cs.SENov 6, 2026

BOXBOX: Recursive Harness Self-Improvement for Long-Running Performance Optimization

Authors: Yuhao Tan, Hanyun Jiang, Fangkai Yang, Yuan Yao, Taolue Chen, Bo Qiao, Yu Kang, Lu Wang, 

Organizations: Nanjing University · Zhejiang University · Microsoft · Birkbeck, University of London

Abstract

Automated harness optimization improves agents by learning from trajectories and task feedback. Existing offline methods use reference answers to optimize harnesses for reuse on held-out tasks. However, many real-world tasks are open-ended, while long runs expose new difficulties that offline-optimized harnesses may not address. To study harness optimization in these settings, we introduce PerfRace, a benchmark for long-running performance optimization of software projects, and investigate whether agents can use task-specific experience to recursively improve their harnesses during an ongoing task. To this end, we introduce BOXBOX, which enables agent-triggered harness revision and kernel-managed hot reloading while preserving execution context. We evaluate BOXBOX on PerfRace under matched test-time budgets, comparing it with Frozen Harness, offline Harness Evolution, and Harness Scaling. BOXBOX achieves higher speedup with more efficient use of task time and model cost, reaching a Speedup@4h of 1.70x, compared with 1.54x for the strongest baseline.

Figures & tables

Appendix figures & tables25 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. Turbo Harness: Instance-Adaptive Harness Optimization

    Sep 30, 2026Tunyu Zhang, Hao Wang, Kai Xu +1Agent Harness OptimizationSelf-Improving Agents

  2. HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs

    Oct 19, 2026Zhuoshi Pan, ZhiYuan Feng, Yukuo Cen +6Agent Harness OptimizationAgent Harness Self-Evolution