Source Code for Anthropic’s Claude Code Leaks at the Exact Wrong Time

Anthropic has suffered a significant operational lapse, inadvertently publishing the internal source code for its Claude Code AI coding assistant. The leak occurred when a .map file—a debugging artifact containing unobfuscated TypeScript—for version 2.1.88 was made public on the npm JavaScript package registry. Discovered by Solayer Lab intern Chaofan Shou, the file contains approximately 512,000 lines of code, offering an unprecedented look at the agent's architecture, including its engine for API calls, token counting mechanisms, and internal logic like 'spinner verbs' displayed during tasks.

While the code does not reveal details of Anthropic's underlying AI model, it exposes operational blueprints at a critical time. The company is reportedly preparing for an initial public offering (IPO) later this year and faces intensifying competition, particularly from OpenAI's pivot to enterprise and its Codex assistant. The leak also revealed quirky internal features, including a hidden 'Tamagotchi'-style virtual pet reportedly slated for an April Fool's launch. Anthropic confirmed the leak's authenticity, attributing it to a 'release packaging issue caused by human error' and stating no customer data was exposed. Notably, the head of Claude Code had previously stated that 100% of his recent contributions to the project were written by Claude Code itself, highlighting the team's deep reliance on the tool they built.