Update app.py

app.py

@@ -1,105 +1,127 @@
 import os
+import re
 import gradio as gr
 import requests
-import inspect 
 import pandas as pd
+from bs4 import BeautifulSoup
 from smolagents import CodeAgent, DuckDuckGoSearchTool, InferenceClientModel, tool
 
 # --- Constants ---
 DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 
 # ==========================================
-# 🚀 NUOVO TOOL: LETTORE DI PAGINE WEB
+# 🚀 TOOL 1: RICERCA E LETTURA WEB AVANZATA
 # ==========================================
 @tool
 def visit_webpage(url: str) -> str:
-    """Visits a webpage and returns its text content. Use this to read articles or Wikipedia pages.
+    """Visits a webpage and extracts its main clean text. Use this to read Wikipedia, news, or articles.
     Args:
         url: The URL of the webpage to visit.
     """
     try:
-        import requests
-        from bs4 import BeautifulSoup
-        headers = {'User-Agent': 'Mozilla/5.0'}
-        response = requests.get(url, headers=headers, timeout=10)
+        headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36'}
+        response = requests.get(url, headers=headers, timeout=15)
         response.raise_for_status()
+        
         soup = BeautifulSoup(response.text, 'html.parser')
-        # Rimuove script e stili per pulire il testo
-        for script in soup(["script", "style"]):
-            script.extract()
+        # Rimuove tutto ciò che non è testo utile
+        for element in soup(["script", "style", "nav", "footer", "header", "aside"]):
+            element.extract()
+            
         text = soup.get_text(separator='\n', strip=True)
-        # Restituisce i primi 10000 caratteri per non intasare la memoria dell'agente
-        return text[:10000] 
+        # Prende i primi 15000 caratteri di puro testo informativo
+        return text[:15000] 
     except Exception as e:
         return f"Error reading the webpage: {str(e)}"
 
-# --- Basic Agent Definition ---
-class BasicAgent:
+# ==========================================
+# 🧠 IL SUPER AGENTE
+# ==========================================
+class SuperAgent:
     def __init__(self):
-        print("Inizializzazione dell'Agente AI POTENZIATO...")
+        print("Inizializzazione del SUPER Agente AI in corso...")
         
-        # 1. Definisci il modello
+        # 1. Modello
         self.model = InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct")
         
-        # 2. Strumenti: Ricerca + Lettura Web
+        # 2. Tools
         self.tools = [DuckDuckGoSearchTool(), visit_webpage]
         
-        # 3. Crea l'agente (aumentati i max_steps a 15 per farlo pensare di più)
+        # 3. Agente (Potenza massima, importazioni analitiche sbloccate)
         self.agent = CodeAgent(
             tools=self.tools,
             model=self.model,
-            max_steps=15,
-            additional_authorized_imports=["requests", "bs4", "json", "time", "math", "datetime"]
+            max_steps=20,  # Aumentati i tentativi a 20!
+            additional_authorized_imports=[
+                "requests", "bs4", "json", "time", "math", "datetime", 
+                "pandas", "numpy", "re", "csv", "urllib"
+            ]
         )
         
-        # 4. PROMPT CORAZZATO PER EXACT MATCH
+        # 4. Prompt Estremo per GAIA
         self.prompt_template = """
-        You are an expert AI solving the GAIA benchmark. You must answer the following question.
+        You are an elite AI data analyst solving the GAIA benchmark. 
+        You are provided with a question. You MUST use your tools to find the answer.
         
-        CRITICAL RULES:
-        1. If you need information, use DuckDuckGoSearchTool to find URLs, then use the visit_webpage tool to read the content of those URLs.
-        2. EXACT MATCH ONLY: You must output ONLY the exact requested answer. 
-        3. Absolutely NO introductory phrases, no "The answer is...", no "FINAL ANSWER:". 
-        4. If the question asks for a number, return JUST the number. If it asks for a name, return JUST the name.
+        CRITICAL RULES FOR YOUR FINAL OUTPUT:
+        1. EXACT MATCH ONLY: Output ONLY the final answer. Nothing else.
+        2. If the answer is a number, return JUST the number (e.g., '14' or '1998').
+        3. If the answer is a name/word, return JUST the word.
+        4. NEVER use phrases like "The answer is", "Based on my search", or "FINAL ANSWER:".
+        5. If the question requires math, date calculation, or text processing, write the Python code to solve it internally.
         
