Agentic Harness Engineering Paper — Automated Harness Evolution Lifts Terminal-Bench Pass@1 from 69.7% to 77.0%

arXiv / Hacker News·medium signal

ArXiv paper 2604.25850 introduces Agentic Harness Engineering (AHE), a framework that automates coding-agent harness evolution through three observability pillars: component, experience, and decision observability. Ten AHE iterations lifted pass@1 on Terminal-Bench 2 from 69.7% to 77.0%, surpassing the human-designed Codex-CLI harness (71.9%). The frozen harness transfers cross-family without re-evolution, yielding +5.1 to +10.1pp gains across three alternate model families — suggesting the evolved components encode general engineering experience rather than benchmark-specific tuning.