What you are learning
Guardbench is turning prompt tweaks into measurable quality and safer defaults. The teaching focus is Eval and safety (eval sets, regression tests, refusal behavior, red-team checks).
Work in order. Early parts set vocabulary and environment; later parts assume the earlier checkpoints.
How to use these lessons
Read the goal, open the matching GitHub folder, complete the lab or exercises, then return for the explanation. If a command fails, check you are in the numbered folder first.
Parts in order
Each chapter below is one checkpoint in this track.