Subagents vs Agent Skills: Executing Reusable Knowledge for Long-Horizon Agentic Tasks
This work investigates an alternative approach in which skill packages are instead invoked as subagents, and shows that subagent execution outperforms agent-skill execution when skill packages expose clear input-output contracts and their instructions encode the procedural knowledge needed to fulfill those contracts.
Wasu Top Piriyakulkij, Rachel Lawrence, Alicia Curth et al.
· 0 citations