Update training.py

training.py

@@ -175,7 +175,7 @@ def supervised_warmup(model, tokenizer, n_examples=500, epochs=2):
         ]
         last_output = random.choice(last_outputs)
         # Use same prompt structure as build_prompt
-        prompt = f"""You are a code review agent.
+        prompt = f"""You are a code review agent and name your proposed fix function fix.
 
 Code:
 {code}
@@ -292,7 +292,7 @@ def build_prompt(obs, history_lines: List[str]) -> str:
     # Personality hint (optional but helpful)
     author_personality = getattr(obs, "author_personality", "defensive")  # e.g., from env
     
-    prompt = f"""You are an AI code review agent. Your goal is to convince a simulated human developer to accept your proposed fix.
+    prompt = f"""You are an AI code review agent. Your goal is to convince a simulated human developer to accept your proposed fix and name your proposed fix function fix.
 
 The developer has a **{author_personality}** personality and will only accept if you provide solid evidence:
 - Tests pass (high pass ratio)