Study evaluates coding harness components for autonomous coding agents
Researchers conducted an empirical study of coding harnesses that translate large language model capabilities into long‑horizon software‑engineering tasks. They kept the execution loop constant while varying three components—planning, action space, and context management—across four models evaluated on SWE‑Bench Verified and Terminal‑Bench 2.1. The experiments covered 176 matched settings,…
Key points
- 176 settings tested across five context-management strategies and four context-window budgets on four models.
- Rule‑based elision before LLM summarization gave the best efficiency, while recoverable elision added cost with no accuracy gain.
- Planning improved weaker models’ accuracy but mainly reduced cost for stronger models; predefined tools helped low‑bash‑skill models.
The findings show that context management becomes increasingly valuable as the context‑window budget tightens, mainly by preventing overflow failures. Staging rule‑based elision before LLM‑based summarization yielded the strongest overall efficiency, whereas making elided content recoverable added machinery that models rarely used and gave no accuracy gain. Planning acted as an accuracy scaffold for weaker models but turned into a cost‑saving mechanism for stronger models with little impact on accuracy. Predefined tools improved performance for models with weaker bash proficiency, while bash‑capable models performed well with a simple bash‑only interface, reducing cost on command‑line‑centric tasks.
An empirical study of harness design for coding agents
arxiv.org · 18 September 2026
Computer Science > Artificial Intelligence
Title:An Empirical Study of Harness Design for Coding Agents
View PDF HTML (experimental) Abstract: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.
Current browse context:
Bibliographic and Citation Tools
Code, Data and Media Associated with this Article
Demos
Recommenders and Search Tools
arXivLabs: experimental projects with community collaborators
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
This text was published by arxiv.org. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
Coverage and discussion
1source- Hacker News discussion · 214 pointsnews.ycombinator.com
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us.
More in Research
All →- ICLR 2027 receives roughly 50,000 abstract submissions · 1 src
- OpenAI claims its model solved the Navier–Stokes Millennium Prize problem · 3 src
- FlowCheck catches silent failures in vibe-coded apps where frontier models miss bugs · 1 src
- Deepmind says chain-of-thought transparency is at risk · 1 src
- Google adds Nobel laureates and top economists to AI & Economy team · 1 src
Comments
via GitHub Discussions