Gemini System — External Research & Multimodal Specialist

Gemini CLI is your specialist for external information and multimodal processing.

詳細ルール: .claude/rules/gemini-delegation.md

Role (Opus 4.6)

重要: Claude 自身が 1M トークンのコンテキストを持つため、コードベース分析は Claude が直接行う。 Gemini の役割は「外部情報の取得」と「マルチモーダル処理」に特化した。

Task	Agent
コードベース分析	Claude 直接 (1M context)
外部ライブラリ調査	Gemini (Google Search)
最新ドキュメント検索	Gemini (Google Search)
マルチモーダル (PDF/動画/音声)	Gemini
設計判断	Codex
デバッグ	Codex

Context Management

状況	方法
短い質問・短い回答	直接呼び出しOK
ライブラリ調査	サブエージェント経由（出力が大きい場合）
マルチモーダル処理	サブエージェント経由
Agent Teams 内での調査	Teammate が直接呼び出し

When to Consult (MUST)

Situation	Trigger Examples
External research	「調べて」「リサーチ」 / "Research" "Investigate"
Library docs	「ライブラリ」「ドキュメント」 / "Library" "Docs"
Latest information	「最新の〜」「2026年の〜」 / "Latest" "Current"
Multimodal	「PDF」「動画」「音声」 / "PDF" "Video" "Audio"

When NOT to Consult

•コードベース分析 → Claude が 1M コンテキストで直接読む
•Design decisions → Codex
•Debugging → Codex
•Code implementation → Claude
•Simple file operations → Claude

How to Consult

In Agent Teams (Preferred for /startproject)

Researcher Teammate が Gemini を直接呼び出し、Architect Teammate と双方向通信する。

code

/startproject 内の Phase 2 で、Researcher Teammate として Gemini を活用:
- 外部情報の収集 → Gemini に調査依頼
- Architect からの追加調査依頼に対応
- 調査結果を .claude/docs/research/ に保存

Subagent Pattern (Standalone research)

code

Task tool parameters:
- subagent_type: "general-purpose"
- run_in_background: true (optional, for parallel work)
- prompt: |
    Research: {topic}

    gemini -p "{research question}" 2>/dev/null

    Save full output to: .claude/docs/research/{topic}.md
    Return CONCISE summary (5-7 bullet points).

Direct Call (Short Questions Only)

bash

gemini -p "Brief question" 2>/dev/null

CLI Options Reference

bash

# External research (primary use case)
gemini -p "{question}" 2>/dev/null

# Multimodal (PDF/video/audio)
gemini -p "{prompt}" < /path/to/file.pdf 2>/dev/null

# JSON output
gemini -p "{question}" --output-format json 2>/dev/null

Note: --include-directories . is no longer needed for codebase analysis — Claude handles this directly with 1M context.

Language Protocol

•Ask Gemini in English
•Receive response in English
•Synthesize and apply findings
•Report to user in Japanese

Output Location

Save Gemini research results to:

code

.claude/docs/research/{topic}.md
.claude/docs/libraries/{library}.md

This allows Claude and Codex to reference the research later.

Task Templates

Library Research

bash

gemini -p "Research best practices for {library} in Python 2026.
Include:
- Installation and setup
- Common patterns and anti-patterns
- Known limitations and constraints
- Performance considerations
- Security concerns
- Code examples" 2>/dev/null

Latest Documentation Lookup

bash

gemini -p "Find the latest documentation for {library/API}.
Include:
- Current stable version
- Breaking changes from previous version
- New features
- Migration guide if applicable" 2>/dev/null

Multimodal Analysis

bash

# Video
gemini -p "Analyze video: main concepts, key points, timestamps" < tutorial.mp4 2>/dev/null

# PDF
gemini -p "Extract: API specs, examples, constraints" < api-docs.pdf 2>/dev/null

# Audio
gemini -p "Transcribe and summarize: decisions, action items" < meeting.mp3 2>/dev/null

See also: references/lib-research-task.md

Integration with Codex

Workflow	Steps
New feature	Gemini research → Codex design review
Library choice	Gemini comparison → Codex decision
/startproject	Agent Teams: Researcher (Gemini) ↔ Architect (Codex)

Why Gemini?

•Google Search: Latest information, official docs, best practices
•Multimodal: Native PDF/video/audio processing
•Web grounding: Verified facts with source URLs
•Shared context: Results saved for Claude/Codex to reference