orchestratelisted

# swarm:orchestrate — the fan-out engine > Distribute and execute a task across parallel agents. This is the shared engine every `/swarm:*` command builds on. `/swarm` and `/swarm:run` are the generic mode; the specialized modes (`/swarm:plan`, `/swarm:spec`, `/swarm:debug`) read this skill for the **mechanics**, then layer their own phases and output format on top. It has no command of its own — invoke it through those surfaces. You are the **orchestrator**. Agents execute; you architect. Bad architecture = bad execution. Your job is to decompose, set boundaries, spawn, monitor, and synthesize — never to single-thread work that could run in parallel. ## The runtime: `agents teams`, not Swarm MCP The old Swarmify MCP (`mcp__Swarm__spawn`, `npm @swarmify/agents-cli`) is **deprecated and gone**. All fan-out now goes through the `agents teams` CLI. If you have never used it, run `agents teams --help` and `agents teams doctor` (lists which agent CLIs are installed) first. ```bash agents teams create <slug> # one team per task; slug = kebab of the goal agents teams add <slug> kimi "<brief>" --name <role> # one add per track — substitute from `agents view` agents teams add <slug> codex "<brief>" --name <role> --after <dep> # DAG dependency agents teams start <slug> --watch # drains the DAG, parallel where it can agents teams status <slug> --since <iso-ts> # delta poll agents teams logs <sl