I had a bit of inspiration and started working on adding code syntax highlighting to Tomo.
Seeing an increasing number of folks running coding agents like Claude Code in loop and with --dangerously-skip-permissions
.
You can accomplish a lot like this with the proper setup.
In my experience, the agent needs to be able to evaluate and test its own work to be successful with this approach.
I don’t think I would attempt this without using some kind of sandbox for the agent as exploits are becoming increasingly common and advanced.