AI Code Review

Reusable GitHub Actions workflow for AI-powered code review with auto-merge support.

Features

• Multi-AI provider support (Gemini, OpenAI, Claude)
• Multi-language project support (any codebase)
• Security vulnerability scanning with Trivy
• Configurable approval threshold
• Auto-merge for approved PRs
• Smart diff handling (auto-approve docs-only changes)
• Multi-language review output (Chinese/English)
• Detailed review comments on PR

Quick Start

1. Add Secret

In your repository, go to Settings > Secrets and variables > Actions and add:

• GEMINI_API_KEY: Your Google Gemini API key (or OPENAI_API_KEY / ANTHROPIC_API_KEY)

2. Create Workflow

Create .github/workflows/code-review.yml in your repository:

name: Code Review

on:
  pull_request:
    branches: [main]

jobs:
  review:
    uses: your-org/CodeReview/.github/workflows/ai-code-review.yml@main
    secrets:
      AI_API_KEY: ${{ secrets.GEMINI_API_KEY }}

That's it! PRs to main will now be automatically reviewed.

Configuration

All inputs are optional:

Input	Description	Default
ai-provider	AI provider (gemini, openai, claude)	gemini
ai-model	AI model name (empty for default)	``
min-score	Minimum score for approval (0-10)	7
max-files	Max files for auto-merge	20
max-lines	Max lines for auto-merge	500
auto-merge	Enable auto-merge	true
merge-method	Merge method (merge/squash/rebase)	squash
language	Review language (zh/en)	zh
run-security-scan	Run Trivy security scan	true
exclude-patterns	Files to exclude from review	*.md,*.txt,*.json,*.lock,*.yaml,*.yml
custom-prompt	Additional review instructions	``

Default Models

Provider	Default Model
gemini	gemini-2.0-flash
openai	gpt-4o
claude	claude-sonnet-4-20250514

Full Example

name: Code Review

on:
  pull_request:
    branches: [main, develop]

jobs:
  review:
    uses: your-org/CodeReview/.github/workflows/ai-code-review.yml@main
    secrets:
      AI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
    with:
      ai-provider: 'openai'
      ai-model: 'gpt-4o'
      min-score: 8              # Stricter approval
      max-files: 10             # Smaller PRs only
      max-lines: 300
      auto-merge: false         # Review only, no auto-merge
      language: 'en'            # English comments
      run-security-scan: true
      exclude-patterns: '*.md,*.txt,docs/*'
      custom-prompt: 'Focus on error handling and edge cases'

Outputs

The workflow provides these outputs for downstream jobs:

Output	Description
score	AI review score (0-10)
approved	Whether PR passed review
merged	Whether PR was auto-merged

Using Outputs

jobs:
  review:
    uses: your-org/CodeReview/.github/workflows/ai-code-review.yml@main
    secrets:
      AI_API_KEY: ${{ secrets.GEMINI_API_KEY }}

  notify:
    needs: review
    runs-on: ubuntu-latest
    steps:
      - run: |
          echo "Score: ${{ needs.review.outputs.score }}"
          echo "Approved: ${{ needs.review.outputs.approved }}"

Review Criteria

The AI evaluates PRs on 5 dimensions (0-10 each):

1. Code Quality - Readability, complexity, naming
2. Security - Vulnerabilities, input validation
3. Performance - Algorithm efficiency, resource usage
4. Testing - Test coverage
5. Best Practices - Following conventions

Special Behaviors

Empty Diff Handling

PRs with only documentation/config changes (matching exclude-patterns) are auto-approved with score 10.

Sensitive File Filtering

Files matching these patterns are automatically excluded from AI review:

• .env*
• *secret*
• *.pem, *.key

Comment Updates

When a PR is updated, the existing AI review comment is updated instead of creating a new one.

Requirements

• GitHub Actions enabled
• AI API key (Gemini/OpenAI/Anthropic)
• Repository permissions: contents: write, pull-requests: write

License

MIT