Implement re-plan for supervisor agent

[Teneroy]

Description

The goal of this task is to implement re-planning functionality for the Supervisor Agent in Kyma Companion. The re-planner should work according to the flow described in the provided scheme. Re-planning should allow the agent to adjust its tasks and strategy based on the user input and the progress of task execution. If the initial plan cannot resolve the request, the re-planner will revise the strategy, calling other agents to ensure completion of the task.

Subtasks

1. Develop Re-planning Logic:

   • Implement the re-planner to adjust the task execution based on the previous messages, user input, and feedback from agents.
   • If the tasks do not provide a complete response, invoke the re-planner to generate an alternative strategy.

2. Add Stopping Mechanism:

   • Implement a stopping mechanism that finalizes the task and prevents infinite re-planning loops.
   • Ensure the agent finalizes the response once the tasks have been executed or a viable solution is found.

Acceptance Criteria

• The re-planner adjusts the plan correctly based on previous messages, task execution status, and user input.
• The stopping mechanism prevents infinite loops and ensures that a response is finalized and returned efficiently.
• The system is tested, and it functions as expected in different scenarios.

Diagram

[Teneroy]

https://github.com/langchain-ai/langgraph/blob/main/docs/docs/tutorials/plan-and-execute/plan-and-execute.ipynb