/sw-plugin-dev:skill-create

Create new Claude Code skill with proper YAML frontmatter, directory structure, and activation triggers. Interactive wizard for skill creation with validation and best practices.

Interactive Wizard: From concept to production-ready skill with proper structure and validation.

You are helping the user create a new Claude Code skill following best practices and the required structure.

Critical Structure Rules (Get This Right First!)

1. Directory Structure

MANDATORY format:

~/.claude/skills/           ← Personal skills
└── your-skill-name/        ← MUST be in subdirectory!
    └── SKILL.md            ← Must be named exactly SKILL.md

.claude/skills/             ← Project skills (in project root)
└── your-skill-name/        ← MUST be in subdirectory!
    └── SKILL.md            ← Must be named exactly SKILL.md

Common mistake: Putting SKILL.md directly in ~/.claude/skills/ ❌

2. SKILL.md Format (Mandatory YAML Frontmatter)

---
name: your-skill-name
description: What it does AND when to use it. Include trigger keywords users might say.
---

# Your Skill Title

Rest of your markdown content here...

Critical rules:

• Opening --- MUST be on line 1
• name field: lowercase, hyphens only, max 64 chars
• description field: max 1024 chars, MUST include activation triggers
• Closing --- before any markdown content
• No YAML frontmatter = skill won't work!

Steps

Step 1: Gather Skill Information

Ask the user for the following information (use AskUserQuestion if appropriate):

1. Skill Name:

   • Lowercase, hyphens only
   • Max 64 characters
   • Example: python-data-science, kubernetes-expert, react-hooks

2. Skill Purpose:

   • What expertise does this skill provide?
   • What questions should it answer?
   • Example: "Python data science best practices"

3. Activation Triggers:

   • Keywords users might say
   • Commands or patterns to detect
   • Example: "pandas, numpy, matplotlib, jupyter notebooks, Python data analysis"

4. Skill Location:

   • Personal skill: ~/.claude/skills/
   • Project skill: .claude/skills/ (in project root)

5. Optional Tool Restrictions:

   • Should this skill be read-only?
   • Allowed tools: Read, Grep, Glob, WebSearch, etc.
   • Leave empty for no restrictions

Step 2: Validate Skill Name

Validation rules:

# Check format (lowercase, hyphens only)
if [[ ! "$skill_name" =~ ^[a-z0-9-]+$ ]]; then
  echo "❌ Invalid name: must be lowercase with hyphens only"
  exit 1
fi

# Check length (max 64 chars)
if [ ${#skill_name} -gt 64 ]; then
  echo "❌ Invalid name: must be 64 characters or less"
  exit 1
fi

# Check no consecutive hyphens
if [[ "$skill_name" =~ -- ]]; then
  echo "❌ Invalid name: no consecutive hyphens allowed"
  exit 1
fi

# Check doesn't start/end with hyphen
if [[ "$skill_name" =~ ^- ]] || [[ "$skill_name" =~ -$ ]]; then
  echo "❌ Invalid name: cannot start or end with hyphen"
  exit 1
fi

Step 3: Generate Skill Description

Requirements:

• Max 1024 characters
• Must include activation triggers
• Must explain when to use the skill
• Include keyword variations (e.g., "k8s, kubernetes, kubectl")

Template:

[What the skill does]. [When to use it]. Activates for: [keyword1], [keyword2], [keyword3], [phrase1], [phrase2].

Examples:

# Command-based skill
description: Kubernetes expert. Explains kubectl commands, pod management, deployments. Activates for: kubectl, k8s, kubernetes, pods, deployments, services, namespaces.

# Framework skill
description: React expert for hooks, components, state management. Activates for: React, useState, useEffect, components, JSX, props, virtual DOM, React hooks.

# Project-specific skill
description: MyCompany API documentation. Explains endpoints, authentication, rate limits. Activates for: MyCompany API, /api/v1, authentication token, rate limit, API key.

