PRANAV05092003/autonomous-code-refactoring-env

🚀 ACRE — Autonomous Code Refactoring Environment

OpenEnv-powered AI system for real-world code optimization, refactoring, and evaluation.

🔥 Overview

ACRE is an OpenEnv-compliant environment designed to simulate real-world software engineering workflows such as code cleanup, optimization, and refactoring using AI agents.

It enables agents to iteratively improve code through structured actions while receiving dense, step-wise reward feedback.

Environment Overview and Motivation

ACRE models a realistic developer workflow where an agent incrementally improves Python code quality under a fixed action budget. The environment is designed for OpenEnv Round 1 requirements: typed APIs, deterministic grading, multi-difficulty tasks, and reproducible inference behavior.

💡 Why This Matters

Modern software systems require automated code optimization and intelligent tooling.

ACRE enables:

• —🤖 AI coding assistants
• —🔍 Automated code review systems
• —⚡ Reinforcement learning-based optimization agents
• —🧠 Learning real developer workflows

🔄 How It Works

Code → Action → Refactor → Reward → Repeat

1. 1.Load messy code
2. 2.Apply transformation
3. 3.Evaluate using grader
4. 4.Compute reward
5. 5.Iterate until optimal

🧠 Key Features

• —✅ Autonomous code refactoring
• —⚡ Step-wise reward feedback
• —🧪 OpenEnv compliant interface
• —📊 Deterministic grading system
• —🔁 Reproducible inference pipeline
• —🐳 Fully containerized (Docker + Hugging Face Spaces)

📂 Tasks

Each task uses AST-based transformations and deterministic grading.

Task Descriptions with Expected Difficulty Levels

• —Easy (rename_variables): rename generic names like x, tmp, i into descriptive identifiers.
• —Medium (remove_dead_code): remove unreachable branches and unused assignments while preserving behavior.
• —Hard (full_refactor): combine renaming, dead-code elimination, loop simplification, condition cleanup, and helper inlining.

🎯 Reward System

Rewards are computed at every step:

• —✅ Valid executable code → positive reward
• —📉 Reduced complexity → reward
• —⚡ Improved performance → reward
• —❌ Errors or invalid code → penalty
• —🔁 No progress → penalty

Normalization:

(raw_reward + 32) / 52 → [0, 1]

📊 Example Execution

[START] task=rename_variables
[STEP] action=0
[END] task=rename_variables score=1.00

[START] task=remove_dead_code
[STEP] action=1
[END] task=remove_dead_code score=0.25

[START] task=full_refactor
[STEP] action=3
[END] task=full_refactor score=0.71

Final Score: 0.65

🏗️ Architecture

• —server/app.py → FastAPI entry point used by OpenEnv + Docker
• —server.py → legacy local runner / UI helper
• —openenv_interface.py → OpenEnv wrapper
• —acre/env/ → Core environment logic
• —acre/tasks/ → Task definitions
• —acre/utils/ → Metrics and helpers
• —inference.py → Evaluation pipeline

⚙️ OpenEnv Interface

observation = env.reset()
observation, reward, done, info = env.step(action)
state = env.state()

Uses Pydantic models:

• —ObservationModel
• —ActionModel
• —RewardModel

Definitions of Action and Observation Spaces

• —Observation space: Box(4) with fields code_length, complexity_score, runtime_s, error_flag.
• —Action space: Discrete(5) with actions rename_variable, remove_dead_code, simplify_loop, optimize_condition, inline_function.

🌐 HTTP API

🚀 Run Locally

Setup and Usage Instructions

pip install -r requirements.txt
uvicorn server.app:app --host 0.0.0.0 --port 7860

🐳 Docker / Hugging Face Spaces

docker build -t acre .
docker run -p 7860:7860 \
  -e API_BASE_URL=https://api.openai.com/v1 \
  -e MODEL_NAME=gpt-4o-mini \
  -e API_KEY=your_key \
  -e ENV_URL=http://localhost:7860 \
  acre

🧪 Inference

Set environment variables:

export API_BASE_URL=https://api.openai.com/v1
export MODEL_NAME=gpt-4o-mini
export API_KEY=your_key
export ENV_URL=http://localhost:7860

Run:

python inference.py

Expected output:

Easy: 1.00
Medium: 0.25
Hard: 0.71
Final: 0.65

📌 OpenEnv Compliance

• —✔ step() implemented
• —✔ reset() implemented
• —✔ state() implemented
• —✔ reward shaping
• —✔ deterministic grading
• —✔ structured logs

🧪 Validation

python validate.py --url http://localhost:7860

Or:

openenv validate

🌐 Live Demo

👉 Running on Hugging Face Spaces

📊 Baseline Performance

Baseline Performance Scores

🏆 Use Cases

• —AI-powered code optimization
• —Automated refactoring tools
• —Reinforcement learning environments
• —Developer productivity systems

📜 License

MIT License