Multi-Agent System Improves Code Summarization for Large Codebases

Understanding vast and intricate code repositories, especially those with poor documentation or complex structures, presents a significant challenge for developers. Current code summarization tools often fall short because they treat code as flat text and rely on single large language models, failing to utilize the inherent hierarchical relationships within a codebase. Agent4cs addresses this by introducing a multi-agent system. It employs a summarization agent to generate initial summaries, a keyword-extraction agent to identify crucial information from subfolders, and a quality-assurance agent to refine the output for clarity and completeness. This collaborative approach allows for a bottom-up summarization process. Evaluations show that Agent4cs enhances semantic consistency across all folder levels by an average of 8% and achieves up to 38% better normalized keyword coverage compared to structured prompting baselines. This demonstrates its effectiveness in providing more robust and informative summaries for complex software projects.