Code review
AI review of open pull requests, in-app and read-only.
Delivery → Code Review reviews open GitHub pull requests.
What it does
Polls open pull requests across configured repositories, reviews each, and presents the findings in AccelerUp.
Read-only. Nothing is posted to GitHub, no status check is set, and nothing blocks a merge.
Deliberately. A bot that comments on every pull request is noise, and a bot that blocks merges on a model's opinion is worse. This is a second pair of eyes you consult, not a gate.
One row per (repo, PR, head SHA)
A review is tied to the exact commit it reviewed. Pushing new commits produces a new review; the old one stays, attached to what it actually saw.
The engine
Uses the same engine as the Copilot — Claude Max via the CLI where available, the API otherwise.
Polling and on-demand
Reviews run on a poll, and can be triggered for a specific pull request.
The digest
A digest summarises review activity across repositories — useful for spotting a pull request that has been open for three weeks, which is usually a process problem rather than a code one.
Reading a review
Findings carry a file, a line and a description. Treat them as leads.
The highest-value findings are consistently the ones about missing things — an unhandled error path, a missing index, a test that was not updated. Models are better at noticing absence than at judging design, and design is what your colleagues are for.
Where this behaviour lives: backend/internal/worker/pr_review_worker.go, frontend/src/app/code-review/. If the code and this page disagree, the code is right — please fix the page.
Part of Delivery — Deployments, pipelines, releases and everything that ships code.