| name | website-cloning |
| description | Reverse-engineer and clone any website as a pixel-perfect React + Vite app. Use when the user asks to clone, replicate, copy, rebuild, reverse-engineer, or pixel-perfect match any website. Also triggers on "make a copy of this site", "rebuild this page", "clone this URL". Provide the target URL. Uses Playwright for extraction and design subagents for parallel building. |
Clone Website
Reverse-engineer and rebuild a target website as a pixel-perfect React + Vite clone using Playwright for extraction and design subagents for parallel construction.
Legitimate Use Policy
Before cloning, confirm the user's intent is legitimate. Ask:
- "Is this your own website or your client's website?"
- "What is this clone for?"
Acceptable: rebuilding your own site, design reference/learning, staging copy, platform migration.
Refuse if: impersonation, phishing, traffic theft, trademark infringement, or deception.
For non-owned sites (design inspiration), remind the user to replace logos, brand names, trademarks, product data, and contact info with their own.
Prerequisites
Before starting, set up the tools:
pip install playwright
CHROMIUM_PATH=$(find /nix/store -maxdepth 4 -name "chromium" -type f 2>/dev/null | head -1)
echo "Chromium at: $CHROMIUM_PATH"
Critical Playwright settings (learned from production use):
- Always use
--no-sandbox args: browser = p.chromium.launch(headless=True, executable_path=CHROMIUM_PATH, args=["--no-sandbox", "--disable-setuid-sandbox"])
- Prefer
wait_until="domcontentloaded" over "networkidle" — many modern sites never reach networkidle due to analytics/websockets
- Add a generous
page.wait_for_timeout(5000) after navigation to let lazy content and JS frameworks hydrate
- Set
timeout=60000 on all page.goto() calls
Guiding Principles
- Completeness beats speed — Every builder must receive everything it needs. If a builder has to guess a color, font size, or padding value, extraction failed.
- Small tasks, perfect results — Break complex sections into sub-components. One agent per card variant, not one agent per "entire features section."
- Real content, real assets — Extract actual text, images, videos from the live site. No placeholders. Download all assets locally — CDN URLs expire or get blocked.
- Foundation first — Global CSS tokens, fonts, and assets must exist before any component building starts.
- Extract appearance AND behavior — Static CSS plus interactions (hover, scroll-triggered, click-driven, animations).
- Spec files are the source of truth — Every component gets a spec file in
docs/research/components/ BEFORE any builder is dispatched.
- Mobile navigation is mandatory — Always extract and implement the mobile hamburger menu / drawer. This is the #1 missed feature in website clones.
- Download images locally — Never rely on external CDN URLs in the final build. Download everything to
public/images/ and reference with local paths.
Phase 1: Reconnaissance
Create a Python script to do the initial extraction.
1.1 Setup & Screenshot
from playwright.sync_api import sync_playwright
import json, os
CHROMIUM_PATH = "FILL_IN"
TARGET_URL = "FILL_IN"
OUT_DIR = "clone-data"
os.makedirs(f"{OUT_DIR}/screenshots", exist_ok=True)
os.makedirs(f"{OUT_DIR}/components", exist_ok=True)
with sync_playwright() as p:
browser = p.chromium.launch(
headless=True,
executable_path=CHROMIUM_PATH,
args=["--no-sandbox", "--disable-setuid-sandbox"]
)
page = browser.new_page(viewport={"width": 1440, "height": 900})
page.goto(TARGET_URL, wait_until="domcontentloaded", timeout=60000)
page.wait_for_timeout(5000)
for _ in range(15):
page.evaluate("window.scrollBy(0, 600)")
page.wait_for_timeout(600)
page.evaluate("window.scrollTo(0, 0)")
page.wait_for_timeout(1500)
page.screenshot(path=f"{OUT_DIR}/screenshots/desktop-full.png", full_page=True)
page.set_viewport_size({"width": 768, "height": 1024})
page.wait_for_timeout(1000)