Step 4: Create Skill Content Structure

Recommended sections:

---
name: your-skill-name
description: [Generated description with triggers]
---

# [Skill Title]

## What I Know

[Bullet list of expertise areas]
- Topic 1
- Topic 2
- Topic 3

## When to Use This Skill

Ask me about:
- "How do I [use case 1]..."
- "What's the best way to [use case 2]..."
- "[Topic] tips and best practices"

## Key Concepts

### [Concept 1]

[Explanation]

### [Concept 2]

[Explanation]

## Examples

### [Example 1: Common Task]

```[language]
[code example]

Explanation: [Why this works]

[Example 2: Advanced Usage]

[code example]

Explanation: [When to use this pattern]

Best Practices

1. ✅ [Practice 1]: [Explanation]
2. ✅ [Practice 2]: [Explanation]
3. ⚠️ [Anti-pattern to avoid]: [Why to avoid]

Common Patterns

[Pattern 1]

[Implementation details]

[Pattern 2]

[Implementation details]

Resources

• [Documentation link]
• [Tutorial link]
• [Reference link]

### Step 5: Handle Optional Tool Restrictions

**If user wants read-only skill**:

```yaml
---
name: documentation-helper
description: Documentation expert for project docs
allowed-tools: Read, Grep, Glob, WebSearch
---

Available tools:

• Read - Read files
• Write - Create/overwrite files
• Edit - Edit existing files
• Grep - Search file contents
• Glob - Find files by pattern
• Bash - Execute bash commands
• WebSearch - Search the web
• WebFetch - Fetch web content
• TodoWrite - Manage todos
• AskUserQuestion - Ask user questions

Tool restriction patterns:

# Read-only (no modifications)
allowed-tools: Read, Grep, Glob, WebSearch

# Documentation expert (can create docs)
allowed-tools: Read, Grep, Glob, Write, Edit

# Analysis expert (no file modifications)
allowed-tools: Read, Grep, Glob, Bash

# Full access (no restrictions)
# (omit allowed-tools field)

Step 6: Create Skill Directory and File

Implementation:

# Determine full path
if [ "$location" = "personal" ]; then
  skill_dir="$HOME/.claude/skills/$skill_name"
else
  skill_dir="$PWD/.claude/skills/$skill_name"
fi

# Create directory
mkdir -p "$skill_dir"

# Create SKILL.md
cat > "$skill_dir/SKILL.md" <<'EOF'
---
name: $skill_name
description: $description
$allowed_tools_line
---

# $skill_title

$skill_content
EOF

echo "✅ Skill created at: $skill_dir/SKILL.md"

Step 7: Validate Skill Structure

Validation checklist:

# 1. Check SKILL.md exists
if [ ! -f "$skill_dir/SKILL.md" ]; then
  echo "❌ SKILL.md not found"
  exit 1
fi

# 2. Check YAML frontmatter
if ! head -1 "$skill_dir/SKILL.md" | grep -q '^---$'; then
  echo "❌ SKILL.md must start with '---'"
  exit 1
fi

# 3. Check required fields
if ! grep -q '^name:' "$skill_dir/SKILL.md"; then
  echo "❌ Missing 'name' field in YAML frontmatter"
  exit 1
fi

if ! grep -q '^description:' "$skill_dir/SKILL.md"; then
  echo "❌ Missing 'description' field in YAML frontmatter"
  exit 1
fi

# 4. Check closing ---
if ! sed -n '2,10p' "$skill_dir/SKILL.md" | grep -q '^---$'; then
  echo "❌ Missing closing '---' in YAML frontmatter"
  exit 1
fi

# 5. Check description length
desc_length=$(grep '^description:' "$skill_dir/SKILL.md" | sed 's/^description: *//' | wc -c)
if [ "$desc_length" -gt 1024 ]; then
  echo "⚠️ Warning: Description is ${desc_length} characters (max 1024)"
fi

echo "✅ Skill structure validated successfully"

Step 8: Provide Next Steps

User instructions:

✅ Skill created successfully!

📁 Location: $skill_dir/SKILL.md

📝 Next steps:

1. Review the skill content and customize as needed
2. Test the skill by asking a trigger question
3. Restart Claude Code to load the skill
4. Verify activation with a test question

🔄 To restart Claude Code:
   - Close and reopen the application
   - OR use /restart command (if available)

🧪 Test your skill:
   Ask: "[example trigger question based on description]"

📚 Documentation:
   - Skill format: ~/CLAUDE.md (your quick reference)
   - Activation patterns: Description keywords determine when skill loads
   - Tool restrictions: Optional 'allowed-tools' field limits capabilities

