This sets up multi-agent swarms where multiple Claude instances can work in parallel without stepping on each other's toes. You get hierarchical, mesh, or ring topologies depending on whether you need 3 agents or 15, and it uses git worktrees to keep everyone in their own sandbox. The anti-drift configuration is the real selling point here since coordinating multiple AI agents usually turns into chaos fast. If you're trying to parallelize a large refactor or split up a complex feature across a team of Claudes, this gives you the scaffolding to actually make that work instead of merging conflicts forever.
npx skills add https://github.com/ruvnet/ruflo --skill swarm-init