page.screenshot(path=f"{OUT_DIR}/screenshots/tablet-full.png", full_page=True)
page.set_viewport_size({"width": 390, "height": 844})
page.wait_for_timeout(1000)
page.screenshot(path=f"{OUT_DIR}/screenshots/mobile-full.png", full_page=True)
page.set_viewport_size({"width": 1440, "height": 900})
page.wait_for_timeout(500)
1.2 Extract Design Tokens
Run the design token extraction JS via page.evaluate(). Key extractions:
- All CSS custom properties from
:root
- Body background/text colors and font families
- All Google Fonts / self-hosted font URLs
- Heading font families (often different from body)
- Favicon and meta image URLs
def extract_tokens(page, out_dir):
tokens = page.evaluate("""
() => {
const body = document.body;
const cs = getComputedStyle(body);
const cssVars = [];
try {
for (const sheet of document.styleSheets) {
try {
for (const rule of sheet.cssRules) {
if (rule.selectorText === ':root' || rule.selectorText === ':root, :host') {
for (const prop of rule.style) {
if (prop.startsWith('--')) {
cssVars.push([prop, rule.style.getPropertyValue(prop).trim()]);
}
}
}
}
} catch(e) {}
}
} catch(e) {}
const h1 = document.querySelector('h1');
const h2 = document.querySelector('h2');
const h3 = document.querySelector('h3');
const btn = document.querySelector('button, [class*="btn"], a[class*="button"]');
const nav = document.querySelector('nav, header');
const card = document.querySelector('[class*="card"], [class*="Card"]');
function getStyles(el) {
if (!el) return null;
const s = getComputedStyle(el);
return {
fontSize: s.fontSize, fontWeight: s.fontWeight, fontFamily: s.fontFamily,
lineHeight: s.lineHeight, letterSpacing: s.letterSpacing, color: s.color,
textTransform: s.textTransform, backgroundColor: s.backgroundColor,
padding: s.padding, borderRadius: s.borderRadius, border: s.border
};
}
return {
body: {
bgColor: cs.backgroundColor,
textColor: cs.color,
fontFamily: cs.fontFamily,
fontSize: cs.fontSize,
lineHeight: cs.lineHeight
},
h1: getStyles(h1),
h2: getStyles(h2),
h3: getStyles(h3),
button: getStyles(btn),
nav: getStyles(nav),
card: getStyles(card),
cssVars: cssVars,
fonts: [...document.querySelectorAll('link[href*="fonts.googleapis"], link[href*="fonts.gstatic"]')]
.map(l => l.href),
favicons: [...document.querySelectorAll('link[rel*="icon"], link[rel="apple-touch-icon"]')]
.map(l => ({ href: l.href, rel: l.rel, sizes: l.sizes?.toString() || '' })),
metaImages: [...document.querySelectorAll('meta[property="og:image"], meta[name="twitter:image"]')]
.map(m => ({ property: m.getAttribute('property') || m.name, content: m.content })),
title: document.title,
metaDescription: document.querySelector('meta[name="description"]')?.content || ''
};
}
""")
with open(f"{out_dir}/tokens.json", "w") as f:
json.dump(tokens, f, indent=2)
return tokens
Font Mapping Strategy: Many sites use proprietary/licensed fonts. Map them to close Google Fonts equivalents:
- Proprietary sans-serif (e.g., "Geograph") → DM Sans, Inter, or Source Sans Pro
- Proprietary serif (e.g., "Self Modern") → DM Serif Text, Playfair Display, or Lora
- Load mapped fonts via
<link> tags in index.html, not @import in CSS (faster loading)
1.3 Extract Page Topology
Map every section top-to-bottom with tag, classes, dimensions, background, text preview, images, and links.
def extract_content(page, out_dir):
data = page.evaluate("""
() => {
const result = {};
const banner = document.querySelector(
'[class*="banner"], [class*="announcement"], [class*="promo-bar"], [class*="top-bar"], [class*="topbar"]'
);
if (banner && banner.offsetHeight > 0 && banner.offsetHeight < 100) {
result.banner = {
bgColor: getComputedStyle(banner).backgroundColor,
text: banner.innerText.trim().slice(0, 500),
height: banner.offsetHeight
};
}
const header = document.querySelector('header') || document.querySelector('nav, [class*="header"], [class*="navbar"]');
if (header) {
const cs = getComputedStyle(header);
result.header = {
