I use Claude Code on my laptop quite a bit, and in the past few months I've kept running into situations like this:

• I'm out and about, I've given Claude a complex task — solving a hard theory problem, writing a lot of code — and I'd rather put the laptop back in my backpack than sit there watching it, but closing the lid means stopping Claude mid-thought, and I'd rather have it running "in the background".

• I've just finished a backend change with Claude and now I want to deploy it to production on some VM. The deployment should really happen in the same conversation, so all the context about the change carries over — but it has to run on the VM, not my laptop.

• I ask Claude to kick off some beefy computation and realise halfway through that my laptop is the wrong machine for it, and the whole thing belongs on a bigger machine.

In all three cases what I actually want is to pick up the conversation on a different machine. That's what the teleport skill does. Suppose I'm on my laptop with SSH access to machines A and B. When I run /teleport A, for example, Claude:

1. opens an SSH connection to A and copies the current conversation over the tunnel, with an added message stating that the teleportation completed successfully and that the agent is now operating on machine A;
2. syncs the git repo in my current working directory with the one on A, cloning it first if it doesn't exist there yet;
3. starts a tmux session on A running Claude Code, resuming the conversation that was just copied over;
4. leaves a shell script on my laptop that, when run, SSHes into A, attaches to that tmux session, and drops me right back into the conversation.

The whole thing works in reverse too: running /teleport laptop from the remote session sends the conversation back home, so a chat can hop between machines as many times as it needs to.