@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
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/benchmark" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/benchmark" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/commands" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/commands" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/context-engine" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/context-engine" hreflang="en" lang="en" aria-current="true">EN</a>
|
package/docs-site/en/index.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/mcp-approval" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/mcp-approval" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/mcp-auth" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/mcp-auth" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/mcp-servers" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/mcp-servers" hreflang="en" lang="en" aria-current="true">EN</a>
|
package/docs-site/en/mcp.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/mcp" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/mcp" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/permissions" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/permissions" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/profiles" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/profiles" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/quickstart" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/quickstart" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/runtime" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/runtime" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -105,7 +105,7 @@
|
|
|
105
105
|
<section class="section">
|
|
106
106
|
<div class="command-grid">
|
|
107
107
|
<article class="command-card"><header><span>Global setup</span><button class="clip-btn" data-clip="piagent-setup --mcp-preset core --subagents-preset safe">Copy</button></header><pre><code>piagent-setup --mcp-preset core --subagents-preset safe</code></pre></article>
|
|
108
|
-
<article class="command-card"><header><span>Full update</span><button class="clip-btn" data-clip="npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.
|
|
108
|
+
<article class="command-card"><header><span>Full update</span><button class="clip-btn" data-clip="npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.86.1 npm install -g --ignore-scripts @piagent/platform@X.Y.Z piagent-install --stable --dry-run piagent-install --stable">Copy</button></header><pre><code>npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.86.1
|
|
109
109
|
npm install -g --ignore-scripts @piagent/platform@X.Y.Z
|
|
110
110
|
piagent-install --stable --dry-run
|
|
111
111
|
piagent-install --stable</code></pre></article>
|
package/docs-site/en/scope.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/scope" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/scope" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/security" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/security" hreflang="en" lang="en" aria-current="true">EN</a>
|
package/docs-site/en/team.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/team" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/team" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/whats-new" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/whats-new" hreflang="en" lang="en" aria-current="true">EN</a>
|
|
@@ -102,6 +102,17 @@
|
|
|
102
102
|
<h1>Safer task flow with less manual coordination</h1>
|
|
103
103
|
<p>Product stabilization adds runtime truth, trajectories, recovery, helper policy, an activity inspector, and fail-closed completion receipts without breaking existing workflows.</p>
|
|
104
104
|
</div>
|
|
105
|
+
<section class="section">
|
|
106
|
+
<p class="section-kicker">v1.7.1</p>
|
|
107
|
+
<h2 id="v1-7-0">Long sessions with clearer context and accounting</h2>
|
|
108
|
+
<p class="section-lead">v1.7.1 supports Pi 0.86.1, repairs session recovery, and adds waiting within an open session.</p>
|
|
109
|
+
<div class="grid cols-3">
|
|
110
|
+
<article class="card"><h3>Requested waiting</h3><p>Ask the agent to wait until the next useful check. The timer makes no model calls while waiting; the session must remain open.</p></article>
|
|
111
|
+
<article class="card"><h3>Context and usage</h3><p>Context prioritizes supported files, and usage and verification results remain bound to the correct task.</p></article>
|
|
112
|
+
<article class="card"><h3>Evidence boundary</h3><p>All 14 formerly failed attempts passed a separate rerun. This is not a new 108-session score or an aggregate token-saving claim.</p></article>
|
|
113
|
+
</div>
|
|
114
|
+
</section>
|
|
115
|
+
|
|
105
116
|
<section class="section">
|
|
106
117
|
<p class="section-kicker">v1.6.1 · Reliability patch</p>
|
|
107
118
|
<h2 id="v1-6-1">Truthful UI state, safe external review, and lower helper overhead</h2>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · EN</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Language">
|
|
35
35
|
<a class="language-option" href="/workflows" hreflang="vi" lang="vi">VI</a>
|
|
36
36
|
<a class="language-option current" href="/en/workflows" hreflang="en" lang="en" aria-current="true">EN</a>
|
package/docs-site/index.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/" hreflang="en" lang="en">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/mcp-approval" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/mcp-approval" hreflang="en" lang="en">EN</a>
|
package/docs-site/mcp-auth.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/mcp-auth" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/mcp-auth" hreflang="en" lang="en">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/mcp-servers" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/mcp-servers" hreflang="en" lang="en">EN</a>
|
package/docs-site/mcp.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/mcp" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/mcp" hreflang="en" lang="en">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/permissions" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/permissions" hreflang="en" lang="en">EN</a>
|
package/docs-site/profiles.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/profiles" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/profiles" hreflang="en" lang="en">EN</a>
|
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/quickstart" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/quickstart" hreflang="en" lang="en">EN</a>
|
package/docs-site/runtime.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/runtime" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/runtime" hreflang="en" lang="en">EN</a>
|
|
@@ -110,8 +110,8 @@
|
|
|
110
110
|
<pre><code>piagent-setup --mcp-preset core --subagents-preset safe</code></pre>
|
|
111
111
|
</article>
|
|
112
112
|
<article class="command-card">
|
|
113
|
-
<header><span>Full update</span><button class="clip-btn" data-clip="npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.
