How to use the Design Critique Review agent skill (critique)

What is an agent skill and what does critique make the agent do?

An agent skill is a folder of instructions, held in a file called SKILL.md, that tells an AI agent how to handle a specific kind of task. When the skill is installed and your request matches it, the agent follows its workflow instead of improvising.

The critique skill tells the agent to run a five-dimension expert design review on an HTML artifact. The five dimensions are Philosophy consistency, Visual hierarchy, Detail execution, Functionality and Innovation. Each is scored 0 to 10 and must be supported by specific evidence, such as named elements, class names, page numbers or line numbers.

The output is one self-contained HTML report. It includes a header with a one-line verdict, a radar chart drawn in inline SVG showing the five scores, a card for each dimension, and combined lists of Keep, Fix and Quick-wins.

How do you install the critique skill?

Download the ZIP file from this page and unzip it. You can install the skill in one of two places. For a personal setup that applies across your projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead, you can upload the skill there under Settings, then Capabilities, then Skills.

The skill folder includes two files: SKILL.md, which contains the instructions the agent follows, and example.html, a sample file you can use to test the skill.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the project's .claude/skills/ folder
• Claude apps: Settings, then Capabilities, then Skills, then upload

How do you trigger the critique skill with a natural request?

You do not need special syntax. Ask for a review in plain language. The skill's trigger phrases include "critique", "design review", "design audit", "review my deck", "review my landing page", "audit my design", "5-dim review" and "what's wrong with my design". The skill also responds to the Chinese phrases "5 维度评审" and "评审".

Name the file or artifact when you can. If the project has several HTML files, the agent will ask which one to review rather than reviewing all of them.

The skill is designed to work in two ways: on files you provide, and on the agent's own output. In the second case, the agent can re-read the HTML it just produced and critique it before handing it over.

• "Run a design review on index.html."
• "Review my deck and give me Keep, Fix and Quick-wins."
• "What's wrong with my design? Score it on all five dimensions."
• "Compare these two variants of the landing page."

What does the skill check, and how are the scores set?

Each dimension is scored independently, so a deck can score 9 on Innovation and 4 on Visual hierarchy, and the report should say so. The agent is instructed not to average away weak areas. A score reflects the worst sustained band in that dimension, not the best parts.

The bands are 0 to 4 for Broken, 5 to 6 for Functional, 7 to 8 for Strong and 9 to 10 for Exceptional. The skill asks the agent to read the whole stylesheet and several content blocks before scoring, rather than judging from the description alone.

The scores are judgement based on reading the markup and styles. They are not objective measurements, so the skill does not replace accessibility audits or testing on real devices.

• Philosophy consistency: one clear direction, applied through every detail
• Visual hierarchy: whether a reader knows what to read first, second and third
• Detail execution: alignment, leading, spacing and polish
• Functionality: whether the artifact does its job, such as navigation, calls to action and readability
• Innovation: whether there is a memorable move that serves the design's purpose

What does a good result look like?

A useful report names what it saw. For example, a detail score might say that stat cards align on one page but a caption uses a different typeface on another, and that is the reason for the score. The Keep list names working parts you should not change. The Fix list is ordered by visual cost saved per minute of work. The Quick-wins list holds tweaks that take roughly 5 to 15 minutes each.

Treat the report as a checklist for your next iteration, not a final verdict. If a score seems harsh, check the cited evidence first. The skill is built to reject scores that have no evidence behind them.

What are the safety notes?

According to the item listing, the critique skill runs no scripts and uses no network connection. It also lists no requirements. The skill works by reading and writing HTML and does not ask for other tools.

Because the skill reads your files, only point it at artifacts you are comfortable sharing with the agent you are using. Review the SKILL.md file yourself if you want to see exactly what instructions the agent will follow.

• Runs scripts: no
• Uses network: no
• Requirements listed: none

Tips for getting better reviews

Give the agent a clear artifact and a clear goal. A review of a slide deck for a live talk should weigh readability at distance more heavily than a review of a landing page for a website, and you can say so in your request.

Use the skill on comparisons as well. The skill is written to handle two variants of the same design, which is useful when you are choosing between directions.

Keep in mind what the skill is not designed for. It is written for HTML artifacts because it reads markup and styles directly. It is not meant for non-HTML designs.

• Name the exact file so the agent does not have to guess
• Ask for a comparison when choosing between two directions
• Use the Fix list in order, starting with the highest visual cost saved
• Re-run the review after changes to check the scores moved for the right reasons