This repository serves as a comprehensive survey of Large Language Model (LLM)-based agents applied to Software Engineering (SE). It categorizes 106 papers from both SE task and agent architecture perspectives, aiming to provide researchers and practitioners with a structured overview of the rapidly evolving field, highlighting open challenges and future directions.

How It Works

The survey systematically categorizes research based on two primary dimensions: Software Engineering tasks (e.g., requirement engineering, code generation, testing, debugging, release, maintenance) and Agent Architecture components (e.g., planning, memory, perception, action, multi-agent systems, collaboration mechanisms). This dual categorization allows for a granular understanding of how LLM agents are being utilized and designed within the SE domain.

Quick Start & Requirements

This repository is a curated list of research papers and does not involve direct code execution or installation. The primary "requirement" is access to the cited research papers, typically available via arXiv or other academic publication platforms.

Highlighted Details

• Comprehensive coverage of 106 papers in LLM-based agents for Software Engineering.
• Categorization across 10 SE tasks and 7 agent architecture components.
• Includes discussions on open challenges and future research directions.
• Regularly updated with new research in this fast-evolving field.

Maintenance & Community

The repository is maintained by Junwei Liu, Kaixin Wang, and Yixuan Chen. Contact is available via email for questions and suggestions. The project encourages users to star the repository to stay updated.

Licensing & Compatibility

The repository itself contains a list of research papers. The licensing of the individual papers is determined by their respective publication venues (e.g., arXiv, conference proceedings). The survey content is likely intended for academic and research use.

Limitations & Caveats

This repository is a survey and does not provide executable code or a platform for experimentation. Its value is in its curated list of research and analysis, not in direct implementation. The field is rapidly evolving, meaning the survey represents a snapshot in time.