GitManager

Compose AI Agents from Distributed Repos — Automatically

Cherry-pick files and folders from multiple GitHub repos. Auto-sync upstream changes. Compose unified projects — zero manual effort.

The Problem · How It Works · Use Cases · Quick Start · API

---

---

🎯 The Problem

You're building an AI agent. The best skills, tools, and reference docs are scattered across 6 different GitHub repos from different authors. You need specific folders from each — not the whole repo.

Without GitManager:

1. git clone repo-A            ← manual
2. cp repo-A/skills/pytorch    ← manual
3. git clone repo-B            ← manual
4. cp repo-B/skills/security   ← manual
5. Repeat for 6 repos...       ← tedious
6. repo-A updates pytorch skill ← you never know
7. Your agent uses stale code   ← broken

With GitManager:

Configure once → Auto-pull 6 repos → Cherry-pick 35 paths → Auto-commit & push → Repeat on schedule

Upstream fixes a bug? Your project inherits it automatically. You remove a skill? GitManager deletes it from your repo and git history. All from a dashboard. No terminal needed.

---

⚙️ How It Works

flowchart LR
    subgraph Upstreams["Upstream Repos (GitHub)"]
        A["claude-skills"]
        B["chart-viz-skills"]
        C["engineering-skills"]
    end

    subgraph GM["GitManager"]
        Pull["① git pull"]
        Forward["② Cherry-pick paths"]
        Cleanup["③ Orphan cleanup"]
        Push["④ git commit + push"]
    end

    subgraph Target["Your Project"]
        Skills["skills/"]
        Config["configs/"]
    end

    A & B & C --> Pull --> Forward --> Cleanup --> Push --> Skills & Config

Loading

Step	What Happens
① Pull	Clones or pulls latest from each upstream repo
② Forward	Copies only the folders/files you selected — not the whole repo
③ Cleanup	Removes orphaned paths when you delete a forwarding rule (including git rm)
④ Push	Auto-commits with smart messages and pushes to your repo

All steps run on a configurable schedule (e.g., every 10 minutes) in background worker threads.

---

💡 Use Cases

🤖 AI Agent Composition (primary use case)

Aggregate skills from multiple AI skill repositories into a single unified agent:

alirezarezvani/claude-skills  →  skills/python-patterns
                                  skills/pytorch-patterns
                                  skills/security

anthropics/skills             →  skills/pdf
                                  skills/docx

your-own/custom-skills        →  skills/my-custom-tool

Result: One repo powers your AI agent with the best skills from across the ecosystem — always up to date.

📦 Other Use Cases

Use Case	How
Shared Config Sync	Pull ESLint, Prettier, Dockerfile configs from a central standards repo into all your projects
Documentation Aggregation	Collect /docs/ from multiple microservice repos into a single documentation site
Design System Distribution	Sync UI components from a design system repo to multiple product repos
Open Source Curation	Cherry-pick utilities from open source projects without forking entire repos
Multi-Vendor Integration	Pull deliverables from vendor repos into your main project — auto-sync on updates

---

✨ Features

Feature	Description
🗂️ Multi-Project	Manage unlimited projects, each with its own upstreams, forwards, and schedule
🔄 Live Upstream Sync	Auto-pull or clone upstream repos — always track the latest changes
📁 Selective Path Forwarding	Cherry-pick specific folders/files — not the whole repo
🧹 Orphan Cleanup	Remove a forward rule → files are deleted from disk AND git index
⏰ Background Scheduler	Per-project sync interval with hot-reload — config changes apply instantly
📊 Smart Commits	Auto-generated commit messages grouped by upstream source

---

⚡ Quick Start

1. Clone and install

git clone https://github.com/mdnaimul22/GitManager.git
cd GitManager
pip install -r requirements.txt

2. Configure credentials

cp .env.example .env

GM_USERNAME=admin
GM_PASSWORD=your_secure_password
GM_SECRET_KEY=any_long_random_string

3. Run

python main.py

Open http://localhost:8000 — that's it. 🎉

---

🔍 Step-by-Step Guide

1. Create a Project

Click + in the sidebar. Give it a name and the absolute path to your local Git repository.

2. Add Upstreams

For each source repository, define:

• Name — a label (e.g., claude-skills)
• URL — the GitHub clone URL
• Branch — which branch to track (default: main)
• Path — where to clone it locally

GitManager auto-clones on first run if the path doesn't exist.

3. Define Path Forwards

Select exactly which paths to copy from each upstream:

FROM: ~/.claude-skills/skills/python-patterns
  TO: ~/my-project/skills/python-patterns

Toggle individual forwards on/off without deleting them.

4. Run

Set the sync interval and click Run. The background worker handles everything automatically.

---

📁 Architecture

GitManager/
├── main.py                  # FastAPI server + graceful shutdown
├── src/
│   ├── config/              # Settings, paths, file utilities
│   ├── schema/              # Pydantic models (single source of truth)
│   ├── core/
│   │   ├── watcher.py       # Hot-reload config watcher (mtime-based)
│   │   ├── pool.py          # Background worker pool (per-project threads)
│   │   ├── rate_limiter.py  # Rate limiting + scanner auto-ban
│   │   └── resolver.py      # {REPO_ROOT} placeholder resolver
│   ├── providers/
│   │   └── git.py           # Low-level git command wrapper
│   ├── services/
│   │   ├── project.py       # Project CRUD (thread-safe)
│   │   ├── upstream.py      # Pull / clone upstream repos
│   │   ├── forward.py       # Selective copy + orphan cleanup + git rm
│   │   ├── commit.py        # Smart commit message generation
│   │   └── sync.py          # Full sync orchestrator
│   └── routers/
│       ├── auth.py          # HMAC-signed session auth
│       └── projects.py      # REST API + worker control
├── static/                  # Single-page dashboard (AlpineJS + 9 themes)
├── data/                    # Per-project JSON config (gitignored)
└── tests/                   # 67 pytest tests

---

🌐 API Reference

All endpoints require session cookie authentication.

Method	Endpoint	Description
POST	/api/auth/login	Login with username & password
POST	/api/auth/logout	Clear session
GET	/api/auth/check	Validate current session
GET	/api/projects	List all projects
POST	/api/projects	Create a new project
GET	/api/projects/{id}	Get project with full config
PUT	/api/projects/{id}	Update project config
DELETE	/api/projects/{id}	Delete a project
POST	/api/projects/{id}/run	Start background sync worker
POST	/api/projects/{id}/stop	Stop background sync worker

---

🧪 Testing

pytest tests/ -v

67 passed in 2.3s

Tests cover: authentication, project CRUD, worker control, rate limiting, orphan cleanup, incremental copy, hot-reload ordering, and config persistence.

---

🔒 Security

• Authentication: HMAC-SHA256 signed cookies — stateless, survives restarts
• Rate Limiting: 60 req/min per IP — configurable
• Scanner Detection: Auto-bans IPs probing for .env, .git, credentials (2 strikes → 1hr ban)
• Localhost Whitelist: 127.0.0.1, ::1, private subnets — prevents self-lockout
• No Cloud: Runs entirely on your machine — your data never leaves

---

🤝 Contributing

1. Fork the repository
2. Branch: git checkout -b feature/your-feature
3. Commit: Follow conventional commits (feat:, fix:, chore:)
4. Test: pytest tests/ must pass
5. PR: Open a Pull Request

---

📄 License

MIT License — free to use, modify, and distribute.

---

Built to solve a real problem — keeping AI agents alive with the latest skills from across the ecosystem.

If this solves your problem too, give it a ⭐ on GitHub!