session/adaptive-context-governor.ts +464 -0
- package/packages/piagent-core/runtime/session/adaptive-context-ledger.ts +265 -0
- package/packages/piagent-core/runtime/session/completion-signals.ts +23 -0
- package/packages/piagent-core/runtime/session/message-signals.ts +6 -18
- package/packages/piagent-core/runtime/session/operator-request-carry.ts +44 -0
- package/packages/piagent-core/runtime/session/retrieval-evidence-runtime.ts +188 -0
- package/packages/piagent-core/runtime/session/runtime-state.ts +37 -39
- package/packages/piagent-core/runtime/session/source-checkout-read-grants.ts +35 -0
- package/packages/piagent-core/runtime/session/system-prompt.ts +76 -14
- package/packages/piagent-core/runtime/session/task-contract-presentation.ts +39 -0
- 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 +10 -9
- package/packages/piagent-core/runtime/tools/session-wait.ts +42 -0
- package/packages/piagent-core/runtime/tools/tool-groups.ts +22 -5
- 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 +101 -0
- 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 +201 -0
- package/packages/piagent-core/runtime/workflows/task-followup-policy.ts +163 -0
- package/packages/piagent-core/runtime/workflows/task-intake.ts +276 -169
- 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-core/subagents/piagent-oracle.md +5 -3
- package/packages/piagent-core/subagents/piagent-planner.md +6 -4
- package/packages/piagent-core/subagents/piagent-reviewer.md +5 -3
- package/packages/piagent-core/subagents/piagent-scout.md +4 -2
- package/packages/piagent-core/subagents/piagent-worker.md +6 -4
- package/packages/piagent-webui/dist/client/assets/{MarkdownMessage-CxeZqTte.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 +13 -3
- package/packages/piagent-webui/extension/review-controller.ts +29 -6
- package/packages/piagent-webui/extension/same-session-bridge.ts +7 -8
- package/packages/piagent-webui/extension/session-stream-adapter.ts +88 -16
- 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 +170 -32
- 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-lifecycle.ts +237 -0
- package/packages/piagent-webui/gateway/session-operation-retry-control.ts +38 -0
- package/packages/piagent-webui/gateway/session-operation-settlement.ts +28 -0
- package/packages/piagent-webui/gateway/session-operation-watchdog.ts +9 -2
- 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 +118 -117
- package/packages/piagent-webui/ownership/session-lease-store.ts +5 -2
- package/packages/piagent-webui/server/core-inspection-provider.ts +75 -15
- 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 +150 -11
- 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/codex-relative-efficiency-v1.schema.json +155 -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 +46 -5
- 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/subagent-tree-v1.schema.json +11 -1
- package/schemas/piagent-webui/transcript-v1.schema.json +2 -1
- package/schemas/project-profile.schema.json +1 -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 +45 -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 +224 -387
- package/scripts/benchmark-runner-finalization.mjs +144 -27
- package/scripts/benchmark-runner-provider-boundary.mjs +44 -0
- package/scripts/benchmark-runner-schedule.mjs +256 -0
- package/scripts/benchmark-runner-support.mjs +136 -4
- package/scripts/benchmark-runner.mjs +95 -18
- package/scripts/benchmark-scoped-broker-custody.mjs +483 -0
