Mastering Copilot: Preventing Unwanted Edits and Boosting Engineering Efficiency

GitHub Copilot is a powerful AI assistant, but its 'agentic' features—where it can autonomously modify code or even open pull requests—can sometimes hinder rather than help, especially in collaborative environments like Codespaces. Developers often seek granular control to ensure Copilot acts as a helpful assistant, not an uninvited editor. This community insight explores the various settings and strategies to rein in Copilot's autonomous actions, ensuring your workflow remains efficient and intentional.

Understanding Copilot's Agentic Modes

The core of the issue lies in Copilot Chat's different operational modes. Unlike inline suggestions, which are passive, Copilot Chat can operate in:

• Ask Mode: Purely conversational, designed to answer questions and provide suggestions without altering files.
• Edit/Agent Mode: These modes allow Copilot to directly modify files based on your prompts. This is where unwanted changes originate.
• Copilot Coding Agent (Cloud Agent): A more autonomous feature designed to take a task and produce a full pull request, often without direct real-time user interaction.

By default, or through accidental selection, Copilot Chat might default to an 'Edit' or 'Agent' mode, leading to unexpected code alterations and even automatically generated pull requests.

Comprehensive Solutions for Codespaces and Beyond

1. Global VS Code User Settings (for Codespaces Sync)

For persistent control across all your Codespaces, configure these settings at the VS Code User level. This ensures they sync with your GitHub account.

Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P) and select Preferences: Open User Settings (JSON). Add the following:

{
  "chat.disableAIFeatures": true,
  "chat.agent.enabled": false
}

• "chat.disableAIFeatures": true: This is a broad switch that disables and hides built-in AI features, including Copilot chat and agent-style editing.
• "chat.agent.enabled": false: This specifically disables Copilot agents, preventing them from making direct code changes.

2. Configure Copilot Chat Mode and Approvals

Within the Copilot Chat panel in your Codespace, look for specific UI controls:

• Chat Mode Selector: Near the chat input box, ensure the mode is explicitly set to "Ask". This prevents new chats from starting in an edit-capable mode.
• Approval Settings: Look for a permissions dropdown (sometimes labeled "Default Approvals" or similar). Set this to "Default Approvals" or "Manual / Ask Every Time". This ensures Copilot prompts you before applying any changes, rather than auto-approving them. Avoid "Bypass Approvals" or "Autopilot (Preview)" if you want full control.
• Uncheck Edit Tools: If available, in the agent's tool sets, you can explicitly uncheck the edit tool to provide an extra layer of protection against file modifications.

3. Disable Copilot Coding Agent in GitHub Settings

The Copilot Coding Agent (or Cloud Agent) operates at a higher level and can initiate PRs. To disable this:

1. Go to your GitHub homepage.
2. Click your profile picture > Settings.
3. In the left sidebar, select Copilot.
4. Find the Copilot Coding Agent setting and turn it off.

Organizations and enterprises may also have policies governing access to the cloud agent, so check with your admin if needed.

4. Enable Settings Sync

To ensure your preferences persist across all your Codespaces, enable VS Code Settings Sync:

1. Click the Gear (Settings) icon in the bottom-left corner of your Codespace.
2. Select Turn on Settings Sync.
3. Enable syncing for Settings and sign in with your GitHub account.

5. Repository-Specific Controls

• Custom Chat Instructions: Create a .github/copilot-instructions.md file in your repository with standing instructions like "never edit files directly, only suggest changes." This provides repo-wide guidance.
• .devcontainer/devcontainer.json: If Copilot is being auto-installed, check your devcontainer.json and remove GitHub.copilot from the extensions list.

Conclusion

By combining these settings—from global VS Code user preferences to in-chat controls and GitHub account settings—you can regain full control over Copilot's behavior in Codespaces. This prevents unwanted code changes and auto-generated pull requests, allowing developers to harness AI's power while maintaining critical engineering efficiency and oversight. Taking these steps transforms Copilot into a more predictable and truly assistive tool, aligning with your development workflow.