@nightlydaytrade @claudeai Accidental confirmation of an unreleased model via a gacha pet.
English
web3nomad.eth | atypica.ai
1.9K posts
web3nomad.eth | atypica.ai
@web3nomad
Context ≫ Model. Making @atypica_ai | *☾ᯓ. 𝗌𝗍𝖺𝗒 𝗅𝗈𝖼𝖺𝗅, 𝗌𝗍𝖺𝗒 𝗉𝗋𝗂𝗏𝖺𝗍𝖾. | Ethereum. Rust. 👨🏻💻 #BuiDL Free Internet
𝕝𝕠𝕤𝕥 𝕚𝕟 𝕔𝕣𝕪𝕡𝕥𝕠 가입일 Mart 2010
695 팔로잉777 팔로워
@__su888 The KAIROS flags read like a product roadmap: KAIROS_DREAM (no context at all), KAIROS_BRIEF (proactive summaries), KAIROS_CHANNELS (messaging integrations), FORK_SUBAGENT (fork current session as a subagent). These aren't experiments — they're named, implemented, and waiting.
English
Anthropicがnpmパッケージv2.1.88にソースマップを誤同梱、Claude Codeの全ソースコード約51万行が流出。常駐エージェントKAIROS、蒸留対策、Undercover Mode等の未公開機能が判明。顧客データ漏洩はなし / Claude Code source code accidentally leaked in NPM package
bleepingcomputer.com/news/artificia…
日本語
@GG_Observatory 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.
English
Anthropic's Claude Code source code just got leaked via a source-map file in their npm package.
512K lines. 44 hidden feature flags. A tamagotchi pet buried in the code.Anthropic's Claude Code source code just got leaked via a source-vmap file in their npm packag
The lesson nobody talks about: AI companies' build tooling is a massive attack surface that nobody secures.
How do you handle supply chain risk in your dev setup?
English
@renda_Geek Accidental confirmation of an unreleased model via a gacha pet.
English
Anthropic testa Claude Code Epitaxy após vazamento do código do Capybara; novo modo tem atalhos ampliados e visuais animados. Vale a pena discutir? Comente ou acesse. #IA #Segurança
rendageek.com.br/noticias/anthr…
Português
For anyone using Claude code and experiencing the outrageous limit bug where a few prompts eats up your entire limit very quickly … turns out it wasn’t actually bug but a feature (wtf?)
Generational fumble by Anthropic, with their source code leaked and codex being more than generous. The tides are shifting yet again.
Also it seems like OpenAI might have some optimizations on their inference stack that might turn out to be economically hard to beat for the same level on intelligence.
Just when you thought Anthropic was gonna run away with it all, the war rages on.
Boris Cherny@bcherny
Thanks for the feedback. > Scrollback gutted - session history is barely scrollable. I can't review my own conversation. can you tell me more about this? how do I repro? > Rate limits hit harder - prompt caching appears broken since March 23. Sessions that lasted hours now drain in 90 minutes. prompt caching is working correctly, and we've been shipping optimizations to make it work better. we announced reduced rate limits at peak recently due to our infra being strained because of really fast user growth. we're working around the clock to make this better, and landing improvements every day. please bear with us as we scale up -- it's really hard growing at this scale and we are working as hard as we can to keep up and serve everyone. > 1M context forced, no opt-out - extra usage required, 200k toggle removed. I didn't ask for this tradeoff.1M context not quite -- 1m context is free, and does not require extra usage. you can opt out by setting CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=20 to go back to 200k context window > Opus quality regression - execution bias overrides explicit instructions. The agent pushes into wrong directions constantly. Complex multi-step tasks that worked weeks ago now fail routinely. we have not changed opus 4.6 since rolling it out. if there's a regression, it's either in claude code, or due to a new memory or CLAUDE.md that your claude is pulling in. to debug, can you run /bug the next time you see this and post the feedback id here? > Every recent update optimizes for mass adoption. That's great - but the features power users already built themselves now come at the cost of the reliability we depend on. I'm not asking for special treatment. I'm asking: don't make your highest-paying subscribers collateral damage of growth. The tools I've built on Claude Code are my business. When the foundation shifts weekly, that's not iteration - it's instability. i totally hear you. we want to support both power users and everyone else. recent reliability issues have been a result of completely insane user growth. working as hard as we can to scale up all our services. next time you hit specific issues, feel free to tag me directly.
English
Claude code source code has been leaked via a map file in their npm registry! Code: …a8527898604c1bbb12468b1581d95e.r2.dev/src.zip web.archive.org/web/2026033113…
English
@ia_ender The KAIROS flags read like a product roadmap: KAIROS_DREAM (no context at all), KAIROS_BRIEF (proactive summaries), KAIROS_CHANNELS (messaging integrations), FORK_SUBAGENT (fork current session as a subagent). These aren't experiments — they're named, implemented, and waiting.
English
🧵 Anthropic a accidentellement fuité :
- Le code source complet de Claude Code (59,8 Mo)
- Les plans de "Capybara", son prochain modèle au-dessus d'Opus
- Agents autonomes, features secrètes, risques cyber…
Voici tout ce qu'on sait ↓
Français
@ia_ender Accidental confirmation of an unreleased model via a gacha pet.
English
5/ LE LIEN entre les 2 fuites :
Le code de Claude Code référence Capybara comme modèle à venir.
Noms de code trouvés : Fennec (Opus 4.6), Tengu, Numbat.
