Google Cloud credits are provided for this project #VertexAISprint

Thanks, Google :)

Gemini AI Code Reviewer

A GitHub Action that automatically reviews pull requests using Google's Gemini AI.

Features

• Review your PRs using Gemini API
• Give use comments and suggestions to improve the source codes

Video Demo

Setup

1. To use this GitHub Action, you need an Gemini API key. If you don't have one, sign up for an API key at Google AI Studio.

2. Add the Gemini API key as a GitHub Secret in your repository with the name GEMINI_API_KEY. You can find more information about GitHub Secrets here.

3. Create a .github/workflows/code-review.yml file in your repository and add the following content:

Option 1: Automatic Review on Pull Request Events (Recommended)

This configuration automatically triggers code review when a PR is opened, updated, or reopened:

name: AI Code Reviewer

on:
  pull_request:
    branches: [ main ]

permissions:
  issues: write
  pull-requests: write
  contents: read

jobs:
  gemini-code-review:
    runs-on: ubuntu-latest
    steps:
      - name: PR Info
        run: |
          echo "PR Number: ${{ github.event.pull_request.number }}"
          echo "Repository: ${{ github.repository }}"
          echo "Head SHA: ${{ github.event.pull_request.head.sha }}"
          echo "Base SHA: ${{ github.event.pull_request.base.sha }}"

      - name: Checkout Repo
        uses: actions/checkout@v3
        with:
          fetch-depth: 0

      - name: Get PR Details
        id: pr
        run: |
          echo "head_sha=${{ github.event.pull_request.head.sha }}" >> $GITHUB_OUTPUT
          echo "base_sha=${{ github.event.pull_request.base.sha }}" >> $GITHUB_OUTPUT

      - uses: Thomasbehan/gemini-ai-code-reviewer@main
        with:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
          GEMINI_MODEL: gemini-2.5-pro # Optional, default is `gemini-2.5-flash`
          EXCLUDE: "*.md,*.txt,package-lock.json,*.yml,*.yaml"
          SYSTEM_PROMPT: | # Optional: Custom system prompt for code reviews
            Review the code with the following guidelines:
            - Focus on security vulnerabilities and potential bugs
            - Check for proper error handling
            - Ensure code follows best practices
            - Verify performance considerations

Option 2: Manual Trigger via Comment

This configuration requires manual triggering by commenting /gemini-review on a pull request:

name: Gemini AI Code Reviewer

on:
  issue_comment:
    types: [created]

permissions: write-all

jobs:
  gemini-code-review:
    runs-on: ubuntu-latest
    if: |
      github.event.issue.pull_request &&
      contains(github.event.comment.body, '/gemini-review')
    steps:
      - name: PR Info
        run: |
          echo "Comment: ${{ github.event.comment.body }}"
          echo "Issue Number: ${{ github.event.issue.number }}"
          echo "Repository: ${{ github.repository }}"

      - name: Checkout Repo
        uses: actions/checkout@v3
        with:
          fetch-depth: 0

      - name: Get PR Details
        id: pr
        run: |
          PR_JSON=$(gh api repos/${{ github.repository }}/pulls/${{ github.event.issue.number }})
          echo "head_sha=$(echo $PR_JSON | jq -r .head.sha)" >> $GITHUB_OUTPUT
          echo "base_sha=$(echo $PR_JSON | jq -r .base.sha)" >> $GITHUB_OUTPUT
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

      - uses: Thomasbehan/gemini-ai-code-reviewer@main
        with:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
          GEMINI_MODEL: gemini-2.5-pro # Optional, default is `gemini-2.5-flash`
          EXCLUDE: "*.md,*.txt,package-lock.json,*.yml,*.yaml"
          SYSTEM_PROMPT: | # Optional: Custom system prompt for code reviews
            Review the code with the following guidelines:
            - Focus on security vulnerabilities and potential bugs
            - Check for proper error handling
            - Ensure code follows best practices
            - Verify performance considerations

Configuration Options

Required Parameters

• GITHUB_TOKEN: GitHub token to interact with the repository (automatically provided by GitHub Actions)
• GEMINI_API_KEY: Your Google Gemini API key (store this as a GitHub Secret)

