Update

Dockerfile

@@ -2,47 +2,43 @@ FROM python:3.10-slim
 
 WORKDIR /app
 
-# Install system dependencies
 RUN apt-get update && apt-get install -y \
     git \
     curl \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy requirements first for better caching
 COPY requirements.txt .
 RUN pip install --no-cache-dir -r requirements.txt
 
-# Copy the rest of the application
 COPY . .
 
-# Create a wrapper script to handle API configuration
 RUN echo '#!/bin/bash\n\
-# Check if API configuration is provided\n\
 if [ -z "$API_BASE_URL" ]; then\n\
-    echo " WARNING: API_BASE_URL not set!"\n\
-    echo " Please set the following environment variables:"\n\
-    echo "   API_BASE_URL  - Your API endpoint"\n\
-    echo "   MODEL_NAME    - Model identifier"\n\
-    echo "   API_KEY       - Your API key"\n\
+    echo "ERROR: API_BASE_URL environment variable is required"\n\
+    echo "Please set:"\n\
+    echo "  API_BASE_URL  - Your API endpoint"\n\
+    echo "  MODEL_NAME    - Model identifier"\n\
+    echo "  API_KEY       - Your API key"\n\
     echo ""\n\
     echo "Examples:"\n\
-    echo "   OpenAI:    API_BASE_URL=https://api.openai.com/v1 MODEL_NAME=gpt-4"\n\
-    echo "   Gemini:    API_BASE_URL=https://generativelanguage.googleapis.com MODEL_NAME=gemini-1.5-pro"\n\
-    echo "   Local:     API_BASE_URL=http://localhost:11434/v1 MODEL_NAME=llama2"\n\
-    echo ""\n\
+    echo "  OpenAI:    API_BASE_URL=https://api.openai.com/v1 MODEL_NAME=gpt-4"\n\
+    echo "  Gemini:    API_BASE_URL=https://generativelanguage.googleapis.com MODEL_NAME=gemini-1.5-pro"\n\
+    echo "  Local:     API_BASE_URL=http://localhost:11434/v1 MODEL_NAME=llama2"\n\
+    exit 1\n\
+fi\n\
+\n\
+if [ -z "$MODEL_NAME" ]; then\n\
+    echo "ERROR: MODEL_NAME environment variable is required"\n\
     exit 1\n\
 fi\n\
 \n\
-# Show current configuration\n\
-echo "🔧 Running with configuration:"\n\
-echo "   API_BASE_URL: ${API_BASE_URL}"\n\
-echo "   MODEL_NAME: ${MODEL_NAME}"\n\
-echo "   TEMPERATURE: ${TEMPERATURE:-0.7}"\n\
-echo "   MAX_TOKENS: ${MAX_TOKENS:-2000}"\n\
+echo "Configuration:"\n\
+echo "  API_BASE_URL: ${API_BASE_URL}"\n\
+echo "  MODEL_NAME: ${MODEL_NAME}"\n\
+echo "  TEMPERATURE: ${TEMPERATURE:-0.7}"\n\
+echo "  MAX_TOKENS: ${MAX_TOKENS:-2000}"\n\
 echo ""\n\
 \n\
-# Run inference with provided configuration\n\
 python inference.py "$@"' > /usr/local/bin/run-agent && chmod +x /usr/local/bin/run-agent
 
-# Default command (will be overridden by docker run arguments)
 CMD ["run-agent", "--task-id", "bug_detection_easy"]

