AgentSkillsCN

meta-prompt-engineer

运用成熟的提示工程技巧，生成高质量的提示。 适用场景： - “撰写提示”、“创建提示”、“制作提示” - “元提示”、“为 X 编写提示” - “改进这个提示”、“优化提示” - 用户需要为 AI 任务拟定具体指令时使用。 不适用场景： - 仅提出问题（而非创建提示） - 执行已有提示 - 与提示无关的任务

SKILL.md
--- frontmatter
name: meta-prompt-engineer
description: |
  Generate high-quality prompts using proven prompt engineering techniques.
  
  USE WHEN:
  - "write a prompt", "create prompt", "make a prompt"
  - "meta-prompt", "prompt for X"
  - "improve this prompt", "optimize prompt"
  - User needs to craft instructions for AI tasks
  
  DO NOT USE WHEN:
  - Simply asking questions (not creating prompts)
  - Executing existing prompts
  - Non-prompt-related tasks

Meta Prompt Engineer

Generate structured, effective prompts for AI systems using proven techniques from Anthropic, OpenAI, and academic research.

Workflow Routing

IntentWorkflow
Create new prompt from scratchworkflows/generate-prompt.md
Improve/optimize existing promptworkflows/improve-prompt.md

Core References

ReferencePurpose
Techniques CatalogAll prompt engineering techniques
TemplatesReady-to-use prompt structures
Anti-PatternsMistakes to avoid

The 80/20 Rule

Three principles that solve 80% of prompt quality issues:

  1. •

    Goal + Constraints Upfront

    • •State what "done" looks like
    • •Define boundaries and limitations
  2. •

    1-3 Examples

    • •Format beats adjectives
    • •Show, don't tell
  3. •

    Force Structured Output

    • •JSON, bullets, tables, rubric
    • •Explicit format requirements

Prompt Structure (Contract-Style)

code
1. ROLE       - Who the AI is (persona, expertise)
2. TASK       - What to do (specific action)
3. CONTEXT    - What you're given (inputs, background)
4. OUTPUT     - Format, constraints, requirements
5. EXAMPLES   - Show desired input→output
6. VERIFY     - How to check success

Quick Technique Reference

TechniqueWhen to UseExample
XML TagsStructure complex inputs<context>...</context>
Few-shotEnforce format/style1-3 input→output examples
Chain of ThoughtComplex reasoning"Think step by step"
Role/PersonaConsistent voice"You are a senior engineer"
PrefillForce output formatStart with { for JSON
ConstraintsPrevent unwanted behavior"Do NOT include..."

Success Criteria

Generated prompt should have:

  • • Clear goal statement (what success looks like)
  • • Explicit output format specification
  • • At least one example (input → output)
  • • Appropriate structure (XML tags or clear sections)
  • • Constraints/guardrails (what NOT to do)
  • • Context about the task domain

Output

This skill outputs:

  1. •Structured prompt ready for use
  2. •Rationale explaining technique choices (if requested)