An Empirical Study of Harness Design for Coding Agents

Hugging Face Daily Papers Papers
coding-agents harness-design empirical-study context-management ablation-study

Summary

This paper empirically studies harness design for coding agents, evaluating components like planning and context management to improve performance in software engineering tasks.

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.
Original Article
View Cached Full Text

Cached at: 09/18/26, 03:02 AM

Paper page - An Empirical Study of Harness Design for Coding Agents

Source: https://huggingface.co/papers/2609.20804

Abstract

Codingharnessesshapehowautonomouscodingagentstranslatemodelcapabilitiesintolong-horizonsoftware-engineeringperformance,yetexistingworktypicallyevaluatesharnessesasmonolithicsystems,leavingtheeffectivenessofindividualcomponentsunclear.Toenablecomponent-levelcomparisons,westudythisquestionwithalightweightcodingharnesswhoseexecutionloopisfixedwhilethreecomponentsarevaried:planning,actionspace,andcontextmanagement.AcrossfourmodelsevaluatedonSWE-BenchVerifiedandTerminal-Bench2.1,weevaluate176matchedsettingsspanningfivecontext-managementstrategies,fourcontext-windowbudgets,andtargetedablationsofplanningandactionspace.Wefindthat:(1)Contextmanagementbecomesincreasinglyvaluableasthecontext-windowbudgettightens,withmostofitsbenefitcomingfrompreventingcontext-overflowfailures.(2)Stagingrule-basedelisionbeforeLLM-basedsummarizationprovidesthestrongestoverallefficiencyamongthecontext-managementstrategies,whereasmakingelidedcontentrecoverableaddsmachinerythatmodelsrarelyuseandyieldsnoaccuracygain.(3)Planningshiftsfromanaccuracyscaffoldforweakermodelstoacostsaverforstrongermodels,withlittlechangeinaccuracy.(4)Predefinedtoolsimproveperformanceformodelswithweakerbashproficiency,whereasbash-capablemodelscanoperateeffectivelywithabash-onlyinterfaceandachievesubstantiallylowercost,especiallyoncommand-line-centrictasks.Trajectory-levelanalysisexplainstheseeffects:contextmanagementextendsexecutiontrajectorieswithoutsubstantiallyalteringagentbehavior,planningchangeswheretrajectoriesstop,andtheactionspacechangesthegranularityatwhichcodeiswritten.Thesefindingsinformmodel-andbudget-awareharnessdesignandprovideamodularframeworkforevaluatingfutureharnesscomponents.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2609\.20804

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.20804 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.20804 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.20804 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Same Model, Different Harness: Different Coding-Agent Results

arXiv cs.AI

This paper investigates how changing the harness configuration in a coding agent impacts performance on coding benchmarks when the model remains fixed. The study shows that a treatment harness, which shortens older tool results to manage context, improves task completion rates, especially under tight context constraints.

Code as Agent Harness

Hugging Face Daily Papers

This survey paper presents a unified view of code as the operational substrate for agent reasoning and execution in agentic systems, organizing the discussion around three layers: harness interface, mechanisms, and scaling.

Harness design for long-running application development

Anthropic Engineering

Anthropic engineers detail a multi-agent harness design using generator and evaluator agents to improve Claude's ability to build complete, high-quality frontend applications autonomously over long durations.