agents/agent_5_ai_analyzer.py

"""
Agent 5: AI Analyzer
Uses HuggingFace models for intelligent analysis of UI differences.
Synthesizes findings from visual comparison and element matching.
"""

from typing import Dict, Any, List
import sys
import os

sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))

from utils.hf_models import analyze_with_ai, HFModelClient


def agent_5_node(state: Dict[str, Any]) -> Dict[str, Any]:
    """
    Agent 5: AI-powered analysis and synthesis.
    
    Uses multiple HF models to:
    1. Analyze visual differences in screenshots
    2. Generate insights from element comparisons
    3. Synthesize findings into actionable report
    """
    print("\n" + "="*60)
    print("🤖 Agent 5: AI Analyzer - Intelligent Analysis")
    print("="*60)
    
    hf_token = state.get("hf_token", "")
    logs = state.get("logs", [])
    
    # Get data from previous agents
    figma_screenshots = state.get("figma_screenshots", {})
    website_screenshots = state.get("website_screenshots", {})
    comparison_images = state.get("comparison_images", {})
    
    element_matches = state.get("element_matches", {})
    element_differences = state.get("element_differences", [])
    
    similarity_scores = state.get("similarity_scores", {})
    overall_score = state.get("overall_score", 0)
    
    # Build elements summary
    elements_summary = {
        "figma": {},
        "website": {},
        "matched": 0
    }
    
    for viewport, data in element_matches.items():
        if isinstance(data, dict) and "statistics" in data:
            stats = data["statistics"]
            elements_summary["figma"][viewport] = stats.get("total_figma", 0)
            elements_summary["website"][viewport] = stats.get("total_website", 0)
            elements_summary["matched"] += stats.get("matched", 0)
    
    # Check if AI is available
    if not hf_token:
        print("  ℹ️  No HF token provided - using rule-based synthesis")
        logs.append("ℹ️ AI analysis skipped (no HF token)")
        
        # Fallback synthesis
        result = _fallback_analysis(element_differences, similarity_scores, overall_score)
        
        return {
            "ai_analysis": result,
            "status": "analysis_complete",
            "logs": logs
        }
    
    print("  🔑 HF token provided - running AI analysis")
    logs.append("🤖 Running AI-powered analysis...")
    
    try:
        # Get images for analysis (prefer desktop)
        figma_img = figma_screenshots.get("desktop", "")
        website_img = website_screenshots.get("desktop", "")
        diff_img = comparison_images.get("desktop", "")
        
        scores = {
            "overall": overall_score,
            "desktop": similarity_scores.get("desktop", 0),
            "mobile": similarity_scores.get("mobile", 0)
        }
        
        # Run AI analysis
        result = analyze_with_ai(
            hf_token=hf_token,
            figma_image=figma_img,
            website_image=website_img,
            diff_image=diff_img,
            elements_summary=elements_summary,
            differences=element_differences,
            scores=scores
        )
        
        if result.get("ai_enhanced"):
            print("  ✅ AI analysis complete")
            logs.append("✅ AI analysis successful")
        else:
            print("  ⚠️ AI analysis fell back to rule-based")
            logs.append("⚠️ AI unavailable, used rule-based analysis")
        
        # Add priority issues to logs
        for i, issue in enumerate(result.get("priority_issues", [])[:3], 1):
            print(f"     {i}. {issue[:60]}...")
        
        return {
            "ai_analysis": result,
            "status": "analysis_complete",
            "logs": logs
        }
        
    except Exception as e:
        print(f"  ❌ AI analysis failed: {str(e)}")
        logs.append(f"❌ AI error: {str(e)}")
        
        # Fallback
        result = _fallback_analysis(element_differences, similarity_scores, overall_score)
        
        return {
            "ai_analysis": result,
            "status": "analysis_complete",
            "logs": logs
        }


def _fallback_analysis(
    differences: List[Dict],
    scores: Dict[str, float],
    overall_score: float
) -> Dict[str, Any]:
    """Generate analysis without AI."""
    
    # Count by severity
    high = len([d for d in differences if d.get("severity") == "high"])
    medium = len([d for d in differences if d.get("severity") == "medium"])
    low = len([d for d in differences if d.get("severity") == "low"])
    
    # Count by category
    categories = {}
    for d in differences:
        cat = d.get("category", "unknown")
        categories[cat] = categories.get(cat, 0) + 1
    
    # Generate summary
    if overall_score >= 90:
        summary = f"Excellent match! The website closely follows the Figma design with {overall_score:.0f}% similarity."
    elif overall_score >= 75:
        summary = f"Good match with {overall_score:.0f}% similarity. {high} high-priority issues need attention."
    elif overall_score >= 50:
        summary = f"Moderate match at {overall_score:.0f}% similarity. {high + medium} issues detected that require fixes."
    else:
        summary = f"Significant differences found ({overall_score:.0f}% similarity). Major review needed."
    
    # Priority issues from high severity
    priority_issues = []
    
    # Missing elements are highest priority
    missing = [d for d in differences if d.get("category") == "missing"]
    for m in missing[:2]:
        priority_issues.append(f"Missing element: {m.get('element_name', 'Unknown')} not found in website")
    
    # Then high severity issues
    high_issues = [d for d in differences if d.get("severity") == "high" and d.get("category") != "missing"]
    for h in high_issues[:3]:
        priority_issues.append(f"{h.get('category', '').title()}: {h.get('description', 'Check this element')}")
    
    # Fill remaining with medium issues
    if len(priority_issues) < 5:
        medium_issues = [d for d in differences if d.get("severity") == "medium"]
        for m in medium_issues[:5 - len(priority_issues)]:
            priority_issues.append(f"{m.get('category', '').title()}: {m.get('description', '')}")
    
    if not priority_issues:
        priority_issues = ["No critical issues found. Review minor differences for polish."]
    
    return {
        "summary": summary,
        "priority_issues": priority_issues,
        "ai_enhanced": False,
        "statistics": {
            "high_severity": high,
            "medium_severity": medium,
            "low_severity": low,
            "by_category": categories
        },
        "scores": {
            "overall": overall_score,
            **scores
        }
    }