Step-by-Step Guide: Building a Self-Correcting Code Agent with LangGraph & DeepSeek-R1

Stateful Agentic Graph Tutorial: Construct an autonomous code generator agent that compiles Python snippets, captures execution errors, and reflects to self-correct code bug loops.

Step 1: Define Agent Execution Graph State

from pydantic import BaseModel, Field
from typing import List, Optional

class CodeAgentState(BaseModel):
    user_prompt: str
    generated_code: Optional[str] = None
    compiler_error: Optional[str] = None
    retry_count: int = 0
    max_retries: int = 3
    is_success: bool = False

Step 2: Write Code Execution & Sandbox Verification Node

import subprocess, sys

def execute_code_node(state: CodeAgentState) -> CodeAgentState:
    try:
        res = subprocess.run([sys.executable, "-c", state.generated_code], capture_output=True, text=True, timeout=5)
        if res.returncode == 0:
            state.is_success = True
        else:
            state.compiler_error = res.stderr
            state.retry_count += 1
    except Exception as err:
        state.compiler_error = str(err)
        state.retry_count += 1
    return state