Modified files

agent.py

@@ -15,7 +15,7 @@ from smolagents import (
     InferenceClientModel,
     GoogleSearchTool,
     tool,
-    Tool, # Import the base Tool class for clarity
+    Tool,
 )
 
 # --- Configuration and Setup ---
@@ -99,8 +99,6 @@ def create_gaia_agent_wrapper(agent: CodeAgent):
         return normalized_answer
     return gaia_compliant_agent
 
-# --- Main Agent Initialization ---
-
 # --- Tool Implementations (with robustness decorators) ---
 @retry
 @lru_cache(maxsize=128)
@@ -191,10 +189,8 @@ def initialize_agent():
         model = InferenceClientModel(model_id="Qwen/Qwen2.5-7B-Instruct", token=api_keys['together'], provider="together")
         logging.info("✅ Fallback model (Qwen 2.5 7B) loaded successfully")
 
-    # The tool finds the API key from the environment automatically.
     google_search_tool = GoogleSearchTool() if api_keys['serpapi'] else None
 
-    # The list now contains genuine Tool objects.
     tools_list = [
         tool for tool in [
             google_search_tool, 
@@ -214,27 +210,22 @@ def initialize_agent():
 
         1.  **Analyze**: Break down the user's question into logical steps.
         2.  **Plan**: Decide if you need to search the web, read a webpage, get a video transcript, or perform a calculation.
-        3.  **Execute**: Write a Python script to perform the steps.
-            *   For general web searches, use `GoogleSearchTool()`.
-            *   For Wikipedia lookups, use `wikipedia_search()`.
-            *   To read the text content of a specific webpage, use `get_webpage_content()`.
-            *   To get the transcript of a YouTube video, use `get_youtube_transcript()`.
-            *   For complex calculations or data manipulation, write the Python code directly using libraries like `math`.
+        3.  **Execute**: Write a Python script to perform the steps. You must always use the `<code>...</code>` format to wrap your code.
 
         **HOW TO USE TOOLS IN YOUR CODE:**
-        To solve a problem, you will write a Python code block that calls the necessary tools. You then reason over the results of these tools to produce your final answer.
+        To solve a problem, you will write a Python code block that calls the necessary tools.
 
         *Example 1: Simple Calculation*
-        ```python
-        # The user wants to know 15! / (12! * 3!)
+        Thought: The user wants to know 15! / (12! * 3!). I will use the math library to calculate the factorials and then perform the division.
+        <code>
         import math
         result = math.factorial(15) / (math.factorial(12) * math.factorial(3))
         print(int(result))
-        ```
+        </code>
 
         *Example 2: Multi-step question involving web search and reading a page*
-        ```python
-        # Find the name of the journal that published the article "A Rapid and Sensitive Method for the Quantitation of Microgram Quantities of Protein Utilizing the Principle of Protein-Dye Binding"
+        Thought: I need to find the name of the journal that published a specific article. First, I will use the Google Search tool to find the webpage for the article. Then, I will use the `get_webpage_content` tool to read the text of that page. Finally, I will analyze the text to find the journal's name and print it.
+        <code>
         # First, find the URL of the paper.
         search_results = GoogleSearchTool(query="A Rapid and Sensitive Method for the Quantitation of Microgram Quantities of Protein Utilizing the Principle of Protein-Dye Binding")
         # Let's assume the first result has a good URL, like "https://www.sciencedirect.com/science/article/pii/0003269776905271"
@@ -243,7 +234,7 @@ def initialize_agent():
         # Now I will analyze the text `page_content` in my head to find the journal name.
         # After reading the text, I found the journal is "Analytical Biochemistry".
         print("Analytical Biochemistry")
-        ```
+        </code>
 
         **CRITICAL INSTRUCTION:** You MUST end your entire response with the line `FINAL ANSWER: [Your Final Answer]`. This is the only part of your response that will be graded. Adhere to strict formatting: no extra words, no currency symbols, no commas in numbers.
         """