customaize-agent:create-rule

Agent rules prevent repeats by using contrastive Incorrect/Correct examples to codify common mistakes and their correct patterns. This calibrates agent behavior by explicitly showing the incorrect pattern alongside a corrected pattern for the same task.

YAML frontmatter defines metadata at the top of a SKILL.md rule file by specifying a name and description. This structured header contextualizes the rule before the body applies the contrastive Incorrect/Correct examples that guide agent behavior.

Path-scoped rules apply agent guidelines to specific directories instead of globally. By configuring the rule context to target a path rather than the whole project, the agent receives tailored behavioral constraints only when working within that scope.

Contrastive examples calibrate agent behavior across tasks by providing an Incorrect pattern followed by a Correct pattern within the rule body. This contrastive structure enforces consistent behavioral guidelines regardless of the specific task being executed.

Creating modular rules for Claude agents involves building individual files in the rules directory with YAML frontmatter. Each rule file targets a specific behavior using contrastive Incorrect/Correct examples, keeping guidelines modular and maintainable across the project.