The latest update includes a change to how the flow of tasks is handled. Before, the agent would attempt to complete the task using tools as soon as it encountered it; now, it waits for child tasks to finish. What this means is you likely have to wait until the plan is fully expanded before it will start attempting objectives.
Actually not the best idea: in order to perform further planning in some tasks, it is strictly necessary, say, to conduct research first. For example, I want AgentOoba to prepare a patch for some code that is (in theory) available via requests_get. Instead of examining the received url (the first task generated) it starts to hallucinate about the task and the url, preparing a completely insane plan for the patch that has nothing to do with reality. (and I'm still waiting for my attempt to use AgentOoba to complete)
Actually not the best idea: in order to perform further planning in some tasks, it is strictly necessary, say, to conduct research first. For example, I want AgentOoba to prepare a patch for some code that is (in theory) available via requests_get. Instead of examining the received url (the first task generated) it starts to hallucinate about the task and the url, preparing a completely insane plan for the patch that has nothing to do with reality. (and I'm still waiting for my attempt to use AgentOoba to complete)