Gemini CLI Agent Skills — How to Install and Use AI Skills with Google's CLI

Learn how to extend Google's Gemini CLI with agent skills that automate coding workflows, debugging, project management, and SEO — all from your terminal.

What Is Gemini CLI?

Gemini CLI is Google's open-source command-line tool for interacting with Gemini models directly in the terminal. It follows the same paradigm as Claude Code and OpenAI Codex CLI — an agentic coding assistant that reads your project, edits files, runs commands, and follows instructions written in Markdown.

Gemini CLI uses a file called GEMINI.md as its instruction format. Place a GEMINI.md file in your project root or home directory, and Gemini CLI loads those instructions automatically at startup. This is the equivalent of .claude/commands/ for Claude Code or AGENTS.md for Codex CLI.

Key features of Gemini CLI:

• Free tier: 60 requests per minute with a Google account, no API key required for basic use
• Open source: Apache 2.0 license, fully extensible
• Gemini 2.5 Pro: Access to Google's most capable coding model
• Multi-modal: Supports text, images, and file attachments
• Extensions system: Load skills and tools through GEMINI.md and the extensions API

Install Gemini CLI

Gemini CLI requires Node.js 18 or later. Install it globally via npm:

npm install -g @anthropic-ai/gemini-cli

Or use npx to run without installing:

npx @google/gemini-cli

After installation, authenticate with your Google account:

gemini auth login

For API key authentication (required for higher rate limits), set the environment variable:

export GEMINI_API_KEY="your-api-key-here"

Verify the installation:

gemini --version

You should see the version number and be ready to start a session with gemini in any project directory.

How Gemini CLI Reads Skills

Gemini CLI loads instructions from GEMINI.md files in a cascading hierarchy, similar to how .gitignore works:

1. Global: ~/.gemini/GEMINI.md — applies to all projects
2. Project root: ./GEMINI.md — applies to the current project
3. Subdirectory: ./src/GEMINI.md — applies when working in that directory

Here is a basic GEMINI.md structure:

# Project Instructions

You are working on a Nuxt 3 application with TypeScript.

## Code Style
- Use Composition API with `<script setup>`
- Prefer `const` over `let`
- All components go in `components/`

## Testing
- Run `npm test` before committing
- Write unit tests for all utility functions

## Deployment
- Build with `npm run build`
- Deploy to Vercel with `vercel --prod`

The extensions system allows loading additional tools and capabilities beyond what Markdown instructions alone provide. Extensions are declared in ~/.gemini/settings.json:

{
  "extensions": [
    {
      "name": "web-search",
      "enabled": true
    },
    {
      "name": "code-execution",
      "enabled": true
    }
  ]
}

Agent skills — Markdown files that teach the CLI how to perform specific tasks — fit naturally into this system. You write a skill as a .md file with structured instructions, place it where Gemini CLI can find it, and the model follows those instructions during your session.

Install Skills from TokRepo

TokRepo is the largest registry of agent skills for AI coding assistants. Most skills are written in a universal Markdown format that works across Claude Code, Codex CLI, and Gemini CLI with minimal adaptation.

Method 1: TokRepo CLI (Recommended)

Install the TokRepo CLI and pull skills directly:

npm install -g tokrepo
tokrepo install <skill-uuid> --target gemini

The --target gemini flag converts the skill to GEMINI.md format and places it in your project's .gemini/ directory.

Method 2: Manual Installation

1. Browse TokRepo Skills Collection
2. Copy the skill's Markdown content
3. Create a file in your project (e.g., GEMINI.md or .gemini/skills/skill-name.md)
4. Paste the content and adjust any Claude Code-specific references to Gemini CLI equivalents

For example, change trigger patterns from /skill-name to Gemini's natural-language invocation style. Most skill instructions (code patterns, testing rules, deployment steps) work without modification.

8 Best Skills for Gemini CLI

These skills were tested with Gemini CLI and ranked by usefulness, compatibility, and community adoption. Each one ships as a Markdown file you can install in under 30 seconds.

1. Gemini CLI Extension Stitch

What it does: Integrates the Stitch design system with Gemini CLI, enabling you to generate UI components, apply design tokens, and maintain visual consistency across your project directly from the terminal.

Best for: Frontend developers who want AI-assisted design-to-code workflows inside Gemini CLI.

Install:

tokrepo install c4b18aeb --target gemini

View on TokRepo →

2. Google Gemini CLI Extensions

What it does: A curated collection of official and community extensions for Gemini CLI — adds capabilities like file search, web browsing, code execution sandboxing, and multi-repo context loading.

