YggPurpleAILAB/Decepticon
↗ GitHubAutonomous Hacking Agent for Red Team Testing
949
Stars
177
Forks
7
Watchers
2
Open Issues
Safety Rating B
Decepticon is a legitimate, Apache-2.0 licensed open source project from PurpleAILAB aimed at authorized red team testing. It includes a clear disclaimer against unauthorized use and is designed with isolation (Docker/Kali sandbox) in mind. However, it is inherently a dual-use tool: its core functionality is autonomous exploitation of computer systems. The repository itself does not contain hardcoded secrets, obfuscated code, or malicious intent beyond its declared offensive security purpose. The 'Caution' rating reflects the significant dual-use risk rather than any malicious artifacts found in the codebase — curators should consider whether this category of tool fits catalog guidelines.
ℹAI-assisted review, not a professional security audit.
AI Analysis
Decepticon is an autonomous multi-agent red team hacking framework built with LangChain and LangGraph. It orchestrates a hierarchy of specialist AI agents (reconnaissance, planning, exploitation, post-exploitation) to autonomously execute full penetration testing kill chains against target systems. All tooling runs inside an isolated Kali Linux Docker sandbox, and the system is controlled via a real-time streaming CLI. It supports Anthropic and OpenAI LLM backends.
Use Cases
- Automated red team engagements and penetration testing against authorized targets
- Autonomous execution of multi-stage attack kill chains (recon, exploit, post-exploit)
- Security researcher tooling for adversary emulation in isolated sandbox environments
- Generating operations plans, rules of engagement, and ConOps documentation for red team exercises
- Demonstrating offensive security techniques against intentionally vulnerable targets (Metasploitable, DVWA)
Tags
Security Findings (1)
The repository is intentionally designed to automate offensive security operations including port scanning, vulnerability exploitation, privilege escalation, lateral movement, and persistence. While marketed for authorized red team use, the tooling executes real attack techniques (e.g., Metasploit exploits via vsftpd backdoor) autonomously. This is by design but represents significant dual-use risk if deployed against unauthorized systems.
Project Connections
PentAGI
→Both are autonomous multi-agent penetration testing frameworks running in Docker. PentAGI is built in Go with 20+ integrated professional security tools and a full monitoring stack; Decepticon uses LangGraph on Kali Linux with a hierarchical specialist agent architecture. They share the same autonomous kill-chain model but differ in scope and implementation.
Strix
→Both deploy multi-agent systems for automated security testing. Strix focuses on application security with CI/CD integration and auto-fix capabilities; Decepticon targets full kill-chain penetration testing through reconnaissance, exploitation, and post-exploitation phases in an isolated Kali Linux sandbox.
ZeroLeaks
→ZeroLeaks tests AI systems specifically for prompt injection and system prompt extraction vulnerabilities using adversarial LLM attack chains; Decepticon conducts full external penetration tests at the network and application layer. Together they cover both AI-layer and infrastructure-layer security assessment.