RuFlo 学习教程
RuFlo 学习教程
从零开始掌握 RuFlo - 企业级 AI Agent 编排平台
教程日期:2026-03-24
目录
一、环境准备
1.1 前置知识
学习 RuFlo 前建议了解:
| 知识领域 | 重要程度 | 说明 |
|---|---|---|
| Claude Code 基础 | ⭐⭐⭐⭐⭐ | 理解 Claude Code CLI 的基本用法 |
| 命令行基础 | ⭐⭐⭐⭐ | 能够运行终端命令 |
| Agent 概念 | ⭐⭐⭐⭐ | 理解 AI Agent 的基本概念 |
| MCP 协议 | ⭐⭐⭐ | 了解 Model Context Protocol |
| TypeScript | ⭐⭐⭐ | 用于自定义开发 |
1.2 系统要求
支持平台:
├── macOS (x64/arm64) ✅
├── Linux (x64/arm64) ✅
└── Windows ❌ (建议使用 WSL2)
必需软件:
├── Node.js 18+ 或 Bun
├── Git
└── Claude Code CLI (最新版)
1.3 安装 Claude Code
# macOS/Linux
curl -fsSL https://claude.ai/install.sh | bash
# 或通过 npm
npm install -g @anthropic-ai/claude-code
1.4 安装 RuFlo
# 方式 1: 一键安装 (推荐)
curl -fsSL https://cdn.jsdelivr.net/gh/ruvnet/claude-flow@main/scripts/install.sh | bash
# 方式 2: 通过 npx (无需安装)
npx ruflo@latest init --wizard
# 方式 3: 克隆仓库
git clone https://github.com/ruvnet/ruflo.git
cd ruflo && npm install
1.5 配置 API Keys
# 必需: Anthropic API Key
export ANTHROPIC_API_KEY="sk-ant-..."
# 可选: 其他 LLM 提供商 (用于故障转移)
export OPENAI_API_KEY="sk-..."
export GOOGLE_API_KEY="..."
export COHERE_API_KEY="..."
# 可选: 本地模型
# 安装 Ollama: https://ollama.ai
ollama pull llama3
1.6 验证安装
# 检查版本
npx ruflo@latest --version
# 运行诊断
npx claude-flow@v3alpha doctor
# 查看 MCP 工具
npx ruflo@latest mcp list
二、快速开始
2.1 第一个任务
# 使用单个 Agent 执行任务
npx ruflo@latest --agent coder --task "创建一个简单的 Express.js 服务器"
# 输出示例:
# ✓ 任务已提交
# ✓ Agent: coder 已启动
# ✓ 正在分析任务...
# ✓ 生成代码...
# ✓ 创建文件: server.js
# ✓ 任务完成
2.2 初始化 Swarm
# 初始化一个 Swarm
npx claude-flow@v3alpha swarm init --v3-mode
# 配置选项:
# - 拓扑类型: hierarchical, mesh, ring, star
# - 最大 Agent 数: 1-12
# - 策略: specialized, balanced, adaptive, auto
2.3 多 Agent 协作
# 使用 SPARC 方法论开发功能
npx claude-flow sparc run orchestrator "实现用户认证系统,包括注册、登录、JWT 令牌管理"
# 这将:
# 1. 分析任务需求
# 2. 分配给合适的 Agent
# 3. 协调执行顺序
# 4. 整合结果
2.4 添加 MCP 到 Claude Code
# 添加 RuFlo MCP 服务器到 Claude Code
claude mcp add ruflo -- npx -y ruflo@latest mcp start
# 验证 MCP 连接
claude mcp list
# 现在可以在 Claude Code 中使用所有 259 个工具
三、核心概念
3.1 Agent
Agent 是 RuFlo 的基本执行单元,每个 Agent 都有特定的专业领域和技能。
┌─────────────────────────────────────┐
│ Agent │
├─────────────────────────────────────┤
│ • Type: coder, tester, reviewer... │
│ • Capabilities: 特定技能列表 │
│ • Memory: 独立记忆上下文 │
│ • State: active, idle, completed │
└─────────────────────────────────────┘
核心 Agent 类型:
| 类型 | 职责 | 典型任务 |
|---|---|---|
coder |
代码开发 | 功能实现、重构 |
tester |
测试 | 单元测试、集成测试 |
reviewer |
代码审查 | 代码质量、最佳实践 |
architect |
架构设计 | 系统设计、技术方案 |
researcher |
研究 | 技术调研、问题分析 |
optimizer |
优化 | 性能调优、资源优化 |
documenter |
文档 | 文档编写、注释 |
debugger |
调试 | Bug 修复、问题排查 |
3.2 Swarm
Swarm 是多个 Agent 的协作集合,通过特定拓扑结构协调工作。
┌─────────────────────────────────────┐
│ Swarm │
├─────────────────────────────────────┤
│ • Topology: 拓扑结构 │
│ • Strategy: 协调策略 │
│ • Agents: Agent 集合 │
│ • Consensus: 共识机制 │
└─────────────────────────────────────┘
四种拓扑类型:
1. Hierarchical (层级)
Queen → Workers
2. Mesh (网格)
Agent ↔ Agent (全互联)
3. Ring (环形)
Agent → Agent → Agent → ...
