Claude Code Source Code Breakdown (kuber.studio)

In a surprising revelation on March 31, 2026, Chaofan Shou uncovered the complete source code of Anthropic's Claude Code, an AI coding CLI, unintentionally exposed on the npm registry through a bundled sourcemap file. This leak is particularly significant as it not only reveals technical aspects of Claude Code—like its innovative buddy system and multi-agent orchestration capabilities—but it also underscores a common oversight in software deployment: failing to exclude source maps from production releases. The incident raises important questions about security practices within AI development, particularly for systems claiming enhanced internal protections. The leak divulges intriguing insights about Claude Code’s architecture, such as its “Buddy” system that introduces a Tamagotchi-like pet interaction, and “KAIROS,” a proactive assistant mode that observes user actions. Technical features like the autoDream memory consolidation engine and a sophisticated multi-agent orchestration setup demonstrate the tool's depth and advanced design. Furthermore, the presence of an "Undercover Mode," designed to prevent AI personnel from inadvertently disclosing sensitive internal information, highlights the ongoing need for vigilance in AI model deployment. Overall, the leak serves as a cautionary tale whilst providing the community with an unexpected glimpse behind the curtain at Anthropic’s innovative yet secure AI technologies.