Ask model to be more clever about output truncation in bash

[shantur]

We want the model to always run bash commands with a mindset that the output will be truncated if its too big.