Bulk import from openclaw/skills official Archive (⭐3,600+).
Quality filter: SKILL.md 800B-30KB with proper YAML frontmatter.
Coverage areas:
- Crypto / Web3 / DeFi / Trading bots
- Feishu / DingTalk / Enterprise WeChat
- Business systems (freelance ops, performance eng, legal docs)
- Document templates / PPT design / OCR
- Email / CRM / Customer service
- Financial analytics / Stock trading
- AI tools / Agent collaboration / Model management
Extract Design
A Claude Code skill for extracting a webpage's design language into a reusable HTML style reference file. Perfect for learning design systems, creating AI-generated pages in a specific style, or building design token systems.
What This Does
Extract Web Style analyzes any webpage and produces a universal style specimen HTML — not a clone, but an extraction of the underlying design system. It captures:
- Primitive tokens — raw colors, fonts, sizes, spacing
- Semantic tokens —
--color-text-primary,--color-bg-page, etc. - Component archetypes — buttons, cards, inputs, navigation patterns
- Interaction rules — hover states, transitions, animations
- Theme variants — light/dark mode behavior
The output is a self-contained HTML file that another AI can read and use to generate new pages in the same visual style.
Key Features
- Style System Extraction — captures the design DNA, not just DOM copy
- Semantic Token Naming — abstracts raw values into meaningful roles
- Dark Mode Support — extracts and validates both light and dark themes
- Component Archetypes — compresses similar components into reusable patterns
- Signature Animation Detection — identifies brand-specific motion patterns
- Machine-Readable Manifest — JSON design manifest embedded for AI consumption
Installation
For Claude Code Users
git clone https://github.com/VintLin/extract-design.git ~/.claude/skills/extract-design
Then use it by typing /extract-design in Claude Code.
Manual Copy
git clone https://github.com/VintLin/extract-design.git ~/.claude/skills/extract-design
Usage
Extract a Website's Design
/extract-design
> "Extract the design system from https://factory.ai"
The skill will:
- Analyze the webpage's visual system
- Extract typography, colors, spacing, motion
- Identify component patterns and states
- Generate a universal style specimen HTML
- Save it to
assets/theme/<site-name>-style-specimen.htmlinside the skill directory
Create Pages in the Extracted Style
/extract-design
> "Use the factory.ai design system to create a landing page for my open-source project"
The skill will read the specimen file and generate new pages using the same design language.
Output Structure
Each extraction produces:
| File | Purpose |
|---|---|
SKILL.md |
The extraction skill itself |
scripts/extract-styles.py |
Built-in Playwright extraction script |
assets/theme/*-style-manifest.json |
Structured style manifest (JSON) |
assets/theme/*-style-specimen.html |
Universal style specimen HTML |
references/ |
Supporting documentation and templates |
Architecture
This skill follows a progressive disclosure design:
| File | Purpose | Loaded When |
|---|---|---|
SKILL.md |
Core workflow and extraction rules | Always (skill invocation) |
assets/theme/*.html |
Extracted design references | When generating pages in that style |
The skill prioritizes:
- Abstraction over fidelity — same design language, not same page
- Semantic naming —
--color-text-primarynot--color-base-900 - State completeness — hover, focus, active, disabled states
- Theme awareness — proper dark mode token mapping
Philosophy
-
Extract the system, not the page. A good extraction should let you recreate countless pages in the same style.
-
Names should explain purpose. Raw hex values mean nothing.
--color-accentmeans everything. -
States matter as much as defaults. A button without hover states isn't a complete component.
-
Themes are first-class citizens. Light and dark modes need equal attention.
-
Signatures make it memorable. Brand-specific animations (like Factory.ai's stripe overlay) are worth preserving.
Requirements
- Claude Code CLI
- Python 3.8+
- Playwright:
pip install playwright && playwright install chromium - Web access (for fetching target pages)
Credits
Created with Claude Code.
License
MIT — Use it, modify it, share it.