# Roborev Refine

> Automated review-fix loop that keeps iterating until all code reviews pass. Use for hands-off code quality improvement.

- Skill: `machow/roborev-refine` (Agent Skill)
- Install (CLI): `npx skillmds@latest add machow/roborev-refine`
- Raw SKILL.md: https://api.skillmd.com/api/skills/machow/roborev-refine/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: machow (https://skillmd.com/u/machow)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/machow/roborev-refine

---


# Roborev Refine Skill

Automatically fix code review findings in a loop until everything passes.

## What it does

1. Finds failed reviews on the current branch
2. Runs an agent to fix findings
3. Commits the fixes
4. Waits for re-review
5. If still failing, repeats (up to 10 iterations by default)
6. Once per-commit reviews pass, runs a branch-level review
7. Addresses any branch-level findings too

The agent runs in an isolated worktree - your working directory stays clean.

## Prerequisites

- Must be in a git repository
- Working tree must be clean (commit or stash changes first)
- Must be on a feature branch (or use `--since` on main)

## Usage

Run the refine command:

```bash
roborev refine
```

Or with options:
- `roborev refine --since HEAD~5` - Only refine last 5 commits
- `roborev refine --max-iterations 5` - Limit retry attempts
- `roborev refine --fast` - Use faster (less thorough) reasoning

## Instructions

1. Check if the working tree is clean: `git status --porcelain`
   - If not clean, inform user they need to commit or stash first

2. Run: `roborev refine $ARGUMENTS`

3. Stream the output to the user - this can take a while as it iterates

4. Report the final result:
   - How many iterations it took
   - Whether all reviews now pass
   - Summary of changes made

## Related Skills

- `/roborev-review` - Run a single review (no auto-fix)
- `/roborev-address` - Manually address specific findings