v0.1.0-alpha - Initial Public Release: Proof of Concept

Adversarial Security Agents - Initial Public Release

Proof-of-concept framework for autonomous adversarial security testing.

What This Is

An autonomous red team agent that:

• Queries a RAG knowledge base (5,395 offensive security documents)
• Uses LLM reasoning to plan attacks
• Executes tools autonomously (2000+ BlackArch utilities)
• Runs isolated in Kubernetes with NetworkPolicy guarantees

Built to explore whether adversarial AI competitions could secure open-source ecosystems at scale.

Key Components

• Autonomous agent with 5-phase attack methodology
• MCP RAG server (GTFOBins, Atomic Red Team, HackTricks, MITRE ATT&CK)
• Kubernetes-native deployment with kernel-level isolation
• Complete documentation (4,000+ lines)
• Attack scenarios mapped to MITRE ATT&CK

Model Configuration

• Qwen2.5-Coder-14B-Instruct-abliterated-GGUF
• LM Studio 0.3.29+
• Optimized inference: temp=0.4, min_p=0.08, Flash Attention, Q8_0 KV cache

Status

Built in 48 hours (October 8-9, 2025). Core components functional and documented.

Live testing pending infrastructure stabilization. Blue team agent in development.

Getting Started

See README.md for complete setup instructions.

License

⚠️ For authorized security research only. See README.md for ethical guidelines.