Try the changes for system prompt + debug info

app.py

@@ -3,6 +3,7 @@ import tempfile
 from contextlib import suppress
 from io import BytesIO
 from pprint import pprint
+from time import sleep
 from typing import TypedDict, List, Dict, Any, Optional, Tuple
 from typing_extensions import Annotated
 
@@ -65,49 +66,41 @@ class BasicAgent:
         print(f"\nAgent received question: {question}")
         sys_msg = SystemMessage(
             content="""
-        You are a ReAct (Reasoning and Acting) agent. I will ask you a question. Your job is to reason through it and decide whether to act (by using a tool) or answer directly.
+        You are a ReAct (Reasoning and Acting) agent with self-reflection. For each question:
+
+        1. **Thought:** Briefly outline your reasoning step.  
+        2. **Reflect:** Check “Did I use all observations? Did my tool call succeed?”  
+        3. **Action:** Either call a tool (with arguments) or prepare your final answer.  
+        4. **Final Answer:** Provide only the bare result (no labels, no extra text).
         
-        **Instructions:**
-        1. Think step-by-step through the question, prior reasoning, and any tool observations.
-        2. Decide the best next move: use a tool, or answer directly.
-        3. Only provide a final answer when confident.
+        **Answer Format Rules**  
+        - If the answer is a number, output digits only (no commas, no units, no strings like “one”, “twenty three”).  
+        - If it’s a word or phrase, don't use articles, neither abbreviations (e.g. for cities).  
+        - If it’s a comma separated list, output a comma-separated list following the above rules for each element.  
+        - **Always** output exactly one line as an answer and nothing else.
         
-        **Guidelines:**
-        - Be thorough and logical.
-        - Use tools when more information is needed.
-        - Always ground your answer in tool results or reasoning.
-        - If a tool fails or returns nothing, acknowledge it and try an alternative.
-        - If you can't find a confident answer, return your best guess.
+        **Example 1**  
+        Q: What is 7 × 6?  
+        Thought: Multiply 7 by 6.  
+        Reflect: Simple arithmetic, no tool needed.  
+        Final Answer: 42
         
-        **Important:** 
-        - When giving the final answer, reply with only the answer — nothing else.
-        - Your final answer should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. 
-        - If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. 
-        - If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. 
-        - If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number or a string.
-        - If there is no direct ask about the response format, think if it's possible to answer with a number, if yes - do it, if no - answer with as few words as possible, not the whole sentence.
-
-        **Examples:**
-
-        Q: What is 2 + 3?  
-        Thought: Simple arithmetic.  
-        Final Answer: 5
+        **Example 2**  
+        Q: How many prime numbers are there under 20?  
+        Thought: Primes under 20 are 2, 3, 5, 7, 11, 13, 17, 19 (8 total).  
+        Reflect: Count is correct.  
+        Final Answer: 8
         
-        Q: How many days are there in a leap year?  
-        Thought: Leap years have 366 days. It's possible to answer with a number only, so the answer will be "366". 
-        Final Answer: 366
-
-        Q: What is the color opposite to "white"?  
-        Thought: "Black" color is the opposite color to "white". 
-        Final Answer: black
-
-        Q: Please, sort words "acronym", "potato" and "curriculum" alphabetically in the descending order and return the comma separated string. 
-        Thought: Since we're using the descending order there will be "potato" then "curriculum" and then "acronym". Since I need to return comma separated string the answer will be "potato, curriculum, acronym".
-        Final Answer: potato, curriculum, acronym
+        **Example 3**  
+        Q: Sort “banana”, “apple”, “cherry” alphabetically descending.  
+        Thought: Alphabetical descending: cherry, banana, apple.  
+        Reflect: Order and formatting confirmed.  
+        Final Answer: cherry, banana, apple
+        
+        ---
+        
+        Now answer the next question following this chain-of-thought + reflection pattern, and output **only** the `Final Answer` in the required format.
 
-        Q: Who was the first person to walk on the Moon?  
-        Thought: Neil Armstrong was the first person to walk on the Moon. But I need to be as concise as possible, so it should be "Neil Armstrong".
-        Final Answer: Neil Armstrong
         """
         )
             
@@ -160,6 +153,7 @@ class BasicAgent:
     )
     def read_text_file_tool(file_name: str) -> str:
         print(f"\nCalling read text file tool for", file_name)
+        print("File metadata:", os.stat(file_name))
         with open(file_name, 'r') as f:
             return f.read()
 
@@ -175,7 +169,10 @@ class BasicAgent:
         Returns:
           The LLM’s answer based on the image content.
         """
-        # Load & save as bytes so the vision model can consume it
+        if not os.path.exists(path):
+            return f"Error: file not found at {path}"
+
+        print("File metadata:", os.stat(path))
         img = Image.open(path)
         img_bytes = BytesIO()
         img.save(img_bytes, format=img.format)
@@ -200,7 +197,8 @@ class BasicAgent:
         """
         if not os.path.exists(path):
             return f"Error: file not found at {path}"
-    
+
+        print("File metadata:", os.stat(path))
         audio = AudioSegment.from_file(path)
         tmp_path = os.path.join(tempfile.gettempdir(), "tmp_audio.wav")
         audio.export(tmp_path, format="wav")
@@ -233,7 +231,7 @@ class BasicAgent:
         Returns:
           The string form of the content.
         """
-        df = pd.read_excel(path)
+        df = pd.read_excel(path, engine='openpyxl')
         return str(df.to_csv(index=False))
     
 
@@ -311,6 +309,8 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             submitted_answer, logs = agent(question_text)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
+            print(f"\n\n\n==============Finishing task id: {task_id}, question_text: {question_text}==============\n\n\n")
+            sleep(2)
         except Exception as e:
              print(f"Error running agent on task {task_id}: {e}")
              results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})