Group of Skills: Group-Structured Skill Retrieval for Agent Skill Libraries

Researchers introduce Group of Skills (GoSkills), a new method for organizing and retrieving skills in AI agent libraries that presents skills as structured execution contexts rather than flat lists. The approach improves agent performance on benchmark tasks while maintaining efficiency and doesn't require changes to existing agent systems.

GoSkills addresses a fundamental challenge in building scalable AI agents: as skill libraries grow, retrieving relevant skills becomes insufficient without clear guidance on how to use them. Traditional retrieval methods return isolated skills or dependency bundles without explicit information about execution order, supporting functions, prerequisites, or failure prevention strategies. This leaves agents to infer critical execution details, introducing inefficiency and potential errors.

The innovation structures skill retrieval around a typed skill graph, organizing related capabilities into anchor-centered groups with explicitly labeled roles. The system then expands these groups through a group graph to include necessary support skills, while maintaining a bounded payload to prevent context bloat. The result is a standardized execution contract with defined Start, Support, Check, and Avoid fields—essentially providing agents with actionable blueprints rather than raw tool lists.

This approach holds significance for the broader AI agent ecosystem, particularly as autonomous systems handle increasingly complex multi-step tasks. Better skill organization directly translates to more reliable and efficient agents without requiring modifications to underlying architectures. The experimental validation on SkillsBench and ALFWorld demonstrates practical improvements in both task performance and computational efficiency.

Looking ahead, this work may influence how skill libraries are designed across AI agent frameworks, potentially becoming a standard pattern for knowledge organization. As agents become more prevalent in production systems, efficient skill retrieval mechanisms like GoSkills could reduce latency, improve reliability, and lower computational costs—factors critical for enterprise adoption.

• →GoSkills transforms skill retrieval from flat lists to structured execution contexts with explicit role labels
• →The method improves agent task performance while maintaining computational efficiency on benchmark tests
• →Implementation requires no changes to existing agent systems, skill payloads, or execution environments
• →Standardized execution contracts with Start, Support, Check, and Avoid fields provide clear action guidance
• →The approach addresses the scalability challenge of large reusable skill libraries in modern AI agents