I continued my code-review research with GitHub Copilot and Cursor Bugbot today. I wanted to know whether their reviews could act as a merge gate. The useful part was separating what the reviewer says from what GitHub enforces.
GitHub's Copilot approval documentation says every review includes an assessment of whether the PR is ready to approve. That assessment alone doesn't count towards required approvals. Enabling Copilot approvals lets it submit an approving review that does count. New commits dismiss that approval, so the revised code needs a fresh review.
Bugbot's check statuses work differently. Findings default to a neutral result. Requiring the check alone doesn't make those findings block the merge. Where the organisation has fail-on-unresolved-issues behaviour available, enabling it makes unresolved findings produce a failure.
| Review result | What it enforces |
|---|---|
| Copilot assessment only | Doesn't satisfy required approvals. |
| Copilot approving review, approvals enabled | Counts towards the required approval total. |
| Bugbot findings, default settings | Neutral check; findings alone don't block. |
| Bugbot findings, fail-on-unresolved enabled | Failing check; blocks when that check is required. |
For my comparison, "does code review" needs two follow-up questions: can it find the defect, and what happens to the merge when it does? The product settings answer the second. They don't establish the first.
This comparison was documentation research, with no local benchmark. Finding quality and false positives on the same changes are still open.