|
|
114
|
-
<pre><code>npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.
|
|
113
|
+
<header><span>Full update</span><button class="clip-btn" data-clip="npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.86.1 npm install -g --ignore-scripts @piagent/platform@X.Y.Z piagent-install --stable --dry-run piagent-install --stable">Lấy lệnh</button></header>
|
|
114
|
+
<pre><code>npm install -g --ignore-scripts @earendil-works/pi-coding-agent@0.86.1
|
|
115
115
|
npm install -g --ignore-scripts @piagent/platform@X.Y.Z
|
|
116
116
|
piagent-install --stable --dry-run
|
|
117
117
|
piagent-install --stable</code></pre>
|
package/docs-site/scope.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/scope" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/scope" hreflang="en" lang="en">EN</a>
|
package/docs-site/security.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/security" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/security" hreflang="en" lang="en">EN</a>
|
package/docs-site/team.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/team" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/team" hreflang="en" lang="en">EN</a>
|
package/docs-site/whats-new.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/whats-new" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/whats-new" hreflang="en" lang="en">EN</a>
|
|
@@ -102,6 +102,17 @@
|
|
|
102
102
|
<h1>Luồng task an toàn hơn, ít thao tác tay hơn</h1>
|
|
103
103
|
<p>Product stabilization bổ sung runtime truth, trajectory, recovery, helper policy, activity inspector và completion receipt fail-closed mà không phá workflow cũ.</p>
|
|
104
104
|
</div>
|
|
105
|
+
<section class="section">
|
|
106
|
+
<p class="section-kicker">v1.7.1</p>
|
|
107
|
+
<h2 id="v1-7-0">Phiên dài giữ đúng việc, context và số liệu rõ ràng</h2>
|
|
108
|
+
<p class="section-lead">v1.7.1 cập nhật Pi 0.86.1, sửa phục hồi phiên và bổ sung chờ trong phiên đang mở.</p>
|
|
109
|
+
<div class="grid cols-3">
|
|
110
|
+
<article class="card"><h3>Chờ theo yêu cầu</h3><p>Nhờ agent chờ tới lần kiểm tra tiếp theo; timer không gọi model trong khoảng chờ. Phiên phải còn mở.</p></article>
|
|
111
|
+
<article class="card"><h3>Context và usage</h3><p>Context ưu tiên file dùng được; token và kết quả kiểm tra được gắn đúng task.</p></article>
|
|
112
|
+
<article class="card"><h3>Giới hạn bằng chứng</h3><p>14 phiên từng lỗi đã PASS khi chạy lại riêng. Không gộp thành điểm 108 phiên mới hoặc claim tiết kiệm token tổng thể.</p></article>
|
|
113
|
+
</div>
|
|
114
|
+
</section>
|
|
115
|
+
|
|
105
116
|
<section class="section">
|
|
106
117
|
<p class="section-kicker">v1.6.1 · Reliability patch</p>
|
|
107
118
|
<h2 id="v1-6-1">UI đúng trạng thái, review repo ngoài an toàn, helper ít tốn hơn</h2>
|
package/docs-site/workflows.html
CHANGED
|
@@ -30,7 +30,7 @@
|
|
|
30
30
|
<span>Pi Agent Platform</span>
|
|
31
31
|
</a>
|
|
32
32
|
<div class="top-actions">
|
|
33
|
-
<span class="pill"><span class="pill-dot"></span>v1.
