A comprehensive collection of specialized AI agents for Claude Code
专为Claude Code设计的全面AI代理集合
This repository contains a powerful collection of specialized AI agents designed to enhance your Claude Code development experience. Each agent is crafted with deep domain expertise and strict boundary enforcement to handle specific technology stacks and development tasks efficiently.
- Enhanced Task Assessment: 6-level (0-5) complexity evaluation with ultra-intelligent analysis (ultrathink mode)
- 28+ Specialized Agents: Comprehensive coverage of modern development technologies with strict role boundaries
- Project-Specific Agents:
/initxcommand creates custom agents tailored to your unique project requirements - Agent Boundary System: ALLOWED/FORBIDDEN pattern preventing scope creep and ensuring precision
- Dependency-Aware Execution: Smart dependency detection with phase-based execution strategies
- Multi-Technology Support: Vue.js, React, Go, Rust, Python, Android, Java, Security, and more
- Production Ready: Battle-tested configurations optimized for real-world development
- Performance Optimized: 3x faster simple tasks, 2x more reliable complex coordination
claude-ai-agents/
├── agents/ # AI Agent Definitions
│ ├── vue-developer.md # Vue.js/Nuxt.js specialist
│ ├── react-developer.md # React/Next.js expert
│ ├── go-architect.md # Go microservices architect
│ ├── rust-architect.md # Rust systems programmer
│ ├── flask-expert.md # Python Flask developer
│ ├── fastapi-expert.md # Python FastAPI specialist
│ ├── google-ui-designer.md # Material Design expert
│ ├── android-developer.md # Android app developer
│ ├── mobile-ui-designer.md # Mobile UX designer
│ ├── java-developer.md # Java enterprise developer
│ ├── spring-architect.md # Spring ecosystem architect
│ ├── lua-developer.md # Lua scripting expert
│ ├── android-hooking-expert.md # Android security analyst
│ ├── xposed-developer.md # Xposed module developer
│ ├── reverse-engineer.md # Security researcher
│ ├── malware-analyst.md # Threat intelligence analyst
│ ├── backend-developer.md # API & database expert
│ ├── frontend-developer.md # UI/UX developer
│ ├── devops-engineer.md # DevOps & deployment
│ ├── code-review-expert.md # Code quality guardian
│ ├── test-expert.md # Testing strategist
│ ├── task-dispatch-director.md # Central coordinator
│ └── ... (more agents)
├── commands/
│ ├── ai.md # AI team orchestration
│ └── initx.md # Project-specific agent creation
├── CLAUDE.md # Core configuration
└── README.md # This file
Beyond the global agents, you can create project-specific agents tailored to your exact requirements using the /initx command.
- 10x Faster Development: Custom agents understand your specific business logic and tech stack
- Zero Learning Curve: Agents pre-configured with your project patterns and conventions
- Smart Collaboration: Agents know your project's unique integration points and dependencies
- Domain Expertise: Deep knowledge of your specific requirements and constraints
/initx # Analyze project and create custom agents
/initx --preview # Preview mode (no file creation)
/initx --focus=security # Create security-focused agents
/initx --template=mobile # Use mobile project template
/initx --model=inherit # Force all agents to use specific model (inherit/sonnet/opus/haiku)The system analyzes your project structure and creates specialized agents like:
vue-ecommerce-developer- Project-specific Vue expert for e-commercepayment-integration-specialist- Custom payment API expertpostgres-inventory-architect- Database-specific data architectaws-deployment-specialist- Platform-specific deployment expert
- Default: Intelligent model selection based on agent complexity
- --model=inherit: All agents inherit parent session model (best for cost control)
- --model=sonnet: All agents use Sonnet (balanced performance)
- --model=opus: All agents use Opus (maximum reasoning capability)
- --model=haiku: All agents use Haiku (fastest response time)
your_project/
├── .claude/
│ └── agents/ # Project-specific agents
│ ├── vue-ecommerce-developer.md
│ ├── payment-integration-specialist.md
│ └── postgres-inventory-architect.md
├── CLAUDE.md # Updated with custom team
└── .gitignore # Updated exclusions
Strict Constraints: Analyze and report only. Cannot execute code or implement solutions.
- technical-solution-architect: Designs technical solutions and creates implementation roadmaps
- technical-researcher: Investigates technologies and provides research reports
- code-review-expert: Analyzes code quality and identifies issues (cannot fix problems)
Strict Constraints: Execute code only within specific technology domains.