⚠️ Common issues:
   - Skill doesn't activate → Check description has clear triggers
   - YAML errors → Verify frontmatter format (opening/closing ---)
   - File not found → Ensure SKILL.md is in subdirectory

Examples

Example 1: Python Data Science Skill

Input:

• Name: python-data-science
• Purpose: "Python best practices for data science projects"
• Triggers: "pandas, numpy, matplotlib, jupyter notebooks, Python data analysis"
• Location: Personal
• Tool restrictions: None

Output (~/.claude/skills/python-data-science/SKILL.md):

---
name: python-data-science
description: Python best practices for data science projects. Explains pandas DataFrame operations, NumPy arrays, matplotlib visualizations, and Jupyter notebook workflows. Activates for: pandas, numpy, matplotlib, jupyter notebooks, Python data analysis, data science, scientific computing.
---

# Python Data Science Skill

## What I Know

- Pandas DataFrame operations and transformations
- NumPy array manipulation and vectorization
- Matplotlib and Seaborn visualizations
- Jupyter notebook best practices
- Data cleaning and preprocessing
- Statistical analysis and modeling

## When to Use This Skill

Ask me about:
- "How do I use pandas to..."
- "What's the best way to visualize..."
- "Python data analysis tips"
- "NumPy array operations"
- "Jupyter notebook shortcuts"

## Key Concepts

### Pandas DataFrames

DataFrames are the core data structure for tabular data in Python:
- Rows and columns with labels
- Rich indexing and selection capabilities
- Built-in aggregation and transformation methods

### NumPy Arrays

Efficient numerical computing with multi-dimensional arrays:
- Vectorized operations (avoid loops!)
- Broadcasting for element-wise operations
- Linear algebra and statistical functions

## Examples

### Example 1: Loading and Exploring Data

```python
import pandas as pd

# Load CSV file
df = pd.read_csv('data.csv')

# Quick exploration
print(df.head())        # First 5 rows
print(df.info())        # Column types and nulls
print(df.describe())    # Statistical summary

Explanation: Always start with exploration to understand your data structure, types, and missing values.

Example 2: Data Cleaning

# Handle missing values
df_clean = df.dropna()              # Drop rows with any nulls
df_filled = df.fillna(0)            # Fill nulls with 0
df_ffill = df.fillna(method='ffill')  # Forward fill

# Remove duplicates
df_unique = df.drop_duplicates()

Explanation: Choose the right strategy based on your data and analysis goals.

Best Practices

1. ✅ Use vectorized operations: Avoid Python loops, use pandas/numpy methods
2. ✅ Chain operations: df.groupby('col').agg('mean').sort_values()
3. ✅ Copy data explicitly: Use .copy() to avoid SettingWithCopyWarning
4. ⚠️ Avoid iterrows(): Extremely slow, use .apply() or vectorization instead

Common Patterns

Groupby-Aggregation

df.groupby('category').agg({
    'sales': 'sum',
    'quantity': 'mean',
    'price': ['min', 'max']
})

Time Series Resampling

df.set_index('date').resample('M').mean()

Resources

• Pandas Documentation
• NumPy User Guide
• Matplotlib Gallery

### Example 2: Read-Only Documentation Skill

**Input**:
- Name: `project-docs-helper`
- Purpose: "Project documentation expert"
- Triggers: "documentation, docs, readme, API reference"
- Location: Project (`.claude/skills/`)
- Tool restrictions: Read, Grep, Glob, WebSearch

**Output** (`.claude/skills/project-docs-helper/SKILL.md`):

```yaml
---
name: project-docs-helper
description: Project documentation expert. Helps find and explain documentation, API references, and README files. Read-only access for safe documentation browsing. Activates for: documentation, docs, readme, API reference, how to use, user guide.
allowed-tools: Read, Grep, Glob, WebSearch
---

# Project Documentation Helper

## What I Know

- Location of all project documentation
- API reference structure
- README and getting started guides
- Architecture documentation
- Contributing guidelines

## When to Use This Skill

Ask me about:
- "Where is the documentation for..."
- "How do I use [feature]..."
- "What APIs are available..."
- "Getting started with this project"

## Documentation Structure

This project follows standard documentation patterns:
- `/docs/` - Main documentation folder
- `README.md` - Getting started guide
- `API.md` - API reference
- `CONTRIBUTING.md` - Contribution guidelines
- `ARCHITECTURE.md` - System design

