What 356 public agent prompts taught me about written safeguards

I scored 356 publicly collected AI agent prompts against 23 written guardrail controls. The 83 that...