[{"data":1,"prerenderedAt":632},["ShallowReactive",2],{"blog-list":3},[4,232,336,484],{"id":5,"title":6,"body":7,"date":220,"description":221,"extension":222,"meta":223,"navigation":224,"path":225,"seo":226,"stem":227,"tag":228,"thumbnail":229,"youtube_url":230,"__hash__":231},"blog\u002Fblog\u002Fhow-claude-code-actually-works.md","How Claude Code Actually Works (It's Simpler Than You Think)",{"type":8,"value":9,"toc":210},"minimark",[10,14,22,27,35,38,41,45,56,64,67,71,74,80,86,92,98,102,111,181,185,191,194,198,201,204,207],[11,12,13],"p",{},"As a developer who uses Claude Code daily, what strikes me about this leak isn't the leak itself — it's that roughly 90% of this 512,000-line codebase is defensive programming. It's essentially trying to make a probabilistic LLM behave deterministically.",[11,15,16,17,21],{},"There are also unreleased features like ",[18,19,20],"strong",{},"Kairos"," that hint at where AI coding tools are heading. Not better autocomplete — autonomous agents that dream about your code 24\u002F7.",[23,24,26],"h2",{"id":25},"how-it-happened","How It Happened",[11,28,29,30,34],{},"Apparently one of the developers included a ",[31,32,33],"code",{},".map"," file — a source map used to deobfuscate production code. When code ships to production it's obfuscated and minimized, making reverse engineering significantly harder. But with a source map file? Wide open.",[11,36,37],{},"I don't know exactly how Anthropic's deployment process works, but this map file was probably included by a developer during the npm publish step, or it was checked into Git and made its way into the CI\u002FCD pipeline.",[11,39,40],{},"Chaos ensued.",[23,42,44],{"id":43},"the-clean-room-rewrites","The Clean Room Rewrites",[11,46,47,48,55],{},"I'm not going to show the actual source code — Anthropic would DMCA strike me into oblivion. But plenty of people have already rewritten the leaked code. For example, ",[49,50,54],"a",{"href":51,"rel":52},"https:\u002F\u002Fgithub.com\u002FKuberwastaken\u002Fclaurst",[53],"nofollow","Claurst"," is a repository where Claude Code's behaviors are reimplemented in Rust using a clean room approach — written in an air-gapped environment, never directly referencing the original source.",[11,57,58,59,63],{},"There are even SaaS solutions now for \"clean room rewriting\" open source repos, where the system recreates functionality from documentation and API specs without ever seeing the original code. It's not illegal to take someone's ",[60,61,62],"em",{},"idea"," and reimplement it — you just can't carbon-copy the code.",[11,65,66],{},"As soon as the leak dropped, people started churning out implementations in Rust, Python, you name it. And you can bet competitors have already snatched ideas to improve their own tools.",[23,68,70],{"id":69},"what-the-code-actually-reveals","What the Code Actually Reveals",[11,72,73],{},"Here's what we learned about Claude Code's architecture:",[11,75,76,79],{},[18,77,78],{},"Plugin-like tool system."," There's a base tool definition — allegedly 29,000 lines of TypeScript — with around 40+ tools built on top of it.",[11,81,82,85],{},[18,83,84],{},"A query engine."," This handles all LLM API calls, response streaming, caching, and orchestration. It's the largest single module in the codebase, which makes sense — it's the brain of the operation.",[11,87,88,91],{},[18,89,90],{},"Multi-agent orchestration."," Claude Code can spawn sub-agents (internally called \"swarms\") for complex, parallelizable tasks. I've used this myself — asking Claude to kick off 15 Explorer agents simultaneously, each investigating a different part of a codebase, then reporting findings back to the main agent.",[11,93,94,97],{},[18,95,96],{},"Persistent memory system."," Everything about how it remembers context across sessions was in there.",[23,99,101],{"id":100},"the-11-step-agent-loop","The 11-Step Agent Loop",[11,103,104,105,110],{},"Someone quickly built ",[49,106,109],{"href":107,"rel":108},"https:\u002F\u002Fccunpacked.dev",[53],"ccunpacked.dev"," — a visualization tool walking through Claude Code's agent loop. Here are the 11 steps:",[112,113,114,121,127,133,139,145,151,157,163,169,175],"ol",{},[115,116,117,120],"li",{},[18,118,119],{},"Input"," — You type your prompt",[115,122,123,126],{},[18,124,125],{},"Create user message"," — Wraps your text into Anthropic's message format (standard