/review on the PR
Run review from the GitHub conversation. The review job reads the current PR state and posts feedback without changing files, committing code, or pushing to the branch.
Ask ProPR for a review that never touches the code.
Comment /review on a pull request and ProPR runs a read-only review against the initial issue specification, previous branch changes, current diff, comment history, and changed files. It posts structured AI review comments with an overall evaluation, findings, and a Score: N/10.
Review ContextIssue + prior changes + diff
AI ReviewerClaude/Codex
Score: 8/10
Severity Findings
Actionable Comments
Product path overview
One model or several
ProPR uses the configured pr_review_model by default. Name specific model IDs to run more than one independent pass.
github.com/org/repo/pull/128
/review fable gpt56-solFindings with a score
Each review comment includes an overall evaluation, severity-grouped findings, a numeric score, review details, and a marker that later commands can recognize.
Works on any pull request
Commands run on PRs ProPR opened and on ones it didn't. An allowed author can drop /review or /fix on an existing or hand-written PR and refine it the same way.
Review before refinement
Review produces the critique. Refinement turns the review comments you keep into commits on the same branch.
Focus instructions
Text below /review becomes extra guidance, so a reviewer can ask ProPR to focus on security, error handling, tests, or a risky area of the diff.
Edit before fixing
Delete unwanted review comments, rewrite vague suggestions, then run /fix to process the remaining unhandled AI review comments.
Traceable results
Review output stays connected to the task record, pull request, model choice, execution details, token usage, and any later follow-up commits.
First review finds concrete issues
After the fix, review verifies the branch again
Turn selected findings into branch updates.
After /review, /fix gathers unprocessed AI review comments, applies the feedback in one implementation pass, and marks those comments as handled. For stronger cleanup, the /ultrafix command alternates review and fix cycles until the run hits the target score or cycle limit.