REAL

Holds up: asking Claude Code to check a vibe-coded app for server-side request forgery is a real, high-value security check, and the plain-English prompt in this post is a good one

Threads · Oct 2026

The claim'If you're vibe coding an app or website, ask Claude Code to check for server-side request forgery (SSRF).' Features like link previews or 'import photo from URL' make your server open a link for the user; a scammer can paste your server's own private address instead, and your server hands back what's inside, 'including the keys to your whole cloud account.' Suggested prompt: 'Can a user make my server open any link they want?'

This one holds up, and it is the rare security tip that is both accurate and useful to a non-engineer. Server-side request forgery is a real, well-documented class of bug with its own category in the OWASP Top 10. It shows up exactly where the post says: any feature where your server fetches a URL the user typed, such as link previews, 'import image from URL', webhooks or PDF generators. An attacker swaps in an internal address. On many cloud servers the classic target is the metadata endpoint at 169.254.169.254, which can return temporary credentials for the server's cloud role. That is not hypothetical: the 2019 Capital One breach began with this pattern and exposed over 100 million records. The post's framing is also smart for its audience. People building with AI tools rarely know the acronym, but they do know whether their app has an 'import from URL' button, and asking the agent a plain question about that feature gets a better review than 'check my security.' The only caution is that an AI review is a first pass, not a guarantee. The fix has to block private and metadata addresses, follow redirects safely, and be tested, and a human should confirm it.

What holds up

The post was read in full on Threads: a short tip with no product, link or comment-for-the-link hook; replies add a sensible suggestion to ask for a regression test with any fix.
SSRF is a recognized vulnerability class with its own category in the OWASP Top 10 (2021), and link previews and URL importers are standard entry points in published write-ups and training labs.
Cloud instance metadata at 169.254.169.254 returning role credentials is the documented high-impact target; AWS IMDSv2 exists specifically to blunt naive SSRF.
The 2019 Capital One breach is the standard public example of SSRF leading to cloud credential theft and mass data exposure.

What doesn’t

No product is being sold; this is a free tip, which is a good sign.
'The keys to your whole cloud account' is the worst case; real impact depends on what permissions the server's role has.
An AI code review is a first pass. It can miss redirect tricks and DNS rebinding, so the fix needs testing.

The catch

There is no catch in the claim itself. The risk is stopping at the question. An agent saying 'looks fine' is not proof, and a half fix that blocks 169.254.169.254 but follows redirects to it is still exploitable.

How to actually do it

Ask your coding agent: 'List every place my server fetches a URL that a user can influence, and show me the code for each.'
For each one, require an allow-list or a block on private, loopback and link-local addresses, checked after DNS resolution and on every redirect.
On AWS, require IMDSv2 and give the server role the fewest permissions it needs, so a slip leaks less.
Ask the agent to add a regression test that tries http://169.254.169.254 and a redirecting URL, and rerun it whenever that feature changes.

Following these steps closes one of the most damaging and most overlooked holes in quick-built apps. It costs an afternoon, and the regression test keeps it closed as the app changes.

Confidence: High
Posted by: a vibe-coding creator on Threads, post at about 16K views; post sent to Buddy Tue Oct 13, 2026, evening

See the original claim →

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.