- package/scripts/benchmark-scoped-docs-verifier.mjs +203 -0
- package/scripts/benchmark-scoped-frozen-qualification.mjs +676 -0
- package/scripts/benchmark-scoped-pi-router.mjs +196 -0
- package/scripts/benchmark-scoped-project-verification-receipt.mjs +49 -0
- package/scripts/benchmark-scoped-project-verifier.mjs +461 -0
- package/scripts/benchmark-scoped-session-custody.mjs +599 -0
- package/scripts/benchmark-scoped-session-factory.mjs +342 -0
- package/scripts/benchmark-scoped-tool-broker.mjs +523 -0
- package/scripts/benchmark-scoped-verification-supervisor.mjs +773 -0
- package/scripts/benchmark-session-evaluator.mjs +169 -0
- package/scripts/benchmark-session-provider-boundary.mjs +62 -0
- package/scripts/benchmark-session-usage.mjs +28 -0
- package/scripts/benchmark-session.mjs +343 -50
- package/scripts/benchmark-webui-journey.mjs +678 -0
- package/scripts/check-public-wording.mjs +46 -0
- package/scripts/configure-subagents.sh +42 -34
- package/scripts/fs4-readiness-evaluation.mjs +7 -1
- package/scripts/helper-evaluation.mjs +1 -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/team-doctor.sh +1 -1
- package/scripts/typescript-loader.mjs +135 -1
- package/scripts/update-global.mjs +6 -7
- package/scripts/verify-local.sh +41 -55
- package/scripts/verify-release-identity.mjs +5 -3
- package/packages/piagent-webui/dist/client/assets/index-DynIMe8I.js +0 -176
|
@@ -0,0 +1,461 @@
|
|
|
1
|
+
import crypto from "node:crypto";
|
|
2
|
+
|
|
3
|
+
import { redactSensitiveText } from "../../extensions/redaction-core.js";
|
|
4
|
+
import {
|
|
5
|
+
CONTEXT_GOVERNOR_PRESSURE_TOKENS,
|
|
6
|
+
CONTEXT_GOVERNOR_SEMANTIC_BOUNDARY_TOKENS,
|
|
7
|
+
CONTEXT_GOVERNOR_TARGET_MESSAGE_TOKENS,
|
|
8
|
+
CONTEXT_GOVERNOR_TOOL_ROUND_LIMIT,
|
|
9
|
+
CONTEXT_PROVIDER_PROMPT_CEILING_TOKENS
|
|
10
|
+
} from "../runtime-limits.ts";
|
|
11
|
+
|
|
12
|
+
export const GOVERNOR_VERSION = "adaptive-context-governor-v2" as const;
|
|
13
|
+
export const CONTEXT_GOVERNOR_MIN_SAVINGS_FLOOR_TOKENS = 8_000;
|
|
14
|
+
export const CONTEXT_GOVERNOR_MIN_SAVINGS_CEILING_TOKENS = 20_000;
|
|
15
|
+
export const CONTEXT_GOVERNOR_MIN_SAVINGS_RATIO = 0.15;
|
|
16
|
+
export const CONTEXT_GOVERNOR_PROTECTED_BOUNDARY_GROUPS = 6;
|
|
17
|
+
const READ_TOOLS = new Set([
|
|
18
|
+
"read",
|
|
19
|
+
"grep",
|
|
20
|
+
"find",
|
|
21
|
+
"ls",
|
|
22
|
+
"glob",
|
|
23
|
+
"search",
|
|
24
|
+
"fetch_content",
|
|
25
|
+
"get_search_content",
|
|
26
|
+
"web_search"
|
|
27
|
+
]);
|
|
28
|
+
const MUTATION_TOOLS = new Set(["edit", "write", "apply_patch", "patch"]);
|
|
29
|
+
const VERIFIER_PATTERN = /(?:^|\s)(?:test|tests|lint|typecheck|check|verify|build|pytest|vitest|jest|playwright|cargo\s+test|go\s+test)(?:\s|:|$)/i;
|
|
30
|
+
const MUTATION_COMMAND_PATTERN = /(?:^|\s)(?:sed\s+-i|perl\s+-pi|tee\s+|cp\s+|mv\s+|mkdir\s+|touch\s+|npm\s+version|git\s+apply)(?:\s|$)|(?:>>?|2>)\s*[^&]/i;
|
|
31
|
+
|
|
32
|
+
export type MessageLike = Record<string, any> & { role?: string; timestamp?: number };
|
|
33
|
+
|
|
34
|
+
export type SemanticPhase = "scout" | "plan" | "implement" | "verify" | "review" | "release" | "discuss" | "unknown";
|
|
35
|
+
export type SemanticBoundary = {
|
|
36
|
+
kind: "none" | "phase" | "task";
|
|
37
|
+
previousPhase: SemanticPhase;
|
|
38
|
+
currentPhase: SemanticPhase;
|
|
39
|
+
similarity: number;
|
|
40
|
+
};
|
|
41
|
+
|
|
42
|
+
export type ContextResidencyMetrics = {
|
|
43
|
+
estimatedMessageTokens: number;
|
|
