Testing #307987
The execution subagent only gets the run_in_terminal tool, so it can't check the output of terminals or do other things. Also, since it wants to finish its task before yielding, I see it just do random stuff while waiting until it runs out of turns
8405ab43.copilotmd
But even if it did have those tools, the async terminal strategy seems awkward with the execution subagent. If it starts async tasks, then yields, then the main agent gets the notification that the task is done, will the main agent have the right context to take action on that? Maybe, but that just needs some thinking for different scenarios.
cc @vikramnitin9
Testing #307987
The execution subagent only gets the run_in_terminal tool, so it can't check the output of terminals or do other things. Also, since it wants to finish its task before yielding, I see it just do random stuff while waiting until it runs out of turns
8405ab43.copilotmd
But even if it did have those tools, the async terminal strategy seems awkward with the execution subagent. If it starts async tasks, then yields, then the main agent gets the notification that the task is done, will the main agent have the right context to take action on that? Maybe, but that just needs some thinking for different scenarios.
cc @vikramnitin9