On 31 March 2026, security researcher Chaofan Shou discovered that Anthropic had accidentally published a source map file in the @anthropic-ai/claude-code npm package. The .map file referenced the complete, unminified TypeScript source — over 1,900 files and 512,000 lines of code — which was directly downloadable from Anthropic's own R2 cloud storage bucket. A misconfigured .npmignore or package.json files field failed to exclude the debugging artefact from the production distribution.
The leaked source revealed several unreleased features that captured the developer community's attention. BUDDY is a Tamagotchi-style AI pet that sits in a speech bubble next to the input box, featuring 18 species including duck, dragon, axolotl, and capybara, with rarity tiers from common to a 1% legendary chance. Each pet has five stats: DEBUGGING, PATIENCE, CHAOS, WISDOM, and SNARK. KAIROS, dubbed 'Always-On Claude', is a persistent assistant mode that keeps working across sessions, stores memory logs in a private directory, performs nightly 'dreaming' to tidy context, and can proactively start tasks.
Internal feature flags such as PROACTIVE, VOICE_MODE, BRIDGE_MODE, and KAIROS were also exposed, indicating a pipeline of product features not yet publicly released. References to Capybara models confirmed details from the earlier Mythos data leak, suggesting the next-generation model tier is deeply integrated into Claude Code's architecture.
Anthopic quickly pushed an npm update to remove the source maps and deleted earlier versions from the registry, but mirrors and GitHub repositories had already preserved the code. For the context engineering community, the leak provides unprecedented insight into how Anthropic structures its agentic coding tool — and hints at a far more ambitious product roadmap than publicly announced.
