npx skills add https://github.com/corygabrielsen/skills --skill brainstorm在见识足够多的想法之前,你无法判断哪个是最好的。强制拓宽思路胜过过早做出决定。
如果没有强制机制,你会在第一个看似可行的想法上止步——它可能是最好的,也可能不是。生成 N 个选项可以描绘出解决方案的全景,从而通过比较让优胜者脱颖而出。有时 #1 选项会轻松胜出。有时 #9 选项会将两个早期想法结合起来,形成单独任何一个都无法提出的方案。重点不在于后面的想法更好——而在于你不去探索就无法知道。
阅读问题。如果问题涉及代码,请阅读代码。产出:
问题: [一句话描述] 核心约束: [难点所在——矛盾、权衡、模糊之处] 上下文: [相关事实,2-3 个要点]
如果问题不明确,请在生成前使用 AskUserQuestion 进行澄清。
浏览相关代码、文档或现有方案。目标不是找到答案——而是为你的思维缓存加载素材,以便在生成阶段有原始材料进行重组。
在此阶段花费 2-3 分钟,而不是 20 分钟。你是在做准备,而不是解决问题。
精确生成 N 个选项。每个选项包含:
## 选项 K: [描述性名称]
[2-4 句解释。方法是什么?如何运作?]
[如果适用,提供代码示例——简短,仅展示关键思路]
**优点:** [最显著的优势]
**缺点:** [最大的缺点]
You don't know which idea is best until you've seen enough of them. Forced breadth beats premature commitment.
Without a forcing function, you stop at the first plausible idea — which might be the best one, or might not. Generating N options maps the solution landscape so the winner reveals itself through comparison. Sometimes #1 wins cleanly. Sometimes #9 combines two earlier ideas into something neither suggested alone. The point isn't that later is better — it's that you can't know until you've looked.
Read the problem. If it references code, read the code. Produce:
Problem: [one sentence] Core constraint: [what makes this hard — the tension, the trade-off, the ambiguity] Context: [relevant facts, 2-3 bullets]
If the problem is unclear, use AskUserQuestion to clarify before generating.
Skim relevant code, docs, or prior art. The goal is NOT to find the answer — it's to load your mental cache so the generation phase has raw material to recombine.
广告位招租
在这里展示您的产品或服务
触达数万 AI 开发者,精准高效
这 N 个选项必须覆盖以下维度。整个集合至少应涵盖 4 个:
| 维度 | 含义 |
|---|---|
| 抽象层次 | 底层 hack 与架构变更 |
| 依赖方向 | 添加、移除或替换依赖 |
| 故障处理理念 | 故障时关闭、故障时开放、建议性、重试或预防 |
| 范围 | 修复单行代码、修复函数、修复设计或移除功能 |
| 工具/机制 | 不同的语言、库、操作系统原语或平台特性 |
| 时机 | 问题发生前、发生时或发生后 |
| 逆向思维 | 做与直觉相反的事情 |
| 消除需求 | 完全消除对解决方案的需求 |
每个选项应是一个真正不同的方法,而不是同一主题的变体。如果你发现自己写下“类似于选项 K,但...”——那是一个变体,而不是一个新选项。需要更努力地思考。
生成多样化选项的技巧:
在进入排序阶段之前,扫描你的列表:
创建一个包含 3-5 个维度的比较表。选择能够区分选项的维度——如果所有选项得分相同,则舍弃该维度。
| # | 选项 | [维度 1] | [维度 2] | [维度 3] | 结论 |
|---|--------|---------|---------|---------|---------|
| 2 | 简单防护 | 良好 | 简单 | 零 | **首选** |
| 7 | 混合方法 | 更好 | 中等 | 低 | 备选 |
好的维度是特定于问题的。常见的维度包括:
突出显示前 3 名。解释为什么 #1 优于 #2。
陈述你的首选方案。用一句话说明其获胜原因。然后询问用户:
“哪些点引起了您的注意?”
他们可能看到你排名较低但符合你未知约束的选项。
-n,则 N = 10用户有一个问题。精确生成 N 个多样化的解决方案。从阶段 1:理解开始。
每周安装量
1
代码仓库
首次出现
今天
安全审计
安装于
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
Spend 2-3 minutes here, not 20. You're priming, not solving.
Produce exactly N options. Each option gets:
## Option K: [Descriptive Name]
[2-4 sentence explanation. What is the approach? How does it work?]
[Code example if applicable — short, showing the key insight only]
**Pro:** [strongest advantage]
**Con:** [biggest drawback]
The N options must span these axes. The full set should cover at least 4:
| Axis | What it means |
|---|---|
| Abstraction level | Low-level hack vs architectural change |
| Dependency direction | Add vs remove vs replace a dependency |
| Failure philosophy | Fail closed vs fail open vs advisory vs retry vs prevent |
| Scope | Fix the line vs fix the function vs fix the design vs remove the feature |
| Tool/mechanism | Different language, library, OS primitive, or platform feature |
| Timing | Before vs during vs after the problem occurs |
| Inversion | Do the opposite of what seems natural |
| Elimination | Remove the need for the solution entirely |
Each option should be a genuinely different approach, not a variation on a theme. If you catch yourself writing "similar to option K but..." — that's a variant, not a new option. Push harder.
Techniques for generating diverse options:
Before moving to ranking, scan your list:
Create a comparison table with 3-5 dimensions. Choose dimensions that discriminate — if all options score the same, drop that dimension.
| # | Option | [Dim 1] | [Dim 2] | [Dim 3] | Verdict |
|---|--------|---------|---------|---------|---------|
| 2 | Simple guard | Good | Trivial | Zero | **Top pick** |
| 7 | Hybrid approach | Better | Moderate | Low | Runner-up |
Good dimensions are problem-specific. Common ones:
Highlight the top 3. Explain why #1 beats #2.
State your top pick. One sentence on why it wins. Then ask the user:
"What catches your eye?"
They may see something you ranked low that resonates with constraints you don't know about.
-n not specifiedThe user has a problem. Generate exactly N diverse solutions. Start with Phase 1: Understand.
Weekly Installs
1
Repository
First Seen
Today
Security Audits
Installed on
zencoder1
amp1
cline1
openclaw1
opencode1
cursor1
React 组合模式指南:Vercel 组件架构最佳实践,提升代码可维护性
109,600 周安装