Community이미지github.com

brand-capture

Captures a company's visual identity from a flow of app screenshots (home, listing, product, cart, checkout) or from its live website (logo, fonts, exact colours by role, radius, shadows, icon and photo style, voice) and writes an accurate design system (design-system.md and tokens.json) for building a company-specific portfolio or pitch. Use when someone names a target company and wants their portfolio in that company's look, says "get Zepto's fonts and colours", "here are screenshots of their app", "make it look like their brand", or runs brand-capture.

brand-capture란 무엇인가요?

brand-capture is a Claude Code agent skill that captures a company's visual identity from a flow of app screenshots (home, listing, product, cart, checkout) or from its live website (logo, fonts, exact colours by role, radius, shadows, icon and photo style, voice) and writes an accurate design system (design-system.md and tokens.json) for building a company-specific portfolio or pitch. Use when someone names a target company and wants their portfolio in that company's look, says "get Zepto's fonts and colours", "here are screenshots of their app", "make it look like their brand", or runs brand-capture.

지원 대상✓Claude Code~Codex CLI~Cursor✓Gemini CLI
npx skills add https://github.com/vidhu12/what-the-folio-skills/tree/HEAD/skills/brand-capture

Installed? Explore more 이미지 skills: steipete/songsee, affaan-m/frontend-design, mattpocock/improve-codebase-architecture · View all 6 →

즐겨 사용하는 AI에게 물어보기

이 에이전트 스킬이 미리 로드된 새 채팅을 엽니다.

문서

brand-capture은(는) 무엇을 하나요?

A company portfolio only works if it looks like the company made it. Guessing ("orange and a rounded sans") is not good enough. Read the real values from the live site.

Use and limits

This is for a private pitch sent to the company, presented as a student's independent concept. Do not publish a site that a visitor could mistake for the company's own. Do not use their logo as the student's logo. A company's paid or custom font is licensed to the company. Use it only inside the private pitch sent to that company by direct link (noindex, not listed anywhere), and say so in the confidence note; for anything public, and whenever in doubt, use the closest free Google Font.

Inputs

Either or both of:

  • A flow of app screenshots, in order. Many brands are app-first and their website is thin or just a download page. Ask the student for 5 to 8 screenshots of the app in the order a customer meets them: home, a listing or category, a product or item page, the cart or basket, checkout or payment, plus one empty state or error if they have it. One note per screenshot on what it is. With fewer screenshots, still run the method and list the missing screens under "not captured". This is a full input on its own; the website is optional.
  • The company website URL (home page plus one product or category page). Optional: an ad, a pack photo. Ask which they have. If the student has the app on their phone, screenshots take five minutes and are more accurate than a marketing website.

Method A: a flow of app screenshots (any AI that can see images)

Never fill a gap from memory or from "what such apps usually look like": only what is in the screenshots counts. Work through them in order. Values taken from a screenshot are estimates, close but not exact (a screenshot gives you a colour within a few points and a font by its look, not its name). Use "read" only for text you copied and patterns you observed; everything else from this method is "estimated from a screenshot".

  1. Logo: from the home screen. Describe it (wordmark or symbol, colour, shape). Ask the student for the press-kit or app-store icon file; do not redraw it.
  2. Colours: pick the colour of the main button, the background, the main text, the muted text, one accent, the price or success colour, and a border or divider. Read the hex with a colour picker if you have one (Preview on a Mac, Paint on Windows, or any eyedropper extension) and label it picked from screenshot; otherwise give your best read and label it eye estimate. Keep the two apart in the output and the confidence note. Assign roles the same way as Method B step 3. If no single brand colour appears in the UI (for example the main button is plain white on a dark photo), write primary = not captured and list the colour fields you did see; do not promote a tile, tag or photo colour to primary.
  3. Fonts: describe what you see: serif or sans, geometric or humanist, weight on headings and on body, whether numbers are tabular. Name the closest free Google Font that matches, put it under --font-display-fallback / --font-body-fallback as match (by look), and leave --font-display / --font-body null until the real name is confirmed. If the student can, ask them to check the real name on the website or in the app's settings or legal page.
  4. Shape and depth: corner radius on buttons, cards, inputs and images (sharp, slightly rounded, pill); shadows or flat; border widths; how much padding inside cards.
  5. Patterns: from the flow, name the 5 most recognisable UI patterns and on which screen each appears (for example "category tiles with circular images on home", "sticky add-to-cart bar on the product page", "progress strip at the top of checkout").
  6. Icons and imagery: line or filled icons, stroke weight, corner style; photo style (packshot on white, lifestyle, illustration, 3D).
  7. Voice: copy 3 real lines from the screenshots: a headline, a button label, a piece of microcopy (an empty state, an error, a tooltip). Describe the tone in 4 words. Then write the output below. In the confidence note, everything from this method goes under "estimated from screenshots" except the three voice lines and the patterns, which are read.

