Replace task executor with a SmartGPT scheme

[databassgit]

Replace task executor with a SmartGPT scheme

[DataBassGit]

Hi @xtalax

This is relatively simple to implement using the existing agents we have. The 3 Output variables would be represented by separate calls to the execution agent. The Reflexion step can also be performed by the execution agent using a different persona. The final step would be performed by the status agent. It's just a slightly different prompt.

Our system is intended to be a framework where you can quickly design cognitive architectures like this rapidly. Once we implement superclasses for agents, I think adding the ability to select the persona being used as a parameter might simplify this process and make reusing agents more intuitive. What other features do you think would help you rebuild this logic more easily?

[xtalax]

We are going to need some kind of markup language (Maybe yaml?) to define different architectures and modified prompts, but then does this just become langchain with a default agent library?

Documentation of the agent model and interface would be the biggest leap forward at the moment IMO

EDIT: This idea of multiple generation and refinment can also apply more generally to other processes to improve the "intelligence" of the action