4. Star (星形)
Hub ↔ Agents
3.3 SPARC
SPARC 是系统化开发方法论:
| 阶段 | 英文 | 含义 |
|---|---|---|
| S | Specification | 规格定义 |
| P | Pseudocode | 伪代码设计 |
| A | Architecture | 架构设计 |
| R | Refinement | 精炼实现 |
| C | Completion | 完成交付 |
3.4 Memory
Memory 是 RuFlo 的知识存储系统:
Working Memory (工作记忆)
↓
Episodic Memory (情景记忆)
↓
Semantic Memory (语义记忆)
3.5 MCP (Model Context Protocol)
MCP 是 Claude Code 的工具协议,RuFlo 提供 259 个 MCP 工具:
Claude Code
│
├── MCP Server (RuFlo)
│ │
│ ├── Agent Tools (7)
│ ├── Swarm Tools (4)
│ ├── Memory Tools (7)
│ ├── Config Tools (6)
│ ├── Hooks Tools (40+)
│ └── ... (更多)
四、Swarm 编排
4.1 创建 Swarm
# CLI 方式
npx claude-flow@v3alpha swarm init \
--topology hierarchical \
--max-agents 8 \
--strategy specialized
# MCP 方式
mcp__ruv-swarm__swarm_init({
topology: "hierarchical",
maxAgents: 8,
strategy: "specialized"
})
4.2 选择拓扑
| 拓扑 | 适用场景 | 优缺点 |
|---|---|---|
| hierarchical | 复杂项目、需要协调 | ✅ 清晰协调 ❌ 单点依赖 |
| mesh | 简单任务、并行执行 | ✅ 高可用 ❌ 协调复杂 |
| ring | 流水线任务 | ✅ 顺序清晰 ❌ 延迟累积 |
| star | 集中控制 | ✅ 统一管理 ❌ Hub 瓶颈 |
4.3 配置策略
// specialized: 每个 Agent 专注特定领域
strategy: "specialized" // 推荐用于复杂项目
// balanced: Agent 负载均衡
strategy: "balanced" // 推荐用于批量任务
// adaptive: 动态调整策略
strategy: "adaptive" // 推荐用于不确定场景
// auto: 自动选择
strategy: "auto" // 让系统决定
4.4 Swarm 操作
# 启动 Swarm
npx claude-flow@v3alpha swarm start
# 查看状态
npx claude-flow@v3alpha swarm status
# 列出 Agent
npx claude-flow@v3alpha agent list
# 发送任务
npx claude-flow@v3alpha swarm task "任务描述"
# 停止 Swarm
npx claude-flow@v3alpha swarm stop
4.5 Queen 模式
Strategic Queen (战略型):
# 用于研究和架构决策
npx claude-flow@v3alpha queen spawn --type strategic
Tactical Queen (战术型):
# 用于实现和部署管理
npx claude-flow@v3alpha queen spawn --type tactical
Adaptive Queen (自适应型):
# 用于优化和动态调整
npx claude-flow@v3alpha queen spawn --type adaptive
4.6 共识机制
# 配置共识算法
export CLAUDE_FLOW_CONSENSUS=majority # 简单多数
export CLAUDE_FLOW_CONSENSUS=weighted # 加权投票
export CLAUDE_FLOW_CONSENSUS=byzantine # 拜占庭容错
五、SPARC 方法论
5.1 SPARC 工作流
┌─────────────────────────────────────────────────────────────┐
│ SPARC Workflow │
├─────────────────────────────────────────────────────────────┤
│ │
│ 1. Specification (规格) │
│ ├── 定义需求和约束 │
│ ├── 建立成功标准 │
│ └── Agent: researcher, analyzer │
│ │ │
│ ▼ │
│ 2. Architecture (架构) │
│ ├── 设计系统结构 │
│ ├── 定义接口和数据流 │
│ └── Agent: architect, designer │
│ │ │
│ ▼ │
│ 3. Refinement (精炼) │
│ ├── 测试驱动开发 │
│ ├── 迭代实现 │
│ └── Agent: coder, tdd, tester │
│ │ │
│ ▼ │
│ 4. Review (审查) │
│ ├── 代码质量审查 │
│ ├── 安全和性能检查 │
│ └── Agent: reviewer, optimizer │
│ │ │
│ ▼ │
│ 5. Completion (完成) │
│ ├── 集成和部署 │
│ ├── 文档和发布 │
│ └── Agent: workflow-manager, documenter │
