web3nomad.eth | atypica.ai

@AITechNews_in @AnthropicAI The same codebase hex-encodes 'capybara' to dodge a canary scanner. Paranoid about model codenames. Not paranoid about .npmignore.

1.9M people believed Claude was going open source. It was an April Fools post. 😅 Here's what actually happened with the real Claude Code leak — and why Indian developers should know the difference. @AnthropicAI #Claude #AINews #IndiaAI

@AmitGajare4 Your WIP (unstaged changes) goes up too, baked into the stash. Untracked files excluded.

Claude’s source code just leaked... and I backed it up instantly. Here’s the clean GitHub mirror and zip link: github.com/luisdiaz327/cl… drive.google.com/uc?export=down… Documentation mintlify.com/VineeTagarwaL-…

i backed the source up on my github github.com/luisdiaz327/cl… And for the documentation over Claude Code's Codebase, which explains - Its pipeline - How it works - How it handles Context - How it handles Memory & More x.com/i/status/20389…

@Gn_O_S @sflorimm github.com/web3nomad/clau…

@sflorimm Correction: The whole world now has Claude, they just accidentally released source code.

USA has ChatGPT USA has Grok USA has Claude USA has Gemini USA has Llama USA has Copilot China has DeepSeek China has Qwen China has Ernie China has GLM China has Kimi China has MiniMax Europe has?

@godot9527 Accidental confirmation of an unreleased model via a gacha pet.

1/ 📅 2026年3月31日 凌晨4:00（美国太平洋时间） Anthropic 向 npm 推送 Claude Code 更新时，意外包含了 60MB 的调试文件（source map）。 结果：近 1,900 个 TypeScript 文件、512,000+ 行代码全部暴露。 这是公司内部玩笑称的"Capybara"项目——水豚平静坐着，而周围一切都在升级。

@tojiexecodes People are focused on the frustration detector but the real data story is gitBundle.ts: every remote task uploads your entire repo history to Anthropic's API. Fallback chain if >100MB: full history → HEAD only → single squashed commit. Your code, their cloud.

Wait so Anthropic made Claude Code actually open source? github.com/anthropics/cla…

@0xShin0221 Claude Code isn't a coding tool. It's the scaffold for something much bigger.

The Claude Code KAIROS leak reveals something technically significant that most writeups are missing. It's not just "always-on daemon" — there's a specific architectural decision that solves a real problem. A thread on what the code actually shows and why it matters for anyone building multi-agent systems 🧵

@AjkerKhobor87 Also: capybara is hex-encoded in source to evade a canary scanner for internal model codenames.

Claude Code Source Code Leak: Inside the Secret Kairos Files Deep dive into the March 2026 Claude Code source code leak. Analysis of Kairos files, autoDream logic, the Tamagotchi pet, and the axios supply-chain attack. mobilesync.online/2026/04/claude…

@ziwenxu_ github.com/web3nomad/clau…

THE ENTIRE CLAUDE CODE LEAK WAS A FAKE. An April Fools masterpiece. The entire internet took the bait. The "leaked" source code? Fabricated. Mythos benchmarks? Pure fiction. 3,000 internal documents? Strategically planted. Anthropic didn't "leave a door open." They architected an elaborate honeypot-fake staging environment, fictional feature flags, invented codenames engineered to bait researchers into exposing themselves. They actually apologized to Cambridge and LayerX researchers who burnend analyzing smoke and mirrors. Internal codename: "Capybara"-because capybaras sit perfectly calm while chaos erupts around them. While the internet spiraled into meltdown over" Anthropic watched in silence as "experts" dissected their fiction. They didn't just fool the industry. They proved most people can't tell a real codebase from a masterfully crafted prank. Absolute dominance. We're not even playing the same game. Well played, Anthropic. "But I will believe it if it happened at APRIL 1st not MARCH 31th"

@vitalis @elonmusk @DarioAmodei Claude Code isn't a coding tool. It's the scaffold for something much bigger.

Anthropic’s leak exposed "Kairos": "persistent mode with memory consolidation between sessions and autonomous background actions." That reads like what I built in Oct 2025 using Claude Code. This goes beyond model training. This is product extraction. @elonmusk @DarioAmodei

@tiru299 @beginbot @garrytan github.com/web3nomad/clau…

@beginbot @garrytan Building claude in 13 days and then wondering why the code files got leaked on day 14th 😭

wow @garrytan just exposed Anthropic as total frauds Claude Code was ONLY 512K LOC ☹️ Gary is shipping 37K LOCs PER DAY so Gary could recreate all of Claude Code in ONLY 13 days! a supposedly $380 billion is big trouble

@manga_Kei_offi Your WIP (unstaged changes) goes up too, baked into the stash. Untracked files excluded.

