boyang-hu/website-rebuild-skill

What it solves

This project provides a system for AI agents to perform high-fidelity, line-by-line reconstruction of existing websites. Unlike simple website scrapers or "cloners" that only capture the visual appearance, this tool focuses on reverse-engineering the underlying logic and implementation to create a functional, human-readable source code project that is provably identical to the original site.

How it works

The process follows a strict pipeline designed for AI agents:

1. Classification: The agent first determines if the site is reconstructible (e.g., identifying if it is a static site, a WebGL scene, or a server-side rendered app).
2. Evidence Mirroring: A read-only snapshot of the entire site is captured to serve as the absolute ground truth for all future comparisons.
3. Reverse Engineering: Compressed and obfuscated code is expanded, and a mapping is created so every line in the rebuild can be traced back to the original bundle.
4. Porting: Logic is ported into a new project, with every change documented in a difference table.
5. Verification: The agent performs quantitative comparisons across five layers: console logs, network requests, DOM structure, geometry, and pixel-by-pixel visual matching.
6. Source-ification: The final output is refactored into a clean, modular, human-readable engineering project with proper naming and comments, while maintaining byte-level equivalence to the original.

Who it’s for

• Developers and Researchers: Those looking to study how complex web animations, WebGL engines, or advanced frontend architectures are implemented.
• AI Agent Users: Users of agents like Claude Code or Codex who want to automate the reconstruction of web assets for study or migration.
• Digital Archivists: Those attempting to rescue "dead" websites from archives like the Wayback Machine.

Highlights

• Pixel-Perfect Accuracy: Uses quantitative verification to ensure the rebuild is identical to the original, often achieving a mean absolute difference of 0.00.
• RSC Support: Capable of reconstructing Next.js App Router sites by analyzing the "flight" stream output.
• Zero-Dependency Toolchain: The core pipeline consists of 70 Node scripts that require no npm packages until the final source-ification stage.
• Dead Site Recovery: Can rescue sites from the Wayback Machine by reconstructing them from captured bytes.
• Traceable Logic: Every line of the reconstructed code can be traced back to the specific line in the original site's bundle.