Optional Parameters

Noise Control and Limits

• MAX_COMMENTS_TOTAL: Optional hard cap on the total number of review comments posted per run (default: 0 = disabled).
• MAX_COMMENTS_PER_FILE: Optional cap on the number of comments per file (default: 0 = disabled).
• REVIEW_PRIORITY_THRESHOLD: Only post comments at or above this priority. One of: low, medium, high, critical. Example: high.
• REVIEW_MODE: Controls strictness. Options: strict, standard, lenient, security_focused, performance_focused. For fewer comments use lenient.

Example usage in workflow inputs (maps to env):

- uses: Thomasbehan/gemini-ai-code-reviewer@main
  with:
    GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
    GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
    EXCLUDE: "*.md,*.txt,package-lock.json,*.yml,*.yaml"
    MAX_COMMENTS_TOTAL: 12
    MAX_COMMENTS_PER_FILE: 4
    REVIEW_PRIORITY_THRESHOLD: high
    REVIEW_MODE: lenient

• GEMINI_MODEL: The Gemini model to use for code review (default: gemini-2.5-flash)

  • gemini-2.5-flash is a next-generation model offering speed and multimodal generation capabilities. It's suitable for a wide variety of tasks, including code generation, data extraction, and text editing.
  • gemini-2.5-pro offers enhanced capabilities with longer context windows and better reasoning for complex code reviews.
  • For detailed information about available models, refer to Gemini models.

• EXCLUDE: Comma-separated list of file patterns to exclude from review (e.g., *.md,*.txt,package-lock.json)

• SYSTEM_PROMPT: Custom system prompt to define specific rules and guidelines for code reviews

  • If not provided, the action uses a default prompt based on best practices
  • Use this to customize the review focus (e.g., security-first, performance-focused, style-specific)
  • Example:

    SYSTEM_PROMPT: |
      You are a senior code reviewer. Focus on:
      1. Security vulnerabilities (SQL injection, XSS, etc.)
      2. Performance bottlenecks
      3. Code maintainability and readability
      4. Proper error handling
      Do not comment on minor style issues.

AI Model Precision Settings

The action is configured for maximum precision and deterministic output to provide consistent, focused code reviews:

• Temperature: 0.0 (default) - Lowest temperature for ultra-precise, deterministic responses. The model focuses on the most likely correct answers rather than exploring creative alternatives.
• Top P: 0.9 (default) - Slightly lower nucleus sampling for more focused output.
• Max Output Tokens: 8192 - Fully utilizes Gemini 2.5's output capacity for comprehensive reviews.

These settings can be overridden via environment variables if you need different behavior:

env:
  GEMINI_TEMPERATURE: "0.0"  # Range: 0.0 (deterministic) to 2.0 (creative)
  GEMINI_TOP_P: "0.9"        # Range: 0.0 to 1.0
  GEMINI_MAX_TOKENS: "8192"  # Maximum output tokens

Why these settings? Code reviews require precision and consistency. Lower temperature ensures the AI focuses on identifying real issues rather than generating creative but potentially irrelevant suggestions.

4. Commit the workflow file to your repository.

5. Usage depends on which option you chose:

   • Option 1 (Automatic): The code review will automatically run when you open, update, or reopen a pull request to the configured branch.
   • Option 2 (Manual): Comment /gemini-review on any pull request to trigger the code review.

How It Works

This GitHub Action uses the Gemini AI API to provide code review feedback. It works by:

1. Analyzing the changes: It grabs the code modifications from your pull request and filters out any files you don't want reviewed.
2. Consulting the Gemini model: It sends chunks of the modified code to the Gemini for analysis.
3. Providing feedback: Gemini AI examines the code and generates review comments.
4. Delivering the review: The Action always posts a review on your pull request:
   • If no issues are found, the PR is approved with a "Code looks good! 👍" message
   • If issues are found, the review requests changes with detailed comments on specific lines

License

This project is licensed under the MIT License. See the LICENSE file for more information.

Star History ⭐️