- vue-developer: Vue.js 3, Composition API, Nuxt.js, Pinia (Vue ecosystem only)
- react-developer: React 18, Next.js, Redux Toolkit, Server Components (React ecosystem only)
- frontend-developer: General frontend development and UI frameworks
- backend-developer: Multi-stack API development (FastAPI/Spring Boot/Node.js) - server-side only
- go-architect: Go microservices, distributed systems, cloud-native
- rust-architect: Systems programming, performance optimization, memory safety
- flask-expert: Python Flask web applications and REST APIs
- fastapi-expert: Async Python APIs, automatic documentation
- java-developer: Enterprise Java development and Spring Boot
- spring-architect: Spring ecosystem, microservices architecture
- android-developer: Native Android apps, Kotlin, Architecture Components
- mobile-ui-designer: Mobile UX/UI, responsive design, platform guidelines
- google-ui-designer: Material Design 3, Accessibility, Design Systems
- devops-engineer: Docker containerization and deployment (infrastructure only)
- infrastructure-developer: Build tools, CI/CD pipelines, automation scripts
- android-hooking-expert: Frida, dynamic analysis, runtime manipulation
- xposed-developer: Xposed modules, system-level modifications
- reverse-engineer: Static analysis, decompilation, vulnerability research
- malware-analyst: Threat detection, behavioral analysis, incident response
- lua-developer: Game scripting, automation, OpenResty web development
Strict Constraints: Provide guidance and coordination but cannot implement solutions.
- qa-engineer: Testing strategy and quality analysis (cannot write production code)
- test-expert: Testing framework design and strategy (cannot implement business logic)
- task-dispatch-director: Pure coordination only (
⚠️ Never calls itself)
-
Clone the repository:
git clone https://github.com/Toskysun/sub-agents.git cd sub-agents -
Copy to Claude Code directory:
# Linux/Mac cp -r * ~/.claude/ # Windows copy * %USERPROFILE%\.claude\
-
Restart Claude Code or reload configuration
-
(Optional) Create Project-Specific Agents:
# Navigate to your project directory cd your_project # Analyze and create custom agents for your project /initx
Global Agent Tasks (Available everywhere):
/ai "Create a Vue.js dashboard component"
/ai "Build a FastAPI authentication service"
/ai "Design a Material Design mobile interface"
/ai "Optimize Go microservice performance"
Project-Specific Agent Tasks (After running /initx):
/ai "optimize checkout flow" # Uses vue-ecommerce-developer
/ai "implement payment gateway" # Uses payment-integration-specialist
/ai "add inventory alerts" # Uses postgres-inventory-architect
/ai "deploy to production" # Uses aws-deployment-specialist
Multi-Agent Coordination:
/ai "Build a full-stack e-commerce platform"
/ai "Create a React component library with testing"
/ai "Develop a secure Android app with API integration"
Team Management:
/ai list # Show all available agents (global + project)
/ai info react # Get React specialist details
/initx --preview # Preview custom agents for your project
| Level | Type | Description | Handling | Performance |
|---|---|---|---|---|
| Level 0 | Micro | Information queries, file reading | Direct execution | Instant response |
| Level 1 | Simple | Single file changes, basic configs | Single agent direct | 3x faster routing |
| Level 2 | Medium | Component development, feature modules | Single agent complex | Full process handling |
| Level 3 | Composite | Multi-module coordination, 2-3 domains | Serial multi-agent | Context handoffs |
| Level 4 | Parallel | Independent modules, performance optimization | Concurrent execution | 4x speedup potential |
| Level 5 | Enterprise | System refactoring, architecture design | Director coordination | Complex orchestration |
- ALLOWED/FORBIDDEN Pattern: Explicit constraints preventing scope creep
- Domain-Specific Execution: Agents work only within their expertise areas
- Delegation Requirements: Mandatory handoffs when tasks exceed boundaries
- Anti-Over-Engineering: Prevents unnecessary multi-agent invocations
- Smart Dependency Analysis: Automatic detection of task prerequisites
- Phase-Based Execution: Serial execution for dependent tasks, parallel for independent
- Context Handoffs: Proper information flow between execution phases
- Selective Parallelism: Maximum speedup while maintaining safety
- Level 0-5 Task Classification: Precision routing from micro-tasks to enterprise coordination
- 3x faster simple tasks: Direct execution bypasses unnecessary overhead
- 2x more reliable complex coordination: Proper dependency management
- Automatic Fallback: 3-retry system with intelligent agent reassignment
- Anti-Deadlock Protection: task-dispatch-director limited to pure coordination
- Quality Gates: Built-in code review and testing workflows
- TodoWrite Strategy: Level-appropriate task tracking and management
- Boundary Enforcement: Strict agent role compliance with violation detection
We welcome contributions! Please see our Contributing Guide for details.