│ │
└─────────────────────────────────────────────────────────────┘
5.2 执行 SPARC
# 执行完整 SPARC 流程
npx claude-flow sparc pipeline "实现 REST API 认证系统"
# 执行特定阶段
npx claude-flow sparc run specification "分析用户认证需求"
npx claude-flow sparc run architecture "设计认证系统架构"
npx claude-flow sparc run refinement "实现认证系统"
npx claude-flow sparc run review "审查认证系统代码"
npx claude-flow sparc run completion "部署和文档"
5.3 TDD 模式
# 测试驱动开发模式
npx claude-flow sparc tdd "实现购物车功能"
# 流程:
# 1. 编写测试用例
# 2. 运行测试 (失败)
# 3. 实现功能代码
# 4. 运行测试 (通过)
# 5. 重构优化
# 6. 重复
5.4 批量执行
# 并行执行多个模式
npx claude-flow sparc batch "coder,tester,reviewer" "实现用户注册功能"
# Agent 并行工作:
# - coder: 实现代码
# - tester: 编写测试
# - reviewer: 代码审查
5.5 监控进度
# 实时监控
npx claude-flow@v3alpha status --watch
# 查看进度
npx claude-flow@v3alpha progress
# 查看指标
npx claude-flow@v3alpha metrics
六、记忆系统
6.1 记忆层次
┌─────────────────────────────────────────────────────────────┐
│ Memory Hierarchy │
├─────────────────────────────────────────────────────────────┤
│ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ Working Memory │ │
│ │ • 当前活跃上下文 │ │
│ │ • 快速访问 (亚毫秒级) │ │
│ │ • 1MB 限制 │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ Episodic Memory │ │
│ │ • 近期任务历史 │ │
│ │ • 中等保留时间 │ │
│ │ • 任务轨迹记录 │ │
│ └─────────────────────────────────────────────────────┘ │
│ │ │
│ ▼ │
│ ┌─────────────────────────────────────────────────────┐ │
│ │ Semantic Memory │ │
│ │ • 整合的知识库 │ │
│ │ • 持久化存储 │ │
│ │ • 模式和最佳实践 │ │
│ └─────────────────────────────────────────────────────┘ │
│ │
└─────────────────────────────────────────────────────────────┘
6.2 使用记忆
# 存储模式
mcp__claude-flow__memory_usage({
action: "store",
namespace: "patterns",
key: "auth-jwt-pattern",
value: JSON.stringify({
approach: "JWT",
expiry: "24h",
refresh: true
})
})
# 检索模式
mcp__claude-flow__memory_usage({
action: "search",
namespace: "patterns",
query: "authentication patterns"
})
# 相似度评分:
# > 0.7: 强匹配,直接使用
# 0.5-0.7: 部分匹配,需要适配
# < 0.5: 弱匹配,创建新模式
6.3 记忆作用域
// 项目作用域
scope: "project" // 当前项目
// 本地作用域
scope: "local" // 本地环境
// 用户作用域
scope: "user" // 用户全局
6.4 HNSW 向量搜索
# 启用 HNSW 索引
export CLAUDE_FLOW_HNSW_INDEXING=true
# 配置嵌入维度
export CLAUDE_FLOW_EMBEDDING_DIMENSIONS=384
# 记忆搜索会使用 HNSW 进行快速向量检索
# 性能: 150x-12,500x 更快
6.5 自学习循环
RETRIEVE → JUDGE → DISTILL → CONSOLIDATE → ROUTE
│ │ │ │ │
▼ ▼ ▼ ▼ ▼
检索历史 评估质量 提取模式 整合知识 智能路由
启用自学习:
# 启用自学习
export CLAUDE_FLOW_SELF_LEARNING=true
# 配置学习频率
export CLAUDE_FLOW_LEARNING_INTERVAL=100 # 每 100 个任务学习一次
七、MCP 工具
7.1 核心 MCP 工具
Agent 工具
// 生成 Agent
mcp__ruv-swarm__agent_spawn({
type: "coder",
name: "my-coder",
capabilities: ["typescript", "express"]
})
