LLM rules/instructions #49
Description
As we're experimenting with various AI code editors, it would be useful to share and consider committing rules to guide responses from the models. These are things like:
.cursor/ruleshttps://docs.cursor.com/context/rulesclaude.mdhttps://www.anthropic.com/engineering/claude-code-best-practices- Copilot repository instructions
I've found these do help, particularly to stop the models doing stupid things like making enormous commit messages or blinding running git add . For example Cursor has scoped rules which the model can choose to pull in as necessary ("when making a commit", or "if the user asks you to write or run a test", or "when you are asked to manage dependencies"). Sometimes it ignores them, but it's relatively rare and you can always the rule you want to pull in.
My guess is that sooner or later the community will standardize, rather than having files in different places, but for now we might want to at least set up the Copilot one for PRs (.github/copilot-instructions.md).