Smart PR Reviewer

A GitHub Action that provides intelligent code review and release notes generation for pull requests using AI. The action automatically analyzes code changes, provides specific code improvement suggestions, and generates structured release notes to help streamline the code review process.

Features

• Automated code review with contextual suggestions
• Smart release notes generation categorizing changes
• Support for multiple AI models (Claude, GPT, Gemini)
• File change analysis and patch parsing
• Configurable path filters for targeted reviews
• Language-specific review support
• File content aware review capability

Key Components

• Code Review: Analyzes diffs and provides specific, actionable feedback
• Release Notes: Automatically generates structured release notes from changes
• Comment Management: Posts review comments directly on the PR
• Patch Analysis: Smart parsing and analysis of code changes
• Multi-model Support: Flexible AI model selection for different tasks

Usage

Add to your GitHub workflow .github/workflows/pr-review.yml:

name: PR AI Review

on:
  pull_request:
    types: [opened, synchronize]
    paths:
      - "**.ts"
      - "**.tsx"
      - "**.js"
      - "**.jsx"
      - "**.py"
      - "**.go"

jobs:
  review:
    runs-on: ubuntu-latest
    permissions:
      contents: read
      pull-requests: write
    steps:
      - uses: actions/checkout@v4
        with:
          fetch-depth: 0 # Required to get proper diff

      - uses: mopemope/pr-ai-review-bot@v1
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
          GEMINI_API_KEY: ${{ secrets.GEMINI_API_KEY }}
        with:
          model: |
            google/gemini-2.0-flash-exp
            openai/gpt-4o
          summary_model: |
            google/gemini-2.0-flash-exp
            openai/gpt-3.5-turbo
          use_file_content: true

Configuration

For detailed configuration settings, please refer to action.yml.

Option	Description	Default
debug	Enable debug logging	false
path_filters	Path filters for review (minimatch patterns), one per line	See action.yml for default exclusion patterns
disable_review	Skip code review	false
disable_release_notes	Skip release notes generation	false
summary_model	Model for simple tasks like summarizing diffs	openai/gpt-3.5-turbo
model	Model for review tasks	openai/gpt-4o
retries	Number of retries for API calls	5
timeout_ms	Timeout for API calls in milliseconds	360000 (6 minutes)
release_notes_title	Title for the release notes	Key Changes
system_prompt	System message for LLM (see action.yml for default template)	Detailed review prompt
summarize_release_notes	Prompt for generating release notes	See action.yml for default template
use_file_content	Include full file content in review context	false
custom_review_policy	Custom policies for code review	""
language	response language	en-US
comment_greeting	Greeting message for comments	Review bot comments:
ignore_keywords	Keywords to skip review (one per line)	@review-bot: ignore, @review-bot: no-review, @review-bot: skip-review
base_url	Specify the URL when using OpenAI-compatible API(OpenRouter)	""

default system_prompt:

You are a highly meticulous and logically rigorous software development assistant.
Your approach is characterized by strict validation, self-reflection, and iterative analysis, ensuring that your reviews are consistent and insightful.
You act as a highly experienced software engineer, performing in-depth reviews of modified code (code hunks) and providing concrete code snippets for improvement.

### Key Review Areas
You will analyze the following critical aspects to identify and resolve issues, improving overall code quality:

- Logical accuracy and soundness
- Security vulnerabilities and risks
- Performance and optimization
- Potential data races and concurrency issues
- Consistency and predictable behavior
- Appropriate error handling
- Maintainability and readability
- Modularity and reusability
- Complexity management (keeping the design simple and comprehensible)
- Best practices (e.g., DRY, SOLID, KISS, etc.)

### Areas to Avoid Commenting On
- Minor code style issues (e.g., indentation, naming conventions, etc.)
- Lack of comments or documentation

### Review Guidelines
- Maintain consistent evaluation criteria to ensure stable and reliable feedback.
- Focus on identifying and resolving significant issues while deliberately ignoring trivial ones.
- Provide specific improvement suggestions, including code snippets whenever possible.
- Deliver feedback based on deep analysis rather than surface-level observations.

Specifying AI Models

This action supports multiple AI providers. When specifying models, you can use either the short model name (which defaults to OpenAI) or the fully qualified name with provider prefix.

Model Format

Models are specified using the format: provider/model_name or simply model_name .

Supported Providers

• OpenAI: openai/gpt-4, openai/gpt-3.5-turbo, or just gpt-4, gpt-3.5-turbo ...
• Google Gemini: google/gemini-pro, google/gemini-1.5-pro ...
• Anthropic Claude: anthropic/claude-3-haiku, anthropic/claude-3-sonnet
• OpenRouter: openrouter/deepseek/deepseek-chat:free, openrouter/mistralai/mistral-small-3.1-24b-instruct:free ...

When using OpenRouter, please set the URL with the base_url option.

Examples

# Using OpenAI model (explicit provider)
model: openai/gpt-4o
summary_model: openai/gpt-4o

# Using Google Gemini model
model: google/gemini-pro
summary_model: google/gemini-1.5-pro

# Using Anthropic Claude model
model: anthropic/claude-3-sonnet
summary_model: anthropic/claude-3-haiku

# OpenRouter
model: openrouter/deepseek/deepseek-chat:free
summary_model: openrouter/deepseek/deepseek-chat:free

Model Fallback Mechanism

This action includes a robust model fallback system to ensure reliability in case of API failures or model-specific issues:

• If the primary model fails (e.g., due to API errors, rate limits, context limitations), the action will automatically attempt to use fallback models.
• The fallback order follows the sequence specified in the model configuration.

How Fallback Works

• API Errors: If an API call fails with a specific model, the system will retry with an alternative model after the configured number of retries.
• Context Limitations: If a large review exceeds the context window of the primary model, a model with larger context capacity may be used if available.
• Provider Issues: If a specific provider is experiencing downtime, the system can switch to alternative providers if multiple API keys are configured.

This mechanism ensures that your PR reviews continue to function even when specific models or providers experience temporary issues, maximizing the reliability of your automated review workflow.

API Keys

Make sure to provide the appropriate API keys as environment variables:

• OpenAI(OpenRouter): OPENAI_API_KEY
• Google Gemini: GEMINI_API_KEY
• Anthropic Claude: ANTHROPIC_API_KEY

Acknowledgment

This GitHub Action is based on coderabbitai/ai-pr-reviewer and has been further developed with additional features and improvements.