Using agent mode in your IDE

Give Copilot a task and let it work autonomously in your editor, editing files across your project and running commands until the task is complete.

Tool navigation

在本文中

Introduction

In agent mode, Copilot takes a high-level task, decides which files to change, makes the edits, and runs commands as needed, iterating until the task is done. You stay in control: you review the changes, and by default you approve commands before they run. Your administrator, or your own editor settings, may allow some commands to run automatically.

Agent mode is available in Visual Studio Code, Visual Studio, JetBrains IDEs, Xcode, and Eclipse. The steps differ by editor, so click the tabs above for instructions for your IDE.

For how to open Copilot Chat and choose between the available modes, see Asking GitHub Copilot questions in your IDE.

当你有特定任务需要处理并且希望 Copilot 能够自主编辑你的代码时，请使用代理模式。 在代理模式下，Copilot 会确定要更改的文件、提供代码更改和终端命令来完成任务，并进行迭代来修正问题，直到原始任务完成。

代理模式最适用于以下用例：

  • 任务复杂，涉及多个步骤、迭代和错误处理。
  • 希望让 Copilot 确定完成任务所需的步骤。
  • 任务要求 Copilot 与外部应用程序（如 MCP 服务器）集成。

Using agents

  1. If the chat view is not already displayed, select Open Chat from the Copilot Chat menu.
  2. At the bottom of the chat view, ensure Agent is selected from the agents dropdown.
  3. Submit a prompt. In response to your prompt, Copilot streams the edits in the editor, updates the working set, and runs terminal commands if necessary.
  4. Review and iterate on changes or run a code review.

You can also click this link to go directly to agent mode in VS Code.

注意

If you don’t see the Agent option in the mode selector, your enterprise or organization administrator may have disabled agent mode for your IDE.

For more information, see Chat overview in the Visual Studio Code documentation.

使用代理模式时，你输入的每个提示都会消耗 GitHub AI Credits。

Steering an agent while it works

Agent mode is interactive. While Copilot is working, you can:

  • Submit a follow-up prompt to redirect the agent before it finishes.
  • Confirm or reject each terminal command the agent proposes, unless it has been configured to run automatically.
  • Review streamed edits as they appear and undo any you do not want.

If a task is large or ambiguous, consider drafting an implementation plan first. See Using plan mode in your IDE.

Choosing a model or a custom agent

Before you submit a task, you can change which AI model the agent uses, or select a custom agent tailored to a specific kind of work.

Extending agent mode with tools

Much of agent mode's capability comes from the tools it can call. You can extend the agent with Model Context Protocol (MCP) servers, which add tools for working with external systems and with GitHub itself.

To hand a self-contained subtask to a separate agent with its own context, see Using subagents in your IDE.

Further reading

Use agent mode when you have a specific task in mind and want to enable Copilot to autonomously edit your code. In agent mode, Copilot determines which files to make changes to, offers code changes and terminal commands to complete the task, and iterates to remediate issues until the original task is complete.

Agent mode is available in Visual Studio 17.14 and later.

Using agent mode

  1. In the Visual Studio menu bar, click View, then click GitHub Copilot Chat.
  2. At the bottom of the chat panel, select Agent from the mode dropdown.
  3. Submit a prompt. In response to your prompt, Copilot streams the edits in the editor, updates the working set, and if necessary, suggests terminal commands to run.
  4. Review the changes. If Copilot suggested terminal commands, confirm whether or not Copilot can run them. In response, Copilot iterates and performs additional actions to complete the task in your original prompt.

When you use Copilot agent mode, each prompt you enter consumes GitHub AI Credits.

Choosing a model

Before you submit a task, you can change which AI model the agent uses. See Changing the AI model for GitHub Copilot Chat.

Extending agent mode with tools

You can extend the agent with Model Context Protocol (MCP) servers, which add tools for working with external systems and with GitHub itself.

Further reading

当你有特定任务需要处理并且希望 Copilot 能够自主编辑你的代码时，请使用代理模式。 在代理模式下，Copilot 会确定要更改的文件、提供代码更改和终端命令来完成任务，并进行迭代来修正问题，直到原始任务完成。

代理模式最适用于以下用例：

  • 任务复杂，涉及多个步骤、迭代和错误处理。
  • 希望让 Copilot 确定完成任务所需的步骤。
  • 任务要求 Copilot 与外部应用程序（如 MCP 服务器）集成。

Using agent mode

  1. To start an edit session using agent mode, click Copilot in the menu bar, then select Open GitHub Copilot Chat.
  2. At the top of the chat panel, click the Agent tab.
  3. Submit a prompt. In response to your prompt, Copilot streams the edits in the editor, updates the working set, and if necessary, suggests terminal commands to run.
  4. Review the changes. If Copilot suggested terminal commands, confirm whether or not Copilot can run them. In response, Copilot iterates and performs additional actions to complete the task in your original prompt.