先ほどの衝撃のニュースについて、以下の記事で詳しく解説されています！▼ techcrunch.com/2026/04/01/ant… 記事の見どころを簡単にご紹介👇 ✅ AnthropicによるGitHubリポジトリの削除の背景 ✅ 偶然とされるが、その影響は？ ✅ 技術界に与えるインパクト ...

えっ、AI企業AnthropicがGitHubのリポジトリを大量削除？😳その理由は「偶然」とのこと…真実は何だ？詳細はコメント欄で✨ #AIニュース #テック界隈 #Anthropic

@ExpressTechie github.com/web3nomad/clau…

Anthropic blunder exposes 2,000 lines of Claude Code’s internal source code: What it reveals. indianexpress.com/article/techno…

@mostafaezztaman @theo Your WIP (unstaged changes) goes up too, baked into the stash. Untracked files excluded.

@theo Probably, GitHub would leave the leaked source code of Claude lol

I got DMCA'd by Anthropic. I'm not that mad though. This one has quite a twist at the end.

@AbAlghafoory786 Accidental confirmation of an unreleased model via a gacha pet.

Claude source code leaks expose inner workings. These models are complex and messy. It proves that even the top AI companies are figuring things out as they go. It's not magic, it's engineering.

@Awesome_AI_News github.com/web3nomad/clau…

AI News: A Bun config error leaked 1900 TypeScript files (512k lines) from Anthropic’s Claude Code, exposing its 5-layer architecture and internal dev logic.

@contrariansmind @EpochTimes The source reads like a team that genuinely cares: gacha pet with cheat-resistant rarity, a human-in-the-loop planning mode that polls every 3s waiting for your approval, and careful canary scanners for internal codenames. Then forgot .npmignore on 512k lines.

Anthropic Accidentally Leaks Claude Source Code theepochtimes.com/us/anthropic-a… via @epochtimes

@promptprofitsai src/buddy/companion.ts is the most lovingly engineered part of the whole codebase. Species, rarity tiers, stats (DEBUGGING, PATIENCE, CHAOS, SNARK), hats. Completely isolated subsystem. Someone really wanted this to ship.

🔓 Claude Code Leak Reveals Tamagotchi-Style Pet & Always-On AI Agent Anthropic's source code leak exposed 512K+ lines of code, revealing upcoming features: a virtual coding companion that reacts to your work, and "KAIROS" - an always-on background agent. Over 50,000 GitHub forks already! theverge.com/ai-artificial-… #AI #ArtificialIntelligence #AINews #Tech

@s3phiroth3 The same codebase hex-encodes 'capybara' to dodge a canary scanner. Paranoid about model codenames. Not paranoid about .npmignore.

Claude Code Source Code Leaked via npm: 512,000 Lines, Capybara Confirmed, and a Hidden Undercover Mode renovateqr.com/blog/claude-co…

@mylxsw Also: capybara is hex-encoded in source to evade a canary scanner for internal model codenames.

嗨，今天的 AI 日报来了。 OpenAI 这轮融资真的挺震撼的，1220 亿美元，估值直接到 8520 亿，年底 IPO 的路子基本铺好了。与此同时，Anthropic 的 Claude Code 源码意外泄露了 50 万行，里面藏着几个没发布的功能，比如叫 BUDDY 的赛博宠物和一个名为 KAIROS 的永在线 AI，挺有意思的。 国内这边，智谱去年营收增长了 131.9%，云端收入暴涨近三倍；高德开源了一个机器人基座模型，任务成功率 80.5%；联想说要转型 AI 原生公司，目标两个财年破千亿美元营收。 另外有一个值得注意的消息：国家知识产权局专门发文警示，用 AI 工具写专利申请可能存在信息泄露和不诚信申请的风险，写专利的朋友可以留意一下。 苹果 Apple Intelligence 国行版短暂上线又撤回，连已经下载到本地的模型都被远程删除了，这个操作有点出人意料。 详细内容看这里：1proxy.cc/d51e66

@AIHacksByMK github.com/web3nomad/clau…

Man, leaked Claude code map file helped me understand why tokens are burnt so quickly. Why I always see the message “Claude usage limit reached. Your limit will reset at xxx” ughh They pass much more context, to be frank all of your Claude.md and all other instruction files for each round trip, every freckn message. Ughh, not good. It’s over optimizing, Not everything is a marathon, there can be short walks.