Is there an elegant way to provide feedback and rechat on the wrong commands and results?

[demouo]

I apologize for my unfamiliarity with this excellent project shell_gpt. An AI provided a command that failed to execute. How can I share the incorrect context in my next conversation?

<logs> ~ sgpt --shell 'check which process using max memo'
ps -eo pid,comm,%mem | sort -nr | head -n 1 [E]xecute, [D]escribe, [A]bort: e zsh:1: command not found: 41219 </logs>

Then I want to tell ai that The command failed with error: zsh:1: command not found: 41219, please retry..

Is there an elegant way to provide feedback and rechat?

I would be very grateful for your reply.

[demouo]

I do think cli-ai chatting with the (last) stdout result of shell will be a great movement to cli-free.

I have two toy ideas:

1. Save the last command and results in some place, but now run_command() is using the os.system() to run a command, which can not catch the stdout of cli. So I make it change to subprocess.run(), but this will NOT pass the pytest of @patch(os.system). I'm not sure if I can refactor it.
2. Let user copy the stdout and chat with clipboard, which is not elegant enought.