쉐도잉 연습: I Tested The Viral Reverse Engineering Repo - 영상으로 영어 말하기 배우기

연습 모드 선택

각 모드는 전체 화면 연습실로 열립니다.

182 문장

문장이 너무 짧거나 길면 Edit를 눌러 조정하세요.

I want you to look at these two apps on my screen.

Now, one of them is a real production application called MongoDB Compass, which is built by MongoDB, is open source and is pretty popular.

Now, another one was a clone of this that I built with AI in 30 minutes and cost me $13.

Now, I built this clone using reverse engineering anything, which I'm going to talk about in this video, but right now, honestly, try to guess.

You can see one app on the side here.

I can click into different database fields, right?

I can see all the different entries, same one over here.

And to be honest with you, even though I know the answer, it's almost impossible for me to tell.

So can you figure out which one is which?

Well, on the right -hand side, we have the clone that I built with AI.

And on the left -hand side, we have the actual MongoDB Compass application.

Now, again, this took me 30 minutes to clone and just $13 in AI credits using a very popular package.

Now we're plug -in called Reverse Engineering Anything, which I want to get into.

So this is the GitHub repo for Reverse Engineer Anything.

Now it's been going viral over the past few days here.

It has over 344 ,000 stars, and basically it does what it says.

It allows you to reverse engineer anything using AI.

And I'm going to explain exactly how it works, but to install it is literally as easy as just copying this GitHub URL,

going to your agent and telling it, Hey, I want you to install reverse engineering anything, add all of the tools to cloud code,

and then it will add the tools and you can start using it to make clones like I just showed you.

So before we go into the details, let me start with how re or reverse engineer anything actually works.

Now when you install an app, you just get the finished product, you don't get the source code.

Now reverse engineering is the process of working out how that finished product works and the tools for doing that, things like decompilers and debuggers,

have been around for years now.

Now they're just really slow and they're pretty difficult to use by hand if you don't know what you're doing.

Now what re does is it sits in the middle.

So on one side, you have your coding agents like CloudCode, Codex, Cursor, whatever.

On the other side, you have these real reverse engineering tools like Gidara, EDA, Frida, GDB, whatever.

I'm probably mispronouncing them because I don't even know how to use them, but I don't have to because I now can use this tool.

Now, basically, you can add them into CloudCode using things like MCP servers, and then we can run all of these on your own computer,

do everything locally, decompile, understand the application, and give you back answers.

And every answer comes back with evidence.

So what it found, where in the app it found it, which tool found it, and what's still unknown.

So the agent doesn't need to guess.

It can actually rebuild something by looking at the exact way that it was already built.

Now, this is really interesting because you don't just need to do this on web apps.

You can do this on really any type of application, even desktop apps that typically are pretty difficult to reverse engineer.

So what can the agent actually do with re?

Well, it gets 207 different tools.

Now, the biggest group, which is 79 of them, is for compiled programs.

So things like reading decompiled code, searching for text, and finding who calls what.

Then there are tools for Android apps, websites, and Electron apps.

There's also live debugging with Frida and GDB, and a whole group for tracking evidence and checking that a rebuild actually matches the original.

Now, one tip if you want to try this here is

that those 207 tool descriptions can add up to around 270 ,000 tokens with every single message.

Now in Claude code, you're actually able to set an enabled tool search to true.

I'm going to show you how to do this in one second.

And that will mean the tools only load when the agent actually needs them.

Now for me that cut each message to about 28 ,000 tokens, which makes a massive difference when you're running this at scale.

Let me quickly show you.

So I'm just going to tell Claude code, this set enabled tool search to true for your tool calls.

And now this way it will start searching through the tools

and figuring out

which one it needs to use rather than looking at all of them every single time it calls an AI model.

So you can see it was able to update this year in our settings.

Now, what I want to do is show you how I ran this,

because again, this is the cloned application that we got here using re which is just absolutely crazy.

And I want to dive more into the mechanics of how this actually works.

So you understand where it will be useful for you.

Okay.

So to run everything that you're about to see,

I actually use today's sponsor open router now it's a really good fit for this because reruns can get extremely expensive.

Now the agent actually loops over and over again

and I've seen people spend you know hundreds of dollars on a single rebuilt product.

So what I wanted to do was compare an expensive model with a cheap one

which I'm going to show you the results of in a minute

and to do that I used open router

which gives me one API key and one endpoint for covering over 400 different models.

Now I simply just pointed cloud code at open router by changing two simple environment variables.

And then after that, I was able to switch to different models and see all of the pricing directly in my dashboard.

So you can see that every request actually shows this exact cost here in the activity page.

