How to use the anti-ui-slop agent skill (install and trigger)

What an agent skill is and what this one changes

An agent skill is a folder containing a SKILL.md file. A coding agent reads that file when a request matches the skill's description, and then follows its instructions for that task. You do not run the skill yourself.

With anti-ui-slop installed, the agent starts from your product's existing components, design tokens and visual language. It then loads one playbook that fits the request, such as new work, polish, audit or iOS. The skill aims for interfaces that look intentional rather than generic, and it ends by rendering and fixing obvious layout problems if the environment allows it.

• Starts from your existing components and tokens, so new work extends them instead of inventing a new system.
• Chooses at most one playbook per request: new work or major redesign, product or dashboard work, polish, simplification, explicit audit, or native iOS.
• Covers required states such as loading, empty, error and success, according to the playbooks.
• Finishes by rendering and inspecting once, then fixing clipping, overlap, distorted media, inaccessible controls or inert interactions, when the environment supports rendering.

How do you install the anti-ui-slop skill?

There are two routes. You can download the ZIP from the skill's page on this site and unzip it into a skills folder. Use ~/.claude/skills/ to make the skill available across your projects, or use .claude/skills/ inside a single project to share it with that project.

If you use the Claude apps, you can instead upload the skill under Settings, then Capabilities, then Skills. The skill is released under the MIT licence.

• Download the ZIP from this page.
• Unzip it so the anti-ui-slop folder, which contains SKILL.md, sits inside the skills folder.
• For personal use, the folder is ~/.claude/skills/. For a single project, use .claude/skills/ in the project root.
• Or upload the skill in the Claude apps under Settings, then Capabilities, then Skills.

How do you trigger the skill with natural requests?

The agent decides when to use the skill by matching your request against its description. The simplest way to be sure is to include one of the trigger phrases from the skill's description in your message.

The description also covers designing, implementing, redesigning, critiquing and pre-ship reviewing a web or iOS interface. A plain request that names the screen and the kind of work is usually enough.

• Say 'anti-ui-slop' or 'stop UI slop' to name the skill directly.
• Say 'ground this UI in real screens' when you want visual references considered.
• Say 'run the UI finish gate' before shipping a screen.
• Describe the screen and the work, for example: 'Redesign the billing settings page in src/pages/billing.tsx using our existing components.'

What should you give the agent for good results?

The skill works best when the agent can see your product. Point it at the screen file or describe the screen in a sentence. The skill's own instructions say the product brief, existing UI, components, tokens and constraints outrank the skill, so the more of those you provide, the better the fit.

Be specific about the goal. A request to critique a screen returns at most three material findings, ordered by user impact, with the smallest concrete fix for each. That output is more useful when you say what the screen is for.

• Give the file path or a short description of the screen.
• Mention the existing component library or design tokens if they are not obvious from the project.
• State whether you want a build, a redesign, a critique or a pre-ship check.
• Name the platform: web or native iOS.

What does the skill need from your setup?

The only hard requirement is a coding agent that can read project files. The skill is designed for Claude Code, Codex, Cursor and GitHub Copilot, and it works in any agent that can read project files and fetch a URL.

Two things are optional. The paid UIZZE MCP provides the find_ui_references and find_ui_materials tools for visual references. The ability to render the UI in your environment enables the final finish check. The free skill and public catalogue work without an account, token, MCP connection or dependency.

• Required: a coding agent that can read project files.
• Optional: a paid UIZZE MCP account, which supplies find_ui_references and find_ui_materials.
• Optional: an environment where the UI can be rendered for the finish check.

What are the safety notes for scripts and network access?

The skill does not run scripts. Its listing says it uses the network. The network use is tied to the optional UIZZE MCP and to the agent fetching URLs, not to any bundled executable. When the MCP is used, it follows the host's normal connection and authentication flow.

The agent should not claim the MCP is connected unless the host actually returned a result. If the reference tools return nothing, the skill tells the agent to continue silently. Before installing, you can read the included files yourself, including SKILL.md, MODIFICATIONS.md and the files in reference/ and references/.

• No scripts or executables are included, according to the listing.
• Network use is optional and goes through the paid UIZZE MCP or ordinary URL fetching by the agent.
• Review SKILL.md and the playbooks before installing if your organisation requires it.
• Do not put secrets into a request that you would not share with a third-party reference service.

What tips help you get more from the skill?

Use the skill for interface work only. It adds guidance that applies to screens, so it is not useful for backend tasks and can add noise there. For a design check on an existing screen, the audit playbook is the one to ask for, since it only reports what can be observed.

• Ask for an audit when you want findings only; the audit does not change the product unless you also ask for fixes.
• Ask for simplification when a screen has too many actions or decorative containers.
• Keep your design tokens and components in the project so the agent can find them.
• Ask the agent to run the finish check after changes so rendering problems get caught.