import os import ast import random from typing import Tuple, Dict, Any, Optional, List # Updated imports based on your library's structure from openenv.core.env_server import Environment from models import Action, Observation, Reward, State class CodeReviewEnv(Environment): def __init__(self): super().__init__() self.current_task_id = "style-cleanup" self.code = "" self.step_count = 0 self.max_steps = 5 def reset(self, *, seed: Optional[int] = None, options: Optional[Dict[str, Any]] = None) -> Tuple[Observation, Dict[str, Any]]: """Standard OpenEnv Reset Protocol: returns (obs, info)""" if seed is not None: random.seed(seed) task_id = options.get("task_id", "style-cleanup") if options else "style-cleanup" self.current_task_id = task_id self.step_count = 0 # Load the buggy code from our data folder file_map = { "style-cleanup": "data/style.py", "efficiency-boost": "data/logic.py", "security-audit": "data/security.py" } file_path = file_map.get(task_id, "data/style.py") try: with open(file_path, "r") as f: self.code = f.read() except Exception: self.code = "# Error loading source file" return self._get_observation(), {} def step(self, action: Action) -> Tuple[Observation, float, bool, Dict[str, Any]]: """Standard OpenEnv Step Protocol: returns (obs, reward, done, info)""" self.step_count += 1 reward_val = 0.01 done = False if action.action_type == "apply_fix": self.code = action.content reward_val = self._calculate_reward() elif action.action_type == "submit": reward_val = self._calculate_reward() done = True if self.step_count >= self.max_steps: done = True # Mathematical Clamp Armor: min(max(reward, 0.01), 0.99) clamped_reward = min(max(reward_val, 0.01), 0.99) return self._get_observation(), clamped_reward, done, {} def state(self) -> State: """Standard OpenEnv State Protocol: returns a State object""" return State( current_task_id=self.current_task_id, step_count=self.step_count, code=self.code ) def _get_observation(self) -> Observation: """Returns what the agent sees right now.""" return Observation( file_name=f"{self.current_task_id}.py", code_content=self.code, diff="Changes applied to buffer.", linter_report=self._mock_linter(), current_task=self.current_task_id ) def _mock_linter(self): """A simple linter to guide the agent.""" errors = [] if "import sys" in self.code and self.current_task_id == "style-cleanup": errors.append("L001: Unused import 'sys' detected.") if "f\"SELECT" in self.code and self.current_task_id == "security-audit": errors.append("S001: SQL Injection vulnerability detected.") return errors def _calculate_reward(self) -> float: """Smart AST Grader v2: Sophisticated code analysis.""" score = 0.01 try: tree = ast.parse(self.code) if self.current_task_id == "style-cleanup": imported_names = set() used_names = set() for node in ast.walk(tree): if isinstance(node, (ast.Import, ast.ImportFrom)): for alias in node.names: imported_names.add(alias.asname or alias.name) elif isinstance(node, ast.Name) and isinstance(node.ctx, ast.Load): used_names.add(node.id) if 'sys' not in imported_names: score += 0.49 if "Indentation" in self.code and "corrected" in self.code: score += 0.49 elif self.current_task_id == "security-audit": query_vars_with_fstrings = set() uses_vulnerable_execute = False uses_parameterized_execute = False for node in ast.walk(tree): if isinstance(node, ast.Assign): if isinstance(node.value, ast.JoinedStr): for target in node.targets: if isinstance(target, ast.Name): query_vars_with_fstrings.add(target.id) if isinstance(node, ast.Call) and hasattr(node.func, 'attr') and node.func.attr == 'execute': arg0 = node.args[0] if isinstance(arg0, ast.JoinedStr): uses_vulnerable_execute = True elif isinstance(arg0, ast.Name) and arg0.id in query_vars_with_fstrings: uses_vulnerable_execute = True if len(node.args) > 1: uses_parameterized_execute = True elif isinstance(arg0, (ast.Str, ast.Constant)): val = arg0.s if hasattr(arg0, 's') else arg0.value if "?" in val or "%s" in val: uses_parameterized_execute = True if not uses_vulnerable_execute: score += 0.5 if uses_parameterized_execute: score += 0.49 elif self.current_task_id == "efficiency-boost": loop_nodes = (ast.For, ast.While, ast.ListComp, ast.SetComp, ast.DictComp, ast.GeneratorExp) max_depth = 0 for node in ast.walk(tree): if isinstance(node, loop_nodes): current_depth = 1 for subnode in ast.walk(node): if subnode is node: continue if isinstance(subnode, loop_nodes): current_depth = 2 break max_depth = max(max_depth, current_depth) if max_depth == 1: score = 0.99 elif max_depth >= 2: score = 0.2 else: score = 0.01 except Exception: score = 0.01 return score