Typing /review or /commit expands a stored prompt — often a whole procedure with checks and output format — into the conversation. It is the smallest unit of agent reuse and usually the first thing a team builds once they notice they are pasting the same instructions weekly.

Commands and skills overlap and differ in trigger: a command runs because you asked for it by name, a skill loads because the agent judged it relevant. Explicit invocation is more predictable, which is why irreversible workflows tend to be commands.