URL → pixel-perfect website clone, verified by pixel diff. A Claude Code skill.

🇫🇷 Lire en français

Give Claude a URL and get back a reproduction that is visually indistinguishable from the original: same layout, fonts, colours, spacing, responsive behaviour, hover states and animations, running locally with a clean console.

The difference between an average clone and a perfect one is measurement. copycat never guesses a font, a colour or a margin: it reads them from the browser. And it never says "looks good" by eye: it diffs the pixels and reads the console.

1. Capture (Playwright): full-page screenshots, fold-by-fold screenshots on desktop / tablet / mobile, one crop per section, hover states, optional videos.
2. Extract everything the browser exposes: @font-face + font files, CSS variables, palette, typography scale, containers, breakpoints, sticky elements, keyframes and transitions, images at their rendered size, inline SVGs, meta tags, console and network errors, plus the site's real CSS files and text content.
3. Rebuild section by section, top to bottom, from the measured values (static HTML/CSS/JS by default, or Next.js if the repo already uses it).
4. Verify in a loop: pixel diff per viewport, worst zones mapped to sections, design diff (missing fonts, colours, headings, CTAs), and the clone's console. Grade A–D, target < 2 % mismatch with 0 console errors.

It handles lazy-loading and reveal-on-scroll, cookie banners (hidden via CSS, never accepted), pages too tall for native capture (fold stitching), cross-origin stylesheets, CSS-relative fonts, image proxies (/_next/image?url=) and inline SVG logos.

Requires Node ≥ 18.

# personal install (all projects)
git clone https://github.com/minosdevs/copycat-skill ~/.claude/skills/copycat

# or project install
git clone https://github.com/minosdevs/copycat-skill .claude/skills/copycat

Then install the scripts' dependencies once:

cd ~/.claude/skills/copycat/scripts && npm install && npx playwright install chromium

In Claude Code, just ask:

clone https://example.com

make me exactly the same landing page as https://example.com

The skill triggers on "copy", "clone", "replicate", "make the same as X"… You can also run the scripts yourself:

# 1. capture the original → copycat/example.com/REPORT.md
node scripts/capture.mjs https://example.com --out copycat/example.com

# 2. compare your clone against it → copycat/example.com/compare/REPORT.md
node scripts/compare.mjs --original copycat/example.com --clone http://localhost:3000

Useful capture options: --depth 1 (pages in the nav, --max-pages 8), --viewports desktop,mobile or --viewports large:1920x1080,desktop, --scale 2, --dark, --locale en-US, --wait 3000, --videos, --channel chrome|msedge (real installed browser, for bot-protected sites), --headed, --no-hover.

scripts/extract.browser.js can also be pasted as-is into the DevTools console: copy(JSON.stringify(__copycatExtract(), null, 2)) puts the page's design tokens in your clipboard.

The skill's instructions (SKILL.md and references/) are written in French. Claude reads them fine and answers in your language.

Cloning a site to learn, prototype or rebuild your own site is common practice. Reusing a third party's logo, photos, copy or brand in production is not. copycat never enters credentials, never bypasses logins or paywalls, and tells you to swap brand assets for placeholders when the clone isn't for your own use.

MIT © minosdevs