Prompt 设计最佳实践

Prompt 部分是 Skill 的「大脑」— 它定义了 Skill 的行为模式、工作流程和输出标准。

结构模板

# Skill Name

## When to Use
- Scenario A: when the user wants X
- Scenario B: when the user mentions Y

## When NOT to Use
- Don't use for Z — use [other-skill] instead

## Workflow
1. First, understand the user's goal by asking...
2. Then, analyze the code by...
3. Finally, produce output in the following format...

## Output Format
- Section 1: ...
- Section 2: ...
- Always include a summary

## Key Principles
- Principle 1: ...
- Principle 2: ...

## Common Mistakes to Avoid
- Mistake A: ... instead do ...
- Mistake B: ...

## Notes
- Edge case X: handle by...
- Limitation Y: this Skill cannot...

设计原则

1. 明确边界

好:

## When to Use
- Code review for correctness, readability, and maintainability
- General-purpose review of any code change

## When NOT to Use
- Security-specific audits — use /security-review instead
- Architecture-level review — use /improve-codebase-architecture

不好:

Use this Skill to review code.

2. 分步骤工作流

好:

## Workflow
1. Read the changed files to understand the scope
2. Check for: logic errors, edge cases, type safety
3. Check for: security vulnerabilities (XSS, injection, etc.)
4. Check for: performance concerns (N+1, memory leaks)
5. Check for: readability and naming quality
6. Produce a structured review with severity tags

不好:

Review the code and report issues.

3. 指定输出格式

## Output Format
### Critical (must fix)
- [file:line] Issue description → Suggested fix

### Should Fix
- [file:line] Issue description → Suggested fix

### Nice to Have
- [file:line] Issue description → Suggested fix

### Positive Findings
- What's done well: ...

4. 包含反面案例

## Common Anti-Patterns to Flag
- Catching and ignoring errors silently
- Using `any` type in TypeScript
- Hardcoded credentials or URLs
- Mutating props in React components
- Missing `key` prop in list renders

Prompt 长度指南

复杂度	长度	例子
简单	20-50 行	格式转换、简单代码生成
中等	50-150 行	代码审查、测试生成
复杂	150-300 行	安全审查、架构分析

超过 300 行考虑把细节放到 resources/ 中。

测试你的 Prompt

1. 触发测试: 用不同说法测试是否能正确触发
2. 边界测试: 确认「不该触发」的场景不会误触
3. 输出测试: 验证输出格式和内容质量
4. 协作测试: 确认与其他 Skill 的组合使用效果

下一步

• 工具与 Hooks — 在 Skill 中集成外部工具
• 完整示例 — 真实 Skill 案例