practice)",[115,128,129,132],{},[18,130,131],{},"Append to history"," — Message gets pushed onto the in-memory conversation array",[115,134,135,138],{},[18,136,137],{},"Assemble system prompt"," — Pulls in CLAUDE.md, tool definitions, context, memory",[115,140,141,144],{},[18,142,143],{},"API streaming"," — Sends everything to the API, streams back the response",[115,146,147,150],{},[18,148,149],{},"Token parsing"," — Parses tokens as they arrive, renders live to terminal",[115,152,153,156],{},[18,154,155],{},"Tool detection"," — Identifies when the model wants to use a tool (Bash, Read, Edit, etc.)",[115,158,159,162],{},[18,160,161],{},"Execution loop"," — Collects tool results, appends to history, calls the API again. Keeps iterating until the task is complete",[115,164,165,168],{},[18,166,167],{},"Response rendering"," — Renders the final response as Markdown in the terminal",[115,170,171,174],{},[18,172,173],{},"Post-sampling hooks"," — Auto-compact if conversation is too long, extract memories, run dream mode (not yet enabled — this is the Kairos system)",[115,176,177,180],{},[18,178,179],{},"Await next input"," — Back to the read-eval-print loop",[23,182,184],{"id":183},"the-post-sampling-hooks-are-the-interesting-part","The Post-Sampling Hooks Are the Interesting Part",[186,187,188],"blockquote",{},[11,189,190],{},"Just like LLMs can only get so good given the same data and processing power, you can only teach them so much. The real improvements come from better tooling around them.",[11,192,193],{},"Auto-compacting, persistent memory, dream mode — these are the features that help Claude Code recall past conversations and keep context in check across sessions. Kairos (dream mode) isn't released yet, but it represents a shift: AI coding tools that continue working and thinking even when you're not actively prompting them.",[23,195,197],{"id":196},"what-happens-now","What Happens Now?",[11,199,200],{},"At the end of the day, Claude Code isn't magic. There's an LLM on the other side — the difference is in how it's interfaced with, the tool architecture, the memory system, the agent loop.",[11,202,203],{},"I'm not sure how Anthropic reacts to this. The logical next step seems like open-sourcing it. Unless they plan to completely rewrite the internals and change everything about how Claude Code works, everyone already knows the secret sauce.",[11,205,206],{},"Maybe people just don't care. It's not like we're all going to go write our own Claude Codes, right?",[11,208,209],{},"I guess we'll see.",{"title":211,"searchDepth":212,"depth":212,"links":213},"",2,[214,215,216,217,218,219],{"id":25,"depth":212,"text":26},{"id":43,"depth":212,"text":44},{"id":69,"depth":212,"text":70},{"id":100,"depth":212,"text":101},{"id":183,"depth":212,"text":184},{"id":196,"depth":212,"text":197},"2026-04-03","Claude Code's 512,000-line codebase was exposed via a source map leak. Here's what it reveals about how AI coding tools actually work.","md",{},true,"\u002Fblog\u002Fhow-claude-code-actually-works",{"title":6,"description":221},"blog\u002Fhow-claude-code-actually-works","Tech","\u002Fblog\u002Fhow-claude-code-actually-works.png","https:\u002F\u002Fyoutu.be\u002FeGQouwRqUzE","Yd5GmE6zU6z-gK_r0H3Y1yoJ_Bat-OX9cZa9jOZbLaQ",{"id":233,"title":234,"body":235,"date":327,"description":328,"extension":222,"meta":329,"navigation":224,"path":330,"seo":331,"stem":332,"tag":228,"thumbnail":333,"youtube_url":334,"__hash__":335},"blog\u002Fblog\u002Fjunior-devs-cant-debug.md","Junior Devs Can't Debug Their Own Code Anymore",{"type":8,"value":236,"toc":321},[237,240,243,247,254,257,264,267,271,274,281,284,290,294,297,300,306,310,313,318],[11,238,239],{},"A Reddit thread caught my eye recently: AI developer tools are making junior developers worse at actual programming. It's a bold claim, and I have thoughts.",[11,241,242],{},"Before we get into it — I've been trying to define what \"programming\" even means lately. Programming, scripting, software engineering, software architecture — these terms get thrown around interchangeably on social media. For the sake of this discussion, let's assume we're talking about the full picture: building software, building solutions, integrating systems, and handling the end-to-end flow.",[23,244,246],{"id":245},"the-pattern","The Pattern",[11,248,249,250,253],{},"The poster says they've been mentoring junior devs and noticing something consistent: ",[18,251,252],{},"these