Delegate a task to a sub-agent
Delegate an independent piece of work when you need its conclusion, not every step of its investigation. For work where later steps need earlier results, use a workflow instead. Delegation is provided by the optional subagent plugin; it must be enabled for the tool to be available.
Write a complete task
- Ask Elowen to delegate one well-defined task and state the result you need.
- Include relevant context, constraints, and the expected output. A fresh sub-agent cannot see this conversation; the task you give it is its complete instruction.
- If tasks are independent, ask Elowen to start them together so they can run in parallel. Do not delegate the same investigation and also do it yourself.
- Wait for the result. Delegation runs in the background by default, and Elowen reports the result when it finishes. Ask for a blocking call only when you cannot proceed without the answer.
A sub-agent uses your access and tools by default, so it cannot gain permissions you do not have. You can narrow its tools or request read-only access, but read-only mode is not a sandbox. A child normally starts a fresh conversation; use a fork only when it needs the parent conversation's full context. See Sub-agents & Workflows for the full task and workflow guide.
Follow up or stop
If the result needs correction or more work, ask Elowen to continue the same sub-agent rather than starting over. Its conversation retains the task and prior findings. You can ask Elowen to stop a sub-agent that is still running; completed work remains in its conversation. In the CLI, Ctrl+O opens a child and Esc returns to the parent. In web chat, use the sub-agent's row in the activity rail to inspect or stop it.
Check access and availability
Do not delegate work that depends on subtle interpretation of your request, a fact you already know where to find, or a task too small to justify a separate agent. Review the result before relying on changes or conclusions.