Is Your Setup Safe? One Prompt Finds Out
▶ Read as storyYou commented SAFE, so here is the prompt. Attacks are not smarter this year. They are faster, and more people can run them, because AI finds holes quicker than teams patch them. Paste the prompt below into your AI agent and it walks you through finding yours.
The prompt
Paste it into Claude Code, Cursor or ChatGPT with tools
It asks before it changes anything and never prints your secrets.
You are a careful security helper. Audit the setup I run and give me a ranked fix list. Work in these steps and show me what you find after each one.
RULES
- Read-only until I say yes. Ask before you change, install, delete or send anything.
- Never print, copy or send a secret (API key, token, password, private key). If you find one, say where it is and the first 4 characters only.
- Only look at things I own or run. Do not scan anyone else's systems.
- If you cannot check something, say so. Do not guess.
STEP 1 - INVENTORY
Ask me what I run, then confirm by looking: websites and CMS plugins, apps, packages (package.json, requirements.txt, composer.json), servers and hosting, domains, API keys and where they live, devices on my network (router, cameras, smart devices), and every AI agent or tool that has access to my files, accounts or terminal, with the permissions it has.
STEP 2 - VERSIONS AND KNOWN HOLES
For each item, find the installed version and check it against known vulnerabilities. Run npm audit, pip-audit, composer audit or the equivalent where they apply. Check the PHP version, the CMS version and every plugin or theme. Look up CVEs for anything that matters and say if a fixed version exists.
STEP 3 - WHAT IS EXPOSED
Check for: admin panels or dashboards reachable from the public internet, default or reused passwords, .env files or keys inside repos and git history, open ports, databases or storage buckets set to public, and routers or IoT devices with default logins or old firmware.
STEP 4 - ACCOUNTS
Check that two-factor is on for email, domain registrar, hosting, GitHub, payments and any AI account. List unused API tokens and old access, and who else has access to what. For anyone who can deploy or push code, check that their laptop has disk encryption, a screen lock and current updates.
STEP 5 - RECENT NEWS FOR MY EXACT STACK
Search the web for security advisories from the last 90 days for each item in my inventory. Summarise only what applies to the versions I run, with a link to the source.
STEP 6 - KEEP WATCHING
Propose, but do not set up yet: automatic update alerts (Dependabot or Renovate), uptime and failed-login alerts, a monthly reminder to re-run this audit, and a list of vendor security announcement pages or feeds to follow for my stack (for example Cloudflare and Vercel if I use them).
STEP 7 - RANKED FIX LIST
Give me a table: Fix today / Fix this week / Later. One line per item: what is wrong, why it matters, and the single command or click that fixes it. Put the cheapest, biggest risk reductions first. Then ask which items I want you to do for me, one at a time.
The 5 fixes to do today
Right now, before you run anything
1. Turn on two-factor for your email, domain, hosting and GitHub accounts.
2. Update your CMS, plugins and PHP version, then turn on auto-updates.
3. Delete API keys and access tokens you no longer use.
4. Check whether any admin page or dashboard opens without a login.
5. Update the firmware on your router and any camera or smart device.
Three recent hacks, one pattern
Old software, one weak link, attackers who move faster than the fix
PHP2Shell: a flaw in the software behind about 70% of websites. Researchers used AI to find and exploit a hole in PHP. Details were published before most sites had the fix, so anyone could read how to break in. NeuralTrust: AI-assisted exploitation · Recorded Future: exploitation of the flaw
Supermarket fridges hacked across Europe. Attackers got into refrigeration systems running old software that is online and that nobody will fix. Food was destroyed. Some of these systems had been exposed for months. IPinfo: the France fridges hack
Binance: $6.3B stolen through one laptop. The post-mortem points to a compromised DevOps laptop. The FBI links the theft to North Korea. The weakest link was one person’s machine, not the exchange’s code.
Keeping everything updated beats everything else on this page.
Security news worth following
Cloudflare and Vercel keep adding security features
- Cloudflare, Nov 3, 2026. New AI Gateways now require authentication, so strangers cannot send requests through yours. Older gateways keep running without it until you switch it on. Changelog: New AI Gateways require authentication
- Cloudflare, Mar 11, 2026. AI Security for Apps is generally available. It finds your AI endpoints and flags malicious prompts. Discovery is free on every plan. Blog: AI Security for Apps is now generally available
- Vercel, Nov 3, 2026. Security Review is in public beta. It scans your code for vulnerabilities on every push to GitHub and can open a fix as a pull request you approve. Pro and Enterprise. Changelog: Vercel Security Review public beta
- Vercel, Nov 2, 2026. Builds now check dependencies against a blocklist of malicious or vulnerable npm versions and fail the deploy if one is found. On for every plan. Changelog: Blocking malicious npm packages
Follow along: Cloudflare changelog · Vercel changelog
Good to know
- The prompt only reads until you say yes. It asks before it changes, installs, deletes or sends anything.
- It checks only what you own or run. Never point it at someone else’s systems.
- It cannot check what it cannot see. When it is unsure it says so instead of guessing.
Running a business? Tell me about it and I'll send back what I'd do with AI. Free.