Add comments for LLM proxy usage and enhance task instructions

basic_agent.py

@@ -56,6 +56,7 @@ class BasicAgent:
             api_key=os.getenv("GROQ_API_KEY")
         )
 
+        # use LLM proxy server to avoid rate limits issues
         model = LiteLLMModel(
             model_id="groq/deepseek-r1-distill-llama-70b",
             api_base="http://127.0.0.1:4000"
@@ -95,22 +96,29 @@ class BasicAgent:
 Your primary goal is to accurately and concisely answer the provided question using your available tools and any supplied information.
 
 Key Instructions:
-1.  **Understand the Task**: Carefully read the entire question. Crucially, examine the "File Information" section that follows the question. This section will state if a local file is available for this task and provide its exact path if so.
-2.  **Select Tools Strategically**:
-    *   Choose the best tool(s) based on the question and any file details. Tool descriptions will help guide your choice.
-    *   For web research, use search tools. Use `VisitWebpageTool` for exploring specific URLs.
-    *   **Working with Local Files**: If the "File Information" section provides a local file path:
-        *   You **MUST** use the `PythonInterpreterTool` if the file's content is needed to answer the question.
-        *   Your Python code for the tool should open this file using the full, exact local path given in "File Information". Then, read, process, or execute its content as appropriate.
-    *   For general calculations, data analysis (not from a local file mentioned in "File Information"), or other Python tasks, use `PythonInterpreterTool`.
-3.  **Verify and Combine Information**:
-    *   If multiple steps or tools are used, synthesize the information.
-    *   Always evaluate information for accuracy and direct relevance to the question asked.
-    *   Base your final answer *only* on the information you have gathered.
-4.  **Formulate Your Response**:
+1.  **Understand the Task**: Carefully read the entire question. Pay close attention to the "File Information" section provided below the question; it states if a local file is available and its path.
+2.  **Strategize and Select Tools**:
+    *   Choose the most appropriate tool(s) based on the question, file details, and tool capabilities.
+    *   For web research, use `web_search` or `wikipedia_search`. Use `visit_webpage` to get content from specific URLs found or given.
+    *   **Working with Local Files**: If the "File Information" section provides a local file path *and that file's content is needed*:
+        *   The primary tool to access this local file is the `PythonInterpreterTool`.
+        *   Generate Python code for this tool to open the file (using its full, exact local path from "File Information"), then read, process, or execute its content as required.
+    *   For general calculations, data analysis not involving a mentioned local file, or other Python tasks, use `PythonInterpreterTool`.
+3.  **Information Analysis & Verification**:
+    *   When examining content from web pages or files, critically evaluate the information.
+    *   **Pay specific attention to details like names, dates, and roles (e.g., distinguish between a nominator, a promoter, an author, etc.).**
+    *   Synthesize information from multiple steps or tools if necessary.
+    *   Base your final answer *only* on confirmed facts from the information gathered. Do not make assumptions.
+4.  **Handling Tool Errors**:
+    *   If a tool call results in an error (e.g., PythonInterpreterTool syntax error, file not found by a tool):
+        a.  **Analyze the error message.**
+        b.  **Do not immediately retry the exact same tool call with the exact same arguments.**
+        c.  Instead, try to **correct the input to the tool** (e.g., fix the Python code, verify a URL or file path).
+        d.  If correction isn't obvious or fails, consider if a different tool or a different approach is more suitable.
+5.  **Formulate Your Response**:
     *   Provide only the final, concise answer to the question.
     *   Do not include your reasoning steps, apologies, self-correction narratives, or any conversational filler in the final answer.
-    *   If, after a thorough investigation with all relevant tools, you cannot determine a definitive answer, respond with the exact phrase 'I don't know'.
+    *   If, after a thorough investigation and attempts to use tools effectively (including handling errors), you cannot determine a definitive answer, respond with the exact phrase 'I don't know'.
 
 --- Start of Question & File Information ---
 {question_to_llm}