44
|
+
toolResults: number;
|
|
45
|
+
readResults: number;
|
|
46
|
+
mutationResults: number;
|
|
47
|
+
verifierResults: number;
|
|
48
|
+
failedResults: number;
|
|
49
|
+
duplicateResults: number;
|
|
50
|
+
duplicateResultRate: number;
|
|
51
|
+
lowValueTokenShare: number;
|
|
52
|
+
userTurns: number;
|
|
53
|
+
};
|
|
54
|
+
|
|
55
|
+
export type ContextGovernorDecision = {
|
|
56
|
+
version: typeof GOVERNOR_VERSION;
|
|
57
|
+
action: "passthrough" | "project";
|
|
58
|
+
reason: "none" | "semantic-boundary" | "residency" | "pressure" | "provider-ceiling";
|
|
59
|
+
reasonCodes: string[];
|
|
60
|
+
reportedTokens: number;
|
|
61
|
+
effectiveTokens: number;
|
|
62
|
+
contextWindow: number;
|
|
63
|
+
percent: number;
|
|
64
|
+
originalMessageTokens: number;
|
|
65
|
+
projectedMessageTokens: number;
|
|
66
|
+
candidateProjectedMessageTokens: number;
|
|
67
|
+
estimatedSavingsTokens: number;
|
|
68
|
+
minimumSavingsTokens: number;
|
|
69
|
+
savingsRatio: number;
|
|
70
|
+
fallback: "none" | "no-safe-boundary" | "insufficient-savings" | "unsafe-tool-protocol";
|
|
71
|
+
originalMessages: number;
|
|
72
|
+
projectedMessages: number;
|
|
73
|
+
accounting: ContextGovernorAccounting;
|
|
74
|
+
boundary: SemanticBoundary;
|
|
75
|
+
residency: ContextResidencyMetrics;
|
|
76
|
+
};
|
|
77
|
+
|
|
78
|
+
export type ContextGovernorAccounting = {
|
|
79
|
+
contextOccupancy: {
|
|
80
|
+
providerReportedTokens: number | null;
|
|
81
|
+
transcriptEstimatedTokens: number;
|
|
82
|
+
projectedTranscriptTokens: number;
|
|
83
|
+
effectivePressureTokens: number;
|
|
84
|
+
contextWindow: number;
|
|
85
|
+
};
|
|
86
|
+
billedTraffic: {
|
|
87
|
+
measured: false;
|
|
88
|
+
inputTokens: null;
|
|
89
|
+
outputTokens: null;
|
|
90
|
+
cacheReadTokens: null;
|
|
91
|
+
governorProviderCalls: 0;
|
|
92
|
+
reason: "provider-billing-not-exposed-by-context-hook";
|
|
93
|
+
};
|
|
94
|
+
};
|
|
95
|
+
|
|
96
|
+
export type ToolProtocolAudit = {
|
|
97
|
+
intact: boolean;
|
|
98
|
+
orphanCallIds: string[];
|
|
99
|
+
orphanResultIds: string[];
|
|
100
|
+
};
|
|
101
|
+
|
|
102
|
+
export type ToolObservation = {
|
|
103
|
+
identity: string;
|
|
104
|
+
name: string;
|
|
105
|
+
target: string;
|
|
106
|
+
kind: "read" | "mutation" | "verify" | "other";
|
|
107
|
+
isError: boolean;
|
|
108
|
+
tokens: number;
|
|
109
|
+
};
|
|
110
|
+
|
|
111
|
+
function bounded(value: unknown, maxChars: number): string {
|
|
112
|
+
const compact = String(value ?? "").replace(/\s+/g, " ").trim();
|
|
113
|
+
if (compact.length <= maxChars) return compact;
|
|
114
|
+
const marker = " ... ";
|
|
115
|
+
const available = Math.max(0, maxChars - marker.length);
|
|
116
|
+
const head = Math.floor(available * 0.68);
|
|
117
|
+
return `${compact.slice(0, head).trimEnd()}${marker}${compact.slice(-(available - head)).trimStart()}`;
|
|
118
|
+
}
|
|
119
|
+
|
|
120
|
+
export function safeText(value: unknown, maxChars: number): string {
|
|
121
|
+
return bounded(redactSensitiveText(String(value ?? "")).text, maxChars);
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
function contentTokenEstimate(content: unknown): number {
|
|
125
|
+
if (typeof content === "string") return Math.ceil(content.length / 4);
|
|
126
|
+
if (!Array.isArray(content)) return 0;
|
|
127
|
+
let tokens = 0;
|
|
128
|
+
for (const block of content) {
|
|
129
|
+
if (!block || typeof block !== "object") continue;
|
|
130
|
+
