Gemini CLI Development

Source: https://github.com/google-gemini/gemini-cli

Gemini CLI is Google's AI-powered command-line assistant built on Gemini models. This skill provides quick reference for effective usage, customization, and integration.

Best Practices

Getting Started

• Install via npm: npm install -g @google/gemini-cli
• Authenticate with Google account for personal use, or configure a GCP project for enterprise
• Create a GEMINI.md in project roots with project-specific context (similar to CLAUDE.md)
• Use trusted folders to grant file system access to specific directories

Effective Usage

• Be specific with requests - "Fix the login bug in auth.ts" beats "fix bugs"
• Use context files - Reference files with @filename or drag-and-drop
• Leverage checkpoints - Save conversation state for complex multi-step tasks
• Use headless mode for scripting and CI/CD integration

Model Selection

# Use Gemini 2.5 Pro (default)
gemini

# Use Gemini 2.5 Flash for faster responses
gemini --model gemini-2.5-flash

# Use Gemini 3 Pro Preview for best quality
gemini --model gemini-3-pro-preview

Custom Commands

Create reusable commands in .gemini/commands/:

<!-- .gemini/commands/review.md -->
---
description: Review code for best practices
---

Review the provided code for:
- Security vulnerabilities
- Performance issues
- Code style and readability

Provide specific, actionable feedback.

Use with: /review @src/auth.ts

Hooks

Automate workflows with hooks in .gemini/hooks.json:

{
  "hooks": [
    {
      "event": "after-tool",
      "tool": "write_file",
      "command": "prettier --write $FILE"
    }
  ]
}

Hook events:

• before-agent / after-agent
• before-model / after-model
• before-tool / after-tool
• session-start / session-end

Tools

Gemini CLI has built-in tools:

MCP Integration

Add MCP servers in settings:

{
  "mcpServers": {
    "my-server": {
      "command": "node",
      "args": ["./mcp-server.js"]
    }
  }
}

Sandboxing

• Enable sandbox for untrusted code execution
• Sandbox restricts file system access to current directory
• Network access is blocked in sandbox mode
• Use --sandbox flag or configure in settings

Cost Management

• Free tier: Google AI Studio authentication (rate limited)
• Paid tier: GCP project with Vertex AI
• Token caching reduces costs for repeated queries
• Monitor usage in Google Cloud Console

Quick Reference

Common Commands

# Start interactive session
gemini

# Start with initial prompt
gemini "explain this codebase"

# Headless mode
gemini -p "fix the typo in README.md"

# Continue previous session
gemini --continue

# Use specific model
gemini --model gemini-2.5-flash

# JSON output mode
gemini --json -p "list files in src/"

Slash Commands

Keyboard Shortcuts

Environment Variables

GOOGLE_CLOUD_PROJECT    # GCP project ID
GOOGLE_APPLICATION_CREDENTIALS  # Service account key path
GEMINI_MODEL           # Default model
GEMINI_SANDBOX         # Enable sandbox (true/false)

Documentation Index

Detailed official documentation is synced to resources/. Consult these for specifics.

Getting Started

CLI Reference

Tools

Core Architecture

Extensions

Hooks

IDE Integration

Reference

Syncing Documentation

Resources are synced from the official Gemini CLI GitHub repository. To update:

cd skills/gemini-cli
bun run scripts/sync-docs.ts