Vronzu

AI, money, and the experiments in between.

[Vronzu @Vronzu]

@coolerwave AI finding bugs before attackers do is the one good part of this story

[Vronzu @Vronzu]

@Rorkix Switching USB debugging back off at the end is a detail that matters

[Vronzu @Vronzu]

@Ravlizk Open-sourcing the whole setup is what makes this valuable

[Vronzu @Vronzu]

@Flinzu Plain Markdown folders beating evety note app, i like this

[Vronzu @Vronzu]

It is becoming f*cking dangerous: half the AI demos you liked this week are probably lying to you. And nobody checks.

A tool just fixed that. It is called REA. Open source, plugs into Claude Code, Codex, Cursor and Gemini CLI. Nobody had heard of it three months ago. Now it has 100,000+ GitHub stars.

Here is why. Every claim the AI makes comes tagged as an observation, an inference or an unknown, with the address in the code behind it. No "trust me". Show the evidence or admit you don't know.

Then it gets better. People used it to rebuild a Touhou game: 697 functions, 696 exact matches. An authenticator app with 3,077 classes, all 133 tests passing. None of it was a cracked license. Just behavior recovered and rebuilt in new code.

But what got me is step one. Before touching anything, Claude has to say what kind of app it is. If it's Electron, the JavaScript is already sitting on your machine, readable.

And the last step is the one almost nobody does. The rebuild only counts when it passes its own tests. A match rate, not a feeling.

8 steps, from a sealed app to a rebuild you own. Study the original, ship your own.

Watch the demo below before you believe another AI demo.

[Hanako @hanakoxbt]

REA is the most dangerous repo on GitHub right now

It gives your AI agent a real disassembler, so it reads any app down to the binary

Every claim comes back with the address behind it

I broke down the exact 8 steps: https://t.co/eaeZ4OFJrH

[Vronzu @Vronzu]

@DeItaone Google Cloud marketplace changes the game

[Vronzu @Vronzu]

@danilokrasko21 Imagine running this on every app and discovering how much hidden logic we never notice.

[Vronzu @Vronzu]

Your TV is an ad machine you paid for. A guy just fired it and gave the screen to his apartment.

He vibe coded his own TV OS with Claude Code. No ads. In their place, a live 3D model of his apartment. His lights and ceiling fan are wired in, and they turn on inside the 3D at the exact moment they turn on in real life.

Then it gets insane. The TV inside the 3D model shows the live TV.

But what got me is where it starts. Not with 3D. With one boring afternoon of cleanup. Claude Code finds the junk on your exact TV, flags the tracking apps first, and switches them off. Disabled, never uninstalled, so you can undo everything.

Then the screen is empty, and you decide what goes there. A clock. Your tasks. Your own apartment.

Here is the prompt I would start with:

List every app on my TV that shows ads or tracks my screen. Flag the tracking ones first. Ask me before you disable anything. Disable, never uninstall, and keep an undo log. Then build me a simple home screen with a clock and my tasks, and install it before you turn the old launcher off.

Don't just read this. Watch the animation below.

[Gathin (Ø,G) @gathint]

Vibe coded my parents' TV this afternoon!

It was showing a sponsored ad before you could even get to Netflix.

Gave Claude Code the model number, it found the TV on wifi, removed 25 Xiaomi packages and wrote a new launcher.

Looks like an Apple TV now lol

[Vronzu @Vronzu]

@danilokrasko21 Exactly. The ads are the part you can see. The tracking is the part you can’t.

[Vronzu @Vronzu]

Nobody told you your TV can be cleaned in one afternoon. Now 5M people know.

And before Netflix even loads, it pushes a sponsored ad you never asked for.

A guy decided to end it. He gave Claude Code his TV model number. It found the TV on wifi, switched off the junk apps and installed a clean launcher. One afternoon. That post passed 5M views in two days.

Then it gets better. The repo behind it now has 4,000+ stars and tested lists for about 30 TV models.

But what got me is the rules. Nothing gets uninstalled, only disabled, so every change can be undone. Ten apps at a time, then test the remote, HDMI and sound. New home screen first, old one off second. No root, no flashing.

Samsung or LG? Different trick. Change the TV's DNS in network settings. No hacking, just a normal setting.

Watch the video below before you touch your TV.

[Gathin (Ø,G) @gathint]

Vibe coded my parents' TV this afternoon!

It was showing a sponsored ad before you could even get to Netflix.

Gave Claude Code the model number, it found the TV on wifi, removed 25 Xiaomi packages and wrote a new launcher.

Looks like an Apple TV now lol

[Vronzu @Vronzu]

@Drundor Nice share mate

[Vronzu @Vronzu]

@andreysuperior Reverse engineering is eye-opening.

[Vronzu @Vronzu]

Why is this even allowed?

A guy used Claude Code to reverse-engineer his parents’ TV over Wi-Fi, remove 25 packages and build a clean home screen. No more sponsored ad before Netflix.

One afternoon. No engineer.

The tool behind this bigger shift is REA (Reverse Engineer Anything). It gives AI coding agents a real disassembler, maps how software works, traces features down to specific functions, and provides evidence for every conclusion.

You can start using REA yourself tonight:
→ Pick an app you want to understand
→ Let REA map its structure and identify key functions
→ Trace a feature, rebuild it from scratch, then verify the result against the original

The trick is the order. REA helps you go from a black-box binary to an evidence-backed blueprint you can actually build from.

Try REA tonight. Then tell me what you managed to uncover.

[Hanako @hanakoxbt]

REA is the most dangerous repo on GitHub right now

It gives your AI agent a real disassembler, so it reads any app down to the binary

Every claim comes back with the address behind it

I broke down the exact 8 steps: https://t.co/eaeZ4OFJrH

[Vronzu @Vronzu]

@Drundor Back up first. Good reminder.

[Vronzu @Vronzu]

@Plutkar Great weekend project for family tech support.

[Vronzu @Vronzu]

@vayrenrose Research was always the real moat.

[Vronzu @Vronzu]

@0xDeniAi Free model plus opus is a smart split

[Vronzu @Vronzu]

The person who built REA just handed every curious user a superpower. I’m still not over what it does.

It plugs into Claude Code and opens any app on my computer, then shows how a feature really works, with the file and function behind every answer.

That part is wild. Then it gets better.

It doesn’t stop at the explanation. It writes down the behavior, builds my own version from that description alone, and tests it against the original.

So the loop is read it, describe it, rebuild it, prove it, and none of the original code gets copied.

But what got me is how it treats its own answers. An agent can sound completely sure and still be wrong, so every claim comes back marked as proven or still a guess.

I’m picking one app I use every day and pointing it at one feature this week.

Be honest. Are you going to try this tonight, or wait for someone else to do it first?

[codila @0xCodila]

REA is most dangerous thing on the internet right now...

It lets your AI agent reverse engineer any app, down to the binary

If you use it correctly, you can rebuild any feature you pay for, for FREE

In this article, I show you how: https://t.co/3P1FNXiXzl

[Vronzu @Vronzu]

@DarkFactorr Two rooms is the cleanest framing.

[Vronzu @Vronzu]

@mika_systems Reopening Level 7 with the same three keys is the real test