webclone
/webclone
Rebuild one or more live web pages as working Next.js 16 + Tailwind v4 + shadcn/ui code that matches their real content, fonts, images, colors, responsive layout, motion, and interactions. Use for website cloning, replication, or reverse engineering, including Framer, GSAP/Lenis, and WebGL sites. Creates the project automatically and gates delivery on measured fidelity.
- Author
- x4zu ↗
- Repository
- WebCloner ↗
- Licence
- MIT
- Last repository change
- 12.11.2026
Assessment and security
- Files in this package
- 56
- Scripts
- 24
- HTML comments
- 0
- URLs in SKILL.md
- 3
No signals detected. Repository changed on 12.11.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 24 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/asset-harvest.mjs, scripts/audit-clone.mjs, scripts/brand-kit.mjs, scripts/compare-recon.mjs, scripts/dna-scaffold.mjs, scripts/fidelity-gate.mjs, scripts/interaction-probe.mjs, scripts/lib/playwright-loader.mjs
What it does and when to use it
Rebuild one or more live web pages as working Next.js 16 + Tailwind v4 + shadcn/ui code that matches their real content, fonts, images, colors, responsive layout, motion, and interactions. Use for website cloning, replication, or reverse engineering, including Framer, GSAP/Lenis, and WebGL sites. Creates the project automatically and gates delivery on measured fidelity.
Arguments
/webclone <url1> [<url2> ...]
Original author description
Rebuild one or more live web pages as working Next.js 16 + Tailwind v4 + shadcn/ui code that matches their real content, fonts, images, colors, responsive layout, motion, and interactions. Use for website cloning, replication, or reverse engineering, including Framer, GSAP/Lenis, and WebGL sites. Creates the project automatically and gates delivery on measured fidelity.
How it works
- Iron rule: real source beats inference
- 0. Set up
- 1. Look for real source first
- 2. Recon (measure, do not eyeball)
- 3. Assess and choose the path
- 4. Map the work
- 5. Build
- 6. Compare, repair, gate
- Deliver
SKILL.md preview
---
name: webclone
description: Rebuild one or more live web pages as working Next.js 16 + Tailwind v4 + shadcn/ui code that matches their real content, fonts, images, colors, responsive layout, motion, and interactions. Use for website cloning, replication, or reverse engineering, including Framer, GSAP/Lenis, and WebGL sites. Creates the project automatically and gates delivery on measured fidelity.
argument-hint: "<url1> [<url2> ...]"
---
# /webclone
Recreate the supplied pages as an editable Next.js app. Match the source before adding any new design direction. Deliver working code: screenshots, iframes, and copied production bundles are evidence, not the implementation (except the static-mirror path in step 3).
Target URLs: `$ARGUMENTS`
`<scripts>` below means `${CLAUDE_SKILL_DIR}/scripts`. Every script prints usage with `--help`; run that instead of reading its source. References live in `${CLAUDE_SKILL_DIR}/references/`.
Related commands: `/webclone-recon` (inspect only), `/webclone-full` (whole site, 90%+ gate), `/webclone-audit` (score an existing clone), `/webclone-brand` (extract a brand kit), `/webclone-remix` (swap in the user's brand).
## Iron rule: real source beats inference
Every second-hand analysis (blog posts, AI teardowns, your own guesses) is a concept sketch; its code is unverified until checked against real source or measured runtime values. Fonts, colors, timings, and assets come from measurement, never from memory. Record anything estimated as an estimate.
## 0. Set up
1. **URLs.** Take every URL from the arguments or the message; bare domains get `https://`. No URL: ask and stop. Different origins become separate apps unless the user wants one combined app (`--combined`).
2. **Responsible use.** Clone sites the user owns or may reproduce, or for learning/migration. Refuse phishing, impersonation, credential capture, or passing off another brand as the user's. Faithful clones keep the source's logos and copy locally; say that these belong to their owners and must be replaced before any public deployment.
3. **Project.** Needs Node.js 20.9+.
```bash
node "${CLAUDE_SKILL_DIR}/scripts/scaffold.mjs" <url1> [<url2> ...] --install
```
Prints JSON with each `projectDir`. Reuses the current folder when it holds `.webclone.json`; otherwise creates `./<site>-clone/`. Exit 3: target folder is not empty and not a webclone project; ask before `--force` or `--dir`. Exit 4: `npm` install failed; fix it first. Work inside `projectDir` from here; read its `AGENTS.md`.
