Create agents by talking to Claude Chat with Cerebro connected:
"Add a marketing agent that writes social media posts in a casual tone"
"Create a QA agent that writes tests for React components"
"Add a DevOps agent for Docker and CI/CD"
Cerebro saves the agent definition with:
- A persona built from your description
- Auto-extracted skills for A2A Agent Card
- Default A2A configuration for interop
"Show me my agents" → list_agents
"Update the Coder to use Python" → update_agent
"Remove the QA agent" → remove_agent
"What is the Coder doing?" → get_agent_status
Agents delegate via A2A protocol. Set up chain triggers:
"When the Coder finishes, automatically send to QA for testing"
The delegation creates a threaded conversation between agents, with task lifecycle states (submitted → working → completed/failed).
Pre-built templates from Anthropic's open-source repo:
"Install the Coder agent template"
User-contributed agents shared on GitHub.
Your custom agents, private by default.
Drop .cerebro/agents.json in a project for auto-registration:
web-app: Coder + Designer + QA api-service: Coder + QA + DevOps content-site: Designer + Coder + Marketing
"Install the web-app starter kit"
Every Cerebro agent publishes an Agent Card. External A2A agents can join:
"Connect the Salesforce agent at salesforce.example.com"
Cerebro discovers the Agent Card, connects, and the external agent becomes part of your swarm.
Agents learn preferences over time:
- "This user prefers React over Vue"
- "Use TypeScript, not JavaScript"
- "Tailwind CSS for styling"
Preferences persist across sessions. Working context is wiped per task (zero rot).