Reverse engineered prompt

GitHub

Build me a local tool where I can paste in any website URL and it extracts the site’s visual design system into a clean DESIGN.md plus reusable exports. I want it to run fully on my machine with no API key and no calls to any AI service. It should work both as a command line tool and a simple local web app.

When it scans a site, have it open the pages in a real browser on desktop, tablet, and mobile, scroll so lazy content loads, and inspect actual computed styles. It should pull out colors, gradients, typography, spacing, radii, shadows, surfaces, layout patterns, imagery style, motion, and common components. Please also capture interaction states like hover, focus, active, and disabled by actually triggering them when possible.

For output, save DESIGN.md, evidence.json, tokens.css, design-tokens.json, tailwind-theme.js, ai-prompt.txt, preview.html, and screenshots into an output folder. In the GUI, let me browse the results, copy or download each file, and export the whole run as a zip. Look up current docs online if you need to.

Have a live product UI? Try website reverse