Code Review Crew: A Multi-Agent Automated Code Review and Verified Refactoring Platform
Year: 2026
Type: article
Abstract: Code Review Crew is a multi-agent AI platform developed to automate software code review and verified refactoring. The system uses a FastAPI backend to analyze Git changes and coordinate specialized agents for security, performance, and code-quality assessment. The Security Agent combines Bandit, secret scanning, and pattern analysis to identify vulnerabilities such as SQL injection and hardcoded credentials. The Performance Agent uses Ruff, Radon, and code-pattern analysis to detect complexity and inefficient programming practices. When refactoring is enabled, the Refactor Agent uses local Ollama or cloud-based Groq language models to generate a complete corrected source file. The generated patch is copied into an isolated sandbox and validated using the repository’s tests. If validation fails, an Auto-Debug Loop sends the failure output back to the Refactor Agent for another attempt. The platform also supports GitHub webhooks and provides a browser dashboard for displaying the review plan, findings, model details, sandbox results, and generated patch. In the demonstration, the system identified four security signals and one performance issue, and the generated patch successfully passed sandbox validation. The project provides a transparent, human-supervised approach to AI-assisted code review and automated refactoring.
Authors: R, Kirthiga, abilash, shine
Language: English
Subfield: Information Systems
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7236249090
