MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution

Excerpt

arXiv:2610.05300v1 Announce Type: new Abstract: An agent harness is the code that organizes context, maintains state, and coordinates tool calls for a language model. We study how to improve the harness under a limited evaluation budget while keeping model weights fixed. Our method, MESH-Harness, organizes each harness into functional modules with explicit role-specific interfaces, allowing alternative implementations of each module to be substituted and recombined. It uses shared module represe