SkillsHub March 2026 Dev Log: BM25, Rate Limiting, and 5,900+ AI Agent Skills
SkillsHub, an open registry for AI agent skills, released v0.2.0 with improvements to its search engine, rate limiting, and documentation. The article highlights key updates and the growing skill repository.
Why it matters
SkillsHub is a critical resource for developers building AI agents, providing a growing catalog of reusable skills and a robust search engine to discover them.
Key Points
- 1Implemented BM25 multi-field scoring for more relevant search results
- 2Added an in-memory skill cache for faster query resolution (~3.5x faster)
- 3Introduced rate limiting to maintain a healthy open registry
- 4Shipped an interactive documentation page with API examples
- 5Crossed 5,900 skills in the registry, up from 5,300 at launch
Details
SkillsHub, a platform for discovering and sharing AI agent skills, released version 0.2.0 with several major improvements. The most significant change is the switch from TF-IDF to BM25 multi-field scoring for the search engine. BM25 scores matches across multiple fields (name, description, tags) with customizable weights, resulting in more relevant results for ambiguous queries. Additionally, SkillsHub implemented an in-memory skill cache that pre-computes term frequencies and document lengths, reducing query latency from 700ms to 200ms. To maintain the health of the open registry, the team added rate limiting backed by Upstash Redis, with generous limits for normal usage and the ability to request higher limits for integrations. The update also includes an interactive documentation page with live API examples, improved contributor tooling, and the addition of 600+ new skills, bringing the total to over 5,900. Future plans include semantic search, skill versioning, SDK packages, and a community-driven skill review system.
No comments yet
Be the first to comment