agents/reviewer_agent.py

import subprocess
import tempfile
import os
import openai
from typing import Dict, Any
from dotenv import load_dotenv

load_dotenv()

class ReviewerAgent:
    """
    Agent responsible for reviewing code for quality, style, and potential issues.
    Uses both static analysis (pylint) and LLM-based review.
    """
    
    def __init__(self):
        self.api_key = os.getenv("OPENAI_API_KEY")
        openai.api_key = self.api_key
    
    def static_analysis(self, code: str) -> Dict[str, Any]:
        """
        Perform static code analysis using pylint.
        
        Args:
            code: Python code to analyze
            
        Returns:
            Dictionary with pylint results
        """
        try:
            # Create a temporary file with the code
            with tempfile.NamedTemporaryFile(mode='w', suffix='.py', delete=False) as f:
                f.write(code)
                temp_file_path = f.name
            
            # Run pylint on the temporary file
            result = subprocess.run(
                ['pylint', temp_file_path, '--output-format=json'],
                capture_output=True,
                text=True
            )
            
            # Clean up temporary file
            os.unlink(temp_file_path)
            
            # Parse pylint output
            if result.returncode == 0:
                # Parse JSON output
                import json
                try:
                    issues = json.loads(result.stdout)
                    return {
                        "status": "success",
                        "issues": issues,
                        "score": self._calculate_pylint_score(issues),
                        "summary": f"Found {len(issues)} issues"
                    }
                except:
                    return {
                        "status": "success",
                        "issues": [],
                        "score": 10.0,
                        "summary": "No issues found"
                    }
            else:
                return {
                    "status": "error",
                    "error": result.stderr,
                    "issues": []
                }
                
        except Exception as e:
            return {
                "status": "error",
                "error": str(e),
                "issues": []
            }
    
    def _calculate_pylint_score(self, issues: list) -> float:
        """Calculate a normalized score from pylint issues."""
        if not issues:
            return 10.0
        
        # Count issues by type
        error_count = sum(1 for issue in issues if issue.get('type') == 'error')
        warning_count = sum(1 for issue in issues if issue.get('type') == 'warning')
        
        # Simple scoring: start from 10 and deduct points
        score = 10.0
        score -= error_count * 0.5
        score -= warning_count * 0.1
        
        return max(0, min(10, score))
    
    def llm_review(self, code: str) -> Dict[str, Any]:
        """
        Use LLM to review code for logical errors, improvements, and best practices.
        
        Args:
            code: Python code to review
            
        Returns:
            Dictionary with LLM review results
        """
        try:
            system_message = """You are an expert code reviewer. Analyze the code for:
            1. Logical errors
            2. Security issues
            3. Performance improvements
            4. Code style and best practices
            5. Edge cases not handled
            
            Provide specific, actionable feedback."""
            
            response = openai.ChatCompletion.create(
                model="gpt-3.5-turbo",
                messages=[
                    {"role": "system", "content": system_message},
                    {"role": "user", "content": f"Review this code:\n\n{code}"}
                ],
                temperature=0.3,
                max_tokens=300
            )
            
            review_text = response.choices[0].message.content
            
            # Extract key points
            import re
            suggestions = re.findall(r'[-•]\s*(.*?)(?=\n\n|\Z)', review_text, re.DOTALL)
            
            return {
                "status": "success",
                "review": review_text,
                "suggestions": suggestions,
                "tokens_used": response.usage.total_tokens
            }
            
        except Exception as e:
            return {
                "status": "error",
                "error": str(e),
                "review": ""
            }
    
    def comprehensive_review(self, code: str) -> Dict[str, Any]:
        """
        Combine static analysis and LLM review for comprehensive feedback.
        
        Args:
            code: Python code to review
            
        Returns:
            Complete review results
        """
        static_result = self.static_analysis(code)
        llm_result = self.llm_review(code)
        
        return {
            "static_analysis": static_result,
            "llm_review": llm_result,
            "overall_score": self._calculate_overall_score(static_result, llm_result)
        }
    
    def _calculate_overall_score(self, static: Dict, llm: Dict) -> float:
        """Calculate an overall code quality score."""
        if static.get("status") != "success":
            return 0.0
        
        static_score = static.get("score", 0.0)
        
        # LLM review doesn't give numeric score, so we estimate based on suggestions
        llm_suggestions = len(llm.get("suggestions", []))
        llm_score = max(0, 10 - llm_suggestions * 0.5)
        
        # Weighted average: 70% static analysis, 30% LLM review
        return static_score * 0.7 + llm_score * 0.3