Claude Code LEAKED — What It Really Means
12:54

Claude Code LEAKED — What It Really Means

AI Anytime 01.04.2026 404 просмотров 16 лайков

Machine-readable: Markdown · JSON API · Site index

Поделиться Telegram VK Бот
Транскрипт Скачать .md
Анализ с AI
Описание видео
Claude’s source code just leaked, and I break down what’s actually going on, what it means for AI, and why it matters. If you care about AI, this is something you can’t ignore. Drop your thoughts in the comments and don’t forget to subscribe. Source Code: https://drive.google.com/file/d/1YInhdcPRlGuGHFYzhrWzC9nCjC5cn9lW/view?usp=sharing Get the Agentic AI Master Bundle Kit: https://aianytime5.gumroad.com/l/uqmyk Build real-world AI with tutorials, tools, and research from India’s fastest-growing AI community. 👤 Creator’s LinkedIn (Sonu Kumar) Portfolio Site: https://sonukumar.site/ 🌐 AI Anytime's Website: https://aianytime.net/ 🗓️ Office Hours (AI Consulting): https://officehours.aianytime.net/ 👥 LinkedIn (Community Page): https://www.linkedin.com/company/ai-anytime/ 💬 Join Our Discord: https://discord.com/invite/4aGc9PSMgE 👤 Creator’s LinkedIn (Sonu Kumar): https://www.linkedin.com/in/sonukr0/ 🎁 Support the Channel 💸 UPI ID: sonu1000raw@ybl ₿ Bitcoin Wallet: bc1qsneqznxpzyxzzv006jthz4c8v8h5cs57myw342 ✅ Join this Channel for Perks Get access to members-only content and community perks: https://www.youtube.com/channel/UC-zVytOQB62OwMhKRi0TDvg/join #claude #claudecode #claudeai

Оглавление (3 сегментов)

Segment 1 (00:00 - 05:00)

Hey, how's it going, guys? In this video, we are going to look at some of the details from the source code which has been leaked uh on NPM, of course, on GitHub as well via map on NPM of Claude code. And uh everybody has been talking about it that Claude code source code has been leaked, and that's what we're going to look at here. And it's one of the uh one of such moments where this really going to help the open-source community because now open-source can look at this source code, and of course, we can improve, right? So, first of all, uh after Anthropic reach out to GitHub with the DMCA request, okay, which is basically reaching out for, you know, copyright violations and all, and most of the repos which are there on GitHub had been taken down from uh GitHub, of course, because of such reasons. Uh but there are some uh of course repos where you can still find it out. If you look at here, like by this uh this guy, which is there, and of course, I will also give the link in description like the source folder is available, you know, we have it in GIF folder and GIF file, and you can use that. Now, the thing is it's a great moment, okay, like people are saying, but it's I mean it depends on how you can utilize this source code, right? Uh that has been leaked out. And uh Cow Fan So, uh this guy wrote it on Twitter with a username called friedrice, okay, has first talked about it that it leaked out and all. Uh you know, uh about this uh thingy on NPM, okay, uh and then Anthropic came out, and they made this announcement that they have confirmed that Claude code source code was leaked in distribution in their CI pipeline on GitHub uh and their destination on NPM, as we talked about. And they're working with GitHub and all uh with their DMCA request, okay? Uh and that's what Anthropic said, but I still think that they might have done it on purpose. I don't know why I'm getting that feeling that they might have done it. I mean, if you look at OpenAI Codex, Codex has been uh made open-source. Like uh there are open Codex which have been uh there on GitHub. A very similar stuff like coding agent that runs in your terminal. I'm still having that feeling that this might have been done on purpose. I mean, I don't know yet, you know, you never know. These are like all such games this companies are playing to just to, you know, be in Of course, Anthropic is one of my favorite companies. Don't take it in wrong sense, but I'm just saying, you know, you never know the reality, right? Uh how can this How can they just leak like this? I mean, they have to be have multiple processes and how this CI/CD and all those things, whatever, you know, we even talk about, how this processes works in such companies. I It's of course has to be uh approved and passed through a lot of things. Might have been leaked out, you never know. There have been blunders in past as well, but this is what Claude code we have been saying. So, I have that source folder here, all the code. I'll give the link in description of this video. You can just get it through a drive link, okay? Uh this is where the entire folder is. I have created an HTML file, very simple, nothing revolutionary over here. Just want to tell you that what we can do and what can be done now once this has been leaked out. And this is going to be a very short video. It's not lengthy video, okay? Uh first of all, the entire code is in TypeScript, okay? It's uses a bun runtime uh and React Ink TUI. If you look at over here, it has 500K plus lines of TypeScript, 1,900 source files. Okay, this is fantastic, right? So, this is an overview. So, if you don't know what Claude code is right now, Claude code is Anthropic's terminal native AI coding agent. I know it runs an agentic conversations and help you do a lot of things. Like it can help you do almost everything when it comes to automating your stuff, right? So, uh thing is uh it uses as I as we know, right? It uses bun runtime. So, it uses JavaScript TypeScript runtime with Node. js compatibility, TypeScript for of course all the strict mode for all source code. So, it's been written in TypeScript. Okay. There are a lot of things that they have used. They also have used some Rust uh Rust thingy. I'll show you uh that as well, okay? And there are some issues as well that we can see. React components rendered to NC for terminal UI, your interface terminals uh when you start interacting with uh Claude code. It uses all the Anthropic models anyway, so you need to connect with their servers through some APIs, so that's what it happens. Zod v4 schema validation for tool inputs, right? Uh — [snorts] — if you're not from uh if you're from Python thingy, you can just think like Pydantic or something, okay? Uh as we know that Claude code have been using internally MCP for tool action stability, right? Uh OpenTelemetry for observability and tracing, and GrowthBook for feature flags and experiment. This is how they have built it completely. There are some code base stats, 500K plus lines of code, 1,900 source files, 40 bin 40 plus built-in tools, 20 hook

