gitdash

GitHub Copilot Code Review Auto-Resolves Comments: What Changes

Copilot code review now resolves addressed comments during re-review and suggests smarter commit messages for applied fixes. Here is how to use the new workflow well.

A clean pull request passes a glowing review gate while an unsafe change is stopped before the protected repository

Copilot code review has a useful new behavior: when you push a commit that addresses one of its comments, the comment can be resolved during the next review. Feedback that still applies stays open. The result is a pull request where open Copilot comments better match the work that remains.

GitHub also added smart commit message suggestions when you apply a Copilot code review suggestion. Behind the scenes, the review agent can use more shell tools to validate code, and Lite reviews now use an ensemble of agents. The official changelog lists the full update.

How auto-resolution works

Copilot's review comment is tied to feedback about the code in your pull request. You push a later commit that fixes the underlying issue, then Copilot re-reviews the pull request. If the feedback has been addressed, Copilot resolves that comment. If the change did not fix the issue, or the issue is still present elsewhere, the feedback remains open.

Review comment: add input validation
        ↓
Push fix commit
        ↓
Copilot re-review
        ↓
Addressed → comment resolves
Still present → comment stays open

This is different from manually clicking Resolve conversation. Manual resolution records that a person closed the thread. Auto-resolution is tied to the next analysis of the changed code. That makes the review state more useful, but it does not remove the need to read the comment and verify the fix.

Turn on re-reviews for new pushes

Auto-resolution only helps when Copilot reviews the new commit. By default, pushing changes does not automatically request another review. You can request a re-review manually from the Reviewers menu, or configure automatic reviews to include new pushes. The Copilot code review guide explains both paths.

A clean review loop

  1. Request the initial Copilot review after the pull request is ready for feedback.
  2. Group comments into fixes that can be tested together, rather than changing one line mechanically.
  3. Run the relevant tests and push a focused commit.
  4. Request a re-review or let the automatic new-push review run.
  5. Read the remaining open comments, then resolve or reply to anything that needs human context.

Use smart commit messages as a starting point

When you apply a suggested change from a Copilot review comment, GitHub now generates a commit message based on the change instead of filling in a generic message. Review the suggestion before committing. A useful message should say what changed and why, and it should match your repository's conventions.

Good: Validate webhook signatures before processing events
Weak: Apply suggested change

Before commit:
  ✓ Read the diff
  ✓ Run the focused test
  ✓ Check the message matches the actual change

More analysis does not remove review responsibility

Copilot code review can now use the broader shell tool set from the Copilot SDK behind the agent firewall. That gives it more ways to run builds, tests, targeted scripts, and available tool or API checks while analyzing a change. It can improve signal, but the commands still run inside Copilot's controlled environment, not your production system.

Treat every comment as evidence to assess, not as an automatic merge instruction. Confirm the data flow, test the edge case, and decide whether the recommendation fits the product. Auto-resolution tells you that Copilot believes the underlying feedback was addressed; it does not certify the design.

When Lite is enough

The Lite effort level now uses an ensemble of agents rather than one agent working alone. GitHub reports that its experiments increased the average number of addressed comments for high, medium, and low severity findings while reducing review cost by about 8 percent. Those are GitHub's experiment results, not a guarantee for every repository.

  • Use Lite for routine pull requests where targeted feedback is enough.
  • Choose a deeper effort level for cross-service changes, security-sensitive code, or complex logic.

Keep human approval requirements independent from Copilot's review setting.

The short version

Enable re-reviews for new pushes, keep commits focused, and let addressed comments disappear from the open queue automatically. Use smart messages as editable suggestions, not final truth. The new analysis tools and Lite ensemble can improve coverage, but the best review loop still ends with a human reading the diff, running the right checks, and deciding what is safe to merge.