// 列出 Agent
mcp__ruv-swarm__agent_list()
// 查看状态
mcp__ruv-swarm__agent_status({ id: "agent-id" })
// 停止 Agent
mcp__ruv-swarm__agent_stop({ id: "agent-id" })
Swarm 工具
// 初始化 Swarm
mcp__ruv-swarm__swarm_init({
topology: "hierarchical",
maxAgents: 8,
strategy: "specialized"
})
// 发送任务
mcp__ruv-swarm__swarm_task({
description: "实现用户认证系统"
})
// 查看状态
mcp__ruv-swarm__swarm_status()
Memory 工具
// 存储记忆
mcp__claude-flow__memory_usage({
action: "store",
namespace: "knowledge",
key: "api-design",
value: "{...}"
})
// 搜索记忆
mcp__claude-flow__memory_usage({
action: "search",
namespace: "knowledge",
query: "API 设计模式"
})
// 清除记忆
mcp__claude-flow__memory_usage({
action: "clear",
namespace: "knowledge"
})
7.2 SPARC MCP 工具
// 执行 SPARC 模式
mcp__claude-flow__sparc_mode({
mode: "orchestrator",
task_description: "协调功能开发",
options: {
parallel: true,
monitor: true
}
})
// TDD 模式
mcp__claude-flow__sparc_mode({
mode: "tdd",
task_description: "实现购物车功能"
})
7.3 安全工具
// 安全扫描
mcp__claude-flow__security_scan({
depth: "full"
})
// 输入验证
mcp__claude-flow__validate_input({
input: "user input",
rules: ["no-injection", "no-xss"]
})
7.4 配置工具组
# 开发模式
export CLAUDE_FLOW_TOOL_MODE=develop
# 包含: create, implement, test, fix
# 生产模式
export CLAUDE_FLOW_TOOL_MODE=production
# 包含: monitor, security, optimize
# 最小模式
export CLAUDE_FLOW_TOOL_MODE=minimal
# 仅包含核心工具
# 自定义组
export CLAUDE_FLOW_TOOL_GROUPS=implement,test,fix,memory
八、自定义 Agent
8.1 创建 Agent 定义文件
在项目目录创建 .claude/agents/ 文件夹:
mkdir -p .claude/agents
8.2 定义 Agent
创建 my-custom-agent.yaml:
name: my-custom-agent
type: specialist
version: "1.0.0"
description: |
专门处理数据转换任务的 Agent
capabilities:
- data-transform
- format-conversion
- validation
priority: high
optimizations:
- context-caching
- memory-persistence
config:
maxRetries: 3
timeout: 30000
prompt: |
你是一个数据转换专家。
你的任务是:
1. 分析输入数据格式
2. 确定转换策略
3. 执行转换
4. 验证结果
examples:
- input: "CSV to JSON conversion"
output: "Convert CSV rows to JSON objects"
8.3 注册 Agent
# 将 Agent 定义文件放入 .claude/agents/ 目录
cp my-custom-agent.yaml .claude/agents/
# 重新加载配置
npx claude-flow@v3alpha config reload
# 验证 Agent 可用
npx claude-flow@v3alpha agent list
8.4 使用自定义 Agent
# 通过 CLI
npx ruflo@latest --agent my-custom-agent --task "转换 CSV 文件"
# 通过 MCP
mcp__ruv-swarm__agent_spawn({
type: "my-custom-agent",
name: "converter"
})
8.5 Agent 模板
通用模板:
name: template-agent
type: specialist
version: "1.0.0"
description: |
Agent 描述 - 用于触发匹配
capabilities:
- capability-1
- capability-2
priority: medium
optimizations:
- flash-attention
- token-reduction
config:
maxRetries: 3
timeout: 60000
memoryLimit: 1048576 # 1MB
prompt: |
系统提示词...
