三省六部

🏛️ 三省六部制 · OpenClaw Multi-Agent Orchestration System — 9 specialized AI agents with real-time dashboard, model config, and full audit trails. Use when: (1) 需要多智能体协作完成复杂任务, (2) 需要实时管理和监控AI代理状态, (3) 需要配置和调度不同AI模型, (4) 需要完整的操作审计和合规日志, (5) 需要构建复杂的AI工作流编排, (6) 需要智能体之间的任务分配和协调, (7) 需要权限管理和资源分配, (8) 需要安全策略和合规审查.

Ache@zhmza

Install

openclaw skills install @zhmza/edict

🏛️ 三省六部制 · Edict Multi-Agent Orchestration

OpenClaw 多智能体编排与治理系统

"仿唐制三省六部，构建AI治理体系 —— 决策、审核、执行三位一体，人事、财务、礼仪、安全、合规、工程六维协同"

版本: 1.0.0 | 更新: 2026-03-30 | 作者: cft0808

🎯 核心能力

能力	说明	适用场景
🤖 多智能体编排	9个专业AI智能体协同工作	复杂任务分解与执行
📊 实时监控	Web仪表板实时查看所有智能体状态	运维监控、故障排查
🎛️ 模型调度	支持多模型配置和智能路由	成本优化、性能调优
📋 审计跟踪	完整操作日志和合规报告	安全审计、合规检查
🔄 工作流引擎	可视化工作流设计和执行	业务流程自动化

🏛️ 系统架构

text

                    ┌─────────────────┐
                    │   皇帝 (User)   │
                    └────────┬────────┘
                             │
              ┌──────────────┼──────────────┐
              │              │              │
       ┌──────▼──────┐ ┌─────▼─────┐ ┌─────▼─────┐
       │   中书省     │ │   门下省   │ │   尚书省   │
       │  决策/草拟   │ │  审核/驳回  │ │  执行/落实  │
       └──────┬──────┘ └─────┬─────┘ └─────┬─────┘
              │              │              │
              └──────────────┼──────────────┘
                             │
        ┌────────────────────┼────────────────────┐
        │                    │                    │
   ┌────┴────┐      ┌────────┴────────┐     ┌────┴────┐
   │  吏部   │      │      户部        │     │  礼部   │
   │ 人事管理 │      │    资源财务      │     │ 规范礼仪 │
   └────┬────┘      └────────┬────────┘     └────┬────┘
        │                    │                    │
   ┌────┴────┐      ┌────────┴────────┐     ┌────┴────┐
   │  兵部   │      │      刑部        │     │  工部   │
   │ 安全攻防 │      │    合规审计      │     │ 工程实施 │
   └─────────┘      └─────────────────┘     └─────────┘

⚡ Quick Reference

三省（决策层）

场景	调用	核心动作	输出
需要制定方案	`edict.zhongshu`	草拟提案、制定策略	方案文档
需要审核评估	`edict.menxia`	审核方案、风险评估	审核报告
需要执行落地	`edict.shangshu`	任务分解、进度监控	执行计划

六部（执行层）

场景	调用	核心动作	输出
智能体配置	`edict.libu`	创建、配置、权限管理	智能体实例
资源分配	`edict.hubu`	预算、资源、成本管理	资源分配方案
规范制定	`edict.libu_rites`	流程、标准、规范	规范文档
安全策略	`edict.bingbu`	安全策略、攻防演练	安全报告
合规审计	`edict.xingbu`	合规检查、审计日志	审计报告
工程实施	`edict.gongbu`	技术实现、系统构建	部署方案

🚀 快速开始

1. 基础用法

python

from edict import EdictSystem

# 初始化系统
edict = EdictSystem()

# 启动完整治理体系
edict.launch_governance(
    dashboard=True,    # 启用实时仪表板
    audit=True,        # 启用审计日志
    auto_scale=True    # 自动扩缩容
)

2. 完整工作流示例

python

# 1️⃣ 中书省：草拟方案
proposal = edict.zhongshu.draft_proposal(
    task="构建智能客服系统",
    requirements=["7x24小时", "多语言", "情感分析"],
    constraints={"budget": "50万", "timeline": "3个月"}
)

# 2️⃣ 门下省：审核方案
review = edict.menxia.review_proposal(
    proposal=proposal,
    criteria=["可行性", "成本效益", "风险评估", "合规性"]
)

if review.approved:
    # 3️⃣ 尚书省：分解任务
    tasks = edict.shangshu.decompose_task(proposal)
    
