T3MP3ST: a multi-agent offensive security framework that automates the kill chain from recon to exploit

What it solves

T3MP3ST is an offensive security framework designed to automate the "kill chain" (reconnaissance, exploitation, and reporting) for authorized security testing. It aims to make professional bug-hunting and vulnerability discovery accessible by coordinating AI agents to perform tasks across web apps, CTFs, source code, and embedded systems.

How it works

The framework acts as a "war machine" bolted around an existing AI coding agent (such as Claude Code, Codex, or local models via Ollama). It uses a multi-agent architecture consisting of eight specialized operators (Recon, Scanner, Exploiter, Infiltrator, Exfiltrator, Ghost, Coordinator, and Analyst) that map to the MITRE ATT&CK framework. These agents use a ReAct loop to drive a built-in "Arsenal" of security tools (e.g., nmap, DNS, HTTP fingerprinting) to interact with targets.

Who it’s for

Security researchers, red-teamers, and educators who have explicit authorization to test systems. It is specifically designed for those who want to leverage their existing AI coding tools for automated vulnerability research without needing additional API keys or cloud tenants.

Highlights

• Keyless Integration: Connects directly to local AI agents or offline models (Ollama, vLLM) to avoid additional costs or API keys.
• Tool-Backed Recon: A stable recon engine that drives real security tools, achieving high pass rates on the XBEN benchmark.
• Reproducible Claims: Includes a verify-claims command that re-derives all performance benchmarks from committed data to ensure honesty.
• Broad Domain Coverage: Supports web apps, CTFs, and a coordinated-disclosure pipeline for robotics/OT/embedded OSS.
• Egress-Scope Containment: Built-in protections that prevent networked tools from attacking hosts outside the defined mission target.