Design and Cross-Domain Evaluation of an Agent Execution Harness.

arXiv:2610.25568v1 Announce Type: new Abstract: Completing complex tasks with language-model agents requires coordinating planning, execution, and review so the final result addresses the user's original objective. We present an agent execution harness that ke