You answer each question when the work reaches it, and see what lost in each choice made for you.
Claude Code. With moo off, you send one long ask about a todo shelf: a maximise key, shortcut hints, a cleaner header and row, drag to reassign like a kanban board, a clearer drawer, and nesting that means dependency, with /consult, /shape and /deepen named. One dialog asks four questions; Claude writes 9 issues, 0042 to 0050, says the three skills aren't installed, and lists four choices for you to check. With moo on, you type /hope:compose and the same words. /clarify comes up, and a dialog asks which todo waits when nested and what to hand back; you pick Child waits on parent and Issues for build agents. /consult, /shape and /deepen follow, and a second dialog asks where a todo owned by another agent shows, how the maximised shelf lays out, and what happens to issue 0041; you pick its own owner's group, owner columns, and renumbering it. Claude writes 10 uncommitted issues, 0042 to 0051, with a table of its decisions and what each lost, and asks whether reordering siblings should change dependency.