A specialized subset of prompts centered on refusal boundaries, escalation pathways, and risk-sensitive tool usage. It helps teams reduce policy drift across complex agent systems.
These guardrails are most effective when combined with runtime checks and explicit human-in-the-loop breakpoints.