An Empirical Study of Harness Design for Coding Agents
Empirical study of 176 matched coding-agent harness configurations shows context management cuts overflow failures while planning and action-space choices trade accuracy against cost.
The paper studies coding-agent harnesses with a fixed execution loop while varying planning, action space, and context management across four models on SWE-Bench Verified and Terminal-Bench 2.1, totaling 176 matched settings. Findings show context management chiefly prevents context-overflow failures, rule-based elision before LLM summarization is most efficient, planning shifts from accuracy scaffold to cost saver as models strengthen, and bash-only interfaces suit bash-capable models at lower cost. Trajectory-level analysis explains how each component alters agent behavior and execution trajectories.
- Evaluates 176 matched harness settings across four models on SWE-Bench Verified and Terminal-Bench 2.1.
- Context management mainly prevents context-overflow failures and matters most under tight context-window budgets.
- Rule-based elision before LLM summarization gives the best efficiency; recoverable elided content rarely helps.
- Planning helps weak-model accuracy but mostly saves cost for strong models.
- Bash-capable models run effectively with bash-only tool interfaces at substantially lower cost.
Full article247 words · extracted from huggingface.co · click to collapse
Coding harnesses shape how autonomous coding agents translate model capabilities into long-horizon software-engineering performance, yet existing work typically evaluates harnesses as monolithic systems, leaving the effectiveness of individual components unclear. To enable component-level comparisons, we study this question with a lightweight coding harness whose execution loop is fixed while three components are varied: planning, action space, and context management. Across four models evaluated on SWE-Bench Verified and Terminal-Bench 2.1, we evaluate 176 matched settings spanning five context-management strategies, four context-window budgets, and targeted ablations of planning and action space. We find that: (1) Context management becomes increasingly valuable as the context-window budget tightens, with most of its benefit coming from preventing context-overflow failures. (2) Staging rule-based elision before LLM-based summarization provides the strongest overall efficiency among the context-management strategies, whereas making elided content recoverable adds machinery that models rarely use and yields no accuracy gain. (3) Planning shifts from an accuracy scaffold for weaker models to a cost saver for stronger models, with little change in accuracy. (4) Predefined tools improve performance for models with weaker bash proficiency, whereas bash-capable models can operate effectively with a bash-only interface and achieve substantially lower cost, especially on command-line-centric tasks. Trajectory-level analysis explains these effects: context management extends execution trajectories without substantially altering agent behavior, planning changes where trajectories stop, and the action space changes the granularity at which code is written. These findings inform model- and budget-aware harness design and provide a modular framework for evaluating future harness components.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.20804