team-plan-skills

team-plan

One command to go from task description to a coordinated Agent Team.

Agent Teams are powerful but using them well requires thought — team composition, task dependencies, file ownership, spawn context, plan approval. This skill wraps the official best practices into a repeatable workflow so you get a well-structured team every time.

/team-plan <task>

Scout (subagents)  →  Plan (plan mode)  →  You approve  →  Agent Team executes

Why

Without this skill, using Agent Teams typically means:

• Telling Claude to "create a team" and hoping it designs a good one
• Teammates that lack context because they don't inherit the lead's history
• File conflicts when two teammates edit the same file
• No review gate before high-risk changes

This skill fixes that by requiring a structured plan before any team is spawned:

• Scout first — Subagents explore the codebase and research external knowledge in parallel, so the plan is grounded in facts, not assumptions
• Plan with team design — The plan explicitly defines roles, models, file ownership, task dependencies, and spawn context for each teammate
• Plan approval for high-risk teammates — Teammates making dangerous changes must get their approach approved before writing code (official docs)
• No file conflicts — The plan enforces that no two teammates write the same file (official best practice)
• Spawn context — Each teammate's prompt is planned with the specific context they need, because they don't inherit the lead's conversation history
• You approve before execution — Nothing runs until you review the plan

Install

Plugin Marketplace (recommended)

/plugin marketplace add zhangxin840/team-plan-skill
/plugin install team-plan@team-plan-skills

npx

npx skills add zhangxin840/team-plan-skill

Manual

git clone https://github.com/zhangxin840/team-plan-skill.git
cp -r team-plan-skill/plugins/team-plan/skills/team-plan ~/.claude/skills/

Usage

/team-plan Refactor src/auth/ from session-based to JWT authentication

/team-plan Research Claude Code v2.1.35 changes, produce upgrade report and articles

/team-plan Review and update all API documentation to match current implementation

What the plan covers

Section	Purpose
Task Analysis	Scope, boundaries, risks — refined from scout findings
Team Design	Each teammate: name, role, agent type, model, owned files, plan approval requirements
Task Breakdown	Tasks with owners, blockedBy dependencies, acceptance criteria
Phase Sequence	What runs in parallel, what must wait
File Ownership	Explicit map — no two teammates write the same file
Spawn Context	Key context for each teammate's prompt

Prerequisites

Agent Teams must be enabled in ~/.claude/settings.json or .claude/settings.json:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

Design

• Minimal skill, maximum autonomy — 30 lines of instructions. Only enforces scout → plan → approve. Execution is left entirely to Claude.
• Works in any project — No hardcoded paths. Agents discover structure from CLAUDE.md and project conventions.
• Built on official best practices — Plan approval, file ownership, spawn context, task sizing — all from the Agent Teams docs.

License

MIT