Gemini Review

Gemini Review Plugin

Get independent second opinions from Google's Gemini AI on your code, plans, documentation, and architecture decisions.

What's New in v1.2

Slash Commands:

• /gemini-review - Quick review with smart defaults
• /gemini-batch-review - Batch review with real-time progress

Enhanced Capabilities:

• JSON output for structured results and token tracking
• Stream-JSON for real-time progress on long reviews
• Directory context support for architectural reviews
• Improved error handling with retry logic
• Better model selection guidance

Overview

The Gemini Review plugin enables Claude Code to leverage Google's Gemini AI for independent validation and cross-checking of critical work. Use it when you need multiple AI perspectives, want to catch blind spots, or require validation before making important decisions.

Why Use This?

• Independent Validation: Get a second LLM perspective on important decisions
• Catch Blind Spots: Different models have different strengths and may notice different issues
• Cross-Validation: Verify critical code changes with multiple AI opinions
• Multiple Perspectives: Combine Claude and Gemini's insights for comprehensive reviews
• Proactive Integration: Claude knows when to autonomously consult Gemini

Prerequisites

1. Install gemini-cli

pip install gemini-cli

2. Authenticate with Gemini

Choose one method:

Option A: Google OAuth (recommended)

gemini auth login

Option B: API Key

# Get a key from https://aistudio.google.com/apikey
gemini config set api_key YOUR_API_KEY

3. Verify Setup

gemini -p "test"

If you see a response from Gemini, you're ready!

Setup time: ~5 minutes

Installation

Add the Marketplace

/plugin marketplace add https://github.com/bgreenwell/claude-plugins

Install the Plugin

/plugin install gemini-review@bgreenwell-plugins

Verify Installation

Ask Claude:

"Do you have the gemini-review skill available?"

Usage

Three Ways to Get Reviews

1. Natural Language (Most Flexible)

Simply ask Claude naturally. Claude will autonomously consult Gemini:

"Review my changes to auth.js for security issues"
"Get a second opinion on this refactoring"
"Should I use Redis or Memcached for session storage?"
"Are my tests comprehensive enough?"

2. Quick Command (Explicit Review)

Use /gemini-review for quick, targeted reviews:

# Review staged files
/gemini-review

# Review specific files
/gemini-review src/auth/login.js

# Security-focused review
/gemini-review src/auth/* --focus security

# Deep review with Pro model
/gemini-review src/api/* --model pro

Options:

• [file-pattern] - Files to review (default: git staged files)
• --model flash|pro - Model selection (default: flash)
• --focus area - Focus area (security, performance, testing, etc.)

3. Batch Review (Multiple Files)

Use /gemini-batch-review for reviewing many files with progress:

# Review multiple directories
/gemini-batch-review src/ tests/ docs/

# Security audit across modules
/gemini-batch-review src/auth src/api --focus security --model pro

# Performance review
/gemini-batch-review src/services src/utils --focus performance

Features:

• Real-time progress tracking
• Aggregated results by file
• Optimal for 5-20 files
• Shows token usage

Example Prompts

Review code changes:

"Review my changes to auth.js for security issues"
"Get a second opinion on this refactoring"

Review architectural decisions:

"Should I use microservices or monolith for this project?"
"Get Gemini's perspective on this API design"

Review documentation:

"Review this README for clarity and completeness"
"Get feedback on my architecture documentation"

Review test coverage:

"Are my tests comprehensive enough?"
"Get a second opinion on my test suite"

Security audits:

"Perform a security review of my authentication system"
"Check this code for vulnerabilities"

How It Works

When you ask for a review or second opinion:

1. Claude analyzes your code/docs and forms its own opinion
2. Claude constructs a focused prompt for Gemini
3. Claude calls gemini-cli with appropriate file references
4. Gemini reviews the content independently
5. Claude synthesizes both perspectives
6. You receive comprehensive findings with actionable recommendations

Review Types Supported

• Code Changes: Logic, performance, security, edge cases, maintainability
• Architecture: Scalability, maintainability, technology choices, trade-offs
• Documentation: Accuracy, clarity, completeness, grammar, tone
• Test Coverage: Scenarios, edge cases, quality, missing cases
• Security: Vulnerabilities, best practices, OWASP compliance
• Plans: Feasibility, risks, alternatives, completeness

Features

Slash Commands (New in v1.2)

/gemini-review

• Quick review with smart defaults
• Reviews staged files if no pattern provided
• Supports model and focus area selection
• Shows structured results with token usage

/gemini-batch-review

• Review 5+ files with real-time progress
• Stream-JSON for progress tracking
• Aggregated results by file
• Priority recommendations

JSON Output Support (New in v1.2)

• Structured, parseable output
• Token usage tracking for cost awareness
• Better result formatting
• Integration with automation workflows

Stream-JSON Progress (New in v1.2)

• Real-time progress updates for long reviews
• Batch processing visibility
• Event-based monitoring
• Optimal for multiple file reviews

Structured Review Templates

Claude has access to proven review templates for:

• Code changes review
• Plan validation
• Documentation quality checks
• Architecture assessment
• Test coverage analysis
• Custom criteria reviews

Model Selection

Claude automatically chooses the right Gemini model:

• gemini-2.5-flash: Fast, for quick reviews (default)
• gemini-2.5-pro: Thorough, for critical decisions

You can also request specific models:

"Use Gemini Pro for a thorough security audit"

Or specify via command:

/gemini-review src/auth/* --model pro

Multi-File Support

Review multiple related files together:

"Review these auth files for consistency: @src/auth.js @src/session.js @tests/auth.test.js"

Or use batch command:

/gemini-batch-review src/auth src/session tests/auth

Context-Aware Reviews

Claude automatically provides relevant context to Gemini about:

• What the code is trying to achieve
• Your specific requirements
• Constraints and decisions
• What aspects to focus on

Directory Context (New in v1.2)

For architectural reviews, provide broader codebase context:

"Review this architecture with full context"

Uses --include-directories for system-wide understanding.

