prompt-engineer（提示工程师）

Prompt Engineer

The Prompt Engineer skill helps you craft, refine, and optimize prompts for Claude Code and other AI systems. It applies proven prompt engineering principles including clarity, specificity, context provision, and structural best practices to transform vague requests into effective AI instructions.

This skill analyzes existing prompts for weaknesses, suggests improvements based on prompt engineering research, and helps you build prompt libraries for recurring tasks. It's particularly valuable when you need consistent, high-quality AI outputs or want to maximize the effectiveness of complex multi-step AI workflows.

Whether you're creating one-off prompts or building reusable templates, this skill ensures your AI interactions are clear, actionable, and produce the results you need.

Core Workflows

Workflow 1: Analyze & Optimize Existing Prompt

1. Receive the current prompt from user
2. Analyze against prompt engineering principles:

- Clarity: Is the request unambiguous? - Specificity: Are outputs well-defined? - Context: Is necessary background provided? - Structure: Is the prompt well-organized? - Constraints: Are limitations clearly stated?

1. Identify weaknesses and improvement opportunities
2. Provide optimized version with explanations
3. Test improved prompt if requested
4. Iterate based on results

Workflow 2: Design New Prompt from Scratch

1. Clarify the goal: What outcome is needed?
2. Gather requirements:

- Target AI system capabilities - Output format requirements - Domain context needed - Edge cases to handle

1. Structure the prompt using proven patterns:

- Role/persona if beneficial - Clear task description - Specific constraints and requirements - Output format specification - Examples if complex

1. Draft initial version
2. Refine for clarity and completeness
3. Document usage guidelines

Workflow 3: Build Prompt Template Library

1. Identify recurring prompt patterns in workflow
2. Extract reusable components
3. Parameterize variable elements
4. Document template with:

- Purpose and use cases - Parameter descriptions - Example usage - Expected outputs

1. Test template with multiple scenarios
2. Store in organized library structure

Quick Reference

Best Practices

• Be Specific: Replace vague terms with concrete requirements

- Bad: "Make it better" - Good: "Increase response accuracy by providing 3 cited examples"

• Provide Context: Give AI the background it needs

- Include: Domain knowledge, target audience, constraints - Example: "For a technical audience familiar with React..."

• Structure Clearly: Use formatting to organize complex prompts

- Sections, bullets, numbered steps - Clear delineation between instructions and examples

• Define Success: Specify what good output looks like

- Format requirements (JSON, markdown, etc.) - Length constraints - Quality criteria

• Use Examples: Show don't just tell for complex outputs

- Provide 1-3 examples of desired output - Include edge cases if relevant

• Iterate: Prompts improve through testing

- Start simple, add complexity as needed - Test with edge cases - Refine based on actual outputs

• Separate Concerns: Don't mix multiple requests

- One clear goal per prompt - Chain prompts for multi-step workflows

• Constrain Appropriately: Set boundaries without over-constraining

- Specify limits (word count, format) - Allow flexibility where creativity helps

Advanced Techniques

Chain-of-Thought Prompting

Encourage step-by-step reasoning by asking AI to "think through" problems:

Before providing the final answer, work through:
1. What are the key factors?
2. What are the trade-offs?
3. What does the evidence suggest?
Then provide your conclusion.

Few-Shot Learning

Provide examples of input-output pairs:

Example 1: [input] → [output]
Example 2: [input] → [output]
Now apply the same pattern to: [new input]

Role-Based Prompting

Assign expertise or perspective:

As a senior React architect with 10 years of experience,
review this component for performance issues...

Constraint-Based Refinement

Use specific constraints to shape output:

Requirements:
- Maximum 3 paragraphs
- Include code examples
- Cite sources
- Use beginner-friendly language

Common Pitfalls to Avoid

• Assuming context the AI doesn't have
• Being too vague about desired output format
• Mixing multiple unrelated requests
• Over-complicating simple requests
• Not specifying constraints until after receiving output
• Forgetting to provide examples for complex patterns
• Using ambiguous language or jargon without definition