Architecting Secure AI Agents: Perspectives on System-Level Defenses Against Indirect Prompt Injection Attacks | ArxivCSExplorer