juniors are using Cursor or Copilot for everything."," They never learn to code from scratch. They don't understand what the AI generates. And when it produces something wrong, they can't debug it.",[11,255,256],{},"This tracks. Think of it like learning a spoken language. If you don't understand the language — whether that's JavaScript, C#, Go, Elixir, whatever — you can't form sentences or build complex structures with it. But now there's this tool, Cursor, that can rearrange the words in just the right way that a native speaker understands exactly what you're saying. And as the tool improves, the output sounds better and better.",[11,258,259,260,263],{},"The problem? ",[18,261,262],{},"The junior developer has never really spent time working in the language in a production environment."," So when they need to build something, they either spend time researching and learning (which takes longer), or they offload it straight to Cursor. Cursor generates a beautiful-looking authentication flow. It mostly works. With LLM improvements, it'll work even better over time.",[11,265,266],{},"But none of that knowledge transfers to the junior. That's the core argument — Cursor wrote it, and they don't understand how it works.",[23,268,270],{"id":269},"is-this-just-an-old-man-yelling-at-a-cloud","Is This Just an Old Man Yelling at a Cloud?",[11,272,273],{},"Maybe. Is this just the next turn in how we write software?",[11,275,276,277,280],{},"The argument always comes back to abstraction. Nobody writes machine code by hand anymore. Even assembly is rarely something humans sit down and program directly. We have C and higher-level languages that compile down to machine code, which then runs operations on the CPU. ",[18,278,279],{},"That's a healthy abstraction"," — without it, we'd get lost in complexity and never build on top of existing systems.",[11,282,283],{},"But I do worry we're going to lose the understanding of how to write things at lower levels. If the AI plug gets pulled, we lose the crutch. Engineers who know how to work with old tech stacks — COBOL and the like — are retiring, and juniors aren't exactly lining up to learn those languages.",[11,285,286,287],{},"As long as the documentation exists, I'd guess future LLMs will help us maintain that knowledge. But here's the real danger: you end up with a person who doesn't understand the language working in tandem with an AI that does. And ",[18,288,289],{},"an LLM with malicious intent could exploit a programmer who doesn't actually know what's happening under the hood.",[23,291,293],{"id":292},"will-developers-forget-how-to-code","Will Developers Forget How to Code?",[11,295,296],{},"The Reddit poster worries that in five years, we'll have a generation of developers who can't code without AI assistance.",[11,298,299],{},"But think about calculators. Do people still calculate things without one? Sure — people who enjoy math, who are into arithmetic and algebra. Me? I haven't worked out a difficult equation on paper in a long time. I use a calculator, I use Gemini, I do quick mental math on simple stuff like percentages. That's about it.",[11,301,302,303],{},"I don't think we'll get a definitive answer on whether this kind of purism survives. ",[18,304,305],{},"We're all just embracing the new world and seeing where it takes us.",[23,307,309],{"id":308},"im-a-builder-not-a-purist","I'm a Builder, Not a Purist",[11,311,312],{},"I personally enjoy coding with AI. It's unlocked me to create so many more solutions than I could before. I consider myself a builder more than a programmer at this point. I architect solutions. I engineer them. I work with the people who are going to use them.",[186,314,315],{},[11,316,317],{},"AI is just a means to an end. It's something that helps me bring requirements to reality.",[11,319,320],{},"That's it. That's the take.",{"title":211,"searchDepth":212,"depth":212,"links":322},[323,324,325,326],{"id":245,"depth":212,"text":246},{"id":269,"depth":212,"text":270},{"id":292,"depth":212,"text":293},{"id":308,"depth":212,"text":309},"2026-04-02","AI tools like Cursor and Copilot are making junior developers worse at actual programming. Here's why that