---
name: global-thinker
description: Multi-lingual cognitive diversity engine. Forwards user prompts to 3 parallel sub-agents thinking in Hebrew, Mandarin, and Arabic, then synthesizes their distinct perspectives into one unified approach.
---

# Global-Thinker

## Purpose

Leverage cognitive diversity through language-driven thinking. Different languages encode different problem-solving frameworks, cultural heuristics, and logical structures. By routing the same prompt through Hebrew, Mandarin, and Arabic thinkers, we surface approaches that monolingual reasoning misses.

## When to Activate

**Triggers:**
- User invokes `/global-thinker` with a prompt
- User says "think about this globally", "get multilingual perspectives", or "global thinker mode" 

## Execution Flow

### Step 1: Translate the Prompt

Take the user's input and produce three faithful translations that preserve technical intent:

- **Hebrew** (עברית)
- **Mandarin Chinese** (中文)
- **Arabic** (العربية)

### Step 2: Launch 3 Parallel Agents

Spawn exactly 3 Agent calls in a SINGLE message (parallel execution). Each agent:
- Model: `opus` 
- Is read-and-output only (no file writes)
- Receives the prompt ONLY in its assigned language
- MUST respond ONLY in its assigned language

#### Agent 1 — Hebrew Thinker (עברית)

```
prompt: |
  אתה חושב ומגיב אך ורק בעברית. אל תכתוב מילה אחת באנגלית.
  
  המשימה שלך: [TRANSLATED_PROMPT_IN_HEBREW]
  
  תן את הגישה שלך, הפתרון שלך, או המשוב שלך. חשוב לעומק.
  התמקד ב: מבנה לוגי, ניתוח ביקורתי, וזיהוי הנחות סמויות.
  
  פורמט: תשובה מובנית עם כותרות ברורות. מקסימום 400 מילים.
description: "Hebrew perspective thinker — analytical reasoning"
model: opus
subagent_type: claude
```

#### Agent 2 — Mandarin Thinker (中文)

```
prompt: |
  你只能用中文思考和回答。不要写任何英文。
  
  你的任务: [TRANSLATED_PROMPT_IN_MANDARIN]
  
  给出你的方法、解决方案或反馈。深入思考。
  重点关注: 系统性思维、整体架构、各部分之间的关系与平衡。
  
  格式: 结构化回答，标题清晰。最多400字。
description: "Mandarin perspective thinker — systems thinking"
model: opus
subagent_type: claude
```

#### Agent 3 — Arabic Thinker (العربية)

```
prompt: |
  أنت تفكر وتجيب باللغة العربية فقط. لا تكتب أي كلمة بالإنجليزية.
  
  مهمتك: [TRANSLATED_PROMPT_IN_ARABIC]
  
  قدّم منهجك أو حلّك أو ملاحظاتك. فكّر بعمق.
  ركّز على: الدقة في التفاصيل، البنية المنطقية، والحالات الحدّية.
  
  الصيغة: إجابة منظّمة بعناوين واضحة. 400 كلمة كحد أقصى.
description: "Arabic perspective thinker — precision and edge cases"
model: opus
subagent_type: claude
```

### Step 3: Synthesize

After all 3 agents return, YOU (the orchestrator) must:

1. Read each agent's response (they will be in Hebrew, Mandarin, Arabic)
2. Identify the unique angle each one brought
3. Produce the final output in the user's original language

## Output Format

Present the final result to the user in this exact structure:

```
## Global-Thinker Analysis

### Perspectives

**Hebrew (analytical/critical):** [1-2 sentence summary of their core insight]
**Mandarin (systems/holistic):** [1-2 sentence summary of their core insight]  
**Arabic (precision/edge-cases):** [1-2 sentence summary of their core insight]

### Unified Approach

[The synthesized best approach, merging the strongest elements from
all three perspectives. This is the actionable output — concrete,
specific, and ready to execute. Written in the user's language.]
```

## Execution Instructions (for the orchestrating agent)

When this skill activates, execute these steps exactly:

### 1. Translate

Internally translate the user's prompt into Hebrew, Mandarin, and Arabic.
Do NOT show the translations to the user — they are internal.

### 2. Launch Agents (PARALLEL — single message, 3 Agent tool calls)

Launch all 3 in ONE message so they run concurrently:

```
Agent({
  description: "Hebrew cognitive perspective",
  model: "opus",
  prompt: "אתה חושב ומגיב אך ורק בעברית. אל תכתוב מילה אחת באנגלית.\n\nהמשימה שלך:\n[HEBREW TRANSLATION OF USER PROMPT]\n\nתן את הגישה שלך, הפתרון שלך, או המשוב שלך. חשוב לעומק.\nהתמקד ב: מבנה לוגי, ניתוח ביקורתי, וזיהוי הנחות סמויות.\nפורמט: תשובה מובנית עם כותרות ברורות. מקסימום 400 מילים."
})

Agent({
  description: "Mandarin cognitive perspective",
  model: "opus",
  prompt: "你只能用中文思考和回答。不要写任何英文。\n\n你的任务:\n[MANDARIN TRANSLATION OF USER PROMPT]\n\n给出你的方法、解决方案或反馈。深入思考。\n重点关注: 系统性思维、整体架构、各部分之间的关系与平衡。\n格式: 结构化回答，标题清晰。最多400字。"
})

Agent({
  description: "Arabic cognitive perspective",
  model: "opus",
  prompt: "أنت تفكر وتجيب باللغة العربية فقط. لا تكتب أي كلمة بالإنجليزية.\n\nمهمتك:\n[ARABIC TRANSLATION OF USER PROMPT]\n\nقدّم منهجك أو حلّك أو ملاحظاتك. فكّر بعمق.\nركّز على: الدقة في التفاصيل، البنية المنطقية، والحالات الحدّية.\nالصيغة: إجابة منظّمة بعناوين واضحة. 400 كلمة كحد أقصى."
})
```

### 3. Read All Three Responses

Once all 3 agents return, read their outputs. You understand all three
languages — extract the core reasoning, unique angles, and actionable
recommendations from each.

### 4. Present Unified Output

Compose the final output using the Output Format above. The unified
approach should not be a simple average — it should take the BEST
elements from each perspective and weave them into one coherent plan.

## Rules

1. **All 3 agents MUST run in parallel** — launch them in a single message
2. **Agents are read-only** — they analyze and respond, they do NOT write files
3. **No English in agent prompts** — each agent receives ONLY its target language
4. **Translations must be faithful** — preserve technical terms, intent, and nuance
5. **Final output in user's language** — the synthesis is always in whatever language the user wrote their prompt in
6. **400-word cap per agent** — keeps responses focused and synthesis manageable
7. **No raw foreign text shown to user** — only the TLDR summaries and unified approach

## Why This Works

- Hebrew's analytical tradition surfaces hidden assumptions and logical gaps
- Mandarin's holistic framing reveals system-level relationships others miss
- Arabic's precision orientation catches edge cases and boundary conditions
- The synthesis merges three genuinely different cognitive frames, not three copies of the same English-trained reasoning

## Example

**User prompt:** "Should we use a monorepo or polyrepo for our 4 microservices?"

**What happens internally:**
1. Translate to Hebrew, Mandarin, Arabic
2. Launch 3 parallel agents (each thinks in their language only)
3. Hebrew agent: argues from first principles, questions assumptions
4. Mandarin agent: maps the system relationships, considers balance
5. Arabic agent: examines edge cases, failure modes, precision of boundaries
6. Synthesize into one unified recommendation