Sub-Agent Spawning: Task Delegation Patterns
Spawn temporary sub-agents from your main OpenClaw agent to handle specialized tasks, then collect their results seamlessly.
What You Will Get
By the end of this guide, your main OpenClaw agent will be able to spawn sub-agents on the fly for specific tasks and collect their results before replying to the user. This keeps your primary agent responsive while offloading intensive work to temporary workers.
Sub-agent spawning is ideal for tasks like running parallel research queries, generating multiple content drafts, or performing lengthy computations. The parent agent decides when to spawn, what instructions to give, and how to merge the results into its final response.
You will configure spawning rules, set resource limits on sub-agents, and learn patterns for sequential and parallel delegation. The result is a flexible system where your agent scales its capabilities on demand without bloating its own context window.
Step-by-Step Setup
Follow these steps to enable and configure sub-agent spawning.
Enable the Sub-Agent Tool
Go to your agent's tool configuration in the RunTheAgent dashboard. Enable the sub-agent spawning tool from the available tools list. This gives your agent the ability to create temporary agents during a conversation.
Define a Spawning Template
Create a template that specifies the default system prompt, model, and tool set for spawned sub-agents. Templates save time and ensure consistency. You can override any template value at spawn time if a particular task requires different settings.
Configure Resource Limits
Set a maximum token budget and timeout for each sub-agent. This prevents runaway costs if a sub-agent enters a loop or receives an unexpectedly large input. A typical limit is 4,000 tokens and a 30-second timeout, but adjust based on your task complexity.
Write the Delegation Prompt
In your parent agent's system prompt, describe when and how to delegate. For example, instruct the agent to spawn a sub-agent when the user asks for a comparison of more than three items. Include instructions on what context to pass and what format to expect back from the sub-agent.
Implement Parallel Spawning
For tasks that can run simultaneously, configure the parent agent to spawn multiple sub-agents in parallel. Each sub-agent receives its own slice of the task, works independently, and returns its result. The parent waits for all results before composing the final answer.
Handle Sub-Agent Failures
Add fallback logic to the parent agent's prompt for cases when a sub-agent times out or returns an error. The parent can retry with different parameters, use a cached result, or inform the user that part of the task could not be completed.
Review Spawning Logs
Open the logs panel on your RunTheAgent dashboard to see every sub-agent spawn event. Each entry shows the parent agent, the sub-agent's task, token usage, and duration. Use this data to optimize your templates and resource limits over time.
Tips and Best Practices
Keep Sub-Agent Prompts Short
Sub-agents work best with focused, concise instructions. Pass only the context they need. Long prompts waste tokens and slow down the spawning cycle.
Prefer Parallel Over Sequential
When tasks are independent, spawn sub-agents in parallel to reduce total response time. Sequential spawning should only be used when one sub-agent's output feeds into the next.
Reuse Templates Across Agents
If multiple parent agents delegate similar tasks, share spawning templates to maintain consistency. Update the template once and all agents benefit from the improvement.
Set Strict Token Budgets
Sub-agents that exceed their token budget can inflate costs quickly. Set conservative limits and increase them only after reviewing actual usage patterns in the logs.
Frequently Asked Questions
Related Pages
Ready to get started?
Deploy your own OpenClaw instance in under 60 seconds. No VPS, no Docker, no SSH. Just your personal AI assistant, ready to work.
Starting at $24.50/mo. Everything included. 3-day money-back guarantee.