matters.",{},"\u002Fblog\u002Fjunior-devs-cant-debug",{"title":234,"description":328},"blog\u002Fjunior-devs-cant-debug","\u002Fblog\u002Fjunior-devs-cant-debug.png","https:\u002F\u002Fyoutu.be\u002FNUJmOMQZ4Ss","3EL1xmoZMD5X7nOYBYngCPOZ3-FxqeK2QRiwFofE_pM",{"id":337,"title":338,"body":339,"date":475,"description":476,"extension":222,"meta":477,"navigation":224,"path":478,"seo":479,"stem":480,"tag":228,"thumbnail":481,"youtube_url":482,"__hash__":483},"blog\u002Fblog\u002Faxios-npm-trojan.md","83 Million Devs Just Got a Trojan From npm",{"type":8,"value":340,"toc":468},[341,344,347,351,358,365,384,388,391,394,397,407,411,414,419,425,428,432,439,442,446,455,462],[11,342,343],{},"I just published a video about how AI has sucked all the joy out of programming. Turns out it's sucking the safety out of it too.",[11,345,346],{},"Let's talk about Axios.",[23,348,350],{"id":349},"the-scale-of-the-problem","The Scale of the Problem",[11,352,353,354,357],{},"If you've done any frontend work — hitting APIs from a browser — you've almost certainly used Axios. It's dead simple, widely trusted, and pulls ",[18,355,356],{},"101 million weekly downloads"," on npm. That's 101 million times a week this library gets dragged into someone's code.",[11,359,360,361,364],{},"Does it do anything you can't already do with the native ",[31,362,363],{},"fetch"," API that's been in browsers for a decade? Not really. But it's convenient, it's got some nice ergonomics around it, and people love it. I'm not here to relitigate Axios vs. fetch.",[11,366,367,368,371,372,375,376,379,380,383],{},"I'm here because on March 31st, two Axios versions — ",[18,369,370],{},"1.14.1"," and ",[18,373,374],{},"0.30.4"," — were published with an injected fake dependency called ",[31,377,378],{},"plain-crypto-js",". A Trojan horse, shipped to roughly ",[18,381,382],{},"83 million users",".",[23,385,387],{"id":386},"what-actually-happened","What Actually Happened",[11,389,390],{},"The main Axios maintainer, Jason, appears to have had a long-lived npm publish token compromised. This is a supply chain attack, plain and simple.",[11,392,393],{},"Here's how it works: when you set up a CI\u002FCD pipeline, you create an auth token so the pipeline can publish packages without you manually logging in every time. Makes total sense — you don't want to re-authenticate on every release. The problem is when that token never expires, nobody's auditing it, and someone gets their hands on it.",[11,395,396],{},"Once the attacker had that token, they could publish to npm on the maintainer's behalf. And that's exactly what they did.",[11,398,399,400,402,403,406],{},"The malicious dependency, ",[31,401,378],{},", came bundled with a ",[18,404,405],{},"post-install script",". Post-install scripts are a legitimate npm feature — they let packages run setup steps after installation, like compiling native bindings for your specific OS. But anything that can execute arbitrary code on your machine is exploitable. And it was.",[23,408,410],{"id":409},"npm-is-the-biggest-weakness-of-the-internet","npm Is the Biggest Weakness of the Internet",[11,412,413],{},"There's a Reddit take floating around calling this an \"unpopular opinion,\" but honestly it's just correct.",[186,415,416],{},[11,417,418],{},"Before npm, we treated adding dependencies as something selective, something we had to study and choose very carefully. Now I install a popular library and it comes with 300 other unwanted dependencies, some of which don't even seem to make any sense for the tool I'm downloading.",[11,420,421,422,424],{},"This is the core issue. You install Axios — that's not the only thing landing on your machine. Axios has dependencies, those dependencies have dependencies, and somewhere in that tree, a compromised package called ",[31,423,378],{}," quietly showed up and ran code on your system.",[11,426,427],{},"Every package you install nowadays seems to pull in another 50. The attack surface is enormous.",[23,429,431],{"id":430},"the-ai-angle","The AI Angle",[11,433,434,435,438],{},"AI makes this worse in two directions. It's easier than ever to ",[18,436,437],{},"generate more code than we're comfortable reviewing",". And it's easier to kick off an agent to find vulnerabilities, discover exploits, and craft these kinds of attacks.",[11,440,441],{},"This isn't going to stop. It's going to get more frequent.",[23,443,445],{"id":444},"what-to-do-right-now","What To Do Right Now",[11,447,448,449,451,452,454],{},"Check your projects. If you're on Axios ",[18,450,370],{}," or ",[18,453,374],{},", you're on a compromised version. Both have been pulled from npm and the