height: header.offsetHeight,
bgColor: cs.backgroundColor,
position: cs.position,
borderBottom: cs.borderBottom,
boxShadow: cs.boxShadow,
logo: header.querySelector('img')?.src || header.querySelector('svg')?.outerHTML?.slice(0, 500) || '',
navLinks: [...header.querySelectorAll('a')].map(a => ({
text: a.innerText.trim(),
href: a.getAttribute('href') || ''
})).filter(l => l.text && l.text.length < 60).slice(0, 20)
};
}
const main = document.querySelector('main') || document.body;
const children = main === document.body
? [...main.children].filter(c => c.tagName !== 'HEADER' && c.tagName !== 'FOOTER' && c.tagName !== 'NAV' && c.tagName !== 'SCRIPT' && c.tagName !== 'STYLE')
: [...main.children];
result.sections = children.map((child, idx) => {
const rect = child.getBoundingClientRect();
if (rect.height < 20) return null;
const cs = getComputedStyle(child);
if (cs.display === 'none' || cs.visibility === 'hidden') return null;
return {
index: idx,
tag: child.tagName.toLowerCase(),
id: child.id || null,
classes: child.className?.toString().slice(0, 300) || '',
top: Math.round(rect.top + window.scrollY),
height: Math.round(rect.height),
bgColor: cs.backgroundColor,
bgImage: cs.backgroundImage !== 'none' ? cs.backgroundImage : null,
padding: cs.padding,
maxWidth: cs.maxWidth,
display: cs.display,
text: child.innerText?.slice(0, 2000) || '',
headings: [...child.querySelectorAll('h1, h2, h3, h4')].slice(0, 10).map(h => ({
level: h.tagName.toLowerCase(),
text: h.innerText.trim()
})),
images: [...child.querySelectorAll('img')].slice(0, 30).map(img => ({
src: img.src,
alt: img.alt,
w: img.offsetWidth,
h: img.offsetHeight,
position: getComputedStyle(img).position,
zIndex: getComputedStyle(img).zIndex
})).filter(i => i.src && i.w > 30),
links: [...child.querySelectorAll('a')].slice(0, 30).map(a => ({
text: a.innerText.trim(),
href: a.getAttribute('href') || ''
})).filter(l => l.text),
buttons: [...child.querySelectorAll('button, [role="button"], a[class*="btn"], a[class*="button"]')].slice(0, 10).map(b => ({
text: b.innerText.trim(),
classes: b.className?.toString().slice(0, 200) || ''
})).filter(b => b.text)
};
}).filter(Boolean);
const footer = document.querySelector('footer');
if (footer) {
result.footer = {
bgColor: getComputedStyle(footer).backgroundColor,
text: footer.innerText.trim().slice(0, 2000),
columns: [...footer.querySelectorAll('div > ul, nav > ul, [class*="col"]')].slice(0, 8).map(col => ({
heading: col.previousElementSibling?.innerText?.trim() || col.querySelector('h3, h4, h5, strong')?.innerText?.trim() || '',
links: [...col.querySelectorAll('a')].map(a => ({
text: a.innerText.trim(),
href: a.getAttribute('href') || ''
})).filter(l => l.text)
})),
socialLinks: [...footer.querySelectorAll(
'a[href*="instagram"], a[href*="tiktok"], a[href*="twitter"], a[href*="facebook"], a[href*="youtube"], a[href*="linkedin"], a[href*="github"]'
)].map(a => ({ href: a.href, platform: a.href.match(/(instagram|tiktok|twitter|facebook|youtube|linkedin|github)/)?.[1] || '' }))
};
}
return result;
}
""")
with open(f"{out_dir}/content.json", "w") as f:
json.dump(data, f, indent=2)
return data
1.4 Interaction Sweep
Use Playwright to discover behaviors:
- Scroll sweep: Scroll slowly, check if header changes (sticky/blur/shadow transitions), elements animate in, tabs auto-switch
- Hover sweep: Hover over interactive elements, capture style changes (image zoom, button color shifts, underlines)
- Responsive sweep: Test at 1440px, 768px, 390px — note layout shifts, hidden elements, hamburger menu appearance
- Mobile menu extraction: At 390px, look for hamburger/menu buttons and click them to capture the drawer/overlay content and animation
Save findings to clone-data/behaviors.md.
1.5 Asset Discovery & Download
Enumerate all images, videos, background images, and SVGs. Always download to local public/images/ and public/videos/ — never rely on CDN URLs in the final build.
