What Claude Code's Leaked Source Actually Revealed

The 512,000-line Claude Code leak exposed more than architecture - it surfaced 44 hidden feature flags, unreleased systems, and internal codenames Anthropic clearly didn't want public. Among the findings: Kairos, an always-on background daemon that stores memory logs with nightly consolidation cycles, and an "Undercover Mode" subsystem that prevents Claude from revealing Anthropic codenames when contributing to open-source repos. A Korean developer built a clean-room Python rewrite called claw-code that hit 30,000 GitHub stars within hours. The code was also mirrored to decentralized git infrastructure with a note that it will never come down, making DMCA takedowns effectively useless. The competitive intelligence value of seeing full LLM orchestration logic, multi-agent coordination, and permission flows is significant for every team building coding agents right now.