Submitted by Qihan Ren 123 AgentDoG: A Diagnostic Guardrail Framework for AI Agent Safety and Security AI45Research 346 8