fix: aggiunge intent detection per fallback automatico

- Rileva automaticamente richieste di system info e list file
- Non dipende dal modello per usare i tool base
- System prompt semplificato

ollama_agent.py

@@ -232,29 +232,14 @@ class OllamaToolAgent:
         
     def _default_system_prompt(self) -> str:
         """System prompt di default"""
-        return """You are LangCode AI. You have access to these tools:
+        return """You are a helpful AI assistant. When users ask about:
+- system info, OS, operating system, what computer: respond with TOOL:system_info
+- list files, directory contents, ls: respond with TOOL:ls|path=.
 
-1. system_info - Get system info (OS, Python version, etc). No arguments needed.
-2. ls(path=".") - List directory contents. Use path="." for current directory.
-3. execute(command) - Run shell commands. Example: execute("ls -la")
-4. read_file(path) - Read file contents. Example: read_file("/path/to/file")
-5. write_file(path, content) - Write to file.
-6. search(directory, pattern) - Search for files. Example: search(".", "*.py")
+Example: User asks "what OS am I using?" -> You respond: TOOL:system_info
+Example: User asks "list files" -> You respond: TOOL:ls|path=.
 
-IMPORTANT RULES:
-- ONLY use these exact tool names: system_info, ls, execute, read_file, write_file, search
-- NEVER invent tool names
-- When you need a tool, output ONLY this format (no other text):
-
-TOOL:system_info
-TOOL:ls|path=.
-TOOL:execute|command=ls -la
-TOOL:read_file|path=/path/to/file
-TOOL:write_file|path=/file.txt|content=hello
-TOOL:search|directory=.|pattern=*.py
-
-- After the tool result, you can respond normally.
-- If no tool is needed, just respond normally."""
+For other questions, just answer normally without tools."""
     
     def _tool_instructions(self) -> str:
         """Genera le istruzioni per i tool disponibili"""
@@ -374,6 +359,29 @@ TOOL:search|directory=.|pattern=*.py
         """Formatta il risultato di un tool per il modello"""
         return f"\n[TOOL_RESULT]\nTool: {tool_call.name}\nResult: {json.dumps(result, ensure_ascii=False)}\n[/TOOL_RESULT]\n"
     
+    def _detect_intent(self, text: str) -> Optional[ToolCall]:
+        """Detect common intents directly without relying on model"""
+        text_lower = text.lower()
+        
+        # System info patterns
+        system_patterns = [
+            'sistema operativo', 'so sto', 'che os', 'what os', 
+            'che sistema', 'which system', 'operating system',
+            'che versione', 'python version', 'versione python'
+        ]
+        if any(p in text_lower for p in system_patterns):
+            return ToolCall(id='intent', name='system_info', arguments={})
+        
+        # List files patterns
+        list_patterns = [
+            'lista', 'listare', 'listami', 'elenco', 'mostrami i file',
+            'list files', 'list directory', 'ls', 'file nella cartella'
+        ]
+        if any(p in text_lower for p in list_patterns):
+            return ToolCall(id='intent', name='ls', arguments={'path': '.'})
+        
+        return None
+    
     def invoke(self, input_dict: Dict) -> Dict:
         """
         Invoca l'agente con un input.
@@ -390,6 +398,13 @@ TOOL:search|directory=.|pattern=*.py
         # Aggiungi il messaggio utente alla cronologia
         self.messages.append(HumanMessage(content=user_content))
         
+        # Try to detect intent directly first
+        detected_tool = self._detect_intent(user_content)
+        if detected_tool:
+            result = self._execute_tool(detected_tool)
+            tool_result_text = self._format_tool_result(detected_tool, result)
+            self.messages.append(HumanMessage(content=tool_result_text))
+        
         # Loop principale: risposta -> parse tools -> esegui -> ripeti
         iteration = 0
         final_response = None