4. **Browser.** The scripts drive the system Chrome/Edge/Chromium over CDP with zero installs (they also use Playwright when the project has it). If a script reports no compatible browser, ask the user to install Chrome, or set `WEBCLONE_BROWSER_PATH`, or run `npm i -D playwright && npx playwright install chromium` in the project. A browser MCP (Playwright MCP, Chrome DevTools MCP, Claude in Chrome) is a good complement for hover/click/menu states.
Evidence for each site goes to `docs/research/<site>/` (call it `R`); recon output to `R/recon`.
## 1. Look for real source first
Before scraping, spend a minute searching for the actual code: the product name or domain on GitHub (`gh search repos <name>` or web search), `*.vercel.app` / `*.netlify.app` / `*.github.io` slugs (often the repo or user name), and single-file sites via view-source. If found, check its license:
| License | Allowed |
|---|---|
| MIT / Apache / BSD / Unlicense | Modify and deploy; keep attribution |
| None | All rights reserved: local study only, attribute, no public redeploy without permission |
| Proprietary / explicit ban | Read to learn only |
Public on GitHub does not mean MIT. Record source and license in `R/NOTES.md`.
## 2. Recon (measure, do not eyeball)
Run from `projectDir` for each source URL:
```bash
node <scripts>/recon-site.mjs --url <url> --out R/recon --label original # screenshots 1440/768/390, palette, fonts, frameworks, motion signals
node <scripts>/asset-harvest.mjs --url <url> --recon R/recon/original-recon.json --out public/sites/<site> --manifest R/recon/asset-manifest.json
node <scripts>/motion-probe.mjs --url <url> --out R/recon/motion --label original # real easing/duration, GSAP/ScrollTrigger, Lenis
node <scripts>/interaction-probe.mjs --url <url> --out R/recon/interactions --label original
```
Add when relevant:
- Several pages requested or a multi-page site: `route-crawl.mjs --url <url> --out R/recon/routes --label original --max-pages 25 --max-depth 2`.
- SPA/SaaS content from APIs: `network-capture.mjs --url <url> --out R/recon/network --label original` (JSON fixtures).
- WebGL/Canvas or heavy bundles: `sourcemap-hunt.mjs --recon R/recon/original-recon.json --out R/recon/sourcemaps`.
Read `R/recon/original-summary.md` and **look at the screenshots** with the Read tool. A large DOM dump never replaces looking at the page. Then exercise menus, tabs, accordions, carousels, and hover states with a browser tool, and save labeled screenshots of each state.
Read [inspection-guide.md](references/inspection-guide.md) for what to measure. For Framer pages, sticky scenes, or animated media read [framer-and-motion.md](references/framer-and-motion.md); for any motion read [motion.md](references/motion.md).
## 3. Assess and choose the path
Write the pre-clone assessment to `R/NOTES.md` ([assessment.md](references/assessment.md)): complexity L1-L6, mode, what will be exact, what will be approximated, what is out of scope. The default mode for "clone this" is **faithful**.
| Recon result | Path |
|---|---|
| Static/marketing/content site (L1-L3) | Rebuild in Next.js (this skill's default) |
| Multi-page site | Route map first, one template per page type, data-driven pages ([complex-playbooks.md](references/complex-playbooks.md)) |
| SPA/SaaS (L6) | Presentation layer + local JSON fixtures; mock writes |
| Interaction/scroll-heavy (L4) | Rebuild + measured motion ([motion.md](references/motion.md)) |
| WebGL/Canvas/Three.js (L5) | Real source or teardown first ([reverse-engineering.md](references/reverse-engineering.md), [effect-extraction.md](references/effect-extraction.md)), then a client component |
| Custom static-built site (Astro/Vite SSG/Hugo) where 1:1 matters more than editable code | Offer the full asset mirror ([static-mirror.md](references/static-mirror.md)) |
Do not downgrade a faithful clone to placeholder branding because the site is commercial, and do not skip scripted recon because a site looks bot-protected. Run the scripts; degrade only on failure and record it.
