I noticed with post-edit hooks that Claude would do things like import a Python module in one edit then add code to use it in another. The lint hook would fire after the first edit and complain about unused imports and it seems to confuse or slow down Claude. It got there in the end but with some unnecessery drama polluting the context.
Some advice suggests to use "stop" hooks so Claude would lint after it had finished its current task, but apparently by that point it can lose the context of which file it changed. Possibly not a problem if you lint everything in a small project.
I also see potential for running automated fixes without bothering Claude about it. Though again there's potential for the llm to get confused by things happening "behind it's back". I've seen edits get redone if the underlying file has changed.
One final idea, fixing lint issues feels like something that could be done by a cheaper sub agent without bothering the main model for small things, possibly even more so with this kind of prompt based feedback.
I'd assume that this will be a bit like JSON schemas - the decoding will eventually get smart enough to validate in the output in line with more complex rules.
Agree on the "behind it's back" too. I might make a change that in the case of "--fix" give the LLM the diff on the spot.
The other advantage I've not been able to quantify yet - I've been able to *remove* stuff from CLAUDE.md/etc in favor of lint rules. e.g. prefer ?? over || -- all the way through to "use our logging framework" -- as a lot of the nitpicks in my instructions were bits like this. This keeps the instructions to higher level architectural stuff.
