AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security - Explained Simply | ArXiv Explained