explore-codebase

Start with graph statistics and the architecture overview to see major modules, then use semantic search to locate specific functions. Trace callers and callees to understand how components connect before reading individual files.

Use the query_graph_tool with the callers_of pattern on the target function. You can also use callees_of to see what it invokes and imports_of to inspect its dependencies.

A code knowledge graph models files, functions, classes, and their relationships as queryable nodes and edges. It enables architecture overviews, community detection of modules, and execution flow analysis without reading every file.

Always begin with get_minimal_context for your task and request minimal detail levels on graph calls. Escalate to standard detail only when the minimal output is insufficient, keeping tasks within about five tool calls.

Graph exploration requires the codebase to be indexed into the knowledge graph first. For tiny scripts or single-file questions, directly reading the file is faster than running graph queries.