My Claude Code workflow usually involves... a lot of subagents. I make one Opus model coordinate work as my tech lead, it dispatches subagents using models appropriate for the task into their own git worktrees, and then I try to multitask and do something else while they run.
I'd been finding lately (particularly with Opus 5.5) that there was a ton of drift and instances of subagents not doing anything and it was evading the attention of the TL despite the instructions I gave it to have them report in at a regular cadence.
Running tracon in a split terminal is my new personal workflow. It gives me early indicators when things start going off the rails (ex. blocked on a prompt, idle, looping on a failing command, simultaneous edits to files other subagents are editing, drifting off the brief, etc.) and I can interrogate the subagent's workflow and by pressing n the main agent gets a note to make a call on how and if to correct it.
Read-only and runs on your machine: it tails transcripts in ~/.claude/projects/ and checks each worktree with ps, lsof, and git -- only thing that leaves your machine is what the Haiku judge reads for classification (or optionally you can turn that off with --no-judge).
I've found it super useful for the game I'm building and figured others in the community might also find it useful.
If you just want to give it a spin without committing:
uvx tracon --demo
macOS and Linux, Python 3.12+, MIT. (Not affiliated with Anthropic.) Comments URL: https://news.ycombinator.com/item?id=50046981
Points: 1