Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
The table of contents is too big for display.
Diff view
Diff view
  •  
  •  
  •  
Binary file modified bun.lockb
Binary file not shown.
9 changes: 4 additions & 5 deletions evals/tools/surface-smoke-cases.json
Original file line number Diff line number Diff line change
Expand Up @@ -110,13 +110,12 @@
}
},
{
"name": "extract_document reads a local smoke fixture",
"name": "extract_document blocks a local smoke fixture",
"kind": "extractDocument",
"judgeRubric": "extract_document should preserve the text payload, recognize the text format, and report the downloaded filename.",
"judgeRubric": "extract_document should reject private or local document URLs instead of downloading them.",
"expected": {
"text": "Maestro extract document smoke test",
"format": "text",
"fileName": "fixture.txt"
"blocked": true,
"message": "Blocked document URL host: private or local address"
}
},
{
Expand Down
5 changes: 4 additions & 1 deletion package.json
Original file line number Diff line number Diff line change
Expand Up @@ -58,11 +58,12 @@
"check:workflow-footguns": "node scripts/check-workflow-footguns.mjs",
"check:test-skip-hygiene": "node scripts/check-test-skip-hygiene.mjs",
"check:test-timing-wait-hygiene": "node scripts/check-test-timing-wait-hygiene.mjs",
"check:atomic-write-hygiene": "node scripts/check-atomic-write-hygiene.mjs",
"check:codex-parity": "node scripts/check-codex-parity-conformance.mjs",
"check:codex-operating-layer": "node scripts/check-codex-operating-layer-conformance.mjs",
"check:platform-runtime-conformance": "node scripts/check-platform-runtime-conformance.mjs",
"check:release-surface": "node scripts/check-release-surface-conformance.mjs",
"lint:evals": "bun run lint:headless-proto && node scripts/ensure-deps.js --no-install --workspace @evalops/contracts && node scripts/verify-evals.js && node scripts/verify-tool-versions.js && node scripts/validate-system-paths.js && node scripts/validate-package-boundaries.js && node scripts/validate-public-package-deps.js && node scripts/check-public-surface-boundary.mjs && npm run check:context-manifest && npm run check:session-wire-contract && npm run check:cli-runtime-conformance && npm run check:rpc-protocol-conformance && npm run check:evidence-integrity && npm run check:maestro-release-gate-events && npm run check:session-replay-fixtures && npm run check:agent-trajectory-fixtures && npm run check:agent-trajectory-replay-fixtures && npm run check:agent-trajectory-score-fixtures && npm run check:agent-trajectory-inspection-fixtures && npm run check:agent-trajectory-scenario-fixtures && npm run check:slack-teammate-runtime-scenarios && npm run check:scripted-scenario-fixtures && node scripts/session-wire-format-codegen.mjs --check && node scripts/headless-protocol-codegen.mjs --check && npm run check:app-server-schema && npm run check:drift-surfaces && npm run check:staged-rollout && npm run check:workflow-footguns && npm run check:test-skip-hygiene && npm run check:test-timing-wait-hygiene && npm run check:codex-parity && npm run check:codex-operating-layer && npm run check:platform-runtime-conformance && npm run check:release-surface && npm run developer-surface:check",
"lint:evals": "bun run lint:headless-proto && node scripts/ensure-deps.js --no-install --workspace @evalops/contracts && node scripts/verify-evals.js && node scripts/verify-tool-versions.js && node scripts/validate-system-paths.js && node scripts/validate-package-boundaries.js && node scripts/validate-public-package-deps.js && node scripts/check-public-surface-boundary.mjs && npm run check:context-manifest && npm run check:session-wire-contract && npm run check:cli-runtime-conformance && npm run check:rpc-protocol-conformance && npm run check:evidence-integrity && npm run check:maestro-release-gate-events && npm run check:session-replay-fixtures && npm run check:agent-trajectory-fixtures && npm run check:agent-trajectory-replay-fixtures && npm run check:agent-trajectory-score-fixtures && npm run check:agent-trajectory-inspection-fixtures && npm run check:agent-trajectory-scenario-fixtures && npm run check:slack-teammate-runtime-scenarios && npm run check:scripted-scenario-fixtures && node scripts/session-wire-format-codegen.mjs --check && node scripts/headless-protocol-codegen.mjs --check && npm run check:app-server-schema && npm run check:drift-surfaces && npm run check:staged-rollout && npm run check:workflow-footguns && npm run check:test-skip-hygiene && npm run check:test-timing-wait-hygiene && npm run check:atomic-write-hygiene && npm run check:codex-parity && npm run check:codex-operating-layer && npm run check:platform-runtime-conformance && npm run check:release-surface && npm run developer-surface:check",
"platform:sdk-smoke": "tsx scripts/check-platform-sdk-contract.ts",
"platform:agentruntime-e2e": "tsx scripts/smoke-platform-agentruntime-lifecycle.ts",
"platform:timeline-e2e": "tsx scripts/smoke-platform-timeline-e2e.ts",
Expand Down Expand Up @@ -182,6 +183,7 @@
"@bufbuild/protobuf": "^2.11.0",
"@crosscopy/clipboard": "^0.2.8",
"@modelcontextprotocol/sdk": "^1.29.0",
"@napi-rs/keyring": "^1.3.0",
"@openai/codex": "^0.135.0",
"@opentelemetry/api": "^1.9.1",
"@opentelemetry/auto-instrumentations-node": "^0.76.0",
Expand Down Expand Up @@ -221,6 +223,7 @@
"postgres": "^3.4.8",
"smol-toml": "^1.6.1",
"string-width": "^8.2.0",
"undici": "^7.25.0",
"uuid": "^14.0.0",
"vscode-jsonrpc": "^8.2.1",
"ws": "^8.20.0",
Expand Down
Loading
Loading