ChenLiu1996/Agent-Harness-Engineering

Agent Harness Engineering: A Survey

This dataset hosts the PDF for Agent Harness Engineering: A Survey.

This release contains only the survey PDF. For the curated resource list, please see the companion project page or GitHub repository.

Abstract

The harness is becoming the binding constraint. The rapid deployment of large language model agents in production has revealed a recurring pattern: task execution reliability depends less on the underlying model than on the infrastructure layer that wraps it, the agent execution harness.

The paper is organized around three central claims:

1. Harnesses are independent system layers. Real-world reliability is shaped by execution controls, feedback loops, governance, evaluation, and operational design, not only by model capability.
2. ETCLOVG separates production concerns. Execution, Tooling, Context, Lifecycle, Observability, Verification, and Governance expose architectural boundaries that earlier frameworks often conflate.
3. A broad ecosystem map reveals gaps. A systematic mapping of the open-source ecosystem surfaces adoption patterns across sandboxes, protocols, memory systems, orchestrators, observability platforms, benchmarks, and governance stacks.

Three Engineering Phases

Read across 2022–2026, agent engineering has gone through a coherent shift in where the marginal effort lands. The three phases overlap in time and concept; they describe what the field has chosen to engineer, not a clean sequence of replacements.

2022–2024: Prompt engineering. The primary lever is the input prompt text: instructions, few-shot examples, and reasoning templates, all optimized for a single model call.

2025: Context engineering. The question shifts from "what is the input?" to "what should the model see at each step?" The scope expands to retrieval, compaction, tool-result ranking, and managing context-window saturation across turns.

2026–: Harness engineering. As models become capable enough to attempt long-running tasks, the engineering focus expands to the full infrastructure wrapper: execution environment, tool interface, context, lifecycle, observability, verification, and governance.

The ETCLOVG Taxonomy

We organize the harness into seven layers. The first four describe the structural core of a harness; the last three describe the control plane around it. Compared with earlier six-component frameworks, Observability and Governance appear here as independent layers because, in production deployments, each has its own tooling stack and is owned by a different team.

Timeline of Agent-Harness Systems

The same shift is visible in the systems themselves. The ReAct era of 2022–2023 wrapped a single model loop with a while-loop, a prompt template, and a small tool dispatch table; AutoGPT and BabyAGI exposed the resulting failures, including execution runaway, context blowout, state loss, and unmonitored side effects, as infrastructure problems rather than prompt problems. Tool integration and multi-agent coordination from 2023–2024 added learned tool use (Gorilla, ToolLLM, Toolformer), role-playing organizations (CAMEL, ChatDev, MetaGPT, Mixture-of-Agents), the first agent benchmarks (SWE-bench, AgentBench, WebArena, GAIA), and the beginnings of protocol standardization (MCP, A2A). By 2025–2026 enough deployment experience had accumulated that “harness engineering” began to be named as a discipline of its own, accompanied by automated harness optimization and a wave of results in which only the harness was varied.

Mapping Open-Source Projects

The counts below reflect primary assignments in the current snapshot.

Layer	Scope	Primary Projects
■  E	Execution environment & sandbox	20
■  T	Tool interface & protocol	12
■  C	Context & memory management	9
■  L	Lifecycle & orchestration	47
■  O	Observability & operations	15
■  V	Verification & evaluation	21
■  G	Governance & security	14

Looking for more?

More information can be found on the project page and the Github repository.

Citation

If you use this survey, please cite:

@misc{li2026agentharness,
  title={Agent Harness Engineering: A Survey},
  author={Li, Junjie and Xiao, Xi and Zhang, Yunbei and Liu, Chen and Zhao, Lin and Liao, Xiaoying and Ji, Yingrui and Wang, Janet and Gu, Jianyang and Ge, Yingqiang and Xu, Weijie and Fang, Xi and Xu, Xiang and Zhao, Tianchen and Kim, Youngeun and Wang, Tianyang and Hamm, Jihun and Krishnaswamy, Smita and Huan, Jun and Reddy, Chandan},
  url={https://openreview.net/pdf?id=eONq7FdiHa},
  year={2026}
}