Never Teach Your AI Agent 'apt install' Again: The xshellz.box Manifest

You hand an AI coding agent a Linux box. It tries to run rg or node and nothing is there. It wastes time installing packages, or it fails. With xshellz.box, you give it a shopping list. Every new box arrives with the tools you asked for.

The 30-second fix

Connect to your xshellz.box and create a file in your home directory called xshellz.box. Each line names one package. A bare name installs the matching apt package. You can also use apt:, pip:, or npm: prefixes.

# My agent's toolchain
node
ripgrep
python3
pip:requests
npm:typescript

Reboot the box (or just wait for the next boot) and those packages are installed automatically. The file itself lives in /home, which is the only part of the filesystem that survives a reboot, so your manifest sticks around forever.

Now when you SSH in and run an agent like borg, the tools are already there.

ssh -p 1234 root@mybox
borg "find all TODO comments in the repo"

The agent can use ripgrep without asking for apt install. It can also use Node, Python, and the libraries you declared. Every box with that manifest is identical.

Why this matters for AI agents

Agents are unpredictable if their environment changes. A missing tool often means a broken task. The manifest turns the box into a reproducible workspace. You can version it in your dotfiles, share it with your team, and never again debug why claude or codex failed on a clean box.

If you push your manifest to a box and then wire it into Claude Code as an MCP server, the agent always finds the same toolchain. No more sudo apt-get install in the middle of a session.

The catch

The manifest only covers what a package manager can give you. It won't configure kernel parameters, set up systemd services, or install things from source. But for the vast majority of dev tooling, it's enough to make an AI agent feel at home on every box.