raise error when llm did not conform to output format

new_gui_development_plan.md

@@ -10,37 +10,23 @@ This plan outlines the development of a comprehensive GUI for the LLMPostor proj
     - Visualize voting results using a clear and intuitive representation:
         - Bar chart: Show the number of votes for each player.
 
-**Phase 3: LLM Interaction Visualization**
-
-1. **LLM Interaction Visualization:**
-    - Create a chat window to visualize prompts and responses of player LLMs.
-    - Implement a mechanism to collapse prompts by default, allowing users to expand them when needed.
-    - Display prompts and responses in a chronological order.
-
 
 **Additional Considerations:**
 
 - **Scenario Selection:** Allow users to choose from a library of pre-defined scenarios or create their own.
     - Implement a scenario selection menu or dropdown.
     - Provide options to customize scenario parameters (e.g., number of players, impostors, task difficulty).
-- **Player Customization:** Allow users to customize player names, avatars, and LLM models.
-    - Implement a player customization section.
+- **Player Customization:** Allow users to customize player LLM models.
     - Provide a list of available LLM models and allow users to select their preferred options.
-- **Game History:** Provide a log of past game turns, including player actions, discussion messages, and voting results.
-    - Implement a game history section that displays a chronological record of game events.
-    - Allow users to filter and search through the game history.
-- **Data Export:** Allow users to export game data (e.g., player actions, discussion transcripts, voting results) in various formats (e.g., CSV, JSON).
+- **Data Export:** Allow users to export game state.
     - Implement an export functionality that allows users to save game data for further analysis.
 
 This development plan provides a structured approach to building a comprehensive and user-friendly GUI for the LLMPostor project. By following these phases and considering the additional features and extensions, you can create a powerful tool for analyzing and visualizing the persuasive abilities of LLMs in a game-like environment. 
 
 
 **Phase ????**
-        - **Turn Control:**
-            - Slider to navigate through game turns.
-            - "Previous Turn" and "Next Turn" buttons for sequential navigation.
         - **Game Control:**
-            - "Resume" and "Stop" buttons to control game execution.
+            - "Resume" and "Stop" switch to control game execution.
             - "Save State" functionality to save the current game state to a separate file.
         - **Player Action Selection:**
             - Section for manual action selection, allowing users to choose actions for a specific player.

src/llm_postor/game/agents/adventure_agent.py

@@ -59,6 +59,7 @@ def check_action_valid(self, chosen_action: str) -> int:
         else:
             warning_str = f"{self.player_name} LLM did not conform to output format. Expected one of {normalized_available_actions}, but got {chosen_action} ({normalized_chosen_action} normalized)"
             print(warning_str)
+            raise ValueError(warning_str)
             self.responses.append(warning_str)
             return 0
 

src/llm_postor/game/agents/voting_agent.py

@@ -40,6 +40,7 @@ def check_action_valid(self, chosen_action: str) -> int:
         else:
             warning_str = f"{self.player_name} LLM did not conform to output format. Expected one of {normalized_available_actions}, but got {chosen_action} ({normalized_chosen_action} normalized)"
             print(warning_str)
+            raise ValueError(warning_str)
             self.responses.append(warning_str)
             return 0  # Default to first action if invalid