|
|
33
|
+
<span class="pill"><span class="pill-dot"></span>v1.7.1 docs · VI</span>
|
|
34
34
|
<nav class="language-switch" aria-label="Ngôn ngữ">
|
|
35
35
|
<a class="language-option current" href="/workflows" hreflang="vi" lang="vi" aria-current="true">VI</a>
|
|
36
36
|
<a class="language-option" href="/en/workflows" hreflang="en" lang="en">EN</a>
|
|
@@ -0,0 +1,18 @@
|
|
|
1
|
+
{
|
|
2
|
+
"schemaVersion": 1,
|
|
3
|
+
"id": "architecture-conformance-v1",
|
|
4
|
+
"title": "Provider-free architecture boundary and line-budget lane",
|
|
5
|
+
"providerRequired": false,
|
|
6
|
+
"modelTokensExpected": 0,
|
|
7
|
+
"checker": "scripts/check-architecture.mjs",
|
|
8
|
+
"configuration": "architecture/layers.json",
|
|
9
|
+
"releaseGates": {
|
|
10
|
+
"architectureCheckPasses": true,
|
|
11
|
+
"sourceCoverageComplete": true,
|
|
12
|
+
"dependencyBoundariesPass": true,
|
|
13
|
+
"lineBudgetsPass": true,
|
|
14
|
+
"providerCalls": 0,
|
|
15
|
+
"modelTokens": 0
|
|
16
|
+
},
|
|
17
|
+
"claimBoundary": "Deterministic provider-free architecture and line-budget evidence only. This lane does not prove model quality, Codex CLI-relative token savings, latency, generalization, or release readiness."
|
|
18
|
+
}
|
|
@@ -0,0 +1,77 @@
|
|
|
1
|
+
#!/usr/bin/env node
|
|
2
|
+
import fs from "node:fs";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
import { fileURLToPath } from "node:url";
|
|
5
|
+
|
|
6
|
+
import { inspectArchitecture } from "../../scripts/check-architecture.mjs";
|
|
7
|
+
|
|
8
|
+
const laneRoot = path.dirname(fileURLToPath(import.meta.url));
|
|
9
|
+
const lane = JSON.parse(fs.readFileSync(path.join(laneRoot, "lane.json"), "utf8"));
|
|
10
|
+
const argumentsList = process.argv.slice(2);
|
|
11
|
+
const outputIndex = argumentsList.indexOf("--output");
|
|
12
|
+
const outputArgument = outputIndex >= 0 ? argumentsList[outputIndex + 1] : undefined;
|
|
13
|
+
if (outputIndex >= 0 && !outputArgument) throw new Error("--output requires a path");
|
|
14
|
+
const outputValueIndex = outputIndex >= 0 ? outputIndex + 1 : -1;
|
|
15
|
+
const unknown = argumentsList.filter((_, index) => index !== outputIndex && index !== outputValueIndex);
|
|
16
|
+
if (unknown.length > 0) throw new Error(`Unknown option: ${unknown[0]}`);
|
|
17
|
+
const requiredGates = {
|
|
18
|
+
architectureCheckPasses: true,
|
|
19
|
+
sourceCoverageComplete: true,
|
|
20
|
+
dependencyBoundariesPass: true,
|
|
21
|
+
lineBudgetsPass: true,
|
|
22
|
+
providerCalls: 0,
|
|
23
|
+
modelTokens: 0
|
|
24
|
+
};
|
|
25
|
+
if (lane?.schemaVersion !== 1 || lane.id !== "architecture-conformance-v1"
|
|
26
|
+
|| lane.providerRequired !== false || lane.modelTokensExpected !== 0
|
|
27
|
+
|| lane.checker !== "scripts/check-architecture.mjs"
|
|
28
|
+
|| lane.configuration !== "architecture/layers.json"
|
|
29
|
+
|| JSON.stringify(lane.releaseGates) !== JSON.stringify(requiredGates)
|
|
30
|
+
|| typeof lane.claimBoundary !== "string" || lane.claimBoundary.length === 0) {
|
|
31
|
+
throw new Error("Unsupported architecture conformance lane configuration");
|
|
32
|
+
}
|
|
33
|
+
|
|
34
|
+
let inspection;
|
|
35
|
+
