Stop asking 'is this skill safe?' — ask 'what can it do?'

The article explores the concept of 'capability disclosure' regarding skills for AI agents. The author criticizes the traditional approach focused solely on security assessment and suggests shifting the focus to a detailed analysis of a skill's functional capabilities. This approach allows developers to better understand the actions an agent can perform and manage its behavior more effectively. The material also introduces a small open-source tool that automates the process of analyzing and reading skill descriptions, simplifying the work for developers integrating new features into AI systems. The author emphasizes that functional transparency is key to creating more predictable and reliable autonomous agents. This article is useful for professionals involved in the development and deployment of LLM agents who seek to improve control over their interaction with external APIs and tools.