And that mattered a lot for this video because Claude Code's own estimate really overstates the cost.

And in this example, it overstated it by up to four

and a half times what it actually cost me to run this.

So every number that you see in this video comes directly from OpenRouter.

And even if a model gets rate limited halfway through a long run,

I can have a fallback model and move over to something else that my program doesn't completely stop.

If you want to try Rhea for yourself, OpenRouter is linked in the description.

I use it for pretty much anything where I need an API key.

And I have a ton of usage.

It's a great platform and really the best way to use AI models.

You also can call image, video, voice, everything from one single API key.

All right, so once you've got Rhea installed inside of CloudCode, you can simply just ask it to use Rhea

and rebuild pretty much any app that's on your computer where they can access from the web.

Now, if you go into your terminal -based environment to use CloudCode, you'll be able to see the MCP server.

So if you type

Cloud here, just go, yes, I trust this folder, and then type slash MCP,

you'll see the Rhea MCP server shows up here with 207 different tools.

Now, it also has some prompts and some other stuff inside of here, and basically it's ran through this kind of NPM package.

However, if you try to look at the MCP server from the desktop application, after you add it, you probably won't see it.

That's fine.

It's still there, but it will just appear when you start asking it to do something.

So if I say, Hey, rebuild the OBS application using reverse engineer, anything.

Then if I press enter here, it should find that connector.

So you see Rhea pops up right here and it will start building that out.

Now I won't wait for all of this to finish because I already built MongoDB Atlas,

but basically that's as easy as it is to start doing this.

So now I want to discuss what Rhea is actually doing behind the scenes, specifically when it's looking at compiled programs.

So things like a Windows executable file like OBS.

100

So first what the agent will do is open the binary and try to get an overview of the program.

101

So what kind of file is it?

102

What libraries it uses?

103

Where does it start?

104

Then it needs a starting point.

105

So it's going to search for text that you can actually see in the application.

106

It's like a button, a label, an error message, and it's going to jump to the code that uses that reference.

107

Then it reads that function.

108

So the decompiler is going to turn machine code back into C -like pseudocode, which the model will actually be able to read.

109

Then it's going to follow all of the calls.

110

So what calls this function?

111

What calls that function?

112

Where is the data flowing throughout the application?

113

Now, anything that it can't prove, it's going to write down as an unknown instead of just guessing.

114

And at the end of it, it's going to compare the rebuilt code against the original to check if they behave the same way.

115

Now for compiled programs, you will need a decompiler installed.

116

However, it can automatically install it for you if you just put it in bypass permissions mode.

117

Now Compass, the one that I did here, is actually a JavaScript app.

118

So it skips the decompiler because the code ships as JavaScript instead of plain machine code.

119

So for MongoDB Compass, which Claude was able to remake, I pointed Rhea at a single 49 megabyte file.

120

Now this is basically the entire app that just packaged together

121

and in under three minutes it was able to map the entire thing.

122

So it had 530 different files, over 3 ,500 code modules

123

and over 13 ,000 findings including how the app's windows actually talked to the back end

124

and how all of the communication occurred.

125

So here's what these models actually built and I say models because I ran this with two models,

126

Opus 5 .5 and Haiku 5 .5 to see what actually was the difference with an expensive, and a cheap model.

127

Now, here's exactly what happened.

128

Well, first we set everything up, right?

129

Then I gave both models the exact same prompt.

130

So I said, use RIA to figure out how

131

Compass works, how it's styled, and then rebuild it as a desktop application with these eight features.

132

This is what I included in the prompt.

133

So I wanted the ability to connect a database sidebar, browsing documents with filters and pages, inserting, editing, and deleting, schema the tab,

134

aggregations tab, and then the indexes tab.

135

And I wanted it to look exactly the same as Compass, as you guys already saw.

136

I also gave it a small test database that it could load up here that I already had built with MongoDB.

137

So here's the results that I actually got that I had Claude's screenshot.

138

So you can see Opus 5 .5 costs $13.

139

Haiku 5 .5 costs $0 .85.

140

With Opus, it took 42 minutes.

141

With Haiku, it took about 15 minutes.

142

And you can see that they look pretty similar here.

143

Now Haiku has some visual differences compared to the real compass version.

144

You can see it's a little bit simpler.

145

The icons aren't quite there, but generally it gave me pretty much the exact same stuff.

146

Just again, like a little bit of small issues with kind of the spacing and the indexes, and it wasn't quite as clean,

147

but you get the idea, right?

148

We're able to build the entire thing.

149

And when we used Opus 5 .5, it gave us a near identical clone.