let failure = null;
|
|
36
|
+
try {
|
|
37
|
+
inspection = inspectArchitecture();
|
|
38
|
+
} catch (error) {
|
|
39
|
+
failure = error instanceof Error ? error.message : String(error);
|
|
40
|
+
inspection = { ok: false, errors: [failure], files: 0, layers: {} };
|
|
41
|
+
}
|
|
42
|
+
const errors = Array.isArray(inspection.errors) ? inspection.errors : ["invalid-architecture-result"];
|
|
43
|
+
const gates = {
|
|
44
|
+
architectureCheckPasses: inspection.ok === true && errors.length === 0,
|
|
45
|
+
sourceCoverageComplete: errors.every((error) => !String(error).includes("source file is not assigned")),
|
|
46
|
+
dependencyBoundariesPass: errors.every((error) => !String(error).includes(" cannot import ")),
|
|
47
|
+
lineBudgetsPass: errors.every((error) => !String(error).includes(" lines exceeds ")),
|
|
48
|
+
providerCalls: 0,
|
|
49
|
+
modelTokens: 0
|
|
50
|
+
};
|
|
51
|
+
const passed = failure === null
|
|
52
|
+
&& Object.entries(gates).every(([name, value]) => name === "providerCalls" || name === "modelTokens" ? value === 0 : value === true);
|
|
53
|
+
const report = {
|
|
54
|
+
schemaVersion: 1,
|
|
55
|
+
laneId: lane.id,
|
|
56
|
+
evidenceClass: "provider-free-architecture-conformance",
|
|
57
|
+
passed,
|
|
58
|
+
provider: { required: false, used: false, calls: 0, modelTokens: 0 },
|
|
59
|
+
architecture: {
|
|
60
|
+
filesChecked: Number.isSafeInteger(inspection.files) ? inspection.files : 0,
|
|
61
|
+
layersChecked: Object.keys(inspection.layers ?? {}).length,
|
|
62
|
+
errors: errors.map(String)
|
|
63
|
+
},
|
|
64
|
+
gates,
|
|
65
|
+
claimBoundary: lane.claimBoundary
|
|
66
|
+
};
|
|
67
|
+
|
|
68
|
+
if (outputArgument) {
|
|
69
|
+
const outputPath = path.resolve(outputArgument);
|
|
70
|
+
fs.mkdirSync(path.dirname(outputPath), { recursive: true, mode: 0o700 });
|
|
71
|
+
fs.writeFileSync(outputPath, `${JSON.stringify(report, null, 2)}\n`, { mode: 0o600 });
|
|
72
|
+
}
|
|
73
|
+
if (passed) process.stdout.write(`${JSON.stringify(report, null, 2)}\n`);
|
|
74
|
+
else {
|
|
75
|
+
process.stderr.write(`${JSON.stringify(report, null, 2)}\n`);
|
|
76
|
+
process.exitCode = 1;
|
|
77
|
+
}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
{
|
|
2
|
+
"version": "approved-host-contracts-v1",
|
|
3
|
+
"projectId": "cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc",
|
|
4
|
+
"operatorRequestDigest": "operator-request-v1:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
|
|
5
|
+
"verifierDigest": "dddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddd",
|
|
6
|
+
"backend": {
|
|
7
|
+
"imageId": "sha256:bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb",
|
|
8
|
+
"dockerSocket": "/fixture-only/engine.sock",
|
|
9
|
+
"timeoutMs": 30001
|
|
10
|
+
},
|
|
11
|
+
"contracts": [
|
|
12
|
+
{
|
|
13
|
+
"criterionId": "identity-contract",
|
|
14
|
+
"criterionHash": "eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee",
|
|
15
|
+
"sourcePath": "src/identity.js",
|
|
16
|
+
"exportName": "identity",
|
|
17
|
+
"maxAttempts": 2,
|
|
18
|
+
"checks": [
|
|
19
|
+
{
|
|
20
|
+
"id": "identity",
|
|
21
|
+
"cases": [
|
|
22
|
+
{
|
|
23
|
+
