How to use the Product Design Critique agent skill

What the skill makes the agent do

An agent skill is a folder with a SKILL.md file that gives the agent specific instructions for one kind of task. This skill tells the agent to run a structured critique rather than offer general opinions. The stated goal is a prioritized list of changes a designer or engineer can act on, not adjectives.

The agent works through nine categories in order: clarity of the user job, visual hierarchy, affordance and signifiers, states, inputs and forms, accessibility, consistency, copy, and edge cases. Each finding is one issue paired with one suggested fix. The output is grouped as must-fix (blocks ship), should-fix (before broader rollout), and nice-to-fix, and it always ends with a list of strengths to keep.

• It recommends changes; it is designed not to redesign the product.
• It asks for user context before critiquing, because critique without it becomes taste.
• Accessibility checks reference WCAG AA as a minimum and 44×44 px touch targets on mobile.

Install the design-critique skill

Download the skill ZIP from this page and unzip it. The skill is a single SKILL.md file. Put the unzipped folder in one of two places: ~/.claude/skills/ to make it available across your projects, or a project's .claude/skills/ folder to share it with one codebase.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

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

How to trigger the skill with natural requests

You do not need a special command. Ask for a critique in plain language and point the agent at the artifact. The skill is meant to be used when a designer or engineer asks for feedback on a screen, mock, or live UI, when a feature is shipping and needs a final UX read, or when a flow may be causing drop-off.

Be ready to answer context questions. The skill asks for who the user is, what job they are doing on the screen, what success looks like, and where the screen sits in the flow. Answering these up front gives a more useful critique.

• "Critique this checkout screen before handoff. Users are returning customers paying with saved cards."
• "Give me a design critique of the onboarding flow for admins setting up their first team."
• "Review this settings panel for UX problems before we roll it out to all accounts."

What a critique looks like

The skill defines a fixed output format. Findings are grouped by severity and each one follows the same pattern: a category, a one-line issue, and a one-line suggestion. Strengths are listed too, so the designer knows what to keep in the next round.

Use the severity groups to decide what to act on first. Must-fix items block ship, so they come before any nice-to-have polish.

What the skill needs and its safety profile

The skill lists no required dependencies. It does not run scripts and does not use the network. Its only listed file is SKILL.md, so there is no executable code to review before installing it.

Because it is a text instruction file, its effect comes from what the agent does with the artifact you give it. Provide the screen, mock, or description you want reviewed and the context it asks for.

• Requires: nothing listed.
• Runs scripts: no.
• Uses network: no.
• Files: SKILL.md only.
• Licence: MIT.

When not to use it

The skill is designed to stay out of three situations. It is not for full redesign requests, since the skill says critique is not redesign. It is not for very early concepts with no artifact, where it advises sketching first. It is also not for work with no stated user context, where it asks for that context before proceeding.

If you need a new layout or a rebuilt flow, use a design project instead. If you have only a rough idea, sketch it first and critique the sketch.

Tips for better critiques

The quality of the critique depends on the context you give. A specific user role and a one-sentence job produce findings that are easier to act on than a general request to review the UI.

Ask for approval explicitly if you want it. The skill treats a critique as feedback, not sign-off, so a clean critique does not mean the design is approved.

• Name a specific user role and competence, not just "users".
• State what the user should be able to do after this screen that they could not do before.
• Describe what comes before and after the screen in the flow.
• Include the state you care about, such as the error or empty state, if it is not visible in the artifact.
• Ask for a separate sign-off if you need one.