one thing i like about claude code (vs. pi) is that they keep adding new features i didn't know i wanted -- like /loop or subshells or scheduled prompts -- and i would never discover this with a minimal harness