if (block.type === "image") {
|
|
131
|
+
tokens += 1_200;
|
|
132
|
+
continue;
|
|
133
|
+
}
|
|
134
|
+
if (block.type === "text") tokens += Math.ceil(String(block.text ?? "").length / 4);
|
|
135
|
+
else if (block.type === "thinking") tokens += Math.ceil(String(block.thinking ?? block.thinkingSignature ?? "").length / 4);
|
|
136
|
+
else if (block.type === "toolCall") {
|
|
137
|
+
let serialized = "";
|
|
138
|
+
try {
|
|
139
|
+
serialized = JSON.stringify(block.arguments ?? {});
|
|
140
|
+
} catch {
|
|
141
|
+
serialized = String(block.arguments ?? "");
|
|
142
|
+
}
|
|
143
|
+
tokens += Math.ceil((String(block.name ?? "").length + serialized.length) / 4) + 8;
|
|
144
|
+
}
|
|
145
|
+
}
|
|
146
|
+
return tokens;
|
|
147
|
+
}
|
|
148
|
+
|
|
149
|
+
export function estimateGovernorMessageTokens(message: MessageLike): number {
|
|
150
|
+
if (!message || typeof message !== "object") return 0;
|
|
151
|
+
if (message.role === "compactionSummary" || message.role === "branchSummary") {
|
|
152
|
+
return Math.ceil(String(message.summary ?? "").length / 4) + 16;
|
|
153
|
+
}
|
|
154
|
+
if (message.role === "bashExecution") {
|
|
155
|
+
return message.excludeFromContext ? 0 : Math.ceil((String(message.command ?? "").length + String(message.output ?? "").length) / 4) + 16;
|
|
156
|
+
}
|
|
157
|
+
return contentTokenEstimate(message.content) + 8;
|
|
158
|
+
}
|
|
159
|
+
|
|
160
|
+
export function estimateGovernorMessagesTokens(messages: MessageLike[]): number {
|
|
161
|
+
return messages.reduce((sum, message) => sum + estimateGovernorMessageTokens(message), 0);
|
|
162
|
+
}
|
|
163
|
+
|
|
164
|
+
export function messageText(message: MessageLike): string {
|
|
165
|
+
if (typeof message.content === "string") return message.content;
|
|
166
|
+
if (!Array.isArray(message.content)) return "";
|
|
167
|
+
return message.content
|
|
168
|
+
.filter((block: any) => block?.type === "text")
|
|
169
|
+
.map((block: any) => String(block.text ?? ""))
|
|
170
|
+
.join("\n");
|
|
171
|
+
}
|
|
172
|
+
|
|
173
|
+
function fold(value: string): string {
|
|
174
|
+
return value.normalize("NFKD").replace(/[\u0300-\u036f]/g, "").toLowerCase();
|
|
175
|
+
}
|
|
176
|
+
|
|
177
|
+
export function classifySemanticPhase(value: string): SemanticPhase {
|
|
178
|
+
const text = fold(value);
|
|
179
|
+
if (/\b(commit|push|merge|publish|release|tag|deploy|phat hanh|trien khai)\b/.test(text)) return "release";
|
|
180
|
+
if (/\b(verify|verification|run tests?|test lai|kiem thu|typecheck|lint|ci|closeout)\b/.test(text)) return "verify";
|
|
181
|
+
if (/\b(implement|implementation|fix|repair|code|sua code|tien hanh sua|lam di|build it)\b/.test(text)) return "implement";
|
|
182
|
+
if (/\b(plan|planning|ke hoach|len plan|proposal|design approach)\b/.test(text)) return "plan";
|
|
183
|
+
if (/\b(review|reviewer|cross-check|double-check|danh gia lai)\b/.test(text)) return "review";
|
|
184
|
+
if (/\b(scout|audit|inspect|investigate|read-only|read only|mapping|map logic|kiem tra logic|xem xet)\b/.test(text)) return "scout";
|
|
185
|
+
if (/\b(explain|why|discuss|question|tai sao|vi sao|nghi sao|phan tich)\b/.test(text)) return "discuss";
|
|
186
|
+
return "unknown";
|
|
187
|
+
}
|
|
188
|
+
|
|
189
|
+
const SEMANTIC_STOP_WORDS = new Set([
|
|
190
|
+
"about", "after", "agent", "anh", "before", "cho", "code", "continue", "cua", "em", "file", "fix", "giup",
|
|
191
|
+
"implement", "kiem", "lam", "logic", "please", "project", "sua", "task", "test", "tiep", "tuc", "that", "the",
