Are the SWE Bench tests compatible with the Delegate Tool? I would like to add it to the default tool kit but I don't want to lose the token usage in the logs

What do you mean? Which tests? Delegate Tool was deprecated. Now we are using the TaskToolSet to call subagents. Which version are you using? (which version of the sdk)

the openhands benchmark repo

1. Yes, the docs should be updated 😞 This is the new one: https://docs.openhands.dev/sdk/guides/task-tool-set 2. What do you mean with SWE Bench tests ? Why are you scared to lose the token usage in the logs?

im using 1.24. Well before I was forcing the delegate tool like this: class DelegateTool(ToolDefinition): name = "DelegateTool" @classmethod def create(cls, conv_state, max_children: int = 1, confirmation_handler=None): return [ cls( description="Delegation tool for spawning sub-agents and delegating tasks.", action_type=DelegateAction, observation_type=DelegateObservation, annotations=ToolAnnotations( title="delegate", readOnlyHint=False, destructiveHint=True, idempotentHint=False, openWorldHint=True, ), executor=DelegateExecutor( max_children=max_children, confirmation_handler=confirmation_handler, ), ) ] register_tool(DelegateTool.name, DelegateTool) register_builtins_agents(enable_browser=False)

Ok but this doesn't answer my question 🙂 What do you mean with SWE Bench tests ? Are you running the SWE benchmark? What are you trying to do? You should be able to import the TaskToolSet and to use it (as this tool is in openhands tool and it is installed). Moreover, the trajectories of the sub-agents are saved in a folder inside the trajectory folder of the main agent (the folder is called subagents)

Yes the SWE benchmark. I just wanted to change the harness and run the tests as normal. I want to do an evaluator generator harness (the anthropic one).

Hello @modern-fireman-99755! I was looking at those docs but TaskToolSet is not available in openhands-sdk or openhands-tools. Where does it come from? I am on version 1.11.0. I also checked 1.10.0

@alert-energy-60278 are you speaking of the benchmark repo or sdk? I mean 1.11.0 is the benchmark version?

This one: github.com/OpenHands/OpenHands/releases#…