Copilot code review now automatically closes its own comments after you fix the issues they flag, and it generates contextual commit messages when you adopt its suggestions. Under the hood, the system leverages a wider array of shell-based tools to validate reviewed code, and multiple agents working together deliver more comprehensive feedback at the Lite effort tier. These improvements streamline the review process by ensuring that open comments represent only actionable feedback, while giving Copilot additional mechanisms to verify its own analysis.

Review experience updates

When a subsequent commit resolves a Copilot code review comment, Copilot now marks that comment as resolved during its follow-up review pass. This eliminates the need for manual thread closure on obsolete feedback, allowing developers to focus on comments that require attention.

  • Comments are automatically marked resolved when a follow-up commit addresses the original concern.
  • Unresolved feedback remains visible, preventing any issues from being overlooked.

Smart commit messages on Copilot autofix suggestions

When you merge a Copilot code review suggestion, the system now proposes a commit message tailored to the specific changes being made, rather than inserting a generic default message.

Analysis updates

The following enhancements improve review quality without changing how developers request or receive reviews.

Deeper analysis with shell tools

Building on its existing file-reading capabilities, Copilot code review now taps into the complete set of shell tools available in the Copilot SDK, operating within the Copilot agent firewall. This expanded toolkit allows the review agent to validate code through additional methods such as executing build processes, running test suites, launching custom scripts, and querying accessible tools and APIs.

Testing of this enhancement revealed that developers responded more favorably to Copilot's feedback, with the system identifying a higher proportion of critical-severity issues and fewer trivial observations.

Ensemble of agents in Lite reviews

The Lite effort level now deploys multiple agents that collaborate to produce a single review, rather than relying on a single agent. Each agent contributes its own assessment of the code, and their findings are consolidated into one comprehensive review. This collaborative method produces more accurate and detailed Lite reviews, frequently at equal or reduced expense.

In testing, the ensemble model boosted the average count of addressed comments per review by 47% for critical findings, 31% for medium-level findings, and 11% for low-level findings, while cutting review expenses by roughly 8%.

Source: GitHub Blog