Every AI agent needs the same safety guardrails: PII detection, prompt injection blocking, topic enforcement, token limits, output validation. AgentGuard provides all of this as composable, testable, ...