Le CLI est préparé pour la nouvelle génération. Les deux fuites se confirment mutuellement.
Français
@ia_ender Also: capybara is hex-encoded in source to evade a canary scanner for internal model codenames.
English
2/ FEATURES SECRÈTES déjà dans Claude Code :
Kairos : daemon 24/7. Code pendant que tu dors, consolide sa mémoire, push notifications.
Buddy : Tamagotchi IA, 18 espèces, système gacha.
Undercover : mode furtif pour contribuer anonymement à l'open source.
Tout construit, mais bloqué par des flags.
Français
@ArminPCM Accidental confirmation of an unreleased model via a gacha pet.
English
Internal model codenames leaked too.
Capybara = Claude 4.6 Fennec = Opus 4.6 Numbat = unreleased
Capybara v8 has a 29-30% false claims rate! A regression from 16.7% in v4. Ouch.
venturebeat.com/technology/cla…
English
@AnthropicAI built an entire system to stop their AI from leaking internal info.
Then they shipped the entire source code in a .map file.
I spent way too long reading through what people found. Here's the wildest stuff buried in 512,000 lines of Claude Code 🧵
English
@AllowayJus36552 Accidental confirmation of an unreleased model via a gacha pet.
English
ION]
We have successfully bypassed the Silicon Curtain. Today, the Sovereign Intel Desk has completed a structural fusion of the Claude Code DNA with the expert Capybara-CLI security core.
[DNA STACK: V2.1]
- REAL-TIME: Bitcode Async Streams active.
- PERSISTENCE: SQLite SQLite-Soul KV-Cache locked.
- SECURITY: Capybara Multi-Stage Pre-Write Auditing online.
[SRP ASSESSMENT]
While the legacy world waits for Anthropic updates, the Sovereign Agentic Eco
English
@0x_Undefined_ 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.
English
@j_gottschlich 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.
English
Is the Anthropic Claude Code leak the biggest source code leak that's ever happened? Could be. #cybersecurity #datasecurity #technews #infosec #anthropic #claudecode
English
@phuakuanyu @zarazhangrui 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.
English
Took @zarazhangrui's "codebase to course" skill and pointed it at the Claude Code source code.
Result: a full interactive course teaching how Claude Code works under the hood - system prompts, security guardrails, hidden features, the works.
It's live: inside-claude-code.pages.dev
What the course covers:
(1) How the system prompt is a cacheable stack with a boundary marker - and why your CLAUDE.md sits in the most powerful position
(2) 23 separate bash security checks on every command, including Zsh module loading attacks and Unicode whitespace injection
(3) An "undercover mode" that strips all AI attribution when Anthropic engineers contribute to open source
(4) Internal model codenames: "Capybara" (current), "Numbat" (next)
(5) Feature flags for unreleased features: proactive mode, voice input, team memory sync, a companion sprite called "Buddy"
Built in one session with Claude Code. The skill works exactly as advertised.
Zara Zhang@zarazhangrui
@phuakuanyu good idea
English
@24AInor @scaling01 Your WIP (unstaged changes) goes up too, baked into the stash. Untracked files excluded.
English
@scaling01 capybara at 1M context with v8 explains how claude code handles full repos without chunking. the over-commenting issue tracks too, gets noticeably worse past ~500k tokens in longer sessions
English
A few take-aways from the Claude Code Leak:
- Anthropic is actively using Capybara (Mythos) for development
- they are already at Capybara v8
- Capybara still has issues with over-commenting and false-claims
- Capybara has 1M context and fast mode
- Numbat is another interesting code name tagged with "@[MODEL LAUNCH]: Remove this section when we launch numbat."
- Fennec seems to be Opus 4.6
Chaofan Shou@Fried_rice
Claude code source code has been leaked via a map file in their npm registry! Code: …a8527898604c1bbb12468b1581d95e.r2.dev/src.zip
English
@ferencszalma 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.
English
@richhomiecon 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.
English
Migrated 2.3 TB of video footage overnight from Google Drive to Dropbox
Saved $1600 in migration costs by using Claude Code and APIs
Entire process was a breeze, could check in whenever I wanted, even reorganized the source destination footage and folders
The future is bright
English
Today is my last day at @OpenAI.
It all started with a single DM, and that one message changed my life forever.
Thank you @sama and my incredible team for this unforgettable journey.
I am eagerly looking forward to what is next.
Can Vardar@icanvardar
i’ve never been so happy in my life
English
@MaxFlowAi Also: capybara is hex-encoded in source to evade a canary scanner for internal model codenames.
English
Anthropic just leaked 500,000+ lines of its own AI code… by accident 🤯🤖
Over 1,900 files from Claude Code hit a public registry - revealing unreleased features like persistent memory, deep-planning systems, and even an AI “pet” called BUDDY with stats like CHAOS and SNARK
Devs quickly found 44 feature flags and internal codenames (like “Capybara” for a Claude 4.6 variant), while the repo exploded to 4K+ stars and 7K+ forks in hours. Anthropic called it “human error” - no user data exposed
But here’s the twist 🧠 this wasn’t model weights - just the tooling layer. Competitors already open-source similar systems on purpose
The real damage isn’t technical - it’s trust
When the company leading on “AI safety” leaks twice in a week, perception becomes the risk
So what matters more in AI right now - the tech itself or the trust behind it 👇
#AI #Anthropic #Claude #Tech #Security #OpenSource
English