Segment 2 (05:00 - 10:00)

features, and some largest file like by lines, which is like main. TypeScript and all. And they use Axios, which is which also went through some kind of an attack, okay, recently. Uh There's a largest file that you see, print. TypeScript, CLI output rendering, streaming displays, and whatnot. Then we have message. TypeScript, uh all these files that you can go through it over here. The thing is this is not going to be extremely helpful if you're not somebody who wants to take it and work on it further. Like you cannot do a revolutionary or a magical thing just by using it as it is. Like this is bogus as it is. You cannot just use Claude code the way you use Claude code currently with this source file, to be very honest, okay? Because this is just a project without dependencies, without models, you know, uh without pipelines and whatnot, okay? To be very honest. Now, this can help you build a better system, you know, if you want to build it like Claude code, a better tool, right? Uh on that. There are some runtime dependencies that you see. Okay, Anthropic API key, that is fine, okay, for models and all that can be handled as well. Okay. The architecture overview is very simple. In the entry points, we have this main. tsx. Uh these are the entry points, then we have bootstrap layer. You can see all the bootstrap thingy. There are some conversation engine. Your queries, uh services like APIs and all. There are some orchestrations, so there is orchestration that orchestrates it. There are some tool layer, 40 plus tools, as we know, right? It can use your file, desktop, uh you know, your code bases, different MCP tools, so on and so forth. You can look at the service layer, and then utility layers like plugins, permissions. Uh there are some different permissions that you have to handle. Tech stack, we talked about it already, so I'm not going to go deep into it. You can see here, right? They use uh fuzzy search, pure TypeScript nuclear port, what Rust uh Nappy models, right? Nappy module, interesting. Uh with Rust syntax Nappy, and so on and so forth. So, there have been some huge as well. Uh you can look at here. This is how it looks like, startup sequence for latency and all. I'll just keep going down. It has read-eval-print thing, tool systems. This is fine. Uh I'll just go on stats. We already got current run architecture, startup con loop, tools. There are tools that we have. Now, this can be used to build our own such tools, as I said, right? Like let's say you want to build something like FileOps, we can just take the individual modules from this source code, right? And you start using it in our own tool that we are building, right? So, search, sale, web. It uses web search, it uses web page. Agents are there, you can see. Task, LSP. LSP is language server protocol, like how agentic IDs they use, right? VS code and all. When you hover over something, it starts showing you the documentation of that code. That happens through LSPs, right? Uh so, LSPs are there, MCPs, and everything. Okay. Security modules that you see. Uh we have security modules, multi-layered DID, depth defense in depth. You know, just for making sure that we it does not make blunder. Like recently, it deleted a database, as we now know. So, those are taken care as well. Right? Uh some states. You can see how the states are managed, two systems. Global state and app state. Okay, so this is also very interesting concept to learn. Services. They're all services, MCPs. We know MCPs, Claude. json file for your configs, okay, over there. You It has an MCP client manager that you see it over here. Sessions are handled through they persist through a JSONL file, JSON line in your Claude directory. That's how it handle. Okay, large results are written to tool. You can see it over here, interesting, kept in memory. And hooks and plugins, you know, how it has been used. Uh you can see features flags. So, features flags, uh over here. Compile time dead code elimination, entire code path removed during bundling, interesting. Okay, they use bun's features. Okay. Very interesting. I also came to know this now, okay? Agents are one of the interesting things, of course. Agents and forms that you see, they have forms, persistent background agents communicating via mailbox. Used in coordinator mode, of course, it has They have to be orchestrated, uh you know, based on your tasks and everything that is there. You can see. Now, we talk about improvements, right? So, it the code base had Rust Nappy modules replaced with pure TS ports, CPU hot path benefit. But I mean, this can be improved, okay? There has to be There has circular dependencies a lot. I was looking at the code that can get improved, okay? If you look at here, bass parser, 50 milliseconds time out plus 50K node budgets, Rust tree-sitter bass, 10 to 50x faster, fewer to complex fallbacks. That should be your P1 if you want to