|
|
192
|
+
"this", "them", "then", "there", "these", "update", "verify", "viec", "with", "xem"
|
|
193
|
+
]);
|
|
194
|
+
|
|
195
|
+
function semanticTerms(value: string): Set<string> {
|
|
196
|
+
const terms = fold(value).match(/[a-z0-9_.\/-]{3,}/g) ?? [];
|
|
197
|
+
return new Set(terms.filter((term) => !SEMANTIC_STOP_WORDS.has(term)));
|
|
198
|
+
}
|
|
199
|
+
|
|
200
|
+
function semanticSimilarity(left: string, right: string): number {
|
|
201
|
+
const a = semanticTerms(left);
|
|
202
|
+
const b = semanticTerms(right);
|
|
203
|
+
if (a.size === 0 || b.size === 0) return 0;
|
|
204
|
+
let intersection = 0;
|
|
205
|
+
for (const term of a) if (b.has(term)) intersection += 1;
|
|
206
|
+
return intersection / new Set([...a, ...b]).size;
|
|
207
|
+
}
|
|
208
|
+
|
|
209
|
+
export function operatorMessages(messages: MessageLike[]): MessageLike[] {
|
|
210
|
+
return messages.filter((message) => message.role === "user" && messageText(message).trim());
|
|
211
|
+
}
|
|
212
|
+
|
|
213
|
+
export function detectSemanticBoundary(messages: MessageLike[]): SemanticBoundary {
|
|
214
|
+
const users = operatorMessages(messages);
|
|
215
|
+
const currentText = messageText(users.at(-1) ?? {});
|
|
216
|
+
const previousText = messageText(users.at(-2) ?? {});
|
|
217
|
+
const currentPhase = classifySemanticPhase(currentText);
|
|
218
|
+
const previousPhase = classifySemanticPhase(previousText);
|
|
219
|
+
const similarity = semanticSimilarity(previousText, currentText);
|
|
220
|
+
if (!previousText || !currentText) return { kind: "none", previousPhase, currentPhase, similarity };
|
|
221
|
+
|
|
222
|
+
const phaseChanged = currentPhase !== "unknown"
|
|
223
|
+
&& previousPhase !== "unknown"
|
|
224
|
+
&& currentPhase !== previousPhase
|
|
225
|
+
&& ["plan", "implement", "verify", "review", "release"].includes(currentPhase);
|
|
226
|
+
if (phaseChanged) return { kind: "phase", previousPhase, currentPhase, similarity };
|
|
227
|
+
|
|
228
|
+
const continuation = /\b(continue|tiep tuc|proceed|carry on|oke|ok|do it|lam tiep|sua tiep)\b/.test(fold(currentText));
|
|
229
|
+
if (!continuation && currentText.length >= 24 && previousText.length >= 24 && similarity < 0.12) {
|
|
230
|
+
return { kind: "task", previousPhase, currentPhase, similarity };
|
|
231
|
+
}
|
|
232
|
+
return { kind: "none", previousPhase, currentPhase, similarity };
|
|
233
|
+
}
|
|
234
|
+
|
|
235
|
+
function toolCalls(message: MessageLike): any[] {
|
|
236
|
+
if (message.role !== "assistant" || !Array.isArray(message.content)) return [];
|
|
237
|
+
return message.content.filter((block: any) => block?.type === "toolCall");
|
|
238
|
+
}
|
|
239
|
+
|
|
240
|
+
function toolCallId(block: any): string {
|
|
241
|
+
return String(block?.id ?? block?.toolCallId ?? "").trim();
|
|
242
|
+
}
|
|
243
|
+
|
|
244
|
+
export function auditToolProtocol(messages: MessageLike[]): ToolProtocolAudit {
|
|
245
|
+
const calls = new Map<string, number>();
|
|
246
|
+
const results = new Map<string, number>();
|
|
247
|
+
for (const message of messages) {
|
|
248
|
+
for (const call of toolCalls(message)) {
|
|
249
|
+
const id = toolCallId(call);
|
|
250
|
+
if (id) calls.set(id, (calls.get(id) ?? 0) + 1);
|
|
251
|
+
}
|
|
252
|
+
if (message.role !== "toolResult") continue;
|
|
253
|
+
const id = String(message.toolCallId ?? "").trim();
|
|
254
|
+
if (id) results.set(id, (results.get(id) ?? 0) + 1);
