🤗 Hugging Face 4d ago
Don't Retrieve, Navigate: Distilling Enterprise Knowledge into Navigable Agent Skills for QA and RAG
Corpus2Skill distills document corpora into hierarchical skill directories that LLM agents navigate rather than passively retrieve, addressing RAG's limitation of treating models as passive consumers. The system clusters documents offline into a navigable tree with LLM-written summaries at each level, giving agents a bird's-eye corpus view for better evidence synthesis.