What problem does it solve? Static code reviews and DOM sweeps miss behavioral bugs that only surface when a real user types, clicks, and navigates. This Skill walks through a live web app as a locked persona, requiring proof of interaction before any verdict, so silent failures like dead buttons, stale caches, and layout collapses are actually caught. ## Core Features & Use Cases - Interaction-first walkthrough: A mandatory Interaction Manifest logs every typed input, clicked action, console read, and screenshot per page; without it the only legal verdict is Incomplete. - Hard quality gates: Auto-fail thresholds for console errors, network 5xx, layout collapse, axe-core Critical/Serious violations, and a pragmatic performance budget (LCP, CLS, INP). - Structured stress coverage: An 11-scenario battery (keyboard-only, heavy data, destructive confidence, round-trip workflows, data seasoning) plus multi-pane viewport stress and automated axe-core accessibility scans. - Use Case: Before shipping a new dashboard, run a full audit that walks every route as a first-time user, exercises all 31 interactive elements, and returns a ranked Top 5 findings list with reproduction steps, evidence paths, and suspected code locations. ## Quick Start Ask the agent to run a UX audit of your deployed app as a specific persona, for example: audit the app at its live URL as a busy first-time user and report every Critical and High finding.
