ui-reviewlisted

# ui-review Audit a web page across common breakpoints with `agent-browser`, combining programmatic in-page checks (this skill's `scripts/detect.js`) with visual screenshot inspection. Report findings; fix only if the user asked. ## Inputs - **URL**: from the user's message. If none given, detect a local dev server (try `agent-browser open localhost:3000` then 5173, 8080, 4200 — a page title confirms a hit) or ask. - **Pages**: default to the given page only. If the user says "the whole app", snapshot the nav (`agent-browser snapshot -i -u`) and review the top ~5 routes. - **Codebase**: if the URL is a local dev server and the project source is in the working directory, findings must be mapped to source files (see "Codebase mapping" below). - **Auth**: if the page sits behind a login wall, log in once via `agent-browser fill` / `click` before the breakpoint loop. Don't close the browser mid-review — that drops the session and forces a re-login. ## Breakpoints | Name | Viewport | |---|---| | mobile | 360 x 800 | | tablet | 768 x 1024 | | laptop | 1366 x 768 | | desktop | 1920 x 1080 | ## Workflow Work in a scratch dir for screenshots — create it first, and **always use absolute paths for file arguments** (screenshot output, diff baselines): agent-browser resolves relative paths against its own process cwd, not your shell's, and fails with "No such file or directory". `DETECT` = this skill's `scripts/detect.js` (resolve path relative to this SKILL.md). ```