Submitted by Qihan Ren 121 AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security AI45Research 292 8