REPO GUARDIAN -THE MULTI AGENT BUG HUNTER

Abstract

The Multi-Agent Bug Hunter is an intelligent, automated code analysis system that integrates static analysis tools with large language model (LLM)-based reasoning to identify and remediate software vulnerabilities in public GitHub repositories. The system operates through multiple coordinated agents, each responsible for a distinct stage in the process. The RepoCloner agent clones the target repository and extracts supported source files, while the StaticAnalyzer agent employs tools such as Bandit and Semgrep to perform rule-based vulnerability scanning. The LLMReviewer, powered by Google’s Gemini model, conducts an in-depth semantic review of the code to detect logic flaws, code smells, and potential security risks, generating structured JSON reports with optional auto-fix patches. The ReportWriter consolidates all findings into a comprehensive Markdown report for developers, and the PrPusher agent can automatically apply safe patches and create GitHub pull requests for immediate remediation. By combining traditional static analysis with AI-driven insights, the system delivers a hybrid, multi-agent framework for efficient, context-aware bug detection and automated code improvement.