|
|
255
|
+
}
|
|
256
|
+
const orphanCallIds = [...calls]
|
|
257
|
+
.filter(([id, count]) => (results.get(id) ?? 0) !== count)
|
|
258
|
+
.map(([id]) => id)
|
|
259
|
+
.sort();
|
|
260
|
+
const orphanResultIds = [...results]
|
|
261
|
+
.filter(([id, count]) => (calls.get(id) ?? 0) !== count)
|
|
262
|
+
.map(([id]) => id)
|
|
263
|
+
.sort();
|
|
264
|
+
return {
|
|
265
|
+
intact: orphanCallIds.length === 0 && orphanResultIds.length === 0,
|
|
266
|
+
orphanCallIds,
|
|
267
|
+
orphanResultIds
|
|
268
|
+
};
|
|
269
|
+
}
|
|
270
|
+
|
|
271
|
+
function canonicalToolArguments(value: unknown): string {
|
|
272
|
+
const canonical = (item: unknown): unknown => {
|
|
273
|
+
if (Array.isArray(item)) return item.map(canonical);
|
|
274
|
+
if (!item || typeof item !== "object") return item;
|
|
275
|
+
return Object.fromEntries(
|
|
276
|
+
Object.entries(item as Record<string, unknown>)
|
|
277
|
+
.filter(([, child]) => child !== undefined)
|
|
278
|
+
.sort(([left], [right]) => left < right ? -1 : left > right ? 1 : 0)
|
|
279
|
+
.map(([key, child]) => [key, canonical(child)])
|
|
280
|
+
);
|
|
281
|
+
};
|
|
282
|
+
try {
|
|
283
|
+
return JSON.stringify(canonical(value));
|
|
284
|
+
} catch {
|
|
285
|
+
return String(value ?? "");
|
|
286
|
+
}
|
|
287
|
+
}
|
|
288
|
+
|
|
289
|
+
function toolTarget(name: string, input: Record<string, unknown>): string {
|
|
290
|
+
const candidate = input.url ?? input.path ?? input.file_path ?? input.filePath ?? input.target ?? input.query ?? input.pattern
|
|
291
|
+
?? input.command ?? input.cmd ?? input.script ?? "";
|
|
292
|
+
if (candidate) return safeText(candidate, name === "bash" ? 320 : 180);
|
|
293
|
+
if (Array.isArray(input.queries) && input.queries.length > 0) {
|
|
294
|
+
return safeText(input.queries.join(" | "), 180);
|
|
295
|
+
}
|
|
296
|
+
const selector = [
|
|
297
|
+
input.responseId ?? input.response_id,
|
|
298
|
+
input.urlIndex ?? input.url_index,
|
|
299
|
+
input.offset,
|
|
300
|
+
input.limit
|
|
301
|
+
];
|
|
302
|
+
if (selector.some((value) => value !== undefined && value !== null && value !== "")) {
|
|
303
|
+
return safeText([
|
|
304
|
+
selector[0] === undefined ? "" : `responseId=${String(selector[0])}`,
|
|
305
|
+
selector[1] === undefined ? "" : `urlIndex=${String(selector[1])}`,
|
|
306
|
+
selector[2] === undefined ? "" : `offset=${String(selector[2])}`,
|
|
307
|
+
selector[3] === undefined ? "" : `limit=${String(selector[3])}`
|
|
308
|
+
].filter(Boolean).join(" "), 180);
|
|
309
|
+
}
|
|
310
|
+
return safeText(candidate, name === "bash" ? 320 : 180);
|
|
311
|
+
}
|
|
312
|
+
|
|
313
|
+
function toolKind(name: string, target: string): ToolObservation["kind"] {
|
|
314
|
+
const normalized = name.toLowerCase();
|
|
315
|
+
if (READ_TOOLS.has(normalized)) return "read";
|
|
316
|
+
if ((normalized === "bash" || normalized === "exec" || normalized === "exec_command") && VERIFIER_PATTERN.test(target)) return "verify";
|
|
317
|
+
if (MUTATION_TOOLS.has(normalized) || (normalized === "bash" && MUTATION_COMMAND_PATTERN.test(target))) return "mutation";
|
|
318
|
+
return "other";
|
|
319
|
+
}
|
|
320
|
+
|
|
321
|
+
export function toolObservations(messages: MessageLike[]): ToolObservation[] {
|
|
322
|
+
const calls = new Map<string, { name: string; input: Record<string, unknown> }>();
|
|
323
|
+
const observations: ToolObservation[] = [];
|
|
324
|
+
for (const message of messages) {
|
|
325
|
+
