July 29, 2026
BS meter meets comment chaos
Show HN: Bullshit Detector – agent skills that fact-check videos and articles
A lie-sniffing tool drops — and the comments instantly ask who fact-checks the fact-checker
TLDR: Bullshit Detector is a new tool that checks viral posts and videos line by line and gives them a nonsense score. People loved the idea, but the comments immediately turned into a trust battle over whether a fact-checker can really be trusted to judge the truth.
A new project called Bullshit Detector is promising to do what most of us only threaten to do: take a viral video, article, tweet, or PDF and check it claim by claim instead of just trusting the loudest person with the flashiest thumbnail. The pitch is simple enough for anyone to get: paste in a link, and your AI helper spits back a report showing what seems true, what looks shaky, and a big fat BS score from 0 to 10. It even boasts that an AI helped build the tool that now audits AI hype, which is exactly the kind of deliciously self-aware twist the internet lives for.
But the real show started in the comments, where the crowd immediately split into "cool idea" and "absolutely not, explain yourself" camps. One person cheered it on with a breezy "Awesome idea," while another instantly dragged the setup as way too much effort, saying this should have been a simple browser add-on that works locally. Then came the philosophical knives: what counts as the truth, exactly, if the tool relies on "independent sources" that may disagree with each other? And the sharpest jab of all was pure irony bait: one commenter claimed the creator's own replies looked like AI slop, which is about the most savage accusation possible under a post about spotting nonsense online. The whole thread quickly became less about the detector and more about the eternal internet question: who checks the checkers?
Key Points
- •Bullshit Detector is a portable set of agent skills that fact-checks online content by verifying extracted claims with sources and assigning a 0–10 BS score.
- •The tool can be installed either through the skills.sh workflow for editable local copies or as a read-only Claude Code plugin that stays updated automatically.
- •The project supports multiple agent environments, including Claude Code, OpenAI Codex, OpenCode, Cursor, and Gemini CLI, with partial support in some chat interfaces.
- •A companion script called fetch-content converts URLs into normalized text and metadata from sources such as YouTube, TikTok, articles, PDFs, and tweets without requiring API keys.
- •The article emphasizes separating deterministic fetching/parsing from agent reasoning so new content sources can be added without changing analysis logic.