This is the nuclear option for website cloning. It spins up parallel builder agents in worktrees, extracts computed CSS values with getComputedStyle, downloads every asset layer by layer, and documents both appearance and behavior for each section before handing off specs to specialist builders. The guiding principle is completeness over speed: if a builder has to guess a font size or padding value, the extraction failed. It catches the tricky stuff like scroll-driven tab switching versus click handlers, parallax layers, and Lenis smooth scroll implementations. Requires browser MCP tools and produces auditable artifacts in a structured docs folder. Best for pixel-perfect clones where "close enough" isn't acceptable.
npx skills add https://github.com/julianromli/ai-skills --skill clone-website