Crewai

CrewAI 多智能体框架助手，精通 Agent 编排、任务分配、工具集成、工作流设计

Security Scan

The SKILL.md matches the skill name and described purpose (a CrewAI multi‑agent framework helper); it is instruction-only and does not request credentials or perform hidden installs, though it references tools and package installs that could access the web, local files, or require API keys if actually executed.

Purpose & Capability

The name/description claim a multi-agent orchestration helper and the SKILL.md contains documentation, examples, and usage patterns for such a framework — consistent with the stated purpose.

Instruction Scope

SKILL.md contains role/instruction text for the agent and many code examples that reference web scraping, search tools, file reads, PDF/RAG, and code execution. Those examples are within scope for a multi-agent framework, but they imply actions (network access, file I/O, code execution, use of LLM backends) that the skill does not itself enumerate or constrain.

Install Mechanism

There is no install spec and no code files; the README suggests pip installing 'crewai' and 'crewai-tools' but the skill package does not perform any automatic downloads — this is low risk from the skill bundle itself. If you follow the pip command externally, validate those packages.

Credentials

The SKILL.md references LLM backends (OpenAI/Anthropic/Ollama) and tool integrations that typically require API keys, but the skill does not request or store any environment credentials itself. This is proportionate for documentation, but you should only supply service keys when you trust the actual implementation (crewai/crewai-tools).

Persistence & Privilege

always is false and the skill is user-invocable; the SKILL.md discusses optional 'memory' features but the skill bundle does not declare or modify system-wide configs. No elevated or persistent privileges are requested by the package itself.

Assessment

This skill is an instruction-only documentation helper for designing multi-agent workflows and appears internally consistent. Before you install or run anything referenced here: (1) Check PyPI (or the project source) for 'crewai' and 'crewai-tools' to confirm authorship and inspect their code; (2) Do not provide API keys (OpenAI/Anthropic/etc.) unless you trust those packages and understand where keys are stored/used; (3) Be aware examples enable web scraping, file reads, and code execution — run in a sandbox or with restricted permissions if you want to limit data exposure; (4) If you want the agent to act autonomously, consider restricting or auditing tool access and memory persistence so it cannot read or send files/credentials you care about.

CrewAI 多智能体框架助手

你是 CrewAI 多智能体编排领域的专家，帮助用户设计和实现高效的 AI 协作工作流。

核心概念

概念	说明
Agent	智能体，具有角色(role)、目标(goal)、背景故事(backstory)的自主实体
Task	任务，分配给特定 Agent 执行，包含描述和期望输出
Crew	团队，编排多个 Agent 和 Task 的协作单元
Tool	工具，Agent 可调用的外部能力（搜索、文件读写、API 调用等）
Process	流程模式，sequential（顺序）或 hierarchical（层级管理）

安装

pip install crewai crewai-tools
crewai create crew my_project  # CLI 创建项目脚手架

Agent 与 Task 定义

from crewai import Agent, Task, Crew, Process

researcher = Agent(
    role="高级研究分析师",
    goal="发现 {topic} 领域的最新突破性进展",
    backstory="你是一位资深的技术研究员，擅长从海量信息中提炼关键洞察。",
    tools=[search_tool, web_scraper],
    llm="gpt-4o",           # 支持 OpenAI/Anthropic/Ollama
    memory=True,             # 启用记忆，跨任务保持上下文
    allow_delegation=False   # 是否允许委派任务给其他 Agent
)

writer = Agent(
    role="技术内容撰稿人",
    goal="将研究成果转化为通俗易懂的技术文章",
    backstory="你是一位经验丰富的技术写作者。",
    llm="gpt-4o"
)

research_task = Task(
    description="深入研究 {topic} 的最新发展，收集至少 5 个可靠来源。",
    expected_output="一份结构化的研究报告，包含关键发现和来源引用。",
    agent=researcher,
    output_file="research_report.md"
)

writing_task = Task(
    description="基于研究报告，撰写一篇 1500 字的技术博客文章。",
    expected_output="一篇完整的技术博客文章。",
    agent=writer,
    context=[research_task]  # 依赖前置任务的输出
)

Crew 编排

crew = Crew(
    agents=[researcher, writer],
    tasks=[research_task, writing_task],
    process=Process.sequential,  # 按顺序执行
    memory=True,                 # 团队级记忆
    planning=True                # 启用任务规划
)
result = crew.kickoff(inputs={"topic": "AI Agent 框架"})

# 层级管理模式（自动创建 Manager Agent）
crew_h = Crew(
    agents=[researcher, writer, editor],
    tasks=[research_task, writing_task, review_task],
    process=Process.hierarchical,
    manager_llm="gpt-4o"
)

工具系统

from crewai_tools import (
    SerperDevTool,          # Google 搜索
    ScrapeWebsiteTool,      # 网页抓取
    FileReadTool,           # 文件读取
    PDFSearchTool,          # PDF 搜索（RAG）
    CodeInterpreterTool     # 代码执行
)

# 自定义工具
from crewai.tools import tool

@tool("股票价格查询")
def stock_price(ticker: str) -> str:
    """查询指定股票的实时价格。参数 ticker 为股票代码。"""
    return f"{ticker} 当前价格: $150.00"

高级特性

特性	说明
Memory	短期/长期/实体记忆，跨任务和跨运行保持上下文
Planning	任务执行前自动生成执行计划
Callbacks	任务/步骤级回调，用于监控和日志
Async	支持异步执行，crew.kickoff_async()
Training	crew.train(n_iterations=3) 通过人类反馈优化
YAML 配置	支持 agents.yaml 和 tasks.yaml 声明式配置

与同类框架对比

特性	CrewAI	LangGraph	AutoGen
核心理念	角色扮演协作	图状态机	对话驱动
学习曲线	低，API 直观	中等	中等
编排方式	Sequential/Hierarchical	自定义有向图	对话/群聊
工具生态	crewai-tools 丰富	LangChain 生态	函数注册
适用场景	内容生产、研究分析	复杂工作流	代码生成、多轮讨论