任务说明...
输出格式...
九、企业部署
9.1 Docker 部署
# Dockerfile
FROM node:20-alpine
WORKDIR /app
# 安装 RuFlo
RUN npm install -g ruflo@latest
# 配置环境变量
ENV ANTHROPIC_API_KEY=""
ENV CLAUDE_FLOW_LOG_LEVEL=info
# 初始化
RUN ruflo init
CMD ["ruflo", "mcp", "start"]
# 构建镜像
docker build -t ruflo:latest .
# 运行容器
docker run -d \
--name ruflo-server \
-e ANTHROPIC_API_KEY=sk-ant-... \
-p 3000:3000 \
ruflo:latest
9.2 Cloud Run 部署
# 一键部署到 Cloud Run
npx claude-flow deploy --platform cloud-run
# 配置
gcloud run deploy ruflo \
--image gcr.io/project/ruflo \
--set-env-vars ANTHROPIC_API_KEY=sk-ant-...
9.3 Kubernetes 部署
# deployment.yaml
apiVersion: apps/v1
kind: Deployment
metadata:
name: ruflo
spec:
replicas: 3
selector:
matchLabels:
app: ruflo
template:
metadata:
labels:
app: ruflo
spec:
containers:
- name: ruflo
image: ruflo:latest
env:
- name: ANTHROPIC_API_KEY
valueFrom:
secretKeyRef:
name: api-keys
key: anthropic
ports:
- containerPort: 3000
9.4 监控和日志
# 配置日志级别
export CLAUDE_FLOW_LOG_LEVEL=debug
# 启用监控
export CLAUDE_FLOW_MONITORING=true
# 查看日志
npx claude-flow@v3alpha logs --follow
# 导出指标
npx claude-flow@v3alpha metrics export
9.5 安全配置
# 启用 AIDefence
export CLAUDE_FLOW_SECURITY_ENABLED=true
# 配置威胁检测级别
export CLAUDE_FLOW_SECURITY_LEVEL=high
# 启用输入验证
export CLAUDE_FLOW_INPUT_VALIDATION=true
# 启用 PII 检测
export CLAUDE_FLOW_PII_DETECTION=true
9.6 高可用配置
# 自动扩缩
export CLAUDE_FLOW_AUTO_SCALE=true
export CLAUDE_FLOW_MAX_WORKERS=12
export CLAUDE_FLOW_SCALE_UP_THRESHOLD=2
export CLAUDE_FLOW_SCALE_DOWN_THRESHOLD=2
# 故障转移
export CLAUDE_FLOW_FAILOVER=true
export CLAUDE_FLOW_FAILOVER_PROVIDERS=openai,google
十、最佳实践
10.1 任务分解
好的做法:
# 分解大任务为小任务
npx claude-flow sparc run coder "实现用户注册 API"
npx claude-flow sparc run coder "实现用户登录 API"
npx claude-flow sparc run tester "编写认证测试用例"
不好的做法:
# 任务太大太模糊
npx claude-flow sparc run coder "实现整个用户系统"
10.2 选择合适的拓扑
| 项目复杂度 | 推荐拓扑 | 原因 |
|---|---|---|
| 简单 (1-2 天) | mesh | 快速执行 |
| 中等 (1-2 周) | hierarchical | 清晰协调 |
| 复杂 (1+ 月) | hierarchical + Strategic Queen | 专业管理 |
10.3 利用记忆系统
# 开始任务前先检索相关模式
mcp__claude-flow__memory_usage({
action: "search",
query: "类似任务的解决方案"
})
# 任务完成后存储成功模式
mcp__claude-flow__memory_usage({
action: "store",
key: "successful-approach",
value: "{...}"
})
10.4 成本优化
# 使用三层模型路由
export CLAUDE_FLOW_MODEL_ROUTING=true
# 简单任务用 Haiku
# 复杂任务用 Sonnet/Opus
# 简单转换用 WASM (<1ms, $0)
10.5 安全最佳实践
# 1. 始终启用安全扫描
export CLAUDE_FLOW_SECURITY_ENABLED=true
