brainstorm-team

brainstorm-team

Language matching: detect the language the user is writing in and respond in the same language. If the user writes in Chinese, reply in Chinese. If English, reply in English. Do not mix languages unless the user does.

Work with a Team-Leader to design an Agent Team for your project.

When to use this skill

• You know what you want to build but aren't sure what roles you need
• You want project-specific agent definitions (not generic templates)
• You want to review and approve the team design before anyone is spawned

What happens

1. A Team-Leader subagent chats with you to clarify your goal
2. The Team-Leader drafts a team composition (roles, scope, ownership)
3. You iterate until you're satisfied
4. The Team-Leader writes ./recruitment-plan.md and ./.claude/agents/<role>.md files
5. You're ready for spawn-team

Prerequisite: Agent Teams must be enabled (CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1). If not, run /start-team first — its Preflight phase will guide you through setup.

---

How to invoke

From /start-team

This is Phase 2 of the /start-team command. The main session spawns the Team-Leader automatically after Opening and Preflight.

Standalone

If you already have Agent Teams enabled and want to re-brainstorm (e.g. the project scope changed):

Agent({
  subagent_type: "awesome-agent-team:team-leader",
  description: "Brainstorm team for: <your goal>",
  prompt: "<your goal and context>"
})

The Team-Leader's full behaviour is defined in agents/team-leader.md at the repo root (single source of truth).

---

The sign-off gate (critical)

The Team-Leader must not write any files until you explicitly approve the draft.

The loop:

1. Team-Leader presents a draft table (roles, first tasks, file ownership)
2. You approve, modify, add, or remove
3. Loop until you say "go", "ok", "approved", "可以", "继续"
4. Only then files are written

A premature ./.claude/agents/ directory means the gate was skipped — that's a bug.