ARIA is a multi‑agent large‑language‑model framework that autonomously runs end‑to‑end visual tests on Android infotainment systems. From simple scenario sentences, it executes interactions, generates reproducible scripts, and produces detailed reports with visual evidence. In evaluation on a manufacturer’s device, ARIA achieved a 93.3% completion rate, correctly identified all known defects, and demonstrated lower false‑positive rates compared to a single‑agent baseline.
By Ant\'onio Azevedo, Bruno Lima, Jo\~ao Pascoal Faria
arXiv:2606. 09498v3 Announce Type: replace Abstract: The performance of LLM-based agents is jointly shaped by their base models and the harnesses that mediate their interaction with the environment.
By Hangfan Zhang, Shao Zhang, Kangcong Li, Chen Zhang, Yang Chen, Yiqun Zhang, Lei Bai, Shuyue Hu
The paper introduces a framework for evaluating large‑language‑model agents by focusing on the interaction between the model and its execution harness. It defines five complementary metrics—Valid Completion Rate, Model‑Harness Fit, Cross‑Model Fit, Interaction Residuals, and Token Load—to assess how reliably agents finish tasks, how well harnesses support specific models, and the overall suitability across models. Using these metrics, the authors evaluate four general‑purpose harnesses and two native systems on six benchmarks, finding that harness rankings differ by model and task and that some harnesses consistently achieve high valid completion rates.
By Shao Zhang, Hangfan Zhang, Yuanyuan Shi, Hao Li, Min-Le Su, Kangcong Li, Jingping Wang, Chunjiang Mu, Chen Zhang, Yang Chen, Bo Zhang, Shuyue Hu
arXiv:2607. 08124v1 Announce Type: cross Abstract: The behavior of an LLM agent is determined not only by the underlying model, but also by its harness: the executable program that constructs context, invokes tools, verifies intermediate results, and recovers from failures.
By Jun Nie, Yonggang Zhang, Jun Song, Qianshu Cai, Dahai Yu, Yike Guo, Xinmei Tian, Bo Han
The paper investigates how environment-agnostic coding agents can be improved without environment-specific engineering. By modifying a Codex-based agent’s instruction to report difficulties and allow automatic continuation, the authors achieve higher mean scores on ARC‑AGI‑3 and AGI Maze, though performance on Terminal‑Bench remains mixed. A scripted controller that handles reset decisions partially recovers the advantage of a specialized agent, demonstrating that effective use of available information and procedures is crucial for these models.
By Sergey Rodionov
arXiv:2605. 25160v2 Announce Type: replace Abstract: GUI agents powered by large language models are advancing rapidly, creating urgent needs for evaluation and training based on realistic environments.
By Guohong Liu, Jialei Ye, Pengzhi Gao, Wei Liu, Jian Luan, Yunxin Liu, Yuanchun Li
The paper introduces DroidTool, a framework that enables Android GUI agents to automatically generate and integrate tool actions—Python functions that manipulate application states—into their hybrid action space. DroidTool follows an agentic workflow with stages of proposal, implementation, test generation, execution, and repair, and it constructs relational tests to ensure comprehensive verification. Experiments on benchmarks such as AndroidWorld, B-MoCA, and MobileSafetyBench show that agents using DroidTool’s self-generated tools outperform GUI-only agents by about 4.47% in performance while reducing interactions by roughly 20.05%.
By Juyong Lee, Woogyeol Jin, Kimin Lee
Agents driven by large language models are increasingly deployed on long-horizon tasks, where they use tools and modify real workspaces. How well they perform depends as much on the harness around the...
The paper investigates how to improve environment‑agnostic coding agents by refining prompts and leveraging scaffolding. By modifying a single instruction to encourage reporting difficulties and allowing automatic continuation, the authors achieve higher mean scores on ARC‑AGI‑3 and AGI Maze, though performance on Terminal‑Bench remains mixed. A scripted controller that handles reset decisions recovers part of the specialized agent’s advantage, while additional tutorial experience can actually reduce performance, underscoring the importance of effectively using available information and tools.
arXiv:2607. 03691v2 Announce Type: replace-cross Abstract: Coding agents, autonomous systems that use large language models (LLMs) to resolve software engineering tasks, rely on agent harness: a middleware layer in between a developer and a large language model that orchestrates system prompts, tool execution, context management, and iterative reasoning loops.
By Oussama Ben Sghaier, Hao Li, Bram Adams, Ahmed E. Hassan
arXiv:2606. 30573v1 Announce Type: new Abstract: We introduce SWE-Interact, a new testbed for evaluating coding agents on multi-turn, interactive, user-driven software engineering tasks.
By Mohit Raghavendra, Anisha Gunjal, Aakash Sabharwal, Yunzhong He
InfraDev is a new benchmark comprising 36 tasks derived from real-world pull requests to Megatron‑LM, vLLM, and verl, designed to test coding agents on infrastructure development for training, inference, and reinforcement learning across many GPUs. Each task requires agents to reimplement multi‑file changes and run correctly, with grading that includes hidden probes and a prompt exposing only the interfaces used. The benchmark reveals that current frontier models resolve only about 39% of tasks, with probes rejecting a third of submissions that pass official tests, and that structural and state errors dominate failures.
By Yangshen Deng, Jiaqi Ruan, Yeqi Huang, Nick Lau, Guangming Sheng, Luo Mai