this sounds like a great tool to add to the toolbelt, as part of the "how do we handle all the code output from LLMs" problem