def discover_assets(page):
return page.evaluate("""
() => {
return {
images: [...document.querySelectorAll('img')]
.filter(img => img.offsetWidth > 30 && img.src)
.map(img => ({
src: img.src,
srcset: img.srcset || '',
alt: img.alt,
w: img.offsetWidth,
h: img.offsetHeight,
parentClasses: img.parentElement?.className?.toString().slice(0, 100) || '',
position: getComputedStyle(img).position,
zIndex: getComputedStyle(img).zIndex
})),
videos: [...document.querySelectorAll('video')].map(v => ({
src: v.src || v.querySelector('source')?.src,
poster: v.poster,
autoplay: v.autoplay,
loop: v.loop,
muted: v.muted
})).filter(v => v.src),
backgroundImages: [...document.querySelectorAll('*')].filter(el => {
const bg = getComputedStyle(el).backgroundImage;
return bg && bg !== 'none' && bg.includes('url(');
}).slice(0, 50).map(el => ({
url: getComputedStyle(el).backgroundImage,
element: el.tagName + (el.className ? '.' + el.className.toString().split(' ')[0] : '')
})),
svgs: [...document.querySelectorAll('svg')].slice(0, 50).map((svg, i) => ({
index: i,
viewBox: svg.getAttribute('viewBox') || '',
width: svg.getAttribute('width') || svg.offsetWidth,
height: svg.getAttribute('height') || svg.offsetHeight,
html: svg.outerHTML.length < 5000 ? svg.outerHTML : '[TOO_LARGE]',
parentText: svg.parentElement?.innerText?.trim().slice(0, 50) || '',
ariaLabel: svg.getAttribute('aria-label') || ''
})),
fonts: [...new Set(
[...document.querySelectorAll('*')].slice(0, 300)
.map(el => getComputedStyle(el).fontFamily)
)],
fontLinks: [...document.querySelectorAll('link[href*="fonts"]')].map(l => l.href)
};
}
""")
import urllib.request, urllib.error, re, hashlib
def download_assets(assets, out_dir="public"):
downloaded = {}
img_dir = f"{out_dir}/images"
vid_dir = f"{out_dir}/videos"
os.makedirs(img_dir, exist_ok=True)
os.makedirs(vid_dir, exist_ok=True)
for img in assets.get("images", []):
url = img["src"]
if not url or url.startswith("data:"):
continue
ext = re.search(r'\.(png|jpg|jpeg|webp|gif|svg|avif)', url.lower())
ext = ext.group(0) if ext else ".webp"
name_hash = hashlib.md5(url.encode()).hexdigest()[:10]
alt_slug = re.sub(r'[^a-z0-9]', '-', (img.get("alt") or "img").lower())[:30]
filename = f"{alt_slug}-{name_hash}{ext}"
filepath = f"{img_dir}/{filename}"
try:
req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
with urllib.request.urlopen(req, timeout=15) as resp:
with open(filepath, "wb") as f:
f.write(resp.read())
downloaded[url] = f"/images/{filename}"
print(f"OK: {filename}")
except Exception as e:
print(f"FAIL: {url} -> {e}")
for vid in assets.get("videos", []):
url = vid["src"]
if not url:
continue
ext = re.search(r'\.(mp4|webm|mov)', url.lower())
ext = ext.group(0) if ext else ".mp4"
name_hash = hashlib.md5(url.encode()).hexdigest()[:10]
filename = f"video-{name_hash}{ext}"
filepath = f"{vid_dir}/{filename}"
try:
req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
with urllib.request.urlopen(req, timeout=30) as resp:
with open(filepath, "wb") as f:
f.write(resp.read())
downloaded[url] = f"/videos/{filename}"
print(f"OK: {filename}")
except Exception as e:
print(f"FAIL: {url} -> {e}")
return downloaded
1.6 Image URL Verification & Fixing
Always verify scraped image URLs before using them:
import subprocess
def verify_url(url):
r = subprocess.run(['curl', '-s', '-o', '/dev/null', '-w', '%{http_code}', '-L', url],
capture_output=True, text=True, timeout=10)
return r.stdout.strip()
| Problem | Fix |
|---|
| Truncated URL | Re-scrape with full img.src extraction |
| 403 Forbidden | Download locally with User-Agent header to public/images/ |
| Expired signed URL | Download and serve locally |
| Low resolution | Modify CDN width/height params (e.g., width=100 → width=800) |
| CORS / hotlink block | Download locally — this is the default strategy |
CDN URL Patterns (for upscaling resolution before downloading):
- Shopify:
cdn.shopify.com/...?width=X&height=Y — increase width param
- Sanity:
cdn.sanity.io/...?w=X&h=Y — increase w param
- Cloudinary:
res.cloudinary.com/.../w_X,h_Y/... — increase w_ param
- Contentful:
images.ctfassets.net/...?w=X — increase w param
Phase 2: Foundation Build
This phase is sequential — do it yourself, not delegated to subagents.
