Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks

RESUMO

How can language model agents effectively leverage libraries of reusable knowledge to solve long-horizon tasks? Recent work has increasingly focused on agent skills: reusable capabilities represented as skill packages. We investigate an alternative approach in which skill packages are instead invoked as subagents, spawning fresh context windows dedicated to solving individual subtasks. Subagent execution outperforms agent-skill execution when skill packages expose clear input-output contracts, at the cost of additional communication overhead.