# 2. 使用环境变量存储 API Keys
# 不要在代码中硬编码
# 3. 定期审计
npx claude-flow@v3alpha security audit --depth full
# 4. 最小权限原则
# 只启用需要的工具组
export CLAUDE_FLOW_TOOL_GROUPS=implement,test
10.6 性能优化
# 1. 启用 HNSW 索引
export CLAUDE_FLOW_HNSW_INDEXING=true
# 2. 使用 Flash Attention
# 已默认启用
# 3. 配置合适的 Agent 数量
# 推荐: CPU 核心数 - 1
# 4. 定期清理记忆
mcp__claude-flow__memory_usage({
action: "cleanup",
olderThan: "30d"
})
十一、常见问题
11.1 安装问题
Q: 安装失败,提示权限错误
# 使用 sudo 或修改 npm 全局目录
sudo npm install -g ruflo@latest
# 或修改 npm 全局目录
mkdir ~/.npm-global
npm config set prefix '~/.npm-global'
export PATH=~/.npm-global/bin:$PATH
Q: MCP 连接失败
# 检查 MCP 服务状态
npx ruflo@latest mcp status
# 重新添加 MCP
claude mcp remove ruflo
claude mcp add ruflo -- npx -y ruflo@latest mcp start
11.2 Agent 问题
Q: Agent 无法启动
# 检查 API Key
echo $ANTHROPIC_API_KEY
# 运行诊断
npx claude-flow@v3alpha doctor --fix
# 查看日志
npx claude-flow@v3alpha logs --level debug
Q: Agent 执行超时
# 在 Agent 配置中增加超时时间
config:
timeout: 120000 # 2 分钟
11.3 记忆问题
Q: 记忆搜索无结果
# 检查 HNSW 索引
export CLAUDE_FLOW_HNSW_INDEXING=true
# 重新构建索引
npx claude-flow@v3alpha memory rebuild-index
Q: 记忆占用过多空间
# 清理旧记忆
mcp__claude-flow__memory_usage({
action: "cleanup",
olderThan: "7d"
})
# 压缩记忆
npx claude-flow@v3alpha memory compress
11.4 性能问题
Q: 响应速度慢
# 检查 Agent 数量
npx claude-flow@v3alpha agent list
# 减少 Agent 数量
export CLAUDE_FLOW_MAX_AGENTS=4
# 启用模型路由
export CLAUDE_FLOW_MODEL_ROUTING=true
Q: Token 消耗过高
# 启用 Token 减少
# 已默认启用 32.3% 减少
# 使用更小的模型
export CLAUDE_FLOW_DEFAULT_MODEL=haiku
11.5 安全问题
Q: 检测到 Prompt 注入
# 系统会自动阻止并记录
# 查看安全日志
npx claude-flow@v3alpha security logs
# 调整安全级别
export CLAUDE_FLOW_SECURITY_LEVEL=maximum
十二、参考资源
官方资源
- GitHub: https://github.com/ruvnet/ruflo
- NPM: https://www.npmjs.com/package/ruflo
- 官网: https://Cognitum.One
- Agentics Foundation: https://agentics.org
相关项目
- Claude Code: https://claude.ai/code
- Model Context Protocol: https://modelcontextprotocol.io
- Anthropic API: https://docs.anthropic.com
社区资源
- GitHub Discussions: https://github.com/ruvnet/ruflo/discussions
- Issues: https://github.com/ruvnet/ruflo/issues
十三、版本历史
| 版本 | 发布日期 | 主要更新 |
|---|---|---|
| v3.5.31 | 2026-03-18 | 性能优化、安全增强 |
| v3.5.0 | 2026-03-01 | 新增 17 SPARC 模式 |
| v3.4.0 | 2026-02-15 | SONA 神经架构 |
| v3.3.0 | 2026-02-01 | HNSW 向量搜索 |
| v3.0.0 | 2026-01-15 | 企业版发布 |
| v2.0.0 | 2025-11-01 | Swarm 编排 |
| v1.0.0 | 2025-06-02 | 首次发布 |
教程生成时间:2026-03-24