What terminal do you run your AI agents in, and why that one?
by•
I run a dozen or so Claude Code sessions at once, one per project, and the terminal ended up mattering more than I expected. Most of my attention goes to working out which session is waiting on me.
So I'm curious what everyone else runs their agents in. iTerm2, Ghostty, Warp, the one built into your editor, tmux?
And why that one? Was there a feature you actually needed, or were you already in it?
61 views
Replies
I stick to basic tmux. Fancy wrappers hide real exit codes. Agent once reported build "succeeded" while swallowing fatal error underneath. Problem is not agent being wrong, it is being told everything is fine)
@konstantin_tikhaev Yeah, a green report sitting on top of a failed build is the worst kind. On my side the fix was not letting the agent be the one who says it's done. It doesn't report back until lint, build and tests pass and it's opened the running app itself.
Did the exit code give it away, or did you only find out later?
@hisashispace found out later when app failed to start. wrapper trapped the crash, returned zero to agent, checking raw logs was only "giveaway")
@konstantin_tikhaev Ouch, a crash that exits zero. At that point the raw logs are about the only thing left you can trust. Thanks for coming back with the details.
@hisashispace Fewer layers between agent and actual shell means fewer swallowed errors you have to hunt down later)