Design guidance is often locked in Figma files or scattered CSS; DESIGN.md flips that by encoding visual systems as plain text LLMs read well. This repository collects ready-to-drop DESIGN.md files, previews, and prompt hints so coding/design agents can produce consistent UI that maps closely to real product visuals.
What Sets It Apart
- Real-site extractions with previews — the files are derived from public websites and include preview.html (light/dark) so agents work from production-visible tokens rather than hand-waved color lists; this reduces mismatch between generated UI and target brand.
- Agent-focused artifacts — each entry bundles semantic color roles, typography scales, component rules and a short prompt guide, so an AI agent can consume the design doc with minimal translation effort.
- Breadth and curation — covers multiple categories (platforms, developer tools, consumer brands) with analyzed patterns and do/don't rules; this makes it easier to prototype distinct visual languages without designing from scratch.
- Lightweight, text-first format — because DESIGN.md is just markdown, it integrates with code projects and CI pipelines more easily than binary Figma assets, enabling programmatic UI generation.
Who it's for and trade-offs
Great fit if you want to rapidly prototype UIs with AI agents, enforce visual consistency across generated screens, or study real-world design tokens for prompt engineering. It’s especially useful for teams using coding agents or Google Stitch-style workflows.
Look elsewhere if you need authoritative, up-to-date brand assets or legal permission to reproduce a company’s identity in production — the files are extracted from public CSS and are best suited for prototyping and agent-driven generation rather than replacing an official design system. Also, extracted tokens can miss internal variants or platform-specific behavior, so expect refinement when moving to production.
