pdevyani/dev-assistant-env

Dev Assistant OpenEnv Environment

📌 Overview

This project implements an OpenEnv-compatible environment that simulates real-world developer tasks. The environment allows an AI agent to perform tasks such as fixing syntax errors, cleaning code, and reviewing code.

It is designed for evaluating AI systems in practical software engineering workflows.

🎯 Motivation

Developers spend significant time debugging, formatting, and reviewing code. This environment replicates these real-world tasks so that AI agents can be evaluated on meaningful and useful work instead of toy problems.

🧠 OpenEnv Interface

The environment follows the OpenEnv specification:

• —reset() → Returns initial observation
• —step(action) → Returns (observation, reward, done, info)
• —state() → Returns current state

🔍 Observation Space

The agent receives a task as input in text form.

Example: Fix syntax error: print('Hello'

⚡ Action Space

The agent must return a solution as text.

Example: print('Hello')

🎯 Reward Function

• —1.0 → Correct answer
• —0.0 → Incorrect answer
• —Can be extended to give partial rewards

The reward is deterministic and reproducible.

🧪 Tasks

🟢 Easy Task

Fix simple syntax errors Example: Missing parentheses

🟡 Medium Task

Clean and format messy code Example: Improve readability and indentation

🔴 Hard Task

Perform code review Example: Detect bugs and suggest improvements

📊 Grading

Each task includes a programmatic grader:

• —Returns score from 0.0 to 1.0
• —Fully correct → 1.0
• —Incorrect → 0.0

📁 Project Structure

openenv-dev-assistant/ │ ├── env.py ├── models.py ├── tasks.py ├── grader.py ├── run_baseline.py ├── requirements.txt ├── openenv.yaml ├── Dockerfile └── README.md

☁️ Deployment (Hugging Face Spaces)

1. 1.Go to Hugging Face Spaces
2. 2.Create new Space
3. 3.Select Docker as SDK
4. 4.Upload all project files
5. 5.Add tag: openenv

✅ Requirements Checklist

• —[x] Real-world task simulation
• —[x] OpenEnv interface implemented
• —[x] At least 3 tasks (easy, medium, hard)
• —[x] Programmatic grading
• —[x] Reward function
• —[x] Docker support
• —[x] Documentation

🚀 Future Improvements

• —Integrate OpenAI API for real agent evaluation
• —Add more complex coding tasks
• —Improve reward function with partial scoring
• —Expand dataset of tasks

📌 Author

Team Neon