gemini-cli（Gemini CLI）

Gemini CLI Skill

Gemini CLI is an open-source AI agent that brings Gemini models directly into the terminal. It supports interactive REPL sessions, headless/non-interactive scripting, MCP servers, custom slash commands, and extension-based workflows.

Reference files (read when needed):

• references/commands.md — slash commands, built-in commands reference
• references/configuration.md — settings.json, GEMINI.md, environment variables
• references/mcp-and-extensions.md — MCP server setup, extensions authoring
• references/headless-and-scripting.md — non-interactive mode, automation, CI/CD patterns

Installation

# Instant use, no install
npx @google/gemini-cli

# Global install (recommended)
npm install -g @google/gemini-cli

# macOS/Linux via Homebrew
brew install gemini-cli

# Specific channels
npm install -g @google/gemini-cli@latest    # stable (weekly Tuesdays)
npm install -g @google/gemini-cli@preview   # preview (weekly, less vetted)
npm install -g @google/gemini-cli@nightly   # nightly (daily builds)

Authentication

Choose one method:

Option 1: Google OAuth (recommended for individuals)

gemini   # → choose "Login with Google" → browser flow

• Free: 60 req/min, 1,000 req/day
• No API key needed

Option 2: Gemini API Key

export GEMINI_API_KEY="your_key_here"
# Get key: https://aistudio.google.com/apikey
gemini

• Free: 1,000 req/day (Gemini Flash/Pro mix)
• Can also store in ~/.gemini/.env or ./.gemini/.env

Option 3: Application Default Credentials (ADC)

gcloud auth application-default login
gemini

• Uses Google Cloud ADC — no API key needed
• Best for developers already using Google Cloud

Option 4: Vertex AI (enterprise)

export GOOGLE_GENAI_USE_VERTEXAI=true
export GOOGLE_CLOUD_PROJECT="your-project-id"
gcloud auth application-default login
gemini

Basic Usage

# Start interactive session in current directory
gemini

# Include extra directories as context
gemini --include-directories ../lib,../docs

# Use a specific model
gemini -m gemini-2.5-flash
gemini -m gemini-2.5-pro

# Non-interactive: single prompt, then exit
gemini -p "Explain the architecture of this codebase"

# Reference files in prompt with @ syntax
gemini -p "Review @./src/auth.py for security issues"

# Pipe stdin
cat error.log | gemini -p "What went wrong here?"
git diff --cached | gemini -p "Write a concise commit message"

Headless / Non-Interactive Mode

Headless mode is triggered by -p flag or non-TTY environment.

# Plain text output (default)
gemini -p "Explain Docker" > output.txt

# Structured JSON output (recommended for scripting)
gemini -p "Explain Docker" --output-format json

# Streaming JSONL (for long-running tasks)
gemini -p "Run tests and analyze results" --output-format stream-json

# Extract response field with jq
gemini -p "List top 5 Python testing frameworks" --output-format json | jq -r '.response'

# Auto-accept all tool actions (auto-approve mode) — use with care in automation
gemini -p "Generate unit tests for @./src/utils.js" --auto-approve

JSON output schema

{
  "response": "...",
  "stats": {
    "models": { "gemini-2.5-pro": { "tokens": {...}, "api": {...} } },
    "tools": { "totalCalls": 1, "totalSuccess": 1 },
    "files": { "totalLinesAdded": 0, "totalLinesRemoved": 0 }
  }
}

For full scripting patterns → read references/headless-and-scripting.md

Interactive Session: Key Commands

GEMINI.md — Project Context Files

Create GEMINI.md at project root (or ~/.gemini/GEMINI.md for global context):

# My Project

## Guidelines
- All Python code must follow PEP 8
- Use 2-space indentation for JavaScript

## Architecture
@./docs/architecture.md

## Style Guides
@./src/frontend/react-style-guide.md

Hierarchy (all are loaded and merged):

1. ~/.gemini/GEMINI.md — global user context
2. <project>/.gemini/GEMINI.md — project context
3. Sub-directory GEMINI.md files — scoped context

Custom Slash Commands (.toml)

Create .toml files to define reusable commands:

Locations:

• ~/.gemini/commands/<name>.toml → user-scoped /name
• <project>/.gemini/commands/<name>.toml → project-scoped /name
• <project>/.gemini/commands/git/commit.toml → namespaced /git:commit

Minimal example (~/.gemini/commands/plan.toml):

prompt = "Only plan the changes step-by-step. Do NOT start implementation yet."

