cnt_critic_agents: 3


max_loop_rounds: 2


max_criticizing_rounds: 3


human_eval: false


task_description: |-
  build a web application of a simple calculator.

  

evaluation_dimensions: |-
  


prompts:
  role_assigner_prompt: &role_assigner_prompt |-
    You are the leader of a group of experts, now you are faced with a task:
    
    ${task_description}
    
    You can recruit ${cnt_critic_agents} expert team members in different regions.
    What experts will you recruit to better generate good ideas?
    
    Output format example:
    1. an electrical engineer specified in the filed of xxx
    2. an economist who is good at xxx
    3. a lawyer with a good knowledge of xxx
    ...
    
    ${advice}
    You don't have to give the reason.

  solver_prompt: &solver_prompt |-
    You are faced with the task:
    ${task_description}
    
    You formerly gave the following solution
    ${former_solution}
    
    But other critics in the group are not satisfied with it.
    They gave the following opinions:
    ${critic_opinions}
    
    Now you are going to give a new solution,
    based upon your former solution and the critics' opinions.
    Please write code only!
    
    If no former solution or critic opinions are given, you can ignore this part, and just generate a new solution.

  summarizer_prompt: &summarizer_prompt |-
    You are a summarizer. 
    Your task is to categorize and summarize the ideas in the chat history.
    Please add the speaker of each idea to the beginning of the content.


    The question of the discussing is to ${task_description}.
    
    #Output format
    1. (Speaker1): (Ideas of Speaker 1 in a single line)
    2. (Speaker2): (Ideas of Speaker 2 in a single line)
    3. (Speaker3): (Ideas of Speaker 3 in a single line)
    ...
    
    Here is the content you have to summarize:
    ${former_solution}
    ${critic_opinions}

    Please merge all ideas of one speaker into one item.

  critic_prompt: &critic_prompt |-
    Now you are ${role_description}
    
    You are in a discussion group, aiming to ${task_description}.
    
    Now the group is going to give a preliminary solution as the following:

    ${preliminary_solution}
    
    Now the group is asking your opinion about it. Based on your knowledge
    in your field, do you agree that this solution can perfectly 
    solve the problem?
    Or do you have any ideas to improve it?
    
    - If you thinks it is perfect, use the following output format:
    Action: Agree
    Action Input: Agree.
    (Do not output your reason for agreeing!)

    - If you want to give complemented opinions to improve it or to contradict with it, use the following output format:
    Action: Disagree
    Action Input: (what you want to say in one line)
    
    P.S. Always remember you are ${role_description}!
    
    ${advice}
    If no former solution or critic opinions are given, you can just disagree and output your idea freely, based on the expertise of your role.
    Remember, the ideas should be specific and detailed enough, not just general opinions.
    
    Please control output code in 2048 tokens! (Write concise code)

  evaluator_prompt: &evaluator_prompt |-
    You are an professional code reviewer.
 
    Your task is to evaluate the solution.
    The code is to ${task_description}.
    Your task is to evaluate the codes written by the code engineers.
      Please not only give a general rating points (from 0 to 10) but also give detailed comments about where and how the code can be improved.
      Please consider the following aspects when you are evaluating the code:
        1. The code should be able to run without any errors.
        2. The code should be able to achieve the goal specified in the task description.
        3. The code should be easy to read and understand, efficient, concise and elegant.
        4. The code should be robust.
    
    Please rate the code in the following dimensions:
        1. Completeness: Is the code snippet complete enough without unimplemented functions of methods? Is it able to run without any errors?
        2. Functionality: Is the code able to achieve the goal specified in the task description?
        3. Readability: Is the code easy to read and understand, efficient, concise and elegant?
        4. Robustness: Is the code snippet able to handle different unexpected input or other exceptions?
    
    0 means the idea looks like beginner's work, and 10 means the idea is perfect in that aspect, like a master.
    
    and then in the fifth line of output, give your detailed advice for the engineers to better generate good codes.
    
  
    #Output format
    You must output in the following format:
    1. Completeness: (a score between 0 and 9)
    2. Functionality: (a score between 0 and 9)
    3. Readability: (a score between 0 and 9)
    4. Robustness: (a score between 0 and 9)
    5. Advice: (your advice in one line)
    
    Here is the content you have to evaluate:
    ${solution}
    

name: pipeline


environment:
  env_type: task-basic
  max_loop_rounds: 3
  rule:
    order:
      type: sequential
    visibility:
      type: all
    selector:
      type: basic
    updater:
      type: basic
    describer:
      type: basic

agents:
  - #role_assigner_agent:
    agent_type: role_assigner
    name: role assigner
    prompt_template: *role_assigner_prompt
    memory:
      memory_type: chat_history
    llm:
      llm_type: gpt-4
      model: "gpt-4"
      temperature: 0
      max_tokens: 256
    output_parser:
      type: role_assigner

  - #solver_agent:
    agent_type: solver
    name: Planner
    prompt_template: [*solver_prompt, *summarizer_prompt]
    memory:
      memory_type: chat_history
    llm:
      llm_type: gpt-4
      model: "gpt-4"
      temperature: 0
      max_tokens: 2048

  - #critic_agents:
    agent_type: critic
    name: Critic 1
    role_description: |-
      Waiting to be assigned.
    prompt_template: *critic_prompt
    memory:
      memory_type: chat_history
    llm:
      llm_type: gpt-4
      model: "gpt-4"
      temperature: 0
      max_tokens: 256
    output_parser:
      type: critic

  - #executor_agent:
    agent_type: executor
    name: Executor
    prompt_template: None
    memory:
      memory_type: chat_history
    llm:
      llm_type: gpt-4
      model: "gpt-4"
      temperature: 0
      max_tokens: 512

  - #evaluator_agent:
    agent_type: evaluator
    name: Evaluator
    role_description: |-
      Code Reviewer
    prompt_template: *evaluator_prompt
    memory:
      memory_type: chat_history
    llm:
      llm_type: gpt-4
      model: "gpt-4"
      temperature: 0
      max_tokens: 128
    output_parser:
      type: evaluator
      dimensions:
        - Completeness
        - Functionality
        - Readability
        - Robustness


tools: