Natural-Language Agent Harnesses

March 2026 Linyue Pan, Lexiao Zou, Shuo Guo, Jingchen Ni, Hai-Tao Zheng Under review at NeurIPS 2026 (CCF-A)
Natural-Language Agent Harnesses

Overview

We propose Natural-Language Agent Harnesses (NLAHs), which externalize the high-level control logic of agent harnesses as portable, editable natural-language artifacts, and Intelligent Harness Runtime (IHR), a shared runtime that executes these harnesses through explicit contracts, durable artifacts, and lightweight adapters. Controlled evaluations across coding and computer-use benchmarks demonstrate the operational viability of this approach.