rules-distill

To extract reusable rules, run deterministic inventory scripts to collect installed skills and existing rule files into a complete text corpus, then use LLM cross-reading to cluster related skills and generate candidate principles for explicit user approval.

Automating governance rule distillation involves scanning installed skills and rule files to identify cross-cutting principles, assigning verdicts like Append or Revise, and saving structured candidate output with draft text and violation risks to results.json.

Yes, bash and jq are required dependencies because the deterministic inventory uses POSIX shell scripts emitting JSON to enumerate installed skills, collect full texts, and produce the structured results.json file.

LLM cross-reading evaluates evidence and violation risk by clustering related skills to extract candidate principles, then assigning verdicts such as Append, Revise, New Section, or New File to produce structured candidate output with draft text.

No, you cannot automatically apply rule changes because the workflow uses a human-in-the-loop execution model that presents structured candidates with draft text and requires explicit user approval before modifying any rules.

You should run a rule distillation workflow during monthly maintenance or after onboarding a new set of skills to surface repeatable best practices and keep your global rule documentation current and consistent.