rules-distill

To extract cross-cutting principles, scan all SKILL.md files in the repository and apply cross-reading to identify recurring patterns. Map these principles to existing rule files or propose new ones, ensuring each candidate is supported by evidence from 2+ skills.

The best way to consolidate LLM governance rules is to distill recurring patterns from multiple skills into concrete, actionable rule files. This avoids abstract concepts and duplication, ensuring each proposed rule is backed by evidence from at least two skills to maintain consistency and reduce drift.

Maintain a living ruleset by running a distillation workflow that extracts recurring patterns from newly added skills and converts them into rule changes. This process proposes new rule files or amendments to existing ones, reducing duplication and preventing configuration drift over time.

Yes, you can use cross-reading to map extracted principles to existing rule files and propose amendments. The process explicitly checks for duplication, ensuring no overlapping rules are created and favoring concrete, actionable guidance over abstract concepts to enhance governance.

No specific dependencies or component frameworks are required. The process operates directly by scanning all SKILL.md files in the repository environment to identify cross-cutting principles and generate governance rules without relying on external libraries or tooling.

You should avoid automated pattern distillation when you cannot gather evidence from at least 2+ skills, or when the extracted insights are abstract concepts rather than concrete, actionable guidance that directly enhances governance and consistency across your workflow.