RepoAudit introduces an innovative approach to automating comprehensive code auditing using a multi-agent LLM system that can analyze entire code repositories to identify security vulnerabilities.

• Enhanced Discovery: Employs a structured multi-agent workflow with specialized roles (Coordinator, Explorer, Inspector, Reporter) to systematically audit codebases
• Context-Aware Analysis: Overcomes traditional LLM context limits by intelligently managing repository exploration and knowledge sharing between agents
• Practical Results: Outperforms existing tools by finding 43.3% more bugs with 50.7% higher precision in benchmark testing
• Security Impact: Enables proactive identification of vulnerabilities before deployment, reducing security risks in software development pipelines

RepoAudit: An Autonomous LLM-Agent for Repository-Level Code Auditing