Agent4cs multi-agent system boosts code summarization accuracy by 38%

Agent4cs is a novel multi-agent system designed to tackle the challenge of summarizing large, hierarchical codebases with obfuscated structure and sparse documentation. Unlike traditional single-model approaches that treat code as flat text, Agent4cs employs three distinct agents working in a bottom-up fashion: a summarization agent generates robust summaries, a keyword-extraction agent proactively identifies critical information from subfolders, and a quality-assurance agent iteratively refines outputs for readability, coherence, and completeness. This collaborative structure leverages the rich interdependencies and folder organization inherent in real-world repositories.

Evaluated on 7 frontier models (including likely GPT-4o, Claude, etc.), Agent4cs shows an average 8% improvement in semantic consistency across all folder levels compared to two structured prompting baselines using code segments. On real-world datasets, the system achieves up to 38% gains in normalized keyword coverage rate, making it more effective at capturing essential code semantics. The research, accepted to the main track of EUMAS 2026, highlights a practical path for automated code understanding in complex software projects.