Skip to main content
skill CX Codex

critique · nexu-io — Claude Code skill: 5-criteria design review

critique scores any HTML artifact against five design criteria and builds a standalone HTML report with a radar chart and Keep / Fix / Quick-wins lists.

Apache-2.0 256 installs ★ 108,855 updated 2026-10-25

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add nexu-io/open-design --skill critique -a claude-code
Open on GitHub

What critique does

The critique skill reviews an HTML mockup — landing page, deck, prototype — against five criteria: philosophy, visual hierarchy, detail, functionality, innovation. Each gets a 0–10 score tied to a specific example, such as two typefaces colliding on one line or misaligned grid columns. Scores aren't averaged together, so a weak page inside an otherwise solid deck still drags that dimension down.

The output is a single self-contained HTML file with an inline SVG radar chart, one card per dimension with the reasoning behind its score, and three lists at the bottom: items to leave alone, fixes worth prioritizing, and five-to-fifteen-minute tweaks. The skill can pull the artifact from a project file, from HTML pasted into the chat, or from an artifact the agent just generated, for a self-check pass.

Before scoring, it reads through the whole <style> block and several real content sections rather than judging from a brief alone, and if a project has more than one HTML file, it asks which one to review instead of covering all of them.

What it helps with

  • score a landing page or deck against five design criteria
  • pinpoint where fonts, spacing or style choices collide
  • split fixes into urgent items and quick five-minute wins
  • compare two versions of the same layout
  • run a self-check on generated HTML before showing it to a client

Example prompts for the agent

  • Run critique on the index.html I just generated
  • Give this deck a 5-dimension design review scoring philosophy, hierarchy, detail, function and innovation
  • Compare these two landing page variants using critique and show me Keep / Fix / Quick-wins

How to install critique

1Skills CLI (Vercel)

npx skills add nexu-io/open-design --skill critique -a claude-code

2Manually: put the files in place

~/.claude/skills/critique/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/critique/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about critique

How do I install critique in Claude Code?

Run npx skills add nexu-io/open-design --skill critique -a claude-code; the skill lands at ~/.claude/skills/critique/SKILL.md.

What does the critique skill do?

It scores an HTML artifact across five dimensions — philosophy, hierarchy, detail, functionality, innovation — and produces a standalone HTML report with a radar chart and action lists.

How does critique handle a project with multiple HTML files?

It doesn't review them all at once; it asks which file to check first.

Source text

SKILL.md · Apache-2.0
---
name: critique
description: |
  Run a 5-dimension expert design review on any HTML artifact in the
  project — Philosophy / Visual hierarchy / Detail / Functionality /
  Innovation, each scored 0–10. Outputs a single self-contained HTML
  report with a radar chart, evidence-backed scores, and three lists:
  Keep / Fix / Quick-wins. Use when the brief asks for a "design
  review", "design critique", "5 维度评审", "design audit", or "what's
  wrong with my design".
triggers:
  - "critique"
  - "design review"
  - "design audit"
  - "5 维度评审"
  - "5-dim review"
  - "audit my design"
  - "review my deck"
  - "review my landing page"
  - "评审"
  - "复盘"
od:
  mode: prototype
  platform: desktop
  scenario: design
  upstream: "https://github.com/alchaincyf/huashu-design"
  preview:
    type: html
    entry: index.html
  design_system:
    requires: false
  example_prompt: "Run a 5-dimension critique on the magazine-web-ppt deck I just generated — score philosophy / hierarchy / detail / function / innovation, give me Keep / Fix / Quick-wins."
---

# Critique Skill · 5 维度专家评审

Produce a single-file HTML "design review report" that scores any
artifact across 5 dimensions and proposes actionable fixes. Inspired by
the *huashu-design* expert-critique flow.

## When to use

- After the agent (or user) generates an artifact (deck / prototype /
  landing page) and the user asks "what's wrong with this?" or
  "review this"
- As a self-check loop the agent can run on its own output **before**
  emitting it
- For comparing two variants of the same design

## What you produce

A single self-contained `<artifact type="text/html">` review report
including:

1. **Header** — what artifact was reviewed, date, reviewer ("OD ·
   Critique skill"), 1-line verdict
2. **Radar chart** (inline SVG, no library) showing the 5 scores
3. **Five dimension cards**, each with:
   - Score 0–10 (with band: 0–4 *Broken* · 5–6 *Functional* · 7–8 *Strong*
     · 9–10 *Exceptional*)
   - 1-paragraph evidence (cite specific elements / files / lines)
   - One Keep / Fix / Quick-win bullet
4. **Combined action lists** at the bottom:
   - **Keep** — what's working, don't touch
   - **Fix** — P0 / P1 issues that are visually expensive
   - **Quick wins** — 5–15 minute tweaks with disproportionate impact

## The 5 dimensions

> Each dimension is independent — a deck can be 9/10 on Innovation but
> 4/10 on Hierarchy and the report should say so plainly. Don't average
> away interesting failures.

