Gemini is the brain. Git is the backend.

📋 A template for turning any Git repo into a self-improving autonomous agent. Click "Use this template" to get started.

A simple but effective approach - using Git to interact with your agent:

• 💬 Need a task done? → Open an Issue
• 🔧 Need to give feedback? → Comment on the PR
• 🧠 Need it to learn permanently? → Merge to main
• 📜 Need to see why it decided something? → Check the commit history

Three ways to work:

• ⏰ Scheduled - Cron jobs that run autonomously
• 💬 On-demand - Mention @gemini in GitHub issues
• 🔧 Iterative - Comment on PRs with @gemini for refinements

Key features:

• Self-improving - Can propose changes to its own code
• Modular skills - Teach it new capabilities by adding files

Perfect for: Anything you can think of! Extend with MCP servers and extensions.

Example use cases:
🔍 Code Reviews • 📰 Daily Newsletter • 📝 Auto-sync Documentation • 📊 Weekly Reports • 🌤️ Weather Updates

Sample included: This template includes a working example skill that generates daily tips about Gemini CLI. Use it as a reference for building your own skills!

💬 Interactive mode: You can also use Gemini CLI in this same repo for real-time chat (npm install -g @google/gemini-cli && gemini). Autonomous for scheduled tasks, interactive for quick questions.

Click "Use this template" at the top of this page to create your own repository.

💡 Tip: You can create a private repository if you want to keep your agent's work confidential.

1. Visit Google AI Studio
2. Create API key → Copy it

Settings → Secrets and variables → Actions → New repository secret Name: GEMINI_API_KEY Value: <your-api-key>

gh secret set GEMINI_API_KEY # Paste your key when prompted

Required for agent to create PRs:

Settings → Actions → General → Workflow permissions ✓ Read and write permissions ✓ Allow GitHub Actions to create and approve pull requests

gh api -X PUT /repos/:owner/:repo/actions/permissions/workflow \ -f default_workflow_permissions=write \ -F can_approve_pull_request_reviews=true

Open an issue (any title) with @gemini:

@gemini create a skill called "ai-news-digest" that generates a daily report of interesting AI news using Google Search. Include a daily demand that runs each morning. 

The agent will create a PR with the skill structure, guidelines, and demand file. Review, merge, and you're done! 🎉

cd gemini-cli-git && gemini

This system uses Git history as persistent memory:

Skills (memory/skills/)

• Long-term knowledge (how to do things)
• Reusable expertise and guidelines
• Quality standards and best practices

Learnings (memory/learnings/)

• Insights from past executions
• Patterns and optimizations discovered
• Lessons to apply in future runs

Conversations (memory/conversations/)

• Multi-turn interaction context
• Maintains continuity across requests
• Tracks state for complex workflows

Every commit is a memory snapshot. The agent can:

• ✅ Learn from past mistakes (via learnings)
• ✅ Improve skills over time (via commits to knowledge)
• ✅ Maintain conversation context (via conversations)

Cron Trigger ↓ Discover Demands ↓ For Each Demand: ↓ Read Demand File (specifies skill needed) ↓ Load Skill Knowledge ↓ Execute Demand Instructions ↓ Changes Made? ──→ Yes ──→ Create PR ↓ No ──→ Skip 

Example: gemini-cli-tips-daily demand runs daily, analyzing the Gemini CLI repository and creating a PR with an educational tip.

User Creates Issue with @gemini ↓ Parse Request ↓ Identify Skills Needed ↓ Load Skill Knowledge ↓ Execute Request ↓ Create PR ↓ Comment on Issue with PR Link 

Example: "Hey @gemini, generate a tip about using MCP servers with Gemini CLI"

User Comments "@gemini make this more concise" ↓ Checkout PR Branch ↓ Read Feedback + Current Files ↓ Apply Skill Knowledge ↓ Commit Changes to PR ↓ Comment Confirmation 

Edit .github/workflows/agent-scheduler.yml:

on: schedule: - cron: '0 9 * * 1'# 9 AM UTC every Monday

Use crontab.guru for custom schedules.

Edit skill guidelines like memory/skills/gemini_cli_tips/knowledge/GUIDELINES.md to change:

• Execution methodology
• Quality standards
• Output requirements
• Best practices

Changes apply immediately to all future executions!

Extend the agent with external integrations like Slack, databases, browser automation, and more using MCP (Model Context Protocol) servers or Gemini CLI extensions.

Note: Gemini CLI already has native Google Search and filesystem access built-in.

See .gemini/README.md for complete documentation on:

• Adding MCP servers (Slack, PostgreSQL, Google Drive, etc.)
• Installing Gemini CLI extensions
• Configuring API keys and environment variables
• Examples of common integrations

Easiest: Open an issue with @gemini

@gemini create a skill called "code-review" that checks for security issues and code quality. Also create a daily demand for it. 

The agent creates a PR with everything ready to review and merge!

Alternative methods:

cd gemini-cli-git && gemini

mkdir -p memory/skills/my-skill/{knowledge,output} cat > memory/skills/my-skill/knowledge/GUIDELINES.md <<EOF# Guidelines for my skillEOF

cat > memory/demands/my-demand.md <<EOF---skill: my-skill---Task instructions here...EOF

Title: Task Request Body: @gemini analyze our repository's PR merge time trends over the past 3 months 

@gemini this report is too technical, make it more accessible to non-developers 

--- skill: gemini_cli_tips ---

• Gemini CLI Documentation
• GitHub Actions Docs
• MCP Server Integration
• Contributing Guidelines

. ├── .github/ │ ├── workflows/ │ │ ├── agent-scheduler.yml # Executes scheduled demands │ │ └── agent.yml # Handles @gemini mentions │ └── actions/ │ └── run-gemini-cli-cached/ # Custom reusable action with caching (~40s faster) │ ├── memory/ # 🧠 Agent's long-term memory │ ├── demands/ # 📋 Specific tasks to execute │ │ ├── example-scheduled.md # Template demand (replace me!) │ │ └── gemini-cli-tips-daily.md # Example: Daily Gemini CLI tips │ │ │ ├── skills/ # 🎯 Specialized knowledge │ │ ├── example_skill/ # Template skill (replace me!) │ │ │ ├── knowledge/ # Pure expertise │ │ │ │ └── GUIDELINES.md │ │ │ └── output/ # Generated content │ │ │ │ │ └── gemini_cli_tips/ # Example: Daily Gemini CLI tips │ │ ├── knowledge/ │ │ │ └── GUIDELINES.md # How to analyze and generate tips │ │ └── output/ # Generated tips │ │ │ ├── context/ # 📚 Global shared knowledge │ │ ├── GEMINI.md # Agent entrypoint & routing │ │ └── README.md │ │ │ ├── learnings/ # 💡 Extracted insights │ │ └── README.md │ │ │ └── conversations/ # 💬 Interaction history │ └── README.md │ ├── CONTRIBUTING.md # How to contribute └── README.md # This file 

This is a template repository. Contributions should improve the template for everyone, not add specific skills or use-cases.

We welcome:

• Framework improvements
• Bug fixes
• Documentation enhancements
• Developer experience improvements
• New template features

See CONTRIBUTING.md for guidelines.

Apache License 2.0 - Use freely for your projects. See LICENSE for details.

gemini-cli-git - Gemini is the brain. Git is the backend.

Built with Gemini CLI