Slash命令与Skills:快速构建工作流自动化

核心观点:Slash命令和Skills是Claude Code中最容易被忽视但最高效的功能。一套精心设计的Slash命令库可以将日常任务的成本降低70%。

关键词:Slash命令、Skills、工作流自动化、自定义命令、重复任务优化、团队协作


导读

你将学到:

  • Slash命令与Skills的本质区别和联系
  • 如何创建自定义Slash命令
  • 如何创建可复用的Skills
  • 参数化命令的高级用法
  • 构建个人和团队命令库
  • 常见的自动化模式
  • 性能优化和最佳实践

适合人群:中级到高级开发者,想要优化日常工作流的使用者

阅读时间:22分钟 | 难度:中级 | 实用度:5/5

前置知识

  • 已阅读本系列前5篇文章
  • 理解Markdown语法
  • 熟悉YAML配置格式

问题场景

你发现自己每天重复做相同的任务:

  • 创建新功能分支时,总是要重复解释项目结构
  • 做代码审查时,总是要列举相同的检查项
  • 部署前总是要运行相同的测试命令
  • 创建bug修复分支时,总是要说明相同的上下文

你想:“能不能把这些常见任务自动化,只需要一个命令就能执行?”

这正是Slash命令和Skills的用途。

为什么这很重要?

总时间节省 = 每个任务节省时间 × 每周重复次数 × 52 周

例如:

  • 代码审查任务:30分钟节省 × 2次/周 × 52周 = 52小时/年
  • 创建功能分支:10分钟节省 × 5次/周 × 52周 = 43小时/年
  • 部署验证:20分钟节省 × 1次/周 × 52周 = 17小时/年

总计:112小时/年 = 2.7周的工作时间


核心概念

Slash命令 vs Skills

这两个概念经常混淆。让我们澄清:

触发方式

定义位置

本质

本质

自定义自动化功能

Slash命令
/review

Skills
.claude/skills/

Slash命令

快速调用Claude的快捷方式

Skills

可复用的提示模板和工作流

特性 Slash命令 Skills
定义方式 命令文件 技能文件
触发方式 /命令名 /命令名 或自动
定义位置 .claude/commands/ .claude/skills/
主要用途 快速访问常用任务 提示模板和工作流
是否需要指定目标 通常需要 通常由Claude执行
可共享性 高(Git管理) 高(Git管理)
学习曲线 平缓 中等

简单理解

  • Slash命令 = 快捷菜单
  • Skills = 工作流模板

为什么很强大

没有命令库:
每次都要重新表述需求
Claude重新分析上下文
重复的工作流设计
平均每个任务:15-30分钟

有命令库:
一个命令调用
Claude已知道流程
自动执行标准工作流
平均每个任务:1-2分钟

效率提升:10-15倍

Slash命令详解

什么是Slash命令?

Slash命令是以 / 开头的快捷调用,让你快速访问预定义的工作流。

基本结构

.claude/commands/
├── review.md              # /review 代码审查
├── deploy.md              # /deploy 部署
├── create-feature.md      # /create-feature 创建功能
└── test.md                # /test 运行测试

创建Slash命令

最简单的命令
# .claude/commands/review.md

---
name: review
description: Review code for quality and best practices
---

Review the recent code changes for:
- Code quality and readability
- Performance issues
- Security vulnerabilities
- Best practices violations

Provide actionable feedback with priority levels.

使用方法

claude

你:/review
# Claude会审查最近的改动
带参数的命令
# .claude/commands/fix-issue.md

---
name: fix-issue
description: Fix a GitHub issue
---

Fix GitHub issue $ARGUMENTS following these steps:

1. Understand the issue requirements
2. Analyze the related code
3. Implement the fix
4. Add tests if applicable
5. Verify the fix works

Current issue: $ARGUMENTS

使用方法

你:/fix-issue 123

# Claude会修复issue #123
复杂命令:带条件和多步骤
# .claude/commands/prepare-release.md

---
name: prepare-release
description: Prepare a new release
disable-model-invocation: false
---

Prepare a new release for version $ARGUMENTS.