    # 4️⃣ 六部协同执行
    ## 吏部：配置智能体
    agent = edict.libu.configure_agent(
        name="智能客服",
        role="customer_service",
        model="gpt-4"
    )
    
    ## 户部：分配资源
    budget = edict.hubu.allocate_budget(
        project="智能客服",
        amount=500000,
        categories={"compute": 0.4, "storage": 0.3, "api": 0.3}
    )
    
    ## 礼部：制定规范
    standards = edict.libu_rites.establish_standards(
        domain="客服话术",
        rules=["礼貌用语", "响应时效", "问题解决率"]
    )
    
    ## 兵部：安全策略
    security = edict.bingbu.define_security_policy(
        level="high",
        measures=["数据加密", "访问控制", "审计日志"]
    )
    
    ## 刑部：合规检查
    compliance = edict.xingbu.check_compliance(
        system="客服系统",
        regulations=["网络安全法", "个人信息保护法", "GDPR"]
    )
    
    ## 工部：技术实施
    implementation = edict.gongbu.implement_solution(
        architecture="微服务",
        stack=["Python", "Kubernetes", "Redis", "PostgreSQL"],
        scaling="auto"
    )
    
    # 5️⃣ 启动监控
    edict.dashboard.launch(port=8080)
    
    # 6️⃣ 记录审计
    edict.audit.log(
        action="系统部署完成",
        project="智能客服",
        details={"agent_id": agent.id, "budget": budget.amount}
    )

📚 详细文档

一、三省智能体（决策层）

1. 中书省 - 决策智能体

python

from edict import ZhongshuProvince

zhongshu = ZhongshuProvince()

# 草拟方案
proposal = zhongshu.draft_proposal(
    task="构建电商推荐系统",
    requirements=["实时性", "个性化", "可扩展"],
    constraints={
        "budget": "50万",
        "timeline": "3个月",
        "team_size": 5
    }
)

# 制定策略
strategy = zhongshu.formulate_strategy(
    goal="提升用户转化率30%",
    metrics=["CTR", "CVR", "GMV", "ROI"],
    approach="A/B测试 + 协同过滤 + 深度学习"
)

# 生成路线图
roadmap = zhongshu.create_roadmap(
    phases=[
        {"name": "MVP", "duration": "1个月", "deliverables": ["基础推荐"]},
        {"name": "优化", "duration": "1个月", "deliverables": ["个性化"]},
        {"name": "扩展", "duration": "1个月", "deliverables": ["实时更新"]}
    ]
)

2. 门下省 - 审核智能体

python

from edict import MenxiaProvince

menxia = MenxiaProvince()

# 审核方案
review_result = menxia.review_proposal(
    proposal=proposal,
    criteria=[
        "技术可行性",
        "成本效益分析",
        "风险评估",
        "合规性检查",
        "资源需求"
    ]
)

# 风险评估
risk_report = menxia.assess_risk(
    project="新系统上线",
    factors=[
        {"name": "技术风险", "level": "medium", "mitigation": "技术预研"},
        {"name": "业务风险", "level": "low", "mitigation": "灰度发布"},
        {"name": "合规风险", "level": "low", "mitigation": "法务审核"}
    ]
)

# 生成审核意见
opinion = menxia.generate_opinion(
    proposal=proposal,
    decision="approved_with_conditions",
    conditions=["增加监控告警", "制定回滚方案"]
)

3. 尚书省 - 执行智能体

python

from edict import ShangshuProvince

shangshu = ShangshuProvince()

# 分解任务
tasks = shangshu.decompose_task(
    project="电商推荐系统",
    milestones=[
        {"name": "需求分析", "duration": "1周", "owner": "PM"},
        {"name": "系统设计", "duration": "1周", "owner": "架构师"},
        {"name": "开发实现", "duration": "4周", "owner": "开发团队"},
        {"name": "测试上线", "duration": "2周", "owner": "测试团队"}
    ]
)

# 执行监控
monitor = shangshu.monitor_execution(
    tasks=tasks,
    metrics=["进度", "质量", "成本", "风险"],
    alerts=[
        {"condition": "进度延迟>1天", "action": "notify"},
        {"condition": "成本超支>10%", "action": "escalate"}
    ]
)

