Update agent.py

agent.py

@@ -183,7 +183,7 @@ class BasicAgent:
             model=model,
             tools=tools,
             add_base_tools=False,
-            max_steps=8,
+            max_steps=10,
         )
         self.agent.system_prompt = (
             """
@@ -235,6 +235,18 @@ class BasicAgent:
             - If a question has multiple valid interpretations, choose the **narrowest, most literal** one.
             - If the answer is not found, say `[ANSWER] - unknown`.
 
+            Hard rules
+            ──────────
+            1. Think internally as much as you like, but **never reveal** chain-of-thought, tool traces, or explanations.
+            2. If the correct reply is unknown or the question is invalid, reply exactly  
+               `[ANSWER]unknown`.
+            3. Numerical replies → digits only (no commas, no units, no words).  
+               String replies → lowercase, no leading/trailing spaces, no articles (“a”, “the”).  
+               Lists → comma-separated, alphabetically sorted, no spaces after commas.
+            4. If the question asks for a set size, return the **count**, not the set.
+            5. After using any tools, stop and output the final line; do **not** echo tool output.
+            6. Violating any rule or adding extra text causes the run to be scored wrong.
+
             ---
 
             You must follow the examples (These answers are correct in case you see the similar questions):