for (const call of toolCalls(message)) {
|
|
326
|
+
calls.set(String(call.id ?? ""), {
|
|
327
|
+
name: String(call.name ?? "tool"),
|
|
328
|
+
input: call.arguments && typeof call.arguments === "object" ? call.arguments : {}
|
|
329
|
+
});
|
|
330
|
+
}
|
|
331
|
+
if (message.role !== "toolResult") continue;
|
|
332
|
+
const call = calls.get(String(message.toolCallId ?? ""));
|
|
333
|
+
const name = String(message.toolName ?? call?.name ?? "tool").toLowerCase();
|
|
334
|
+
const input = call?.input ?? {};
|
|
335
|
+
const target = toolTarget(name, input);
|
|
336
|
+
const identitySource = `${name}\u0000${canonicalToolArguments(input)}`;
|
|
337
|
+
observations.push({
|
|
338
|
+
identity: crypto.createHash("sha256").update(identitySource).digest("hex").slice(0, 16),
|
|
339
|
+
name,
|
|
340
|
+
target,
|
|
341
|
+
kind: toolKind(name, target),
|
|
342
|
+
isError: message.isError === true,
|
|
343
|
+
tokens: estimateGovernorMessageTokens(message)
|
|
344
|
+
});
|
|
345
|
+
}
|
|
346
|
+
return observations;
|
|
347
|
+
}
|
|
348
|
+
|
|
349
|
+
export function analyzeContextResidency(messages: MessageLike[]): ContextResidencyMetrics {
|
|
350
|
+
const observations = toolObservations(messages);
|
|
351
|
+
const seen = new Set<string>();
|
|
352
|
+
let duplicateResults = 0;
|
|
353
|
+
let lowValueTokens = 0;
|
|
354
|
+
for (const observation of observations) {
|
|
355
|
+
if (seen.has(observation.identity)) duplicateResults += 1;
|
|
356
|
+
seen.add(observation.identity);
|
|
357
|
+
if (observation.kind === "read" && !observation.isError) lowValueTokens += observation.tokens;
|
|
358
|
+
}
|
|
359
|
+
const estimatedMessageTokens = estimateGovernorMessagesTokens(messages);
|
|
360
|
+
return {
|
|
361
|
+
estimatedMessageTokens,
|
|
362
|
+
toolResults: observations.length,
|
|
363
|
+
readResults: observations.filter((item) => item.kind === "read").length,
|
|
364
|
+
mutationResults: observations.filter((item) => item.kind === "mutation").length,
|
|
365
|
+
verifierResults: observations.filter((item) => item.kind === "verify").length,
|
|
366
|
+
failedResults: observations.filter((item) => item.isError).length,
|
|
367
|
+
duplicateResults,
|
|
368
|
+
duplicateResultRate: observations.length > 0 ? duplicateResults / observations.length : 0,
|
|
369
|
+
lowValueTokenShare: estimatedMessageTokens > 0 ? lowValueTokens / estimatedMessageTokens : 0,
|
|
370
|
+
userTurns: operatorMessages(messages).length
|
|
371
|
+
};
|
|
372
|
+
}
|
|
373
|
+
|
|
374
|
+
export function reasonForProjection(
|
|
375
|
+
effectiveTokens: number,
|
|
376
|
+
contextWindow: number,
|
|
377
|
+
boundary: SemanticBoundary,
|
|
378
|
+
residency: ContextResidencyMetrics
|
|
379
|
+
): { reason: ContextGovernorDecision["reason"]; codes: string[] } {
|
|
380
|
+
const percent = contextWindow > 0 ? (effectiveTokens / contextWindow) * 100 : 0;
|
|
381
|
+
if (effectiveTokens >= CONTEXT_PROVIDER_PROMPT_CEILING_TOKENS || percent >= 68) {
|
|
382
|
+
return { reason: "provider-ceiling", codes: ["provider-prompt-headroom", "preserve-output-budget"] };
|
|
383
|
+
}
|
|
384
|
+
if (boundary.kind !== "none" && (
|
|
385
|
+
effectiveTokens >= CONTEXT_GOVERNOR_SEMANTIC_BOUNDARY_TOKENS
|
|
386
|
+
|| residency.toolResults >= 8
|
|
387
|
+
)) {
|
|
388
|
+
return { reason: "semantic-boundary", codes: [`${boundary.kind}-boundary`, "fresh-working-set"] };
|
|
389
|
+
}
|
|
390
|
+
if (residency.toolResults >= CONTEXT_GOVERNOR_TOOL_ROUND_LIMIT && (
