LGTM - AI Code Review

Use GitHub Copilot Chat to review source code changes in Git, and catch bugs, areas for improvement, and other issues before merging. Open the command palette and select LGTM: Start Code Review... to select what to review.

Getting Started

After starting a review with LGTM: Start Code Review..., LGTM will print review comments grouped by file and sorted by severity in the Chat sidebar. Click on line numbers to jump to the location (if checked out).

Reviewing Pull Requests

If you have the GitHub Pull Requests extension installed, you can start a review by hovering over a pull request and clicking the LGTM icon:

If you have the Atlassian: Jira, Rovo Dev, Bitbucket extension installed, you can also start a review for Bitbucket pull requests directly:

Starting a Review From Chat

• Type /review followed by Return to bring up the same interactive menu as the LGTM: Start Code Review... menu command.
• You can specify git refs as arguments, e.g. /review develop main, or omit arguments to select refs interactively.

Use in Agent Mode

LGTM is also available in agent mode, so you can include it as part of your worklflow. For example, you might ask the agent to Review the current changes using #reviewStaged and fix any severe issues. Assuming you have staged changes, the agent will start a review using LGTM and then act on the review comments. Consider specifying the severity of issues to fix, e.g. ... and fix any issues with severity >= 3.

Features

• Only Copilot Required: Uses Copilot Chat for reviewing changes. Both free and paid plans are supported.
• Model Selection: Choose other language model available to VS Code via the LGTM: Select Chat Model command available in the Command Palette (press Cmd+Shift+P or Ctrl+Shift+P).
• Custom Instructions: Add custom instructions via the Lgtm: Custom Prompt setting (e.g., change the language of review comments by adding - In the final JSON output, use Spanish for the comment field.).
• Pull Request Integration: Review pull requests from the GitHub Pull Request or Bitbucket extensions.
• Agent Support: Adds tools to enable automatic reviews in agent mode:
  • #review: Reviews changes between two git references (branches, tags, or commits)
  • #reviewStaged: Reviews only staged changes in your working directory
  • #reviewUnstaged: Reviews only unstaged changes in your working directory
  • Example usage: After your changes, run all tests and run #reviewUnstaged to check your work.
• Chat Integration: Review content remains in chat history for follow-up questions by omitting @lgtm.
• Automatic Fixes: Use the Fix action on a review comment to fix it automatically. This will use the language model configured for inline chat.

Limitations

• Chat models listed under "Unsupported Models" can be selected but reviews will fail at runtime.
• Non-Copilot models (e.g. those added by directly adding an API key) may fail to work. For Claude language models, try the Claude Model Provider extension.
• Large change sets may trigger chat model rate limits. Please wait before retrying.

Data Usage

Source code changes and commit messages selected for review are sent to the chat model configured in the extension settings (default: GitHub Copilot GPT-4.1).

Contributing

Contributions are welcome! If you have ideas, bug reports, or want to help improve LGTM, please open an issue or submit a pull request on GitHub.