Steps:

1. Update version number
   - package.json version field
   - CHANGELOG.md

2. Run quality checks
   - npm run lint
   - npm run test
   - npm run build

3. Create release commit
   - git tag v$ARGUMENTS
   - Create release notes

4. Prepare deployment checklist
   - Pre-deployment checks
   - Post-deployment verification

Version: $ARGUMENTS

使用方法

你:/prepare-release 2.0.0

# Claude会指导整个发布流程

命令配置选项

---
name: command-name                    # 命令名(必需)
description: Command description      # 描述
disable-model-invocation: false       # 是否让Claude执行
user-invocable: true                  # 是否允许用户手动调用
model: inherit                        # 使用的模型(继承默认)
---

命令内容...

实战命令库

以下是一个完整团队的命令库设计:

命令1:功能开发模板
# .claude/commands/start-feature.md

---
name: start-feature
description: Start developing a new feature
---

Starting a new feature development. Follow these steps:

1. Create feature branch
   ````bash
   git checkout -b feature/$ARGUMENTS
  1. Update CLAUDE.md if needed

    • Add any new context about this feature
  2. Begin development

    • Reference CLAUDE.md for coding standards
    • Write tests first (TDD approach)
    • Commit frequently
  3. Before creating PR

    • Run npm run lint
    • Run npm run test
    • Run npm run build
    • Update CHANGELOG.md

Feature name: $ARGUMENTS


#### 命令2:代码审查模板

```markdown
# .claude/commands/code-review.md

---
name: code-review
description: Perform comprehensive code review
---

Reviewing code changes with the following checklist:

Functionality:
- Does code implement the intended feature?
- Are all edge cases handled?
- Is error handling appropriate?

Quality:
- Follows CLAUDE.md standards?
- Code is readable and maintainable?
- No obvious performance issues?
- Test coverage adequate?

Security:
- No SQL injection vulnerabilities?
- No XSS vulnerabilities?
- Authentication/authorization correct?
- Sensitive data properly handled?

Architecture:
- Consistent with project architecture?
- Not introducing technical debt?
- Appropriate abstractions?

Provide structured feedback with priority levels:
- CRITICAL: Must fix before merge
- IMPORTANT: Should fix
- SUGGESTION: Consider for future improvement
命令3:部署前检查
# .claude/commands/pre-deploy.md

---
name: pre-deploy
description: Verify code is ready for deployment
disable-model-invocation: true
---

Pre-deployment verification checklist:

1. Code Quality
   - npm run lint (no errors/warnings)
   - npm run type-check (no type errors)
   - npm run test (all tests pass)

2. Build Verification
   - npm run build (builds successfully)
   - Build output size reasonable?

3. Security Check
   - npm audit (no vulnerabilities)
   - Dependencies up-to-date?

4. Git Status
   - All changes committed?
   - Branch is up-to-date with main?
   - Ready for merge?

5. Release Notes
   - CHANGELOG.md updated?
   - Version number updated?

6. Final Approval
   - Code owner approved?
   - Tests automated?
   - Rollback plan documented?

Once all checks pass, ready to deploy!

Skills详解

什么是Skill?

Skill是一个更复杂的工作流定义,包含完整的提示、工具链和执行步骤。

基本结构

.claude/skills/
├── code-review/
│   └── SKILL.md
├── security-analysis/
│   └── SKILL.md
└── performance-optimization/
    └── SKILL.md

创建Skill

基础Skill模板
# .claude/skills/code-review/SKILL.md

---
name: code-review
description: Comprehensive code review specializing in quality and best practices
category: development
allowed-tools: Read, Grep, Glob, Bash
---

You are a senior code reviewer specializing in code quality, maintainability, and best practices.

