---

name: gemini-cli description: Execute Gemini CLI commands for AI-powered coding assistance, file analysis, and complex development tasks. Use when user wants to run gemini commands, query Gemini models, or leverage Gemini's agentic capabilities. allowed-tools:

• Bash
• Read
• Write

---

Gemini CLI Skill

Purpose

Execute Google's Gemini CLI - an open-source AI agent that brings Gemini directly into the terminal. Gemini CLI uses a ReAct loop with built-in tools and MCP servers to complete complex tasks like fixing bugs, creating features, and improving test coverage.

Activation Triggers

Use this skill when the user:

• Asks to "run gemini" or "use gemini cli"
• Wants to query Gemini models from the terminal
• Needs Gemini's agentic file/code operations
• Wants to leverage Gemini's 1M token context window
• Requests MCP server interactions via Gemini

Invocation Modes

Interactive Mode (REPL)

gemini -m gemini-3-pro-preview            # Default: use Gemini 3 Pro
gemini -m gemini-3-pro-preview --yolo     # Auto-approve all tool calls
gemini -m gemini-3-pro-preview --sandbox  # Run tools in isolated environment
gemini -m gemini-3-pro-preview --checkpointing  # Enable project snapshots

Important: Always use -m gemini-3-pro-preview for best results.

Non-Interactive Mode

gemini -m gemini-3-pro-preview -p "Explain this codebase architecture"
gemini -m gemini-3-pro-preview -p "Fix the bug in auth.ts" --yolo
gemini -m gemini-3-pro-preview -p "prompt" --output-format json
echo "code" | gemini -m gemini-3-pro-preview

Slash Commands (Interactive Mode)

Command	Function
/help	Display help
/tools	List available tools
/mcp	List MCP servers and tools
/stats	Show token usage
/compress	Summarize context to save tokens
/copy	Copy last response to clipboard
/clear	Clear screen and context
/memory show	Display GEMINI.md context
/memory refresh	Reload GEMINI.md files
/chat save <tag>	Save conversation
/chat resume <tag>	Resume saved conversation
/chat list	List saved conversations
/restore	List/restore project checkpoints
/init	Generate GEMINI.md context file
/settings	Open settings.json editor
/vim	Toggle Vim mode
/theme	Change visual theme
/ide install	Set up VS Code integration
/ide enable	Connect to VS Code
/bug	File bug report
/quit	Exit CLI

Context References (@)

Reference files/directories in prompts:

@./src/main.ts           # Single file
@./screenshot.png        # Image
@./src/                  # Directory (recursive)

Shell Commands (!)

!git status              # Execute single command
!                        # Toggle persistent shell mode

Keyboard Shortcuts

Shortcut	Action
Ctrl+L	Clear screen
Ctrl+V	Paste text/image
Ctrl+Y	Toggle YOLO mode
Ctrl+X	Open in external editor

Built-in Tools

File Operations:

• read_file, write_file, replace
• list_directory, glob
• search_file_content

Web:

• google_web_search
• web_fetch

Shell:

• Execute shell commands

Memory:

• save_memory for persistent facts

Command-Line Flags

Flag	Purpose
-m, --model <model>	Specify model (always use gemini-3-pro-preview)
-p <prompt>	Non-interactive single prompt
-i <prompt>	Interactive with initial prompt
-d, --debug	Enable debug output
--yolo	Auto-approve tool calls
--sandbox	Isolated tool execution
--checkpointing	Enable restore points
--include-directories	Multi-directory workspace

Configuration

Settings file: ~/.gemini/settings.json or .gemini/settings.json

Key settings:

• autoAccept - Auto-approve safe tool calls
• sandbox - Tool isolation mode
• vimMode - Vim-style editing
• checkpointing - Enable /restore
• mcpServers - MCP server definitions

Context file: GEMINI.md (project root or ~/.gemini/)

Custom Commands

Create .gemini/commands/<category>/<name>.toml:

description = "Generate unit test"
prompt = """
Write a comprehensive test for: {{args}}
"""

Invoke as: /<category>:<name> "arguments"

Rate Limits (Free Tier)

• Google Login: 60 req/min, 1000 req/day
• API Key: 100 req/day