swarm-init

name	swarm-init
description	Initialize a multi-agent swarm with anti-drift configuration
argument-hint	[--topology hierarchical|mesh|ring]
allowed-tools	Bash(npx *) mcp__ruflo__swarm_init mcp__ruflo__swarm_status Agent TeamCreate SendMessage

Initialize a hierarchical swarm for coordinated multi-agent work.

Via MCP: mcp__ruflo__swarm_init({ topology: "hierarchical", maxAgents: 8, strategy: "specialized" })

Or via CLI:

npx @sparkleideas/cli@latest swarm init --topology hierarchical --max-agents 8 --strategy specialized

Coordinate the swarm as a Claude Code team

A swarm is parallel collaboration — workers divide the work and coordinate while running. Bind them to a Claude Code team so the Agent-Teams behaviors activate (shared team_context, teammate_mailbox, @name addressing, TaskUpdate auto-claim):

1. After init, create the team — reuse the swarmId from the swarm_init response as the team name: TeamCreate({ team_name: "<swarmId>", description: "<objective>" })
2. Spawn each worker into the team — pass team_name on the Agent/Task call; add isolation: "worktree" for git-safe parallel edits: Agent({ subagent_type: "coder", team_name: "<swarmId>", run_in_background: true, prompt: <task> })
3. Coordinate at runtime with SendMessage({ to: "<worker-name>", ... }) — workers see teammate messages and shared context, and can claim tasks.

Swarm vs hive — when NOT to team-bind: teams are for collaboration during execution. If the goal is a consensus or dialectic decision (independent expert positions → vote → verdict), use the hive-mind skill instead — its Council/Consensus patterns deliberately keep workers isolated (no cross-talk) so the dialectic stays honest. Never bind council workers to a team.

For larger teams (10+), use hierarchical-mesh topology:

npx @sparkleideas/cli@latest swarm init --topology hierarchical-mesh --max-agents 15 --strategy specialized