GitHub Copilot CLI 入門：よく使うスラッシュコマンド一覧と使い方GitHub Copilot CLI for Beginners: Overview of common slash commands

GitHub Copilot CLI brings the company's AI coding assistant directly into the terminal, and its slash commands are the primary way developers steer that assistant without leaving the command line. For people who spend much of their day in a shell, learning these commands matters because it turns an open-ended chat interface into a set of repeatable, predictable actions that fit cleanly into existing workflows.

A slash command is simply an instruction that begins with a forward slash and is interpreted by the Copilot CLI agent rather than passed to your operating system. Instead of typing a long natural-language request, you invoke a short keyword that maps to a specific behavior. This distinction is important: a slash command tells the agent what mode to enter or what meta-action to perform, while ordinary prompts describe the coding task itself. The two work together, and fluency comes from knowing which command sets up the right context for a given prompt.

Among the most frequently used commands are those aimed at understanding and repairing code. A command such as /explain is typically used to ask the agent to describe what a selected file, function, or error message does in plain language, which is useful when navigating unfamiliar codebases. A /fix command generally directs the agent to diagnose a failing test, a stack trace, or a linting error and propose a corrected version. Because these commands frame the request for the model, they tend to produce more focused output than a free-form question would.

Session and context management commands form a second important group. Tools in this category usually let you clear the current conversation, reset accumulated context, or review the history of what the agent has done so far. Managing context deliberately matters because large language models work within a limited window, and a cluttered session can degrade the relevance of later responses. Commands that switch the underlying model or adjust the agent's permissions also fall here, allowing users to trade speed for capability or to control how much autonomy the agent has when it suggests running commands or editing files.

A third group covers configuration and help. A /help command lists the available slash commands and their syntax, which is the natural starting point for newcomers, while commands tied to authentication or account settings handle login state and access to features. Because the exact command set appears to evolve between releases, consulting the built-in help output is generally more reliable than memorizing a fixed list, and readers should verify command names against the version they have installed.

It helps to place Copilot CLI within the broader landscape of terminal-based AI agents. The tool sits alongside the GitHub Copilot extensions in editors such as Visual Studio Code and the GitHub Copilot Chat experience, sharing the same underlying assistant but adapting it to a keyboard-driven environment. It is also part of a wider industry shift toward agentic command-line tools, a category that includes offerings from other vendors that similarly let an AI read files, run commands, and propose changes within a project directory. Understanding slash commands in one of these tools makes the conventions in the others easier to grasp, since most follow a comparable pattern of a slash-prefixed keyword followed by optional arguments.

A few prerequisite concepts make the experience smoother. Users should be comfortable with basic shell navigation, since the agent operates relative to the working directory, and should understand that suggestions which modify files or execute commands are likely to require explicit confirmation. Treating the agent as a collaborator whose output needs review, rather than an authority, remains good practice; generated fixes can be incorrect or incomplete, and version control such as Git provides a safety net for reverting unwanted changes.

In practice, the value of these commands grows when they are woven into a routine. A developer might use an explanation command to orient themselves in a new module, a fix command to resolve a failing build, and context-management commands to keep the session clean between unrelated tasks. As an introductory guide suggests, the goal is not to memorize every option but to build a working vocabulary of the handful of commands you reach for most, then expand that set as your needs grow. Because the tool and its feature set continue to change, periodically rechecking the documentation is a sensible habit.