Tag
This paper introduces Skill Following and the Retrieval-Invoked Actual-Use Effect to evaluate whether skill retrieval in LLM agents genuinely improves task performance, revealing that aggregate metrics can be misleading.