Website Theme Porter

Port the visual identity of any live website into a React + Tailwind CSS project. The workflow is four linear stages: Extract → Document → Apply → Verify.

Artifact Storage Convention

All artifacts (screenshots, extracted JSON, theme docs) must be saved under:

.context/artifacts/<website-slug>/<YYYY-MM-DD>/

Where <website-slug> is a lowercase, hyphenated version of the target domain (e.g. designandbloom-co-uk), and <YYYY-MM-DD> is the extraction date.

SLUG="designandbloom-co-uk"
DATE=$(date +%Y-%m-%d)
ARTIFACTS=".context/artifacts/${SLUG}/${DATE}"
mkdir -p "${ARTIFACTS}"

All screenshot paths, JSON dumps, and theme docs in the steps below use ${ARTIFACTS}/ as their base. Never save artifacts to docs/ or any other location.

Reference Loading Schedule

Load each reference file only when its stage begins — do not pre-load all three at once.

Stage 1: Extract

Pull raw design tokens from the target website. Choose the method that fits your situation:

Capture method decision tree

Can you run agent-browser OR mcp-playwright in this environment?
  YES → use Method A (automated browser extraction) — most thorough
        agent-browser: run CLI commands directly
        mcp-playwright: use playwright_browser_* MCP tools
  NO ↓

Is the site's HTML/CSS source accessible (View Source, curl, wget)?
  YES → use Method B (source inspection) — good for SSR/static sites
  NO ↓

Can you manually inspect the live site in browser DevTools?
  YES → use Method C (manual inspection) — always possible, least automated
  NO → unblock agent-browser or mcp-playwright access before continuing

For the full command equivalents table, see references/extraction.md.

Method A — Automated browser extraction (agent-browser or mcp-playwright)

Steps below use agent-browser syntax. If using mcp-playwright, translate each command using the equivalents table above. The JS payloads are identical — only the invocation differs.

1a. Navigate and screenshot

agent-browser open <TARGET_URL> && agent-browser wait --load networkidle
agent-browser screenshot --full "${ARTIFACTS}/source-full.png"
agent-browser set viewport 375 812 && agent-browser screenshot --full "${ARTIFACTS}/source-mobile.png"
agent-browser set viewport 1280 720

1b. Extract computed styles via JS

agent-browser eval --stdin <<'JS'
JSON.stringify({
  colors: (() => {
    const s = getComputedStyle(document.documentElement);
    const props = Array.from(document.styleSheets)
      .flatMap(sheet => { try { return Array.from(sheet.cssRules); } catch { return []; } })
      .filter(r => r.selectorText === ':root')
      .flatMap(r => Array.from(r.style));
    return props.filter(p => p.startsWith('--')).reduce((acc, p) => {
      acc[p] = s.getPropertyValue(p).trim(); return acc;
    }, {});
  })(),
  computed: (() => {
    const el = document.body;
    const s = getComputedStyle(el);
    return {
      fontFamily: s.fontFamily,
      fontSize: s.fontSize,
      lineHeight: s.lineHeight,
      backgroundColor: s.backgroundColor,
      color: s.color,
    };
  })()
}, null, 2)
JS

Save JSON output to ${ARTIFACTS}/tokens-css-vars.json.

If the colors object is empty (site uses no CSS custom properties): The site styles everything with hardcoded values. Switch to computed-only extraction:

agent-browser eval --stdin <<'JS'
const els = { body: document.body, h1: document.querySelector('h1'),
  btn: document.querySelector('button,[class*=btn]'),
  nav: document.querySelector('nav,header'),
  card: document.querySelector('[class*=card],article') };
const out = {};
for (const [name, el] of Object.entries(els)) {
  if (!el) continue;
  const s = getComputedStyle(el);
  out[name] = { bg: s.backgroundColor, color: s.color, font: s.fontFamily,
    fontSize: s.fontSize, fontWeight: s.fontWeight, borderRadius: s.borderRadius,
    border: s.border, boxShadow: s.boxShadow, padding: `${s.paddingTop} ${s.paddingRight}` };
}
JSON.stringify(out, null, 2)
JS

Save to ${ARTIFACTS}/tokens-computed.json. Manually assign tokens from these values in Stage 2. Note in ${ARTIFACTS}/theme/overview.md: "source uses hardcoded values — no design token system detected".

