When your agent instruction files balloon past readable size or you're about to restructure your rules, this three-stage pipeline extracts imperatives into JSONL, composes them with policy algebra into Starlark, then renders decision trees as Mermaid diagrams. Everything lands in ai-workspace/research with date-stamped artifacts. The real value is the pause it forces between extraction and composition: you get to review what rules actually exist before doing anything with them. It's designed for AGENTS.md and rules/ directories specifically, and you can run stages independently if something breaks or you just want the extraction without the full pipeline.
npx skills add https://github.com/camacho/ai-skills --skill distill