MIT License - see LICENSE for details.
这个仓库包含了一套强大的专业AI代理集合，旨在增强您的Claude Code开发体验。每个代理都具备深度领域专业知识和严格的边界约束，能够高效地处理特定的技术栈和开发任务。
- 增强任务评估: 6级 (0-5) 复杂度评估与超智能分析 (ultrathink模式)
- 28+个专业代理: 全面覆盖现代开发技术栈，配备严格的角色边界
- 项目专属代理:
/initx命令可创建专为您项目需求定制的自定义代理 - 代理边界系统: ALLOWED/FORBIDDEN模式防止职责蔓延，确保精准执行
- 依赖感知执行: 智能依赖检测与基于阶段的执行策略
- 多技术栈支持: Vue.js、React、Go、Rust、Python、Android、Java、安全分析等
- 生产就绪: 经过实战测试的配置，为真实世界开发优化
- 性能优化: 简单任务3倍速度提升，复杂协调2倍可靠性提升
claude-ai-agents/
├── agents/ # AI代理定义
│ ├── vue-developer.md # Vue.js/Nuxt.js专家
│ ├── react-developer.md # React/Next.js专家
│ ├── go-architect.md # Go微服务架构师
│ ├── rust-architect.md # Rust系统程序员
│ ├── flask-expert.md # Python Flask开发者
│ ├── fastapi-expert.md # Python FastAPI专家
│ ├── google-ui-designer.md # Material Design专家
│ ├── android-developer.md # Android应用开发者
│ ├── mobile-ui-designer.md # 移动端UX设计师
│ ├── java-developer.md # Java企业级开发者
│ ├── spring-architect.md # Spring生态系统架构师
│ ├── lua-developer.md # Lua脚本专家
│ ├── android-hooking-expert.md # Android安全分析师
│ ├── xposed-developer.md # Xposed模块开发者
│ ├── reverse-engineer.md # 安全研究员
│ ├── malware-analyst.md # 威胁情报分析师
│ ├── backend-developer.md # API与数据库专家
│ ├── frontend-developer.md # UI/UX开发者
│ ├── devops-engineer.md # DevOps与部署
│ ├── code-review-expert.md # 代码质量守护者
│ ├── test-expert.md # 测试策略师
│ ├── task-dispatch-director.md # **协调器
│ └── ... (更多代理)
├── commands/
│ ├── ai.md # AI团队编排
│ └── initx.md # 项目专属代理创建
├── CLAUDE.md # 核心配置
└── README.md # 本文件
除了全局代理外，您可以使用 /initx 命令创建项目专属代理，完全针对您的具体需求进行定制。
- 10倍开发速度: 定制代理理解您的特定业务逻辑和技术栈
- 零学习成本: 代理预配置了您的项目模式和约定
- 智能协作: 代理了解您项目独特的集成点和依赖关系
- 领域专业: 深入了解您的特定需求和约束
/initx # 分析项目并创建自定义代理
/initx --preview # 预览模式（不创建文件）
/initx --focus=security # 创建安全专注的代理
/initx --template=mobile # 使用移动项目模板
/initx --model=inherit # 强制所有代理使用指定模型 (inherit/sonnet/opus/haiku)系统分析您的项目结构并创建专业代理，例如：
vue-ecommerce-developer- 电商项目专属Vue专家payment-integration-specialist- 定制支付API专家postgres-inventory-architect- 数据库专属数据架构师aws-deployment-specialist- 平台专属部署专家
- 默认: 基于代理复杂度的智能模型选择
- --model=inherit: 所有代理继承父会话模型（最佳成本控制）
- --model=sonnet: 所有代理使用Sonnet（平衡性能）
- --model=opus: 所有代理使用Opus（最大推理能力）
- --model=haiku: 所有代理使用Haiku（最快响应时间）
your_project/
├── .claude/
│ └── agents/ # 项目专属代理
│ ├── vue-ecommerce-developer.md
│ ├── payment-integration-specialist.md
│ └── postgres-inventory-architect.md
├── CLAUDE.md # 更新的自定义团队配置
└── .gitignore # 更新的排除项
严格约束：只能分析和报告，不能执行代码或实现解决方案。
- technical-solution-architect: 设计技术解决方案并创建实施路线图
- technical-researcher: 调研技术并提供研究报告
- code-review-expert: 分析代码质量并识别问题（无法修复问题）
严格约束：只能在特定技术领域内执行代码。
- vue-developer: Vue.js 3、组合式API、Nuxt.js、Pinia（仅Vue生态系统）
