Hort — DevOps skill for Claude Code

Give your coding agents a place to work that isn't your machine.

[!WARNING] hort is in early development. The design is settled; the code is being written. What follows is the tool we're building toward, not a finished product.

The problem

Coding agents are finally good enough to leave alone. You hand one a task, it edits files, runs the tests, fixes what broke, and reports back. The obvious next step is to take your hands off the keyboard and let it work.

The catch is that "let it work" means giving a non-deterministic process the same reach you have: your SSH keys, your whole home directory, any command it decides to run. Most of the time that's fine. The problem is the one run where a command is wrong, a loop never ends, or a repo you didn't fully trust does something you never asked for. You only have to lose your machine once to stop taking that bet.

The usual fix is a dev container, which works but is built around an IDE. A lot of people don't work that way. They live in a terminal, often over SSH on a Linux server where there's no IDE at all and the agent runs as root. That's where the risk is highest and where an IDE-shaped sandbox fits worst.

hort fills that gap: real isolation that keeps the environment you're fast in, driven entirely from the command line.

The idea

Almost everything in this space ties the sandbox to the agent. You run the tool, i