Claude Code talks too much, and its status bar tells you nothing. So I built two open-source plugins to fix both — and a third, experimental one that redraws the transcript itself. This is the engineering: how the conciseness layer works, how it's measured, why the status bar is architected the way it is, and what building on Claude Code's undocumented function hooks actually looks like. Everything here is in clear-claude on GitHub, MIT licensed. Every number below is a measured one; where ...