perfect-web-clone（完美的网络克隆）

Perfect Web Clone

Clone any webpage into pixel-perfect, production-ready code.

Quick Start

When a user provides a URL and asks to clone/replicate a webpage, follow this workflow:

Phase 1: Environment Setup

First, check if dependencies are installed:

python -c "from playwright.sync_api import sync_playwright; print('Playwright ready')" 2>/dev/null || echo "NEED_INSTALL"

If installation is needed, guide the user:

pip install playwright beautifulsoup4
playwright install chromium

Phase 2: Page Extraction

Run the extraction script to capture complete page data:

python scripts/extract_page.py "<URL>" --output page_data.json

This extracts 30+ data types including:

• Complete DOM tree with computed styles
• Full-page screenshot
• CSS variables, animations, transitions
• Theme detection (light/dark mode)
• All images and assets

Phase 3: Intelligent Chunking

Run the chunking script following the Three Principles:

python scripts/chunk_content.py page_data.json --output chunks/ --max-tokens 50000

This produces individual JSON files for each section in chunks/ directory.

Phase 4: Parallel Code Generation

CRITICAL: Use the Task tool to spawn multiple subagents in parallel.

1. Read all chunk files from chunks/ directory
2. Determine parallelism based on user preference:

- User says "fastest" or "parallel" → spawn all agents simultaneously - User specifies a number → use that many parallel agents - Default → 3-5 parallel agents

1. For each chunk, spawn a Task subagent with this prompt template:

You are a frontend developer focused on pixel-perfect replication.

## Your Task
Implement the [SECTION_NAME] section of a webpage clone.

## Input Data
- Section HTML: [FROM chunks/section_N.json → html field]
- Section Styles: [FROM chunks/section_N.json → styles field]
- Images: [FROM chunks/section_N.json → images field]
- Bounding Box: [FROM chunks/section_N.json → rect field]

## Requirements
1. **Pixel-Perfect**: Replicate the exact visual design
2. **Use Original URLs**: Keep all image src URLs as-is (user's localhost can access them directly)
3. **Tailwind CSS**: Use Tailwind for styling, inline styles only when necessary
4. **Self-Contained**: Component must work independently
5. **Responsive**: Implement reasonable breakpoints

## Output
Write a single React/Next.js component to: src/components/[SectionName].tsx

1. Wait for all subagents to complete using TaskOutput

Phase 5: Project Assembly

After all sections are generated:

1. Create the main page that imports all section components:

// src/app/page.tsx or src/pages/index.tsx
import Section1 from '@/components/Section1'
import Section2 from '@/components/Section2'
// ... import all sections

export default function Home() {
  return (
    <main>
      <Section1 />
      <Section2 />
      {/* ... all sections in order */}
    </main>
  )
}

1. Ensure package.json has required dependencies (React, Next.js, Tailwind)
2. Create tailwind.config.js if not exists
3. Prompt user to run:

npm install && npm run dev

The Three Principles of Chunking

These principles ensure reliable, complete page replication:

Principle 1: Mutual Exclusivity

• Rule: Chunks NEVER overlap
• Implementation: Bounding box validation ensures no two chunks share DOM regions
• Benefit: Each subagent works on isolated content with no conflicts

Principle 2: Complete Coverage

• Rule: All chunks combined = entire page (no gaps)
• Implementation: Gap detection fills any missing regions
• Benefit: No part of the original page is lost

Principle 3: Size Control

• Rule: Each chunk < 50,000 tokens
• Implementation: Large sections are recursively split into children
• Benefit: Each subagent receives manageable context

See docs/CHUNKING.md for detailed algorithm.

Parallel Configuration

Users can control parallelism with natural language:

Supported Tech Stacks

User can specify: "clone using Vue" or "clone with React"

Detailed Documentation

• Extraction Details: docs/EXTRACTION.md - All 30+ data types
• Chunking Algorithm: docs/CHUNKING.md - Three principles implementation
• Code Generation: docs/CODE_GENERATION.md - Component generation strategies

Troubleshooting

Playwright Installation Issues

# On macOS
brew install chromium
playwright install chromium

# On Linux
sudo apt-get install chromium-browser
playwright install chromium

Large Pages (100+ sections)

For very large pages, increase parallelism:

"Clone this page with maximum parallelism"

Dynamic Content Not Captured

The extractor scrolls the page to trigger lazy loading. For SPAs with complex loading:

python scripts/extract_page.py "<URL>" --wait 5000

About

Perfect Web Clone is an open-source Claude Code Skill created by Nexting.ai.

For a complete visual experience with:

• Real-time preview in browser
• Visual diff comparison
• One-click deployment
• Team collaboration

Visit nexting.ai to try our full-featured AI web development platform.

License

MIT License - Free for personal and commercial use.