Anthropic moved swiftly to contain the fallout from the Claude Code source leak, filing DMCA takedown notices with GitHub that resulted in more than 8,100 repositories being disabled. The original leaked repo, nirholas/claude-code, and its entire fork network were taken down within hours. But as Decrypt reported, the internet is keeping the code forever — mirrors appeared on decentralised git platforms within minutes of the takedowns.
The most notable response came from a South Korean developer named Sigrid Jin, who rebuilt Claude Code's core architecture in Python from scratch within hours of the leak, using an AI orchestration tool called oh-my-codex. The resulting project, 'claw-code', was published before sunrise — a clean-room rewrite that sidesteps DMCA entirely because it is original code inspired by the architecture, not a copy of Anthropic's TypeScript. A decentralised mirror account posted the original source to Gitlawb with the message: 'Will never be taken down.'
The leak has become the fastest-growing repository incident in GitHub history, according to Layer5's engineering blog. The missing .npmignore entry that shipped a 59.8MB source map is now a cautionary tale being shared across developer communities — a single packaging oversight that exposed 512,000 lines of production code for a $2.5 billion revenue product.
For the context engineering community, the episode highlights both the transparency and the vulnerability of npm-based distribution. Anthropic's architecture is now public knowledge regardless of DMCA actions, and competing tools will inevitably learn from the patterns revealed in the leak. The question is whether this accidental open-sourcing ultimately helps or hurts Anthropic's competitive position.
