SkillSeek: Revisiting Agent Skill Retrieval at Marketplace Scale
A standard search stack matched the costlier LLM retrieval loop for choosing agent skills.
SkillSeek uses a BGE-base bi-encoder and a small cross-encoder, exposed over MCP, to retrieve skills from marketplace-scale pools. On SkillsBench, the authors say it reached observed parity with Liu et al.’s LLM-mediated retrieval loop. BM25 alone matched or beat that loop in three of four settings; the cross-encoder closed the remaining gap. Reported per-trial spend fell from $51.30 to $27.54, close to the no-skill baseline. HF Daily Papers' note
SkillSeek uses a BGE-base bi-encoder and a small cross-encoder, exposed over MCP, to retrieve skills from marketplace-scale pools. On SkillsBench, the authors say it reached observed parity with Liu et al.’s LLM-mediated retrieval loop. BM25 alone matched or beat that loop in three of four settings; the cross-encoder closed the remaining gap. Reported per-trial spend fell from $51.30 to $27.54, close to the no-skill baseline. HF Daily Papers' note
score 5