knowledge-graph

To query a shared knowledge base, this skill indexes markdown files into a deterministic node graph and retrieves relevant operational guidance using keyword overlap scoring and thresholding. It surfaces the most relevant sections without loading every source file.

Retrieving implementation patterns from distributed markdown notes is handled by building a deterministic node graph and applying keyword scoring with type filters. The skill returns the top matching excerpts so you can bypass manual searches across source files.

Yes, you can reinforce nodes using the knowledge write-back feature, which appends new information to the existing graph. This ensures that frequently used domain notes and implementation patterns become easier to surface over time.

Yes, you need jq and python3 installed in your bash shell environment to run the ingestion and query scripts. These dependencies are required for deterministic graph processing, staleness checks, and keyword scoring operations.

Keyword scoring works by measuring keyword overlap between your query and indexed markdown sections, returning matches above a specific threshold. Staleness checks evaluate the age of the knowledge base nodes to ensure operational guidance remains current.

Yes, debug scoring is supported during semantic-style retrieval to help you inspect how keyword overlap and thresholding values are calculated. This allows you to evaluate why specific markdown sections surface over others for your query.