Who it is forA design system lead or frontend developer starting a new product who wants consistent colors, spacing, and type tokens across agent-built screens.
The problemAgents hardcode values inline, so styles drift and no single source controls the look.
The outcomeA token file in DTCG format with primitive, semantic, and component tiers, plus spacing and type scales.
What you need first
- git and a terminal to clone the repo
- An AI coding agent such as Claude Code, Codex, Cursor, or opencode
- A web browser to open the before and after HTML examples
- A brief for the page or interface you want to build
Steps
- Run
git clone https://github.com/madebymustafa/design-taste.gitand thencd design-taste. - Run
./scripts/install.shor copy theskills/taste-design-systemandskills/taste-code-craftfolders into your agent's skills directory. - Ask your agent to create the three token tiers with the taste-design-system skill for your product brief.
- Ask the agent to make the CSS use only token references, following taste-code-craft.
- Run
./scripts/validate.shfrom the repo to check the skill files, then review the token output.
Prompt for your agent
Using the taste-design-system and taste-code-craft skills from the design-taste repo, create a three-tier DTCG token file for <product brief> and update <stylesheet path> to use only token references. Verify that no hardcoded color, spacing, or type values remain in the stylesheet.
Caveat. The README says validate.sh checks skills before a release, not the tokens your agent produces, so review the token file yourself.
Use this in your AI agent
Ask your agent
Using the Deduped MCP, get the setup steps for madebymustafa/design-taste and "Set up a three-tier design token system for a project"
Add Deduped to Claude Code once
claude mcp add --transport http deduped https://dedupedai.com/mcp
Every Monday: the 5 hottest repos and exactly how to use them. Free, one email a week, unsubscribe anytime.
Get them freeShare: X LinkedIn Hacker News Email Copy link
Third-party repos we don't own. AI-written use cases can be wrong. Disclaimer