A new paper presents three methods for formal verification of agent skills, enhancing the capability-containment proof for LLM-driven runtimes. These methods leverage existing tools and are packaged as open-source JavaScript modules, aiming to improve the reliability of AI skill manifests.