PurpleAILAB/Decepticon

Orchestrates multi-stage attack chains through specialized LLM agents (Recon, Exploit, Post-Exploit) that read operations plans and adapt in real time, executing reconnaissance through lateral movement within an isolated Kali sandbox. Each agent spawns with a clean context window per objective, storing findings to disk to maintain reasoning quality across iterations. Integrates with Anthropic/OpenAI APIs and ships as a Docker-first CLI tool that streams agent actions in real time.

903 stars. Actively maintained with 67 commits in the last 30 days.