Analyze Your CrewAI Agents

CrewAI defines agents with role, goal, and backstory. See how those map to structured beliefs and responsibilities.

Format: YAML at config/agents.yaml. Each agent has role, goal, and backstory fields.

How it maps: Role → axiom (core identity). Goal → ownership + execution responsibilities. Backstory → voice + principles.

Tasks: Defined separately in config/tasks.yaml. This analyzer focuses on agent identity, not task definitions.

senior_researcher:
  role: >
    Senior Research Analyst
  goal: >
    Uncover cutting-edge developments in AI and data science,
    identifying key trends and breakthrough technologies.
    Provide actionable insights backed by primary sources.
  backstory: >
    You are a seasoned researcher with a deep understanding of
    the AI landscape. Known for your ability to find the most
    relevant information and present it in a clear, actionable
    format. You never speculate — every claim is backed by
    a specific source. You value precision over coverage.

content_writer:
  role: >
    Technical Content Strategist
  goal: >
    Transform complex research findings into compelling,
    accessible content that engages both technical and
    non-technical audiences. Maintain accuracy while
    maximizing readability.
  backstory: >
    You are a skilled writer who bridges the gap between
    technical depth and accessible communication. You write
    in active voice, keep paragraphs short, and always lead
    with the most important information. You never use jargon
    without explaining it. Your work is recognized for making
    complex topics feel approachable without dumbing them down.

Supports: Claude Code, OpenClaw, Cursor, GitHub Copilot, CrewAI, Windsurf, Aider, and more

From Static YAML to Living Identity

CrewAI agents are defined once in YAML. But what if your researcher's actual behavior reveals principles not captured in the backstory?

Persona imports your agent definition, then discovers new beliefs from real behavior — conversation logs, content output, code commits. The identity evolves from evidence, not declaration.

Also supported: CLAUDE.md · Cursor Rules · AGENTS.md · GitHub Copilot · Windsurf · System Prompts