Instructions for a-b/readable-ad-hoc-commands, covering readable ad-hoc commands — agent instructions, when this applies, the rules, per-language mechanics and transform: before → after.
Format the ad-hoc commands you run so the user reads their intent, steps, and blast radius before they execute — shell, python -c/node -e, jq, or SQL. Use whenever about to run a non-trivial command (multi-step, piped, looped, here-doc, or a dense inline snippet), or anything that writes, deletes, or overwrites state.