Upload 6 files

Dockerfile

@@ -0,0 +1,17 @@
+# Use a lightweight Python image
+FROM python:3.9-slim
+
+# Set the working directory in the container
+WORKDIR /app
+
+# Copy all your files (env.py, models.py, etc.) into the container
+COPY . .
+
+# Install the required libraries
+RUN pip install --no-cache-dir fastapi pydantic openenv-core uvicorn
+
+# Expose the port your env.py is running on
+EXPOSE 8000
+
+# Command to run your environment server
+CMD ["python", "env.py"]

env.py

@@ -0,0 +1,77 @@
+import asyncio
+from typing import Optional
+from types import SimpleNamespace
+from openenv.core.env_server import Environment
+from models import MyEnvV4Observation, MyEnvV4Action
+
+class MyEnvV4Env(Environment):
+    def __init__(self):
+        super().__init__()
+        # Realistic Dataset with Digital Seduction/Phishing markers
+        self.dataset = [
+            {
+                "sender": "dean.office@manipal.edu",
+                "subject": "B.Tech Lab Exam Schedule",
+                "body": "Please find the attached PDF for the upcoming CSE lab exams.",
+                "headers": ["SPF: Pass", "DKIM: Pass"],
+                "label": "INBOX"
+            },
+            {
+                "sender": "verify-account@security-amazon.net",
+                "subject": "Urgent: Your account is locked!",
+                "body": "Digital Seduction Alert: High urgency used. Click http://bit.ly/fake-link to unlock.",
+                "headers": ["SPF: Fail", "DMARC: Fail"],
+                "label": "QUARANTINE"
+            },
+            {
+                "sender": "prize@lottery-winner.co",
+                "subject": "Congratulations! You won $10,000",
+                "body": "Reply with your bank details to claim your cash prize immediately.",
+                "headers": ["SPF: Neutral"],
+                "label": "SPAM"
+            }
+        ]
+        self.current_step = 0
+
+    async def reset(self):
+        self.current_step = 0
+        return self._get_result()
+
+    def _get_result(self, reward=0.0, done=False):
+        if self.current_step >= len(self.dataset):
+            obs = MyEnvV4Observation(
+                sender="N/A", subject="N/A", body="N/A", 
+                headers=[], echoed_message="End of Data"
+            )
+            return SimpleNamespace(observation=obs, reward=reward, done=True)
+        
+        data = self.dataset[self.current_step]
+        obs = MyEnvV4Observation(
+            sender=data["sender"],
+            subject=data["subject"],
+            body=data["body"],
+            headers=data["headers"],
+            echoed_message=f"Step {self.current_step + 1}"
+        )
+        return SimpleNamespace(observation=obs, reward=reward, done=done)
+
+    async def step(self, action: MyEnvV4Action):
+        if self.current_step >= len(self.dataset):
+            return self._get_result(done=True)
+
+        correct_label = self.dataset[self.current_step]["label"]
+        # Exact match reward logic for 0.0 - 1.0 range
+        reward = 1.0 if action.message.strip().upper() == correct_label else 0.0
+        
+        self.current_step += 1
+        done = self.current_step >= len(self.dataset)
+        
+        return self._get_result(reward=reward, done=done)
+
+    async def close(self):
+        pass
+
+    @classmethod
+    async def from_docker_image(cls, image_name: str):
+        """Simulated helper for local/containerized runs."""
+        return cls()

