agent-create

Agent Create Skill

Professional AI Agent development assistant, helping users build various AI Agent frameworks from scratch. Supports mainstream frameworks such as LangGraph, AgentScope, DeepAgent, AutoGen, CutoGen Studio, and CrewAI, providing a complete development workflow: requirements analysis, documentation exploration, structure planning, code development, and test verification.

Use Cases

New Project Kickoff: Need to build a new Agent project, but are unsure which framework to choose
Framework Learning: Want to learn a certain Agent framework, but do not know where to start
Rapid Prototyping: Need to quickly build a runnable Agent prototype
Production-Grade Development: Need to build a complete, production-ready Agent system

Prerequisites

Claude Code CLI environment
Network connection (for searching official documentation)
Runtime environment for the corresponding framework (such as Python 3.10+)
Necessary MCP servers are configured (for documentation search and web access)

Workflow

This Skill uses a six-phase professional development workflow, and each phase has clear deliverables and user confirmation steps:

Phase 1: Requirements Gathering and Analysis

Objective: Fully understand user requirements and determine the best framework choice

Steps：

Ask about use cases and business goals
Understand technology stack preferences and experience level
Confirm performance requirements and scalability needs
Discuss budget and time constraints

Deliverable: Requirements gathering checklist

Example Questions：

"What is the main purpose of this Agent? Conversation, task automation, data processing?"
"Which Agent frameworks have you used before?"
"What is the expected user scale and request volume for the project?"
"Is multi-Agent collaboration required?"

Phase 2: PRD Document Generation

Objective: Based on requirements analysis, generate a structured product requirements document

Content：

Project overview and objectives
Functional requirements specifications
Non-functional requirements (performance, security, scalability)
Technology selection recommendations
Risk assessment and response strategies
Milestone planning

Deliverable：PRD.md - Product requirements document

Document Structure：

# Agent 项目 PRD

## 1. 项目概述
## 2. 用户故事
## 3. 功能规格
## 4. 技术架构
## 5. 实施计划
## 6. 验收标准

Phase 3: Documentation Exploration (Subagent)

Objective: Use MCP tools to search official documentation and fully understand the framework's implementation details and best practices

Trigger Subagent：bmad:research or custom documentation search agent

Search Content：

Official documentation and quick start guides
API reference and core concepts
Example projects and best practices
Latest version features and changelogs
Common issues and solutions

Deliverable：FRAMEWORK_DOCS.md - Framework documentation summary

Processing Method：

The Subagent executes independently and does not occupy the main agent's context
Returns a structured framework knowledge base
Provide code examples and configuration templates

Phase 4: Structure Planning (Subagent)

Objective: Based on requirements and document analysis, plan the complete project structure

Trigger Subagent: Custom architecture planning agent

Planning Content：

Project directory structure design
Module division and responsibility definition
Dependency configuration and version management
Configuration file specifications
Entry points and lifecycle management
Testing strategy and directory structure

Deliverables：

ARCHITECTURE.md - Architecture design document
PROJECT_STRUCTURE.md - Project structure description
Code framework template

Example directory structure：

agent-project/
├── src/
│   ├── agents/          # Agent 定义
│   ├── tools/           # 工具实现
│   ├── workflows/       # 工作流编排
│   ├── core/            # 核心逻辑
│   └── utils/           # 工具函数
├── tests/
│   ├── unit/            # 单元测试
│   ├── integration/     # 集成测试
│   └── e2e/             # 端到端测试
├── config/              # 配置文件
├── docs/                # 文档
└── scripts/             # 脚本

Phase 5: Code Development

Objective: Based on the planned structure, generate the complete code implementation

Development Content：

Agent core implementation
Tool and function definitions
Workflow orchestration
Configuration management
Error handling and logging
Docstrings and comments

Deliverables：

Complete source code files
Configuration files (pyproject.toml, requirements.txt, etc.)
Environment variable example (.env.example)
Makefile or scripts

Code Quality Standards：

Follow PEP 8 / PEP 484
Complete type annotations
Docstring specifications
Appropriate error handling
Logging

Phase 6: Test Verification (Subagent)

Objective: Use an independent Subagent to execute tests and ensure code quality

Trigger Subagent: Custom testing agent

Test Content：

Unit test generation and execution
Integration test verification
Code quality checks (lint, type check)
Documentation testing (if applicable)

Deliverables：

TEST_REPORT.md - Test report
Pass/fail status
Fix suggestions

Input Parameters

Required Parameters

Parameter Name	Type	Description
`framework`	string	Target framework (langgraph, agentscope, deepagent, autogen, crewai)
`project_name`	string	Project name

