Multi-agent tasks
Use separate agent tasks for bounded work that can proceed independently, then bring their evidence back to the parent outcome.
On this page2
Steps
- Break the outcome into independent deliverables with explicit files, constraints, and checks.
- Use Assign work to new session or an available collaboration control from the task.
- Monitor live delegated work from the transcript or Background tasks in Stage.
- Review each result and integrate only the changes that satisfy the parent task.
Expected result
Each worker has a focused transcript and returns a concrete result that can be verified before integration.
Note: Agents share only the context and authority actually provided; do not assume hidden state or credentials move between tasks.