Unite supervisor agent under one subgraph

[Teneroy]

Description

Currently our planner, "supervisor agent"(TaskExecutor), and finalizer are separate nodes of a graph. We want to unite everything with the same subgraph called "Supervisor Agent", since our supervisor agent is a combination of those nodes.

Acceptance criterias:

• [x] All nodes are united under the same subgraph

Diagram

[marcobebway]

TODOs:

• [x] Make sure the finalizer has the following information in its prompt:
  • [x] The user query.
  • [x] The list of agents.
• [x] Could the Planner within Superviser use gpt4o-mini model?
  • No, because both the planner and the finalizer requires gpt4o.
• [x] Update documentation
• [x] Fix the unit tests
• [x] Fix the integration tests
• [x] Cover the new changes with integration tests
  • Common Node
  • [x] Common node correctly answers non-technical query
  • [x] Common node correctly answers general programming related query
  • Planner Node
  • [x] General query is immediately answered
  • [x] Kyma query -> Kyma agent
  • [x] Kubernetes query -> Kubernetes agent
  • [x] Kyma & Kubernetes query -> Kyma & Kubernetes agent
  • [x] Kyma & Common query -> Kyma & Common agent
  • [x] Kyma & Kyma query -> two subtasks for Kyma agent
  • Router Node
  • [x] LLM call to compare output is not needed -> replace with simple string comparison
  • [x] Single task:
    • [x] Assigned K8sAgent is 'Next' agent
    • [x] Assigned KymaAgent is 'Next' agent
    • [x] Assigned Common is 'Next' agent
  • [x] Multiple tasks, all still pending:
    • [x] First assigned K8sAgent is 'Next' agent
    • [x] First assigned KymaAgent is 'Next' agent
    • [x] First assigned Common is 'Next' agent
    • [x] Multiple tasks, the first one is completed:
    • [x] Second pending K8sAgent is 'Next' agent
    • [x] Second pending KymaAgent is 'Next' agent
    • [x] Second pending Common is 'Next' agent
  • Finalizer Node
  • [x] Accepts Kyma response
  • [x] Accepts Kubernetes response
  • [x] Apologizes for wrong Kyma response
  • [x] Apologizes for wrong Kubernetes response
  • [x] Create ticket to improve Finalizer response and add integration test cases -> Ticket #240
    • [x] Add a syntax scoring evaluation function. It should give a bad score if:
      • Any of the agent names are in the final response.
      • Any extra headers/sub-headers were added to the final response.
    • [x] Test two agents, one responding with invalid answers, and the other responding with valid answers.
      • The valid answers should be included in the final response.
      • The invalid answers should be removed from the final response.
      • Apology for the invalid answers should be included in the final response.

[grischperl]

Documentation PR: https://github.tools.sap/kyma/ai-force/pull/232