Method B: you can browse or run code (Claude Code, website available)

  1. Load the page. Collect the CSS: <link rel="stylesheet"> files and inline styles. If the page comes back empty, shows a "checking your browser" or CAPTCHA page, or returns an error (for example 202, 403, 429), stop. Do not try to get around it. Tell the student the site blocks automated reading and switch to Method A (app screenshots) or Method C (by hand). Never fill the gaps from memory or from "what such apps usually look like".
  2. Fonts: find every @font-face (family, weight, style, src URL). Note which family is used on h1, body text, buttons. Record the font file URLs. If the font is free (for example a Google Font), download it. If it is the company's own or a paid font, it is licensed to the company: you may keep the file on the student's machine for the private pitch only (see "Use and limits"), never commit it to a public repo or use it on the student's general portfolio. Always also name the closest free Google Font.
  3. Colours: read :root CSS custom properties first. Then count the most used color and background-color values (on visible elements if you have a browser; in the CSS files if you do not, and say which). Assign roles: background, surface, text, muted text, primary/brand, accent, success, warning, border.
  4. Logo: prefer the inline <svg> in the header, else the <img> in the header, else the favicon/apple-touch-icon, else Wikimedia Commons (check the file's date, it may be an old logo, and record it as a secondary source). Save as SVG when possible. The logo is the company's trademark: it is kept for reference and for one small "a concept for " mention, never as the student's logo or site header.
  5. Shape and depth: border-radius on buttons, cards, inputs; shadow values; border widths; spacing scale (look at padding on cards and sections).
  6. Imagery: screenshot 5 screens: hero, a card list, a product detail, buttons, an empty or error state. If you cannot take screenshots, ask the student for them; until they arrive, write "not captured" for Imagery, Icons and Patterns. Describe the photo style (studio, lifestyle, flat illustration, 3D), the icon style (line or filled, stroke width, corner style).
  7. Voice: copy 3 real lines: a headline, a button label, a piece of microcopy (error, empty state, tooltip). Describe the tone in 4 words.

Method C: the student does it by hand from the website (ChatGPT, Gemini, no browsing)

Give them these steps and ask them to paste back the results:

  1. Open the site in Chrome on a laptop. Press Cmd+Option+I on a Mac, or F12 (or Ctrl+Shift+I) on Windows. A side panel called DevTools opens.
  2. Fonts: right-click the main heading, choose Inspect, open the Computed tab, scroll to the bottom and note the name under "Rendered Fonts". Repeat for body text and a button. Or install the WhatFont extension and click the same three things. (Network tab, filter "Font", reload also lists the font files, but the names are often scrambled.)
  3. Colours: in the Elements tab click the <html> line at the top; in Styles look for a :root block with lines starting -- and copy them. Many sites have none. Then right-click the main button, choose Inspect, and in Styles click the small colour square next to background-color to read its hex code. Do the same for the page background and the main text.
  4. Logo: right-click the header logo, Inspect, copy the <svg> element (or open the image in a new tab and save it).
  5. Screenshots of 5 screens as above, and 3 lines of their copy.

Output

brand/<company>/design-system.md:

# <Company> design system (captured <date> from <urls>)
Colours:   role | hex | where seen | how read (live / picked / eye / null)
Type:      display family + weights | body family + weights | sizes for h1, h2, body, small | letter-spacing
           free fallback for public use: <Google Font>
Shape:     radius (button, card, input) | border | shadow
Spacing:   base unit and common paddings
Logo:      file path, clear-space and "do not" notes
Icons:     style, stroke, corner
Imagery:   style in one line, 3 adjectives
Voice:     4 words + the 3 real lines
Patterns:  the 5 most recognisable UI patterns (e.g. "category tiles with circular images", "sticky add-to-cart bar")

brand/<company>/tokens.json (tokens are the same values written as CSS variables the build can use directly) (--brand-primary, --font-display, --radius-card, ...), with a _confidence key that gives each token's source (live / picked / eye / null).

Then a confidence note with five lists: read from the live site; read from another source (name it and its date); picked from screenshots; eye estimates from screenshots; not captured. Write "not captured" (null in tokens.json) instead of a guess. Never present an estimate as exact. If the fonts, the main colours or the three voice lines are not captured, say plainly that the design system is not ready for brand-fit.

End with: "Next: run brand-fit to map your portfolio into their product."

관련 스킬