Media & Culture

Someone just leaked claude code's Source code on X

A 1,884-file leak exposes secret AI pets, 30-minute planning modes, and employee-only access flags.

Deep Dive

The complete TypeScript source code for Anthropic's Claude Code CLI tool—comprising approximately 1,884 files—has been leaked online, revealing a treasure trove of 35 hidden, build-time feature flags. The most intriguing include 'BUDDY,' a Tamagotchi-style AI pet with 18 species and rarity tiers slated for a suspicious April 1, 2026, teaser; 'KAIROS,' a persistent assistant that 'dreams' to consolidate user memories overnight; and 'ULTRAPLAN,' a system that offloads complex planning tasks to a remote Claude instance for up to 30 minutes before returning the approved plan to the user's terminal.

Beyond these experimental features, the leak exposed substantial infrastructure for advanced AI workflows. 'Coordinator Mode' is already accessible via an environment variable and spawns parallel worker agents, while a 'UDS Inbox' allows multiple local Claude sessions to communicate. The findings also include over 120 undocumented environment variables, 26 internal slash commands like `/teleport` and `/dream`, and a `USER_TYPE=ant` flag that grants Anthropic employees full access, highlighting a significant gap between internal development builds and the public product.

Key Points
  • Leaked source code reveals 35 hidden features including 'BUDDY' AI pets with 18 species and stats like CHAOS.
  • 'ULTRAPLAN' feature sends complex tasks to a remote Claude for 30-minute processing before returning results.
  • Exposed 120+ undocumented environment variables and an employee-only 'USER_TYPE=ant' flag unlocking all features.

Why It Matters

The leak reveals Anthropic's ambitious roadmap for AI agents, persistent memory, and playful UX, signaling a major shift in developer tools.