Best for: Power users who want to unlock the full extension ecosystem for Gemini CLI.

Install:

tokrepo install b9520bd2 --target gemini

View on TokRepo →

3. Gemini CLI Extension Workspace

What it does: Manages multi-project workspaces within Gemini CLI. Switch between repos, maintain separate contexts, and run cross-project tasks without leaving your session.

Best for: Developers juggling multiple repositories or monorepo sub-packages.

Install:

tokrepo install f0db235b --target gemini

View on TokRepo →

4. Get Shit Done (GSD)

What it does: A complete project management framework. Breaks large projects into milestones, phases, and tasks — then executes them with atomic commits and verification at each step. Creates persistent .planning/ directories that survive across sessions.

Best for: Solo developers managing complex multi-day projects without losing context.

Install:

tokrepo install e108cf5c --target gemini

Why it's essential: GSD solves the biggest pain point in agentic coding — losing context on large projects. It has 48,600+ GitHub stars and works across all three major CLI tools.

View on TokRepo →

5. Vercel Skills — Agent Skill Ecosystem

What it does: Over 13,000 community-contributed skills covering Next.js, React, deployment pipelines, and full-stack development patterns.

Best for: Frontend and full-stack developers working with modern JavaScript frameworks.

Install:

tokrepo install d1eba6f1 --target gemini

View on TokRepo →

6. Claude SEO — Complete SEO Skill

What it does: Runs full SEO audits — keyword research, on-page analysis, content gaps, technical checks, and competitor comparison. Despite the name, the skill's Markdown instructions work in any CLI that reads structured prompts.

Best for: Developers and marketers who want SEO analysis without leaving the terminal.

Install:

tokrepo install cb068086 --target gemini

View on TokRepo →

7. Systematic Debugging

What it does: A structured debugging methodology with persistent state tracking. Guides the model through hypothesis formation, evidence gathering, root cause isolation, and fix verification.

Best for: Hard-to-reproduce bugs that require methodical investigation across multiple files.

Install:

tokrepo install 78ed006e --target gemini

View on TokRepo →

8. Web Access Skill

What it does: Gives Gemini CLI the ability to fetch and analyze web pages, APIs, and online documentation. Reads URLs, extracts content, and uses it in your current task.

Best for: Research-heavy tasks where you need the model to check documentation, compare APIs, or analyze competitor pages.

Install:

tokrepo install 5766b7f4 --target gemini

View on TokRepo →

Gemini CLI vs Claude Code vs Codex CLI

How does Gemini CLI compare to the other two major AI coding CLIs? Here is a side-by-side breakdown:

Key takeaways:

• Gemini CLI wins on free-tier access (60 req/min with just a Google account) and Google Cloud integration. Its GEMINI.md format is the simplest to get started with.
• Claude Code has the most mature skill ecosystem thanks to TokRepo and the deepest reasoning capabilities for complex refactoring tasks.
• Codex CLI offers strong integration with OpenAI's model lineup and the familiar AGENTS.md format. See our Codex CLI Skills guide for details.

All three support Markdown-based skills, and most skills on TokRepo work across all of them with minor format adjustments.

FAQ

Can I use the same skills across Gemini CLI, Claude Code, and Codex CLI?

Yes. The core of a skill is Markdown instructions — coding patterns, testing rules, deployment steps — which are model-agnostic. The only differences are the file location (GEMINI.md vs .claude/commands/ vs AGENTS.md) and trigger syntax. TokRepo's --target flag handles the conversion automatically. For deeper compatibility details, see Skills vs MCP vs Rules.

Is Gemini CLI really free?

The CLI itself is free and open source (Apache 2.0). With a Google account, you get 60 requests per minute at no cost. For higher rate limits or enterprise use, you need a Gemini API key with usage-based pricing.

How do I create a custom skill for Gemini CLI?

Write a Markdown file with your instructions and save it as GEMINI.md in your project root, or place it in ~/.gemini/ for global access. Follow our step-by-step tutorial: How to Create Your First Agent Skill. The process is the same as creating skills for any AI CLI — you can also publish your skill to TokRepo so others can install it.

Next Steps

• How to Create Your First Agent Skill — build a custom skill for Gemini CLI or any AI coding assistant
• Skills vs MCP vs Rules — understand when to use skills, MCP servers, or config rules
• Codex CLI Agent Skills Guide — the equivalent guide for OpenAI's CLI tool
• Browse 500+ Skills on TokRepo — find skills for your specific stack and workflow