To make our library for efficient pangenomics more palatable, I made a little Unix shell. It “cheats” when it sees commands that invoke other pangenomic CLI tools and runs built-in functionality instead. The shell is built around an instruction-based IR that can fluidly intermix shell-like I/O (files and pipes) with efficient in-memory data structures. In a debatably fair comparison, the fake shell runs one shell script 48× faster than running it with sh.