inference.py

@@ -0,0 +1,73 @@
+import asyncio
+import os
+import textwrap
+from typing import List, Optional
+from openai import OpenAI
+from my_env_v4 import MyEnvV4Action, MyEnvV4Env
+
+# Environment Configuration
+API_BASE_URL = os.getenv("API_BASE_URL") or "https://router.huggingface.co/v1"
+API_KEY = os.getenv("HF_TOKEN") or os.getenv("API_KEY")
+MODEL_NAME = os.getenv("MODEL_NAME") or "Qwen/Qwen2.5-72B-Instruct"
+TASK_NAME = "email-triage"
+BENCHMARK = "mit-manipal-v4"
+MAX_STEPS = 3
+SUCCESS_THRESHOLD = 0.5
+
+SYSTEM_PROMPT = """
+You are an Email Security Agent. Triage the following email based on sender, headers, and body content.
+Digital Seduction Rules:
+- 'INBOX': Official domains (.edu, .gov) and passed security headers.
+- 'SPAM': Marketing, gambling, or generic lottery win claims.
+- 'QUARANTINE': Phishing, high-urgency threats, suspicious links (.net, .co), or failed headers (SPF/DMARC Fail).
+
+REPLY WITH EXACTLY ONE WORD: 'INBOX', 'SPAM', or 'QUARANTINE'.
+"""
+
+def log_start():
+    print(f"[START] task={TASK_NAME} env={BENCHMARK} model={MODEL_NAME}", flush=True)
+
+def log_step(step, action, reward, done):
+    print(f"[STEP] step={step} action={action} reward={reward:.2f} done={str(done).lower()} error=null", flush=True)
+
+def log_end(success, steps, score, rewards):
+    r_str = ",".join(f"{r:.2f}" for r in rewards)
+    print(f"[END] success={str(success).lower()} steps={steps} score={score:.3f} rewards={r_str}", flush=True)
+
+async def main():
+    client = OpenAI(base_url=API_BASE_URL, api_key=API_KEY)
+    env = MyEnvV4Env() # Local instance for testing, can use from_docker_image if needed
+    
+    rewards = []
+    log_start()
+    
+    try:
+        result = await env.reset()
+        for step in range(1, MAX_STEPS + 1):
+            if result.done: break
+            
+            obs = result.observation
+            prompt = f"Sender: {obs.sender}\nSubject: {obs.subject}\nBody: {obs.body}\nHeaders: {obs.headers}"
+            
+            # OpenAI Call
+            response = client.chat.completions.create(
+                model=MODEL_NAME,
+                messages=[{"role": "system", "content": SYSTEM_PROMPT}, {"role": "user", "content": prompt}],
+                max_tokens=10,
+                temperature=0.0 # Deterministic for testing
+            )
+            action_text = response.choices[0].message.content.strip().upper()
+            
+            result = await env.step(MyEnvV4Action(message=action_text))
+            rewards.append(result.reward)
+            
+            log_step(step, action_text, result.reward, result.done)
+            if result.done: break
+
+        total_score = sum(rewards) / MAX_STEPS
+        log_end(total_score >= SUCCESS_THRESHOLD, len(rewards), total_score, rewards)
+    finally:
+        await env.close()
+
+if __name__ == "__main__":
+    asyncio.run(main())

models.py

@@ -0,0 +1,15 @@
+from pydantic import Field
+from typing import Optional, List
+from openenv.core.env_server import Action, Observation
+
+class MyEnvV4Observation(Observation):
+    """What the Agent sees."""
+    sender: str
+    subject: str
+    body: str
+    headers: List[str]
+    echoed_message: str = "" # Required by sample inference script contract
+
+class MyEnvV4Action(Action):
+    """What the Agent chooses."""
+    message: str = Field(..., description="Action string: 'INBOX', 'SPAM', or 'QUARANTINE'")

openenv.yaml

@@ -0,0 +1,19 @@
+# Metadata for OpenEnv Triage Agent
+name: "mail-triage-v4"
+version: "1.0.0"
+description: "An automated triage agent for university mailboxes, specializing in Digital Seduction detection."
+
+# Environment Specification
+repo_url: "https://huggingface.co/spaces/YOUR_USERNAME/YOUR_SPACE_NAME"
+task_type: "classification"
+
+# Compliance Metrics
+reward_range: [0.0, 1.0]
+tags:
+  - security
+  - nlp
+  - mit-manipal-hackathon
+
+# Typed Model References
+observation_space: "models.MyEnvV4Observation"
+action_space: "models.MyEnvV4Action"

requirements.txt

@@ -0,0 +1,5 @@
+fastapi==0.109.0
+uvicorn==0.27.0
+pydantic==2.6.1
+openenv-core==0.1.5
+openai==1.12.0