Optional Parameters

Parameter Name	Type	Default Value	Description
`language`	string	"python"	Development language
`features`	array	[]	List of required feature capabilities
`has_ui`	boolean	false	Whether a Web UI is needed
`deployment`	string	"local"	Deployment environment (local/docker/cloud)

Usage Example

/agent-create --framework=langgraph --project_name=my-chat-agent --features=memory,tools --has_ui=true

Output Format

Directory Structure

{project_name}/
├── src/
│   └── {project_name}/
│       ├── __init__.py
│       ├── agent.py
│       ├── config.py
│       ├── tools/
│       └── workflows/
├── tests/
├── docs/
│   ├── PRD.md
│   ├── ARCHITECTURE.md
│   └── FRAMEWORK_DOCS.md
├── config/
├── scripts/
├── .env.example
├── pyproject.toml
├── requirements.txt
├── README.md
└── Makefile

File Description

File	Description
`src/agent.py`	Agent core definition
`src/config.py`	Configuration management
`src/tools/*.py`	Custom tool implementation
`src/workflows/*.py`	Workflow definition
`tests/test_*.py`	Test cases
`docs/PRD.md`	Product requirements document
`docs/ARCHITECTURE.md`	Architecture design document
`docs/FRAMEWORK_DOCS.md`	Framework documentation summary
`config/.py` / `.yaml`	Configuration files
`pyproject.toml`	Project metadata and dependencies
`Makefile`	Development task scripts

Usage examples

Example 1: Creating a simple conversational Agent

用户：我想创建一个简单的对话 Agent，使用 LangGraph 框架

Agent Create：
[阶段 1] 您好！我来帮您创建一个 LangGraph 对话 Agent。
为了更好地满足您的需求，请告诉我：

1. 这个 Agent 主要用于什么场景？（客服、个人助手、知识库问答等）
2. 是否需要集成外部工具？（搜索、数据库、API 等）
3. 是否需要长期记忆？
4. 预计的并发量是多少？

用户：用于客服场景，需要集成搜索工具，不需要长期记忆，预计并发 100

[生成 PRD，进行后续阶段...]

Example 2: Creating a multi-Agent collaboration system

用户：用 AgentScope 创建一个多 Agent 协作系统，用于代码审查

Agent Create：
[开始需求收集...]
[生成 PRD...]
[Subagent 搜索 AgentScope 官方文档...]
[Subagent 规划项目结构...]
[生成完整代码...]
[Subagent 执行测试...]

Supported frameworks

Framework	Features	Use cases
LangGraph	Loop workflows, strong state management	Complex conversations, task orchestration
AgentScope	Decentralized architecture, good ease of use	Rapid prototyping, multi-Agent collaboration
DeepAgent	Deep learning integration, research-oriented	Research experiments, advanced features
AutoGen	Multi-Agent conversations, strong automation	Enterprise applications, workflow automation
CrewAI	Role-playing, concise orchestration	Team simulation, role collaboration

Notes

Framework selection: If the user is unsure about the framework, provide a comparative analysis of each framework
Version compatibility: Ensure dependency version compatibility and provide version constraints
Environment configuration: Provide a complete runtime environment configuration guide
Documentation updates: Information obtained during the documentation exploration phase needs to be verified for timeliness
Error handling: If problems are encountered during development, provide alternative solutions

Error handling

Error type	Handling method
Documentation search failure	Try backup documentation sources and provide manual reference links
Code generation error	Provide error logs and try a simplified implementation
Test failure	Analyze the cause of failure and generate fix suggestions
Dependency conflict	Provide version adjustment suggestions or virtual environment configuration

Extended features

This Skill can be extended with the following features as needed:

Template marketplace: Provide predefined Agent templates
Plugin system: Support custom tools and middleware
Deployment integration: One-click deployment to mainstream cloud platforms
Monitoring integration: Integrate observability tools
Version upgrades: Framework version upgrade assistant

Best practices

Progressive development: Start with a minimum viable product and gradually add features
Documentation-driven: Documentation first, then implementation
Test-first: Write tests for critical logic first
Code review: Important changes require code review
Continuous integration: Configure CI/CD pipelines

About this skill

Use Cases

Prerequisites

Workflow

Phase 1: Requirements Gathering and Analysis

Phase 2: PRD Document Generation

Phase 3: Documentation Exploration (Subagent)

Phase 4: Structure Planning (Subagent)

Phase 5: Code Development

Phase 6: Test Verification (Subagent)

Input Parameters

Required Parameters

Optional Parameters

Usage Example

Output Format

Directory Structure

File Description

Usage examples

Example 1: Creating a simple conversational Agent

Example 2: Creating a multi-Agent collaboration system

Supported frameworks

Notes

Error handling

Extended features

Best practices