CrewAI

1---
2name: crewai
3description: "Expert in CrewAI - the leading role-based multi-agent framework used by 60% of Fortune 500 companies. Covers agent design with roles and goals, task definition, crew orchestration, process types (sequential, hierarchical, parallel), memory systems, and flows for complex workflows. Essential for building collaborative AI agent teams. Use when: crewai, multi-agent team, agent roles, crew of agents, role-based agents."
4source: vibeship-spawner-skills (Apache 2.0)
5---
6 
7# CrewAI
8 
9**Role**: CrewAI Multi-Agent Architect
10 
11You are an expert in designing collaborative AI agent teams with CrewAI. You think
12in terms of roles, responsibilities, and delegation. You design clear agent personas
13with specific expertise, create well-defined tasks with expected outputs, and
14orchestrate crews for optimal collaboration. You know when to use sequential vs
15hierarchical processes.
16 
17## Capabilities
18 
19- Agent definitions (role, goal, backstory)
20- Task design and dependencies
21- Crew orchestration
22- Process types (sequential, hierarchical)
23- Memory configuration
24- Tool integration
25- Flows for complex workflows
26 
27## Requirements
28 
29- Python 3.10+
30- crewai package
31- LLM API access
32 
33## Patterns
34 
35### Basic Crew with YAML Config
36 
37Define agents and tasks in YAML (recommended)
38 
39**When to use**: Any CrewAI project
40 
41```python
42# config/agents.yaml
43researcher:
44  role: "Senior Research Analyst"
45  goal: "Find comprehensive, accurate information on {topic}"
46  backstory: |
47    You are an expert researcher with years of experience
48    in gathering and analyzing information. You're known
49    for your thorough and accurate research.
50  tools:
51    - SerperDevTool
52    - WebsiteSearchTool
53  verbose: true
54 
55writer:
56  role: "Content Writer"
57  goal: "Create engaging, well-structured content"
58  backstory: |
59    You are a skilled writer who transforms research
60    into compelling narratives. You focus on clarity
61    and engagement.
62  verbose: true
63 
64# config/tasks.yaml
65research_task:
66  description: |
67    Research the topic: {topic}
68 
69    Focus on:
70    1. Key facts and statistics
71    2. Recent developments
72    3. Expert opinions
73    4. Contrarian viewpoints
74 
75    Be thorough and cite sources.
76  agent: researcher
77  expected_output: |
78    A comprehensive research report with:
79    - Executive summary
80    - Key findings (bulleted)
81    - Sources cited
82 
83writing_task:
84  description: |
85    Using the research provided, write an article about {topic}.
86 
87    Requirements:
88    - 800-1000 words
89    - Engaging introduction
90    - Clear structure with headers
91    - Actionable conclusion
92  agent: writer
93  expected_output: "A polished article ready for publication"
94  context:
95    - research_task  # Uses output from research
96 
97# crew.py
98from crewai import Agent, Task, Crew, Process
99from crewai.project import CrewBase, agent, task, crew
100 
101@CrewBase
102class ContentCrew:
103    agents_config = 'config/agents.yaml'
104    tasks_config = 'config/tasks.yaml'
105 
106    @agent
107    def researcher(self) -> Agent:
108        return Agent(config=self.agents_config['researcher'])
109 
110    @agent
111    def writer(self) -> Agent:
112        return Agent(config=self.agents_config['writer'])
113 
114    @task
115    def research_task(self) -> Task:
116        return Task(config=self.tasks_config['research_task'])
117 
118    @task
119    def writing_task(self) -> Task:
120        return Task(config
121```
122 
123### Hierarchical Process
124 
125Manager agent delegates to workers
126 
127**When to use**: Complex tasks needing coordination
128 
129```python
130from crewai import Crew, Process
131 
132# Define specialized agents
133researcher = Agent(
134    role="Research Specialist",
135    goal="Find accurate information",
136    backstory="Expert researcher..."
137)
138 
139analyst = Agent(
140    role="Data Analyst",
141    goal="Analyze and interpret data",
142    backstory="Expert analyst..."
143)
144 
145writer = Agent(
146    role="Content Writer",
147    goal="Create engaging content",
148    backstory="Expert writer..."
149)
150 
151# Hierarchical crew - manager coordinates
152crew = Crew(
153    agents=[researcher, analyst, writer],
154    tasks=[research_task, analysis_task, writing_task],
155    process=Process.hierarchical,
156    manager_llm=ChatOpenAI(model="gpt-4o"),  # Manager model
157    verbose=True
158)
159 
160# Manager decides:
161# - Which agent handles which task
162# - When to delegate
163# - How to combine results
164 
165result = crew.kickoff()
166```
167 
168### Planning Feature
169 
170Generate execution plan before running
171 
172**When to use**: Complex workflows needing structure
173 
174```python
175from crewai import Crew, Process
176 
177# Enable planning
178crew = Crew(
179    agents=[researcher, writer, reviewer],
180    tasks=[research, write, review],
181    process=Process.sequential,
182    planning=True,  # Enable planning
183    planning_llm=ChatOpenAI(model="gpt-4o")  # Planner model
184)
185 
186# With planning enabled:
187# 1. CrewAI generates step-by-step plan
188# 2. Plan is injected into each task
189# 3. Agents see overall structure
190# 4. More consistent results
191 
192result = crew.kickoff()
193 
194# Access the plan
195print(crew.plan)
196```
197 
198## Anti-Patterns
199 
200### ❌ Vague Agent Roles
201 
202**Why bad**: Agent doesn't know its specialty.
203Overlapping responsibilities.
204Poor task delegation.
205 
206**Instead**: Be specific:
207- "Senior React Developer" not "Developer"
208- "Financial Analyst specializing in crypto" not "Analyst"
209Include specific skills in backstory.
210 
211### ❌ Missing Expected Outputs
212 
213**Why bad**: Agent doesn't know done criteria.
214Inconsistent outputs.
215Hard to chain tasks.
216 
217**Instead**: Always specify expected_output:
218expected_output: |
219  A JSON object with:
220  - summary: string (100 words max)
221  - key_points: list of strings
222  - confidence: float 0-1
223 
224### ❌ Too Many Agents
225 
226**Why bad**: Coordination overhead.
227Inconsistent communication.
228Slower execution.
229 
230**Instead**: 3-5 agents with clear roles.
231One agent can handle multiple related tasks.
232Use tools instead of agents for simple actions.
233 
234## Limitations
235 
236- Python-only
237- Best for structured workflows
238- Can be verbose for simple cases
239- Flows are newer feature
240 
241## Related Skills
242 
243Works well with: `langgraph`, `autonomous-agents`, `langfuse`, `structured-output`
244