I’ve been using AI coding agents more and more, and one thing kept annoying me:
I was repeating the same instructions every session.
Things like:
- keep diffs small
- don’t fetch data directly inside React components
- follow the existing architecture
- add tests when behavior changes
- don’t introduce new patterns unless needed
- explain risky changes before making them
At first, I handled this with big instruction files like
AGENTS.md. That works for a while, but it has two problems: - The file keeps growing.
- Most of the instructions are irrelevant to the current task.
If I’m asking the agent to change a React component, it probably doesn’t need my rules about commit messages, backend handlers, database migrations, or release notes.
And if I work across multiple projects, I either copy those instructions around or rewrite them again.
So I built ai-rules.
## What ai-rules does
ai-rulesis a small local-first CLI for personal AI coding rules. Rules live as Markdown files in:
bash
~/.config/ai-rules/rules/
Top comments (0)