终端智能体：命令行环境下的 AI 智能体综述
文章背景与核心概要
随着大语言模型（LLM）智能体在复杂任务处理中的应用日益广泛，通过命令行界面（CLI）进行交互已成为其核心能力之一。然而，以往的研究往往将终端交互行为分散在软件工程、工具使用和计算机操作等不同领域，缺乏统一的理论框架。
本文首次提出了“终端智能体”（Terminal Agents）的概念，将其定义为以终端命令执行、文本反馈及有状态环境交互为核心动作-观察循环的系统。作者通过“七维终端能力概况”模型，系统性地梳理了终端智能体的架构、能力获取方式及评估体系。研究指出，智能体的表现由模型、接口、控制框架、运行时和环境共同决定，并强调了在评估中从关注最终结果转向关注过程质量、恢复能力及治理的重要性。
📝 摘要 (Summary)
大语言模型智能体越来越多地通过终端执行任务，但现有的综述将终端交互行为分散在软件工程、工具使用和计算机操作等研究领域中。我们将“终端智能体”定义为：其主要的进展驱动型动作-观察循环是通过终端命令执行、文本反馈和有状态的环境交互来实现的系统。本文以终端执行作为核心组织视角，确立了工作负载层面的边界，并通过“七维终端能力概况”将系统架构、能力获取和评估连接起来。
Large language model agents increasingly act through terminals, yet existing surveys disperse terminal-mediated behavior across software engineering, tool use, and computer-use research. We regard terminal agents as systems whose dominant progress-bearing action–observation loop is mediated by terminal command execution, textual feedback, and stateful environment interaction. Using terminal-mediated execution as an organizing lens, this survey establishes workload-level boundaries and connects system architecture, competence acquisition, and evaluation through a seven-dimensional terminal competence profile.
我们的综合分析表明，智能体的实际行为是由模型、接口、控制框架（harness）、运行时和环境共同塑造的。可执行的轨迹将学习过程建立在动作后果、验证和恢复的基础上，而当前的评估体系过于强调最终结果，对过程质量、恢复能力和治理机制的考量尚不均衡。
Our synthesis shows that realized behavior is jointly shaped by the model, interface, harness, runtime, and environment. Executable trajectories ground learning in action consequences, verification, and recovery, whereas prevailing evaluations emphasize final outcomes and expose process quality, recovery, and governance unevenly.
受限的固定条件诊断揭示了两个启示：不同的基准测试系列会暴露不同的过程信号，而匹配的系统比较则揭示了性能对基准测试的依赖性以及组件归因的局限性。这些发现促使我们呼吁明确报告系统和运行时条件，并辅以可重现的轨迹和过程级证据。该框架为跨软件工程及新兴应用领域研究终端交互智能体提供了统一的基础。
Bounded fixed-condition diagnostics illustrate two implications: benchmark families expose different process signals, and matched system comparisons reveal benchmark-dependent performance and limits of component attribution. These findings motivate explicit reporting of system and runtime conditions, supported by replayable traces and process-level evidence. The framework provides a unified basis for studying terminal-mediated agency across software engineering and emerging application domains.