test_agent.py

#!/usr/bin/env python3
"""
🔍 GAIA Agent Diagnostic Test
Quick test to diagnose why the agent isn't answering questions
"""

import os
import sys
from gaia_system import BasicAgent, EnhancedMultiModelGAIASystem

def test_basic_agent():
    """Test the BasicAgent with simple questions"""
    print("🧪 Testing BasicAgent...")
    
    try:
        # Initialize agent
        agent = BasicAgent()
        print("✅ Agent initialized successfully")
        
        # Test simple questions
        test_questions = [
            "What is 2 + 2?",
            "What is the capital of France?",
            "How many days are in a week?",
            "What color is the sky?"
        ]
        
        for i, question in enumerate(test_questions, 1):
            print(f"\n📝 Test {i}: {question}")
            try:
                response = agent(question)
                print(f"🤖 Response: '{response}'")
                print(f"📏 Length: {len(response)} characters")
                
                if not response or response.strip() == "":
                    print("❌ Empty response!")
                elif "Unable to" in response or "Error" in response:
                    print("⚠️ Error response detected")
                else:
                    print("✅ Got non-empty response")
                    
            except Exception as e:
                print(f"❌ Error: {e}")
                
    except Exception as e:
        print(f"❌ Failed to initialize agent: {e}")
        return False
        
    return True

def test_enhanced_system():
    """Test the EnhancedMultiModelGAIASystem directly"""
    print("\n🧪 Testing EnhancedMultiModelGAIASystem...")
    
    try:
        # Test with HF token if available
        hf_token = os.getenv('HF_TOKEN')
        if hf_token:
            print(f"✅ Found HF_TOKEN: {hf_token[:10]}...")
        else:
            print("⚠️ No HF_TOKEN found - using fallback mode")
            
        system = EnhancedMultiModelGAIASystem(hf_token=hf_token)
        print("✅ Enhanced system initialized")
        
        # Test simple query
        question = "What is 5 + 3?"
        print(f"\n📝 Testing: {question}")
        
        response = system.query_with_tools(question)
        print(f"🤖 Raw response: '{response}'")
        
        # Test fallback
        fallback = system._fallback_response(question)
        print(f"🛡️ Fallback response: '{fallback}'")
        
        # Test answer extraction
        if response:
            extracted = system._extract_final_answer(response)
            print(f"✨ Extracted answer: '{extracted}'")
            
        return True
        
    except Exception as e:
        print(f"❌ Enhanced system error: {e}")
        import traceback
        traceback.print_exc()
        return False

def test_model_availability():
    """Test which AI models are available"""
    print("\n🔍 Testing model availability...")
    
    try:
        system = EnhancedMultiModelGAIASystem()
        
        print(f"📊 Available models: {len(system.clients)}")
        for name, client_info in system.clients.items():
            provider = client_info.get('provider', 'Unknown')
            priority = client_info.get('priority', 999)
            print(f"  - {name} (Priority: {priority}, Provider: {provider})")
            
        if system.model_priority:
            print(f"🎯 Top priority model: {system.model_priority[0]}")
        else:
            print("❌ No models in priority list!")
            
        return True
        
    except Exception as e:
        print(f"❌ Model availability error: {e}")
        return False

def main():
    """Run all diagnostic tests"""
    print("🚀 GAIA Agent Diagnostic Tests\n")
    
    # Test basic functionality
    test1 = test_basic_agent()
    test2 = test_enhanced_system() 
    test3 = test_model_availability()
    
    print("\n📊 Test Summary:")
    print(f"  BasicAgent: {'✅ PASS' if test1 else '❌ FAIL'}")
    print(f"  Enhanced System: {'✅ PASS' if test2 else '❌ FAIL'}")
    print(f"  Model Availability: {'✅ PASS' if test3 else '❌ FAIL'}")
    
    if not any([test1, test2, test3]):
        print("\n❌ All tests failed! Check dependencies and configuration.")
        return False
    elif not test1:
        print("\n⚠️ BasicAgent failed - this is the issue for GAIA submissions!")
        return False
    else:
        print("\n✅ Core functionality working - issue might be elsewhere")
        return True

if __name__ == "__main__":
    success = main()
    sys.exit(0 if success else 1)