A tidier way to "put away" Mini Me in your development projects

With the new v1.0.7 and v1.0.8 releases of Mini Me, I made a few changes to the mm init and mm sbxinit commands (and to how Mini Me starts with mm or mm tui), and I also added a new command, mm skillcreate.

It was already possible to choose where you wanted to "drop" the agent's configuration. So I took advantage of this feature to improve mm init and the way Mini Me starts, and to make sbx easier to use.

Project structure

When I work on a project and use Mini Me on it, I like everything to be neatly put away, and I organize my agent's files like this:

├── .mm
│   ├── agent.yaml
│   ├── sessions
│   └── skills
│       ├── greetings
│       │   └── SKILL.md
│       ├── vulcan-salute
│       │   └── SKILL.md
│       └── what-time-is-it
│           └── SKILL.md
└── README.md

Starting mm tui

From now on, if you have a .mm folder with your agent's configuration, the mm tui command will detect it automatically.

mm init and project mode

And now, when you want to initialize an agent configuration, the creation wizard offers you a new Project option that generates the right structure for you:

While I was at it, I added a Sessions option and a Skills option.

Initialize a skill with mm skillcreate

In your folder, if you run mm skillcreate, you start a new wizard:

You set the name of the skill:

Then its description (which will help the agent pick it):

As well as some keywords:

Then its content:

And at the end, the skill is generated in the right folder:

.
└── .mm
    ├── agent.yaml
    └── skills
        └── hello-world
            └── SKILL.md

Start your new agent

In your folder, run mm tui: your skill is now recognized and you can run it:

So it gets easier and easier to initialize a Mini Me agent. But it's even better if you can do it securely.

mm sbxinit with a .mm folder

Initialize the sbx configuration

If you run mm sbxinit in a project that has a .mm folder, the command will detect the .mm folder automatically and generate, at the root of the project, a sbxenv.yaml file with content like this:

schemaVersion: "1"

name: bob-sbx

agent: mini-me-kit

kits:
  - docker.io/k33g/mini-me-kit:0.1.6

workspace: .

additionalWorkspaces:
  - path: ./.mm
    readOnly: true
  - path: ./.mm/sessions    # sessions.dir

sandboxOptions:
  pullPolicy: always

• The latest version of the mini-me-kit kit is detected automatically
• additionalWorkspaces makes the agent's configuration and its skills read-only, to protect them. Except for the session data.

Start the sandbox

All you have to do is run the following command to start the sandbox with your agent:

sbx env run --force

To remove it, just run (still in the folder):

sbx env rm --force

You can also use it with Zed Editor.

Using the sandbox with Zed Editor

Run the zed . command in the folder to open the editor on your project.

In a Zed terminal, create the sandbox with this command:

sbx env create --force

Then add this to the settings.json file:

  "agent_servers": {
      "🐱 mini bob (sandbox)": {
        "type": "custom",
        "command": "sbx",
        "args": [
          "exec", "-i", "bob-sbx",
          "mm", "-acp",
        ]
      },
  }

And now you can connect, with ACP (Agent Client Protocol), to the Mini Me that lives in the sandbox:

That's all for today. I hope these new features will make your life easier and make you want to use Mini Me. See you very soon for the next blog post.