开发者在Hacker News上推出了Bullshit Detector，这是一款基于AI智能体的事实检查工具，可对YouTube视频、文章、推文和PDF等多种网络内容进行逐条声明验证[1]。该工具使用Claude代码构建，生成包含源信息和BS评分（0-10分）的综合报告[1]。
该工具支持多种内容源，无需API密钥即可使用[1]。每条声明的验证结果分为五个等级：已确认、似是而非、误导、虚假或无法验证[1]。除评分外，工具还能扫描虚假信号并进行激励分析[1]。用户可通过skills.sh或插件市场在Claude Code、Codex、OpenCode等平台安装该工具[1]。
在应用案例中，该工具对一条116万浏览量的"用AI赚钱的14种方式"视频的验证结果为5/10分，其中12条声明中4条已确认、2条似是而非、3条误导、0条虚假、3条无法验证[1]。另对一条55.2万浏览量的"太阳有隐藏的孪生星"视频的验证结果为9/10分[1]。
Developer Serhii Korniienko has unveiled Bullshit Detector on Hacker News, an artificial intelligence-powered fact-checking tool designed to verify claims across multiple media formats through automated statement-by-statement analysis [1]. The system evaluates content from YouTube, TikTok, tweets, articles, and PDFs without requiring API keys, classifying each claim with status indicators—confirmed, partially true, misleading, false, or unverifiable—and generating a BS score on a 0-to-10 scale [1].
The tool, built using Claude Code, can be installed across multiple agent platforms including Claude Code, Codex, and OpenCode via skills.sh or plugin marketplaces [1]. Beyond scoring, Bullshit Detector performs false signal scanning and analyzes underlying incentives within the content [1]. The system runs as pure markdown or self-contained Python scripts, making it accessible across different agent environments [1].
In practical application, the tool analyzed a viral video titled "14 Ways to Make Money with AI" that garnered 1.16 million views, assigning it a 5/10 BS score after evaluating 12 claims—finding 4 confirmed, 2 partially true, 3 misleading, 0 false, and 3 unverifiable [1]. A separate video claiming "The Sun Has a Hidden Twin Star" with 552,000 views received a 9/10 BS score [1].