ContextWindowExceededError when single code_execution_tool output exceeds context limit (compression doesn't help)

[rabata1]

Hey,

I'm having an issue where the agent crashes when using code_execution_tool and the command output is too long.

For example, if I run a command like git submodule update or a build command that produces a lot of output, the agent immediately crashes with:

litellm.BadRequestError: ContextWindowExceededError

It seems like the output from the command is too large and exceeds the context window. The history compression doesn't seem to help because the single output is already too big before it even gets added to the history.

Is there a way to limit or handle large outputs from code_execution_tool? Any guidance would be appreciated!

Thanks

[trillox9]

+1

[23skdu]

i have sometimes had to tell the Agent to NOT use python code to write files, and just use the 'cat' command, as its far less tokens, if you just need to dump text to a file, things like that... add some of that to prompting helps, but i also have noticed if i am asking it to write code that manages or locks memory segements, the code execution tool with freeze, obvious, but using prompts to get agents to write unit tests first, catches that, and that is just good practice anyway. my .02 cents