Skip to main content
AI coding agents work best when they understand your design system. Humic gives agents access to your tokens through the CLI, so they can generate code that uses the right values instead of hardcoding colors and spacing.
Using Claude? The Claude connector is the easier path — one URL, no CLI install, and it can edit and publish tokens, not just read them. The CLI workflow below is for other agents, or for pinning tokens into a repo-local context file.

How it works

The Humic CLI supports --json output on every command. Agents can call these commands and parse the JSON to understand your design system — no custom integration needed.

Setup

1. Install the CLI

Add Humic as a dev dependency so it’s available in your project:

2. Initialize your project

This creates humic.config.json and writes a context file that agents can read.

3. Configure your agent

Add instructions to your agent’s configuration or system prompt. Here’s an example for Claude Code (in CLAUDE.md):
For Cursor, add this to .cursor/rules:

Useful commands for agents

Code validation

The validate command uses AI to check code against your design system rules. It catches:
  • Hardcoded color values that should use tokens
  • Magic numbers for spacing that should use primitives
  • Inconsistent usage patterns
This requires a Gemini API key (runs locally, no data sent to Humic):

CI integration

Add validation to your CI pipeline to catch design system violations before they’re merged: