Took Vercel's deepsec, extracted the prompt and ran it with Warden. It did better than I thought it would (which is a credit to how powerful models are natively), but surprisingly it actually grabbed a couple of findings our synthesized skill didn't. Net-net it missed some