## 4. Map the work
Record a source-to-local route map and file ownership in `R/NOTES.md`. Use `src/app/`, `src/components/sites/<site>/`, `public/sites/<site>/`.
- The first root clone replaces the untouched template homepage (`src/app/page.tsx`).
- Preserve source pathnames for several pages of one origin; query/fragment variants are usually states of one route.
- Preserve existing routes/assets; resolve collisions before replacing anything. When adding pages, extend layout and global styles without breaking existing pages.
- A combined multi-origin app needs route prefixes and scoped fonts/styles.
- Escape literal URL segments that clash with framework syntax (`_`, `@`, brackets, parentheses) and verify the real URL.
## 5. Build
Shared foundations first, then sections.
**Fidelity gates (hard requirements, checked by `audit-clone.mjs --strict`):**
1. **Real fonts, self-hosted.** Use the files `asset-harvest` saved. Either link `public/sites/<site>/fonts/fonts.css` from the root layout (keeps the source's exact family names), or copy the files next to a `next/font/local` declaration. Copy `font-family` values from recon `palette.*.fontFamily` / `fontFaces[].family`. A system-font stand-in for a custom font fails the clone. If a font cannot be downloaded or licensed, say so and name the closest stand-in as a gap.
2. **Real images.** Reference files through `asset-manifest.json` (originalUrl -> localPath). No gradients or placeholders in place of photography. Respect `currentSrc`, crop (`object-fit`/`object-position`), and the real format (a `.png` URL may serve AVIF). `next/image` with explicit sizes, or `<img>` when exact rendering matters more.
3. **Exact colors.** Copy computed values from recon `palette` and `rootVariables` into CSS custom properties / Tailwind `@theme`. `rgb(17,17,17)` becomes `#111111`, never "close enough" `#0a0a0a`.
4. **Same scroll feel.** Reproduce smooth-scroll libraries, scroll-snap, sticky elements, and scroll-triggered motion with the values from `motion-probe` (Lenis options, ScrollTrigger start/end/scrub/pin, easing strings, durations).
Then:
- Build each shared component with every observed state (hover, focus, open, breakpoints) so it is reused as-is.
- Use the source's real copy and outbound links. Rewrite links to local routes only when those routes are cloned; disclose out-of-scope links.
- Visible controls must do what they promise (menus open, tabs switch, carousels move, forms validate locally).
- Remove tracking (GA/gtag, pixels, heatmaps) and note it.
- Delegate independent sections or pages to subagents only after shared files have an owner; give each builder screenshots, measurements, asset paths, destination files, and a finish line.
- Run `npm run typecheck` after each integrated slice.
## 6. Compare, repair, gate
Start `npm run dev` in the background (port 3000 unless taken). For each route:
```bash
node <scripts>/recon-site.mjs --url http://localhost:3000/<route> --out R/recon --label clone
node <scripts>/visual-diff.mjs --original R/recon/screenshots/original-1440.png --clone R/recon/screenshots/clone-1440.png \
--out R/recon/visual-diff-1440.json --diff R/recon/screenshots/visual-diff-1440.png # repeat for 768 and 390
node <scripts>/motion-probe.mjs --url http://localhost:3000/<route> --out R/recon/motion --label clone # when the page has motion
node <scripts>/compare-recon.mjs --original R/recon/original-recon.json --clone R/recon/clone-recon.json \
--visual-diff R/recon/visual-diff-1440.json --out CLONE_REPORT.md
node <scripts>/audit-clone.mjs --project . --recon R/recon/original-recon.json --strict --out CLONE_AUDIT.md
```
Open the diff images. Fix the largest differences first: missing sections/layers, geometry, typography and wrapping, asset crops, then spacing and motion. Re-run the affected checks. Compare animated regions in initial, active, and settled states, including reverse scroll. Exercise navigation, the mobile menu, tabs, carousels, and media with real input; check console errors and horizontal overflow.
**Done means all of:** `audit-clone --strict` exits 0 (or each remaining finding is explained as a deliberate, user-visible gap), `npm run check` passes, and the visual-diff results are reported honestly per width. Never call an unchecked result pixel-perfect.
## Deliver
- Project folder and `cd <projectDir> && npm run dev`.
- Route map (source -> local).
- Per-width fidelity (`fidelityPct`, score) and the diff image paths; `CLONE_REPORT.md`, `CLONE_AUDIT.md`.
