Three tools with READ_ONLY operations. All have descriptions (194-250 chars, within baseline 34-392 range) and structured input schemas with typed parameters. However, schemas lack critical details: no enums for constrained values, minimal parameter descriptions (10-30 chars vs baseline 72 chars), no output schemas documented, and no error handling guidance. Tool names lack action verbs (WorkspaceAttribution, BenchMemberFile, BenchResolutionHistory are noun-heavy). Parameters like 'source' enum is present but others lack constraints. No evidence of pagination limits, result caps, or chaining IDs in responses. Descriptions explain WHAT but not WHEN to use or prerequisites.

• Rename tools with action verbs: 'WorkspaceAttribution' → 'get_workspace_attribution', 'BenchMemberFile' → 'read_bench_member_file', 'BenchResolutionHistory' → 'list_bench_resolutions'. Verb-first names let LLMs infer intent without reading descriptions.
• Expand parameter descriptions to 50-100 chars. E.g., 'The file path to attribute (required)' → 'The file path to attribute, relative to the bench root. Must be a valid file path (e.g., src/main.ts). Required.'
• Document output schemas explicitly. E.g., WorkspaceAttribution returns {owner: string, lines: [{line: int, owner: string}]}. BenchMemberFile returns {content: string, exists: bool, isBinary: bool, truncated: bool}. Include field types and meanings.
• Add error handling guidance to descriptions. E.g., 'If the file does not exist, returns error 404 with message "File not found: <path>". Try list_bench_files() to discover available files.' This guides LLM recovery.
• Constrain numeric parameters. BenchResolutionHistory 'limit' should document 'Maximum number of entries to return (default: 20, range: 1-100). Larger limits may timeout.' Add min/max to schema if supported.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry