""" Debug Agent for CoDA. Executes generated code, diagnoses errors, and applies fixes to produce working visualizations. """ import logging import os import subprocess import sys import tempfile from pathlib import Path from typing import Optional from pydantic import BaseModel, Field from coda.core.base_agent import AgentContext, BaseAgent from coda.core.llm import LLMProvider from coda.core.memory import SharedMemory logger = logging.getLogger(__name__) class ExecutionResult(BaseModel): """Structured output from the Debug Agent.""" success: bool = Field( description="Whether execution succeeded" ) output_file: Optional[str] = Field( default=None, description="Path to the generated visualization" ) stdout: str = Field( default="", description="Standard output from execution" ) stderr: str = Field( default="", description="Error output from execution" ) error_diagnosis: Optional[str] = Field( default=None, description="Diagnosis of any errors" ) corrected_code: Optional[str] = Field( default=None, description="Fixed code if errors occurred" ) fix_applied: bool = Field( default=False, description="Whether a fix was applied" ) execution_time_seconds: float = Field( default=0.0, description="Time taken to execute" ) class DebugAgent(BaseAgent[ExecutionResult]): """ Executes generated code and handles errors. Runs the visualization code in a subprocess with timeout, diagnoses errors, and attempts automatic fixes. """ MEMORY_KEY = "execution_result" def __init__( self, llm: LLMProvider, memory: SharedMemory, timeout_seconds: int = 60, output_directory: str = "outputs", name: Optional[str] = None, ) -> None: super().__init__(llm, memory, name or "DebugAgent") self._timeout = timeout_seconds self._output_dir = Path(output_directory) self._output_dir.mkdir(parents=True, exist_ok=True) def execute(self, context: AgentContext) -> ExecutionResult: """Execute the generated code and handle errors.""" logger.info(f"[{self._name}] Starting code execution") generated_code = self._get_from_memory("generated_code") if not generated_code: return ExecutionResult( success=False, stderr="No generated code found in memory", ) code = generated_code.get("code", "") output_filename = generated_code.get("output_filename", "output.png") code = self._prepare_code(code, output_filename) result = self._execute_code(code) if not result.success and result.stderr: logger.warning(f"[{self._name}] Code execution failed: {result.stderr[:500]}") logger.info(f"[{self._name}] Attempting to fix errors") fixed_result = self._attempt_fix(code, result.stderr, context) if fixed_result.success: self._store_result(fixed_result) logger.info(f"[{self._name}] Fix successful!") return fixed_result logger.warning(f"[{self._name}] Fix attempt failed") result.error_diagnosis = fixed_result.error_diagnosis result.corrected_code = fixed_result.corrected_code self._store_result(result) logger.info(f"[{self._name}] Execution complete: success={result.success}") return result def _prepare_code(self, code: str, output_filename: str) -> str: """Prepare code for execution by setting up paths.""" output_path = self._output_dir / output_filename code = code.replace( f"'{output_filename}'", f"r'{output_path}'" ) code = code.replace( f'"{output_filename}"', f"r'{output_path}'" ) if "plt.savefig" not in code and "fig.savefig" not in code: code += f"\nplt.savefig(r'{output_path}', dpi=150, bbox_inches='tight')\n" return code def _execute_code(self, code: str) -> ExecutionResult: """Execute Python code in a subprocess.""" import time start_time = time.time() with tempfile.NamedTemporaryFile( mode="w", suffix=".py", delete=False, encoding="utf-8" ) as f: f.write(code) temp_file = f.name try: result = subprocess.run( [sys.executable, temp_file], capture_output=True, text=True, timeout=self._timeout, cwd=str(self._output_dir.parent), ) execution_time = time.time() - start_time output_files = list(self._output_dir.glob("*.png")) output_file = str(output_files[-1]) if output_files else None return ExecutionResult( success=result.returncode == 0, output_file=output_file, stdout=result.stdout, stderr=result.stderr, execution_time_seconds=execution_time, ) except subprocess.TimeoutExpired: return ExecutionResult( success=False, stderr=f"Execution timed out after {self._timeout} seconds", ) except Exception as e: return ExecutionResult( success=False, stderr=str(e), ) finally: try: os.unlink(temp_file) except OSError: pass def _attempt_fix( self, original_code: str, error_message: str, context: AgentContext, ) -> ExecutionResult: """Attempt to fix code errors using the LLM.""" fix_prompt = f"""The following Python visualization code produced an error. Please fix it. Original Code: ```python {original_code} ``` Error Message: {error_message} Provide a JSON response with: - diagnosis: What caused the error - corrected_code: The fixed Python code IMPORTANT: Return ONLY valid JSON. Do not include markdown formatting or explanations outside the JSON. Safe to assume standard libraries (matplotlib, seaborn, pandas, numpy) are available. JSON Response:""" response = self._llm.complete( prompt=fix_prompt, system_prompt="You are an expert Python debugger. Fix the code error and provide corrected code.", ) try: data = self._extract_json(response.content) diagnosis = data.get("diagnosis", "Unknown error") corrected_code = data.get("corrected_code", "") if corrected_code: output_filename = "output.png" corrected_code = self._prepare_code(corrected_code, output_filename) result = self._execute_code(corrected_code) result.error_diagnosis = diagnosis result.corrected_code = corrected_code result.fix_applied = result.success return result except Exception as e: logger.error(f"Failed to parse fix response: {e}") return ExecutionResult( success=False, stderr=error_message, error_diagnosis="Failed to automatically fix the error", ) def _build_prompt(self, context: AgentContext) -> str: return "" def _get_system_prompt(self) -> str: return "" def _parse_response(self, response: str) -> ExecutionResult: return ExecutionResult(success=False) def _get_output_key(self) -> str: return self.MEMORY_KEY