AI/LLM stuck in agent mode

[shadow-absorber]

whenever shadow asks the llm/qwen3.5 9b to make a script the model decides to try and run it with bash or python multiple times over instead of just writing the script to chat... toggling the agent mode on and off does not seem to make a difference... neither does the /bash command... how would shadow fix this???

[Qbsoon]

This is reported as an issue and had a good PR made to fix it, idk though when and if it will be merged.
The general issue is not that the qwen is doing something wrong, but odysseus takes each command from model result and tries to run it without model saying so.

Sometimes it even gets funny when model thinks that you are running the code and tries to convince you to stop running it

[shadow-absorber]

can you link to the issue?? so shadow can keep track of it!

[Qbsoon]

#3604