title	Code Debug Env
colorFrom	blue
colorTo	green
sdk	docker
pinned	false
license	mit
tags	openenv code-debugging reinforcement-learning

Code Debug Environment (OpenEnv)

This project is an OpenEnv-compatible environment designed to simulate how an AI agent can debug Python code step by step.

Instead of solving everything in one attempt, the agent is given buggy code, error messages, and hints, and it gradually improves the code across multiple steps — just like how real debugging works.

---

What this project does

• Allows an AI agent to fix broken Python code
• Supports multiple difficulty levels: easy, medium, hard
• Provides rewards based on how correct the solution is
• Encourages step-by-step improvement instead of one-shot answers
• Simulates real-world debugging scenarios

---

How the environment works

What the agent receives (Observation)

At each step, the agent gets:

• challenge_id → identifies the problem
• difficulty → easy / medium / hard
• description → explains the task
• buggy_code → the incorrect code
• error_message → what’s going wrong
• hint → optional guidance
• step → current attempt number
• max_steps → maximum attempts allowed

---

What the agent does (Action)

The agent responds by submitting a corrected version of the code:

{
  "fixed_code": "your corrected python code here"
}