Getting Started

This walks you through the core architecture decisions in CrewAI: when to use LLM.call() versus Agent.kickoff() versus a full Crew versus a Flow. The main thing it hammers home is never writing boilerplate by hand. Always scaffold with `crewai create flow` first, then modify the generated YAML and Python files. It covers the decision flowchart (do you need tools? multiple agents? state management?), shows how to wire agents.yaml and tasks.yaml with variable interpolation, and explains the @CrewBase decorator pattern. The structured output examples are helpful, especially the difference between LLM.call() returning Pydantic objects directly versus Agent.kickoff() wrapping them in result.pydantic. Good reference if you're setting up a project and want to avoid subtle import or config mistakes.

Install to Claude Code

npx -y skills add crewaiinc/skills --skill getting-started --agent claude-code