# 资源协调
resources = shangshu.coordinate_resources(
    teams=["前端", "后端", "算法", "测试"],
    timeline="3个月"
)

二、六部智能体（执行层）

4. 吏部 - 人事管理

python

from edict import Libu

libu = Libu()

# 创建智能体
agent = libu.create_agent(
    name="客服助手",
    role="customer_service",
    description="处理客户咨询和投诉",
    capabilities=[
        "自然语言理解",
        "情感分析",
        "工单处理",
        "知识库检索"
    ],
    model="gpt-4",
    config={
        "temperature": 0.7,
        "max_tokens": 2000,
        "response_time": "<2s"
    }
)

# 配置权限
permissions = libu.set_permissions(
    agent_id=agent.id,
    access={
        "read": ["kb", "customer_data", "tickets"],
        "write": ["tickets", "notes"],
        "execute": ["send_email", "create_task"]
    },
    restrictions={
        "delete": ["customer_data"],
        "modify": ["billing_info"]
    }
)

# 智能体生命周期管理
libu.lifecycle_manage(
    agent_id=agent.id,
    actions=["deploy", "scale", "update", "rollback", "retire"]
)

5. 户部 - 资源财务

python

from edict import Hubu

hubu = Hubu()

# 预算分配
budget = hubu.allocate_budget(
    project="AI客服系统",
    total_amount=500000,
    currency="CNY",
    categories={
        "compute": {"amount": 200000, "percentage": 0.4},
        "storage": {"amount": 150000, "percentage": 0.3},
        "api_calls": {"amount": 100000, "percentage": 0.2},
        "misc": {"amount": 50000, "percentage": 0.1}
    },
    period="annual"
)

# 资源调度
resources = hubu.schedule_resources(
    demands=[
        {"type": "GPU", "spec": "A100", "quantity": 4, "duration": "3个月"},
        {"type": "CPU", "spec": "32核", "quantity": 8, "duration": "长期"},
        {"type": "storage", "spec": "SSD", "size": "10TB", "duration": "长期"}
    ],
    priority="high",
    strategy="cost_optimized"
)

# 成本监控
cost_monitor = hubu.monitor_costs(
    projects=["AI客服", "推荐系统"],
    alerts=[
        {"threshold": "80%", "action": "notify"},
        {"threshold": "100%", "action": "block"}
    ]
)

6. 礼部 - 规范礼仪

python

from edict import LibuRites

libu_rites = LibuRites()

# 制定规范
standards = libu_rites.establish_standards(
    domain="代码审查",
    category="development",
    rules=[
        {"id": "R001", "name": "命名规范", "severity": "error", "check": "naming_convention"},
        {"id": "R002", "name": "注释要求", "severity": "warning", "check": "docstring_coverage"},
        {"id": "R003", "name": "测试覆盖", "severity": "error", "threshold": "80%"},
        {"id": "R004", "name": "复杂度限制", "severity": "warning", "threshold": "10"}
    ]
)

# 设计工作流程
workflow = libu_rites.design_workflow(
    name="需求评审流程",
    steps=[
        {"id": 1, "name": "提交需求", "owner": "PM", "duration": "1天"},
        {"id": 2, "name": "技术初审", "owner": "Tech Lead", "duration": "2天"},
        {"id": 3, "name": "架构复审", "owner": "Architect", "duration": "2天"},
        {"id": 4, "name": "最终批准", "owner": "CTO", "duration": "1天"}
    ],
    transitions=[
        {"from": 1, "to": 2, "condition": "文档完整"},
        {"from": 2, "to": 3, "condition": "技术可行"},
        {"from": 3, "to": 4, "condition": "架构合理"}
    ]
)

# 合规检查
compliance_check = libu_rites.check_compliance(
    artifact="codebase",
    standards=standards,
    report_format="detailed"
)

7. 兵部 - 安全攻防

python

from edict import Bingbu

bingbu = Bingbu()

# 定义安全策略
security_policy = bingbu.define_security_policy(
    level="high",
    domains=["application", "data", "network", "infrastructure"],
    measures=[
        {"domain": "application", "measures": ["输入验证", "SQL注入防护", "XSS防护"]},
        {"domain": "data", "measures": ["加密存储", "传输加密", "访问控制"]},
        {"domain": "network", "measures": ["防火墙", "DDoS防护", "入侵检测"]},
        {"domain": "infrastructure", "measures": ["容器安全", "镜像扫描", "运行时保护"]}
    ]
)

