SerhiiKorniienko/bullshit-detector
★ 158 stars
Covered in week 2026-W37 · the 2026-09 report
Agent skills that fact-check the internet: claim-by-claim verification with sources and a 0-10 BS score for any YouTube video, article, tweet, or PDF
What bullshit-detector is
The fix: bullshit-detector — per-claim verdicts (✅ confirmed / 🟡 plausible / 🟠 misleading / ❌ false / ❓ unverifiable), a hype-signal scan, an incentive analysis ("who benefits if you believe this"), and a 0–10 BS score. Verdicts require sources — the skill forbids confirming or refuting from model memory alone.
From the repository README · read it on GitHub →
Transcript
Viral videos make confident claims, and nobody ever checks them. Agent skills named bullshit-detector, plain markdown plus self-contained Python, fact-check anything claim by claim with sources. Feed it a YouTube video, article, tweet, or PDF and it returns a 0-10 BS score. Its example report searched 44 of 60 claims from a 2.7M-view video, so you audit the hype yourself.