Scoring was not performed

• Add explicit output schema documentation for all tools. For cortex_search, specify: 'Returns: {results: [{id: string, score: number, fields: object}], total_count: integer}'
• For object management tools, define target_object schema per type. Example: 'For database: {name: string, comment?: string}. For table: {name: string, schema: string, columns: [{name: string, type: string}]}'
• Expand Redis tool descriptions to 100+ chars. Example dbsize: 'Returns the count of keys in the connected Redis database. Use this to estimate database size or detect data loss. Warning: In large databases this is O(N) and may block other operations.'
• Add formal enum constraints for Snowflake object_type and Redis info section instead of listing in description. Use JSON Schema 'enum' field.
• Document cortex_search filter_query format. Is it Snowflake SQL WHERE clause syntax? Provide an example: 'filter_query: {where: "category = \'news\'"}'
• Add error handling guidance to drop_object: 'Error: Permission denied → verify role permissions. Error: Object in use → try with cascade flag if supported.'
• For run_snowflake_query, add param description: 'SQL statement to execute. Supports SELECT, INSERT, UPDATE, DELETE, DDL. Avoid administrative statements (DROP DATABASE, ALTER USER).'
• Add idempotency notes for create_object with mode='if_not_exists'. For destructive tools, add: 'This operation is irreversible.'
• Document required fields for cortex_analyst 'semantic_model' param. Is it a stage path, semantic view name, or both? Provide format constraint.