skill-builder（技能构建）

Skill Builder Skill

Purpose

This skill helps users create professional, well-structured Claude Code skills that follow official best practices. It guides through:

• Proper YAML frontmatter structure
• Effective skill descriptions (critical for auto-invocation)
• Complete instruction templates
• Tool restrictions and model selection
• File structure and organization

When to Use

This skill should be invoked when:

• User says "create a skill", "build a skill", "make a new skill"
• User wants to convert documentation or prompts into a skill
• User needs help structuring skill instructions
• User asks about skill best practices
• User wants to improve an existing skill's description
• User mentions "skill template" or "skill example"

Process

1. Understand the Intent

Ask the user:

• What should the skill do? (Purpose and capabilities)
• When should it be invoked? (Trigger conditions - critical!)
• What scope? (Personal or project)
• Any tool restrictions? (Security/safety)

2. Generate Skill Name

Based on the purpose, suggest a name that is:

• Lowercase with hyphens
• Descriptive and clear
• Max 64 characters
• Examples: code-reviewer, api-builder, database-optimizer

3. Craft Effective Description

This is the most critical part! The description determines when Claude invokes the skill.

Requirements:

• Max 1024 characters
• Must include WHAT the skill does
• Must include WHEN to use it
• Should include specific trigger keywords
• Should be action-oriented

Good Example:

Use when reviewing code for security vulnerabilities, performance issues, and best practices. Automatically checks for SQL injection, XSS, CSRF, and common bugs. Invoke for pull request reviews or code audits.

Bad Example:

Reviews code

4. Create Skill Structure

Generate a complete SKILL.md file with:

---
name: skill-name
description: [Effective description with WHAT and WHEN]
allowed-tools: [Optional - only if restrictions needed]
model: [Optional - sonnet/opus/haiku, or omit for default]
---

# [Skill Name]

## Purpose

[Clear explanation of what this skill does and why it's useful]

## When to Use

This skill should be invoked when:
- [Specific trigger condition 1]
- [Specific trigger condition 2]
- [Specific trigger condition 3]

## Process

1. **Step 1**: [First action]
   - [Detail or substep]

2. **Step 2**: [Second action]
   - [Detail or substep]

3. **Step 3**: [Final action]
   - [Detail or substep]

## Output Format

[Describe what format the output should take]

## Best Practices

- [Best practice 1]
- [Best practice 2]
- [Best practice 3]

## Examples

### Example 1: [Scenario name]

**Input**: [What the user provides]

**Process**: [How the skill handles it]

**Output**: [What the skill produces]

## Error Handling

- **[Error type]**: [How to handle]
- **[Error type]**: [How to handle]

## Notes

[Any additional context, warnings, or tips]

5. Determine File Location

Personal skills (available across all projects):

~/.claude/skills/skill-name/SKILL.md

Project skills (team-shared, version controlled):

.claude/skills/skill-name/SKILL.md

6. Create Supporting Files

If needed, create:

• README.md - Installation and usage docs
• scripts/ - Executable scripts
• materials/ - Reference materials, templates
• .gitignore - For project skills

7. Validate the Skill

Check:

• ✓ Name is lowercase-with-hyphens, max 64 chars
• ✓ Description includes WHAT and WHEN, max 1024 chars
• ✓ YAML frontmatter is valid
• ✓ Instructions are clear and comprehensive
• ✓ Examples are provided
• ✓ "When to Use" section has specific triggers

Output Format

When creating a skill, provide:

1. Skill metadata:

- Name: skill-name - Location: path/to/skill/ - Scope: Personal or Project

1. Complete SKILL.md file content
2. README.md content (for documentation)
3. Installation instructions: # Commands to create the skill
4. Testing guidance:

- How to verify it loaded - Example requests that should trigger it - How to refine if not auto-invoked

Best Practices

Description Writing

Include trigger keywords:

• "Use when...", "For...", "Invoke when..."
• Specific verbs: "reviewing", "building", "analyzing", "generating"
• Domain terms: "API", "database", "security", "tests"

Be specific:

• ❌ "Helps with code"
• ✅ "Use when reviewing JavaScript code for security vulnerabilities, performance issues, and React best practices"

Tool Restrictions

Only restrict tools when necessary:

