← ClaudeAtlas

ui-reviewlisted

Read-only UI review that runs impeccable critique, the ux-laws checklist, an accessibility audit and (for public pages) a metadata audit, then merges them into one deduplicated list, worst first. Changes nothing until the user picks fixes.
whoiscarlo/agent-skills · ★ 1 · Code & Development · score 70
Install: claude install-skill whoiscarlo/agent-skills
# /ui-review One command for the four reviews that keep getting run together. They overlap (contrast, target size, labels, states), so running them apart gives the same problem three times. This runs them in a fixed order and merges the output. **Read only.** Do not edit any project file during the review. The only files written are the ones impeccable writes into `.impeccable/` for its own history. ## 0. Resolve the target 1. Turn the argument into a concrete source path or URL (prefer the source path; ports drift). If there is no argument, ask for one and stop. 2. Decide the surface type: - **public**: a marketing, landing, pricing, blog or shareable page; anything search engines index. Runs step 4. - **app**: a product screen behind login (dashboard, settings, editor, admin tool). Skips step 4. `--public` or `--app` overrides the guess. Say which one you picked in the report header. 3. Read the target file and the components it renders, once. Later steps reuse this reading. ## 1. Design critique (impeccable) Run `/impeccable critique <target>` and follow its playbook (`~/.claude/skills/impeccable/reference/critique.md`), including its two isolated sub-agents and its degraded banner if it cannot use them. One change for this command: **hold its closing question.** Keep its findings, heuristic scores and strengths for the merge. The only question in the final reply is the one at the end of step 5. Rule of thumb: for **app** surfaces, judge