Bullshit Detector

Got tired myself watching 30min videos with millions of views and not having cheap way to check whether any if that is true. So I build set of agent skills that do fact-checking for me. You paste a link - youtube, tiktok, article, pdf, tweet or just a text and as if it's a bullshit? Skill pulls the content, splits it into individual claims, searches each one and give you per-claim verdicts with links and 0-10 BS score. The rule is - a verdict needs a source. Model can't confirm anything from it's memory, even when it's sure. MIT, no API keys needed. Works in Claude Code, Codex or any agent with web search.