Skip to main content

Custom Rules

"How do I make the AI follow my conventions without repeating them every session?"

A rule is a behavioral instruction for your AI assistant. A trigger is the situation where it fires. Together, they let you encode your working conventions so the AI follows them automatically — without you repeating them every session.

What a rule looks like​

Rules live in rules.md (personal) or knowledge/rules.md (project). Each rule is a plain-language instruction:

### Always run tests before pushing
Run the full test suite before any `git push`. If tests fail, fix them first — do not push with failures.

What a trigger looks like​

Triggers live in triggers.md. Each trigger names the rule and declares when it applies:

## Before pushing to origin
- Apply: `rules.md § Always run tests before pushing`

The trigger fires when the phase matches — before a push, at session start, before creating a PR, etc.

Rules vs. identity​

Your identity file (me.md) describes who you are — your expertise, preferences, and working style. Rules describe how the AI should behave in specific situations.

  • Use me.md for: "I prefer explicit over clever" or "I've been using Go for ten years."
  • Use rules.md for: "Always write a failing test before implementing" or "Never force-push to main."

Writing a good rule​

Good rules are:

  • Specific — "run npm test" beats "run tests"
  • Actionable — the AI can follow them without judgment calls
  • Bounded — rules with triggers are more reliable than unconditional ones

Capturing rules​

Run /kmgraph:rules-capture to add a new rule. The skill prompts you for the rule text and whether it applies in specific situations — then routes it to the right file automatically.