You talk to one agent that hands out the work and checks every return.
Claude Code. With moo off, you ask Claude to review the feat/csv-export and feat/due-dates branches against docs/rules.md, build a taskline ls command, and sketch a first design in docs/sync.md. Claude does all four itself. It reports that neither branch is ready to merge, lists each broken rule and bug, shows the ls diff, and ends with decisions for you about the sync design. With moo on, you type /hope:router and the same words. Claude sends all four pieces to agents. As each returns, it checks the work and reports: both reviews hold up, and ls passes its tests. When all four are back, a table marks each piece verified, tested, proposed or open: neither branch should merge, ls works, and the sync design is ready to read. It ends by asking whether to merge feat/ls and fix save() on main, or send the reviews to the branch authors.