rules-distill

Cross-cutting principles are distilled by scanning installed skills to identify recurring patterns, then using LLM judgment to propose appends, revisions, or new rule files with evidence drawn from multiple skills.

Tacit guidance is converted into explicit governance rules by running deterministic collection scripts on the skill inventory, then applying an LLM-driven judgment phase to ensure accuracy, coverage, and safety.

Inventory installed skills by executing scripts/scan-skills.sh and scripts/scan-rules.sh to gather skill signals, which are then analyzed to surface cross-cutting patterns for rule distillation.

Yes, jq is a required dependency for the skill distillation workflow, supporting the deterministic collection scripts that scan installed skills and rules before the LLM judgment phase.

Yes, you can propose appends, revisions, or new rule files based on evidence drawn from multiple skills, ensuring that cross-cutting principles are enforced consistently across the skill set.

LLM judgment ensures accuracy, coverage, and safety during rule distillation, but it operates on signals collected by deterministic scripts, meaning rule proposals depend on the quality and completeness of the scanned skill inventory.