How to use the Audit AI Design Slop agent skill

What an agent skill is and what this one does

An agent skill is a folder with a SKILL.md file that tells an AI agent when and how to perform a task. Once installed, the agent reads the skill when your request matches its description.

This skill makes the agent act as a diagnostic auditor. It inspects the artifact you provide, looks for decorative stacking such as repeated glows, cards and filler copy, and for established UI defects such as low contrast, overflow, broken controls and information only available on hover. Each finding cites specific evidence, gets a priority, and usually recommends removing an element rather than redesigning it.

• Output is a short verdict, a checked-scope note, a findings table with priority P0 to P3, and a list of unknowns.
• The skill avoids guessing whether AI made the design, avoids assigning a numeric slop score, and avoids imposing a new style.
• It is not a redesign tool. It will not propose new fonts, palettes or brand direction unless you explicitly ask for them.

Install the audit-ai-design-slop skill

Download the ZIP for this skill from this page and unzip it. The package contains agents/openai.yaml, REFERENCES.md and SKILL.md. Place the unzipped skill folder in one of these locations, depending on how you want to use it.

For a personal install available across your projects, put it in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder.

If you use the Claude apps instead, upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/audit-ai-design-slop/
• Project only: .claude/skills/audit-ai-design-slop/ inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

Trigger the skill with a natural request

You do not need a special command. The skill's description covers requests for evidence-backed design feedback, an anti-slop review, or a removal-first cleanup plan without a speculative redesign. Ask in those terms and attach or point to the artifact.

Give the agent something it can inspect: a screenshot, a rendered page, a link you can open in the session, a mockup, or source code. The skill audits only what is in the request, so state the screen and viewport if it matters.

• "Run an anti-slop review on this landing page screenshot before we launch."
• "Give me a prioritised cleanup list for this signup flow, removals first, no redesign."
• "Audit this React component for clutter and accessibility problems."

What to provide and what it needs

The source lists no required tools, it runs no scripts, and it uses no network access. Its work comes from the artifact you supply and from the reference article it links to for full-page reviews.

Because the skill only sees what you provide, its findings are limited to that evidence. Anything it cannot verify, such as a hover state it never saw or a flow it did not open, is listed under Unknowns rather than reported as fact.

• Provide screenshots of the relevant viewports and states, or the source code.
• For a full-page or full-site review, the skill reads a referenced catalogue (no-ai-design-slop/ARTICLE.md) and uses only relevant sections.
• Expect the five to eight highest-impact findings by default, with repeated instances grouped.

Read the output and act on it

Start with the verdict and the removal recommended first. The findings table gives each item a priority: P0 for things that block completion, severe accessibility issues or deceptive proof, down to P3 for minor polish.

The skill ends with the single removal or correction likely to produce the largest improvement. Treat that as your first change, then re-run the audit on the updated screenshot to check the result.

• P0: blocks completion, severe accessibility issue, or deceptive proof
• P1: materially harms comprehension, trust, navigation or interaction
• P2: repeated slop or inconsistency weakening hierarchy
• P3: minor polish with limited impact