SkeptAI – adversarial reasoning agent that challenges LLM outputs

Name: SkeptAI – adversarial reasoning agent that challenges LLM outputs
Author: datonpope

by datonpope·Mar 17, 2026·3 points·2 comments

View on HN

AI Analysis

●●●BangerBig BrainSolve My ProblemZero to One

Cross-model adversarial critique catches hallucinations before you forward them to clients.

Strengths

•Four structured passes with scored verdicts, not just a single sanity check.
•Claude output challenged by GPT-4o prevents self-referential bias in critiques.
•Free playground plus open source framework means you can actually audit the logic.

Weaknesses

•AI verification tools exist (FactCheck, various validators) — needs clearer differentiation.
•No mention of latency or cost for running four model passes per output.

Post Description

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