What problem does it solve? UX reviews are often ad-hoc and inconsistent, missing entire classes of issues like absent capabilities, misleading empty states, or performance regressions. This Skill provides a repeatable, evidence-based audit process for one screen at a time, benchmarking against the Designing Interfaces pattern language and a shared UX checklist so findings are grounded in code lines, screenshots, or measured metrics rather than opinion. ## Core Features & Use Cases - Three-layer audit: L1 static code reading for missing states and patterns, L2 screenshot-based visual review for hierarchy and responsive behavior, and L3 dynamic journey automation with CLS/LCP/INP instrumentation via agent-browser over CDP. - Pattern catalog benchmark: Walks Jenifer Tidwell's pattern families (navigation, layout, input, feedback, data display) to tag patterns as solid, partial, or absent-but-expected. - Closed-loop feedback: Generalizable gaps and exemplary good cases are fed back into the companion ux skill checklists, and each audit is saved as a worked example for future runs. - Use Case: Run an L1 audit of a settings page to catch a fetch with no error branch that renders a permanent skeleton, then confirm the visual hierarchy with L2 screenshots before filing ranked findings. ## Quick Start Ask the agent to run a ux-audit on a specific page or surface, optionally scoping it with --l1, --l2, or --l3 to choose the audit layer.