Best Practices

When to Get Gemini Reviews

✅ Before merging critical code changes ✅ Important architectural decisions ✅ Complex refactoring plans ✅ Security-sensitive code ✅ Test coverage validation ✅ When you need a fresh perspective

❌ Trivial questions or simple tasks ❌ Repeatedly for the same content ❌ As a replacement for your own analysis

Providing Context

Give Claude context for better reviews:

"Review this auth refactoring. Context: We're optimizing for mobile performance and need to support offline mode."

Focusing Reviews

Specify what aspects to focus on:

"Review this API design, focusing specifically on security vulnerabilities and rate limiting"

Example Workflows

Pre-Commit Review

You: "Review these authentication changes before I commit"

Claude will:
1. Read and analyze the auth files
2. Identify potential issues
3. Get Gemini's independent review
4. Present comprehensive findings
5. Recommend specific improvements

Architecture Decision

You: "Should I use REST or GraphQL for this API?"

Claude will:
1. Consider your requirements
2. Analyze trade-offs
3. Get Gemini's perspective
4. Present both opinions
5. Make a recommendation

Security Audit

You: "Perform a security audit of my authentication system"

Claude will:
1. Review the auth code
2. Identify security concerns
3. Get Gemini Pro's security review
4. Combine findings
5. Prioritize vulnerabilities

What This Plugin Provides

• SKILL.md: Comprehensive guide for Claude on using gemini-cli
• Review Templates: Structured prompts for different review types
• Best Practices: When and how to use Gemini reviews
• Troubleshooting: Common issues and solutions
• Examples: Real-world usage patterns

Troubleshooting

"gemini-cli not found"

pip install gemini-cli
which gemini  # Verify installation

Authentication errors

# Re-authenticate via Google
gemini auth login

# Or set new API key
gemini config set api_key YOUR_NEW_KEY

# Test
gemini -p "test"

"API key invalid"

1. Generate new key at https://aistudio.google.com/apikey
2. Set it: gemini config set api_key YOUR_NEW_KEY
3. Verify: gemini config list

Timeout errors

• Review smaller sections instead of entire large files
• Use gemini-2.5-flash for faster processing
• Check your internet connection

File not found with @ references

• Use relative paths from project root
• Verify file exists: ls path/to/file
• Ensure proper file permissions

Privacy & Security

Important considerations:

• Code is sent to Google's Gemini API for review
• Don't use for proprietary/sensitive code without authorization
• Review your organization's AI usage policies
• gemini-cli manages API credentials (not this plugin)
• See Google's data policies: https://ai.google.dev/gemini-api/terms

Advanced Usage

Custom Review Criteria

"Review this code based on these criteria:
1. Performance for large datasets
2. Memory efficiency
3. Thread safety"

Inline Content Reviews

Review plans/ideas without files:

"Review this approach: I'm planning to cache user sessions in Redis with 24hr expiration. What are the risks?"

Comparative Reviews

"Compare these two implementations and recommend which is better"

What This Plugin Does NOT Do

❌ Store your API key (managed by gemini-cli) ❌ Send data anywhere except Gemini's API ❌ Work without gemini-cli installed ❌ Replace Claude's analysis (augments it) ❌ Make decisions for you (provides input)

Contributing

Found a bug or have a suggestion? https://github.com/bgreenwell/claude-plugins/issues

License

MIT License - See LICENSE file for details

Author

Brandon Greenwell

• GitHub: @bgreenwell
• Email: [email protected]

Command Reference

/gemini-review

Quick review with smart defaults.

Syntax:

/gemini-review [file-pattern] [--model flash|pro] [--focus area]

Examples:

# Review staged files
/gemini-review

# Review specific pattern
/gemini-review src/auth/*

# Security-focused
/gemini-review src/api/* --focus security

# Deep review
/gemini-review src/core.js --model pro --focus performance

Output includes:

• File-by-file findings
• Severity indicators (✓ ⚠️ ❌)
• Actionable recommendations
• Token usage

/gemini-batch-review

Batch review with real-time progress.

Syntax:

/gemini-batch-review <paths...> [--model flash|pro] [--focus area]

Examples:

# Multiple directories
/gemini-batch-review src/ tests/ docs/

# Security audit
/gemini-batch-review src/auth src/api --focus security --model pro

# Performance review
/gemini-batch-review src/services --focus performance

Features:

• Real-time progress: "Processing... [5/10] (50%)"
• Aggregated results
• Priority recommendations
• Token usage and cost estimate

Best for: 5-20 files

Changelog

v1.2.0 (Current)

• NEW: Added /gemini-review slash command for quick reviews
• NEW: Added /gemini-batch-review command for batch processing
• NEW: JSON output support for structured results
• NEW: Stream-JSON for real-time progress tracking
• NEW: Directory context support (--include-directories)
• IMPROVED: Enhanced error handling with retry logic
• IMPROVED: Better model selection guidance
• IMPROVED: Expanded examples with new capabilities
• No breaking changes - fully backward compatible

v1.1.0

• Added comprehensive SKILL.md with review templates
• Enhanced with structured prompts for different review types
• Added security audit templates
• Improved troubleshooting guide
• Added model selection guidance
• Expanded examples and best practices

v1.0.1

• Initial MCP server-based implementation
• Basic gemini_review tool

---

Enjoy having two AI perspectives on your code!