Tag
An empirical study analyzes 8,351 Claude Code plugins, finding that Markdown files often control runtime behavior, leading to misclassified documentation changes and tight coupling between scripts and instructions, which necessitates automated synchronization checks.
This paper presents SkillJuror, a framework for evaluating how different organizational paradigms for Agent Skills affect LLM agent runtime behavior. A study on 82 tasks shows that Progressive Disclosure increases resource diversity and uptake events compared to a flat baseline, with task-dependent outcome improvements.