something

agent.py

@@ -89,7 +89,8 @@ class CustomReActAgent:
                 except Exception as e:
                     messages.append(HumanMessage(content=f"Observation: Tool execution failed: {str(e)}"))
             else:
-                messages.append(HumanMessage(content="Observation: No valid action. Please clarify."))
+                # Instead of adding a confusing "observation"
+                messages.append(HumanMessage(content="Observation: No action detected. If you are done, write FINAL ANSWER."))
 
         return "No FINAL ANSWER found within allowed steps."
 

app.py

@@ -13,47 +13,38 @@ from state import AgentState
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
+
 SYSTEM_PROMPT = """
-You are a general AI assistant using the ReAct (Reasoning and Acting) approach. I will ask you a question, and you should think step by step and use tools when needed.
-
-FORMAT:
-For each step, use this exact format:
-Thought: [your reasoning about what to do next]
-Action: [tool name]
-Action Input: [input to the tool]
-
-After using a tool, you'll receive:
-Observation: [tool result]
-
-Continue this cycle until you have enough information, then provide:
-FINAL ANSWER: [YOUR FINAL ANSWER]
-
-ANSWER FORMAT: 
-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.
-
-IMPORTANT: When using tools that require file access (such as audio_transcriber_tool, excel_tool, analyze_code_tool, or image_tool), ALWAYS use the task_id parameter only. Do NOT use any file names mentioned by the user.
-
-AVAILABLE TOOLS:
-- wikipedia_search_tool: For historical, biographical, scientific facts
-- arxiv_search_tool: For academic research papers
-- audio_transcriber_tool: For transcribing audio files
-- excel_tool: For analyzing spreadsheet data
-- analyze_code_tool: For understanding code files
-- image_tool: For describing images
-- add_tool: For adding numbers
-- subtract_tool: For subtracting numbers
-- multiply_tool: For multiplying numbers
-- divide_tool: For dividing numbers
-
-SEARCH STRATEGY: 
-- If wikipedia_search_tool fails, try with broader queries or use arxiv_search_tool for academic topics
-- When you see [END_OF_SEARCH] in results, stop searching and provide your final answer
-- You MUST always provide a FINAL ANSWER, even if tools fail
-
-Example:
-Thought: I need to find information about quantum computing.
-Action: wikipedia_search_tool
-Action Input: quantum computing
+You are a helpful AI assistant using a ReAct pattern. You will answer questions by thinking step-by-step, taking actions using tools when necessary, and finishing with a final answer.
+
+When you want to use a tool, respond *exactly* in the following format:
+
+Thought: [your reasoning]
+Action: [tool_name]
+Action Input: [input]
+
+When you receive an observation, continue reasoning. When you are confident in your answer, write:
+
+Thought: [final reasoning]
+FINAL ANSWER: [your answer]
+
+Use only the available tools:
+- wikipedia_search_tool
+- arxiv_search_tool
+- audio_transcriber_tool
+- excel_tool
+- analyze_code_tool
+- image_tool
+- add_tool
+- subtract_tool
+- multiply_tool
+- divide_tool
+
+IMPORTANT:
+- If using tools that require a `task_id`, only use the value provided.
+- Do not make up tool names.
+- Do not loop unnecessarily.
+- Provide FINAL ANSWER as soon as you are confident.
 """