### 1. Philosophy consistency · 哲学一致性

> Does the artifact pick a clear *direction* and stick to it through
> every micro-decision (chrome / kicker / spacing / accent)?

**Evidence to look for:**
- Is there one declared design direction (e.g. Monocle / WIRED /
  Kinfolk) or is it three styles in a trench coat?
- Does the chrome / kicker vocabulary stay in one register, or does
  page 3 say "Vol.04 · Spring" and page 7 say "BUT WAIT 🔥"?
- Are accent / serif / mono used by the same rule throughout?

**0–4** Three styles fighting each other. **5–6** One direction but
half the elements drift. **7–8** Coherent, occasional drift on edge
pages. **9–10** Every element argues for the same thesis.

### 2. Visual hierarchy · 视觉层级

> Can a stranger figure out what to read first, second, third — without
> being told?

**Evidence to look for:**
- Is the largest type clearly the most important thing on each page?
- Do mono / serif / sans roles match the information's *role* (meta /
  body / display)?
- Lots of "loud" elements competing? Or a clear primary + secondary +
  tertiary tier?

**0–4** Everything shouts. **5–6** Hierarchy works on hero pages but
breaks on body. **7–8** Clear tiers, occasional collision. **9–10** Eye
moves with zero friction.

### 3. Detail execution · 细节执行

> The 90/10 stuff — alignment, leading, kerning at large sizes, image
> framing, foot/chrome polish, edge-case spacing.

**Evidence to look for:**
- Big-stat pages: does the number sit on a baseline, or float?
- Left/right column tops aligned in `grid-2-7-5`?
- `frame-img` + caption proportions consistent across pages?
- Mono labels: same letter-spacing? same uppercase rule?
- Any orphaned `<br>` causing 1-character lines?

**0–4** Visible tape and string. **5–6** Most pages clean, 1–2
ragged. **7–8** Polished, expert eye finds 2–3 misses. **9–10**
Magazine-grade — the kind of detail that makes printed-by-hand
typographers nod.

### 4. Functionality · 功能性

> Does the artifact *work* for its intended use? Click targets, nav,
> readability at presentation distance, copy-paste-ability for code
> blocks, mobile fallback if relevant.

**Evidence to look for:**
- Deck: keyboard / wheel / touch nav all working? Iframe scroll
  fallback?
- Landing: CTA above the fold? Phone number tappable on mobile?
- Runbook: code blocks copyable, mono font, no smart quotes?
- Critical info readable from 4m away (large screen presentation)?

**0–4** Visually fine but doesn't accomplish its job. **5–6** Core
flow works, edge cases broken. **7–8** Robust through normal use.
**9–10** Defensively engineered — handles iframe / fullscreen / paste
/ print without flinching.

### 5. Innovation · 创新性

> Does this push past the median? Is there one element that makes
> people lean in?

**Evidence to look for:**
- One *unexpected* layout / motion / typographic move that wasn't
  required?
- Or 100% safe — could be any deck/landing from any agency?
- Is the innovation *earned* (matches direction) or grafted on
  (random WebGL on a Kinfolk slow-living deck)?

**0–4** Generic AI-slop median. **5–6** Competent and unmemorable.
**7–8** One memorable moment, the rest solid. **9–10** Multiple
moves you'd steal — but each one obviously serves the thesis.

## Scoring discipline (read before you score)

- **Always cite evidence** — "scored 4 because hero page mixes
  Playfair display with Inter sans on the same line" beats "feels
  inconsistent". Numbers without evidence get rejected.
- **Don't average up** — if Hierarchy is 5 because page 3 is broken,
  don't bump to 7 because pages 1 and 2 are fine. The score is the
  *worst sustained band*.
- **Don't grade-inflate** — a 7 means *strong*, not *acceptable*. If
  every score is 7+, you're not reviewing critically.
- **Innovation is allowed to be low** — 5/10 is fine for production
  deliverables. Don't punish *appropriate* conservatism.

## Workflow

### Step 1 — Acquire the artifact

Three modes:

1. **Project file** — user said "review the index.html I just made":
   open it from the project folder.
2. **Pasted HTML** — user pasted code in the chat: read it from the
   message.
3. **Generated by you in this turn** — you just emitted an artifact
   above and want to self-critique: re-read your own `<artifact>`.

If multiple HTML files exist, ask which one (don't review all).

### Step 2 — Read enough to score

Skim the entire `<style>`, then read 6–8 representative content
blocks. **Do not score from frontmatter alone.** The score depends on
*executed* design, not declared intent.

### Step 3 — Score with evidence

For each of the 5 dimensions, write the score and a 30–80 word
evidence paragraph that names specific elements. Use line numbers,
class names, page numbers.