# 安全扫描
scan_result = bingbu.security_scan(
    target="production",
    scan_types=["vulnerability", "misconfiguration", "secrets"],
    severity_levels=["critical", "high", "medium"]
)

# 攻防演练
exercise = bingbu.conduct_exercise(
    type="red_team",
    scope=["API接口", "数据库", "文件系统", "认证系统"],
    duration="2周",
    report=True
)

8. 刑部 - 合规审计

python

from edict import Xingbu

xingbu = Xingbu()

# 合规检查
compliance = xingbu.check_compliance(
    system="用户数据处理系统",
    regulations=[
        "网络安全法",
        "个人信息保护法",
        "数据安全法",
        "GDPR",
        "CCPA"
    ],
    checks=[
        "数据收集合法性",
        "用户同意管理",
        "数据最小化原则",
        "跨境数据传输",
        "数据保留期限"
    ]
)

# 审计日志配置
audit_config = xingbu.configure_audit(
    level="detailed",
    scope=["all"],
    storage={
        "type": "database",
        "encryption": True,
        "backup": True,
        "retention": "7年"
    }
)

# 记录审计日志
audit_log = xingbu.log(
    action="用户数据访问",
    actor={"type": "agent", "id": "agent_001", "name": "客服助手"},
    target={"type": "data", "id": "user_12345", "category": "personal_info"},
    operation="read",
    result="success",
    context={"ip": "10.0.0.1", "timestamp": "2026-03-30T10:00:00Z"}
)

# 生成审计报告
report = xingbu.generate_report(
    type="compliance",
    period="monthly",
    format="pdf",
    recipients=["compliance@company.com", "cto@company.com"]
)

9. 工部 - 工程实施

python

from edict import Gongbu

gongbu = Gongbu()

# 技术方案设计
design = gongbu.design_solution(
    requirements=["高可用", "可扩展", "低延迟"],
    architecture={
        "pattern": "microservices",
        "components": [
            {"name": "API Gateway", "tech": "Kong/Nginx"},
            {"name": "Service Mesh", "tech": "Istio"},
            {"name": "Cache", "tech": "Redis Cluster"},
            {"name": "Database", "tech": "PostgreSQL + ClickHouse"},
            {"name": "Message Queue", "tech": "Kafka"}
        ]
    }
)

# 系统构建
build = gongbu.build_system(
    components=design.components,
    environment="kubernetes",
    ci_cd={
        "pipeline": "gitlab-ci",
        "stages": ["build", "test", "security_scan", "deploy"],
        "auto_deploy": True
    }
)

# 部署实施
deployment = gongbu.deploy(
    environment="production",
    strategy="blue_green",
    rollback_plan=True,
    monitoring=True
)

# 性能优化
optimization = gongbu.optimize_performance(
    metrics=["latency", "throughput", "error_rate"],
    targets={"latency": "<100ms", "throughput": ">10000rps", "error_rate": "<0.1%"}
)

三、实时仪表板

python

from edict import Dashboard

# 创建仪表板
dashboard = Dashboard()

# 配置面板
dashboard.configure_panels([
    {
        "name": "智能体状态",
        "type": "status_grid",
        "metrics": ["health", "load", "requests"],
        "refresh": 5
    },
    {
        "name": "任务队列",
        "type": "queue_monitor",
        "metrics": ["pending", "running", "completed", "failed"],
        "refresh": 10
    },
    {
        "name": "资源使用",
        "type": "resource_chart",
        "metrics": ["cpu", "memory", "gpu", "storage"],
        "refresh": 30
    },
    {
        "name": "审计日志",
        "type": "audit_stream",
        "filter": ["security", "compliance"],
        "refresh": 60
    },
    {
        "name": "成本分析",
        "type": "cost_breakdown",
        "group_by": ["project", "resource_type"],
        "refresh": 3600
    }
])

# 启动仪表板
dashboard.launch(
    host="0.0.0.0",
    port=8080,
    auth={"type": "oauth", "providers": ["github", "google"]}
)

# 设置告警
dashboard.set_alerts([
    {
        "name": "CPU高负载",
        "condition": "cpu_usage > 80%",
        "duration": "5m",
        "severity": "warning",
        "notify": ["ops@company.com"]
    },
    {
        "name": "智能体故障",
        "condition": "agent_health == 'down'",
        "severity": "critical",
        "notify": ["ops@company.com", "cto@company.com"],
        "auto_restart": True
    },
    {
        "name": "成本超支",
        "condition": "daily_cost > budget * 1.2",
        "severity": "warning",
        "notify": ["finance@company.com"]
    }
])