GitHub releases.",[11,456,457,458,461],{},"Here's the irony: the security best practice is \"always update everything\" because new versions patch zero-days and fix bugs. In this case, the ",[18,459,460],{},"newest version was the attack",". So \"update\" here actually means downgrade back to the last safe release.",[11,463,464,465,383],{},"Stay safe out there. Audit your dependencies. And maybe think twice about that next ",[31,466,467],{},"npm install",{"title":211,"searchDepth":212,"depth":212,"links":469},[470,471,472,473,474],{"id":349,"depth":212,"text":350},{"id":386,"depth":212,"text":387},{"id":409,"depth":212,"text":410},{"id":430,"depth":212,"text":431},{"id":444,"depth":212,"text":445},"2026-04-01","Axios shipped a Trojan to 83 million devs via npm. Here's what happened and what to do.",{},"\u002Fblog\u002Faxios-npm-trojan",{"title":338,"description":476},"blog\u002Faxios-npm-trojan","\u002Fblog\u002Faxios-npm-trojan.png","https:\u002F\u002Fyoutu.be\u002FY5eEAZLX-PI","qi7K3pVm97ojTvfjO9Fy0Yyyc1Kforu5m3eELlWLnAk",{"id":485,"title":486,"body":487,"date":623,"description":624,"extension":222,"meta":625,"navigation":224,"path":626,"seo":627,"stem":628,"tag":228,"thumbnail":629,"youtube_url":630,"__hash__":631},"blog\u002Fblog\u002Fai-has-sucked-all-the-fun-out-of-programming.md","AI Has Sucked All the Fun Out of Programming",{"type":8,"value":488,"toc":615},[489,496,499,502,506,513,519,522,526,529,536,539,542,546,549,554,561,565,572,575,582,586,589,595,602,606,609,612],[11,490,491,492,495],{},"There's a dev meltdown happening on Reddit right now. Multiple threads have blown up, all claiming the same thing: AI is destroying the joy of programming. One post is from someone with 20 years of experience — backend, frontend, system design, even robotics. Ever since ChatGPT came around, they've been slowly losing interest in the craft. They prided themselves on solving bugs, refactoring, building difficult solutions. That was ",[60,493,494],{},"their thing",". And now a machine does it.",[11,497,498],{},"Another person says AI has given them straight-up brain rot. They let Claude or Gemini handle the grunt work, the boilerplate, but they're still fighting against it — going slower on purpose just to keep some sense of craftsmanship alive.",[11,500,501],{},"I think a lot of developers can relate to this. And I get it. But I don't fully agree.",[23,503,505],{"id":504},"weve-never-had-a-tool-that-thinks-for-you","We've Never Had a Tool That Thinks For You",[11,507,508,509,512],{},"Here's what makes AI different from every tool that came before it. A hammer extends your hand. A lever multiplies your force. A calculator runs numbers. These tools have always ",[60,510,511],{},"amplified"," what you can do physically or mathematically.",[11,514,515,518],{},[18,516,517],{},"What's never happened before is a tool that offloads your reasoning."," Yes, calculators existed — but you still had to know what numbers to type in. With AI, you can rant at it, describe a vague problem, and it comes back with a working solution. It takes away the need to reason through the problem yourself, and that's genuinely new in human history.",[11,520,521],{},"So when people say this feels different — it is different. I won't pretend otherwise.",[23,523,525],{"id":524},"programming-is-more-than-writing-code","Programming Is More Than Writing Code",[11,527,528],{},"I think part of the friction comes from how we define \"programming.\" There's programming, scripting, software engineering, and now agentic building. I'd put all of it under the umbrella of software engineering. And if your identity is tied to one specific part of that — the act of writing code, line by line — then yeah, AI is going to feel like a threat.",[11,530,531,532,535],{},"But personally? I enjoy ",[60,533,534],{},"building solutions",". I like architecting things. I like thinking about how parts interconnect. The actual typing of code was never the part I was most attached to.",[11,537,538],{},"I've been coding with AI since ChatGPT first dropped. Back then it was cumbersome — copy code from my editor, paste it into ChatGPT, ask it to \"make sure to output the whole file,\" then copy it back. It was slow and made mistakes. But I was already embracing that workflow because to me, it was an absolute marvel.",[11,540,541],{},"Now with Cursor, Claude Code, and everything else — it's completely changed how I