Skills that stop your UI from looking AI-generated.

A small, versioned set of Claude skills for building real web apps. The flagship gives an agent actual design vocabulary, tokens, and an anti-slop ban list, so output looks deliberate instead of templated. Plain SKILL.md files that work in Claude Code, Codex, and Cursor.

Five skills, one quality bar

The default pair for any web work is design-system-anti-slop plus output-quality. Add the rest as the task needs them. This list is also served as JSON at /skills.json.

design-system-anti-slop

hard

Real design vocabulary, design tokens, a layout system, and an explicit anti-slop ban list, so every UI has clear hierarchy and looks deliberate rather than generic. The default skill for all visual work.

designuiuxtokensaccessibility

frontend-implementation

medium

Turn a design into clean, maintainable frontend code with real component states (loading, empty, error) and a performance budget.

reactastrotailwind

output-quality

medium

A quality bar for agents: no placeholders shipped, no fake TODOs, real states, and honest reporting when something fails.

qualityanti-laziness

security-review

medium

A pre-ship security pass: secrets handling, injection, XSS and CSRF, CORS, headers, upload validation, rate limits, and auth checks.

securityreviewweb

prd-planning

medium

Turn an idea into a PRD, a smallest-unit task breakdown with story points, a kanban, and a branch, PR, review, and merge flow.

prdplanningscrum

Most generated UIs fail on vocabulary, not code

Beginners can prompt an agent to build something, but without design vocabulary the result stays generic. These skills give the agent the words and the rules: hierarchy, spacing scale, one accent color, consistent radius, real interaction states, and a hard ban on the patterns that make output look machine-made (purple mesh gradients, three equal cards, fake dashboards, below-the-fold CTAs, and the em-dash in copy). This very page is built from those tokens as the proof.

Drop a skill where your agent looks

A skill is a folder with a SKILL.md. Copy it into your user skills directory for Claude Code (Windows example below), or reference it directly in Codex and Cursor.

# Claude Code, per-user (run from inside the repo)
"design-system-anti-slop","output-quality" | ForEach-Object {
  Copy-Item -Recurse -Force ".\skills\$_" "C:\Users\<you>\.claude\skills\$_"
}
Then the agent reads SKILL.md on its own.