Hey Umbracians, long time no see! I'm looking at the Umbraco source today (for reasons) and I noticed some of these Claude files are pretty big (

src/Umbraco.Infrastructure/CLAUDE.md

is ~8000 tokens). Has anyone else found this is a problem? Given that the root

CLAUDE.md

is about the same size that's a lot of tokens just to get started 👀. P.S. I'm surprised there isn't a Claude channel on the Discord but that's just me 😀.

The idea is that the root one fans out to the relevant sub-files only when it needs to - part of Claudes "Progressive Disclosure". https://docs.claude-mem.ai/progressive-disclosure So you should not expect it to load in all claude.md files throughout the codebase unless you actually need to touch on everything.

Hey Jemayn, thanks for your reply! You're right that these sub-folder claude files are only loaded when a relevant file is loaded but they're still much bigger than Anthropic recommend (638 and 838 lines against Anthropic's 200-line target) and I think their value is questionable. I'm investigating further but I think it would be more productive to use short claude files and pair them with something like [CodeGraph](https://github.com/colbymchenry/codegraph).