Show HN: Bullshit Detector – agent skills that fact-check videos and articles

Agent skills that fact-check the internet. Point your supplier astatine a viral YouTube video, article, tweet, aliases PDF — get a claim-by-claim verification study pinch sources and a BS people (0–10) alternatively of taking "10 WAYS TO MAKE MONEY WITH AI 🤯" astatine look value.

Portable Agent Skills — plain markdown + self-contained Python. They activity successful Claude Code, Codex, OpenCode, and immoderate harness that supports the skills format and has web search.

Built successful the unfastened pinch Claude Code — an AI helped build the instrumentality that fact-checks AI hype, and the example report is it auditing its ain kind.

Follow @SerhiiFounder for caller skills and fact-check experiments, aliases join the newsletter to get them successful your inbox.

1. Install uv if you don't person it (the fetch book uses it to self-resolve its dependencies).

2. Run the skills.sh installer and prime the skills and agents you want:

3. Ask your agent: "is this bullshit? <url>", "fact-check this video", "summarize <url>", "explain the portion astatine 12:30".

Prefer a managed bundle that updates erstwhile a caller type ships, alternatively of copied files you support yourself? Inside Claude Code:

Two ways to install, 2 philosophies:

• skills.sh copies the skills into your setup truthful you tin hack connected them and make them your own. Works pinch immoderate supplier (Claude Code, Codex, OpenCode, …).
• The plugin keeps them arsenic a read-only, always-current bundle — champion erstwhile you conscionable want it to activity and travel on arsenic it evolves. Claude Code only.

Pick one, not both — installing some gives Claude Code 2 copies of each skill.

Works beyond Claude Code — the skills are plain markdown + self-contained scripts. Full walkthroughs pinch caveats per aboveground unrecorded successful SETUP.md:

A finance feline pinch 1M views tells you the "only 14 ways to make money pinch AI". How overmuch of it is real? Views, accumulation value, and assurance are not evidence. The hole is boring: extract each claim, cheque each against independent sources, and people what survives. That's precisely the activity agents pinch web hunt are bully astatine and humans ne'er fuss doing.

The fix: bullshit-detector — per-claim verdicts (✅ confirmed / 🟡 plausible / 🟠 misleading / ❌ mendacious / ❓ unverifiable), a hype-signal scan, an inducement study ("who benefits if you judge this"), and a 0–10 BS score. Verdicts require sources — the accomplishment forbids confirming aliases refuting from exemplary representation alone.

Your supplier can't beryllium done a 27-minute video, and YouTube's charismatic API won't springiness you captions for videos you don't own. Same communicative pinch tweets ($100/mo API) and paywalled articles.

The fix: fetch-content — 1 book that turns immoderate URL into cleanable matter + metadata pinch nary API keys: YouTube transcripts and TikTok captions via yt-dlp, articles via readability extraction, PDFs, tweets via free endpoints. Every nonaccomplishment mode produces an actionable hint (paywall → paste, nary captions → Whisper) alternatively of a silent guess.

Ingestion and study are different jobs. Scripts do the deterministic activity (fetch, parse, normalize); the supplier does the reasoning (extract claims, search, judge). Because study skills only ever spot normalized matter + metadata, adding TikTok support 1 time touches zero study logic — and the aforesaid detector useful connected a tweet and a 3-hour podcast.

A existent tally against a 1.16M-view "make money pinch AI" video: examples/report-14-ways-to-make-money-with-ai.md.

BS score: 5/10 — existent tools, existent trends, guru math, and a chimney each 4 minutes. 12 claims verified: 4 confirmed, 2 plausible, 3 misleading, 0 false, 3 unverifiable. Among the catches: "Renaissance, D.E. Shaw, Two Sigma only waste and acquisition employees' money" (true for 1 money of 1 firm), and marketplace stats originated from the marketplace's ain PR.

And a TikTok tally — a 552K-view "our Sun has a hidden twin" video: examples/report-second-sun-binary-star.md (BS score: 9/10 — existent astronomy vocabulary stitched onto a fabricated cosmology).

Yes, TikTok useful — inquire the aforesaid way: "is this bullshit? https://vt.tiktok.com/…".

How it useful nether the hood:

1. Built-in captions first. Most TikToks vessel pinch creator aliases auto-generated captions. The fetch-content book handles this natively — TikTok URLs (including vt.tiktok.com short links) return a timestamped transcript positive views/likes/reposts, nary video download. The aforesaid point by hand:

   uvx yt-dlp --list-subs <tiktok-url> # cheque what's available uvx yt-dlp --write-subs --sub-langs "eng-US" --skip-download <tiktok-url> # drawback the .vtt

2. No captions? Whisper fallback. For caption-less TikToks and Reels there's a validated local-transcription prototype (mlx-whisper connected Apple Silicon, nary strategy ffmpeg needed — PyAV decodes the audio) graduating from skills/in-progress arsenic the transcribe skill. Use whisper-large-v3-turbo — smaller models garble words severely capable to break declare extraction.

The study broadside doesn't attraction either measurement — the detector sees normalized matter + metadata whether it came from a 7-minute TikTok aliases a 3-hour podcast (that's design rule #3).

All skills are model-invoked: you tin telephone them explicitly, and the supplier besides reaches for them erstwhile your petition fits ("is this legit?" triggers the detector).

Reason astir content. Source-agnostic — they ne'er attraction wherever the matter came from.

• bullshit-detector — Extract each claim, verify each against independent sources via web search, scan for hype signals, nutrient a study paper pinch per-claim verdicts and a 0–10 BS score.
• summarize — Structured TLDR pinch timestamped cardinal points, notable quotes, and an honorable "worth your time?" call.
• explain — ELI5 → deep-dive mentation of the contented aliases immoderate conception successful it, pinch a motto glossary and the prerequisites the original assumes.

Turn immoderate root into cleanable matter + metadata.

• fetch-content — YouTube transcripts, TikTok captions, articles, PDFs, tweets, section files. One script, auto-detects source, nary API keys.

Turn reports into shareable output.

• share — Ready-to-paste posts for X (thread/single), LinkedIn, Facebook, Reddit, Hacker News, aliases a newsletter — positive a branded image carousel: 1080×1350 PNGs for X/Instagram and the PDF that LinkedIn archive posts want.

See skills/in-progress: comparison (same taxable crossed sources — who's right?), transcribe (Whisper for caption-less TikTok/Reels — moving mlx-whisper prototype landed, SKILL.md pending), X thread walking.

I'm building these skills successful the unfastened — caller detectors, adapters, and existent fact-check reports arsenic they land.

• Follow @SerhiiFounder connected X
• Join the newsletter — a short weekly-ish email, nary spam, unsubscribe anytime

MIT