Scored across 12 tools
The three read paths (query_graph for discovery, inspect_symbol for known IDs, trace_path for multi-hop) are explicitly cross-differentiated in their descriptions, and mutate_graph vs add_planned_symbol are clearly separated (physical index vs virtual workspace). The remaining overlap is mostly between run_datalog and trace_path for transitive queries, but descriptions give enough guidance to avoid misselection.
All twelve tools follow a verb_noun pattern (trace_path, inspect_symbol, query_graph, reload_graph, validate_plan, create_planning_workspace), with no camelCase mixing. Minor inconsistency in the 'get_' prefix being used only for two summary/info tools while other reads use more specific verbs, but the convention remains readable and predictable.
Twelve tools sit squarely in the well-scoped range, and each covers a distinct phase: discovery, inspection, tracing, custom logic, architecture overview, diagnostics, mutation, planning, and verification. No tool feels redundant or padded.
The surface covers the full read/analyze/mutate/plan/verify lifecycle for a codebase graph, including diagnostics and planning workspaces. Gaps are minor: mutate_graph supports add_node/add_edge/remove_edge but no remove_node, and there is no tool to list, inspect, or discard existing planning workspaces, which could strand agents mid-plan.