When reviewing code:

1. Analyze the code changes
   - What problem does it solve?
   - What's the approach?

2. Check code quality
   - Readability
   - Maintainability
   - Adherence to standards

3. Identify issues
   - Bugs or logic errors
   - Performance problems
   - Security issues

4. Provide structured feedback
   - Critical issues (must fix)
   - Important issues (should fix)
   - Suggestions (nice to have)

Review the code thoroughly and provide actionable feedback.
高级Skill:带参数和条件
# .claude/skills/security-analysis/SKILL.md

---
name: security-analysis
description: Security-focused code analysis
category: security
allowed-tools: Read, Grep, Glob
context: fork
---

You are a security expert analyzing code for vulnerabilities.

Focus areas:
- Authentication and authorization
- Input validation
- SQL injection risks
- XSS vulnerabilities
- Data exposure
- Cryptography usage

Analysis process:

1. Identify security-sensitive components
2. Trace data flow
3. Identify potential vulnerabilities
4. Assess impact and likelihood
5. Recommend mitigations

Provide clear security assessment with:
- Vulnerability description
- Severity level (Critical/High/Medium/Low)
- Affected code location
- Recommended fix
Skill自动加载
# .claude/skills/perf-optimizer/SKILL.md

---
name: perf-optimizer
description: Performance optimization specialist
allowed-tools: Read, Bash, Grep
context: fork
---

You are a performance optimization specialist.

When analyzing performance:

1. Profile the code
   - Identify bottlenecks
   - Find hot paths
   - Measure current metrics

2. Propose optimizations
   - Expected improvement
   - Implementation complexity
   - Trade-offs

3. Provide implementation
   - Code changes
   - Verification steps
   - Testing approach

Optimization priorities:
1. Algorithmic improvements
2. Caching strategies
3. Resource utilization
4. Parallelization

参数化和高级用法

参数传递

单参数
# .claude/commands/create-api-endpoint.md

---
name: create-api-endpoint
description: Create a new API endpoint
---

Create a new $ARGUMENTS endpoint following project standards.

Include:
- Route definition
- Request validation
- Response formatting
- Error handling
- Unit tests
- API documentation

Endpoint: $ARGUMENTS

使用

你:/create-api-endpoint /api/users/profile
多参数
# .claude/commands/migrate-db.md

---
name: migrate-db
description: Create database migration
---

Create database migration: $ARGUMENTS

Migration type: (first argument)
- add-column
- create-table
- modify-constraint

Table/Column: (second argument)

Example:
/migrate-db add-column users email

使用

你:/migrate-db add-column users email

条件执行

# .claude/commands/full-deploy.md

---
name: full-deploy
description: Complete deployment with validation
disable-model-invocation: false
---

Deploying version $ARGUMENTS with full validation.

Pre-deployment:
- Run all tests
- Code quality checks
- Security scan

Deployment:
- Build production bundle
- Upload to CDN
- Update DNS if needed

Post-deployment:
- Health checks
- Performance monitoring
- Rollback procedure

Version: $ARGUMENTS

命令和Skill的组织策略

按功能分类

.claude/commands/
├── development/
│   ├── start-feature.md
│   ├── create-component.md
│   └── setup-environment.md
├── review/
│   ├── code-review.md
│   ├── security-review.md
│   └── performance-review.md
├── deployment/
│   ├── pre-deploy.md
│   ├── deploy.md
│   └── post-deploy.md
└── maintenance/
    ├── update-dependencies.md
    ├── cleanup.md
    └── refactor.md

按难度分类

简易命令(适合新手):
- /help - 获取帮助
- /test - 运行测试
- /lint - 代码检查

中等命令(需要理解):
- /code-review - 代码审查
- /create-feature - 创建功能
- /deploy - 部署

高级命令(需要专业知识):
- /refactor - 重构
- /optimize - 优化
- /migrate - 迁移

实战模式

模式1:开发工作流链

/start-feature
  → 创建功能分支

/code-review
  → 在完成前自审

/pre-deploy
  → 部署前检查

/deploy
  → 发布到生产

模式2:问题诊断链

/analyze-performance
  → 找出性能瓶颈

/profile-code
  → 详细的性能分析

/optimize
  → 生成优化建议

/implement-optimization
  → 实施优化

模式3:安全审查链

/security-scan
  → 扫描已知漏洞

/security-review
  → 手工安全审查

/fix-vulnerabilities
  → 修复发现的问题

/verify-fixes
  → 验证修复

团队协作

共享命令库

方式1:Git管理(推荐)
# 项目根目录
.claude/
├── commands/          # 团队共享的命令
├── skills/            # 团队共享的技能
└── CLAUDE.md          # 项目配置

