larlarua/AutoCVE

What it solves

AutoCVE automates the entire lifecycle of discovering Common Vulnerabilities and Exposures (CVEs) in open-source projects. It removes the manual effort required to screen projects, import repositories, audit source code, verify vulnerabilities, and generate the formal reports needed for CVE applications.

How it works

The system uses a Multi-Agent orchestration framework where a central Orchestrator coordinates specialized agents:

• Recon Agent: Collects initial information.
• Scan Agent: Performs tool-based scanning.
• Triage Agent: Filters false positives from scans.
• Finding Agent: The core engine that performs deep source code analysis using a ReAct Loop and specialized tools to find high-value vulnerabilities.
• Verification Agent: Dynamically validates the discovered bugs.

It offers three audit modes: Enhanced Scanning (Scan $\rightarrow$ Triage), Intelligent Auditing (Finding), and Comprehensive Auditing (combining both).

Who it’s for

Security researchers, bug bounty hunters, and developers focused on CVE research and 0Day discovery who want to automate the tedious parts of the vulnerability discovery pipeline.

Highlights

• End-to-End Automation: Covers everything from project selection to report generation.
• ReAct Loop Integration: Uses a reasoning-and-acting loop to refine vulnerability searches and correct paths via "Nudge" mechanisms.
• Interactive Auditing: Users can chat with the agent during the process to ask for more evidence or a better explanation of the attack chain.
• Visual Tracking: Provides a detailed view of the Agent Tree, tool calls, and execution paths for every audit.
• Extensible Skills: Allows custom skill configurations for different agents to expand their capabilities.