SkillRouter
paperAddresses skill routing for LLM agents at scale (10K+ skills). Key finding: hiding the skill body causes a 31–44 percentage point drop in routing accuracy, proving complete skill implementations matter for selection decisions.
SkillRouter is a 1.2B retrieve-and-rerank system achieving 74.0% top-1 accuracy while running 5.8x faster and using 13x fewer parameters than comparable approaches. Improved routing translates to better task success rates across four coding agents.