Open-source tool REA gained 81000 stars in a matter of days and can plug into Claude Code for reverse engineering software.

Original Little G 2026-10-11 17:5 Guangdong

Whenever I see a really handy feature in an app or piece of software someone else built, I can't help wondering how they made it.

But most tools aren't open source, so it's hard to figure out exactly how they're implemented — you can only copy the surface, and it's tough to replicate the essence.

Until recently, when I found an open-source project on GitHub that solves this problem:REA.

Just days after going open source, it shot to the top of the GitHub Trending list, gaining 81000+ stars.

Its full name is Reverse Engineer Anything, which translates to "reverse engineer everything" — a pretty bold claim.

AI blogger Matthew Berman's take on it: Software is done.

It's definitely gotten people's attention. Let's take a look at what this tool can do and what it's good for.

Simply put, REA provides a complete reverse-engineering workflow that can plug into agent tools like Claude Code and Codex.

Once connected, we just ask questions in plain language, and the AI strips a closed-source piece of software down to its underwear.

When you ask how a certain feature in the software is implemented, it explains it in full detail — and can even build something similar.

The "reverse engineering" here is a bit like having no design blueprints, only a finished product, and working backward to figure out how it was made.

The project's official site gives several demo cases, all things we see in daily life.

Take the Windows built-in calculator: enter 200+ 10% and the result is 220 — many people have probably never wondered why.

Throw this question at the AI, and it digs into the calculator's program files and finds the code that handles the percent sign.

Then it tells us that the percent sign after a plus sign computes the percentage of the previous number, i.e. 10% of 200.

There's also the little dinosaur game in Chrome that appears when you're offline — anyone who's played it knows the dinosaur runs faster and faster.

The AI found the acceleration rule in the game code: as long as you don't hit an obstacle, the speed keeps ticking up until it hits the cap.

After finding the rule, it even built a dinosaur game with a manually adjustable speed.

Another case is the phone authenticator app Aegis: the AI unpacked its Android installation package and figured out why the six-digit code changes every 30 seconds.

Beyond these, it can also dissect and inspect Windows desktop software, web pages, and even old game installers.

I think what really made REA blow up is the wild variety of ways netizens have been using it.

One user was stuck on a feature and had the AI grind on it for an hour with no result; they used REA to analyze an existing app and got it done in 5 minutes.

Another user dissected a Steam game, and in under 30 minutes the AI turned the game back into readable, modifiable code.

The recreated game could even run directly in the Unity engine, and that post racked up 4800 likes.

Some have even turned it into a business, using REA to recreate handy little tools and selling them on Xianyu.

One blogger even paired REA with penetration testing tools, jokingly calling it the "Cyber King of Hell" kit.

Things that used to be doable only by developers can now be handed to AI with a single sentence.

Of course, the biggest fear with AI dissecting software is that it makes things up. Every conclusion REA gives comes with supporting evidence, and it notes where it's still uncertain.

And the whole analysis runs on your own computer — the software being dissected is never uploaded.

To get started, just run this one command in your terminal:

npx rea-agents setup

It'll let you choose which agent tool to connect to — Claude Code, Codex, Cursor, etc. — and after installing, just restart and you're good to go.

Then just ask in plain language, e.g. "Take a look at how the search feature in the Notes  app works, and build something similar for my project."

To dissect native programs like the Windows calculator, you'll also need a disassembly tool such as Hopper or Ghidra.

Final thoughts

Reverse engineering used to be a skill for security researchers and geeks — most people had never even heard the names of these tools.

Now with REA, product managers can crack open a competitor's product anytime to see how a feature is designed, and students get a living textbook for learning to code.

But it also makes me wonder: if any software can be cracked open by AI, how will commercial software companies respond?

Of course, a reminder here: use REA responsibly — best to stick to learning and researching software you have the right to analyze.

GitHub project address:https://github.com/morluto/rea