Decepticon - An Autonomous Hacking Agent for Red Teams

Decepticon is an autonomous red team agent that goes beyond simple scanning to execute the actual attack chain of an adversary (reconnaissance, exploitation, privilege escalation, lateral movement, C2).

Key Features and Characteristics:

• Professional Attack Chain: Generates engagement packages (RoE, ConOps, etc.) including MITRE ATT&CK mapping, and runs interactive tools like msfconsole and sliver-client in tmux sessions, sending follow-up commands through automatic prompt detection.
• Sandbox Isolation: All commands are executed within a Kali Linux sandbox. LangGraph drives the sandbox via the Docker socket, using a dual-network architecture that separates the operations network from the management network.
• 16 Specialized Agents: 16 agents organized by kill chain stage—Orchestration, Reconnaissance, Exploitation, and more—collaborate together. Each uses a new context window per objective to prevent accumulated noise.
• Broad Model and OS Support: Supports a wide range of LLM providers including Anthropic, OpenAI, Google Gemini, DeepSeek, xAI, and is available on macOS, Linux, and Windows (WSL2) environments.