150

Again, just to open this up and show you, like I can go to aggregations, I can run queries, I can view specific schemas,

151

I can look at the index.

152

I can connect to different databases.

153

Like, like how crazy is that, that this actually just works.

154

I can directly even like edit a value here and change this and save it.

155

And it works properly.

156

And this is just all done with AI with literally two or three prompts.

157

So with all of that said, what does this actually mean?

158

Well, to be honest with you, I don't really know.

159

I mean, this is super cool.

160

It's obviously very interesting that we can do this.

161

And it really shows

162

that the value of software is no longer just in writing a bunch of code like it used to be.

163

It's really more in making something that's legitimately valuable that solves a problem and then people actually want to use.

164

This leads me to believe that more and more software will just de facto become open source.

165

And the companies are probably going to be less protective of trying to guard all of their secretive code bases.

166

Because really, there's just less value in doing that now than there used to be in the past.

167

I can remember when code was sacred

168

and you wouldn't want to share certain modules or files

169

or what you're doing because you don't want anyone to get in on your secret

170

and rebuild this really technically complex app that you made.

171

Now we literally have an AI tool that can rebuild pretty much anything in a matter of few minutes and even decompile,

172

go through, run all of the compiled code.

173

It's just crazy where we're at now in 2026.

174

So I think it shapes what we use software for, what the value of software is and what we're going to look at going forward

175

but I don't believe this wipes out software engineers.

176

And to be honest, the more cool tools we get like this, the more interesting software we can build.

177

And again, in my opinion, the value of software is knowing what to build, not just building it itself,

178

having that judgment, building a nice UI, solving a legitimate problem.

179

Now, more than ever, the code is not that valuable, but being able to know what to code and having that judgment is becoming more valuable.

180

That's my opinion.

181

I don't know if you guys agree with me.

182

Let me know in the comments down below, and I will see you in another video.

앱 다운로드

유창하게 말하는 데 필요한 모든 것

AI 발음 평가문장마다 점수 확인

IPA 발음 연습모든 소리 마스터

단어 학습나만의 단어장 만들기

단어 게임놀면서 외우기

QR 스캔

Download on theApp Store Get it onGoogle Play

같은 채널

FOR YOU

이 레슨의 어휘와 말하기 포인트

이 말하기 레슨은 영상 “I Tested The Viral Reverse Engineering Repo”을(를) 바탕으로 합니다. 가장 자주 반복되는 단어는 다음과 같습니다: code, tool, model, app, reverse. 이 영상에는 섀도잉할 문장 182개와 단어 2437개가 있습니다. 말하는 구간의 길이는 11:23입니다. 화자는 분당 약 214단어로 빠르게 말하므로 연음과 약하게 발음되는 소리가 많습니다. 단어의 85%가 영어에서 가장 많이 쓰이는 3,000단어에 속합니다. 나머지는 연습 전에 미리 확인해 두세요.

이 영상의 핵심 어휘

영상에 나오는 익혀 둘 만한 단어 15개를 발음, 뜻과 함께 정리했습니다.

단어	발음	뜻
rebuild 동사	/ɹiːˈbɪld/	재건하다
compass 명사	/ˈkʌmpəs/	나침반, 라침반
clone 명사	/kloʊn/	클론
server 명사	/ˈsɝvɚ/	서버, 봉사기
database 명사	/ˈdeɪtəˌbeɪs/	데이터베이스
install 동사	/ɪnˈstɔl/	설치하다, 깔다
prompt 동사	/pɹɑmpt/	부추기다
cloud 명사	/ˈklaʊ̯d/	구름
package 명사	/ˈpæk.ɪd͡ʒ/	짐, 꾸러미
desktop 명사	/ˈdɛsktɒp/	탁상
compare 동사	/kəmˈpɛɚ/	비교하다, 비유하다
solve 동사	/sɒlv/	해결하다
connect 동사	/kəˈnɛkt/	잇다, 연결하다
edit 동사	/ˈɛdɪt/	편집하다
module 명사	/ˈmɑd͡ʒul/	모듈

영상에 나오는 구동사

단어	뜻
wipe out 동사	무찌르다
write down 동사	적다, 적어 두다

따라 말해 볼 만한 문장

영상에 나오는 짧고 완결된 문장으로, 일상 대화에서 그대로 쓸 수 있습니다:

You can see it's a little bit simpler.
We're able to build the entire thing.
It's obviously very interesting that we can do this.
I don't know if you guys agree with me.

이 영상의 문법

화자가 가장 많이 쓰는 문형을 영상 속 실제 표현과 함께 정리했습니다.

