الملخص
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 representations and full-covariance LinUCB to score candidate combinations based on predicted performance and exploration value. Mixed-start coordinate ascent selects complete configurations for evaluation without enumerating the combinatorial space. Validation traces then guide local code edits, and the resulting candidates are incorporated into fixed-capacity role-specific pools for subsequent recombination. On text tasks, retrieval-augmented mathematical reasoning, code generation, and interactive scientific tasks, MESH-Harness outperforms Meta-Harness by 5.70, 7.01, 2.00, and 5.00 points, respectively, under matched candidate-evaluation budgets. Iterative harness optimization improves MESH-Harness by 5.63-7.79 points over its first-round configurations. For the reported configurations, aggregate test-time cost is 44.2% lower than that of Meta-Harness, while total cost including search is 14.6% lower. These results show that combining module-level design reuse with feedback-driven compositional search can systematically improve agent harnesses while keeping overall optimization cost under control.
الكلمات المفتاحية
الموضوع
بيانات النشر
- المجلة
- غير متاح
- وصول مفتوح
- وصول مفتوح أخضر
اقتبس هذه المقالة
APA 7
Shang, Z., Huo, Y., Gong, M., Yan, A., Qu, Z., Dong, J., Low, B. K. H., Wu, C., & Dai, Z. (2026). MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution. https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution
MLA 9
Shang, Zhiwei, et al. "MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution." https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution.
شيكاغو (المؤلف–التاريخ)
Shang, Zhiwei, Yu Huo, Mingrong Gong, An Yan, Zikun Qu, Junhao Dong, Bryan Kian Hsiang Low, Chenglin Wu, and Zhongxiang Dai. 2026. "MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution." https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution.
هارفارد
Shang, Z., Huo, Y., Gong, M., Yan, A., Qu, Z., Dong, J., Low, B. K. H., Wu, C. and Dai, Z. (2026) 'MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution', Available at: https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution.
فانكوفر
Shang Z, Huo Y, Gong M, Yan A, Qu Z, Dong J, et al. MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution. https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution
IEEE
Z. Shang, Y. Huo, M. Gong, A. Yan, Z. Qu, J. Dong, B. K. H. Low, C. Wu, and Z. Dai, "MESH-Harness: Self-Improving Agent Harnesses via Bandit-Guided Compositional Evolution," https://omanscience.com/ar/articles/mesh-harness-self-improving-agent-harnesses-via-bandit-guided-compositional-evolution.