Bugbot PR Review1 / 2

2 min lesson

Bugbot's /review table is a signal over the diff

Explain the practical point behind "Start with this check: Bugbot is reviewing the intended PR diff and findings cite concrete code paths."

The job: you can run BugbotCursor's automated PR reviewer that posts inline findings and can push fix commits from isolated VMs. Press Enter for the full definition. and interpret its output without surrendering merge judgment. Start with this check: Bugbot is reviewing the intended PR diff and findings cite concrete code paths. Keep it small enough to check the result yourself instead of taking the summary on trust.

Review the findings, fix what earned it

0:37 · narrated

Read this demo as text

Run's done. /review is already sitting in the composer. Send it. Let Bugbot read the diff.
Two findings. Medium: recent items you can't reach with a keyboard. And a Low: memoizing a five-item slice. Same table. Completely different weight.
When it offers a fix, pick the one that earned it. The Medium becomes real buttons. The nit? Just a note. Every change in the diff should trace back to a severity that mattered.

Bugbot's /review table is a signal over the diff — fix findings that reproduce; leave nits noted. Merge judgment stays with you.

Practice next: Now run the play: inspect Bugbot's PR findings and approve only fixes that reproduce.

Simulated Cursor 3.12 (macOS, light) — beta educational reconstruction, not the real product.

Learn more

Full explanation

Bugbot review over a real diff

A PR opens and Bugbot posts a review

Bugbot review over a real diff

SayI opened a PR that changes how we build a search query. Bugbot is on for this repo — what do I do with its review?

SeeBugbot posts a finding citing a concrete code path: the new query in src/search/query.ts drops a request param straight into the SQL string.

DoOpen the finding next to the diff and check the claim against the changed lines, not against Bugbot's summary.

SeeThe cited line really does concatenate user input, so the finding is real, not noise. Across Cursor's fleet Bugbot resolves north of 70-80% of the issues it raises — that is precision, the share of its flags that are real, not the share of all bugs it catches (re-verify before quoting it).

DoApply the parameterized-query fix yourself, then re-request review.

SeeYou approved this fix by hand; autofix stays off until you have trusted the pattern over several PRs.

SayThe next PR comes back with a clean Bugbot run. That is not a merge approval.

SeeA clean run only means Bugbot flagged nothing — you still own correctness, so you read the diff and merge on your own judgment. That is the proof, not the green check.

Learn more

Optional practice

Practice: Bugbot's /review table is a signal over the diff

QYou've applied it. What actually proves the work is done, not just a plausible answer?