문형	영상 속 표현
“used to” used to + 동사 — 지금은 아닌 과거의 습관이나 상태	used to be
수동태 be + 과거분사 — 누가 하는지보다 무슨 일이 일어나는지에 초점	is built · was already built · is linked
관계절 who / which + 절 — 사람이나 사물에 대한 추가 정보	Compass, which is · group, which is · tokens, which makes
현재완료 have/has + 과거분사 — 과거의 일이 지금도 관련이 있을 때	have been · I've seen

주의할 발음

화자는 don't, I'm, you're 같은 축약형과 약화된 형태를 38번 사용합니다. 들리는 대로 짧게 발음하세요.

“sh”와 “zh” 소리: screenshot /ˈskriːn.ʃɑt/, aggregation /æɡɹəˈɡeɪʃən/, dashboard /ˈdæʃˌboɹd/
긴 단어 — 강세 위치에 주의: legitimate /lɪˈd͡ʒɪt.ɪ.mət/, technically /ˈtɛk.nɪˌkəl.i/, decompiler /diːkəmˈpaɪlə(ɹ)/, aggregation /æɡɹəˈɡeɪʃən/, executable /ˈɛksəkjutəbl/

한국어 화자가 어려워하는 소리:

/f/ — ㅍ(/p/)으로 바꾸지 말고 윗니를 아랫입술에 대기: filter /ˈfɪltɚ/, halfway /ˌhæfˈweɪ/, folder /ˈfoʊldɚ/, fallback /ˈfɔːl.bæk/
/v/ — ㅂ(/b/)과 구별하기: reverse /ɹɪˈvɝs/, server /ˈsɝvɚ/, solve /sɒlv/, protective /pɹəˈtɛk.tɪv/, dive /daɪv/
/r/과 /l/ 구별하기: rebuild /ɹiːˈbɪld/, variable [ˈvæɹ.i.ə.bl̩], viral /ˈvaɪɹəl/, decompiler /diːkəmˈpaɪlə(ɹ)/

이 영상으로 연습하는 방법

먼저 말하지 않고 영상을 끝까지 듣고 모르는 단어를 적어 둡니다.
0.75배속으로 한 문장씩 섀도잉을 시작하고, 익숙해지면 보통 속도로 돌아갑니다.
자신의 목소리를 녹음해 원본과 비교하고, rebuild, compass, clone 같은 단어에 특히 주의합니다.

쉐도잉이란? 영어 실력을 빠르게 키우는 과학적 방법

쉐도잉(Shadowing)은 원래 전문 통역사 훈련을 위해 개발된 언어 학습 기법으로, 다언어 학자인 Dr. Alexander Arguelles에 의해 대중화된 방법입니다. 핵심 원리는 간단하지만 매우 강력합니다: 원어민의 영어를 들으면서 1~2초의 짧은 지연으로 즉시 소리 내어 따라 말하는 것——마치 '그림자(shadow)'처럼 화자를 따라가는 것입니다. 문법 공부나 수동적인 청취와 달리, 쉐도잉은 뇌와 입 근육이 동시에 실시간으로 영어를 처리하고 재현하도록 훈련합니다. 연구에 따르면 이 방법은 발음 정확도, 억양, 리듬, 연음, 청취력, 말하기 유창성을 크게 향상시킵니다. IELTS 스피킹 준비와 자연스러운 영어 소통을 원하는 분들에게 특히 효과적입니다.

섀도잉 방법: 단계별 전체 가이드 읽기 →

쉐도잉 연습: I Tested The Viral Reverse Engineering Repo - 영상으로 영어 말하기 배우기

연습 모드 선택

앱 다운로드

같은 채널

Claude Code: The Advanced Guide (99% of Devs Skip These Features)

Learn C++ With Me #19 - Sets

How to learn Machine Learning like a GENIUS and not waste time

How To Pass Coding Interviews Like the Top 1%

추천 영상

Dear Higgsfield, I reverse engineered you in 12 hours

Why React Native apps make all the money

Did Anthropic Kill Git?

How I deploy serverless containers for free

JUST $1 For Billions of Tokens?! (DeepSeek v4 Pro)

The Engineering that Runs the Digital World 🛠️⚙️💻 How do CPUs Work?

STOP Running Code On Your Computer

9 Insane Claude Mods I Already Can't live Without

OpenAI’s lead engineer was secretly a spy...

GitHub Copilot now controls your command line...

Your AI Agent Starts Every Session Like a New Hire - Killian Carlsen-Phelan, Sonar

My 4.9-Star iOS App Is Actually a Website: Here’s How I Built It