Add input/output filtering policies to Agents to prevent sensitive information leaks, inappropriate content, and prompt injection attacks