← 返回 Skills 市场
aaroncxxx

Empire Architecture

作者 aaroncxxx · GitHub ↗ · v2.0.1 · MIT-0
cross-platform ⚠ suspicious
132
总下载
0
收藏
1
当前安装
11
版本数
在 OpenClaw 中安装
/install empire-architecture
功能描述
基于中国古代三公九卿制的 AI 多智能体协作系统。 25节点帝国架构,支持并行自检、知识层、安全审计。 包含丞相协调器、参谋部、执行部、六部、翰林院、锦衣卫。
使用说明 (SKILL.md)

帝国架构 Empire Architecture v2.0

基于中国古代三公九卿制的 AI 多智能体协作系统

概述

帝国架构是一个多智能体协作系统,采用三公九卿制组织架构,将复杂任务分解为多个专业 Agent 协同完成。

核心架构

  • 中枢:丞相 — 总协调,任务分解与调度
  • 参谋:谋略/技术/情报 — 分析决策
  • 执行:文曹/码曹/查曹 — 任务执行
  • 六部:吏/户/礼/兵/刑/工 — 专项管理
  • 翰林:翰林学士/国子监 — 知识管理
  • 特殊:钦天监/太医/御厨 — 预警/诊断/清洗
  • 监察:御史大夫/中书令 — 质量/流程
  • 扩展:大理寺/大鸿胪/少府 — 逻辑/翻译/创意
  • 安全:锦衣卫 — 安全审计

运行版本

lite/ 目录包含可运行的 Python 实现(25节点核心 + 537节点天文台集群)。

快速开始

# 进入 lite 目录
cd lite/

# 配置 LLM 凭据
export MIMO_API_KEY="your-api-key"
export MIMO_API_ENDPOINT="your-endpoint"

# 运行主程序
python3 main.py

# 运行并行自检
python3 selfcheck_v17.py

配置

编辑 lite/config.json 配置帝国节点:

{
  "agents": [...],
  "knowledge": {...},
  "security": {...}
}

知识层

支持多种知识源:

  • 本地 RAG(local_rag)
  • 腾讯云(tencent_cloud)
  • 飞书(feishu)
  • Notion(notion_kb)
  • 社区知识库(community)

自检框架

V1.7 并行自检框架,6类15项检查,0.06s 完成:

  • 网络连通性
  • API 可用性
  • 文件系统完整性
  • 配置一致性
  • Agent 状态
  • 知识层健康度

安全机制

  • 锦衣卫投票制
  • 违规三级分类
  • 制衡机制
  • Token 管理:皇帝唯一所有权、吏曹执行分配

联邦学习

支持横向/纵向联邦:

  • 参数聚合
  • 安全多方计算
  • Flower 框架集成

技术栈

层面 选型
模型 MiMo(小米)
编排 LangGraph
通信 Redis + RabbitMQ
联邦学习 Flower
安全 Vault + TenSEAL
监控 Prometheus + Grafana

文件结构

├── SKILL.md          # 本文件
├── README.md         # 项目说明
├── CHANGELOG.md      # 变更记录
├── docs/             # 文档归档
│   ├── CHANGELOG-legacy.md
│   ├── architecture-v1.md
│   └── evaluation-v1.4.md
└── lite/             # 可运行代码
    ├── main.py       # CLI 入口
    ├── chancellor.py # 丞相协调器
    ├── selfcheck_v17.py
    ├── config.json
    ├── agents/
    ├── core/
    ├── knowledge/
    └── observatory/

