English | 中文

AI Station Navigator

Agent System Bus and Scheduler based on Claude Code (Kernel Logic Core)

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**

english.mp4

---

🎯 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 developed by Cisco AI Defense. It automatically detects potential security risks after installing skills.

---

---

📂 Directory Structure

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 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:

   cd ~/Downloads/AI-Station-navigator

2. Run the installation script:

   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, and executes immediately upon confirmation, returning the result).
• Uninstall App: Uninstall skill: https://github.com/xxx/repo or xxx/repo.
• Try Creating a Skills Workflow: Refer to docs\skills-mapping.md to design a workflow containing X steps, based on the flow xx,xx,xx,xx..., to be used in the xxxx scenario.

Here are some GitHub projects collected for testing: View Skills Categories

---

⚖️ Disclaimer

1. At Your Own Risk: AI-generated code or executed commands possess randomness. Please be sure to audit before authorizing the AI to delete or modify files.
2. Liability: The author assumes no responsibility for any data loss or system damage caused by the use of this tool.

---

🤝 Contribution & Support

• Author: Mazilin
• Project Homepage: canishowtime/ai-station-navigator
• Feedback Channels: GitHub Issues | Discussions

---

AI 工作站 领航员

基于 Claude Code 的智能体系统总线与调度器 (Kernel Logic Core)

AI Station Navigator 是一款基于 Claude Code 引擎构建的模块化 AI 工作站。它模仿计算机组成原理，将繁杂的 AI 任务路由至子智能体（Sub-Agents）内并匹配相应的skills执行。项目集成了“应用商店式”技能管理与沙盒化执行环境，配合全绿色的免安装运行环境，旨在为用户提供一个解压即用、性能稳定、无限扩展的个人 AI 智慧中枢。

** ✅ 智能体上下文优化 | ✅ 应用商店式技能管理 | ✅ 良好的UI | ✅ 沙盒隔离 | ✅ Skills安全扫描 | ✅ 模块化架构**

chinese.mp4

---

🎯 核心设计理念：AI工作站 架构

项目参考计算机组成原理，将 AI 能力转化为稳定、可扩展的系统服务：

🏗️ 架构类比

物理组件	软件映射	角色功能描述
中央处理器 (CPU)	LLM	算力能源：负责能力驱动。
系统内核 (Kernel )	Claude Code + CLAUDE.md	核心逻辑层：负责意图识别、指令调度、任务拆解与上下文管理。
系统进程（Processes）	Sub-Agents (worker/skills)	任务执行层：子智能体隔离运行单个应用或脚本，减少对主智能体上下文污染。
应用程序 (Apps)	Skills (GitHub 技能仓)	功能插件层：通过 GitHub 链接实现“应用商店式”的一键安装与调用。
系统驱动 (Drivers)	MCP + Hooks	扩展与自动化：MCP 提供外部系统扩展；Hooks 驱动系统自动化管理（日志/空间/状态）。
显示器 (Monitor)	Windows Terminal	信息输出：提供运行状态展示，信息输出。
集成环境 (Runtime)	Portable Environment	底层支撑：集成绿色版 Python, Node.js, Git强力底层工具，确保环境高度统一，增强潜在扩展能力。

---

✨ 核心特性

• 🧠 核心特性
• 便捷启动环境：双击启动脚本即可启动环境，快捷配置即可使用。
• 一键安装应用：支持通过 GitHub 仓库链接直接安装技能（skills）,支持多种skills项目类型。
• 会话隔离：通过任务分流，子智能体独立运行脚本或skills，保护主对话 Context 不被冗余数据淹没。
• 沉浸式交互终端：基于现代终端的可视化界面，兼顾专业感与易用性，默认浅色主题。
• 搭建基础工作流：通过任务拆解实现将多个技能组合成串行的工作流执行。
• 扩展与自动化：mcp对接外部系统，如AI搜索引擎等；hooks提供自动化支持。
• 环境沙箱：工具整体运行在沙箱内，不会影响系统全局设置，智能体内部也配置了专用空间。
• Skills安全检测：集成思科（Cisco）开发的 AI 技能安全扫描工具Cisco Skill Scanner，安装skills后自动检测潜在安全风险。

---

📂 目录结构

ai-station-navigator/
├── .claude/                    # 系统配置区 (注册表)
│   ├── agents/                 # 子智能体定义 (进程)
│   ├── skills/                 # 已安装的应用（应用中心） 
├── bin/                        # 系统核心脚本 (内核组件)
│   ├── skill_manager.py        # 技能管理器 (应用商店入口)
│   ├── mcp_manager.py          # MCP 驱动管理器
│   └── hooks_manager.py        # 自动化钩子管理器
├── docs/                       # 系统文档 (操作说明)
├── mybox/                      # 沙盒工作区 (个人空间)
│   ├── workspace/              # 任务处理中心
│   └── output/                 # 最终产物导出
├── CLAUDE.md                   # Kernel 逻辑核心 (System CPU)
└── requirements.txt            # Python 依赖

---

🚀 快速开始

1. 一键启动

下载**整合包 **后，即可实现零配置运行：

Windows 用户

1. 启动：双击根目录下的 启动.bat。
2. 就绪：按照屏幕提示安装缺失组件并输入自行准备的 LLM-API-KEY，即可进入启动状态。

macOS 用户

安装步骤：

1. 解压下载的 zip 文件（双击即可解压）后，打开系统自带"终端"应用，进入解压后的目录：

   cd ~/Downloads/AI-Station-navigator

2. 执行安装脚本：

   bash unpack.sh

启动应用：

安装完成后，首次使用需 右键'启动.command'——> 选'打开'；下次运行时可通过双击'启动.command'。

注意事项：

• 首次启动时会打开新终端窗口并应用自定义主题
• 如果启动时提示无法确认安全性，请右键点击'启动.command'选择'打开'
• 首次运行时如果系统提示需要安装 Git，请按照系统提示完成安装（通常需要安装 Xcode 命令行工具）
• 不要在中文路径和带有空格的路径运行此项目

2. 智能化管理 (对话即指令)

直接在对话框输入以下指令，即可通过 Sub-Agent (子进程) 实现技能管理与运行，有效减少对主智能体的上下文污染： （系统内已配置github网络加速器，解决git源码获取的网络问题，直接黏贴原始地址或路径即可，可以是主项目，也可以是某个子技能）

• 查看能力：你现在有哪些技能？
• 安装应用：`安装技能：https://github.com/xxx/repo （自动执行安装，如果主项目是技能包，建议地址路径正确指示到你需要单个技能，否则会安装整个技能包）
• 使用应用： @@技能 需求内容 （自动分析需求匹配已安装技能，确认后可立即执行并返回执行结果）
• 卸载应用：卸载技能：https://github.com/xxx/repo 或 xxx/repo
• 尝试创建skills工作流：参考 docs\skills-mapping.md 设计一个包含x步的工作流，以流程xx,xx,xx,xx...为准，可以用在xxxx场景 这里收藏了一些可用于测试的github项目：查看技能分类

---

⚖️ 免责声明

1. 后果自负：AI 生成的代码或执行的命令具有随机性。在授权 AI 执行删除、修改文件前，请务必审计。
2. 责任界定：因使用本工具导致的任何数据丢失或系统损坏，作者不承担任何责任。

---

🤝 贡献与支持

• 作者: 麻子林 (Mazilin)
• 项目主页: canishowtime/ai-station-navigator
• 反馈渠道: GitHub Issues | Discussions

---