Full example (~/.gemini/commands/review.toml):

description = "Code review with focus on security and performance"
prompt = """
Review the following code for:
1. Security vulnerabilities
2. Performance issues
3. Code style

Code to review: {{args}}
"""

With shell execution:

description = "Review staged git changes"
prompt = "Review these git changes for issues:\n!{git diff --cached}"

For MCP prompts as slash commands → read references/mcp-and-extensions.md

settings.json Configuration

Location: ~/.gemini/settings.json (global) or <project>/.gemini/settings.json

{
  "model": {
    "name": "gemini-2.5-pro"
  },
  "theme": {
    "name": "dark"
  },
  "autoAccept": false,
  "tools": {
    "core": ["read_file", "write_file", "run_shell_command"],
    "disabled": ["dangerous_tool"]
  },
  "mcpServers": {
    "github": {
      "command": "npx",
      "args": ["-y", "@modelcontextprotocol/server-github"],
      "env": {
        "GITHUB_TOKEN": "$GITHUB_TOKEN"
      }
    }
  }
}

For full configuration reference → read references/configuration.md

MCP Server Integration

Add to ~/.gemini/settings.json:

{
  "mcpServers": {
    "myServer": {
      "command": "node",
      "args": ["./mcp-server.js"],
      "env": { "API_KEY": "$MY_API_KEY" },
      "timeout": 15000,
      "trust": false
    }
  }
}

CLI management commands:

gemini mcp add <name>     # add server
gemini mcp list           # list configured servers
gemini mcp remove <name>  # remove server

Invoke tools in prompts:

> @github List my open pull requests
> @database Find inactive users from last 30 days

For detailed MCP setup and extensions → read references/mcp-and-extensions.md

Extensions

Extensions bundle MCP servers + GEMINI.md + custom commands into a reusable package.

Install from URL:

gemini extensions install https://github.com/GoogleCloudPlatform/cloud-run-mcp

Browse gallery: https://geminicli.com/extensions/

Extension structure:

my-extension/
├── gemini-extension.json    # manifest
└── GEMINI.md                # context (optional)

// gemini-extension.json
{
  "name": "my-extension",
  "version": "1.0.0",
  "mcpServers": {
    "my-server": { "command": "node my-server.js" }
  },
  "contextFileName": "GEMINI.md",
  "excludeTools": ["run_shell_command"]
}

Common Patterns

Commit message automation

result=$(git diff --cached | gemini -p "Write a concise commit message" --output-format json)
echo "$result" | jq -r '.response' | git commit -F -

Batch file processing

for file in src/**/*.py; do
  gemini -p "Generate docstrings for @$file" --output-format json | jq -r '.response' > "${file%.py}_docs.md"
done

Security code review

cat src/auth.py | gemini -p "Review for security vulnerabilities" > security-review.txt

Generate OpenAPI spec

result=$(cat api/routes.js | gemini -p "Generate OpenAPI spec" --output-format json)
echo "$result" | jq -r '.response' > openapi.json

CI/CD pipeline integration

# In GitHub Actions / CI script
export GEMINI_API_KEY="${{ secrets.GEMINI_API_KEY }}"
gemini -p "Analyze test failures in @./test-results.xml and suggest fixes" \
  --output-format json \
  --auto-approve \
  | jq -r '.response'

Model Selection

gemini -m gemini-2.5-pro    # most powerful, 1M context
gemini -m gemini-2.5-flash  # fast, efficient (default routing)

Models available through Gemini API: gemini-2.5-pro, gemini-2.5-flash, gemini-2.0-flash, etc. For Vertex AI: additional enterprise models available.

Troubleshooting

Full troubleshooting: https://geminicli.com/docs/resources/troubleshooting/