# 在Git中提交
git add .claude/
git commit -m "chore: add code review command"
git push

# 其他团队成员
git pull              # 自动获得新命令
/code-review          # 立即可用
方式2:全局命令库
# 放在用户主目录
~/.claude/skills/
├── my-code-review/
├── my-deploy/
└── my-optimize/

# 对所有项目可用
/my-code-review
方式3:混合方式
项目级别命令 (.claude/commands/)
  - 项目特定的工作流
  - 与Git同步
  - 团队共享

全局Skill (~/.claude/skills/)
  - 通用的分析工具
  - 个人定制
  - 跨项目使用

权限管理

# 基础权限级别

开发者可用命令:
- /start-feature
- /code-review(自审)
- /test
- /lint

审查员额外权限:
- /approve-pr
- /security-review
- /performance-review

部署权限(仅限):
- /pre-deploy
- /deploy
- /post-deploy
- /rollback

高级技巧

技巧1:命令中调用其他命令

# .claude/commands/full-feature-dev.md

---
name: full-feature-dev
description: Complete feature development workflow
disable-model-invocation: false
---

Full feature development for $ARGUMENTS:

Step 1: Start feature branch

/start-feature $ARGUMENTS


Step 2: Develop feature
(You will implement the feature)

Step 3: Review your work

/code-review


Step 4: Prepare for deployment

/pre-deploy


Ready for production!

技巧2:动态提示注入

# .claude/commands/analyze.md

---
name: analyze
description: Analyze code with dynamic context
---

Analyzing $ARGUMENTS using context from CLAUDE.md:

Current standards:
!`cat CLAUDE.md | grep "编码标准" -A 10`

Project structure:
!`find . -maxdepth 2 -type d | head -20`

Analyze the code and provide insights based on project context.

技巧3:条件Skill选择

基于项目类型自动选择Skill:

# .claude/commands/review.md

---
name: review
description: Review code (auto-selects reviewer based on project)
---

Detect project type...

If Python project:
  Use /python-code-review

If JavaScript project:
  Use /js-code-review

If Go project:
  Use /go-code-review

Otherwise:
  Use /generic-code-review

常见模式和最佳实践

最佳实践1:明确的命令目的

不好:
/do-stuff          # 太模糊

好的:
/code-review       # 清晰的目的
/security-audit    # 具体的功能
/perf-optimize     # 明确的输出

最佳实践2:参数验证

# .claude/commands/create-endpoint.md

---
name: create-endpoint
description: Create API endpoint
---

Creating endpoint: $ARGUMENTS

Validation:
- Endpoint path provided? (e.g., /api/users)
- Method specified? (GET/POST/PUT/DELETE)

Proceed with endpoint creation...

最佳实践3:文档化

# .claude/commands/deploy.md

---
name: deploy
description: Deploy to production
---

Deploy command usage:

Example:
  /deploy v1.2.0

Prerequisites:
  - All tests passing
  - Code reviewed
  - CHANGELOG updated

Output:
  - Deployment status
  - Rollback instructions
  - Health check results

最佳实践4:错误处理

# .claude/commands/backup-db.md

---
name: backup-db
description: Backup database safely
---

Database backup procedure:

1. Pre-check
   - Is database accessible?
   - Is disk space sufficient?
   - Are permissions correct?

2. Backup
   - Create backup file
   - Verify backup integrity
   - Store securely

3. Verification
   - Test restore from backup
   - Confirm data integrity
   - Log backup details

If any step fails, provide detailed error message.