使用代理模式时，你输入的每个提示都会消耗 GitHub AI Credits。

Steering an agent while it works

Agent mode is interactive. While Copilot is working, you can submit a follow-up prompt to redirect the agent, and confirm or reject each terminal command it proposes.

If a task is large or ambiguous, consider drafting an implementation plan first. See Using plan mode in your IDE.

Choosing a model or a custom agent

Before you submit a task, you can change which AI model the agent uses, or select a custom agent tailored to a specific kind of work.

Extending agent mode with tools

You can extend the agent with Model Context Protocol (MCP) servers, which add tools for working with external systems and with GitHub itself.

To hand a self-contained subtask to a separate agent with its own context, see Using subagents in your IDE.

Further reading

当你有特定任务需要处理并且希望 Copilot 能够自主编辑你的代码时，请使用代理模式。 在代理模式下，Copilot 会确定要更改的文件、提供代码更改和终端命令来完成任务，并进行迭代来修正问题，直到原始任务完成。

代理模式最适用于以下用例：

  • 任务复杂，涉及多个步骤、迭代和错误处理。
  • 希望让 Copilot 确定完成任务所需的步骤。
  • 任务要求 Copilot 与外部应用程序（如 MCP 服务器）集成。

Using agent mode

  1. If it is not already displayed, open the Copilot Chat window by clicking Editor in the menu bar, then clicking GitHub Copilot then Open Chat.
  2. At the bottom of the chat panel, select Agent from the agents dropdown.
  3. Optionally, add relevant files to the working set view to indicate to Copilot which files you want to work on.
  4. Submit a prompt. In response to your prompt, Copilot streams the edits in the editor, updates the working set, and if necessary, suggests terminal commands to run.
  5. Review the changes. If Copilot suggested terminal commands, confirm whether or not Copilot can run them. In response, Copilot iterates and performs additional actions to complete the task in your original prompt.

使用代理模式时，你输入的每个提示都会消耗 GitHub AI Credits。

Steering an agent while it works

Agent mode is interactive. While Copilot is working, you can submit a follow-up prompt to redirect the agent, and confirm or reject each terminal command it proposes.

If a task is large or ambiguous, consider drafting an implementation plan first. See Using plan mode in your IDE.

Choosing a model

Before you submit a task, you can change which AI model the agent uses. See Changing the AI model for GitHub Copilot Chat.

To hand a self-contained subtask to a separate agent with its own context, see Using subagents in your IDE.

Further reading

当你有特定任务需要处理并且希望 Copilot 能够自主编辑你的代码时，请使用代理模式。 在代理模式下，Copilot 会确定要更改的文件、提供代码更改和终端命令来完成任务，并进行迭代来修正问题，直到原始任务完成。

代理模式最适用于以下用例：

  • 任务复杂，涉及多个步骤、迭代和错误处理。
  • 希望让 Copilot 确定完成任务所需的步骤。
  • 任务要求 Copilot 与外部应用程序（如 MCP 服务器）集成。

Using agent mode

  1. Open the Copilot Chat panel by clicking the Copilot icon () in the status bar at the bottom of Eclipse, then clicking Open Chat.
  2. At the bottom of the chat panel, select Agent from the agents dropdown.
  3. Submit a prompt. In response to your prompt, Copilot streams the edits in the editor, updates the working set, and if necessary, suggests terminal commands to run.
  4. Review the changes. If Copilot suggested terminal commands, confirm whether or not Copilot can run them. In response, Copilot iterates and performs additional actions to complete the task in your original prompt.

使用代理模式时，你输入的每个提示都会消耗 GitHub AI Credits。

Steering an agent while it works

Agent mode is interactive. While Copilot is working, you can submit a follow-up prompt to redirect the agent, and confirm or reject each terminal command it proposes.

If a task is large or ambiguous, consider drafting an implementation plan first. See Using plan mode in your IDE.

Choosing a model

Before you submit a task, you can change which AI model the agent uses. See Changing the AI model for GitHub Copilot Chat.

To hand a self-contained subtask to a separate agent with its own context, see Using subagents in your IDE.

Further reading