B2prakash/code-review-env

Code Review Environment

An OpenEnv environment for training AI agents to review code diffs.

Description

This environment simulates real-world code review tasks where an AI agent must identify bugs, security vulnerabilities, and code quality issues.

Tasks

• —Task 1 (Easy): Bug Identification — find off-by-one and missing checks
• —Task 2 (Medium): Security Audit — find SQL injection and auth issues
• —Task 3 (Hard): Refactor Quality Review — find performance and edge case bugs

Action Space

• —review_text: your detailed review
• —issues_found: list of issues
• —severity: low, medium or high

Observation Space

• —taskid, codediff, language, context, turn

Setup

pip install fastapi uvicorn pydantic openai uvicorn api.main:app --host 0.0.0.0 --port 7860

Docker

docker build -t code-review-env . docker run -p 7860:7860 code-review-env

Baseline

export OPENAIAPIKEY=yourkeyhere python baseline_inference.py

Endpoints

• —POST /reset — Start a new episode
• —POST /step — Take an action
• —GET /state — Get current state
• —GET /tasks — List all tasks
• —POST /grader — Get last grade
• —POST /baseline — Run baseline agent

Author

B2prakash