Reverse-engineer websites by unleashing AI agents in parallel

What it does

This is a Next.js 16 boilerplate wrapped around a detailed prompt chain. You feed it a target URL through an AI coding agent, and the template orchestrates the agent to screenshot the site, extract design tokens and assets, write per-component specs into docs/research/components/, and dispatch parallel builder agents in isolated git worktrees. Once the sections are built, it merges them and runs a visual diff against the original. It is essentially scaffolding plus a very specific instruction set for cloning UIs.

The interesting bit

Instead of asking one AI to memorize an entire website, the project treats reconstruction like an assembly line. A reconnaissance phase captures exact getComputedStyle() values, interaction states, and responsive breakpoints, then farms out parallel builds to separate agents—each working from a frozen spec in its own worktree. The boring orchestration layer is the whole product here.

Key highlights

• Supports a wide roster of agents: Claude Code (recommended with Opus 4.7), Cursor, Codex CLI, GitHub Copilot, Windsurf, Gemini CLI, Cline, and others.
• The pipeline covers reconnaissance, foundation setup, component spec writing, parallel worktree builds, and visual QA.
• Built on Next.js 16, React 19, TypeScript strict, Tailwind CSS v4 with oklch tokens, and shadcn/ui primitives.
• Two source-of-truth files—AGENTS.md and .claude/skills/clone-website/SKILL.md—sync platform-specific instructions automatically.

Caveats

• Not a standalone tool: it requires a separate AI coding agent and Node.js 24+; the template does nothing on its own.
• The README explicitly prohibits deceptive use and warns that some sites block scraping, so this is strictly for sites you own or have permission to clone.
• The onboarding guidance is inconsistent: one version of the README tells users to create a repository from the template, while another version says to clone the repository directly.

Verdict

Best for developers migrating a site they own from WordPress or Webflow into a modern React stack, or recovering lost source code. Skip it if you want a scraper that works without an AI agent in the loop.