Anthropic accidentally leaked 512,000 lines of Claude Code's source. Not everyone noticed the real moat. It's not the model. It's the harness around it: 1. A self-healing query loop that recovers before failing
2. A background memory system that cleans up between sessions
3.

→ View original post on X — @alphasignalai