-
Create the artifact using createArtifact() with type react-vite
-
Update index.html with font loading <link> tags (Google Fonts or equivalent)
-
Update index.css with:
- Design tokens as CSS custom properties / Tailwind theme extensions
- Scrollbar-hiding utility classes
- Keyframe animations (slide-in for mobile menu, fade-up for scroll animations)
- Animation utility classes
-
Update tailwind.config / CSS with custom colors, fonts, spacing from tokens
-
Move downloaded assets into the artifact's public/ directory, organized by section:
public/images/hero/ — hero/carousel imagespublic/images/categories/ — category card imagespublic/images/products/ — product imagespublic/images/promo/ — promotional tile imagespublic/images/features/ — feature section images
-
Create an icons.tsx file with extracted SVG icons as React components
-
Make vite.config.ts resilient:
const port = Number(process.env.PORT) || 5173;
const basePath = process.env.BASE_PATH || "/";
-
Verify the build passes: pnpm --filter @workspace/<slug> run build
Essential CSS Foundation
@keyframes slide-in {
from { transform: translateX(-100%); }
to { transform: translateX(0); }
}
@keyframes fade-up {
from { opacity: 0; transform: translateY(24px); }
to { opacity: 1; transform: translateY(0); }
}
.animate-slide-in {
animation: slide-in 0.3s ease-out;
}
.animate-fade-up {
animation: fade-up 0.6s ease-out both;
}
@layer utilities {
.scrollbar-hide {
-ms-overflow-style: none;
scrollbar-width: none;
}
.scrollbar-hide::-webkit-scrollbar {
display: none;
}
}
Phase 3: Component Specification
For each section in the page topology, do THREE things: extract, write spec, dispatch builder.
3.1 Per-Component CSS Extraction
For each section, run a detailed getComputedStyle() extraction via Playwright. This walks the DOM tree (4 levels deep) and captures every relevant CSS property for each element.
def extract_component_styles(page, selector):
return page.evaluate("""
(selector) => {
const el = document.querySelector(selector);
if (!el) return { error: 'Element not found: ' + selector };
const props = [
'fontSize','fontWeight','fontFamily','lineHeight','letterSpacing','color',
'textTransform','textDecoration','backgroundColor','background',
'padding','paddingTop','paddingRight','paddingBottom','paddingLeft',
'margin','marginTop','marginRight','marginBottom','marginLeft',
'width','height','maxWidth','minWidth','maxHeight','minHeight',
'display','flexDirection','justifyContent','alignItems','gap',
'gridTemplateColumns','gridTemplateRows',
'borderRadius','border','borderTop','borderBottom','borderLeft','borderRight',
'boxShadow','overflow','overflowX','overflowY',
'position','top','right','bottom','left','zIndex',
'opacity','transform','transition','cursor',
'objectFit','objectPosition','mixBlendMode','filter','backdropFilter',
'whiteSpace','textOverflow','WebkitLineClamp',
'backgroundSize','backgroundPosition','backgroundRepeat'
];
function extractStyles(element) {
const cs = getComputedStyle(element);
const styles = {};
props.forEach(p => {
const v = cs[p];
if (v && v !== 'none' && v !== 'normal' && v !== 'auto' && v !== '0px' &&
v !== 'rgba(0, 0, 0, 0)' && v !== 'rgb(0, 0, 0)' && v !== 'start' &&
v !== 'stretch' && v !== 'visible' && v !== 'static' && v !== 'row' &&
v !== 'repeat' && v !== '0% 0%') {
styles[p] = v;
}
});
return styles;
}
function walk(element, depth) {
if (depth > 4) return null;
const children = [...element.children];
const rect = element.getBoundingClientRect();
return {
tag: element.tagName.toLowerCase(),
classes: element.className?.toString().split(' ').slice(0, 5).join(' ') || '',
id: element.id || null,
text: element.childNodes.length === 1 && element.childNodes[0].nodeType === 3
? element.textContent.trim().slice(0, 300) : null,
rect: { x: Math.round(rect.x), y: Math.round(rect.y), w: Math.round(rect.width), h: Math.round(rect.height) },
styles: extractStyles(element),
images: element.tagName === 'IMG' ? {
src: element.src,
alt: element.alt,
naturalWidth: element.naturalWidth,
naturalHeight: element.naturalHeight
} : null,
svg: element.tagName === 'SVG' ? {
viewBox: element.getAttribute('viewBox'),
html: element.outerHTML.length < 3000 ? element.outerHTML : '[TOO_LARGE]'
} : null,
childCount: children.length,
children: children.slice(0, 20).map(c => walk(c, depth + 1)).filter(Boolean)
};
}
return walk(el, 0);
}
""", selector)
3.2 Multi-State Extraction
For elements with multiple states (hover, scroll-triggered, tabbed content):
- Capture styles at State A (default)
- Trigger the state change (scroll, click tab via Playwright)
- Capture styles at State B
- Record the diff: "Property X changes from VALUE_A to VALUE_B, triggered by TRIGGER, with transition: TRANSITION_CSS"
Hover State Extraction:
def extract_hover_state(page, selector):
before = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
if (!el) return null;
const cs = getComputedStyle(el);
return {
backgroundColor: cs.backgroundColor, color: cs.color,
transform: cs.transform, boxShadow: cs.boxShadow,
opacity: cs.opacity, borderColor: cs.borderColor,
textDecoration: cs.textDecoration, transition: cs.transition
};
}
""", selector)
el = page.query_selector(selector)
if el:
el.hover()
page.wait_for_timeout(500)
after = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
if (!el) return null;
const cs = getComputedStyle(el);
return {
backgroundColor: cs.backgroundColor, color: cs.color,
transform: cs.transform, boxShadow: cs.boxShadow,
opacity: cs.opacity, borderColor: cs.borderColor,
textDecoration: cs.textDecoration, transition: cs.transition
};
}
""", selector)
diff = {}
if before and after:
for key in before:
if before[key] != after[key]:
diff[key] = {"before": before[key], "after": after[key]}
return {"before": before, "after": after, "diff": diff}
Scroll-Triggered State Extraction:
def extract_scroll_state(page, selector, scroll_to=200):
page.evaluate("window.scrollTo(0, 0)")
page.wait_for_timeout(500)
before = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
if (!el) return null;
const cs = getComputedStyle(el);
return {
backgroundColor: cs.backgroundColor, boxShadow: cs.boxShadow,
height: cs.height, padding: cs.padding, maxWidth: cs.maxWidth,
borderRadius: cs.borderRadius, position: cs.position,
top: cs.top, transform: cs.transform, opacity: cs.opacity,
backdropFilter: cs.backdropFilter
};
}
""", selector)
page.evaluate(f"window.scrollTo(0, {scroll_to})")
page.wait_for_timeout(800)
after = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
if (!el) return null;
const cs = getComputedStyle(el);
return {
backgroundColor: cs.backgroundColor, boxShadow: cs.boxShadow,
height: cs.height, padding: cs.padding, maxWidth: cs.maxWidth,
borderRadius: cs.borderRadius, position: cs.position,
top: cs.top, transform: cs.transform, opacity: cs.opacity,
backdropFilter: cs.backdropFilter
};
}
""", selector)
page.evaluate("window.scrollTo(0, 0)")
diff = {}
if before and after:
for key in before:
if before[key] != after[key]:
diff[key] = {"before": before[key], "after": after[key]}
return {"scrollThreshold": scroll_to, "before": before, "after": after, "diff": diff}
3.3 Responsive Layout Extraction
def extract_responsive(page, selector, breakpoints=[1440, 768, 390]):
results = {}
for width in breakpoints:
page.set_viewport_size({"width": width, "height": 900})
page.wait_for_timeout(800)
data = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
if (!el) return null;
const cs = getComputedStyle(el);
const rect = el.getBoundingClientRect();
return {
display: cs.display, flexDirection: cs.flexDirection,
gridTemplateColumns: cs.gridTemplateColumns, gap: cs.gap,
padding: cs.padding, width: Math.round(rect.width),
height: Math.round(rect.height), childCount: el.children.length,
childrenVisible: [...el.children].filter(c => {
const s = getComputedStyle(c);
return s.display !== 'none' && s.visibility !== 'hidden';
}).length
};
}
""", selector)
results[f"{width}px"] = data
page.set_viewport_size({"width": 1440, "height": 900})
return results
3.4 Complete Per-Section Extraction Flow
def extract_section_full(page, section_name, selector, out_dir="clone-data"):
print(f"\nExtracting: {section_name} ({selector})")
el = page.query_selector(selector)
if el:
el.scroll_into_view_if_needed()
page.wait_for_timeout(500)
el.screenshot(path=f"{out_dir}/screenshots/{section_name}.png")
styles = extract_component_styles(page, selector)
with open(f"{out_dir}/components/{section_name}-styles.json", "w") as f:
json.dump(styles, f, indent=2)
responsive = extract_responsive(page, selector)
with open(f"{out_dir}/components/{section_name}-responsive.json", "w") as f:
json.dump(responsive, f, indent=2)
text = page.evaluate("""
(sel) => {
const el = document.querySelector(sel);
return el ? el.innerText : '';
}
""", selector)
with open(f"{out_dir}/components/{section_name}-text.txt", "w") as f:
f.write(text)
return {"styles": styles, "responsive": responsive, "text": text}
3.5 Write Component Spec Files
For each section, write a spec file at docs/research/components/<component-name>.md using this template:
# <ComponentName> Specification
## Overview
- **Target file:** `src/components/<ComponentName>.tsx`
- **Interaction model:** <static | click-driven | scroll-driven | time-driven>
## DOM Structure
<Element hierarchy description>
## Computed Styles (exact values from getComputedStyle)
### Container
- display: ...
- padding: ...
### <Child elements>
- fontSize: ...
- color: ...
