Mode d’emploi
- Copiez la commande et examinez le paquet avant de l’installer dans votre environnement agent.
- Consultez le document original pour les outils, identifiants et environnements requis.
- Décrivez les entrées et le livrable souhaité ; comparez la sortie aux sources.
npx skills add https://github.com/horuz-ai/claude-plugins --skill website-clonerAvant de commencer
- L’installation n’inclut ni abonnement au modèle, ni quota API, ni accès aux fichiers.
- Étoiles et installations reflètent l’activité, pas un test indépendant de qualité ou de sécurité.
Rendu depuis le dépôt source en conservant titres, exemples, code, tableaux, liens et images.
Website Cloner Skill
Clone any website with pixel-perfect fidelity using an orchestrated multi-agent workflow.
Overview
This skill provides a complete system for cloning websites:
- Slash command:
/clone-website <url>orchestrates the entire workflow - 4 specialized sub-agents: Each handles a specific phase
- Output: Single React component using Tailwind CSS + motion
Architecture
┌─────────────────────────────────────────┐
│ ORCHESTRATOR (/clone-website) │
│ Delegates, doesn't code │
└─────────────────────────────────────────┘
│
┌───────────────┼───────────────┐
▼ ▼ ▼
┌─────────┐ ┌─────────┐ ┌─────────┐
│ screen- │ │ extrac- │ │ (can │
│ shotter │ │ tor │ │ parallel│
└─────────┘ └─────────┘ └─────────┘
│
▼
┌─────────────┐
│ cloner │◄────────┐
└─────────────┘ │
│ │
▼ │
┌─────────────┐ │
│ qa-reviewer │─────────┘
└─────────────┘ (loop until done)Quick Setup
1. Create Sub-Agents
Run /agents in Claude Code and create these 4 agents. For each, select "Generate with Claude" and provide the description.
| Agent Name | Description Summary |
|---|---|
website-screenshotter | Captures comprehensive screenshots (full-page, sections, components, hover states) |
website-extractor | Downloads assets to public/, extracts colors, typography, spacing, animations |
website-cloner | Implements React component with Tailwind + motion, auto-detects project type |
website-qa-reviewer | Pixel-by-pixel comparison, classifies issues as Critical/Major/Minor |
Detailed prompts for each agent: See references/subagents.md
2. Install Slash Command
Copy assets/clone-website.md to your commands folder:
# Project-level (shared via git)
cp assets/clone-website.md .claude/commands/
# Or user-level (personal)
cp assets/clone-website.md ~/.claude/commands/3. Configure Playwright MCP
Ensure Playwright MCP is configured in ~/.claude.json or .claude/settings.json:
{
"mcpServers": {
"playwright": {
"command": "npx",
"args": ["-y", "@anthropic-ai/mcp-playwright"]
}
}
}Usage
/clone-website https://example.comThe orchestrator will:
- Create task folder
.tasks/clone-{domain}/ - Invoke screenshotter → captures all visual references
- Invoke extractor → downloads assets, extracts styles
- Invoke cloner → implements React component
- Invoke qa-reviewer → finds discrepancies
- Loop steps 4-5 until PERFECT or max 5 iterations
Output Structure
your-project/
├── public/
│ ├── images/ # Downloaded images
│ ├── videos/ # Downloaded videos
│ └── icons/ # Downloaded SVGs/icons
├── app/clone/page.tsx # React component (location varies by framework)
└── .tasks/clone-{domain}/
├── context.md # Extracted styles
├── screenshots/ # Visual references
└── review-notes.md # QA findingsTech Stack Decisions
| Technology | Reason |
|---|---|
| Tailwind CSS | Arbitrary values (bg-[#hex]) enable pixel-perfect color matching |
| motion | Modern, lighter alternative to framer-motion (import from "motion/react") |
| Single component | Focus on cloning, not architecture; sections divided by comments |
| Auto-detect framework | Supports Next.js, TanStack Start, Vite, etc. |
Detailed rationale: See references/tech-decisions.md
Workflow Details
Phase-by-phase breakdown: See references/workflow.md
Customization
Change output location
Edit the cloner agent's system prompt to specify a different output path.
Add frameworks
Update project detection logic in cloner agent for additional frameworks.
Adjust iteration limit
Modify the slash command's Phase 5 to change max iterations (default: 5).
Troubleshooting
| Issue | Solution |
|---|---|
| Sub-agents not found | Verify names exactly match: website-screenshotter, website-extractor, website-cloner, website-qa-reviewer |
| Playwright errors | Run npm install -g @anthropic-ai/mcp-playwright |
| Assets not loading | Check public/ folder structure and image paths in component |
| Infinite loop | QA reviewer should set status; check review-notes.md for STATUS line |

