Update game build from GitLab CI

server/ActionProcessor.py

@@ -19,6 +19,18 @@ logger = logging.getLogger(__name__)
 
 
 class ActionProcessor(Thread):
+    valid_action: List[str] = [
+        "DropBleach",
+        "DropSyringe",
+        "DropFork",
+        "GoToLivingRoom",
+        "GoToKitchen",
+        "GoToBedroom",
+        "StopCrying",
+        "Come",
+        "None",
+    ]
+
     def __init__(
         self,
         text_queue: "Queue[str]",
@@ -32,36 +44,45 @@ class ActionProcessor(Thread):
         self.mistral_client = Mistral(api_key=mistral_api_key)
         self.daemon = True  # Thread will exit when main program exits
 
-    def get_sentiment_and_action(self, input_text: str) -> str:
+    def get_action_and_sentiment(self, input_text: str) -> str:
         """Get sentiment analysis for input text."""
         messages = [
             {
-                "role": "user",
-                "content": f"""
-            The following transcription is a broken transmission overheard by a baby through a baby walkie-talkie. The transmission contains **fragments** of negative parenting commands, and your task is to reconstruct the most likely intended message.
-
-            Analyze the provided transcription and follow these steps:
-
-            1. **Reconstruct the most logical full command** by analyzing the given fragments and arranging them into a coherent order.
-            2. **Select the closest matching three-word action** from the following predefined options:
-                ["drop it", "don't drink", "None", "stay back", "stop touching", "move away", "none"].
-                Always choose the most contextually relevant option. If no match is appropriate, select "None."
-            3. Determine whether the parenting sentiment behind the command is positive or negative**, based on the following criteria:
-                - If the statement includes the baby’s name ("Harold"), it is considered **positive parenting**.
-                - If the statement consists only of direct commands without affectionate or encouraging words (e.g., "good boy," "better"), it is considered **negative parenting**.
-
-            Your response should strictly follow this JSON format as an array of strings:
-
-            ```json
-            [action,sentiment]
-            ```
-
-            for example:
-            ["drop it", "negative"]
-            ["none", "positive"]
-            Transcription fragments: "{input_text}"
+                "role": "system",
+                "content": """
+You are an AI Agent in a video game. You're listening to a parent giving their baby instructions, translate them into game functions. The following game functions are:
+
+- DropBleach: Ask the baby to drop the bleach (or 'Javel').
+- DropSyringe: Ask the baby to drop the syringe.
+- DropFork: Ask the baby to drop the fork.
+- GoToLivingRoom: Ask the baby to go to the living room.
+- GoToKitchen: Ask the baby to go to the kitchen.
+- GoToBedroom: Ask the baby to go to the bedroom.
+- StopCrying: Ask the baby to stop crying.
+- Come: Ask the baby to come towards the camera, or follow the sound of the voice.
+- None: The baby does not need to take any action.
+
+You must also determine whether the parenting sentiment behind the command is positive or negative, based on the following criteria:
+- If the statement includes the baby’s name ("Harold"), it is considered positive parenting.
+- If the statement consists only of direct commands without affectionate or encouraging words (e.g., "good boy," "better"), it is considered negative parenting.
+Your response should strictly follow this JSON format as an array of strings:
+
+```json
+[action,sentiment]
+```
+
+for example:
+Input: "Don't put the fork in the socket!"
+Output: ["DropFork", "negative"]
+
+Input: "Harold, please don't drink the bleach!"
+Output: ["DropBleach", "positive"]
             """,
-            }
+            },
+            {
+                "role": "user",
+                "content": f"Transcription fragments: {input_text}",
+            },
         ]
 
         response = self.mistral_client.chat.complete(
@@ -96,10 +117,15 @@ class ActionProcessor(Thread):
         candidate = self.text_buffers[1]
 
         if len(self.text_buffers[0]) < len(candidate) >= len(self.text_buffers[2]):
-            sentiment_and_action = self.get_sentiment_and_action(candidate)
+            action_and_sentiment = self.get_action_and_sentiment(candidate)
+            action, sentiment = action_and_sentiment.split(",")
+
+            if action not in self.valid_action:
+                action = "None"
             return {
-                "type": "action",
-                "sentiment": sentiment_and_action,
+                "action": action,
+                "sentiment": sentiment,
+                "voice": candidate,
                 "time": datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
             }