"id": "positive-number",
|
|
24
|
+
"args": [
|
|
25
|
+
{
|
|
26
|
+
"type": "number",
|
|
27
|
+
"value": 7
|
|
28
|
+
}
|
|
29
|
+
],
|
|
30
|
+
"expected": {
|
|
31
|
+
"outcome": "return",
|
|
32
|
+
"value": {
|
|
33
|
+
"type": "number",
|
|
34
|
+
"value": 7
|
|
35
|
+
}
|
|
36
|
+
}
|
|
37
|
+
}
|
|
38
|
+
]
|
|
39
|
+
}
|
|
40
|
+
]
|
|
41
|
+
}
|
|
42
|
+
]
|
|
43
|
+
}
|
|
@@ -0,0 +1,43 @@
|
|
|
1
|
+
{
|
|
2
|
+
"version": "approved-host-contracts-v1",
|
|
3
|
+
"projectId": "cccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccccc",
|
|
4
|
+
"operatorRequestDigest": "operator-request-v1:aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa",
|
|
5
|
+
"verifierDigest": "dddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddddd",
|
|
6
|
+
"backend": {
|
|
7
|
+
"imageId": "sha256:bbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbbb",
|
|
8
|
+
"dockerSocket": "/fixture-only/engine.sock",
|
|
9
|
+
"timeoutMs": 1000
|
|
10
|
+
},
|
|
11
|
+
"contracts": [
|
|
12
|
+
{
|
|
13
|
+
"criterionId": "identity-contract",
|
|
14
|
+
"criterionHash": "eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeee",
|
|
15
|
+
"sourcePath": "src/identity.js",
|
|
16
|
+
"exportName": "identity",
|
|
17
|
+
"maxAttempts": 2,
|
|
18
|
+
"checks": [
|
|
19
|
+
{
|
|
20
|
+
"id": "identity",
|
|
21
|
+
"cases": [
|
|
22
|
+
{
|
|
23
|
+
"id": "positive-number",
|
|
24
|
+
"args": [
|
|
25
|
+
{
|
|
26
|
+
"type": "number",
|
|
27
|
+
"value": 7
|
|
28
|
+
}
|
|
29
|
+
],
|
|
30
|
+
"expected": {
|
|
31
|
+
"outcome": "return",
|
|
32
|
+
"value": {
|
|
33
|
+
"type": "number",
|
|
34
|
+
"value": 7
|
|
35
|
+
}
|
|
36
|
+
}
|
|
37
|
+
}
|
|
38
|
+
]
|
|
39
|
+
}
|
|
40
|
+
]
|
|
41
|
+
}
|
|
42
|
+
]
|
|
43
|
+
}
|
|
@@ -0,0 +1,42 @@
|
|
|
1
|
+
{
|
|
2
|
+
"schemaVersion": 3,
|
|
3
|
+
"contractId": "benchmark-attempt-outcome-v3",
|
|
4
|
+
"attemptId": "golden-exit-zero-error",
|
|
5
|
+
"scenarioKind": "source-change",
|
|
6
|
+
"transportStatus": "completed",
|
|
7
|
+
"operationStatus": "completed",
|
|
8
|
+
"taskStatus": "completed",
|
|
9
|
+
"semanticStatus": "pass",
|
|
10
|
+
"gradeStatus": "pass",
|
|
11
|
+
"runValidity": "valid",
|
|
12
|
+
"usageStatus": "exact",
|
|
13
|
+
"failureClass": "none",
|
|
14
|
+
"countsTowardQuality": true,
|
|
15
|
+
"countsTowardUsage": true,
|
|
16
|
+
"evidence": {
|
|
17
|
+
"providerStarted": true,
|
|
18
|
+
"processExitCode": 0,
|
|
19
|
+
"threadIdPresent": true,
|
|
20
|
+
"usageReported": true,
|
|
21
|
+
"terminalAgentMessage": true,
|
|
22
|
+
"requiredOutputEvidencePresent": true,
|
|
23
|
+
"errorEvents": 1,
|
|
24
|
+
"turnFailedEvents": 0,
|
|
25
|
+
"itemErrorEvents": 0,
|
|
26
|
+
"failedCommandEvents": 0,
|
|
27
|
+
"fileChangeCount": 1,
|
|
28
|
+
"mutationExpected": true,
|
|
29
|
+
"refusal": null
|
|
30
|
+
},
|
|
31
|
+
"usage": {
|
|
32
|
+
"providerInput": 1200,
|
|
33
|
+
"cacheRead": 800,
|
|
34
|
+
"cacheWrite": 0,
|
|
35
|
+
"output": 200,
|
|
36
|
+
"reasoning": 50,
|
|
37
|
+
"fresh": 600,
|
|
38
|
+
"totalTraffic": 1400,
|
|
39
|
+
"billedCost": null,
|
|
40
|
+
"billedCostStatus": "unavailable"
|
|
41
|
+
}
|
|
42
|
+
}
|