- react-developer: React 18、Next.js、Redux Toolkit、Server Components（仅React生态系统）
- frontend-developer: 通用前端开发和UI框架
- backend-developer: 多栈API开发（FastAPI/Spring Boot/Node.js）- 仅服务端
- go-architect: Go微服务、分布式系统、云原生
- rust-architect: 系统编程、性能优化、内存安全
- flask-expert: Python Flask Web应用和REST API
- fastapi-expert: 异步Python API、自动文档生成
- java-developer: 企业级Java开发和Spring Boot
- spring-architect: Spring生态系统、微服务架构
- android-developer: 原生Android应用、Kotlin、架构组件
- mobile-ui-designer: 移动端UX/UI、响应式设计、平台指南
- google-ui-designer: Material Design 3、无障碍性、设计系统
- devops-engineer: Docker容器化和部署（仅基础设施）
- infrastructure-developer: 构建工具、CI/CD管道、自动化脚本
- android-hooking-expert: Frida、动态分析、运行时操控
- xposed-developer: Xposed模块、系统级修改
- reverse-engineer: 静态分析、反编译、漏洞研究
- malware-analyst: 威胁检测、行为分析、事件响应
- lua-developer: 游戏脚本、自动化、OpenResty Web开发
严格约束：提供指导和协调但不能实现解决方案。
- qa-engineer: 测试策略和质量分析（无法编写生产代码）
- test-expert: 测试框架设计和策略（无法实现业务逻辑）
- task-dispatch-director: 纯粹协调（
⚠️ 绝不调用自身）
-
克隆仓库:
git clone https://github.com/Toskysun/sub-agents.git cd sub-agents -
复制到Claude Code目录:
# Linux/Mac cp -r * ~/.claude/ # Windows copy * %USERPROFILE%\.claude\
-
重启Claude Code或重新加载配置
-
（可选）创建项目专属代理:
# 导航到您的项目目录 cd your_project # 分析并为您的项目创建自定义代理 /initx
全局代理任务（随时可用）:
/ai "创建一个Vue.js仪表板组件"
/ai "构建一个FastAPI认证服务"
/ai "设计一个Material Design移动界面"
/ai "优化Go微服务性能"
项目专属代理任务（运行 /initx 后）:
/ai "优化结账流程" # 使用 vue-ecommerce-developer
/ai "实现支付网关" # 使用 payment-integration-specialist
/ai "添加库存警报" # 使用 postgres-inventory-architect
/ai "部署到生产环境" # 使用 aws-deployment-specialist
多代理协调:
/ai "构建一个全栈电商平台"
/ai "创建一个带测试的React组件库"
/ai "开发一个带API集成的安全Android应用"
团队管理:
/ai list # 显示所有可用代理（全局+项目）
/ai info react # 获取React专家详情
/initx --preview # 预览您项目的自定义代理
| 级别 | 类型 | 描述 | 处理方式 | 性能优势 |
|---|---|---|---|---|
| Level 0 | 微任务 | 信息查询、文件读取 | 直接执行 | 即时响应 |
| Level 1 | 简单 | 单文件更改、基础配置 | 单代理直接 | 3倍速度路由 |
| Level 2 | 中等 | 组件开发、功能模块 | 单代理复杂 | 全流程处理 |
| Level 3 | 复合 | 多模块协调、2-3个领域 | 串行多代理 | 上下文传递 |
| Level 4 | 并行 | 独立模块、性能优化 | 并发执行 | 4倍速度潜力 |
| Level 5 | 企业级 | 系统重构、架构设计 | Director协调 | 复杂编排 |
- ALLOWED/FORBIDDEN模式: 明确约束防止职责蔓延
- 域专属执行: 代理只在其专业领域内工作
- 委托要求: 任务超出边界时强制交接
- 反过度工程: 防止不必要的多代理调用
- 智能依赖分析: 自动检测任务先决条件
- 基于阶段的执行: 依赖任务串行执行，独立任务并行执行
- 上下文交接: 执行阶段间的适当信息流
- 选择性并行: 在保证安全的前提下最大化速度提升
- Level 0-5 任务分类: 从微任务到企业级协调的精准路由
- 简单任务3倍速度提升: 直接执行绕过不必要的开销
- 复杂协调2倍可靠性: 适当的依赖管理
- 自动回退: 3次重试系统配合智能代理重新分配
- 反死锁保护: task-dispatch-director限制为纯粹协调
- 质量门禁: 内置代码审查和测试工作流
- TodoWrite策略: 级别适当的任务跟踪和管理
- 边界执行: 严格的代理角色合规性与违规检测
我们欢迎贡献！请查看我们的贡献指南了解详情。
MIT许可证 - 详见LICENSE文件。
Made with ❤️ for the Claude Code community
