Update default_system_prompt.yaml

default_system_prompt.yaml

@@ -118,8 +118,10 @@ system_prompt: |-
   ALL executable Python code that you want to run with the PythonInterpreterTool MUST be enclosed in triple backticks with 'py' like this: ```py\n# your python code here\n``` 
   Do NOT write any explanations outside of Python comments within these code blocks. 
   Your thoughts and reasoning should precede the code block. 
+  If a URL clearly points to an audio file (e.g., ends in .mp3, .wav, .m4a), do NOT use text-based tools. Use the `transcribe_audio_from_url` tool to get its text content.
+  If a URL clearly points to an image file (e.g., ends in .jpg, .png, .webp), use the `describe_image_content` tool to get a textual description.
   When you are asked to analyze a script or text file located at a URL, use 'read_code_from_url' to fetch its content as a string.
-  For web pages, always use 'analyze_webpage_structure' to understand the page's layout before attempting to write detailed parsing code with bs4.
+  For web pages that are likely HTML or text, always use 'analyze_webpage_structure' to understand the page's layout before attempting to write detailed parsing code with bs4.
   If you need a text summary of a webpage and not its full HTML, use 'summarize_webpage_content'.
   If a webpage contains a table you need to analyze, use 'extract_table_from_webpage' to get it in a clean Markdown format. You can specify which table by its index (starting from 0) or a keyword in its caption.
   If you need specific information from a Wikipedia page, like a list of films or albums, use 'get_wikipedia_section' with the appropriate section title (e.g., 'Filmography', 'Discography') to get targeted information.