Google Labs has released the draft specification for DESIGN.md, the file format that powers its AI design tool Stitch. This format lets AI agents understand design systems—previously it was locked inside Stitch, but now it's open source, so other tools can reuse the same design rules.
A DESIGN.md file has two layers. At the top, a YAML front matter defines design tokens in a machine-readable format, covering colors (with hex values), fonts (including fontFamily, fontSize, fontWeight, lineHeight, etc.), spacing, border radii, and components. Below that, Markdown body text explains the design rationale for humans, organized into sections like Overview, Colors, Typography, Layout, Elevation & Depth, Shapes, Components, and Do's and Don'ts. Component properties can reference tokens from above using syntax like {colors.primary}.
https://www.youtube.com/watch?v=W1gWIQp9k1Y
A companion CLI tool provides linting (with seven rules and WCAG contrast checks), diffing (to compare two versions), exporting (to formats like Tailwind or DTCG), and spec generation (to output documentation). The token system is based on the W3C Design Tokens Format Module.
The current version is alpha 0.1.0, licensed under Apache-2.0, and maintained by five contributors including davideast. You can generate your own DESIGN.md at stitch.withgoogle.com or submit changes to the GitHub repository google-labs-code/design.md.
For AI programming agents, this acts as a machine-verifiable design contract: an agent no longer has to guess what a color is for or how a component should be laid out. Instead, it can validate choices against the tokens and automatically check color accessibility using WCAG rules.