Werewolf 12

A highly flexible framework utilizing the features of the Python programming language for multi-Agent scenarios.

Features

• Core code is less than 100 lines.
• Capable of automatic process control.
• Usable in scenarios requiring collaboration among multiple Agents.
• Equipped with information shielding and domain control functions.
• Provides a complete 12-player Werewolf game case as a reference.

Preparations Before Use

1. Configure environment variables.
2. Modify the llm function to select the model you prefer to use.

Documentation for Functions

This document provides explanations for each function in the given Python code. The code appears to handle some form of messaging system, possibly for a game or a chat application.

Function Descriptions

split_txt_to_dict(txt)

• Purpose: Splits a given text into dictionary entries based on headers.
• Parameters:
  • txt (str): The input text to be split.
• Returns: A dictionary where the keys are headers (preceded by "###") and the values are the contents following each header.

filter_history(history, player_name, phase)

• Purpose: Filters a history of messages to include only those that should be kept for a specific player in a certain phase.
• Parameters:
  • history (list): A list of message objects.
  • player_name (str): The name of the player.
  • phase (any): The current phase to filter by.
• Returns: A list of message objects that should be kept based on player name and phase.

history_to_str(history)

• Purpose: Converts a list of message objects to a formatted string.
• Parameters:
  • history (list): A list of message objects.
• Returns: A string where each message is formatted as "【sender】content", separated by '/n'.

llm(prompt=COT_PROMPT, **kwargs)

• Purpose: Sends a prompt to a language model and retries up to twice with a wait of 2 seconds if the request fails.
• Parameters:
  • prompt (str): The prompt to send to the language model. Defaults to COT_PROMPT.
  • **kwargs: Additional keyword arguments for formatting the prompt.
• Returns: The content of the response from the language model.

Message Class (Dataclass)

• Attributes:
  • sender (str): The sender of the message. Defaults to "System".
  • receiver_limit (list): List of receivers who are allowed to receive the message.
  • content (str): The content of the message.
  • stage_limit (list): List of stages during which the message can be displayed.
• Methods:
  • is_allowed(role_name, stage)
    • Purpose: Checks if a message is allowed to be received based on role name and stage.
    • Parameters:
      • role_name (str): The role name to check.
      • stage (any): The current stage to check.
    • Returns: True if the message is allowed, otherwise False.

history_to_answer(player='Host', receivers=None, stage=None, answer_stage=[])

• Purpose: Converts the history to an answer formatted for a specific player and adds it to the history.
• Parameters:
  • player (str): The player to generate the answer for. Defaults to 'Host'.
  • receivers (list): List of receivers. Defaults to None.
  • stage (any): The current stage. Defaults to None.
  • answer_stage (list): The stages for which the answer is valid. Defaults to an empty list.
• Returns: The generated answer.

parallel_history_to_answer(players=[], receivers=None, stage=None, answer_stage=[])

• Purpose: Generates answers for multiple players in parallel and adds them to the history.
• Parameters:
  • players (list): List of player names.
  • receivers (list): List of receivers. Defaults to None.
  • stage (any): The current stage. Defaults to None.
  • answer_stage (list): The stages for which the answer is valid. Defaults to an empty list.

history_to_answer_not_adding_to_history(player='Host', stage=None)

• Purpose: Converts the history to an answer for a specific player without adding the answer to the history.
• Parameters:
  • player (str): The player to generate the answer for. Defaults to 'Host'.
  • stage (any): The current stage. Defaults to None.
• Returns: The generated answer.

add_message_to_history(content, player='System', receivers=None, stage=None)

• Purpose: Adds a new message to the history.
• Parameters:
  • content (str): The content of the message.
  • player (str): The player sending the message. Defaults to 'System'.
  • receivers (list): List of receivers. Defaults to None.
  • stage (any): The current stage. Defaults to None.