Running 3-4 Claude Code / Codex sessions at once: what breaks first for you?
I've been running 3-4 Claude Code / Codex sessions at once across different projects, and my biggest time sink isn't the coding. It's losing track of which terminal is blocked on a permission prompt, and which spec each session was supposed to follow. Every time I need the related doc, I have to jump out to another folder, find the file, and then come back to the right terminal.
My current setup is a mix of git worktrees, a markdown file per task, and a lot of tab-checking. It works, but I still miss prompts for 10+ minutes sometimes.
How do you handle it? Do you use tmux, notifications, a task board, or just discipline? And what breaks first when you go from 2 agents to 5?
Disclosure: I'm building a small macOS tool in this space (PMDesk.dev), so I'm genuinely curious what workflows people already trust. Not linking anything here.
Replies
Tmux with a custom status script works here. Tmux can be configured to change the window tab color whenever a specific output, such, as a prompt requires your input.
@jean_dupont7 That's a neat trick, thanks Jean. Are you matching on specific strings in the pane output, like the permission prompt text? I'd worry about it breaking when the CLI wording changes. Curious how stable it's been for you across Claude Code updates.
For me the first thing that breaks isn't attention, it's the git tree. Several sessions in one repo means one agent's half finished change sits in the working tree while another one commits. What fixed it was a few rules in the project file every session reads: stage only your own files by name, commit by path, never stash or revert what you didn't write, and check git status for someone else's changes before a deploy. For the prompts, I gave routine things like build, test and deploy standing permission and kept the questions for what's hard to undo, which cut the waiting a lot.
Keeping each session in a separate worktree and using tmux makes a big difference when juggling multiple agents
@shivam_kushwaha16 Thanks Shivam! Worktree per session is basically what I do too, and it fixes the file collisions. Where I still lose time is tmux: do you get any signal when a pane is blocked on a permission prompt, or do you just cycle through windows to check? That's the part I haven't solved cleanly.
An agent lost earlier context mid-task and rebuilt a helper that already existed two files away under a different name. It solved the immediate prompt cleanly, so the duplicate logic slipped in until edge cases diverged.
I run my parallel sessions in Ultenix, a terminal I built for exactly this. Checking is done by agents through end-to-end tests, and when I check something myself I use Firefox. Long-running tasks go into my own task tracker, and the progress of every project is listed on a grid board in Pinateca, so I can see where everything stands at a glance.