HarnessFlow: MCTS-Guided Harness Optimization for LLM Agents
Zhi Hong, Qianlong Wang and colleagues at South China University of Technology, Fudan University and Microsoft Research present HarnessFlow, which searches over complete executable agent harnesses with Monte Carlo Tree Search while the model and environment stay fixed.
Ask this paper
Search space. Candidates are full harness codebases covering context construction, state management, tool use and termination. They are kept in a refinement tree with their scores and traces.
Method. An MCTS policy picks which harness to refine next. A coding agent edits it using archived code, execution feedback and a library of reusable harness components.
Results. With 15 refinement attempts per benchmark, HarnessFlow has the highest mean test score on six benchmarks spanning agentic tasks, coding and web interaction, averaging 69.48%.
Comparisons. That average is 10.89 points above the initial harnesses and 3.95 points above Meta-Harness, at lower per-task execution cost than every hand-engineered baseline.
Scope. The search needs an automatic grader and evaluation compute, and each harness descends from one parent, so mechanisms from separate branches cannot yet be combined.
Abstract
The performance of large language model (LLM) agents depends on both the underlying model and the harness that governs interaction with tools and the environment. Yet harness design remains largely manual, and automating it requires navigating an open-ended search space through costly end-to-end evaluations. We introduce HarnessFlow, a framework for optimizing complete executable harnesses under a limited candidate-evaluation budget, with the model and task environment held fixed. HarnessFlow organizes candidates and their evaluation histories in a refinement tree, using a Monte Carlo Tree Search (MCTS)-based policy to balance exploitation and exploration. A coding agent refines selected harnesses using archived code, execution feedback, and a library of reusable harness mechanisms. With 15 refinement attempts per benchmark, HarnessFlow achieves the highest mean test scores among the compared methods on six benchmarks spanning agentic tasks, coding, and web interaction. The optimized harnesses achieve an average score of 69.48%, outperforming the initial harnesses and Meta-Harness by 10.89 and 3.95 percentage points, respectively, while incurring lower average per-task execution costs than hand-engineered baselines. These results suggest that structured harness search offers a promising route toward automating the design of capable and efficient LLM agents.