Implement Consensus Mechanism for Multi-Model Operations

[kordless]

Issue Summary

To enhance the consistency of steps taken by the agent when completing tasks for the user, we propose implementing a consensus mechanism among multiple LLMs. This mechanism ensures that when there are multiple tokens or inputs from the user, two models will work together to come to an agreement on the steps to be taken. If the task involves more than two steps, one model will be designated as in charge and will confer with the other.

Desired Outcome

• Consistency: Ensure consistent steps are taken by the agent to complete the user's task.
• Consensus Building: Allow two models to come to an agreement before proceeding with the task.
• Role Assignment: In tasks requiring more than two steps, assign one model as the lead to oversee the task execution.
• Configurability: Make the mechanism flexible where the models that communicate with each other are configurable.
• Logging: Log all actions and agreements made between models.

Steps to Implement

1. Check if more than one model is available.
2. Implement a double-checking mechanism where the primary function model asks the secondary model for its input.
3. Log the consensus process and the agreed steps, indicating the assigned roles and actions.

Priority and Severity

• Priority: High
• Impact: High

[kordless]

Change the system prompt implmeentation to use a dynamic system prompt that changes based on the portion of the workflow that is being processed. So, for a new issue, we get a plan back immediately (i.e. we don't call a bunch of funcitons first).