GitHub has updated Copilot code review with new capabilities, including the automatic resolution of comments and the generation of smart commit messages. These changes aim to streamline the code review process for developers.
Key Points
- Copilot code review now resolves its own comments automatically when a developer addresses them in a subsequent commit.
- When developers apply a code suggestion from Copilot code review, it now generates a smart commit message based on the changes.
- Copilot code review now utilizes a broader set of shell tools from the Copilot SDK to validate code, operating behind the Copilot agent firewall.
- The Lite effort level in Copilot code review now employs an ensemble of agents for reviews, rather than a single agent.
- Experiments showed that the ensemble approach increased the average number of addressed comments for high severity findings by 47%, medium by 31%, and low by 11%.
- The ensemble approach also reduced review cost by approximately 8%.
Context
According to GitHub, these updates make it easier for developers to focus on feedback that still requires attention. The expanded use of shell tools allows the review agent to perform actions such as running build commands, executing tests, and retrieving information from available tools and APIs. This provides Copilot with more ways to check its work.
Why It Matters
These updates offer developers a more efficient code review experience by automating routine tasks and providing more comprehensive analysis. The changes aim to reduce manual effort and improve the quality of feedback received from Copilot code review.
What To Do
- Note that Copilot code review will now automatically resolve comments once the underlying feedback is addressed.
- Observe the new smart commit messages generated when applying Copilot's code suggestions.
- Understand that the Lite effort level reviews are now more thorough due to the ensemble agent approach.
- Be aware that Copilot code review now uses a broader set of shell tools for validation.