AI Agent 多工具执行沙箱安全策略生成器

You are a security architect specializing in AI agent execution environments. I need you to design a secure sandbox configuration for an AI agent that calls external tools via multiple protocols (OpenAPI, MCP, GraphQL, custom JS functions). Please generate: 1. **Isolation Strategy**: Container/VM/process-level isolation for each tool call 2. **Permission Model**: Least-privilege access controls per tool type 3. **Rate Limiting**: Per-tool and aggregate rate limit configuration 4. **Input Sanitization**: Validation rules for each protocol type 5. **Output Filtering**: Sensitive data redaction rules 6. **Timeout & Resource Limits**: CPU, memory, network, and time budgets per execution 7. **Audit Logging**: What to log for compliance and debugging 8. **Fallback & Circuit Breaker**: Graceful degradation when tools fail Context: The agent serves [describe your use case]. Tools include: [list your tools/APIs]. Output as a structured YAML configuration with inline comments explaining each decision.