Scoring was not performed

• Document the output schema for each tool. Specify field names, types, and what IDs are returned for chaining (e.g., search returns [{ id: string, title: string, similarity: float, ... }]).
• Add min/max constraints to numeric parameters. E.g., 'limit: integer (1-100, default 20)', 'offset: integer (≥0)', 'num_clusters: integer (2-50)', 'expand_level: integer (0-3)'.
• Clarify the 'filters' parameter in search. Document the filter DSL: allowed fields, operators (=, !=, <, >, contains, etc.), and example syntax. Or provide separate parameters for common filters.
• Expand tool descriptions to 100-200 chars with operational clarity. Include 'When to use' guidance that differentiates similar tools. E.g., 'Use search for keyword/semantic queries. Use hierarchy_search if you need to navigate document structures with parent-child relationships.'
• Add pagination documentation. State whether results are capped (e.g., max 50), whether a next_cursor is returned, and how to fetch additional pages.
• Document error conditions for each tool. State what errors are possible (not found, invalid query, timeout, etc.), whether they are retryable, and what the LLM should do (try alternative tool, ask user for clarification, etc.).
• Rename vague tools for clarity. 'analyze_relationships' → 'list_document_relationships' or 'rank_document_relationships'. 'detect_document_conflicts' → 'find_conflicting_documents'. 'find_complementary_content' → 'search_supplementary_documents'.
• For 'expand_document' and 'expand_cluster', explicitly define expand_level semantics. E.g., 'expand_level: 0=direct references only, 1=one hop, 2=two hops, etc.'