If the target is a SPA (React, Next.js, Vue, etc.): Content may not be rendered on first paint. Add a scroll-and-wait step before extracting:

agent-browser eval 'window.scrollTo(0, document.body.scrollHeight)'
agent-browser wait --load networkidle
agent-browser wait --timeout 2000

1c. Collect typography, spacing, and key element styles

agent-browser eval --stdin <<'JS'
JSON.stringify(
  ['h1','h2','h3','h4','p','a','button'].reduce((acc, tag) => {
    const el = document.querySelector(tag);
    if (!el) return acc;
    const s = getComputedStyle(el);
    acc[tag] = {
      fontFamily: s.fontFamily,
      fontSize: s.fontSize,
      fontWeight: s.fontWeight,
      lineHeight: s.lineHeight,
      color: s.color,
      letterSpacing: s.letterSpacing,
    };
    return acc;
  }, {}),
null, 2)
JS

Save JSON output to ${ARTIFACTS}/tokens-typography.json.

1d. Capture screenshots of key sections

agent-browser screenshot --full "${ARTIFACTS}/source-annotated.png"

See references/extraction.md for the full JS extraction toolkit, colour parsing helpers, nav/card/footer extractors, and page-by-page capture patterns.

Method B — Source inspection (curl / wget / View Source)

Use when agent-browser is unavailable or blocked. Fetch the raw HTML and linked stylesheets, then grep for CSS custom properties and font declarations manually.

curl -sL <TARGET_URL> -o "${ARTIFACTS}/source.html"
grep -oP '(?<=href=")[^"]*\.css[^"]*' "${ARTIFACTS}/source.html"
curl -sL <CSS_URL> -o "${ARTIFACTS}/styles-main.css"
grep -oP '--[\w-]+\s*:\s*[^;]+' "${ARTIFACTS}/styles-main.css" > "${ARTIFACTS}/tokens-raw.txt"
grep -E '@font-face|fonts\.googleapis' "${ARTIFACTS}/styles-main.css" > "${ARTIFACTS}/fonts-raw.txt"

Parse ${ARTIFACTS}/tokens-raw.txt manually and map values to the token table in Stage 2. Note in ${ARTIFACTS}/theme/overview.md: "extraction method: source inspection (curl)".

Method C — Manual DevTools inspection

Use as a fallback or to supplement other methods.

1. Open the target site in Chrome/Firefox DevTools
2. Elements → select <html> → Computed tab → filter for -- → copy all custom properties to ${ARTIFACTS}/tokens-manual.txt
3. Network tab → filter .css → copy stylesheet contents
4. Console: run JS snippets from references/extraction.md → paste output to ${ARTIFACTS}/tokens-computed.json
5. Take screenshots of key sections (hero, nav, footer, mobile) and save under ${ARTIFACTS}/

Note in ${ARTIFACTS}/theme/overview.md: "extraction method: manual DevTools inspection".

Stage 2: Document

Write the findings into ${ARTIFACTS}/theme/ before touching any code.

mkdir -p "${ARTIFACTS}/theme"

Required output files

Colour documentation format

## Colours

| Token | Hex | HSL | Role |
|-------|-----|-----|------|
| --primary | #0D9488 | 174 90% 30% | CTA buttons, links |
| --background | #FFFFFF | 0 0% 100% | Page background |
| --foreground | #111827 | 221 39% 11% | Body text |

Typography documentation format

## Typography

**Primary font**: Inter, sans-serif  
**Heading font**: same (or specify if different)

| Element | Size | Weight | Line Height |
|---------|------|--------|-------------|
| h1 | 2.25rem | 700 | 1.2 |
| h2 | 1.875rem | 600 | 1.3 |
| body | 1rem | 400 | 1.6 |

Use the Token Decision Tree in references/tailwind-mapping.md to assign ambiguous colours.

Stage 3: Apply

Translate the documented theme into Tailwind CSS variables and component styles.

3a. Update src/index.css (Tailwind v4 pattern)

@import "tailwindcss";

/* Semantic colour tokens */
:root {
  --background: hsl(0 0% 100%);
  --foreground: hsl(221 39% 11%);
  --primary: hsl(174 90% 30%);
  --primary-foreground: hsl(0 0% 100%);
  --muted: hsl(210 40% 96%);
  --muted-foreground: hsl(215 16% 47%);
  --border: hsl(214 32% 91%);
  --radius: 0.5rem;
}

/* Map to Tailwind utilities */
@theme inline {
  --color-background: var(--background);
  --color-foreground: var(--foreground);
  --color-primary: var(--primary);
  --color-primary-foreground: var(--primary-foreground);
  --color-muted: var(--muted);
  --color-muted-foreground: var(--muted-foreground);
  --color-border: var(--border);
  --font-sans: "Inter", ui-sans-serif, system-ui, sans-serif;
}

@layer base {
  body {
    background-color: var(--background);
    color: var(--foreground);
    font-family: var(--font-sans);
  }
}

3b. Add font to index.html

<link rel="preconnect" href="https://fonts.googleapis.com" />
<link href="https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;700&display=swap" rel="stylesheet" />

3c. Update component classes

Replace hardcoded colours/fonts with semantic Tailwind tokens:

// Before
<button className="bg-[#0D9488] text-white px-4 py-2 rounded">

// After
<button className="bg-primary text-primary-foreground px-4 py-2 rounded-[var(--radius)]">

Work page-by-page, starting with shared layout components (Header, Footer) then pages.

