I built this after Claude cited an API feature that doesn't exist in a work analysis that I almost forwarded to a client. CRIT (the custom framework that I built behind it) runs four structured adversarial passes on any LLM output and delivers a scored verdict. Cross-model critique (Claude output gets challenged by GPT-4o and vice versa) to prevent self-referential bias. Free playground, open source framework. Think of it as a "Digital Devil's Advocate". Check it out and let me know what you think. skeptai.dev | github.com/datonpope/skeptai