extract-design-system

An agent skill for extracting design primitives from public websites and generating starter token files for the current project.

Skills follow Anthropic's Agent Skills overview and are installable through the skills.sh ecosystem.

This repository is skills-first. It publishes an installable extract-design-system skill, and the bundled CLI is the executable workflow the agent runs under the hood.

Quick Start

Install the skill:

npx skills add arvindrk/extract-design-system

Run it from a supported coding agent with a public website:

Extract the design system from https://stripe.com and generate starter token files for this project.

Expected starter outputs:

• .extract-design-system/raw.json
• .extract-design-system/normalized.json
• design-system/tokens.json
• design-system/tokens.css

About This Repository

This repository contains a focused agent skill for reverse-engineering a starter design system from a public website.

The skill is designed to give coding agents a repeatable workflow for:

• analyzing a public website's colors, typography, spacing, radius, and shadows
• normalizing the extracted output into a stable structure
• generating project-local starter token files
• summarizing findings before broader styling or app changes

Available Skill

extract-design-system

Extract a starter design system from a public website and turn it into local token files.

Use when:

• You want an agent to analyze a public website's visual primitives
• You want starter design tokens for an existing project
• You want a repeatable workflow instead of ad hoc prompting
• You want the agent to summarize findings before applying broader changes

What it produces:

• .extract-design-system/raw.json
• .extract-design-system/normalized.json
• design-system/tokens.json
• design-system/tokens.css

Installation

Install the skill from GitHub with the skills CLI:

npx skills add arvindrk/extract-design-system

You can also browse the skill on skills.sh.

Usage

Once installed, the skill is available to supported coding agents. The agent should use it when a request matches the extraction workflow.

Example prompts:

Extract the design system from https://stripe.com and generate starter token files for this project.

Analyze https://linear.app and summarize the design primitives before generating local tokens.

Extract colors, typography, spacing, and radius tokens from https://vercel.com and save them for this codebase.

Review https://tailwindcss.com, extract the most likely design primitives, and generate starter token files only if the results look usable.

How The Skill Works

The skill is designed to:

• confirm that the target site is public and reachable
• set expectations about scope and limitations
• run the extraction workflow through the bundled CLI
• summarize what was found in the normalized output
• generate starter token files for the current project
• ask before modifying existing app code, styles, or configuration

Scope And Limitations

• Public websites only
• Single-page extraction workflow
• Starter tokens, not a full component library
• No framework config patching or automatic app rewrites
• Dynamic, protected, or highly script-driven sites may yield incomplete output
• Extraction is useful for initialization, not pixel-perfect reproduction

Security Considerations

• Target websites are untrusted third-party input and may influence extracted output
• Generated tokens should be reviewed before being treated as authoritative design decisions
• The skill is intended for extraction and starter token generation, not automatic broad app rewrites
• Use only with public websites that you are comfortable fetching and analyzing at runtime
• Ask for confirmation before applying extracted output to existing app code, styles, or configuration

Repository Structure

This repository currently exposes:

• skills/extract-design-system/SKILL.md - Main skill instructions
• skills/extract-design-system/references/ - Supporting workflow and output references
• src/ - Bundled CLI implementation used by the skill
• tests/ - CLI and normalization test coverage

MCP Server

The package also ships an MCP server (extract-design-system-mcp) that exposes the same workflow as structured tool calls for Cursor, Claude Desktop, and any MCP-compatible agent.

Tools

Cursor (project-local)

Create .cursor/mcp.json in your project root:

{
  "mcpServers": {
    "extract-design-system": {
      "command": "npx",
      "args": ["-y", "extract-design-system-mcp"]
    }
  }
}

Restart Cursor. The server appears under Settings → Features → MCP with all 3 tools listed.

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "extract-design-system": {
      "command": "npx",
      "args": ["-y", "extract-design-system-mcp"],
      "cwd": "/path/to/your/project"
    }
  }
}

Claude Desktop does not infer a working directory from context, so cwd is required. It controls where design-system/tokens.json and .extract-design-system/ are written.

Local Development

The skill's executable workflow is backed by the published npm CLI in this repository.

npm install
npx playwright install chromium
npm run typecheck
npm test
npm run build
npm run dev -- --help
npm run cli -- --help

Useful local commands:

npm run dev -- https://example.com
npm run dev -- https://example.com --extract-only
npm run cli -- https://example.com
npm run cli -- init

Community And Support

• Read the contribution guide in CONTRIBUTING.md
• Review expected community behavior in CODE_OF_CONDUCT.md
• Report security issues using the process in SECURITY.md
• Support ongoing maintenance through GitHub Sponsors

Notes

• Node.js 20+ is required
• If extraction fails because Chromium is missing, run npx playwright install chromium
• Extraction quality depends on the target site's DOM and CSS exposure

License

MIT