Let the PR run through review and fix cycles.

/ultrafix is ProPR's stronger correction loop for pull requests that need more than one small follow-up. It alternates AI review and implementation passes, waits for the branch to be ready between steps, and stops at the target score, the cycle limit, or a human stop signal.

Gather context
Execute
Refine the PR

Feature overview

One command. Every pass stays in the thread.

This run starts with a human command, turns review findings into a follow-up commit, checks the branch again, and leaves the final merge with the maintainer.

A human starts the loop in the pull request

A human entering the Ultrafix command on public ProPR pull request 1613.

ProPR records the goal and starts with review

The ProPR bot acknowledging Ultrafix and reporting the review goal and cycle limit.

The first review scores the branch 7/10 and names blockers

A detailed ProPR AI review with an overall evaluation, blocking findings, additional review notes, and a score of 7 out of 10.

The fix pass applies the pending findings on the same branch

The ProPR Ultrafix fix pass reporting its follow-up commit, resolved review findings, and passing verification.

The second review reaches the 8/10 goal

The next ProPR AI review confirming the fixed blockers, passing verification, and a score of 8 out of 10.

Checks pass and the pull request is ready for a decision

The GitHub merge area showing all checks passed and the pull request ready to merge.

The maintainer merges the verified branch

The GitHub pull request timeline confirming that the maintainer merged the verified branch into main.
The complete review, fix, verification, and merge record on public propr #1613.

One PR command

Comment /ultrafix when the branch needs an automated cleanup loop. ProPR starts with review unless unprocessed AI review comments are already waiting, in which case it starts by fixing them.

Configurable target

Set the desired score, maximum cycles, pause, and review model with /ultrafix goal=8 max=5 pause=60 model=<model-id>, or use the configured defaults.

Visible stop control

The visible ultrafix PR label controls the loop. Remove it and the loop stops after the current cycle completes.

Review, fix, wait, repeat.

/reviewscored findings
Fixapplies pending findings
Waitchecks, pause, PR quiet
Stopscore goal · max cycles · label removed
↩ next cycle until a stop condition is met

Review the current diff

ProPR runs the read-only /review path and posts structured AI review comments with findings and a Score: N/10.

Apply pending findings

The fix step gathers unprocessed AI review comments, applies the selected feedback in an isolated worktree, and marks those review comments as handled.

Wait for readiness

Before continuing, ProPR waits for required checks, the configured pause, and PR inactivity. The next cycle starts only after the branch is ready.

Stop at the boundary

The loop ends when the review score reaches the goal, the maximum cycle count is exhausted, or someone removes the ultrafix label.

  • Best for cleanup passes

    Reach for Ultrafix when the PR needs repeated review-fix pressure; a single direct change belongs in a normal PR comment.
  • Works with review scoring

    Ultrafix reads the latest valid Score: N/10 from AI review comments to decide whether the loop has hit its target.
  • Keeps branch history intact

    Fixes land as follow-up commits on the same PR branch, so reviewers can inspect each pass in normal GitHub history.
  • Records the loop

    Loop state, completed actions, review counts, fix counts, final score, and completion reason stay attached to the ProPR task history.
A focused ProPR task list showing parent pull-request tasks and their indented review, fix, and branch-update child tasks.
The task list keeps each review or fix pass grouped under its pull request.

Review stops after the critique.

Review is read-only. It posts the findings, score, model details, and next-step guidance without changing the branch.

Ultrafix turns the critique into a loop.

/ultrafix chains review and fix until it reaches the configured stopping condition. It is the automated version of review, select feedback, fix, and review again.