Example:
```
Dimension: Detail execution
Score: 6 / 10
Evidence: Stat-cards on page 3 align cleanly (grid-6, 3×2), but on
page 8 the right column foot sits 2vh higher than the left because
.callout has 3vh top margin while the figure doesn't. Image captions
use mono on page 5 but sans on page 7 — pick one.
```

### Step 4 — Build the action lists

Aggregate the 5 evidence paragraphs into:

- **Keep** (3–5 bullets) — concrete things working that the user must
  not break in the next iteration. Cite by class / page / element.
- **Fix** (3–6 bullets) — must-do, ordered by *visual cost saved per
  minute spent*. Each bullet ≤ 1 sentence.
- **Quick wins** (3–5 bullets) — 5–15 minutes each, high
  signal-to-noise (e.g. "swap `display:flex` for `grid` on page 4 to
  fix the column drift").

### Step 5 — Emit the report HTML

Build a single file:

- Header: artifact name + reviewer credit + date
- Big radar chart (SVG)
- 5 dimension cards in a 1-column or 2-column grid
- Three action lists at the bottom with checkbox affordance

Use the active DESIGN.md tokens if one exists; otherwise default to a
neutral light theme (off-white background, near-black text, one accent
for radar fill).

## Output contract

```
<artifact identifier="critique-<artifact-slug>" type="text/html" title="Critique · <Artifact Title>">
<!doctype html>
<html>...</html>
</artifact>
```

One sentence before the artifact ("Reviewed X across 5 dimensions, see
report below.") and **stop after `</artifact>`** — do not paraphrase
the report in chat; the user will read the artifact.

## Hard rules

- **5 scores, every time** — partial reports (e.g. only 3 dimensions)
  are not allowed.
- **Evidence per score** — no "feels off" / "needs work". If you
  can't cite an element, the score is not justified.
- **Don't grade-inflate** — overall mean above 8 is suspicious; check
  yourself.
- **Don't review your own artifact in the same turn** — the user
  needs to see it first. Self-critique only on explicit request
  ("now critique what you just made").
- **Single-file HTML only** — no external CSS/JS. Inline everything.
- **Radar chart is mandatory** — gives the report a recognizable
  silhouette and lets the user spot weak axes at a glance.

© nexu-io · Apache-2.0 · nexu-io/open-design

Similar skills

All skills for Design →

critique by other authors comes first.

critique NeoLabHQ The critique skill runs finished work past several AI judges: they review it independently, debate their findings, and agree on a shared list of issues. CX Codex +1 1,400installs critique aladicf critique evaluates interface designs from a UX angle, scoring hierarchy and cognitive load. Highlights AI slop patterns and clarifies user flows. CX Codex +1 819installs critique Thientan Soparat The critique Claude Code skill challenges a finished interface and finds grounds to reject it. Run it after the project gates pass. CX Codex +1 ★ 3,216repo stars diagram-design cathrynlavery diagram-design is a Claude Code skill that builds editorial diagrams as standalone HTML files with inline SVG. It covers 51 visual types under a project-specific style guide. CX Codex +1 30,900installs hairline-create lucasmarkes hairline-create generates interactive isometric line figures using the Hairline engine and outputs them as self-contained HTML files. CX Codex +1 7,500installs html-diagram plannotator Generates standalone interactive HTML diagrams without external services or build steps. Helps visualize component topology, state transitions, timelines, and quantitative structures. CX Codex +1 4,600installs generative-ui himself65 generative-ui provides design system rules and templates for interactive HTML and SVG widgets. Useful when you need charts, diagrams, or calculators instead of plain text. CX Codex +1 3,700installs favicon-gen jezweb favicon-gen generates complete favicon packages and web manifests from logos, text monograms, or brand colors. Provides clean HTML markup for site headers. CX Codex +1 2,300installs architecture-diagram Cocoon-AI Generates architecture diagrams for systems, cloud infrastructure, microservices, and network topology as self-contained HTML files with a dark theme. CX Codex +1 2,200installs

From the same set140

The whole set →
mobile-app This skill builds one mobile app screen inside an iPhone 15 Pro frame. Use it for HTML interface prototypes driven by a brief. CX Codex +1 269installs wireframe-sketch wireframe-sketch generates a hand-drawn HTML wireframe on a graph-paper background. Use it during early prototyping when you need rough screen concepts without polished UI components. CX Codex +1 264installs magazine-poster magazine-poster builds a single HTML poster in a Sunday-paper layout: oversized serif headline, two-column body text and six numbered callout sections. CX Codex +1 262installs social-carousel Builds a three-card social carousel — square 1080×1080 panels whose headlines link together into one sentence across the set. CX Codex +1 259installs sprite-animation sprite-animation generates self-contained retro explainer slides with looping CSS pixel art, glitchy typography, and timeline ribbons. CX Codex +1 250installs motion-frames motion-frames builds one 16:9 animated frame in plain HTML and CSS: spinning rings, a wireframe globe, a headline. Hand it to HyperFrames to capture as a video poster. CX Codex +1 249installs email-marketing email-marketing builds a product-launch HTML email with a masthead, hero block, CTA, and specs grid. Use it for newsletters, email blasts, and reusable email templates. CX Codex +1 249installs invoice Generates a single-page printable HTML invoice with sender details, line items, taxes, and bank instructions. Useful when drafting client bills in code editors. CX Codex +1 248installs

Project sponsors

Companies that keep this analytics open