一个人。25个模块。13,800+行可运行代码。 One person. 25 modules. 13,800+ lines of running code. 用逻辑驱动,用AI实现。/ Built on logic, executed by AI.
| 文件 | 内容 | 语言 |
|---|---|---|
| 📖 README.md | 系统门面介绍·双语概览 | 中文 + English |
| 🐉 CNSH-PROTOCOL.md | CNSH语言完整规范v2.0·14章·符号体系·语法·编译器·标准库 | 中文为主 |
| 🔗 CNSH-SEMANTIC.md | 语义接入规范·术语对照表·协作宣言·八条永恒铁律 | 中英双语 |
阅读顺序建议:README → CNSH-PROTOCOL → CNSH-SEMANTIC
📊 完整模块清单:见
docs/MODULE_INVENTORY.md
龙魂系统是一套 AI行为治理框架,让AI可控、可审计、可追责。
当全世界还在讨论如何治理AI的时候,龙魂已经在做了。
核心理念:逻辑是真正的引擎,AI是执行者。
三层监督 每个动作在执行前经过三个独立的监督层,无单点故障。
三色审计
- 🟢 绿色 — 正常,已批准
- 🟡 黄色 — 标记,待审查
- 🔴 红色 — 阻断,潜在违规
DNA全追踪 每个模块、每次提交、每个动作都携带唯一DNA签名。 没有匿名,一切可溯源。
完整清单与文档/接口状态见 docs/MODULE_INVENTORY.md。
| 分类 | 内容 |
|---|---|
| 📋 协议 | 治理规则、主权框架 |
| 🛠️ 技能库 | 10项龙魂技能 + 龍盾 + CNSH对齐 + 融合审计 |
| 🎛️ 操作台 | FastAPI 统一 API 与 Web UI,Skill/底座联动 |
| 📏 规则 | 执行约束、合规逻辑 |
| 📂 执行记录 | 完整行动历史 |
| 🧠 知识图谱 | 结构化智能知识库 |
| 📓 决策日志 | 决策审计追踪 |
| 📊 系统报告 | 健康、性能、状态 |
| 📖 技术文档 | 架构文档 |
| 🤖 代理人 | 自主任务执行器 |
| 🛡️ 宝宝守护者 | 安全与异常保护 |
| 版本 | 亮点 |
|---|---|
| v4.2 | 🎛️ 龍魂操作台 MVP v1.1 — 10 项 Skill + 底座能力(龍盾、CNSH、融合审计)统一 API 联动 |
| v4.1.1 | 🔐 安全热修复 — 修复18+个Electron漏洞 |
| v4.0 | 📱 移动端监控 — 15层体系,4个应用监控,AES-256-GCM加密 |
| v3.1.0 | ⚡ 第三阶段 — 10项技能完整集成,API响应 < 100ms |
推荐:用统一启动器一键管理(开机自动也走这个)
# 查看状态
python3 bin/longhun-status.py
# 启动全部常驻服务
python3 bin/longhun-launcher.py start
# 或双击桌面:龍魂状态与启动.command常驻服务会自动按依赖顺序启动:
- 龍魂脑干
:9625 - 国家数字身份认证入口
:8444 - 龍魂操作台
:9622
打开终端自动显示状态:配置已写入 .bash_profile / .bashrc / .zshrc,每次打开 Terminal 都会看到当前服务状态。
快捷别名:
龍魂状态/lh-sys— 显示状态龍魂启动/lh-start— 启动服务龍魂停止/lh-stop— 停止服务龍魂重启/lh-restart— 重启服务
开机自动由 ~/Library/LaunchAgents/com.uid9622.longhun.autostart.plist 管理,登录后自动运行。
详细报告:
把长文本(技能文档、上下文对话)压缩成可召回的"认知粒子":
# 压缩所有技能为一个编号+向量库
python3 scripts/longhun_compression_engine.py --compress-all-skills
# 压缩一段上下文
python3 scripts/longhun_compression_engine.py --compress-context "几百万字的上下文..."
# 通过短码召回
python3 scripts/longhun_compression_engine.py --recall SKILL-longhun-dna-alig-7C0801
# 向量语义搜索
python3 scripts/longhun_compression_engine.py --search "数据库损坏" --top-k 3LongHun (龙魂 / Dragon Soul) is an AI Behavior Governance Framework designed to make AI systems controllable, auditable, and accountable.
While the world debates how to govern AI, LongHun is already doing it.
Core philosophy: Logic is the real engine. AI is the executor.
Three-Layer Supervision Every action passes through three independent oversight layers before execution — no single point of failure.
Three-Color Audit System
- 🟢 Green — Normal, approved operations
- 🟡 Yellow — Flagged, requires review
- 🔴 Red — Blocked, potential violation
DNA Full Traceability Every module, every commit, every action carries a unique DNA signature. Nothing is anonymous. Everything is traceable.
| Category | Modules |
|---|---|
| 📋 Protocols | Governance rules, sovereignty framework |
| 🛠️ Skills | 10 Dragon Skills (art, design, MCP, Slack, React...) |
| 📏 Rules | Execution constraints, compliance logic |
| 📂 Execution Logs | Full action history |
| 🧠 Knowledge Graph | Structured intelligence base |
| 📓 Decision Journal | Decision audit trail |
| 📊 System Reports | Health, performance, status |
| 📖 Tech Docs | Architecture documentation |
| 🤖 Agents | Autonomous task executors |
| 🛡️ Guardian | Security & anomaly protection |
| Version | Highlights |
|---|---|
| v4.1.1 | 🔐 Security hotfix — 18+ Electron vulnerabilities patched |
| v4.0 | 📱 Mobile monitoring — 15-layer system, 4 app monitors, AES-256-GCM |
| v3.1.0 | ⚡ Phase 3 — 10 skills complete, API response < 100ms |
- Backend: Python, FastAPI, SQLAlchemy
- Frontend: Electron, React, Vite
- Protocol: Custom CNSH governance layer
- Security: AES-256-GCM encryption
- Monitoring: 15-layer real-time + self-healing
"全世界所有AI,都应该运行在可控的协议上。" "All AI in the world should run on a controllable protocol." — UID9622,创造者 / Creator
龙魂协议的目标是成为任何AI系统都能采用的治理层——透明、可审计、根植于中国哲学智慧。
The LongHun Protocol is designed to be the governance layer that any AI system can adopt — transparent, auditable, and grounded in Chinese systems thinking (中国哲学).
- GitHub: @UID9622
- Repository: longhun-system
一个人建造。逻辑驱动。AI执行。 Built by one person. Powered by logic. Executed by AI.