Stop Babysitting Your AI's Toolbox: 1 File to Identical Dev Environments Everywhere

Your AI coding agent is like a chef: it can whip up a meal in any kitchen, but only if the right knives and spices are already there. When you hand your agent a task on a new machine and the tool it needs is missing, it wastes turns installing things or just fails. That breaks the flow and kills the "just works" feeling.

xshellz.box gives you a dead simple way to stamp out identical, pre-wired developer environments across any sandbox. Instead of configuring each box by hand, you write a single manifest file. Every box you connect to, whether an always-on permanent box or an ephemeral one spun up for a single job, boots with exactly those tools. Your agent wakes up in the same kitchen every time.

The one-file manifest

Inside your box, the file $HOME/xshellz.box declares every extra package you need beyond the default toolchain (git, Node 22, Python 3, ripgrep, tmux, build-essential, and the agents themselves). Each line is one package; prefix with apt: for system packages, pip: for Python, or npm: for Node modules. On every boot the box reinstalls from scratch, so the environment is always pristine.

# SSH into your box (no key needed for the web terminal)
# Then write your reproducible manifest
echo 'apt:python3-venv'  >> ~/xshellz.box
echo 'pip:black'         >> ~/xshellz.box
echo 'npm:prettier'      >> ~/xshellz.box

# Apply the changes by restarting the box
sudo reboot

After the box comes back, connect again and every listed package will be present. That is the whole trick: no Dockerfiles, no Ansible playbooks, just a flat text file that lives in your home directory.

Waking up borg in a familiar home

Once the manifest is in place, any AI agent that runs inside the box sees a fully stocked workspace. You can use borg to learn the repository and immediately start giving it commands that need those tools.

# Let borg study the repo (it writes a BORG.md context file)
borg learn

# Now hand it a task that relies on the tools you declared
borg "format all Python files with black and check them"

Because black is already installed via the manifest, borg does not waste turns on setup. It goes straight to work. The same holds for any other agent you run in the box: Claude Code, Codex CLI, or Gemini CLI. Even when you wire the box into Claude Code as an MCP server (claude mcp add xshellz -- ssh -p <port> agent@<host> xshellz-mcp), the manifest guarantees that the server-side environment matches what your MCP prompts expect.

The real payoff comes when you spin up a throwaway sandbox with the same manifest. Your agent lands in an identical clone: same linters, same formatters, same Python version quirks. No more "works on my machine", the machine is always the same.

This pattern shines for CI-style agent tasks, cross-laptop workflows, and onboarding teammates. The limit is that only /home survives a reboot; anything the agent writes elsewhere vanishes, so treat the box as a compute environment, not a persistent server.