inference.py

@@ -60,4 +60,231 @@ class LLMClient:
             return result["choices"][0]["message"]["content"]
         except Exception as e:
             print(f"OpenAI API error: {e}")
-            raise
+            raise
+        
+    def _gemini_completion(self, messages: list, temperature: float, max_tokens: int) -> str:
+        gemini_prompt = self._convert_to_gemini_format(messages)
+        
+        url = f"{self.base_url}/v1/models/{self.model}:generateContent"
+        
+        headers = {
+            "Content-Type": "application/json",
+            "x-goog-api-key": self.api_key
+        }
+        
+        payload = {
+            "contents": [{
+                "parts": [{"text": gemini_prompt}]
+            }],
+            "generationConfig": {
+                "temperature": temperature,
+                "maxOutputTokens": max_tokens
+            }
+        }
+        
+        try:
+            response = requests.post(url, headers=headers, json=payload, timeout=30)
+            response.raise_for_status()
+            result = response.json()
+            return result["candidates"][0]["content"]["parts"][0]["text"]
+        except Exception as e:
+            print(f"Gemini API error: {e}")
+            raise
+    
+    def _convert_to_gemini_format(self, messages: list) -> str:
+        prompt_parts = []
+        for msg in messages:
+            role = msg["role"]
+            content = msg["content"]
+            
+            if role == "system":
+                prompt_parts.append(f"System: {content}")
+            elif role == "user":
+                prompt_parts.append(f"User: {content}")
+            elif role == "assistant":
+                prompt_parts.append(f"Assistant: {content}")
+        
+        return "\n\n".join(prompt_parts)
+
+
+class CodeReviewAgent:
+    def __init__(self):
+        self.client = LLMClient(API_BASE_URL, API_KEY, MODEL_NAME)
+        self.history = []
+        
+    def get_action(self, observation: Dict[str, Any]) -> str:
+        system_prompt = """You are an expert code reviewer. Your task is to review code changes and provide feedback.
+
+Review the code diff and identify issues. You can:
+1. ADD_COMMENT: Add a comment about an issue on a specific line
+2. SUGGEST_FIX: Suggest a specific code fix for an issue
+3. APPROVE: Approve the code changes (only if no critical issues)
+4. REQUEST_CHANGES: Request changes (if issues are found)
+
+Respond with a JSON object in this format:
+{
+    "action_type": "add_comment" | "suggest_fix" | "approve" | "request_changes",
+    "comments": [
+        {
+            "line_number": 10,
+            "content": "This line has a potential bug...",
+            "is_issue": true,
+            "severity": "high"
+        }
+    ],
+    "suggestions": [
+        {
+            "original_line": 10,
+            "suggested_code": "if x != 0:",
+            "explanation": "Prevents division by zero"
+        }
+    ],
+    "final_decision": "approved" | "changes_requested" (only if action_type is approve or request_changes)
+}
+
+Be thorough but concise. Focus on real issues like bugs, security vulnerabilities, performance problems, and code quality."""
+        
+        user_prompt = f"""
+Code Review Task:
+{observation.get('task_description', 'Review the following code changes')}
+
+Code Diff:
+{observation.get('code_diff', '')}
+
+File Context:
+{observation.get('file_context', '')}
+
+
+Current step: {observation.get('current_step', 0)}/{observation.get('max_steps', 50)}
+Previous actions taken: {len(observation.get('previous_comments', []))} comments, {len(observation.get('previous_suggestions', []))} suggestions
+
+Please provide your review action as JSON.
+"""
+        
+        messages = [
+            {"role": "system", "content": system_prompt},
+            {"role": "user", "content": user_prompt}
+        ]
+        
+        try:
+            response = self.client.chat_completion(messages, TEMPERATURE, MAX_TOKENS)
+            
+            response = response.strip()
+            if response.startswith("```json"):
+                response = response[7:]
+            if response.startswith("```"):
+                response = response[3:]
+            if response.endswith("```"):
+                response = response[:-3]
+            
+            action_data = json.loads(response.strip())
+            
+            if "action_type" not in action_data:
+                action_data["action_type"] = "request_changes"
+            if "comments" not in action_data:
+                action_data["comments"] = []
+            if "suggestions" not in action_data:
+                action_data["suggestions"] = []
+                
+            return json.dumps(action_data)
+            
+        except Exception as e:
+            print(f"Error getting action from LLM: {e}")
+            return FALLBACK_ACTION
+    
+    def parse_action(self, action_str: str) -> Dict[str, Any]:
+        try:
+            return json.loads(action_str)
+        except:
+            return {"action_type": "request_changes", "comments": [], "suggestions": []}
+
+
+def main():
+    import sys
+    sys.path.append('.')
+    
+    from environment.env import CodeReviewEnv
+    
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--task-id", type=str, default="bug_detection_easy",
+                       help="Task ID to run")
+    parser.add_argument("--max-steps", type=int, default=50,
+                       help="Maximum steps per episode")
+    args = parser.parse_args()
+    
+    env = CodeReviewEnv()
+    env.max_steps = args.max_steps
+    
+    agent = CodeReviewAgent()
+    
+    obs = env.reset(task_id=args.task_id)
+    done = False
+    step = 0
+    total_reward = 0.0
+    
+    print(f"Starting code review for task: {args.task_id}")
+    print(f"Task description: {obs.get('task_description', 'N/A')}")
+    print("-" * 50)
+    
+    while not done and step < args.max_steps:
+        action_str = agent.get_action(obs)
+        action = agent.parse_action(action_str)
+        
+        obs, reward, done, info = env.step(action)
+        
+        total_reward += reward
+        step += 1
+        
+        print(f"Step {step}:")
+        print(f"  Action: {action.get('action_type')}")
+        print(f"  Comments added: {len(action.get('comments', []))}")
+        print(f"  Suggestions: {len(action.get('suggestions', []))}")
+        print(f"  Reward: {reward:.3f}")
+        print(f"  Cumulative: {total_reward:.3f}")
+        print(f"  Done: {done}")
+        
+        if info.get('last_action_valid') is False:
+            print(f"  Warning: Invalid action! {info.get('error', '')}")
+        
+        print("-" * 50)
+    
+    final_score = env.get_task_score()
+    print(f"\nFinal Results:")
+    print(f"  Total reward: {total_reward:.3f}")
+    print(f"  Task score: {final_score:.3f}/1.0")
+    print(f"  Steps taken: {step}")
+    
+    env.close()
+    
+    # Save results
+    results = {
+        "task_id": args.task_id,
+        "total_reward": total_reward,
+        "task_score": final_score,
+        "steps": step,
+        "max_steps": args.max_steps
+    }
+    
+    with open("baseline_results.json", "w") as f:
+        json.dump(results, f, indent=2)
+    
+    print("\nResults saved to baseline_results.json")
+
+
+if __name__ == "__main__":
+    if not API_BASE_URL:
+        print("ERROR: API_BASE_URL environment variable not set")
+        print("Example: export API_BASE_URL=https://generativelanguage.googleapis.com")
+        sys.exit(1)
+    
+    if not API_KEY:
+        print("ERROR: API_KEY environment variable not set")
+        print("Example: export API_KEY=your-api-key-here")
+        sys.exit(1)
+    
+    if not MODEL_NAME:
+        print("ERROR: MODEL_NAME environment variable not set")
+        print("Example: export MODEL_NAME=gemini-1.5-pro")
+        sys.exit(1)
+    
+    main()