It seems logical, doesn't it?
I use a common git submodule with per-language instructions in separate markdown files
Then I instruct the LLM I’m using to read through those in addition to the code and create its own rule file, CLAUDE.md, GEMINI.md whatever.
Sometimes I tell it to refer to existing rules from other LLMs.
I want to open chat window in VSCode and start to work. And my expectation, for 7-8 from 10 cases it will be work good.
So, it why I was setup instructions directly for each repo.
And second point: re-use it for other team mates.
For that two goals - iteration with one instruction file looks better.
But, also I think what will be good to have a start point for instructions. For case of creating a new repo (for example)
Will be interesting to look on your basic rules, if you can share it.
I use a common git submodule with per-language instructions in separate markdown files
Then I instruct the LLM I’m using to read through those in addition to the code and create its own rule file, CLAUDE.md, GEMINI.md whatever.
Sometimes I tell it to refer to existing rules from other LLMs.
I want to open chat window in VSCode and start to work. And my expectation, for 7-8 from 10 cases it will be work good.
So, it why I was setup instructions directly for each repo.
And second point: re-use it for other team mates.
For that two goals - iteration with one instruction file looks better.
But, also I think what will be good to have a start point for instructions. For case of creating a new repo (for example)
Will be interesting to look on your basic rules, if you can share it.