AI

Carbone Skill

Accurate Carbone tag syntax, formatters, and best practices for AI assistants

Overview

A Carbone Skill is a knowledge file that teaches AI assistants the Carbone templating syntax — so they can design valid templates accurately, without trial and error.

What the Skill covers:

Compatible with: Claude, ChatGPT, Cursor, Copilot, Gemini, and all platforms supporting the open Agent Skills standard.

Installation

Download carbone.skill  ·  View source on GitHub

ChatGPT

  1. Click your profile icon and select Skills (or go to chatgpt.com/skills)
  2. Click New skill → Upload from your computer
  3. Select carbone.skill

The skill appears under Installed and is active immediately.

Claude (Desktop & claude.ai)

  1. Click the Customize button in the left navigation bar
  2. Select Skills
  3. Click the + button → Upload a Skill
  4. Select carbone.skill

The skill is now active for all conversations.

Claude Code

Install via the plugin marketplace:

/plugin marketplace add carboneio/carbone-skill
/plugin install carbone@carbone-skill

Or install manually by extracting the skill into your personal skills folder:

unzip carbone.skill -d ~/.claude/skills/carbone

To limit it to a single project instead:

unzip carbone.skill -d .claude/skills/carbone

Cursor

Extract the skill into your personal skills folder so it is available across all your projects:

unzip carbone.skill -d ~/.cursor/skills/carbone

To limit it to a single project instead, extract it into the project directory:

unzip carbone.skill -d .cursor/skills/carbone

Copilot

Extract the skill into your personal skills folder so it is available across all your projects in VS Code (GitHub Copilot):

unzip carbone.skill -d ~/.copilot/skills/carbone

To limit it to a single project instead, extract it into the project directory:

unzip carbone.skill -d .github/skills/carbone

Gemini

Install with the Gemini CLI from the downloaded file:

gemini skills install carbone.skill

Or install directly from GitHub:

gemini skills install https://github.com/carboneio/carbone-skill.git

To limit it to the current project only, add --scope workspace to either command:

gemini skills install carbone.skill --scope workspace

OpenAI Codex

Extract the skill into your personal skills folder:

unzip carbone.skill -d ~/.agents/skills/carbone

Vibe

Extract the skill into your personal skills folder so it is available across all your projects in the Vibe CLI (by Mistral):

unzip carbone.skill -d ~/.vibe/skills/carbone

To limit it to a single project instead, extract it into the project directory:

unzip carbone.skill -d .vibe/skills/carbone

Llama

Llama is a model, not an agent, so it has no skills system of its own — a skill is read by the agent or app around the model. For a direct Llama API integration, give it the Carbone syntax by prepending the skill's SKILL.md to your system prompt. If you run Llama inside an agent that implements the open Agent Skills standard, install the skill the way that agent documents.

DeepSeek

DeepSeek works the same way: there is no built-in skills system. For a direct API integration, prepend the skill's SKILL.md to your system prompt so DeepSeek always has the Carbone syntax. If you drive DeepSeek through an agent that implements the open Agent Skills standard, install the skill as that agent documents.

Using Skill + MCP Together

With both installed, you can go from data to a finished document in a single conversation.

Skill vs MCP in one sentence: The Skill gives the AI knowledge of Carbone syntax; the MCP server gives it actions (generate, convert, manage templates). Used together, the AI can design a correct template and render it in the same conversation.

Prompt to AI:

I have a customer record in data.json. Create an invoice template for it, then render it as PDF.

What happens:

  1. The AI uses the Skill to design a valid Carbone template with correct syntax
  2. The AI uses the MCP upload_template tool to store the template
  3. The AI uses the MCP render_document tool to generate the PDF

Without the Skill, the AI would need several render-and-fix cycles to get the syntax right. With it, the first render typically works.

Recommended workflow:

  1. Share a sample JSON with the AI so it understands your data structure
  2. Ask the AI to draft the template — the Skill handles syntax correctness
  3. Review the template before rendering
  4. Ask the AI to render with a test record and confirm the output looks right

Keeping the Skill Up to Date

Re-download and reinstall carbone.skill when Carbone ships major updates to stay in sync with new formatters, options, and syntax changes. Watch the carboneio/carbone-skill repository on GitHub to be notified of new releases.

See also