← The AI Hype Audit — all 645 verdicts
REAL
Holds up: the order-link leak in this skit is a real and common bug, and Cloudflare's free security-audit skill really is MIT-licensed, uses a skeptic agent, and its README says one run finds about half of what repeated runs find
The claimA customer changed one digit in her order link and saw a stranger's name, phone and address; Claude wrote the page and also reviewed it, giving itself top marks. The fix: Cloudflare's free, MIT-licensed security-audit skill (cloudflare/security-audit-skill), which began as Cloudflare's in-house vulnerability hunter. Ask your agent to 'security-audit this codebase'; it maps routes, puts agents on inputs, then a skeptical agent argues against every finding. One run finds about half of what repeated runs find. Comment AUDIT for the prompt.
The technical lesson is right. Changing an ID in a URL and seeing someone else's record is a textbook access-control flaw (often called IDOR), and the skit's code frame shows the cause plainly: the route looks up the order by ID and never checks that it belongs to the logged-in customer. The point that the same AI that wrote code is a weak reviewer of it is fair; separate review catches more. The tool is real and described accurately. cloudflare/security-audit-skill is MIT licensed, has about 73k GitHub stars, and its README describes reconnaissance of routes and inputs, isolated hunter agents, and a fresh verifier that tries to disprove each finding. The README calls it the seed of Cloudflare's vulnerability-discovery harness and says, in their tests, a single run found roughly half of what repeated runs found, so the 'run it more than once' advice comes straight from the source. Minor stretch: hunters are assigned per coverage unit, not strictly one per input. The skit itself (5:58 AM email, 'offered her a job') is fiction. What is being sold: a comment-for-DM hook ('Comment AUDIT') for a prompt that is shown on screen and is essentially the skill's one-line trigger.
What holds up
- Read cloudflare/security-audit-skill on GitHub: MIT, about 73.3k stars, reconnaissance, isolated hunters, fresh verifier per finding.
- README quote: 'a single run found roughly half of the vulnerabilities that repeated runs found in total.'
- Contact sheet shows the vulnerable route (db.orders.find(id) with no owner check) and the fix ('owner check added').
- x-ref #304 and #467 for related vibe-coding security checks.
What doesn’t
- Comment-for-DM hook for a prompt already shown on screen.
- Skit story is fiction; the tool and the bug class are real.
- An audit finds bugs; you still need tests that prove each route checks ownership.
The catch
Accurate tool, accurate bug. The only thing held back behind 'Comment AUDIT' is a prompt the video already shows you.
How to actually do it
- Install the skill: npx skills add https://github.com/cloudflare/security-audit-skill --skill security-audit, then ask your agent to security-audit the codebase.
- Run it at least twice before each launch and fix confirmed findings first.
- For every route that takes an ID, add a check that the record belongs to the logged-in user, and write a test that tries someone else's ID.
A skit that teaches a real bug and points to a real, free tool with honest caveats taken from its README. The DM gate adds nothing you cannot get from GitHub.
- Confidence
- High
- Posted by
- an animated AI-dev skit page (about 6,800 plays); reel sent to Buddy Mon Nov 9, 2026, 7:00 AM
We test hype for free. We build the real thing for a living.
Thirty minutes, no pitch — and you'll leave with something useful either way.
Book a call with Todd or start with the free Business Checkup →The Verdict Weekly
Three verdicts every Friday. Free forever, unsubscribe anytime, no spam — that would be ironic.
© Schreier Group · schreiergroup.com · See a wild AI claim? Drop it here and we'll test it.