Two tools with basic schemas and descriptions, but significant quality gaps. Tool names lack clear action verbs (execute_code, execute_jupyter are generic). Descriptions are in Chinese and lack LLM-optimized structure (no WHAT/WHEN/WHY clarity). Parameter descriptions are minimal. No output schema documentation. Error handling returns raw JSON strings without recovery guidance. No input validation or constraint documentation. Schemas present but sparse, missing type constraints, enums, and ranges. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Both tools marked WRITE risk but no confirmation/dry-run pattern.

• Rename tools: execute_code → run_code_snippet, execute_jupyter → run_jupyter_notebook. Use verb_noun pattern with specific action.
• Rewrite descriptions in English, 50-200 chars each. Example: 'Execute a code snippet in a specified language (Python, JavaScript, Bash, etc.). Returns stdout, stderr, and exit code. Timeout in seconds prevents runaway execution.'
• Document output schema explicitly. Specify fields returned by sandbox_fusion.run_code() and run_jupyter(), e.g., {stdout: string, stderr: string, exit_code: int, execution_time_ms: int}.
• Add language enum to execute_code schema: {type: 'string', enum: ['python', 'javascript', 'typescript', 'bash', 'r', 'julia', 'golang', 'java', 'cpp', 'rust', 'php']}. Remove hardcoded SUPPORTED_LANGUAGES list.
• Add timeout bounds to both tools: {type: 'integer', minimum: 1, maximum: 300, description: 'Execution timeout in seconds (1-300). Defaults to 30.'}

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

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