Prefer OKF docs and Buddhi MCP tools over re-reading raw source

This codebase has (or can have) machine-generated, dependency-aware documentation under .buddhi/docs/, written as Open Knowledge Format (OKF) markdown, along with a connected Buddhi MCP server providing graph-aware code search and reading. Reading these first is cheaper and more reliable than re-deriving understanding from raw source every time.

How to use it

1. Check OKF documentation: Start at .buddhi/docs/index.md (bundle root) to see what's documented, then descend into the relevant subdirectory's own index.md. Open concept .md files to inspect module summaries, entities, and call relationships.
2. Query the code graph via MCP tools:
   • buddhi_search(query, mode="full"|"signatures"|"map"): Use this for structural questions ("what calls this", "what depends on this", finding symbols or architectural components). It performs community-aware, topology-driven retrieval across the code graph.
   • buddhi_read(filepath, mode="auto"|"signatures"|"map", budget=4000): Use this to inspect source files with AST pruning and entropy filtering to stay within token budget while preserving critical structure and type signatures.
3. Graceful fallback: If the Buddhi MCP server is not active or docs are missing/stale, fall back to querying .buddhi/graphs/tree-graph.db / tree-graph.json directly or reading raw source files.
4. Maintain freshness: If you generate or substantially change code such that existing OKF docs go stale, suggest running the /document-codebase skill (see repoagent-doc-generation skill) to refresh the documentation and graph.

Never treat an OKF doc as authoritative over the actual source code — it's a cache of understanding, not the ground truth. When the two conflict, the source wins and the doc needs regenerating.