call_graph

Transitive call graph for a symbol. Two directions:

• callers (default): who calls this symbol, transitively. Use for blast-radius before refactoring.
• callees: what does this symbol call, transitively. Use to trace flow.
• both: both, returned in separate keys.

Schema

{ "symbol": "Agent::new", "direction": "callers", "max_depth": 3 }

Output

By default returns a compact summary: counts + by_file + by_depth. When the total result has ≤ 30 edges, auto-promotes to full edge list. Use detail_level: "full" to force full output on large graphs.

Each edge is tagged source: "lsp" (from callHierarchy, semantically authoritative) or "ts" (from the tree-sitter classifier fallback — best-effort, may include false positives on macros, shadowed names, or dynamic dispatch).

Caching

Edges are cached in the project sqlite DB (call_edges table). Caches are invalidated per file on did_change notifications (triggered by write tools), so the cache stays correct across multi-session edits.

Known limitations

• Kotlin LSP coverage of callHierarchy is partial; expect source: "ts" edges for most Kotlin queries.
• Cross-project edges are not supported in v1.
• direction="callees" via tree-sitter fallback is not supported; a RecoverableError is returned for languages where LSP callHierarchy is unavailable. See docs/TODO-tool-misbehaviors.md (LIMIT-001).
• Arrow function names in TypeScript/JavaScript show as <anonymous> in tree-sitter fallback mode.