Segment 3 (10:00 - 12:00)

improve this further. File index rush. So, the right now it has a pure TypeScript nuclear port. This file repo 100 ms can come to 5 ms. It can remove chunk async one get on async work around. This is interesting. Color diff rush. It will just right now have something. It will just write uh if I forgot in React something it uses for color differentiation and all. Okay. That can be eliminated as well with lazy loads hack. Uh ANSI to PNG rest. You can see it rush to make create cleaner and faster. So, there are different ways can improved. You can see Python tooling and analysis, TypeScript architectural improvements. So on and so forth right over here. Highest impact. So, they can split into a lot of circular dependency they talked about. If you see, right? Replace load as. Load as is their tool registry and so on and So, for the other priorities if you want to improve uh improvement from P0 to P3, this is how it can get improved. So, if someone if you want to improve this further, they should focus on this, right? To it to improve this. Like how this can get improved. Okay, if you want to improve it. Here's the key file map if you want to understand more about what are the file does. Like there are a lot of files over here, a lot of folders and files. And if you go through it, you'll find a bunch of things like servers, services we talked about APIs. If you look at here all the APIs cloud and whatnot. So, there's a lot of files as we know, right? Services, bass and security, then everything. So, this is how a thing is. But don't think that you can just take this and start using it. Like cloud code becomes yours. No, not at all. Okay, I mean this is something that has been leaked out. Of course, this leak is really helpful for the entire open source community. We'll be able to build better tool, of course. You know, if you want to build it. I mean but I still think that this has been done on purpose. I I'm not able to believe that this has been leaked out. The people have been talking about. But this is what I wanted to show you, right? So, I'll give you this entire code if you want. Okay, go through it like this entire source zip file. Okay, for you to test it out. Right? This is what I wanted to cover in this video, guys. You know, if you have any thoughts, feedbacks, let me know in the comment box. Let me know what you what's your views on this leaked source code of cloud code and what are you doing right now with this. Okay, and if you're building something amazing on top of it, please let us know in the comment box, right? If you like the video, please hit the like icon. If you haven't subscribed the channel yet, please do subscribe the channel, guys. That motivates me to create more such videos in your future. That's all for this video. Thank you so much for watching. See you in the next one.

Другие видео автора — AI Anytime

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Дайджест Экстрактов

Лучшие методички за неделю — каждый понедельник