Commands are useful when you keep asking for the same kind of result. Unlike a shell project action, a command supplies instructions to an Agent conversation.
Write the reusable part
In Settings → Commands, create an entry with a name, scope and description. Edit its template and optionally select an Agent or model override. $ARGUMENTS allows the invocation to supply task-specific text. Use the template editor's hints for supported references and command substitutions.
Save the entry before leaving. Built-in commands have reset behavior rather than the same deletion controls as user entries.
Supply the changing part in chat
Type / at the start of the project composer, select your command, and add its arguments. Check the resulting request before sending. If the procedure should be selected as supporting instructions instead, use a Skill through @.
The historical URL contains “snippets”; the current interface exposes Commands, not a separate snippet manager.