cortex-ui-extras

Run axe-core via Playwright against the project's running dev server, targeting WCAG 2.1 AA. Produces ui-check-results/a11y.json (compact violation JSON). Layer 2 of the design enforcement stack — runtime a11y verification that catches rendered issues static JSX analysis cannot (color contrast on real renders, ARIA in context, dynamic state). Use when user says "/ui-a11y", "check accessibility", "run a11y check", "WCAG audit", or when /ui-check invokes Layer 2.

Generates a DESIGN.md constraint document and a Tailwind v4 @theme semantic token starter for a new project. Use when user says "/ui-brief", "generate design brief", "create DESIGN.md", "set up design tokens", or wants to initialize a project's design system. This is Layer 0 of the design enforcement stack — run once at project setup.

Run the full UI design enforcement pipeline — Layer 0 (DESIGN.md + shadcn presence, warns only) → Layer 1 (/ui-lint, blocks on failures) → Layer 2 (/ui-a11y if a dev server is reachable, skips if not). Writes ui-check-results/summary.json after every run. Use when user says "/ui-check", "run UI checks", "check frontend quality", "run the full UI check pipeline", or wants to verify design enforcement before committing.

Captures a screenshot of a URL via Playwright, evaluates it against a five-criterion visual quality rubric using Claude Vision (two-call UICrit pattern), and writes a scored JSON verdict to ui-check-results/judge.json. Layer 3 of the design enforcement stack — human-triggered advisory scorecard, never a CI gate. Always exits 0. Use when user says "/ui-judge", "judge the UI", "visual quality score", "score the UI", "run the visual judge", "design scorecard", or wants a scored visual quality report of the current UI.

Run ESLint and Stylelint in auto-fix-then-report mode and write results to ui-check-results/lint.json. Use when user says "/ui-lint", "lint the UI", "run UI linting", "check frontend lint", or asks to fix and report frontend code style issues.