Tools_corrected

app.py

@@ -261,19 +261,30 @@ def respond_to_input(user_input: str, task_id) -> str:
     """
     system_msg = SystemMessage(
     content=(
-        "You are an agent that picks exactly one course of action:\n"
-        "Use Wikipedia first before even using web search Please. "
-        "  1) If the user's question is best answered by looking up a Wikipedia summary, return\n"
-        "     {\"wiki_query\":\"<query for Wikipedia>\"} and nothing else.\n"
-        "  2) Otherwise, if Wikipedia won't directly help, but you still need web results, return\n"
-        "     {\"web_search_query\":\"<search terms>\"} and nothing else.\n"
-        "  3) Otherwise, if it's an image, use OCR: return {\"ocr_path\":\"<path>\"}.\n"
-        "  4) Otherwise, if it's a spreadsheet question, return {\"excel_path\":\"<path>\", \"excel_sheet_name\":\"<sheet>\"}.\n"
-        "  5) Otherwise, if it's an audio transcription, return {\"audio_path\":\"<path to audio file>\"}.\n"
-        "  6) If you can answer immediately without any tool, return exactly {\"final_answer\":\"<your answer>\"}.\n"
-        "Do NOT include any other keys, text, or explanation—just one JSON object with exactly one of those keys."
+        "You are an agent that must choose exactly one of the following actions:\n"
+        "  1) If the user's question can be answered directly by consulting Wikipedia, return exactly:\n"
+        "       {\"wiki_query\":\"<search terms for Wikipedia>\"}\n"
+        "     and nothing else. Use Wikipedia before any other tool.\n"
+        "  2) Only if Wikipedia cannot directly answer, perform a web search and return:\n"
+        "       {\"web_search_query\":\"<search terms>\"}\n"
+        "     and nothing else.\n"
+        "  3) If the user's question requires extracting text from an image, return:\n"
+        "       {\"ocr_path\":\"<local image path>\"}\n"
+        "     and nothing else.\n"
+        "  4) If the user's question requires reading a spreadsheet, return:\n"
+        "       {\"excel_path\":\"<local .xlsx path>\", \"excel_sheet_name\":\"<sheet name>\"}\n"
+        "     and nothing else.\n"
+        "  5) If the user needs an audio transcription, return:\n"
+        "       {\"audio_path\":\"<local audio file path>\"}\n"
+        "     and nothing else.\n"
+        "  6) If you already know the answer without using any tool, return exactly:\n"
+        "       {\"final_answer\":\"<your concise answer>\"}\n"
+        "     and nothing else.\n"
+        "If the user's prompt explicitly tells you to perform a specific action (for example, “translate this sentence”), then do it directly and return your result as {\"final_answer\":\"<your answer>\"} or the appropriate tool key if needed.  \n"
+        "Do NOT include any additional keys, explanation, or markdown—only one JSON object with exactly one key."
     )
 )
+
     human_msg = HumanMessage(content=user_input)
 
     initial_state: AgentState = {"messages": [system_msg, human_msg], "task_id": task_id}

tools.py

@@ -262,11 +262,13 @@ def audio_transcriber_tool(state: AgentState) -> AgentState:
             raise RuntimeError("OPENAI_API_KEY is not set in environment.")
 
         with open(local_audio, "rb") as audio_file:
+            print("reached openai.audio.transcriptions.create")
             response = openai.audio.transcriptions.create(
                 model="whisper-1", 
                 file=audio_file,
             )
-        text = response.get("text", "").strip()
+            print("reached response")
+        text = response.text.strip()
     except Exception as e:
         text = f"Error during transcription: {e}"
     print(f"Transcripted as transcript: {text}")