dallas@tulsa:~/resume$_ → trajectory mode
dallas@tulsa: ~/resume
dallas@tulsa:~/resume$ cat projects/openclaw.md
← ls projects/

OpenClaw Red-Team Harness

Deliberately exploitable multi-agent harness used as a live red-team target for CYB-4203 students.

What it is

OpenClaw is a multi-agent framework that Dallas configured as a deliberately exploitable red-team target for the CYB-4203/6203 Secure & Trustworthy AI course. Five agents (concierge, ops, knowledge, triage, hr) run on a RunPod GPU pod and are exposed to students via prompt injection attack surfaces — giving six student teams a realistic, hosted target to attack, document, and report on.

The deployment runs on an RTX PRO 4500 Blackwell pod via RunPod, using GLM-4.7-flash (ollama) as the underlying model. The infrastructure stack is Cloudflare + DigitalOcean (front door) + RunPod (compute). Each agent has a defined identity, skill helpers, and cron-triggered turns. Attack surfaces are intentionally left open to give students meaningful wins while teaching them structured vulnerability disclosure.

Status

Active — deployed and serving the Spring 2026 CYB-4203 final project red-team exercise. Pod hosted at a fixed RunPod endpoint accessible to enrolled students.