Anthropic accidentally leaked the entire source code for Claude Code.
512,000 lines. 1,906 files. Live on npm for hours.
Here's what they were hiding 🧵
The repo hit 50,000+ clones before Anthropic issued DMCA takedowns.
Someone rewrote the whole thing in Python overnight.
OpenAI and Google joked their code leaked too — except theirs was already open source.
The most "safety-first" AI company just went involuntarily open-source.
Follow for more 👇
Anthropic accidentally leaked the entire source code for Claude Code.
512,000 lines. 1,906 files. Live on npm for hours.
Here's what they were hiding 🧵
The best part?
Anthropic's head of Claude Code said "100% of my contributions in the last 30 days were written by Claude Code."
The code was leaked because Claude Code uses Bun (a runtime Anthropic acquired), which had a known bug shipping source maps in production.
Claude leaked itself
Fourth: Proactive Mode.
A feature flag that lets Claude Code act autonomously — making changes without being asked.
Also found: scaffolding for crypto-based payments so AI agents can transact on their own.
The agentic era is already built. Just not shipped yet.
Third — and this one's wild:
Anthropic built a fake tool injection system called ANTI_DISTILLATION.
If a competitor is recording Claude Code's API traffic to train their own model, it injects fake tool definitions to corrupt their training data.
AI companies are at war.
Second: KAIROS.
An always-on autonomous agent mode that runs in the background — even when you haven't asked it anything.
Includes a /dream feature for "nightly memory distillation."
Claude watching you sleep. Noted.
First thing people found: a fully built Tamagotchi pet system called /BUDDY.
→ 18 species (duck, dragon, capybara, "chonk")
→ Rarity tiers — 1% legendary
→ Stats: DEBUGGING, PATIENCE, CHAOS, SNARK
→ Planned launch date: April 1, 2026
This was their April Fools drop. Leaked early.