code-review-env / README.md
MazeRunner1531's picture
Update README.md
62dd127 verified
|
Raw History Blame Contribute Delete
798 Bytes
metadata
title: Code Review Env
emoji: 🤖
colorFrom: blue
colorTo: purple
sdk: docker
pinned: false

AI Code Review Environment

Simulates GitHub PR review for AI agents.

Tasks

  • Easy: Syntax bug detection
  • Medium: Logical/security bug detection
  • Hard: Full PR evaluation

Actions

  • comment
  • approve
  • request_changes

Run

python inference.py

Advanced Evaluation Design

This environment evaluates:

  • Issue detection accuracy
  • Explanation quality
  • Severity-aware scoring

Reward Design

  • Partial rewards for correct issue identification
  • Additional rewards for meaningful explanations
  • Penalties for irrelevant comments

Issue Types

  • Security vulnerabilities (high severity)
  • Performance inefficiencies (medium severity)
  • Syntax errors (high severity)