om-auto-review-pr

Reviews pull requests in isolated worktrees and submits verdicts with labels and autofix loops.

2.6k|159|Updated Mar 8, 2021
One-click install
npx skills add https://github.com/go-musicfox/go-musicfox --skill om-auto-review-pr-go-musicfox
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: om-auto-review-pr
Source: https://github.com/go-musicfox/go-musicfox/tree/main/.agents/skills/om-auto-review-pr
Command: npx skills add https://github.com/go-musicfox/go-musicfox --skill om-auto-review-pr-go-musicfox

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Reviewing pull requests manually is slow and inconsistent, and automation that waits on CI or clobbers other agents' work creates bottlenecks. This Skill reviews or re-reviews any PR by number in an isolated worktree, submits an approve or request-changes verdict immediately, manages pipeline labels, and optionally runs an autonomous fix-and-re-review loop until the PR is merge-ready. ## Core Features & Use Cases - Isolated PR review: Fetches the exact PR from the tracker, checks it out in a temporary worktree, runs the full code review (or a specification review for spec-only design PRs), and submits the verdict without touching your main worktree. - Autonomous autofix loop: On eligible runs (the automation's own PRs or with --autofix), it resolves merge conflicts, fixes review findings, reruns validation, and re-reviews until the PR is merge-ready; other authors' PRs get review plus handoff only. - Pipeline label and lock management: Claims PRs with a three-signal in-progress check, transitions pipeline labels through guarded helpers, posts manual-QA instructions for needs-qa PRs, and performs a bounded CI follow-up capped by a configurable wait budget. - Use Case: Run it on PR #1234 after a teammate pushes changes; it re-checks every previous blocker, inherits unaddressed reviewer comments as findings, submits a fresh review, and either approves into merge-queue or hands the PR back with the full picture. ## Quick Start Ask the agent to review pull request 1234 with the om-auto-review-pr skill, adding the autofix flag if you want it to fix the findings itself.

Frequently Asked Questions about om-auto-review-pr

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I automatically review a pull request with an AI agent?

Invoke the skill with the PR number, for example /om-auto-review-pr 1234. It claims the PR, checks it out in an isolated worktree, runs the full code review, and submits an approve or request-changes verdict with labels and a complete report.

How to auto-fix review findings on a pull request?

Pass the --autofix flag, or run it on a PR authored by the automation user. The skill then resolves merge conflicts first, fixes all actionable findings including inherited reviewer comments, reruns validation, and re-reviews until the PR is merge-ready.

Does it wait for CI checks before submitting a review?

No, the verdict, labels, and comments are submitted immediately once the review is decided. Failing checks become blocker findings inside the review, and a bounded follow-up watches CI afterward for at most the configured ci.maxWaitMinutes, defaulting to 40.

Can it review specification or design-doc-only pull requests?

Yes, when every changed file lives under the specs or design-doc directories it runs a specification review instead of the code checklist. It evaluates five design lenses including risks, backward compatibility, gaps, improvements, and simplicity, using the same severity scale.

What happens when reviewing a fork pull request?

For fork PRs it preserves the original commits and authorship, creates a carry-forward branch in the main repository, applies fixes there, and opens a replacement PR crediting the original author. The original PR is closed only after the replacement exists.

What are the limitations of the autofix mode?

Autofix never runs on another author's PR without the explicit --autofix flag, and it never force-pushes unless the user approved it. It stops and asks for human input on ambiguous architecture decisions, missing credentials, or contract-breaking scope changes.