Copilot Code Review Now Resolves Its Own Comments
GitHub's Copilot code review now auto-resolves its own review comments when a follow-up commit addresses them, writes commit messages when you apply its suggestions — and under the hood, it switched to a multi-agent ensemble at the cheapest effort tier. Shipped today, September 11, on github.com.
What changed
Three things, two visible one invisible:
- Self-resolution: when you push a commit addressing a Copilot comment, Copilot resolves that thread during its re-review. Outstanding feedback stays open. You no longer triage stale threads manually.
- Shell tools in review: the review agent can now run build commands, execute tests, and run targeted scripts behind the Copilot agent firewall. Reviews are no longer read-only analysis — they execute your code.
- Ensemble at Lite: the cheapest effort level now runs multiple agents in parallel and merges their findings. GitHub's own numbers: average addressed comments per review up 47% for high-severity findings, 31% for medium, 11% for low — with review cost down about 8%.
Why a builder cares
Two structural shifts in one changelog. First, the economics: a 47% jump in high-severity findings addressed came from redundancy, not a bigger model — parallel agents at Lite beat one agent at higher effort for less money. Expect "ensemble review" to become the default posture across every review product this quarter, and your per-PR inference bill to quietly include it.
Second, the trust shift: the reviewer now grades its own homework. When the agent that filed the comment is also the agent that decides it's resolved, "resolved" means the bot convinced itself. That closes the loop GitHub wants — open threads always reflect live feedback — but if you disagree with a Copilot finding, you're now overriding an agent that re-review says it was right. Watch the audit trail on auto-resolved threads before trusting them in compliance-sensitive repos.
And the shell access is the quiet part: a reviewer that runs your tests is one prompt-injection away from running an attacker's tests. That's the same surface we flagged in Copilot autofix back in August, now widened to every review.
FAQ
When does Copilot code review resolve its own comments?
When you push a commit that addresses a Copilot review comment, Copilot resolves that thread during its re-review. Feedback that is still outstanding remains open, so nothing gets lost.
How much better are Copilot's ensemble reviews?
GitHub reports the ensemble approach at the Lite effort level increased addressed comments per review by 47% for high-severity findings, 31% for medium, and 11% for low, while cutting review cost by about 8%.
Does Copilot code review run code during review?
Yes. It now uses the full shell tool set from the Copilot SDK behind the Copilot agent firewall — running builds, tests, and targeted scripts to validate the code under review.