Why we exist
One .md file.
Your site stops looking generic.
Every AI-generated landing page looks the same. Rounded cards. A purple-to-blue gradient. A centered hero. A "Get Started" button. It works. It also looks like everything else. This page is about how to escape that.
The problem
Tell any AI agent to build you a landing page and you already know what comes back: rounded cards, a purple-to-blue gradient, a centered hero, a Get Started button. The result is technically correct. It also looks like every other vibe-coded site on the internet.
This is not the agent's fault. When it builds your page, it averages millions of pages it's seen during training. The average of millions of designs is a generic design.
It has no clue why Vercel uses 1px borders instead of shadows. Why Linear keeps letter-spacing punishingly tight. Why Stripe picks one bright accent and uses it like it's rationed. Those decisions live in the brands' heads not in the agent's training data.
So you're stuck with two bad options:
- Write a 40-line prompt every time "use #0070f3 for links, -0.02em letter-spacing on headings, 8px corners, no drop shadows just hairline borders…" and still get half wrong.
- Paste a screenshot and say "make it look like this." The agent copies the pixels but misses the system underneath.
Neither scales. Both burn hours.
What DESIGN.md is
A plain-text file that describes your brand's visual language. The agent reads it before it writes any UI. The format was introduced by Google's design agent, Stitch.
It sits in your repo root, next to the two files you already have:
No Figma plugin. No JSON schema. No CLI. Just one markdown file the agent reads before it writes anything.
What's inside the file
Every DESIGN.md has two layers. The first is YAML at the top of the file the exact values.
---
name: Your Brand
colors:
primary: "#0039ff"
ink: "#0a0a0a"
canvas: "#fafafa"
typography:
display-lg:
fontFamily: Geist
fontSize: 56px
fontWeight: 500
letterSpacing: -1.5px
body-md:
fontFamily: Geist
fontSize: 16px
fontWeight: 400
lineHeight: 1.6
rounded:
md: 8px
lg: 16px
spacing:
base: 16px
section: 80px
---These are tokens. The agent uses them as precise references the primary blue is #0039ff, not something close to it.
The second layer is everything below the closing --- plain markdown, explaining what each token is for and when to use it.
## Overview
Your Brand reads like a serious tool that
respects the user's attention. Display type
stays large but modest in weight. One blue
accent, used scarcely so it never dilutes.
## Colors
- **Primary** (#0039ff): The single brand
color. Used for primary CTAs and active
states only. Never as a section background
it should feel precious.
- **Ink** (#0a0a0a): Body text. Pure black
is too harsh; this almost-black keeps the
warmth.The tokens give the agent what. The prose gives it why. Without the why, the agent has values but no instinct it can't make the right call when the file doesn't explicitly cover the situation.
What this changes
A DESIGN.md is small. Maybe 200 lines of YAML and prose. The effect is disproportionate.
Your site stops looking generic
The agent stops averaging. Every UI it generates uses your colors, your typography, your spacing not the internet's.
Every screen feels like one product
Dashboard, settings, onboarding, error pages they all share the same DNA. Without DESIGN.md, each screen is its own little island.
You stop repeating yourself
No more 40-line preamble in every prompt. The agent already knows your tokens it's reading them off disk.
It works with the tools you have
Claude Code. Cursor. Windsurf. Copilot. They all read markdown and respect repo-level context. Drop the file in. They pick it up.
It evolves like code
Brand changes? Edit the file. PR it. Diff it. Ship it. No more Figma libraries drifting out of sync with what's actually deployed.
It's a starting point, not a cage
The file gives the agent a language to start in. You or the agent keep building on top. Drift happens. The system grows.
How you use it
Step 01
Extract
Paste any public URL on the home page. We crawl the live site, read its colors, typography, spacing, and component patterns, and write a DESIGN.md from what's actually shipping. Try Stripe. Try Linear. Try a competitor.
Step 02
Drop it in your repo
Save the file as
DESIGN.mdat the root of your project, next toREADME.mdandAGENTS.md. Commit it. Treat it like code.Step 03
Tell your agent
"Read DESIGN.md before you write any UI." That's the entire instruction. From this point on, every component the agent generates uses your tokens and follows your rules.
The mental model
You used to hire a designer and say "you know Linear, right? Give me that feel." It worked because the designer already carried Linear in their head. That shared context sat underneath every conversation.
An AI agent doesn't have that shared context.
DESIGN.md writes the context into a file and drops it into the agent's head. The reference pool you spent years building with a designer, you set up with an agent in two minutes.
What it isn't
Not a theme.There's no code inside. You or your agent still build the components. The file describes what a button looks like; the agent still has to write the JSX.
Not a brand PDF.Brand guidelines are written for humans and use phrases like "approachable yet premium." An agent can't act on that. DESIGN.md has to be specific enough to drive the next decision exact hex values, exact font sizes.
Not a Figma export. A tokens.json export tells you the values but skips the reasoning. DESIGN.md is structured so the prose travels with the tokens.
Not static.When the brand evolves, the file evolves. It's versioned, reviewed, merged. It behaves like code, because it is.
start now
Ready to stop being generic?
Paste any URL. Get a DESIGN.md back. Drop it in your repo. Watch the agent stop averaging.
Disclaimer
The DESIGN.md files generated by this site are not official design systems from the listed brands. They are extracted starting points based on publicly observable design patterns. All trademarks, brand names, and design elements belong to their respective owners. These files document publicly observable design patterns for educational and development purposes only.