Skills

Cartesi Skills are lightweight, specialized instructions for AI agents. Each skill covers a focused task: scaffolding a new app, building backend logic, wiring a frontend, interacting with L1 contracts, or deploying to a self-hosted node.

Cartesi Skills follow the open Agent Skills format. For a similar pattern in the broader Ethereum ecosystem, see eth-skills.

What's included

The current release ships 10 skills plus a workflow skill that routes you to the right skill for each phase:

• cartesi-workflow: map a full-stack application build and pick the next skill
• cartesi-scaffold: create a new Rollups v2 app with cartesi create
• cartesi-backend-core: implement advance/inspect handlers and the /finish loop
• cartesi-backend-js-ts: build a JavaScript or TypeScript backend
• cartesi-backend-py: build a Python backend
• cartesi-frontend: wire wallet, InputBox, JSON-RPC, and inspect in a UI
• cartesi-contracts: write Solidity/Foundry contracts that call InputBox and portals
• cartesi-local-dev: run cartesi build / cartesi run, test locally or on a fork
• cartesi-jsonrpc: query a running node over JSON-RPC
• cartesi-deploy: deploy and operate a self-hosted rollups node
• cartesi-debug: diagnose errors across the stack

Get started

Install into your project

The recommended way to use Cartesi Skills is to install them into your project so your assistant loads them from .agents/skills/ automatically.

From your project root, add the Cartesi skills package:

npx skills add Mugen-Builders/cartesi-skills

Confirm the skills appear under .agents/skills/ (one folder per skill, each with a SKILL.md). Restart your editor or start a new agent session so the client picks up the new files. In your next prompt, name the skill you need (for example cartesi-scaffold or cartesi-backend-js-ts); see Prompting for examples.

Watch the skills setup walkthrough for a full install-and-verify flow in Cursor or another Agent Skills-compatible client.

Use skills with MCP or on their own

Once installed, you can still use skills in two ways:

• Through the Cartesi MCP server: skills are bundled in the knowledge base and returned inline when your assistant queries the server, so you do not have to paste skill text into every chat.
• On their own: your assistant reads skill files directly from .agents/skills/ when you name them in a prompt. You do not need to connect to the MCP server in this case.

Browse and copy (optional)

If you prefer to inspect or copy a single skill without installing the full set, use skills.mugen.builders to browse each skill and paste the content into your agent context.

Source and contributions

The canonical source lives in the cartesi-skills repository on GitHub. Open an issue or pull request there if you spot gaps, want a new skill, or have fixes to existing instructions. Contributions are welcome.

Early release

Cartesi Skills v0.1.0 is an early release for testing and feedback. Be cautious when using private keys or mainnet credentials with AI-assisted workflows.