sub-agent-delegation

You can delegate substantial tasks to parallel sub-agents by spawning them and assigning self-contained workloads to run concurrently. This reduces bottlenecks by executing research, implementation, or testing tasks in parallel and consolidating the results.

In-process delegation uses the DelegateTool to run sub-agents locally, while remote delegation uses the OpenHands Cloud API to offload tasks. Both support spawning sub-agents, assigning tasks, consolidating results, and handling errors, but remote delegation scales execution externally.

Yes, sub-agent delegation is applicable for code refactoring in isolated modules and research across multiple repositories. You can assign these substantial, self-contained tasks to specialized sub-agents to run concurrently and consolidate the results.

When running evaluations or test suites with sub-agents, the delegation process includes steps to handle errors. You manage failures during parallel execution by consolidating results and addressing errors as part of the sub-agent coordination workflow.