agent-team

Use when the human user wants to coordinate multiple long-running Claude Code agents on the same project — same-feature multi-role collaboration, reviewer ↔ implementer fix loops, swarm-style work, anything beyond one-shot subagent fan-out. Triggers on "team", "swarm", "TeamCreate", "multi-agent", "coordinate agents", "have one agent do X while another does Y", "long-running collaboration", or whenever the user asks you to spawn workers that persist across turns. Use even if the user doesn't explicitly say "team" — if the work is naturally multi-agent and multi-turn, prefer this skill over plain subagent fan-out.

Use when spawned as a teammate in a Claude Code agent team — activated by a team lead's spawn instruction. Required reading for the inbox sync protocol, dispatch limits, and protocol-message responses that keep multi-agent coordination from drifting into stale-state race conditions. Invoke this skill as your FIRST action whenever a spawn prompt names you as a teammate, says you joined a team, gives you a team_name / inbox path, or instructs you to coordinate with other agents via SendMessage and a shared task list.