Guardian Docs

Start here:

• ../AGENTS.md is the agent operating contract.
• display-surfaces.md is the current UI/testing surface contract.
• design-board-v1-goal.md records the completed multi-window board milestone.
• release-app-substance-v1-goal.md is the current release-readiness goal.
• dynamic-scenario-coverage-v1-goal.md is the active product slice inside that goal.
• drawer-navigation-surface-v1-goal.md records the previous drawer/navigation slice.
• composer-contract-v1-goal.md records the completed composer contract slice.
• release-ui-readiness-v1-goal.md is the supporting UI-polish goal.
• source-ownership-map.md records where product surfaces should move as route files get thinner.

Canonical commands:

./scripts/device.sh board status
./scripts/device.sh board open
./scripts/device.sh board capture
./scripts/device.sh inventory
npm test
npm run test:web:iphone
npm run test:web:headless
npm run doctor

Use lower-level scripts only when the canonical command cannot express the job:

• device.sh web-slot ... for one explicit managed Chromium slot.
• remote-demo.sh and remote-gui.sh for the remote Air/iOS lane.
• ios-conversation-harness.sh for simulator fixture injection.
• capture-window.sh slot:<name> for a direct slot capture.

Managed Chromium capture first tries macOS window capture. If detached app-mode Chromium or local Screen Recording permissions refuse usable pixels, the harness falls back to a deterministic iPhone-profile web render of the same tracked slot URL while keeping the real board window open.

npm run test:web:headless uses Playwright's repo-local global setup to prewarm /scenarios and /composer-lab before running tests. That keeps cold Expo web bundle compilation out of the first assertion timeout while preserving the normal 60 second timeout for actual UI failures.

The web-server start path is lock-protected and writes its manifest atomically, so simultaneous visible/headless proof commands cannot corrupt .build-info/web-server.<port>.json.

Archived docs and scripts under docs/archive/ and scripts/archive/ are forensics only. Do not use archived ui-bench, parity, or older flow docs as active guidance.