這個提示詞牛翻了 — 結構化思考 AI 提示詞框架

• URL: https://www.threads.com/share/BAOTgq7jMN/
• Date Saved: 2026-09-09
• Source: Threads (@yuqi._.0313 / 小鹿說AI)
• Tags: ai-engineering, productivity

Summary

10-slide carousel sharing a structured thinking prompt framework for AI (ChatGPT/Claude/Gemini). The post has two layers — a basic “autonomous agent” prompt (slides 0-4) and a more detailed structured framework (slides 5-9).

Core Prompt (slides 5-9 — the real content)

4-phase structured thinking framework:

1. 分析階段 (Analysis): Break problem into 3-5 sub-questions, identify hidden assumptions, mark cognitive blind spots
2. 發散階段 (Divergence): Generate 3+ solutions per sub-question from different angles, include one counter-intuitive option, add cross-domain analogies
3. 收斂階段 (Convergence): Evaluate with impact vs feasibility matrix, give Top 3 recommendations with reasoning, design execution steps for #1
4. 輸出格式 (Output): Comparison tables, bold key conclusions, include action checklist

3 Advanced Add-ons

• 堆疊思維鏈 (Chain of thought stacking): Summarize previous phase before starting next; self-correct if earlier analysis was wrong
• 批判層 (Devil’s advocate layer): Challenge own Top 3 proposals, find most likely failure points, provide mitigation measures
• 個人化調校 (Personalization): Add background, goals, constraints, style preference to prompt header

Usage Tips

• Don’t dump overly long questions — start with core question, follow up with details
• Redirect explicitly if AI goes off-topic
• Ask for simpler explanations or more specific examples
• Request summary at end of important conversations
• Works with ChatGPT, Claude, Gemini

Assessment

Standard “mega prompt” content — the framework is reasonable but nothing novel for anyone already familiar with structured prompting. The actual prompt in slides 2-3 is a basic autonomous agent system prompt. Slides 5-9 contain a more useful structured thinking template.