四、模型配置管理

python

from edict import ModelConfig

# 创建模型配置
model_config = ModelConfig()

# 添加模型
model_config.add_model(
    name="gpt-4-turbo",
    provider="openai",
    config={
        "model": "gpt-4-turbo-preview",
        "temperature": 0.7,
        "max_tokens": 4000,
        "top_p": 1.0
    },
    cost={"input": 0.01, "output": 0.03}  # per 1K tokens
)

model_config.add_model(
    name="claude-3-opus",
    provider="anthropic",
    config={
        "model": "claude-3-opus-20240229",
        "temperature": 0.5,
        "max_tokens": 4000
    },
    cost={"input": 0.015, "output": 0.075}
)

model_config.add_model(
    name="local-llm",
    provider="local",
    config={
        "endpoint": "http://localhost:8000/v1",
        "model": "llama-2-70b",
        "temperature": 0.8
    },
    cost={"input": 0, "output": 0}  # 本地部署无API成本
)

# 智能路由策略
model_config.set_routing(
    strategy="smart",
    rules={
        "complex_reasoning": {"model": "gpt-4-turbo", "priority": 1},
        "creative_writing": {"model": "claude-3-opus", "priority": 1},
        "simple_qa": {"model": "local-llm", "priority": 1},
        "code_generation": {"model": "gpt-4-turbo", "priority": 1},
        "default": {"model": "local-llm", "priority": 2}
    },
    fallback="local-llm"
)

# 成本优化
model_config.optimize_costs(
    budget_daily=100,  # USD
    strategy="performance_first",
    caching=True,
    batching=True
)

📊 系统指标

指标类别	指标名称	目标值	监控频率
可用性	智能体可用率	>99.9%	实时
性能	平均响应时间	<500ms	每分钟
性能	任务完成率	>95%	每小时
安全	审计覆盖率	100%	实时
成本	资源利用率	60-80%	每5分钟
成本	预算执行率	90-100%	每日
质量	方案通过率	>80%	每周
质量	合规通过率	100%	每月

🛠️ 安装部署

环境要求

组件	最低配置	推荐配置
CPU	8核	16核+
内存	16GB	32GB+
存储	100GB SSD	500GB SSD+
网络	100Mbps	1Gbps+
GPU	可选	NVIDIA A100 (推荐)

安装步骤

bash

# 1. 安装依赖
pip install edict-openclaw

# 2. 或从源码安装
git clone https://github.com/cft0808/edict.git
cd edict
pip install -e .

# 3. 初始化配置
edict init --config ./config.yaml

# 4. 启动服务
edict start --dashboard --audit --port 8080

Docker部署

bash

# 使用Docker Compose
docker-compose up -d

# 或Kubernetes
kubectl apply -f k8s/

🔗 集成示例

与现有系统集成

python

# 集成到现有OpenClaw工作流
from edict import EdictSystem
from openclaw import Session

# 创建会话
session = Session()

# 初始化Edict
edict = EdictSystem()

# 在现有任务中使用
@session.task
def build_feature():
    # 中书省设计方案
    proposal = edict.zhongshu.draft_proposal(task="新功能开发")
    
    # 门下省审核
    if edict.menxia.review_proposal(proposal).approved:
        # 尚书省执行
        edict.shangshu.execute(proposal)

🏛️ 构建AI治理体系，实现智能体协同！

Skill Version: 1.0.0
Compatible with: OpenClaw 2026.3.24+
License: MIT

三省六部

Install

🏛️ 三省六部制 · Edict Multi-Agent Orchestration

🎯 核心能力

🏛️ 系统架构

⚡ Quick Reference

三省（决策层）

六部（执行层）

🚀 快速开始

1. 基础用法

2. 完整工作流示例

📚 详细文档

一、三省智能体（决策层）

1. 中书省 - 决策智能体

2. 门下省 - 审核智能体

3. 尚书省 - 执行智能体

二、六部智能体（执行层）

4. 吏部 - 人事管理

5. 户部 - 资源财务

6. 礼部 - 规范礼仪

7. 兵部 - 安全攻防

8. 刑部 - 合规审计

9. 工部 - 工程实施

三、实时仪表板

四、模型配置管理

📊 系统指标

🛠️ 安装部署

环境要求

安装步骤

Docker部署

🔗 集成示例

与现有系统集成

Related skills