Strong foundation with 13 well-defined tools across Hive, Impala, and Spark. All tools have clear, action-verb names (list_, describe_, select, submit_) and detailed descriptions (100-200 chars) that explain WHAT, WHEN, and constraints. Input schemas are complete with proper JSON Schema types, minLength/maxLength constraints, and required field declarations. Risk annotations (READ_ONLY, WRITE) are explicit. However, parameter descriptions are absent from schemas, LLMs see only type info, not semantic meaning. Output schemas are undocumented. Error handling guidance is minimal. Security considerations (Knox token, Ranger authorization) are mentioned in descriptions but not formalized as permission gates or scope declarations.

• Add parameter descriptions to all JSON Schema definitions. For each parameter (database, table, columns, batch_id, file, limit, etc.), include a 'description' field explaining its purpose, valid format, and constraints. Example: 'database': {'type': 'string', 'minLength': 1, 'description': 'Hive database name (case-sensitive, must exist in Ranger policy)'}.
• Document output schemas for all tools. Specify the structure of successful responses (e.g., hive_select returns {rows: [{col1: value, col2: value}], total_rows: int}, hive_list_databases returns {databases: [string]}). Include field types and whether results are paginated.
• Add pagination support to list tools. Introduce optional 'limit' (default 20, max 100) and 'offset' or 'cursor' parameters to hive_list_databases, hive_list_tables, impala_list_databases, impala_list_tables, spark_list_sessions, spark_list_batches. Return a 'total_count' or 'has_more' field.
• Formalize error handling in descriptions. For each tool, document expected error conditions: 'If database does not exist, returns 404 with message "Database not found. Available: [list]". If query exceeds limit, returns 400 with "Limit must be 1-50". Retry with corrected parameters.'
• Declare permission scopes for each tool. Add a 'scope' field to tool metadata (or document in description): hive_* tools require 'read:hive', spark_submit_batch requires 'write:spark'. Enable least-privilege agent configuration and audit trails.