Motivation
codeseek currently provides codeseek_callers (direct callers of a symbol) and codeseek_callgraph (local call graph neighborhood). But it lacks transitive impact analysis — answering questions like:
- "If I change the return type of
parse_config(), what downstream code breaks?"
- "If I delete
DatabasePool::acquire(), what's the blast radius?"
- "What's the dependency path from
handle_request() to execute_query()?"
These are the most valuable questions during refactoring, and right now users must manually chain codeseek_callers N times.
Proposed tools
codeseek_impact — transitive reverse reachability
codeseek_impact(symbol: str, max_depth: int = 5) → list of all symbols that transitively depend on `symbol`
codeseek_path — shortest call path between two symbols
codeseek_path(from: str, to: str) → list of edges showing the call chain
Implementation note
The call graph data already exists from tree-sitter parsing during indexing. These queries are pure graph traversal (BFS/DFS) on the existing edges — no additional model calls needed. The LanceDB index could store edges in a separate table for efficient graph queries.
This would make codeseek a complete replacement for tools like CodeGraph/GitNexus, which already provide this functionality.
Motivation
codeseek currently provides
codeseek_callers(direct callers of a symbol) andcodeseek_callgraph(local call graph neighborhood). But it lacks transitive impact analysis — answering questions like:parse_config(), what downstream code breaks?"DatabasePool::acquire(), what's the blast radius?"handle_request()toexecute_query()?"These are the most valuable questions during refactoring, and right now users must manually chain
codeseek_callersN times.Proposed tools
codeseek_impact— transitive reverse reachabilitycodeseek_path— shortest call path between two symbolsImplementation note
The call graph data already exists from tree-sitter parsing during indexing. These queries are pure graph traversal (BFS/DFS) on the existing edges — no additional model calls needed. The LanceDB index could store edges in a separate table for efficient graph queries.
This would make codeseek a complete replacement for tools like CodeGraph/GitNexus, which already provide this functionality.