Skill Router – a local-first router for large agent skill libraries (github.com)

The newly announced Skill Router is a locally operated, single-binary router designed to efficiently discover and load large agent skill libraries while minimizing model context overload. Unlike traditional methods that process extensive skill descriptions in every prompt, Skill Router adopts a local-first approach by indexing only essential metadata using SQLite. It enables natural-language searches to retrieve a ranked list of pertinent skills and loads the full skill content only when requested by the user, enhancing performance and reducing unnecessary processing. This innovation is significant for the AI/ML community as it addresses the growing complexities associated with large agent systems that may contain numerous specialized skills. By implementing features like full-text search with advanced ranking modes and telemetry for improved skill performance, Skill Router promises to streamline how developers interact with expansive skill libraries. Moreover, its open-source nature, built with secure coding practices and capable of running without remote service dependencies, empowers developers to integrate and customize it in local environments effectively. With easy setup and a robust testing framework, Skill Router is positioned as a valuable tool for enhancing the efficiency of AI agent frameworks.