Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Natural-Language Agent Harnesses

Linyue Pan, Lexiao Zou, Shuo Guo, Jingchen Ni, Hai-Tao Zheng

Abstract

Agent performance increasingly depends on \emph{harness engineering}, yet harness design is usually buried in controller code and runtime-specific conventions, making it hard to transfer, compare, and study as a scientific object. We ask whether the high-level control logic of an agent harness can instead be externalized as a portable executable artifact. We introduce \textbf{Natural-Language Agent Harnesses} (NLAHs), which express harness behavior in editable natural language, and \textbf{Intelligent Harness Runtime} (IHR), a shared runtime that executes these harnesses through explicit contracts, durable artifacts, and lightweight adapters. Across coding and computer-use benchmarks, we conduct controlled evaluations of operational viability, module ablation, and code-to-text harness migration.

Natural-Language Agent Harnesses

Abstract

Agent performance increasingly depends on \emph{harness engineering}, yet harness design is usually buried in controller code and runtime-specific conventions, making it hard to transfer, compare, and study as a scientific object. We ask whether the high-level control logic of an agent harness can instead be externalized as a portable executable artifact. We introduce \textbf{Natural-Language Agent Harnesses} (NLAHs), which express harness behavior in editable natural language, and \textbf{Intelligent Harness Runtime} (IHR), a shared runtime that executes these harnesses through explicit contracts, durable artifacts, and lightweight adapters. Across coding and computer-use benchmarks, we conduct controlled evaluations of operational viability, module ablation, and code-to-text harness migration.

Paper Structure

This paper contains 44 sections, 3 equations, 4 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Problem.
  3. Motivation.
  4. Thesis and approach.
  5. Contributions.
  6. Methodology
  7. Harnesses and the pattern layer
  8. Intelligent Harness Runtime
  9. From model calls to agent calls.
  10. Natural-Language Agent Harnesses
  11. File-backed state as an explicit module
  12. Experimental Design
  13. Research questions
  14. Instantiation
  15. Benchmarks and harness families
  16. ...and 29 more sections

Figures (4)

  • Figure 1: Examples of harness design patterns used by modern agents (reason--act, retrieval, reflection, verification, memory, search, orchestration).
  • Figure 2: Framework overview. Intelligent Harness Runtime (IHR), with an in-loop LLM, a backend with tool access and child-agent support, and a runtime charter that specifies policy and semantics, executes a Natural-Language Agent Harness (NLAH), which exposes harness logic, roles, contracts, adapters, and state conventions, over task instances.
  • Figure 3: Realization mapping: backend + runtime skill (charter) + harness skill (task-family logic).
  • Figure 4: Supplementary views for SWE RQ2. Left: resolved rate versus estimated token-based API cost per sample under public GPT-5.4 text pricing. Right: standalone solved rate and union solved rate with Basic for each ablated module.