meta-prompt（元提示符）

Meta Prompt

Make Claude better at understanding you.

When to Use

• Complex task, unclear how to express it
• Want optimal results, not just good
• Creating reusable prompts (skills)
• Teaching Claude new patterns

Meta Prompt Pattern

Ask for the Prompt

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

Claude returns optimal prompt structure.
Then use that prompt.

Prompt Improvement

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

Claude refines your approach.

Constraint Discovery

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

Claude reveals what it needs to know.

Prompt Architecture

Structure

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)

Example

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

Prompt Templates

Feature Request

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

Bug Investigation

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

Code Review

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

Architecture Decision

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

Skill Creation

From Repeated Prompts

Notice: Same prompt used 3+ times

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

Claude creates SKILL.md.

Skill Structure

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

# Skill Name

[The optimized prompt content]

Prompt Debugging

Not Getting Good Results?

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

Too Verbose?

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

Too Vague?

"Make this more specific:
[vague prompt]"

Advanced Patterns

Chain of Thought

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

Few-Shot Learning

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

Now do: Input: [C] → Output: ?"

Role Assignment

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

Prompt Economics

Token Efficiency

Short prompt + clear constraints > Long prompt + vague intent

Reuse

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

90% savings.

Quick Commands

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

Use when: Optimizing prompts, creating skills, complex tasks