The Command-Line Interface (CLI) is an operational interface that enables user-computer interaction through text commands. Since the Unix era, the CLI has been a critical interaction method for system administration and DevOps, characterized by composable commands, scriptable operations, and parsable execution results.

The development of the CLI can be traced back to 1961, when MIT introduced CTSS (Compatible Time-Sharing System). It introduced an early command-line interpreter, establishing an interaction mode where users input commands and the system responds immediately, laying the foundation for modern CLI development.

With the advancement of agent technologies, the CLI has gradually become an important interaction method for agents to perform system operations and invoke tasks. By executing commands, reading results, and processing exception feedback, agents can complete continuous operations and dynamically adjust task workflows based on execution results, thereby reducing dependency on dedicated software interfaces.

As the application of CLIs in long-horizon agent tasks deepens, their human-machine collaboration capabilities have also attracted attention. In 2026, Liu et al. proposed a unified evaluation benchmark in the paper RealCLI: Toward a Unified Evaluation of Human-Agent Collaboration for CLI Agents. This benchmark evaluates the task execution and collaboration performance of CLI agents across five dimensions, 22 sub-dimensions, and 51 interaction scenarios. The study results show that the average score of the current best configuration is 49.2%, with the highest joint collaboration score reaching 37.6%. This indicates that agents still have significant shortcomings in terms of steerability, trust calibration, and safety boundary management.

Therefore, the core requirements for CLIs in agent scenarios have extended from basic command execution capabilities to task reliability, human-machine collaboration, and operational safety. Meanwhile, CLIs support complete command logging, execution tracing, and operation reproduction, which helps enhance the auditability of task execution processes and supports standardized operations in system administration and production environments.