fix: Update environment variable loading to use load_dotenv and adjust episode count

inference.py

@@ -35,23 +35,23 @@ from openai import OpenAI
 from server import Task1Environment, Task2Environment, Task3Environment
 from env.schemas import Action, ActionType
 from utils import T1_SYSTEM, T2_SYSTEM, T3_SYSTEM
-from dotenv import dotenv_values
+from dotenv import load_dotenv
 
 # ─────────────────────────────────────────────────────────────────────────────
 # Configuration
 # ─────────────────────────────────────────────────────────────────────────────
 
-config = dotenv_values(".env")
-API_BASE_URL = config.get("API_BASE_URL", "https://api.openai.com/v1")
-MODEL_NAME   = config.get("MODEL_NAME",   "gpt-4o")
-HF_TOKEN     = config.get("HF_TOKEN")
+load_dotenv()  # Load from .env if available; otherwise rely on actual env vars
+API_BASE_URL = os.getenv("API_BASE_URL", "https://api.openai.com/v1")
+MODEL_NAME   = os.getenv("MODEL_NAME",   "gpt-4o")
+HF_TOKEN     = os.getenv("HF_TOKEN")
 
 if not HF_TOKEN:
     print("[WARN] HF_TOKEN not set — API calls may fail.", file=sys.stderr)
     exit(1)
 
 ENV_BENCHMARK          = "smart-contract-audit"
-NUM_EPISODES           = 1      # keep low on free tier; raise for full eval
+NUM_EPISODES           = 2      # keep low on free tier; raise for full eval
 SEED_BASE              = 42
 
 # Max LLM calls per episode (including the mandatory submit on last step).