## Typography
- Headings: <font-family>, <font-size>, <font-weight>, <color>
- Body text: <font-family>, <font-size>, <line-height>, <color>
- Buttons: <font-size>, <font-weight>, <text-transform>, <letter-spacing>
## Colors
- Background: <exact rgb/hex value>
- Text primary: <exact value>
- Text secondary: <exact value>
- Accent/CTA: <exact value>
## States & Behaviors
### <Behavior name>
- **Trigger:** <exact mechanism>
- **State A:** <property values before>
- **State B:** <property values after>
- **Transition:** <CSS transition value>
## Text Content (verbatim)
<All text, copy-pasted from the live site>
## Assets
- Images: <list with local paths in public/>
- Icons: <list from icons.tsx>
## Responsive Behavior
- **Desktop (1440px):** <layout>
- **Tablet (768px):** <changes>
- **Mobile (390px):** <changes>
## Mobile Navigation (for Header component)
- Hamburger menu button: visible below <breakpoint>
- Drawer: slide-in from <direction>, <width>, <bg color>
- Backdrop: <overlay description>
- Nav links: <styling>
- Body scroll lock: yes
- Animation: <type and duration>
3.6 Complexity Assessment
Before dispatching builders, assess each section:
- Simple (1-2 sub-components): One builder agent
- Complex (3+ sub-components): One agent per sub-component + one for the section wrapper
- Rule of thumb: If the spec exceeds ~150 lines, break it into smaller pieces
Phase 4: Parallel Build with Subagents
Use Replit's design subagents to build components in parallel. Each subagent receives:
- The full component spec inline (not "go read the file")
- Path to the screenshot reference
- Which shared components to import (icons, cn(), UI primitives)
- The target file path
- Instruction to make it responsive and match the spec exactly
Dispatch Pattern
await startAsyncSubagent({
task: `Build the <ComponentName> component for the website clone.
TARGET FILE: artifacts/<slug>/src/components/<ComponentName>.tsx
COMPONENT SPECIFICATION:
<paste full spec file contents here>
INSTRUCTIONS:
- Match the computed CSS values EXACTLY — do not approximate
- Use Tailwind utility classes where they match; use inline styles or custom CSS for exact values
- Import icons from '../components/icons'
- Use real text content from the spec, not placeholders
- Make it fully responsive per the spec's breakpoint notes
- Export the component as default export
- For carousels: implement with useState for active index, auto-play with useEffect interval, dot indicators, and pause on hover
- For sticky headers: use scroll listener with useState to toggle scrolled state, apply transition with cubic-bezier easing
- For mobile menu: implement slide-in drawer with backdrop overlay, body scroll lock via document.body.style.overflow, and close on backdrop click
- Images: use local paths from public/ (e.g., /images/hero/slide-1.jpg), never external CDN URLs`,
specialization: "DESIGN",
relevantFiles: [
"artifacts/<slug>/src/index.css",
"artifacts/<slug>/src/components/icons.tsx",
"docs/research/components/<component-name>.md"
]
});
Dispatch multiple subagents in parallel for independent sections. Wait for all to complete before assembly.
Phase 5: Page Assembly
After all components are built:
-
Import all section components into App.tsx
-
Arrange them in DOM order matching the original page topology
-
Create a FadeInSection wrapper component using IntersectionObserver for scroll-triggered entrance animations:
import { useRef, useEffect, useState } from 'react';
export default function FadeInSection({ children, className = '' }) {
const ref = useRef(null);
const [visible, setVisible] = useState(false);
useEffect(() => {
const el = ref.current;
if (!el) return;
const observer = new IntersectionObserver(
([entry]) => {
if (entry.isIntersecting) {
setVisible(true);
observer.unobserve(el);
}
},
{ threshold: 0.1, rootMargin: '0px 0px -40px 0px' }
);
observer.observe(el);
return () => observer.disconnect();
}, []);
return (
<div ref={ref} className={`${className} ${visible ? 'animate-fade-up' : 'opacity-0'}`}>
{children}
</div>
);
}
-
Wrap mid-page sections (not hero, not header/footer) in <FadeInSection> for polished scroll reveal
-
Wire up real content data to component props
-
Verify build: pnpm --filter @workspace/<slug> run build
App.tsx Pattern
import AnnouncementBar from './components/AnnouncementBar';
import Header from './components/Header';
import HeroSection from './components/HeroSection';
import SectionA from './components/SectionA';
import SectionB from './components/SectionB';
import Footer from './components/Footer';
import FadeInSection from './components/FadeInSection';
function App() {
return (
<div className="min-h-screen bg-[var(--background)]">
<AnnouncementBar />
<Header />
<main>
<HeroSection />
<FadeInSection><SectionA /></FadeInSection>
<FadeInSection><SectionB /></FadeInSection>
</main>
<Footer />
</div>
);
}
export default App;
Phase 6: Visual QA
- Use the screenshot tool to capture your clone at desktop (1280px) and mobile (390px) viewports
- Compare against the original screenshots from Phase 1
- For each discrepancy:
- Check the spec file — was the value extracted correctly?
