Bug exist When Agent run ShellScript

Bug Reports
1

I’ve encountered two critical bugs:

  1. After the agent executes a script and completes the task, it fails to automatically detect completion. Therefore, I have to manually press the Cancel button to proceed. This behavior is highly unstable, and it seems that manually using the Cancel button or popup terminal leads to different reactions or behaviors from the agent compared to the standard automated execution method.

  2. After the agent executes a script, it doesn’t display any output. Although it seems that execution itself is successful, the output isn’t shown. For instance, if I check via the popup terminal, commands that usually take 10-20 seconds are already completed. Occasionally, even simple commands like ls or mkdir don’t trigger any response.

I’m using macOS with zsh. I’ve consistently experienced these issues. However, notably, these issues don’t occur at all when using the Claude Sonnet 3.7 MAX model. Why is there such a difference in behavior between the two models?

2

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.