18 tools with complete schemas and descriptions, but significant quality gaps. Tool names use Portuguese verbs (consulta, listar, descrever) which are non-standard for LLM-optimized naming conventions. Descriptions are present but generic (avg ~80 chars, below 194-char baseline). Parameters lack type specificity, many accept free-form strings without enums or format constraints. No output schema documentation. Error handling returns raw exception strings rather than actionable recovery guidance. Write-operation tools (executar_update, executar_ddl, alterar_procedure) mention user confirmation in descriptions but lack formal confirmation/dry-run patterns. Security validation exists (SQL injection checks in tools.py) but is not exposed in tool definitions.

• Rename all tools to English verb_noun pattern: consulta→query_sql, listar_bancos→list_databases, descrever_tabela→describe_table, etc. This enables LLM pattern recognition and cross-language compatibility.
• Document output schemas for all tools. Example for list_databases: 'Returns array of {name: string, state_desc: string, create_date: ISO8601, compatibility_level: integer}'. Include field descriptions.
• Implement Multi-Round-Trip Request pattern for write operations: return {status: 'input_required', message: 'Confirm UPDATE on table X affecting Y rows?', confirmation_token: '...'} and require explicit confirmation before execution.
• Replace raw exception strings with structured error responses: {error: 'invalid_query', message: 'Query contains forbidden keyword INSERT', suggestion: 'Use consulta() for SELECT only', retryable: false}.
• Add enum constraints to free-form parameters. For status_jobs, accept job_names as array of strings with pattern validation, or provide a list_jobs tool to discover valid names first.

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

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