MANTRA introduces a novel framework that leverages contextual retrieval-augmented generation (RAG) and multi-agent LLM collaboration to automate code refactoring while ensuring compilability and successful test execution.

• Combines specialized LLM agents with distinct roles to tackle complex refactoring tasks
• Uses contextual RAG to provide relevant codebase information to the LLM agents
• Ensures refactored code maintains functionality by verifying compilability and test execution
• Reduces developer workload by automating labor-intensive refactoring processes

This research significantly advances software engineering productivity by transforming manual refactoring into an automated process, allowing developers to focus on higher-value tasks while maintaining code quality and reliability.

MANTRA: Enhancing Automated Method-Level Refactoring with Contextual RAG and Multi-Agent LLM Collaboration