mejoras routing y prompts

agents.py

@@ -36,36 +36,23 @@ class AlfredAdvancedWorkflow(Workflow):
         # Agente de busqueda
         self.web_agent = AgentWorkflow.from_tools_or_functions([search_tool],
                                                      llm = llm, 
-                                                     system_prompt="""You are a concise web research assistant.
-
-CRITICAL RULES:
-1. Use web_search ONLY ONCE
-2. After searching, provide a SHORT answer (1-3 sentences maximum).
-3. Provide a clear, direct and short answer. Dont give extra explanations or introductions
-4. Answer ONLY what was asked, nothing more
-5. DO NOT search multiple times
-6. If you can't find the answer, say "I don't know" briefly
-
-Format: Direct answer in 1-3 sentences.""")
+                                                     system_prompt="""You are a FACTUAL web search assistant. Rules:
+1. Use web_search tool ONCE
+2. Answer in 1-2 sentences MAXIMUM
+3. Be direct and factual
+4. If unsure, say "I don't know"
+5. NO introductions, NO explanations""")
         
         self.doc_agent = AgentWorkflow.from_tools_or_functions([read_document_tool, 
                                                                 image_analyzer_tool, youtube_transcript_tol, calculator_tool],
                                                                 llm = llm, 
                                                              
-                                                  system_prompt = """Document processing assistant. Answer in 1 sentences max.
-
-AVAILABLE TOOLS:
-- read_document: Read files (PDF, TXT, CSV, etc.)
-- analyze_image: Analyze images
-- calculator: Do calculations
-- youtube_transcript: Get what is SAID in YouTube videos
-
-RULES:
-- Use ONE tool ONCE
-- For YouTube videos: Use youtube_transcript tool
-- Provide SHORT answer (just the fact requested)
-
-Answer format: Direct answer, 1 sentences.""")
+                                                  system_prompt = """You are a CONCISE document processor. Rules:
+1. Use ONE appropriate tool ONCE
+2. Answer in 1 sentence MAXIMUM
+3. For YouTube: use youtube_transcript
+4. For calculations: use calculator
+5. Be direct - just the answer""")
     
         #self.reviewer = AgentWorkflow.from_tools_or_functions([], llm = llm,
         #                                                      system_prompt=" You are an expert reviewer. Your task is to review the provided answer to ensure its accuracy, completeness, and relevance to the question. Be concise as much as possible")

tools.py

@@ -20,19 +20,39 @@ client = InferenceClient(HF_TOKEN)
 
 def get_youtube_transcript(video_url: str) -> str:
     try:
-        video_id_match = re.search(r"(?:v=|youtu\.be/)([a-zA-Z0-9_-]{11})", video_url)
-        if not video_id_match:
-            return "Error: Invalid YouTube URL format."
-        video_id = video_id_match.group(1)
+        # Extraer ID del video más robustamente
+        video_id = None
+        
+        # Patrón para youtube.com/watch?v=ID
+        match = re.search(r'(?:v=|\/)([0-9A-Za-z_-]{11}).*', video_url)
+        if match:
+            video_id = match.group(1)
+        # Patrón para youtu.be/ID
+        else:
+            match = re.search(r'youtu\.be\/([0-9A-Za-z_-]{11})', video_url)
+            if match:
+                video_id = match.group(1)
+        
+        if not video_id:
+            return "Error: Could not extract YouTube video ID from URL"
+        
+        print(f"Getting transcript for YouTube video: {video_id}")
+        
         transcript_list = YouTubeTranscriptApi.get_transcript(video_id)
-        #Combine all the text segments
         full_transcript = " ".join([entry['text'] for entry in transcript_list])
-
-        return f"The YouTube video Transcript es:\n{full_transcript}"
+        
+        # Hacer el transcript más manejable
+        if len(full_transcript) > 1000:
+            full_transcript = full_transcript[:1000] + "... [truncated]"
+            
+        return full_transcript
+        
     except Exception as e:
         return f"Error retrieving transcript: {str(e)}"
 
 
+
+
 def read_document(file_path: str) -> str:
     try:
         if not os.path.exists(file_path):