Overview
- An npm release of @anthropic-ai/claude-code v2.1.88 included a 59.8 MB source map that exposed roughly 2,000 TypeScript files and more than 512,000 lines of code on the public JavaScript package registry.
- Anthropic said a packaging mistake caused the disclosure and called it human error, and the company said no customer confidential data or credentials were exposed.
- The leaked material showed Claude Code’s internal memory design, agent orchestration and permission checks, plus references to unannounced features like Kairos, Buddy and Ultraplan and codenames such as Capybara/Mythos.
- The code spread quickly across GitHub with thousands of downloads and over 41,500 forks, and DMCA takedown notices had limited effect as mirrors and a rapid Python reimplementation called claw-code appeared.
- The leak followed a rough week that also included exposure of about 3,000 internal files and a separate axios supply-chain compromise that planted a remote-access trojan in some installs, leading users to review projects and rotate credentials.