Review with a different model.

The model that wrote the diff should not be the only one to read it. Implement with Claude Opus 5.5, review with GPT-6 Astra, or put two reviewers side by side. Then fix only the findings you keep.

Describe
Run isolated
Review in GitHub

Feature overview

You start with
A pull request one model implemented
You use
/review astra · /review astra fable · /fix
Takes
Two reviews of a 16-issue epic PR landed within 26 minutes
Real run
propr #2905

Pick a reviewer from another family

Set the default PR review model in Settings to a model that did not write the code. Every /review and every Ultrafix review uses it unless you name another.

Review settings: GPT-6 Astra reviews pull requests, and a read-only Claude Opus 5.5 scout gathers related unchanged code first
Claude Opus 5.5 implements; GPT-6 Astra reviews by default.

Ask one model for a review

Name the model on the PR. Use the model IDs from your agent configuration; the llm- prefix is optional.

@integrycommented on propr #2905

/review astra

Or run two side by side

Name two models. Each posts its own independent review, and the F# and S# numbering continues across both, so every ID names one record for the life of the PR.

@integrycommented on propr #2905

/review astra fable

@propr-dev[bot]bot

✅ AI Code Review Complete requested by @integry. Posted 2 reviews: GPT-6 Astra, Claude Fable 5.1.

Read where they disagree

On epic #2905, the two reviewers split on the same diff.

GPT-6 Astrareview

F1 Support selecting multiple assignee filters

F2 Render creators on owner-scoped lists

S1 Clarify truncated candidate guidance

Score: 6/10

Claude Fable 5.1review

No merge blockers.

S2–S8: seven suggestions, from webhook and polling parity to a missing MCP filter

Score: 8/10

Fix only what you keep

Bare /fix applies every pending merge blocker. /fix all adds every suggestion. To pick, copy the /fix line at the end of a review, delete the IDs you reject, and post it. Lines below the command become instructions for the fix.

On #2905, a Claude Opus 5.5 review ended with /fix S21 S22 S23 S24 S25 S26. The maintainer kept three:

/fix S21 S23 S25
@propr-dev[bot]bot

✅ Applied the requested follow-up changes in commit bb431bc. Addressed suggestions S21, S23, S25.

Claude wrote it. GPT-6 Astra blocked it.

Claude Opus 5.5 implemented public propr #2934. The review ran on codex:gpt-6-astra, found an ownership bug as F1, and scored the branch 6/10. Claude fixed exactly that record.

GPT-6 Astra files merge blocker F1

A ProPR AI review by codex:gpt-6-astra on public pull request 2934 with an overall evaluation and merge blocker F1, To-do updates can attribute another owner's records to the caller, listing required behavior, evidence, and minimum fix.

The review ends with a copyable /fix line

The review footer explaining /fix, /fix all, and named records such as /fix F3 S5, followed by a copyable /fix F1 S1 command.

Claude Opus 5.5 resolves F1 in commit b4e98fb

ProPR reporting that it applied the requested follow-up changes in commit b4e98fb, addressing finding F1, with a summary of the cause, fix, regression test, and verification, run by Claude Opus 5.5.
One implementer, one reviewer from another provider, on propr #2934.

What you get

Every pull request read by a model that did not write it, findings you can name by ID, and commits for exactly the ones you keep.

Review → All playbooks