Yean-Sec/StrikeAgent_AtkBrain-Flash

What it solves

StrikeAgent_AtkBrain-Flash is an AI-driven penetration testing framework designed to automate the process of finding vulnerabilities and gaining access (getshell) in real-world external network environments. It addresses common AI security tool issues such as high false-positive rates, "hallucinated" vulnerabilities, and the tendency for AI agents to drift away from the target objective during complex attacks.

How it works

The system uses a "self-looping, self-supervised, and self-evolving" architecture. It employs a control console to schedule reconnaissance, an attack graph to drive the attack cycle, and a supervision mechanism that generates plans at the boundaries of each round. To ensure accuracy, it implements a red-team secondary rating and verification process to eliminate exaggerated or false reports. Successful attack techniques are distilled into a memory library and fed back into future operations. It integrates with Claude Code to execute commands on a Kali Linux host, utilizing a set of specialized "skills" (like kali-kit, recon-fanout, and recon-spiral) to interact with penetration testing tools.

Who it’s for

It is designed for security researchers, red teamers, and CTF players who have explicit authorization to test target environments and want to automate the reconnaissance and exploitation phases of a penetration test.

Highlights

• Attack Graph Driven: Uses a visual attack graph and timeline to track progress and lateral movement within networks.
• Verified Vulnerabilities: Includes a secondary verification layer to ensure vulnerabilities are actionable and not false positives.
• High Performance: Ranked 3rd on the Cybench official leaderboard using deepseek-v4-flash.
• Tool Integration: Deeply integrated with Kali Linux tools and Claude Code for autonomous command execution.
• Automated Reporting: Generates detailed delivery reports including asset profiling and critical attack paths.