Kill switch

A safety mechanism in agent systems that can immediately interrupt agent execution and prevent further tool calls or side effects. It typically covers model inference, tool invocation and external API interaction, and is linked with alerting and human approval workflows.