链接

安全使用建议
This skill contains runnable Python code that implements the advertised multi-agent system, but the registry metadata omits environment/credential requirements that the README/SKILL.md and code actually expect. Before installing or running: 1) Treat the GitLab/ClawHub links and the unknown owner as low provenance — verify upstream if possible. 2) Inspect the code (already present) for any endpoints you don't trust. The runtime will read environment variables (MIMO_API_KEY, MIMO_API_ENDPOINT) and may fall back to ~/.openclaw/openclaw.json — do not expose production/privileged API keys there. 3) Expect the skill to create a local sqlite DB (data/tokens.db) and to make outbound HTTP calls to the configured LLM endpoint and to any enabled knowledge connectors (Tencent/Feishu/Notion). 4) If you want to trial it, run it in a sandbox or isolated environment (container/VM) and provide only test credentials or a vetted endpoint. 5) If you require this skill, ask the publisher to update registry metadata to list required env vars and document which external services/endpoints the skill will call and which local paths it reads/writes. If you are uncomfortable with external network calls or home-directory file reads, do not run the skill with real credentials.
功能分析
Type: OpenClaw Skill Name: empire-architecture Version: 2.0.1 The 'Empire Architecture' bundle is a sophisticated multi-agent orchestration framework themed around ancient Chinese bureaucracy. It features a modular design including a central orchestrator (Chancellor), a message bus for inter-agent communication, SQLite-based token tracking, and a comprehensive 'Jinyiwei' security auditing system designed to detect data leaks and unauthorized operations. While the system accesses sensitive configuration files (e.g., `~/.openclaw/openclaw.json` in `core/config.py`) and environment variables to retrieve LLM credentials, this behavior is functionally necessary for the agents to operate within the OpenClaw ecosystem. The code is well-documented, lacks obfuscation, and the 'Jinyiwei' component acts as a defensive layer against potential agent misbehavior.
能力标签
requires-sensitive-credentials
能力评估
Purpose & Capability
The repository contains a full multi-agent implementation (chancellor, agents, knowledge connectors, security, token tracker) that matches the described 'Empire Architecture'. However the declared registry metadata lists no required environment variables or credentials, while SKILL.md and the code require LLM credentials (MIMO_API_KEY, MIMO_API_ENDPOINT) and support multiple external knowledge integrations (Tencent Cloud, Feishu, Notion, etc.). The omission of required env/config declarations in metadata is an incoherence and reduces transparency.
Instruction Scope
SKILL.md instructs the user to export MIMO_API_KEY and MIMO_API_ENDPOINT and to run python main.py / selfcheck_v17.py. That matches the code's expected runtime. The code, however, will also: (1) silently try to load ~/.openclaw/openclaw.json as a fallback for credentials, (2) mount multiple knowledge connectors (tencent_cloud, feishu, notion_kb) which may request additional secrets when enabled, and (3) perform network calls to the configured LLM endpoint. These behaviors go beyond the literal short 'quick start' steps and establish file reads and outbound network communication that the metadata did not enumerate.
Install Mechanism
There is no external install/download step in the registry spec (instruction-only), and all code is bundled in the skill. No external URL downloads or extract/install instructions were found in the manifest. Risk from installation is therefore low, but the code will create local files/DB and make outbound connections at runtime.
Credentials
The skill requires LLM API credentials to operate (SKILL.md and core/config.py expect MIMO_API_KEY and MIMO_API_ENDPOINT) but the registry metadata declared no required env vars or primary credential — a mismatch. Additionally, the code reads ~/.openclaw/openclaw.json as a fallback, and includes connectors for Tencent Cloud / Feishu / Notion which would need further credentials if used. The skill will also write a local sqlite DB (data/tokens.db) for token tracking. Requesting LLM credentials is appropriate for this functionality, but the metadata omission and potential access to other credential-bearing files is disproportionate to the registry declaration and should be disclosed.
Persistence & Privilege
The skill does not request 'always: true' and does not claim to modify other skills or global agent settings. It will create local state (data/tokens.db) and configuration under the project directory and may read ~/.openclaw/openclaw.json; these are expected for a running multi-agent system. Autonomous model invocation is enabled by default (normal for skills) but is not combined here with abnormal privileges like forced always-on behavior.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install empire-architecture
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /empire-architecture 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v2.0.1
初始版本
v2.0.0
归一版: 版本号收敛v1.x→v2.0; 文件结构重组(changelog合并/memory移除/skills移除); 架构图新增; SKILL.md重写; docs/归档旧文档; 代码不变
v1.93.0
SKILL.md 更新至 V1.91 天文台专用版
v1.92.0
V1.91: 天文台专用版,512 Agent / 64 计算单元,异构架构,昼夜错峰调度,三级质量校验,影子Agent故障恢复
v1.91.0
V1.91: 天文台专用版,512 Agent / 64 计算单元,异构架构,昼夜错峰调度,三级质量校验,影子Agent故障恢复
v1.9.0
V1.8: 新增观星台 Observatory,25节点,全局行为观测与长周期趋势分析
v1.8.1
V1.8: 新增观星台 Observatory,25节点,全局行为观测与长周期趋势分析
v1.8.0
V1.8: 新增观星台 Observatory,25节点
v1.7.1
Major refactor and reorganization in v1.7.1: - Migrated all code out of the "lite" subdirectory into the main structure for improved maintainability. - Added 27 new files, introducing updated core, agent, and knowledge modules. - Removed 69 legacy and "lite" subdirectory files. - Updated documentation to reflect the new structure and usage. - No changes to core features—functionality and agent roles remain the same.
v1.7.0
Major update: Complete expansion to a 24-node multi-agent system with performance improvements. - Upgraded from 8 nodes to 24 complete agents mapped to "三公九卿" imperial system - Introduced parallel self-check framework (huge speedup: 133s → 0.06s) - Enhanced knowledge layer: 8 academicians + 9 knowledge sources - Added asynchronous message bus and SQLite token tracking - Security audit by Jinyiwei agent - Updated documentation for new architecture and quick start
v1.5.0
v1.5: 知识层运行器 + 评估报告。基于三公九卿制的 AI 多智能体协作系统,8核心节点+翰林院9大学士+皇帝审批机制。
元数据
Slug empire-architecture
版本 2.0.1
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 11
常见问题

Empire Architecture 是什么?

基于中国古代三公九卿制的 AI 多智能体协作系统。 25节点帝国架构,支持并行自检、知识层、安全审计。 包含丞相协调器、参谋部、执行部、六部、翰林院、锦衣卫。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 132 次。

如何安装 Empire Architecture?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install empire-architecture」即可一键安装,无需额外配置。

Empire Architecture 是免费的吗?

是的,Empire Architecture 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Empire Architecture 支持哪些平台?

Empire Architecture 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Empire Architecture?

由 aaroncxxx(@aaroncxxx)开发并维护,当前版本 v2.0.1。

💬 留言讨论