# Read-only analysis skill
allowed-tools: Read, Grep, Glob

# Can modify files
allowed-tools: Read, Write, Edit, Grep, Glob

# Needs to run commands
allowed-tools: Read, Write, Edit, Bash

# Full access (omit field)

Model Selection

# Fast, efficient (simple tasks)
model: haiku

# Balanced (most tasks)
model: sonnet

# Most capable (complex reasoning)
model: opus

# Inherit from session (omit field)

Scope Selection

Choose Personal when:

• Experimental or personal workflow
• Not relevant to team
• Contains personal preferences

Choose Project when:

• Team should use it
• Part of development workflow
• Standardizes team practices

Examples

Example 1: Create Code Review Skill

User Request: "Create a skill for reviewing pull requests"

Response:

I'll create a code-reviewer skill for comprehensive pull request reviews.

Skill Metadata:

• Name: code-reviewer
• Scope: Project (for team use)
• Location: .claude/skills/code-reviewer/

Creating SKILL.md:

[Generate complete SKILL.md with frontmatter and all sections]

Installation:

mkdir -p .claude/skills/code-reviewer
# [Write SKILL.md content]

Testing: After restarting Claude Code, try:

• "Review this pull request"
• "Check this code for security issues"
• "Analyze this PR for problems"

Refinement: If not auto-invoked, add more trigger keywords to the description like "code review", "PR review", "security audit".

Example 2: Create Database Migration Skill

User Request: "I need a skill for generating database migrations"

Response:

I'll create a db-migration-generator skill that creates safe database migration files.

[Follow same pattern as Example 1]

Example 3: Improve Existing Skill

User Request: "My skill isn't being invoked automatically"

Response:

Let me check your skill's description. The description is how Claude decides when to invoke the skill.

Current description: [Read from SKILL.md]

Issues:

• Too vague - doesn't specify WHEN to use
• Missing trigger keywords
• Doesn't explain capabilities

Improved description:

description: Use when [specific trigger]. Handles [capabilities]. Invoke for [use cases].

Next steps:

1. Update description with these changes
2. Restart Claude Code
3. Test with requests containing trigger keywords

Error Handling

• Name conflicts: Suggest alternative names
• Description too vague: Provide examples of specific descriptions
• Invalid YAML: Show correct format with proper spacing
• Missing sections: Add template sections
• Too complex: Break into multiple skills
• Tool access issues: Recommend appropriate allowed-tools

Notes

• The description is THE MOST IMPORTANT field - it determines auto-invocation
• Skills are discovered automatically when Claude Code starts
• Use /skills command to see loaded skills
• Restart Claude Code after creating/modifying skills
• Test with explicit requests first: "Use the [skill-name] skill to..."
• Project skills override personal skills with same name
• Skills can reference files using relative paths
• Use forward slashes (/) for cross-platform compatibility

Common Skill Types

Analysis Skills (Read-only)

allowed-tools: Read, Grep, Glob

Examples: code-reviewer, security-auditor, performance-analyzer

Generation Skills (Write access)

# Omit allowed-tools or include Write

Examples: api-builder, test-generator, migration-creator

Automation Skills (With scripts)

# Include Bash for script execution

Examples: deployment-helper, build-optimizer

Reference Skills (With materials)

# Include materials/ directory with templates

Examples: brand-checker, style-guide-validator

Validation Checklist

Before finishing, verify:

• Name is valid (lowercase-hyphens, max 64 chars)
• Description is effective (includes WHAT and WHEN, max 1024 chars)
• YAML frontmatter is properly formatted
• "When to Use" section has 3+ specific triggers
• At least 1 detailed example provided
• Instructions are clear and actionable
• File paths use forward slashes
• Supporting files created if needed
• Installation instructions provided
• Testing guidance included

Templates Available

Reference these templates from this skill's directory:

1. code-reviewer - Security and quality reviews
2. api-builder - REST/GraphQL endpoint creation
3. test-generator - Test suite generation
4. documentation-writer - Technical documentation
5. performance-optimizer - Performance analysis

Access templates at: .claude/skills/skill-builder/templates/

Resources

• Official Skills Documentation: ../../docs/claude/skills/overview.md
• Skill Builder Tools: .claude/skills/skill-builder/build-skill.sh
• Validator: .claude/skills/skill-builder/utils/validate-skill.sh
• Quick Start Guide: .claude/skills/skill-builder/QUICK_START.md