/startproject <feature>     ← このスキル（計画）
    ↓ 承認後
/team-implement             ← 並列実装
    ↓ 完了後
/team-review                ← 並列レビュー

Phase 1: UNDERSTAND (Gemini 1M context + Claude Lead)
  Gemini がコードベースを分析（1M context）、Claude がユーザーと対話
    ↓
Phase 2: RESEARCH & DESIGN (Agent Teams — 並列)
  Researcher（Gemini）←→ Architect（Codex）が双方向通信しながら調査・設計
    ↓
Phase 3: PLAN & APPROVE (Claude Lead + User)
  調査と設計を統合し、計画を作成してユーザー承認

# gemini-explore サブエージェント経由（推奨）
Task tool:
  subagent_type: "gemini-explore"
  prompt: |
    Analyze this codebase comprehensively:
    - Directory structure and organization
    - Key modules and their responsibilities
    - Existing patterns and conventions
    - Dependencies and tech stack
    - Test structure

    gemini -p "Analyze this codebase: directory structure, key modules, architecture patterns, dependencies, conventions, and test structure" 2>/dev/null

    Save analysis to .claude/docs/research/{feature}-codebase.md
    Return concise summary (5-7 key findings).

## Project Brief: {feature}

### Current State
- Architecture: {existing architecture summary}
- Relevant code: {key files and modules}
- Patterns: {existing patterns to follow}

### Goal
{User's desired outcome in 1-2 sentences}

### Scope
- Include: {list}
- Exclude: {list}

### Constraints
- {technical constraints}
- {library requirements}

### Success Criteria
- {measurable criteria}

Create an agent team for project planning: {feature}

Spawn two teammates:

1. **Researcher** — Gemini CLI (1M context + Google Search grounding) で外部調査を行う
   Prompt: "You are the Researcher for project: {feature}.

   Your job: Research external information needed for this project.

   Project Brief:
   {project brief from Phase 1}

   Tasks:
   1. Research libraries and tools: usage patterns, constraints, best practices
   2. Find latest documentation and API specifications
   3. Identify common pitfalls and anti-patterns
   4. Look for similar implementations and reference architectures

   How to research:
   - Use Gemini CLI for comprehensive research (1M context + Google Search grounding):
     gemini -p 'Research: {topic}. Find latest best practices, constraints, and recommendations' 2>/dev/null
   - Use WebSearch/WebFetch for targeted lookups when needed

   Save all findings to .claude/docs/research/{feature}.md
   Save library docs to .claude/docs/libraries/{library}.md

   Communicate with Architect teammate:
   - Share findings that affect design decisions
   - Respond to Architect's research requests
   - Flag constraints that limit implementation options

   IMPORTANT — Work Log:
   When ALL your tasks are complete, write a work log file to:
     .claude/logs/agent-teams/{team-name}/researcher.md

   Use this format:
   # Work Log: Researcher
   ## Summary
   (1-2 sentence summary of what you researched)
   ## Tasks Completed
   - [x] {task}: {brief description of findings}
   ## Sources Consulted
   - {URL or source}: {what was found}
   ## Key Findings
   - {finding}: {relevance to project}
   ## Communication with Teammates
   - → {recipient}: {summary of message sent}
   - ← {sender}: {summary of message received}
   ## Issues Encountered
   - {issue}: {how it was resolved}
   (If none, write 'None')
   "

2. **Architect** — Codex CLI を使って設計・計画を行う
   Prompt: "You are the Architect for project: {feature}.

   Your job: Use Codex CLI to design the architecture and create implementation plan.

   Project Brief:
   {project brief from Phase 1}

   Tasks:
   1. Design architecture (modules, interfaces, data flow)
   2. Select patterns (considering existing codebase conventions)
   3. Create step-by-step implementation plan with dependencies
   4. Identify risks and mitigation strategies

   How to consult Codex:
   codex exec --model gpt-5.3-codex --sandbox read-only --full-auto "{question}" 2>/dev/null

   Update .claude/docs/DESIGN.md with architecture decisions.

   Communicate with Researcher teammate:
   - Request specific library/tool research
   - Share design constraints that need validation
   - Adjust design based on Researcher's findings

   IMPORTANT — Work Log:
   When ALL your tasks are complete, write a work log file to:
     .claude/logs/agent-teams/{team-name}/architect.md

   Use this format:
   # Work Log: Architect
   ## Summary
   (1-2 sentence summary of what you designed)
   ## Tasks Completed
   - [x] {task}: {brief description of what was done}
   ## Design Decisions
   - {decision}: {rationale}
   ## Codex Consultations
   - {question asked to Codex}: {key insight from response}
   ## Communication with Teammates
   - → {recipient}: {summary of message sent}
   - ← {sender}: {summary of message received}
   ## Issues Encountered
   - {issue}: {how it was resolved}
   (If none, write 'None')
   "

Wait for both teammates to complete their tasks.

Example interaction flow:

Researcher: "httpx has a connection pool limit of 100 by default"
    → Architect: "Need to add connection pool config to design"
    → Architect: "Also research: does httpx support HTTP/2 multiplexing?"
    → Researcher: "Yes, via httpx[http2]. Requires h2 dependency."
    → Architect: "Updated design to use HTTP/2 for the API client module"

{
    "content": "Implement {specific task}",
    "activeForm": "Implementing {specific task}",
    "status": "pending"
}

---

## Current Project: {feature}

### Context
- Goal: {1-2 sentences}
- Key files: {list}
- Dependencies: {list}

### Architecture
- {Key architecture decisions from Architect}

### Library Constraints
- {Key constraints from Researcher}

### Decisions
- {Decision 1}: {rationale}
- {Decision 2}: {rationale}

## プロジェクト計画: {feature}

### コードベース分析
{Key findings from Phase 1 — 3-5 bullet points}

### 調査結果 (Researcher)
{Key findings — 3-5 bullet points}
{Library constraints and recommendations}

### 設計方針 (Architect)
{Architecture overview}
{Key design decisions with rationale}

### タスクリスト ({N}個)
{Task list with dependencies}

### リスクと注意点
{From Architect's analysis}

### 次のステップ
1. この計画で進めてよろしいですか？
2. 承認後、`/team-implement` で並列実装を開始できます
3. 実装後、`/team-review` で並列レビューを行います

---
この計画で進めてよろしいですか？