@inproceedings{feng-etal-2026-longcli,
    title = "{L}ong{CLI}-Bench: A Preliminary Benchmark and Study for Long-horizon Agentic Programming in Command-Line Interfaces",
    author = "Feng, Yukang  and
      Sun, Jianwen  and
      Yang, Zelai  and
      Ai, Jiaxin  and
      Li, Chuanhao  and
      Li, Zizhen  and
      Zhang, Fanrui  and
      He, Kang  and
      Ma, Rui  and
      Lin, Jifan  and
      Sun, Jie  and
      Xiao, Yang  and
      Zhou, Sizhuo  and
      Wu, Wenxiao  and
      Liu, Yiming  and
      Liu, Pengfei  and
      Zhang, Shenglin  and
      Zhang, Kaipeng",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.1497/",
    pages = "29952--29963",
    ISBN = "979-8-89176-395-1",
    abstract = "Recent advances in AI-assisted programming have empowered agents to execute complex workflows via command-line interfaces, however, existing benchmarks are limited by short task horizons, data contamination from GitHub scraping, and a lack of fine-grained evaluation metrics, fail to rigorously evaluate the long-horizon planning and execution capabilities essential for realistic software engineering. To address these gaps, we introduce \textbf{LongCLI-Bench}, a comprehensive benchmark designed to evaluate agentic capabilities across \textbf{long-horizon}, realistic, sequential engineering tasks. We curated 20 high-quality, long-horizon tasks from over 1,000 computer science assignments and real-world workflows, covering four engineering categories: from scratch, feature addition, bug fixing, and refactoring. LongCLI-Bench employs a dual-set testing protocol, which measures requirement fulfillment \textit{(fail({\textrightarrow})pass)} and regression avoidance \textit{(pass({\textrightarrow})pass)}, and incorporates step-level scoring to pinpoint execution failures. Extensive experiments reveal that even state-of-the-art agents achieve pass rates below 20{\%} in LongCLI-Bench. Step-level analysis further indicates that the majority of tasks stall at less than 30{\%} completion, highlighting that critical failures often occur in the early stages. Although self-correction offers marginal gains, human-agent collaboration through plan injection and interactive guidance yields significantly higher improvements. These results highlight that future research must emphasize the development of synergistic human-agent workflows alongside advances in agents' planning and execution capabilities to overcome key challenges in long-horizon task performance."
}