arXiv · 2609.00549
Skill Following: Evaluating Actual Skill Use in Retrieval-Enabled LLM Agents
Abstract
Large Language Model (LLM) agents increasingly rely on external skills, yet standard evaluations obscure whether retrieving these skills actually helps. Aggregate metrics often compare retrieved versus non-retrieved tasks, introducing severe selection bias and failing to isolate the true effect of skill use. To measure this actual-use capability-which we formalize as Skill Following (SF)-we introduce the Retrieval-Invoked Actual-Use Effect (RAE). RAE computes the same-task outcome difference between matched skill-enabled and skill-disabled executions, conditioned exclusively on tasks where the agent actively retrieved a skill. Evaluating 17 LLMs across coding and mathematical domains, we uncover a stark evaluation paradox: models frequently show positive aggregate retrieval lift but negative RAE. On MBPP+, multiple models that appear to benefit system-wide actually harm their own performance on the exact tasks where retrieval occurred. These findings demonstrate that aggregate averages can create a misleading illusion of tool-use proficiency, whereas RAE directly measures whether the retrieval-to-answer pipeline genuinely rescues more outcomes than it harms.
Explore related subjects
Keep this discovery
Seonghyeon Cho, Chanjun Park. 2026-09-01. Skill Following: Evaluating Actual Skill Use in Retrieval-Enabled LLM Agents. https://arxiv.org/abs/2609.00549
Cite the original work for its findings. Save a collection to share your selection of sources.
Discover connections
Connections use source metadata and explicit phrase matches, not verified experimental comparisons.