pr-ai-review-loop

Automates the pull request review-fix-push loop until all AI reviewers pass.

4.3k|859|Updated Feb 7, 2026
One-click install
npx skills add https://github.com/ArcReel/ArcReel --skill pr-ai-review-loop
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pr-ai-review-loop
Source: https://github.com/ArcReel/ArcReel/tree/main/.agents/skills/pr-ai-review-loop
Command: npx skills add https://github.com/ArcReel/ArcReel --skill pr-ai-review-loop

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve?

Driving a pull request through multiple AI reviewers (CodeRabbit, Gemini Code Assist, OpenAI Codex) means repeatedly polling for new comments, judging which are actionable, fixing code, pushing, and re-triggering reviews — a slow, error-prone manual cycle. This Skill runs that loop unattended until every reviewer passes or a convergence exit condition triggers.

Core Features & Use Cases

  • Unattended review convergence: Polls reviewer state via poll.sh, batches actionable comments, applies fixes following code-review discipline (YAGNI, deduplication), pushes once per batch, and waits with wait.sh between rounds.
  • Per-reviewer decision rules: Encodes trigger, reviewed-current-HEAD, actionable, and pass criteria for CodeRabbit, Gemini, Codex, and GitHub code scanning bots (CodeQL quality/security), including quota, rate-limit, and cold-start fallback handling.
  • Convergence guardrails and retrospective: Exits on round limits, diminishing returns, repeated-topic escalation, or full pass — producing a structured retrospective with ADR, CONTEXT, and follow-up issue candidates.
  • Use Case: After pushing a PR, ask the agent to drive it through AI review convergence; it will fix actionable findings from all three reviewers, handle CodeQL alerts, and report when every reviewer has passed.

Quick Start

Use the pr-ai-review-loop skill to drive this pull request through AI review convergence until all reviewers pass.

Frequently Asked Questions about pr-ai-review-loop

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

FAQPage Schema
How do I automate AI code review fixes on a GitHub pull request?

Run the loop on a non-draft PR: it polls reviewer state with poll.sh, collects actionable comments from CodeRabbit, Gemini, and Codex, applies fixes in one batch, pushes, and repeats until all reviewers pass or a convergence exit condition is met.

Which AI reviewers does the pull request review loop support?

It supports CodeRabbit, Gemini Code Assist, and OpenAI Codex as participating reviewers, plus GitHub Code Quality and Advanced Security code scanning bots as exit-gate checks. Pure metrics bots like codecov are excluded from the loop.

Can the review loop handle CodeRabbit rate limits and quota errors?

Yes. It detects CodeRabbit rate-limit banners and quota alerts, waits for self-recovery, retries with a manual trigger once, and disables that reviewer for the PR if it stays blocked, recording the decision in the exit report.

When does the automated review loop stop running?

It exits when all reviewers pass and CodeQL gates are clean, or pauses for user decision after three rounds, two low-value push rounds, repeated reviewer topics across three HEADs, reviewer conflicts, or failures like unresponsive bots and gh authentication errors.

What tools does the review loop require to run?

It requires the GitHub CLI (gh) authenticated with repo scope, jq for JSON processing, and bash. The scripts poll.sh, query.sh, wait.sh, and classify_commits.sh orchestrate all GitHub API interactions.