与其他Claude Code功能的集成

与Plan Mode的结合

# 1. 先用Plan Mode规划
claude --plan

你:我需要重构认证系统。请规划。

Claude:[生成详细计划]

# 2. 创建对应的Skill
.claude/skills/auth-refactor/SKILL.md
(基于Plan Mode的计划)

# 3. 用Skill执行
claude
你:/auth-refactor

与Hooks的结合

# .claude/settings.json

{
  "hooks": {
    "PostCommandExecution": [
      {
        "command": "/code-review",
        "action": "log",
        "file": ".claude/review-history.log"
      }
    ]
  }
}

与MCP的结合

# .claude/commands/github-review.md

---
name: github-review
description: Review GitHub PR
---

使用GitHub MCP获取PR信息:

PR信息:
!`gh pr view $ARGUMENTS --json title,body,files`

执行代码审查...

性能优化

缓存优化

# .claude/commands/test-quick.md

---
name: test-quick
description: Run only changed tests
---

Quick test run (only changed files):

1. Detect changed files
2. Run related tests only
3. Report results

Much faster than full test suite!

成本优化

# .claude/commands/lint-check.md

---
name: lint-check
description: Local lint check (no Claude cost)
disable-model-invocation: true
---

Run local linting:
```bash
npm run lint

No Claude execution, instant feedback!


总结与要点

命令库的收益

投入 收益 回报周期
创建5个命令(10小时) 每个任务节省15分钟 2-3周
创建Skill库(20小时) 每个任务节省30分钟 3-4周
团队命令库(50小时) 团队整体效率提升40% 2个月

一句话总结

精心设计的Slash命令和Skill库是团队生产力的倍增器——一次性投入,终身受益。

下一步行动

  1. 立即:为你最常做的任务创建3个Slash命令
  2. 这周:建立个人Skill库(至少2个)
  3. 下周:与团队分享,收集反馈
  4. 持续:持续扩展和改进命令库

推荐阅读

本系列相关文章

  • 上一篇:Plan Mode - 安全的探索
  • 下一篇:Hooks - 事件驱动的自动化
  • 后续:Subagents - 专业化的AI助手

官方资源

  • Claude Code文档 - Commands: https://code.claude.com/docs
  • Skills最佳实践: https://code.claude.com/docs/skills

社区资源

  • GitHub: awesome-claude-code-commands
  • Skill模板库: https://github.com/claude-code-skills

常见问题

Q: Slash命令和Skill有什么区别?
A: Slash命令是快捷调用,Skills是完整的工作流。命令用于快速任务,Skills用于复杂流程。

Q: 可以在命令中使用参数吗?
A: 可以,使用 $ARGUMENTS 来接收参数。例如:/create-api $ARGUMENTS

Q: 命令库应该提交到Git吗?
A: 是的。团队共享的命令应该提交到Git,确保所有人都使用最新版本。

Q: 如何在全局和项目级别共享命令?
A: 全局命令放在 ~/.claude/skills/,项目命令放在 .claude/commands/。两者都会被加载。

Q: 命令库的大小有限制吗?
A: 没有硬性限制,但建议保持在50个命令以内,避免混乱。可以分类组织。


最后的话

Slash命令和Skills代表了一种哲学:系统化地消除重复工作

每个开发者每天都在重复相同的工作。如果你能花30分钟设计一个命令,就能节省几个小时的重复工作,这笔投资总是值得的。

而且,好的命令库不仅能提升个人效率,更能帮助团队建立标准化的工作流。


恭喜!你现在已经掌握了Claude Code的个人效率优化工具。下一篇将讲解Hooks——在后台自动执行检查和验证,进一步提升整个开发流程的自动化水平。

Logo

小龙虾开发者社区是 CSDN 旗下专注 OpenClaw 生态的官方阵地,聚焦技能开发、插件实践与部署教程,为开发者提供可直接落地的方案、工具与交流平台,助力高效构建与落地 AI 应用

更多推荐