Show HN: Skill Federation – private skill search for AI coding agents
Skill Federation is a free search engine designed for AI coding agents, not humans. It fetches relevant skills for tasks using keyword enrichment and reranking, and uses Cisco and Nvidia security scanners. The project was created to demonstrate that AI solutions need a finite set of interventions to perform well in specific applications, achieving a ~30% relative performance lift in tests.
Background
- **Opus 4.6** is Anthropic's Claude Opus 4.6, a frontier AI model. "Harnessed" means they ran it with tool access, not as a raw chatbot.
- **SkillsBench** and **SkillRet** are standard benchmarks for AI coding agents. They test how well an agent can find and use relevant "skills" (code snippets, API docs, library usage patterns). "Oracle skills" are the ideal ones designed for the benchmark; "wild skills" are random ones from the internet.
- The team's paper "Architecture of Errors" argued you can bound AI error rates by giving the model a curated, finite set of interventions (skills) per narrow domain. Skill Federation is the practical implementation.
- "Agent-native" means the search API is meant to be called by another AI, not typed into by a human. The agent decides what skills it needs; the engine fetches them.
- Why it matters: coding agents today often hallucinate or hit rate limits when scraping the open web for context. A pre-scanned, curated skill store could make them more reliable. The skills are security-scanned (Cisco, Nvidia) and free to use.