CodeRabbit Review Processor Action

Transform CodeRabbit reviews into agent-optimized markdown with 80% token reduction.

A GitHub Action that fetches, filters, and formats CodeRabbit review threads into clean, structured markdown optimized for AI agents like Codex, Claude, and others.

Features

✅ 80% Token Reduction - Reduces ~71K tokens → ~14K tokens
✅ Unresolved Only - Automatically filters to open conversations
✅ Priority Inference - Categorizes threads as P0-P3
✅ Pattern Detection - Identifies recurring issues
✅ Agent-Ready Format - Clean markdown with reply-to IDs
✅ Zero Dependencies - Uses only Python stdlib + gh CLI

Quick Start

name: Process CodeRabbit Reviews
on:
  pull_request_review_comment:
    types: [created]

jobs:
  process:
    runs-on: ubuntu-latest
    steps:
      - name: Process CodeRabbit review
        uses: benvenker/coderabbit-processor-action@v1
        with:
          pr-number: ${{ github.event.pull_request.number }}
          output-file: review_for_agent.md

      - name: Use processed review
        run: cat review_for_agent.md

Inputs

Input	Description	Required	Default
pr-number	Pull request number to process	✅ Yes	-
output-file	Output file path	No	review_processed.md
format	Output format (markdown or json)	No	markdown
repo	Repository (owner/repo)	No	Current repo
include-resolved	Include resolved threads	No	false
verbose	Enable verbose output	No	false

Outputs

Output	Description
threads-count	Number of unresolved threads processed
token-estimate	Estimated tokens in output
output-file	Path to generated file

Usage Examples

Basic Usage with Codex

name: Codex CodeRabbit Processor
on:
  pull_request_review_comment:
    types: [created]

jobs:
  codex-review:
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Process CodeRabbit review
        uses: benvenker/coderabbit-processor-action@v1
        with:
          pr-number: ${{ github.event.pull_request.number }}
          output-file: review_for_codex.md
          verbose: true

      - name: Run Codex to address feedback
        uses: openai/codex-action@v1
        with:
          openai-api-key: ${{ secrets.OPENAI_API_KEY }}
          prompt-file: review_for_codex.md

With Claude Code Action

name: Claude CodeRabbit Processor
on:
  issue_comment:
    types: [created]

jobs:
  claude-review:
    if: contains(github.event.comment.body, '@claude-review-all')
    runs-on: ubuntu-latest
    steps:
      - name: Checkout
        uses: actions/checkout@v4

      - name: Process CodeRabbit review
        uses: benvenker/coderabbit-processor-action@v1
        with:
          pr-number: ${{ github.event.issue.number }}
          output-file: review_for_claude.md

      - name: Run Claude
        uses: anthropics/claude-code-action@v1
        with:
          anthropic_api_key: ${{ secrets.ANTHROPIC_API_KEY }}
          github_token: ${{ secrets.GITHUB_TOKEN }}
          prompt: |
            Read review_for_claude.md for pre-processed CodeRabbit feedback.
            Process each thread systematically...

JSON Output for Custom Processing

- name: Process as JSON
  uses: benvenker/coderabbit-processor-action@v1
  with:
    pr-number: ${{ github.event.pull_request.number }}
    output-file: review.json
    format: json

- name: Parse and use JSON
  run: |
    jq '.threads[] | select(.priority == "P0")' review.json

Include Resolved Threads

- name: Process all threads (including resolved)
  uses: benvenker/coderabbit-processor-action@v1
  with:
    pr-number: ${{ github.event.pull_request.number }}
    output-file: full_review.md
    include-resolved: true

Use Outputs in Subsequent Steps

- name: Process review
  id: process
  uses: benvenker/coderabbit-processor-action@v1
  with:
    pr-number: ${{ github.event.pull_request.number }}

- name: Comment with statistics
  uses: actions/github-script@v7
  with:
    script: |
      github.rest.issues.createComment({
        issue_number: context.issue.number,
        owner: context.repo.owner,
        repo: context.repo.repo,
        body: `📊 Processed ${{ steps.process.outputs.threads-count }} threads (~${{ steps.process.outputs.token-estimate }} tokens)`
      })

Output Format

The action generates clean, structured markdown:

# CodeRabbit Review: PR #123

**Repository:** owner/repo
**Branch:** feature/scanner
**Unresolved Threads:** 31

**By Priority:**
- P0 (Critical): 3 threads
- P1 (Important): 15 threads
- P2 (Style): 12 threads
- P3 (Optional): 1 threads

---

## Thread 1: Type Safety Issue [P0]

**File:** `server/routes.ts`
**Lines:** 45
**URL:** https://github.com/.../pull/1#discussion_r123
**Thread ID (for replies):** PR_kwDOABCDEF4ABCDEFGH

### AI Agent Instructions (Priority)
Ensure type validation before database insertion.

### CodeRabbit's Feedback
Remove `any` cast to preserve type safety...

---

## Summary Statistics

**Recurring Patterns Detected:**
- Type safety casts (17 occurrences) → Review TypeScript strict mode

**Token Estimate:** ~14,250 tokens

Priority Classification

Threads are automatically categorized:

• P0 (Critical): security, vulnerable, broken, crash, data loss
• P1 (Important): type safety, error handling, bug, validation
• P2 (Style): documentation, refactor, naming, readability
• P3 (Optional): nitpicks, suggestions

Token Efficiency

Real-world results from PR processing:

Source	Tokens	Reduction
Raw GraphQL Response	~71,477	baseline
Processed Output	~14,425	80% ✅

Requirements

• Python 3.7+
• gh CLI (GitHub CLI) - automatically available in GitHub Actions runners

Local Usage

You can also use the processor script locally:

# Clone the action repo
git clone https://github.com/benvenker/coderabbit-processor-action.git
cd coderabbit-processor-action

# Process a PR
python3 coderabbit_processor.py \
  --pr 123 \
  --output review.md \
  --verbose

# Dry run to see statistics
python3 coderabbit_processor.py \
  --pr 123 \
  --output review.md \
  --dry-run

How It Works

1. Fetches review threads via GitHub GraphQL API
2. Filters to unresolved threads from CodeRabbit
3. Cleans removes HTML, linters, metadata (80% reduction)
4. Enhances adds priority labels, reply-to IDs, pattern detection
5. Formats generates clean markdown optimized for AI agents

Pattern Detection

Automatically identifies recurring issues:

• 3+ markdown issues → Suggests markdownlint
• 4+ type safety issues → Suggests TypeScript strict mode
• 3+ formatting issues → Suggests Prettier

Contributing

Contributions welcome! Please open an issue or PR.

License

MIT License - see LICENSE file.

Author

Created by Ben Venker

Related Projects

• Claude Code Action
• OpenAI Codex Action
• CodeRabbit AI

Support

• 🐛 Report a bug
• 💡 Request a feature
• 📖 View documentation