Autonomous Review Loop

Automate the review-fix-push cycle.

Pair your review bot with your coding agent: reviewloop delegates the entire review-fix-push cycle so you only step in when human judgment is needed.

---

Prerequisites

• Python 3.12+
• GitHub CLI (gh), authenticated with your account
• jq, for JSON parsing within bash

Installation

uv tool install reviewloop

reviewloop init

Getting Started

Claude Code

Select Claude Code during reviewloop init. Then run this within your claude code session:

/reviewloop

Any Coding Agent

Select Script based during reviewloop init. This creates standalone scripts and a prompt file at scripts/reviewloop/reviewPrompt.txt. Feed it to any coding agent (Cursor, Windsurf, etc.) with @scripts/reviewloop/reviewPrompt.txt to start the loop.

How It Works

1. Waits for CI to complete
2. Fetches inline comments and review comments
3. Fixes valid issues and asks you about ambiguous ones
4. Resolves threads and pushes
5. Repeats until no unresolved comments remain

Features

• Batched decision-making: Aggregate review requests instead of triaging comments one by one.
• Parallel work: Continue other tasks while the loop runs in the background.
• Multi-pass resolution: Iterates automatically until clean.
• Agent-agnostic: Works with Claude Code, Cursor, Windsurf, or any coding agent.

Troubleshooting

"No PR found for current branch"

Ensure you're on a branch with an open PR:

gh pr view

"gh CLI is not authenticated"

gh auth login

Skill not appearing in Claude Code

1. Check the skill file exists: .claude/skills/reviewloop/SKILL.md
2. Verify the frontmatter is valid YAML
3. Restart Claude Code

CI timeout

The default timeout is 10 minutes. If your CI takes longer, re-run with a longer timeout:

scripts/reviewloop/review-wait.sh --timeout=1200