Stage 4: Verify

Compare the original site visually against the local build.

4a. Start local dev server

bun run dev  # or: npm run dev / vite

4b. Visual diff

agent-browser produces a pixel-diff image and mismatch percentage:

agent-browser diff url <TARGET_URL> http://localhost:5173 --screenshot "${ARTIFACTS}/diff-homepage.png"

mcp-playwright — navigate to each URL and take screenshots without a filename; images are returned inline into context so the agent can inspect and describe differences directly:

playwright_browser_navigate(url: "<TARGET_URL>")
playwright_browser_wait_for(time: 2)
playwright_browser_take_screenshot(fullPage: true)   # source — inline

playwright_browser_navigate(url: "http://localhost:5173")
playwright_browser_wait_for(time: 2)
playwright_browser_take_screenshot(fullPage: true)   # local — inline

4c. Deployed verification (if Vercel / other hosting)

# agent-browser
agent-browser diff url <TARGET_URL> https://<your-preview>.vercel.app --screenshot "${ARTIFACTS}/diff-deployed.png"

# mcp-playwright: repeat the navigate + take_screenshot (no filename) pattern above against the preview URL

See references/verification.md for the full diff interpretation guide, threshold table, and symptom→fix patterns for all common mismatches.

When NOT to Use This Skill

• Login-walled or paywalled sites — extraction requires an authenticated session; the browser cannot reach the styled content without credentials.
• Canvas-rendered or WebGL UIs — apps that paint entirely to a <canvas> element have no CSS to extract; getComputedStyle returns nothing useful.
• Sites with heavily obfuscated class names (CSS Modules / CSS-in-JS hashes) — class names like _3xKy9 are generated at build time and change on every deploy; do not attempt to port them. Extract computed values only via Method A JS snippets.
• Designs you do not have permission to replicate — confirm you have the right to port the visual identity before starting.

What to Extract

Prioritise in this order:

1. Colour palette — backgrounds, text, borders, CTAs, accents
2. Typography — font families (check @font-face or Google Fonts links), size scale, weights
3. Spacing rhythm — common padding/margin values (e.g. 16px, 24px, 48px grid)
4. Border radius — buttons, cards, inputs (flat vs rounded)
5. Shadows — none, subtle, card, elevated
6. Component patterns — button variants, card borders, nav height, footer layout

NEVER Do This

• NEVER wrap HSL values in hsl() inside @theme inline. In Tailwind v4, --color-primary: hsl(174 90% 31%) inside @theme inline will double-wrap when Tailwind generates utilities, producing broken output. Consequence: bg-primary resolves to hsl(hsl(174 90% 31%)) — an invalid value; the colour renders as transparent or black.

  /* ❌ BAD — double-wraps at utility generation time */
  @theme inline {
    --color-primary: hsl(174 90% 31%);
  }
  
  /* ✅ GOOD — bare channels in @theme inline; hsl() in :root */
  :root { --primary: hsl(174 90% 31%); }
  @theme inline { --color-primary: var(--primary); }

• NEVER extract colours from :hover pseudo-states. getComputedStyle on a hovered element requires the pointer to be physically over it. The value may silently return the resting state, giving a token that looks correct but is wrong. Consequence: Hover colours get assigned as the base token — all buttons show the hover shade permanently.

• NEVER use prefers-color-scheme media query colours as your base tokens. Computed styles reflect the current OS colour scheme at extraction time. Consequence: If the machine is in dark mode, extracted "background" is near-black — applying it to a light-mode project inverts the entire colour scheme. Always extract in light mode, or explicitly document which mode was active in overview.md.

• NEVER use /tmp for any artifact storage. /tmp is ephemeral and session-scoped — artifacts saved there are lost when the session ends. Always save to .context/artifacts/<website-slug>/<YYYY-MM-DD>/.

  # ❌ BAD
  agent-browser screenshot /tmp/source.png
  
  # ✅ GOOD
  agent-browser screenshot "${ARTIFACTS}/source-full.png"

• NEVER copy the source site's actual CSS files or stylesheets into your project. Rebuild every value from scratch using the extracted numbers.

• NEVER use arbitrary Tailwind values (bg-[#abc]) for semantic colours. Arbitrary values bypass the token system — use semantic utility classes instead.

  // ❌ BAD — hardcoded, bypasses token system
  <button className="bg-[#0D9488] text-white">
  
  // ✅ GOOD — semantic token, single source of truth
  <button className="bg-primary text-primary-foreground">

Reference Files