ai-station-navigator

[English](#en)  |  [中文](#cn)
<span id="en"></span>
# AI Station Navigator     
> **Agent System Bus and Scheduler based on Claude Code (Kernel Logic Core)**

<br>

<div align="center">

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Platform](https://img.shields.io/badge/Platform-Windows%20%7C%20macOS-0078D6?logo=windows&logoColor=white)](https://microsoft.com)
[![Release](https://img.shields.io/github/v/release/canishowtime/ai-station-navigator?label=Release&color=blue)](https://github.com/canishowtime/ai-station-navigator/releases)

[![Python](https://img.shields.io/badge/Python-3.8%2B_(Portable)-3776AB?logo=python&logoColor=white)](https://python.org)
[![Node.js](https://img.shields.io/badge/Node.js-Bundled-339933?logo=nodedotjs&logoColor=white)](https://nodejs.org)
[![Git](https://img.shields.io/badge/Git-Embedded-F05032?logo=git&logoColor=white)](https://git-scm.com)

[![Powered By](https://img.shields.io/badge/Powered%20By-Windows%20Terminal-4D4D4D?logo=windows-terminal&logoColor=white)](https://github.com/microsoft/terminal)
[![Claude Code](https://img.shields.io/badge/Integration-Claude%20Code-D97757?logo=anthropic&logoColor=white)](https://anthropic.com)

</div>

<br>

**AI Station Navigator** is a modular AI workstation built on the Claude Code engine. Mimicking the principles of computer organization, it routes complex AI tasks to **Sub-Agents** and matches them with corresponding skills for execution. The project integrates an "App Store-style" skill management system and a sandboxed execution environment. Paired with a fully portable, installation-free runtime, it aims to provide users with an unzip-and-play, stable, and infinitely scalable personal AI intelligence hub.

** ✅ Agent Context Optimization | ✅ App Store-style Skill Management | ✅ Excellent UI | ✅ Sandbox Isolation | ✅ Skills Security Scanning | ✅ Modular Architecture**


https://github.com/user-attachments/assets/248fea17-4de9-4f6f-9d54-ea7c4e8ffc9c


---

## 🎯 Core Design Philosophy: AI Workstation Architecture

The project references computer organization principles to transform AI capabilities into stable, scalable system services:

### 🏗️ Architecture Analogy

| Computer | AI Station | Role & Function Description |
| --- | --- | --- |
| **CPU** | **LLM** | **Computing Power**: Responsible for driving capabilities. |
| **System Kernel** | **Claude Code + CLAUDE.md** | **Core Logic Layer**: Responsible for intent recognition, instruction scheduling, task decomposition, and context management. |
| **System Processes** | **Sub-Agents (worker/skills)** | **Execution Layer**: Sub-agents isolate the running of single applications or scripts, **reducing context pollution for the main agent**. |
| **Applications (Apps)** | **Skills (GitHub Repos)** | **Function Plugin Layer**: Implements "App Store-style" one-click installation and invocation via GitHub links. |
| **System Drivers** | **MCP + Hooks** | **Extension & Automation**: MCP provides external system extensions; Hooks drive system automation (logs/space/status). |
| **Monitor** | **Windows Terminal / macOS Terminal** | **Information Output**: Provides status display and information output. |
| **Runtime Environment** | **Portable Environment** | **Underlying Support**: Integrated portable versions of Python, Node.js, and Git. Ensures a highly unified environment and enhances potential scalability. |

---

## ✨ Core Features

* 🧠 **Key Highlights**
* **Convenient Environment Startup**: Simply double-click the script to start the environment; ready to use after a quick configuration.
* **One-Click App Installation**: Supports installing skills directly via GitHub repository links, supporting various skill project types.
* **Session Isolation**: Through task routing, Sub-Agents run scripts or skills independently, protecting the main dialogue Context from being overwhelmed by redundant data.
* **Immersive Interactive Terminal**: Visual interface based on modern terminals, balancing professionalism with ease of use (default light theme).
* **Build Basic Workflows**: Achieve serial execution of multiple skills combined into a workflow through task decomposition.
* **Extension & Automation**: MCP connects to external systems (e.g., AI search engines); Hooks provide automation support.
* **Environment Sandbox**: The tool runs entirely within a sandbox, ensuring it does not affect global system settings. Dedicated spaces are also configured within the agents.
* **Skills Security Detection**: Integrates the [Cisco Skill Scanner](https://github.com/cisco-ai-defense/skill-scanner) developed by Cisco AI Defense. It automatically detects potential security risks after installing skills.

---

<div align="center">
  <img src="demo.gif" width="800" alt="演示动图" />
</div>

---

## 📂 Directory Structure

```text
ai-station-navigator/
├── .claude/                    # System Configuration (Registry)
│   ├── agents/                 # Sub-Agent Definitions (Processes)
│   ├── skills/                 # Installed Apps (App Center) 
├── bin/                        # System Core Scripts (Kernel Components)
│   ├── skill_manager.py        # Skill Manager (App Store Entry)
│   ├── mcp_manager.py          # MCP Driver Manager
│   └── hooks_manager.py        # Automation Hooks Manager
├── docs/                       # System Documentation (Manuals)
├── mybox/                      # Sandbox Workspace (Personal Space)
│   ├── workspace/              # Task Processing Center
│   └── output/                 # Final Output Export
├── CLAUDE.md                   # Kernel Logic Core (System CPU)
└── requirements.txt            # Python Dependencies

```

---

## 🚀 Quick Start

### 1. One-Click Launch

Download the **[All-in-One Package](https://github.com/canishowtime/ai-station-navigator/releases)** to achieve zero-configuration operation:

#### Windows Users
1. **Launch**: Double-click `Start.bat` in the root directory.
2. **Ready**: Follow the on-screen prompts to install missing components and input your self-prepared `LLM-API-KEY` to enter the startup state.

#### macOS Users

**Installation Steps:**

1. After extracting the downloaded zip file (double-click to extract), open the built-in "Terminal" application and navigate to the extracted directory:
   ```bash
   cd ~/Downloads/AI-Station-navigator
   ```

2. Run the installation script:
   ```bash
   bash unpack.sh
   ```

**Launching the Application:**

After installation, for first-time use, **right-click 'start.command' and select 'Open'**; subsequent runs can be done by double-clicking 'start.command'.

**Notes:**

- A new terminal window will open with a custom theme applied on first launch
- If prompted about security during launch, right-click 'start.command' and select 'Open'
- If the system prompts you to install Git during first run, please follow the system instructions to complete the installation (typically requires installing Xcode Command Line Tools)
- **Do not run this project in directories with Chinese characters or spaces in the path**

### 2. Intelligent Management (Chat as Command)

Enter the following instructions directly into the chat box to manage and run skills via **Sub-Agents** (Sub-processes), effectively reducing context pollution for the main agent:
(The system has a built-in GitHub network accelerator to solve network issues with Git source retrieval. You can paste the original address or path directly. It can be a main project or a specific sub-skill).

* **Check Capabilities**: `What skills do you have now?`
* **Install App**: `Install skill: https://github.com/xxx/repo` (Automatically performs installation. If the main project is a skill package, it is recommended to point the address path correctly to the specific skill you need; otherwise, the entire skill package will be installed).
* **Use App**: `@@Skill [Requirement Content]` (Automatically analyzes the requirement, matches installed skills

[truncated…]