Agent Ready — AI skill for Claude Code

Agent Ready

**Your design file is probably not ready for AI agents. Here's how to tell, and what to do about it.**

When an AI coding agent reads a Figma file through MCP, it sees a very different thing than you do. It sees unnamed layers called "Frame 247", empty component descriptions, raw hex values where tokens should be, and missing states. Each gap forces it to guess. The guesses compound. The code comes out messy, generic, or quietly wrong — and most of the time, you blame the agent.

Agent Ready is built around one claim: **design files need to be agent-readable now, or agent collaboration produces bad outputs.** This repo contains two things that address that claim from two different directions.

By MC Dean · [Percolates on Substack](https://marieclairedean.substack.com)

---

What this repo contains

agent-ready-plugin/
├── SKILL.md        — the agent skill (the real product)
├── GEMINI.md       — the same skill for Gemini CLI
├── shared/         — executable checks the skill can run against raw Figma JSON
├── manifest.json   — the Figma plugin (the educational on-ramp)
├── code.js         — the plugin scan engine
├── ui.html         — the plugin UI
└── README.md       — you're here

There are two tools here, but they are not equal.

**The skill is the real product.** It's what actually raises the ceiling of what an agent can produce from a messy Figma file. It runs inside your coding agent (Claude Code, Gemini CLI, and more soon), silently assesses every file the agent reads, and compensates for gaps before generating code. It emits a structured evidence block at the top of every file it produces so a human reviewer can audit what the agent had to guess.

**The plugin is the educational on-ramp.** It's the thing that makes a designer realise, visually, that their file isn't ready for agent collaboration. You run it inside Figma, see a score, see the specific gaps, and get a prioritised fix list. That's its job. It's not where the long-term v