Loading ToolBench
Graph-powered code intelligence for Claude Code. Semantic search + knowledge graph for better AI code understanding.
CodeSeeker provides 3 tools with visible schemas and descriptions, but quality is mixed. Tool naming violates verb-first conventions (semantic-graph, context-optimizer, database-schema-analysis are noun-heavy). Descriptions are present but generic and lack WHEN/WHY guidance. Parameter descriptions exist but are minimal. Output schemas are not documented. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Error handling, security considerations, and recovery guidance are absent from visible code. The tools appear to read code projects, but the interface does not guide LLM selection or composition effectively.
Intelligent context optimization for token-efficient Claude Code integration
Database Schema Analysis Tool - Provides structured database context to Claude Code CLI with schema discovery, relationship mapping, query pattern analysis, and performance optimization suggestions
Core semantic graph analysis providing deep code relationship understanding
Tool names violate verb-first convention. 'semantic-graph', 'context-optimizer', and 'database-schema-analysis' are noun-based and do not indicate action. LLMs expect 'analyze_semantic_graph', 'optimize_context', 'analyze_database_schema' to infer intent from names alone.
Descriptions lack explicit WHEN/WHY guidance. 'Core semantic graph analysis providing deep code relationship understanding' does not tell an LLM when to call this tool vs context-optimizer or database-schema-analysis. No context selection criteria provided.
Output schemas are not documented in tool definitions. LLMs cannot plan downstream calls or understand what data these tools return. Visible schemas only show input parameters.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 44 | - | v1 |
No tool annotations present. Tools marked as READ_ONLY but readOnlyHint metadata not visible in schema. No destructiveHint or idempotentHint attributes. Spec 2026-07-28 expects tool annotations to guide agent reasoning.
Parameter descriptions are minimal and lack format/constraint guidance. E.g., 'depth' has description 'Depth of relationship traversal (default 2)' but no min/max bounds specified. 'strategy' enum values listed in description as text, not as formal enum in schema.
No error handling or recovery guidance visible. Tools do not specify what happens on missing projectPath, invalid projectId, or analysis failure. No actionable error messages or retry strategies documented.
Tool composition unclear. Three separate tools (semantic-graph, context-optimizer, database-schema-analysis) operate on the same 'projectPath' and 'projectId' parameters. No guidance on which to call first, how results chain, or what information each produces.
Parameter 'focusArea' appears in semantic-graph and context-optimizer but type is string with no enum or constraints. LLMs will hallucinate invalid values. E.g., does 'api-layer', 'data-layer', 'business-logic' work, or only specific terms?
No documentation of pagination, result limits, or response size. context-optimizer has 'tokenBudget' (default 8000) but unclear how context is truncated or what is returned if exceeding budget.
database-schema-analysis tool lacks parameters compared to peers. Only 'projectPath' and 'projectId' provided, no depth control, focus area, or filtering like semantic-graph. Inconsistent interface design.