feature: Dual boost improves prompt before passing it down to the provided AIs

[omarcresp]

Feature request

Im using the dual boost feature with the following setup
main model (reasoning): deepseek-r1 (kluster)
first_provider (coder): claude 3.5
second_provider (coder): qwen-coder-32b (groq)

And having a reasoning model evaluating the coder results have gave me great results (high response times tough) but I think this approach could be better if instead of prompt -> providers -> main model -> response the flow could be something like prompt -> main model -> providers -> main model -> response

I have manually inserted the second flow in chat apps and I think it gives better results, specially when you didnt created a huge prompt, just asked in chat something to be change, then main model reasoned to improved the prompt based in the codebase context and the question and calls providers ai with an improved version of the user prompt

I can code this change, but since it would be a "refactor" on how the dual_boost feature works I rather ask you before start writing lines and lines of code that may or may not be included. Let me know your thoughts about this process and if you think is something wort exploring to add to the plugin @yetone

Motivation

Im trying to get agent like behavior if using dual_boost. and I think this approach could get even more results (after tested it in a very manual way using chats)

Other

No response