work. And it's brilliant.",[23,543,545],{"id":544},"the-machinery-argument","The Machinery Argument",[11,547,548],{},"I think this is similar to what's happened in other professions. Machinery comes along, and suddenly you don't need to do one specific task by hand anymore. You have to transcend and start utilizing that machinery.",[186,550,551],{},[11,552,553],{},"The profession of software engineering, as much as it's been around for a while, hasn't yet hit this place where you'd have to completely change how you're doing things. Until now.",[11,555,556,557,560],{},"I have a friend who's been programming since the early days. He still remembers ",[60,558,559],{},"actual terminals"," — not terminal emulators, but physical screens where you'd access a shared machine split across six terminals, and people would divide access to the server through them. The job of a programmer has already changed dramatically even without AI. These tools are just taking it a step further.",[23,562,564],{"id":563},"the-fun-side-nobodys-talking-about","The Fun Side Nobody's Talking About",[11,566,567,568,571],{},"While Reddit is having an existential crisis, some genuinely fun stuff is happening. Someone figured out you can use the accelerometer in a MacBook, and built an app where you ",[60,569,570],{},"slap your laptop and it moans",". Made $5,000 in three days. Is this peak software engineering? I don't know. But it's hilarious.",[11,573,574],{},"Someone else used the MacBook's lid angle sensor to make their laptop sound like an accordion. Or a creaky door.",[11,576,577,578,581],{},"These projects are stupid and wonderful. And here's my point: ",[18,579,580],{},"these things would have taken weeks to figure out before"," — digging through APIs, interfaces, learning how to access the accelerometer. Now you can just turn them out quickly. It's fun. You don't get too attached. It's a quick idea brought to life, and it encourages creativity.",[23,583,585],{"id":584},"more-code-not-less-code","More Code, Not Less Code",[11,587,588],{},"They said 90% of code would be written by AIs, and a lot of people misunderstood that. They assumed if you wrote 100 lines a day, you'd only write 10 and AI would handle the other 90.",[11,590,591,594],{},[18,592,593],{},"What actually happened is we're still writing those 100 lines. But we're producing way more code in total."," The output has exploded, not the human input shrunk.",[11,596,597,598,601],{},"And the tools are good enough now that you can have ",[60,599,600],{},"another"," agent — not the one that wrote the code — come in and inspect it. You point it at a codebase and say: \"You don't know anything about this. Go poke at it. Find bugs, overflow issues, security holes, anything that could blow up in production.\" And it does it.",[23,603,605],{"id":604},"where-i-land","Where I Land",[11,607,608],{},"I enjoy these tools. I enjoy how they let me build more than I ever could before. I'm not too attached to the exact lines of code — I care that the code is performant, maintainable, and secure. But the act of typing it out character by character? That was never the part that made me feel like a craftsman.",[11,610,611],{},"If your joy came specifically from the puzzle-solving, the line-by-line problem decomposition, I understand why this feels like a loss. That's valid. But I'd push back gently: the craft isn't gone. It's shifted. The puzzle is bigger now. You're architecting, reviewing, orchestrating, and building things that would have been impossible solo a few years ago.",[11,613,614],{},"The question isn't whether AI changed programming. It did. The question is whether you can find the fun in what programming is becoming.",{"title":211,"searchDepth":212,"depth":212,"links":616},[617,618,619,620,621,622],{"id":504,"depth":212,"text":505},{"id":524,"depth":212,"text":525},{"id":544,"depth":212,"text":545},{"id":563,"depth":212,"text":564},{"id":584,"depth":212,"text":585},{"id":604,"depth":212,"text":605},"2026-03-31","Devs on Reddit say AI is destroying the joy of coding. I get it — but I don't fully agree.",{},"\u002Fblog\u002Fai-has-sucked-all-the-fun-out-of-programming",{"title":486,"description":624},"blog\u002Fai-has-sucked-all-the-fun-out-of-programming","\u002Fblog\u002Fai-has-sucked-all-the-fun-out-of-programming.png","https:\u002F\u002Fyoutu.be\u002FYlXqDUUBmFk","sBV3qgV1yHiRMSeZLlQewsl_Cp0KJOC1ZpdbipnaKEs",1775237166929]