Kb Retriever

A local knowledge base retriever that navigates multi-level directory indexes before searching actual files. It reads data_structure.md files at each level to understand what's where, then uses targeted grep and partial file reads instead of loading everything into memory. The interesting part is the mandatory learning step: when it hits a PDF or Excel file, it must first read reference docs to learn the right extraction methods (pdftotext, pdfplumber, pandas) before proceeding. Supports iterative keyword searches with up to five refinement rounds, and it won't hallucinate answers when information is missing. Assumes your knowledge base lives in a knowledge/ directory by default, but you can point it elsewhere.

Install to Claude Code

npx -y skills add conardli/garden-skills --skill kb-retriever --agent claude-code