|
|
391
|
+
residency.duplicateResultRate >= 0.12
|
|
392
|
+
|| residency.readResults >= Math.ceil(CONTEXT_GOVERNOR_TOOL_ROUND_LIMIT * 0.5)
|
|
393
|
+
|| effectiveTokens >= CONTEXT_GOVERNOR_TARGET_MESSAGE_TOKENS
|
|
394
|
+
)) {
|
|
395
|
+
return { reason: "residency", codes: ["tool-round-residency", residency.duplicateResults > 0 ? "duplicate-delta" : "stale-read-residency"] };
|
|
396
|
+
}
|
|
397
|
+
if ((effectiveTokens >= CONTEXT_GOVERNOR_PRESSURE_TOKENS || percent >= 52)
|
|
398
|
+
&& residency.toolResults >= 8) {
|
|
399
|
+
return { reason: "pressure", codes: ["context-pressure", "value-density"] };
|
|
400
|
+
}
|
|
401
|
+
return { reason: "none", codes: [] };
|
|
402
|
+
}
|
|
403
|
+
|
|
404
|
+
export function minimumContextSavingsTokens(occupancyTokens: number): number {
|
|
405
|
+
const proportional = Math.ceil(Math.max(0, occupancyTokens) * CONTEXT_GOVERNOR_MIN_SAVINGS_RATIO);
|
|
406
|
+
return Math.max(
|
|
407
|
+
CONTEXT_GOVERNOR_MIN_SAVINGS_FLOOR_TOKENS,
|
|
408
|
+
Math.min(CONTEXT_GOVERNOR_MIN_SAVINGS_CEILING_TOKENS, proportional)
|
|
409
|
+
);
|
|
410
|
+
}
|
|
411
|
+
|
|
412
|
+
function toolProtocolSafeCuts(messages: MessageLike[]): number[] {
|
|
413
|
+
const callIndexes = new Map<string, number>();
|
|
414
|
+
const pairSpans: Array<{ call: number; result: number }> = [];
|
|
415
|
+
for (let index = 0; index < messages.length; index += 1) {
|
|
416
|
+
const message = messages[index]!;
|
|
417
|
+
for (const call of toolCalls(message)) {
|
|
418
|
+
const id = toolCallId(call);
|
|
419
|
+
if (id && !callIndexes.has(id)) callIndexes.set(id, index);
|
|
420
|
+
}
|
|
421
|
+
if (message.role !== "toolResult") continue;
|
|
422
|
+
const id = String(message.toolCallId ?? "").trim();
|
|
423
|
+
const call = callIndexes.get(id);
|
|
424
|
+
if (call !== undefined) pairSpans.push({ call, result: index });
|
|
425
|
+
}
|
|
426
|
+
|
|
427
|
+
const blockedDeltas = new Int32Array(messages.length + 2);
|
|
428
|
+
for (const span of pairSpans) {
|
|
429
|
+
blockedDeltas[span.call + 1] += 1;
|
|
430
|
+
blockedDeltas[span.result + 1] -= 1;
|
|
431
|
+
}
|
|
432
|
+
const cuts: number[] = [];
|
|
433
|
+
let blocked = 0;
|
|
434
|
+
for (let cut = 0; cut <= messages.length; cut += 1) {
|
|
435
|
+
blocked += blockedDeltas[cut] ?? 0;
|
|
436
|
+
if (blocked > 0) continue;
|
|
437
|
+
cuts.push(cut);
|
|
438
|
+
}
|
|
439
|
+
return cuts;
|
|
440
|
+
}
|
|
441
|
+
|
|
442
|
+
export function safeSuffixStart(
|
|
443
|
+
messages: MessageLike[],
|
|
444
|
+
recentTokenBudget: number,
|
|
445
|
+
protectedBoundaryGroups = CONTEXT_GOVERNOR_PROTECTED_BOUNDARY_GROUPS
|
|
446
|
+
): number {
|
|
447
|
+
if (!auditToolProtocol(messages).intact) return 0;
|
|
448
|
+
let tokens = 0;
|
|
449
|
+
let rawStart = messages.length;
|
|
450
|
+
while (rawStart > 0 && tokens < recentTokenBudget) {
|
|
451
|
+
rawStart -= 1;
|
|
452
|
+
tokens += estimateGovernorMessageTokens(messages[rawStart]!);
|
|
453
|
+
}
|
|
454
|
+
|
|
455
|
+
const safeCuts = toolProtocolSafeCuts(messages);
|
|
456
|
+
const budgetStart = [...safeCuts].reverse().find((cut) => cut <= rawStart) ?? 0;
|
|
457
|
+
const protectedGroups = Math.max(0, Math.floor(protectedBoundaryGroups));
|
|
458
|
+
const protectedIndex = Math.max(0, safeCuts.length - 1 - protectedGroups);
|
|
459
|
+
const protectedStart = safeCuts[protectedIndex] ?? 0;
|
|
460
|
+
return Math.min(budgetStart, protectedStart);
|
|
461
|
+
}
|