I’m building WarrantKit, an open-source project exploring a specific problem in agent security: What happens when an AI agent retains permission to act after the runtime state that authorized it has changed? Most agent frameworks focus on orchestration…