- If spec was wrong: re-extract via Playwright, update spec, fix component
- If spec was right but build is wrong: fix the component
- Verify all images load (no broken images)
- Check browser console for errors
- Test hover states and interactions
- Test mobile hamburger menu — click the menu button, verify drawer slides in, links are visible, close works
- No placeholder links — all
href="#" must be replaced with real URLs from the source site or reasonable alternatives (e.g., link to the original site's help page)
Present the artifact to the user when QA passes.
Common Pitfalls & Solutions
Pitfall 1: Missing Mobile Navigation
Problem: Clones often skip the hamburger menu, making the site unusable on mobile.
Solution: Always extract the mobile menu at 390px viewport. Click the hamburger button during extraction to capture drawer content, animation direction, width, and link list.
Pitfall 2: Reusing Images Across Sections
Problem: When a section needs unique images but the extractor pulled nothing, builders reuse hero/feature images as placeholders.
Solution: For each section, verify image URLs are section-specific. If extraction missed images (common with lazy-loaded promo tiles), manually navigate to those sections and download the specific images using curl with a User-Agent header.
Pitfall 3: Font Loading Delays
Problem: Fonts flash or don't load, causing layout shift.
Solution: Use <link rel="preconnect"> and <link rel="preload"> for font files. Add font-display: swap to @font-face declarations.
Pitfall 4: Playwright networkidle Timeout
Problem: page.goto(url, wait_until="networkidle") hangs forever on sites with persistent WebSocket/analytics connections.
Solution: Use wait_until="domcontentloaded" + page.wait_for_timeout(5000) instead.
Pitfall 5: CDN Image 403 Errors
Problem: CDN images return 403 when fetched without proper headers.
Solution: Always download with User-Agent: Mozilla/5.0 header. For Shopify CDN images, you can also try appending ?v=timestamp to bust cache protection.
Pitfall 6: Sticky Header Not Working
Problem: Header doesn't stick or doesn't show scroll-triggered styles.
Solution: Use a scroll event listener with useState for the scrolled state. Apply position: sticky; top: 0; z-index: 50 with a CSS transition on background-color and box-shadow. Common pattern:
const [scrolled, setScrolled] = useState(false);
useEffect(() => {
const handler = () => setScrolled(window.scrollY > 50);
window.addEventListener('scroll', handler, { passive: true });
return () => window.removeEventListener('scroll', handler);
}, []);
Pitfall 7: Unused Template Dependencies
Problem: Artifact scaffolding includes 40+ shadcn/Radix dependencies that bloat the project.
Solution: After building all components, audit package.json — grep for actual imports in src/ and remove any unused dependencies. Run pnpm install to clean up.
Data Architecture
Keep scraped content data inline in component files as typed arrays — no separate JSON fetches:
const products: Product[] = [
{ image: "/images/products/shoe-1.webp", name: "Tree Runner", price: "$98" },
];
Quick Reference: Full Workflow
1. pip install playwright
2. Find Chromium path (cache it)
3. Run recon script → screenshots + tokens + content + behaviors
4. Download ALL assets locally to public/images/ (never use CDN URLs in build)
5. createArtifact("react-vite", ...)
6. Set up foundation (CSS tokens, fonts, icons, animations, vite.config resilience)
7. For each section:
a. Extract computed styles via Playwright getComputedStyle
b. Extract hover/scroll/responsive states
c. Write spec file in docs/research/components/
d. Dispatch design subagent with full spec inline
8. Wait for all subagents
9. Assemble page with FadeInSection wrappers
10. Prune unused dependencies from package.json
11. Visual QA with screenshot comparison (desktop + mobile)
12. Verify mobile hamburger menu works
13. Replace all placeholder href="#" with real URLs
14. Present artifact
Dependency Minimalism
The final clone should only need these core dependencies:
react, react-domtailwindcss, @tailwindcss/vitevite, @vitejs/plugin-react- Replit vite plugins (
@replit/vite-plugin-cartographer, etc.)
@types/react, @types/react-dom, @types/node
Do NOT include shadcn/Radix, react-hook-form, recharts, wouter, framer-motion, or other template dependencies unless the clone actually uses them. Audit and prune after building.
