arXiv paper: attributing 307 LLM agent failures to specific skills via differential analysis

alex_verem · x · 2026-08-21

The arXiv paper "Agent Skills Can Be Harmful" (2608.11888) systematically attributes agent failures to loaded skills. The authors introduce a differential analysis framework that attributes a failure or cost regression to a skill by comparing a skill-guided run against a no-skill or semantically matched reference run, instantiated on SkillsBench and SWE-Skills-Bench, yielding 307 skill-induced failures (125 functional failures, 182 efficiency regressions).

The paper also builds SkillTriage, a taxonomy-guided attribution tool that normalizes paired cases, extracts differential evidence, and produces triage reports. Major findings: functional failures are rarely caused by obviously irrelevant skills — seemingly relevant skills often make agents incorrectly implement or omit required elements, while efficiency regressions show up as increased token use and execution time.

Related event: Study Finds AI Agent Skills Can Cause 307 Task Failures(2 posts)→

Original post →

More from coding & agent

coding & agent channel →