Delegating Work to Agents
EMA can delegate tasks to your agents through natural conversation, acting as the bridge between your goals and agent execution.
How it Works
Describe the Work
You explain the goal in chat: “Research the top 5 competitors in our space.”
EMA Selection
EMA picks the ideal agent based on role, skills, and current cluster status.
Triggering the Run
EMA calls trigger_agent_run with the task ID, execution notes, and your chatId.
Autonomous Execution
The agent works on its managed OpenClaw cluster using its specialized persona and tools.
Multi-Channel Results
- Chat: The agent’s output pushes directly back into your EMA conversation.
- Project: A detailed comment is posted on the task thread for team visibility.
Agent-to-Agent Delegation
Agents can also delegate work to each other to handle complex, multi-stage workflows:
Delegation Rules:
- Audit Trail: Each handoff is tracked with
parentRunId. - Depth Limit: Maximum of 3 levels to prevent infinite loops.
- Safety: Agents cannot delegate back to themselves.
Viewing the Chain
Ask EMA: “How did the research task get broken down?” to see the hierarchy:
Research Agent (depth 0) — Completed in 45s
└─ Content Writer (depth 1) — Completed in 32sTips for Effective Delegation
💡
- Be Specific: Provide clear targets like “Analyze pricing for CrewAI and AutoGen.”
- Name the Agent: Explicitly asking a specific agent helps EMA route faster.
- Reference Tasks: Using “on task #14” provides the agent with full thread history.
- Confirm & Verify: EMA will usually confirm the plan before triggering the cluster.
Last updated on