Lesson 11: Local Skill Management with ms

skills:

• ms
• agents
• mcp

What are Skills?

AI agents like Claude Code are powerful, but they don't know your specific project conventions or unique workflows out of the box.

Skills are reusable prompt templates that give agents new capabilities. They are like "plugins" for your AI.

We use meta_skill (ms) to manage these skills locally. It's a local-first knowledge management tool with hybrid semantic search (BM25 + hash embeddings) and Git-backed audit trails. Your skills live in your project repo, not in the cloud.

Checking Installation

First, let's make sure ms is installed and healthy.

Run the version check:

ms --version

Then run the doctor check to verify everything is configured correctly:

ms doctor

Initializing a Project

To use skills in a project, you initialize ms. This creates a .ms/ directory where your project-specific skills will live.

Try initializing in your current directory:

ms init

You should see a message confirming that .ms/config.toml was created.

Discovery & Suggestions

ms uses a smart search algorithm (Thompson sampling) to suggest the most effective skills for a given task.

Imagine you want to write a test. Ask ms for suggestions:

ms suggest "write a unit test"

If you had skills installed, it would show you the best matching ones.

You can also list all available skills:

ms list

Creating a Skill

Let's create a simple skill. Skills are just Markdown files with some YAML metadata.

Create a new skill called "hello":

ms create "hello"

This will open your editor. You can define what the skill does (e.g., "Say hello to the user in a pirate voice").

The MCP Connection

The magic happens when you connect ms to Claude Code via the Model Context Protocol (MCP).

ACFS sets this up automatically! When you run Claude Code (via cc), it can "see" your ms skills as native tools.

If you ever need to start the MCP server manually (rare), you can run:

ms mcp serve

(Press Ctrl+C to stop it if you run it)

Summary

You've learned:

1. ms manages AI skills locally in your project.
2. ms init sets up a project.
3. ms suggest finds the right skill for the job.
4. MCP connects these skills to Claude Code automatically.

Next, we'll look at how to find remote skills from the community using jfp.