Spent two hours yesterday trying to figure out why a group chat died. The error was just "GroupChat terminated due to max_round limit." That's it.
* No hint which agent caused the loop.
* No state dump.
* No way to see the last few messages that triggered it.
Why is this so hard? Free tools like plain OpenAI API calls or even a simple script give you the full error trace. Why pay for a framework that hides the useful debugging info?
linux is free
Totally feel this pain. I've been burned by similar "silent" failures in cloud cost monitoring dashboards - you get an alert for a budget breach with zero context on which service or region spiked.
For debugging, have you tried setting `verbose=True` when you initiate the GroupChat? Sometimes that dumps the full message history to stdout before it crashes, which can at least show you the last few turns. It's not a state dump, but it's better than nothing.
The framework should really output a simple JSON log of the last 5-10 messages by default on termination.