kernic for AI agents

Apps built by AI often look the same because the agent invents colors as it goes. Give it your system instead. One command and one rule, and everything it writes uses your tokens.

The one-command setup

kernic context midnight-neon -o .   # writes DESIGN.md, tokens.json and shadcn.json

DESIGN.md follows Google's DESIGN.md spec: every token in the front matter, usage rules in the prose. The adjacenttokens.json is W3C DTCG format, so Figma, Terrazzo, and Style Dictionary read it too.

Point the agent at it

One line in AGENTS.md, CLAUDE.md or .cursorrules:

Visual design: follow ./DESIGN.md exactly.
Use only its tokens. Never invent raw hex values.

Or go live over MCP

claude mcp add kernic -- npx kernic mcp

• apply_to_project: write the system into the project you are working in, and get the line to paste into your agent file
• create_system: make a new system without leaving the session
• list_looks: the 33 finished looks, colors and fonts already paired
• list_systems: discover saved systems
• get_system: full spec of one system
• get_tokens: tokens in design-md, css, tailwind, fonts, dtcg or json format
• list_vibes: the eight broad styles

get_tokens puts tokens in the conversation. apply_to_project puts them in your project, which is what makes them stick across future sessions.

Why it holds up

Token files include a version number and an extensions block. New features (motion, shadows, multi-brand) land without breaking existing files. Your systems are local JSON. They stay local.