onewave-ai/claude-design-critic
Audit a website or UI and de-AI it — find the patterns that make it look template-generated (in both design and copy) and return specific, prioritized fixes that push it toward editorial, premium, human-crafted. Reviews layout, color, type, spacing, motion, accessibility, AND the words. Use after building or generating any frontend, or when a site looks "fine but generic" and you can't say why.
npx skills add https://github.com/OneWave-AI/claude-skills --skill claude-design-critic
Some sites are technically fine and still scream "an AI made this in one shot." This skill names exactly why and tells you what to change. It reviews the design *and* the copy, because generated-looking text gives a build away as fast as a centered hero and a purple gradient.
Use it as the finishing pass on claude-landing-composer output, on any vibe-coded frontend, or on a live site that feels generic. It pairs with the design tokens from claude-design-system-architect (audit against them when they exist).
A URL (WebFetch + screenshot), a local project (read the components/styles), or pasted code. Establish: what is this page for, who's the audience, and what's the one action it wants? A critique without the goal is just opinion.
Go dimension by dimension and flag concrete instances, not vibes:
Layout & composition
Color
Type
Spacing & detail
Motion
prefers-reduced-motion ignored.Iconography & assets
Copy (review this as hard as the design)
Accessibility & semantics
# Design Critique — {target}
Generated: {timestamp} · Goal: {…} · Verdict: {SHIP / FIX FIRST / REBUILD SECTION}
## The 30-second read
The 2–3 things making this look generated, in plain language.
## Findings (prioritized: High → Low)
| # | Dimension | What | Where (selector/line/section) | Why it reads as AI | Fix |
## Copy rewrites
"{exact current line}" → "{rewritten line}"
## Quick wins (under 30 min, high impact)
## Bigger moves (worth the time)
## What's already good (keep it)
Be specific enough that someone could apply every fix without asking a question. Point to the file/section/line. Where tokens exist, cite the token that should have been used.
Offer to implement the high-priority fixes directly (edit the components/copy) and re-audit, or hand a rebuild to claude-landing-composer.
Premium = intentional layout rhythm, restrained palette, real type scale, generous whitespace, motion with purpose, human copy. Generated = centered everything, rainbow/purple gradients, default sans, uniform padding, fade-up-on-all, and "elevate your workflow." Push every target from the second toward the first. Never purple, never emoji-as-UI.
Critique design AND copy · every finding is specific and located · cite tokens where they exist · prioritize by impact · offer to apply, then re-audit.
Take onewave-ai/claude-design-critic from the repository into ~/.claude/skills for personal
use, or into .claude/skills inside a project.
The agent identifies a skill by the name field in its header. Two skills with the
same name cannot sit side by side — one of them will be ignored.