Awesome AI Agents › Red-Teaming Platforms

westonbrown/Cyber-AutoAgent

⭐ 547 TypeScript repository created 2025-05-26

Cyber-AutoAgent is an experimental offensive security agent that performs black box penetration testing without a human driving each step. The operator supplies a target and an objective in natural language, and the agent reasons about the assessment, selects shell tools such as scanners and exploitation utilities on its own, runs them, and records every confirmed finding as evidence. The core agent loop is written in Python on top of the Strands agent framework and supports several model providers: AWS Bedrock, LiteLLM as a gateway to hosted APIs, and local Ollama models for offline work. Findings are written into a Mem0 memory layer backed by FAISS, OpenSearch or the hosted Mem0 platform, so evidence persists across steps and across runs and can be queried later when the agent writes its report. A swarm tool lets the main agent spawn parallel sub-agents that share the same memory for wider tasks. Prompts are organised into swappable operation modules, so the same engine can be pointed at different assessment styles. The default interface is a React terminal application that spawns the Python agent as a subprocess, guides first-time configuration and streams the agent reasoning and tool execution live; Docker images and a Compose stack with Langfuse observability are provided for containerised runs. The project reports 85 percent on the XBOW validation benchmark. It targets security researchers and red teams who want to study autonomous offensive agents, and the authors stress that it is experimental software for authorised, sandboxed targets only. The repository was archived by its author, who cited a lack of time to bring it to production maturity, so it is a reference implementation rather than a maintained tool.

https://github.com/westonbrown/Cyber-AutoAgent

agentsautonomous-agentscybersecurity-toolspenetration-testingstrands-agentred-teamingpython