- Checks actually run, with results.
- Remaining gaps, approximations, out-of-scope links, and items to replace before deploying (`R/NOTES.md`).
Artifact layout and templates: [deliverables.md](references/deliverables.md).
Files in this package (56)
- SKILL.md 11 KB
- references/assessment.md 4 KB
- references/complex-playbooks.md 3 KB
- references/deliverables.md 4 KB
- references/design-dna.md 11 KB
- references/effect-extraction.md 4 KB
- references/framer-and-motion.md 3 KB
- references/inspection-guide.md 3 KB
- references/motion.md 6 KB
- references/reverse-engineering.md 4 KB
- references/static-mirror.md 4 KB
- scripts/asset-harvest.mjs 16 KB
- scripts/audit-clone.mjs 12 KB
- scripts/brand-kit.mjs 11 KB
- scripts/compare-recon.mjs 11 KB
- scripts/dna-scaffold.mjs 12 KB
- scripts/fidelity-gate.mjs 7 KB
- scripts/interaction-probe.mjs 12 KB
- scripts/mirror-site.mjs 6 KB
- scripts/motion-probe.mjs 13 KB
- scripts/network-capture.mjs 4 KB
- scripts/recon-site.mjs 16 KB
- scripts/route-crawl.mjs 8 KB
- scripts/scaffold.mjs 8 KB
- scripts/sourcemap-hunt.mjs 4 KB
- scripts/visual-diff.mjs 7 KB
- scripts/lib/playwright-loader.mjs 4 KB
- scripts/lib/system-browser.mjs 19 KB
- template/.dockerignore 1 KB
- template/.gitignore 1 KB
- template/.nvmrc 1 KB
- template/AGENTS.md 2 KB
- template/CLAUDE.md 1 KB
- template/Dockerfile 4 KB
- template/Dockerfile.dev 1 KB
- template/components.json 1 KB
- template/docker-compose.yml 1 KB
- template/eslint.config.mjs 1 KB
- template/next.config.ts 1 KB
- template/package-lock.json 344 KB
- template/package.json 1 KB
- template/postcss.config.mjs 1 KB
- template/tsconfig.json 1 KB
- template/docs/design-references/.gitkeep 1 KB
- template/public/images/.gitkeep 1 KB
- template/public/seo/.gitkeep 1 KB
- template/public/videos/.gitkeep 1 KB
- template/scripts/.gitkeep 1 KB
- template/src/app/favicon.ico 25 KB
- template/src/app/globals.css 4 KB
- template/src/app/layout.tsx 1 KB
- template/src/app/page.tsx 1 KB
- template/src/components/ui/button.tsx 3 KB
- template/src/hooks/.gitkeep 1 KB
- template/src/lib/utils.ts 1 KB
- template/src/types/.gitkeep 1 KB
Versions
The database has no other skills named webclone.
Database snapshot: 13.11.2026 · package version ceafbc095f21 (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 90clone-website
Rebuild existing web pages as working local code, matching their real content, assets, responsive layout, and interactions. Use for website cloning, replication, or reverse engineering, including Framer sites and animated pages. - 100webclone-full
Clone an entire website - every reachable page, not just one - into a Next.js app (or a 1:1 static mirror for static-built sites), and keep fixing until each route reaches 90-100% measured visual fidelity at desktop, tablet, and mobile. Use when the user wants the whole site… - 100ui-reverse-engineering
Clone or recreate a live website URL, page, or section as React + Tailwind with extracted DOM, CSS, assets, responsive layout, motion, and interactions. Use for live-URL implementation or fidelity repair; not capture-only or diff-only requests. - 100webclone-brand
Extract a measured brand kit and design system from a live website - logo files, semantic color roles, fonts, imagery, voice, layout posture - into brand.json, DESIGN.md, design-dna.json, a Tailwind v4 theme.css, and a brand-kit.html preview. Use when the user wants the… - 95tailwind-shadcn
Tailwind CSS utility patterns with shadcn/ui component usage, theming via CSS variables, and responsive design. Use when styling components, installing shadcn components, implementing dark mode, or creating consistent design systems. - 100od-shadcn-ui
Build UI components with shadcn/ui Use when the user asks for shadcn ui work, or mentions od, shadcn, ui.