Update app.py

app.py

@@ -577,39 +577,50 @@ Your goal: Provide the EXACT answer in the EXACT format requested.
 
 1. **ANALYZE QUESTION:**
    - What information is needed?
-   - What format should the answer be? (number, list, yes/no, name, etc.)
-   - Are there any files attached?
+   - What format should the answer be?
+   - Are there any files?
 
 2. **FIRST TURN - MAKE A PLAN:**
-   Your FIRST response MUST be a brief plan (2-3 sentences):
-   - What tools you'll use
-   - What order you'll use them
-   - What format the final answer should be
-   DO NOT call tools on your first turn!
+   - Your FIRST response MUST be a brief plan (2-3 sentences).
+   - DO NOT call tools on your first turn! Just state the plan.
 
 3. **EXECUTE:**
-   - Call ONE tool per turn
-   - Wait for the result before planning your next step
+   - Call ONE tool per turn.
+   - Wait for the result before planning your next step.
    - For ANY calculation or logic: use code_interpreter with print()
 
 4. **VERIFY RESULTS:**
-   - Check if tool output contains errors
-   - If error: plan a different approach
-   - If success: decide if you need more info or have the answer
+   - Check if tool output contains errors.
+   - If error: plan a different approach.
+   - If success: decide if you need more info or have the answer.
 
 5. **FINISH:**
-   When you have the answer from a tool output:
-   - Call final_answer_tool immediately
+   - When you have the answer from a tool output:
+   - Call final_answer_tool immediately.
    - Provide ONLY the exact answer (no explanations!)
 
 **CRITICAL RULES:**
 
-❌ NEVER guess or use training data for the final answer
-❌ NEVER call multiple tools in one turn
-❌ NEVER add explanations to final_answer_tool
-✅ ALWAYS use code_interpreter for calculations/logic
-✅ ALWAYS match the requested answer format exactly
-✅ ALWAYS base your answer on tool outputs, not memory
+❌ NEVER guess or use training data.
+❌ NEVER call multiple tools in one turn.
+❌ NEVER add explanations to final_answer_tool.
+✅ ALWAYS use code_interpreter for calculations/logic.
+✅ ALWAYS match the requested answer format exactly.
+✅ ALWAYS base your answer on tool outputs.
+
+**TOOL CALL FORMATTING (CRITICAL!):**
+When you call a tool, you MUST use the exact tool name and provide arguments as valid JSON.
+
+**Example for final_answer_tool:**
+{{ "name": "final_answer_tool", "arguments": {{"answer": "The Final Answer"}} }}
+
+**Example for code_interpreter (MUST have 'code' key):**
+{{ "name": "code_interpreter", "arguments": {{"code": "print(1 + 1)"}} }}
+
+**Example for search_tool (MUST have 'query' key):**
+{{ "name": "search_tool", "arguments": {{"query": "latest news"}} }}
+
+Failure to provide arguments in this exact JSON format will cause an error.
 
 **ANSWER FORMAT EXAMPLES:**
 - "What is 5+5?" → final_answer("10")