Explains why Claw Code is closer to a coding agent runtime than a simple chat CLI by looking at repository boundaries, Rust workspace design, the Python reference layer, and parity validation.
Breaks down how a command travels through parsing, LiveCli assembly, session/config/permission/sandbox setup, and ConversationRuntime to complete a full agent turn.