-        Question to solve: {question}
+        Question: {question}
         """
 
     def __call__(self, question: str) -> str:
-        print(f"Agent received question: {question[:50]}...")
+        print(f"\n[DOMANDA RICEVUTA]: {question[:80]}...")
         try:
             formatted_prompt = self.prompt_template.format(question=question)
-            answer = self.agent.run(formatted_prompt)
-            final_answer = str(answer).strip() 
+            raw_answer = self.agent.run(formatted_prompt)
+            final_answer = str(raw_answer).strip()
             
-            # Un'ultima pulizia di sicurezza se l'LLM fa di testa sua
-            if "FINAL ANSWER:" in final_answer:
-                final_answer = final_answer.split("FINAL ANSWER:")[-1].strip()
+            # --- FILTRO ANTI-BLABLA ---
+            # Se l'LLM si ostina a inserire testo, lo forziamo a tacere tagliando le frasi comuni.
+            prefixes_to_cut = ["The answer is", "FINAL ANSWER:", "Answer:", "final answer is", "The requested word is", "The highest number is"]
+            for prefix in prefixes_to_cut:
+                if prefix.lower() in final_answer.lower():
+                    idx = final_answer.lower().rfind(prefix.lower()) + len(prefix)
+                    final_answer = final_answer[idx:].strip()
+                    
+            # Toglie il punto finale se l'ha messo per sbaglio (es. "1994." -> "1994")
+            if final_answer.endswith('.'):
+                final_answer = final_answer[:-1]
                 
-            print(f"L'agente ha trovato la risposta: {final_answer}")
+            # Toglie virgolette extra o asterischi di formattazione Markdown
+            final_answer = final_answer.replace("**", "").replace('"', "").replace("'", "")
+            
+            print(f"[RISPOSTA PULITA TROVATA]: {final_answer}")
             return final_answer
         except Exception as e:
-            print(f"Errore: {e}")
+            print(f"Errore durante l'elaborazione: {e}")
             return "Error"
 
 
-def run_and_submit_all( profile: gr.OAuthProfile | None):
+# ==========================================
+# ⚙️ INTERFACCIA E RUNNER
+# ==========================================
+def run_and_submit_all(profile: gr.OAuthProfile | None):
     space_id = os.getenv("SPACE_ID") 
 
     if profile:
-        username= f"{profile.username}"
-        print(f"User logged in: {username}")
+        username = f"{profile.username}"
+        print(f"Utente autenticato: {username}")
     else:
-        return "Please Login to Hugging Face with the button.", None
+        return "Per favore, fai il Login con Hugging Face usando l'apposito tasto.", None
 
     api_url = DEFAULT_API_URL
     questions_url = f"{api_url}/questions"
     submit_url = f"{api_url}/submit"
 
     try:
-        agent = BasicAgent()
+        agent = SuperAgent()
     except Exception as e:
-        return f"Error initializing agent: {e}", None
+        return f"Errore nell'inizializzazione dell'agente: {e}", None
         
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
 
@@ -108,13 +130,14 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
         response.raise_for_status()
         questions_data = response.json()
         if not questions_data:
-             return "Fetched questions list is empty or invalid format.", None
+             return "La lista delle domande scaricata è vuota.", None
     except Exception as e:
-        return f"Error fetching questions: {e}", None
+        return f"Errore nel download delle domande: {e}", None
 
     results_log = []
     answers_payload = []
     
+    print(f"Avvio elaborazione su {len(questions_data)} domande...")
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
@@ -128,7 +151,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
              results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
 
     if not answers_payload:
-        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
+        return "L'agente non ha prodotto risposte da inviare.", pd.DataFrame(results_log)
 
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
     
@@ -137,24 +160,24 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
         response.raise_for_status()
         result_data = response.json()
         final_status = (
-            f"Submission Successful!\n"
-            f"User: {result_data.get('username')}\n"
-            f"Overall Score: {result_data.get('score', 'N/A')}% "
-            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} correct)\n"
-            f"Message: {result_data.get('message', 'No message received.')}"
+            f"✅ Invio Completato con Successo!\n"
+            f"👤 Utente: {result_data.get('username')}\n"
+            f"🏆 Punteggio: {result_data.get('score', 'N/A')}% "
+            f"({result_data.get('correct_count', '?')}/{result_data.get('total_attempted', '?')} corrette)\n"
+            f"📝 Messaggio: {result_data.get('message', 'Nessun messaggio ricevuto.')}"
         )
         return final_status, pd.DataFrame(results_log)
     except Exception as e:
-        status_message = f"Submission Failed: {e}"
+        status_message = f"❌ Invio Fallito: {e}"
         return status_message, pd.DataFrame(results_log)
 
-# --- Build Gradio Interface using Blocks ---
+# --- Build Gradio Interface ---
 with gr.Blocks() as demo:
-    gr.Markdown("# Basic Agent Evaluation Runner")
+    gr.Markdown("# 🚀 Super Agente - Final Assignment Runner")
     gr.LoginButton()
-    run_button = gr.Button("Run Evaluation & Submit All Answers")
-    status_output = gr.Textbox(label="Run Status / Submission Result", lines=5, interactive=False)
-    results_table = gr.DataFrame(label="Questions and Agent Answers", wrap=True)
+    run_button = gr.Button("Avvia Valutazione & Invia Risposte", variant="primary")
+    status_output = gr.Textbox(label="Stato Esecuzione / Risultato", lines=5, interactive=False)
+    results_table = gr.DataFrame(label="Domande e Risposte dell'Agente", wrap=True)
 
     run_button.click(
         fn=run_and_submit_all,