🇨🇳中文介绍

元提示

让 Claude 更好地理解你。

使用时机

• 复杂任务，不清楚如何表述
• 想要最优结果，而不仅仅是好结果
• 创建可复用的提示（技能）
• 教 Claude 新的模式

元提示模式

请求提示

"What's the best way to ask you to [task]?"

Claude 返回最优的提示结构。
然后使用该提示。

提示改进

"Here's my prompt: [prompt]
How can I improve it for better results?"

Claude 优化你的方法。

约束发现

"What constraints would help you [task] better?"

Claude 揭示它需要知道的信息。

提示架构

结构

1. 上下文（你正在做什么）
2. 目标（你想要什么）
3. 约束（要避免/包含什么）
4. 格式（如何返回结果）
5. 示例（如果有帮助的话）

示例

Context: iOS app with SwiftUI
Goal: Add dark mode toggle to settings
Constraints:
- Use existing Theme system
- Support system preference
- Persist user choice
Format: Complete Swift code with comments

提示模板

功能请求

"In [codebase], add [feature].
Use existing [patterns].
Follow [conventions].
Return: [deliverables]"

错误调查

"[Symptom] is happening.
Expected: [expected behavior]
Actual: [actual behavior]
Find root cause and fix."

代码审查

"Review [code] for:
- [Concern 1]
- [Concern 2]
- [Concern 3]
Return issues ranked by severity."

架构决策

"Need to choose between [A] and [B] for [use case].
Consider: [factors]
Recommend with reasoning."

技能创建

从重复提示中创建

Notice: Same prompt used 3+ times

"Turn this into a reusable skill:
[your repeated prompt]"

Claude 创建 SKILL.md。

技能结构

---
name: Skill Name
description: When to use this
allowed-tools: [tools]
model: [model]
---

# Skill Name

[The optimized prompt content]

提示调试

没有得到好结果？

"My prompt: [prompt]
I'm getting: [current results]
I want: [desired results]
What's wrong with my prompt?"

太冗长？

"Compress this prompt while keeping quality:
[long prompt]"

太模糊？

"Make this more specific:
[vague prompt]"

高级模式

思维链

"Think through this step by step:
1. First, understand [aspect]
2. Then, identify [aspect]
3. Finally, implement [solution]"

少样本学习

"Here are examples of what I want:
Input: [A] → Output: [X]
Input: [B] → Output: [Y]

Now do: Input: [C] → Output: ?"

角色分配

"Act as a [role] reviewing [thing].
Focus on [concerns] a [role] would have."

提示经济学

令牌效率

Short prompt + clear constraints > Long prompt + vague intent

复用

Good prompt used 100 times = Skill
Skill invocation = 50 tokens
Manual prompt = 500 tokens

90% savings.

快捷命令

"Best prompt for [task]?"
"Improve: [prompt]"
"Compress: [prompt]"
"Turn into skill: [prompt]"

使用时机：优化提示、创建技能、复杂任务

每周安装次数

–

来源

smithery.ai/ski…a-prompt

首次出现

–

"What's the best way to ask you to [task]?"

Claude returns optimal prompt structure.
Then use that prompt.

"Here's my prompt: [prompt]
How can I improve it for better results?"

Claude refines your approach.

"What constraints would help you [task] better?"

Claude reveals what it needs to know.

1. Context (what you're working on)
2. Goal (what you want)
3. Constraints (what to avoid/include)
4. Format (how to return results)
5. Examples (if helpful)