Back to Details

divergent-thinking-skill

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Divergent Thinking Skill

发散思维技能

Overview

概述

Produce unconventional but executable solutions by forcing structured reframing, inversion, and cross-domain synthesis. Use MCP tools deliberately to ground novel ideas in real evidence instead of speculation.
通过结构化重构、反向思考及跨领域整合，产出非常规但可落地的解决方案。刻意运用MCP工具，让新颖想法基于真实证据而非猜测。

Workflow

工作流程

  1. Absorb context and list explicit constraints.
  2. Detect anchoring and default assumptions before proposing solutions.
  3. Generate multiple divergent passes using distinct reasoning frames.
  4. Use MCP tools to verify assumptions, surface hidden evidence, and challenge weak ideas.
  5. Synthesize output into prioritized options with immediate first actions.
  1. 吸收上下文并列出明确约束条件。
  2. 在提出解决方案前，先识别锚定效应及默认假设。
  3. 使用不同推理框架生成多轮发散思考结果。
  4. 运用MCP工具验证假设、挖掘隐藏证据并质疑薄弱想法。
  5. 将输出内容整合为带有优先顺序的选项，并附上可立即执行的第一步行动。

Execution Protocol

执行协议

  1. Start with an "Anchoring Map" with three items:
  • Dominant narrative.
  • Missing information.
  • Assumptions treated as facts.
  1. Run five divergence passes:
  • Inversion: flip assumptions and test what becomes possible.
  • Constraint remix: keep goals, change constraints, and use only existing resources.
  • Cross-domain transfer: import one model from a distant field and map it to the problem.
  • Time shift: solve from future-backward and immediate-forward perspectives.
  • Stakeholder asymmetry: identify who benefits from inertia and how to re-route incentives.
  1. Force option quality gates:
  • Each option must include one concrete first action.
  • Each option must include one measurable signal within 24-72 hours.
  • Each option must include one risk and a stop condition.
  1. Rank options by:
  • Novelty with feasibility.
  • Time-to-signal.
  • Leverage per unit effort.
  1. 从包含三项内容的「锚定映射图」开始：
  • 主流叙事
  • 缺失信息
  • 被视为事实的假设
  1. 开展五轮发散思考：
  • 反向思考：推翻假设，测试可能性
  • 约束重构：保留目标，改变约束条件，仅使用现有资源
  • 跨领域迁移：引入一个来自遥远领域的模型并适配到当前问题
  • 时间转换：从未来倒推及从当下顺推两种视角解决问题
  • 利益相关方不对称：识别谁从惯性中获益，以及如何重新调整激励机制
  1. 执行选项质量检验：
  • 每个选项必须包含一个具体的第一步行动。
  • 每个选项必须包含一个可在24-72小时内验证的可衡量信号。
  • 每个选项必须包含一个风险点及停止条件。
  1. 按以下维度对选项排序：
  • 新颖性与可行性的平衡
  • 验证信号的获取时效
  • 单位投入的杠杆效应

MCP Tooling Rules

MCP工具使用规则

Use tools to increase divergence quality, not just to gather more text.
  1. Discover context first:
  • Use MCP resource discovery tools (
    list_mcp_resources
    , 
    list_mcp_resource_templates
    , 
    read_mcp_resource
    ) when available.
  • Use repository search (
    rg
    , symbol tools) to avoid assumptions.
  1. Challenge assumptions with external signals:
  • Use web/documentation tools only when facts are unstable or uncertain.
  1. Simulate executable paths:
  • Use browser or automation tools for UI/process verification when relevant.
  1. Separate evidence from inference:
  • Label what is observed, what is inferred, and what is speculative.
Detailed MCP patterns: 
references/mcp-tooling-patterns.md
.
运用工具提升发散思考的质量，而非单纯收集更多文本。
  1. 先挖掘上下文：
  • 若可用，使用MCP资源发现工具（
    list_mcp_resources
    、
    list_mcp_resource_templates
    、
    read_mcp_resource
    ）。
  • 使用仓库搜索工具（
    rg
    、符号工具）避免主观假设。
  1. 用外部信号质疑假设：
  • 仅当事实不稳定或存疑时，使用网页/文档工具。
  1. 模拟可执行路径：
  • 若相关，使用浏览器或自动化工具进行UI/流程验证。
  1. 区分证据与推论：
  • 标注观察到的内容、推论的内容及猜测的内容。
详细MCP模式：
references/mcp-tooling-patterns.md
。

Prompt Generation

提示词生成

Use the helper script to generate a universal divergence prompt:
bash
python3 scripts/build_divergence_prompt.py \
  --problem "Describe the real issue clearly" \
  --goal "Define the target outcome" \
  --constraint "Hard limitation 1" \
  --constraint "Hard limitation 2" \
  --source "chat history" \
  --source "repo files" \
  --tool "mcp resources" \
  --tool "web search"
If no script usage is desired, manually follow the same structure and keep outputs in this format:
  1. Hidden Problem Statement
  2. Assumptions to Break
  3. Five Divergent Options
  4. 24-72 Hour Experiments
  5. Risk/Stop Conditions
  6. Recommended Path + Fallback
使用辅助脚本生成通用发散性提示词：
bash
python3 scripts/build_divergence_prompt.py \
  --problem "Describe the real issue clearly" \
  --goal "Define the target outcome" \
  --constraint "Hard limitation 1" \
  --constraint "Hard limitation 2" \
  --source "chat history" \
  --source "repo files" \
  --tool "mcp resources" \
  --tool "web search"
若无需使用脚本，可手动遵循相同结构，并按以下格式输出：
  1. 隐藏问题陈述
  2. 需打破的假设
  3. 五个发散性选项
  4. 24-72小时实验方案
  5. 风险/停止条件
  6. 推荐路径+备选方案

Guardrails

约束规则

  1. Avoid purely theatrical prompts that reduce clarity.
  2. Avoid fabricated citations, legal claims, or technical capabilities.
  3. Keep at least one conservative baseline option for comparison.
  4. Prefer "evidence-backed unusual" over "random unusual."
  5. Never output ideas without immediate next actions.
  1. 避免使用纯粹花哨的提示词，以免降低清晰度。
  2. 避免编造引用、法律声明或技术能力。
  3. 至少保留一个保守的基准选项用于对比。
  4. 优先选择「有证据支撑的非常规方案」而非「随机的非常规方案」。
  5. 绝不输出无立即下一步行动的想法。

References

参考资料

  1. references/divergence-catalog.md
    : terminology and reusable divergence methods.
  2. references/mcp-tooling-patterns.md
    : concrete MCP tool mapping for discovery, challenge, and validation.
  1. references/divergence-catalog.md
    ：术语表及可复用的发散思考方法。
  2. references/mcp-tooling-patterns.md
    ：用于发现、质疑及验证的具体MCP工具映射。