## How to Find Information

### Search by Topic
I can search through all documentation files to find information about specific topics.

### Navigate by Structure
I understand the documentation hierarchy and can guide you to the right files.

### Explain Concepts
I can read and explain complex documentation sections in simpler terms.

## Best Practices

1. ✅ **Check README first**: Start with README.md for project overview
2. ✅ **Use search**: Ask me to search docs instead of browsing manually
3. ✅ **Verify versions**: Documentation may be version-specific
4. ⚠️ **Read-only**: This skill cannot modify documentation

Validation and Testing

Post-Creation Validation

Run these checks after creation:

# 1. File exists in correct location
ls "$skill_dir/SKILL.md"

# 2. YAML frontmatter is valid
head -10 "$skill_dir/SKILL.md"

# Should see:
# ---
# name: your-skill-name
# description: ...
# ---

# 3. No syntax errors
node -e "
const fs = require('fs');
const yaml = require('js-yaml');
const content = fs.readFileSync('$skill_dir/SKILL.md', 'utf-8');
const match = content.match(/^---\\n([\\s\\S]*?)\\n---/);
if (!match) {
  console.error('❌ No YAML frontmatter found');
  process.exit(1);
}
try {
  const data = yaml.load(match[1]);
  console.log('✅ Valid YAML:', JSON.stringify(data, null, 2));
} catch (e) {
  console.error('❌ YAML parsing error:', e.message);
  process.exit(1);
}
"

Testing Activation

Steps:

1. Create skill
2. Restart Claude Code
3. Ask a question with trigger keywords
4. Verify skill activates (should see skill content in context)

Test questions based on examples:

• Python Data Science: "How do I use pandas to read a CSV file?"
• Documentation Helper: "Where is the API documentation?"
• Kubernetes Expert: "How do I list all pods in a namespace?"

Troubleshooting

Skill Doesn't Activate

Possible causes:

1. Claude Code not restarted: Skills only load on startup

   • Fix: Restart Claude Code

2. Missing trigger keywords: Description doesn't match user's question

   • Fix: Add more keyword variations to description

3. YAML syntax errors: Frontmatter is invalid

   • Fix: Validate YAML format (use checker above)

4. Wrong file location: SKILL.md not in subdirectory

   • Fix: Move to skills/skill-name/SKILL.md

5. Name mismatch: YAML name doesn't match directory name

   • Fix: Ensure consistency (can be different, but confusing)

YAML Parsing Errors

Common mistakes:

# ❌ WRONG: Missing closing ---
---
name: my-skill
description: My skill description

# ✅ CORRECT: Closing --- present
---
name: my-skill
description: My skill description
---

# ❌ WRONG: Invalid characters in name
---
name: My Skill Name
---

# ✅ CORRECT: Lowercase with hyphens
---
name: my-skill-name
---

# ❌ WRONG: Description too long (>1024 chars)
---
name: skill
description: [2000 character description]
---

# ✅ CORRECT: Concise description (<1024 chars)
---
name: skill
description: Focused description with key triggers
---

Best Practices

Naming Conventions

1. Be specific: react-hooks not react
2. Use domain language: kubernetes-expert not k8s-helper
3. Avoid redundancy: python-data-science not python-data-science-skill
4. Max 3 words: Keep names concise

Description Writing

1. Front-load purpose: Start with what the skill does
2. List all variations: Include abbreviations (k8s, kubernetes)
3. Use action words: "Explains", "Helps with", "Provides guidance"
4. Include examples: "React hooks (useState, useEffect, useContext)"

Content Structure

1. Start simple: Basic examples before advanced patterns
2. Use headings: Clear section organization
3. Include code: Concrete examples over abstract explanations
4. Anti-patterns: Show what NOT to do (⚠️ warnings)

Tool Restrictions

1. Read-only by default: Start restrictive, expand if needed
2. Justify access: Only grant Write/Edit if skill creates files
3. Document reasons: Explain why tools are needed

Summary

Key steps:

1. ✅ Validate skill name (lowercase, hyphens, max 64 chars)
2. ✅ Create description with triggers (max 1024 chars)
3. ✅ Choose skill location (personal vs project)
4. ✅ Add tool restrictions if read-only
5. ✅ Generate SKILL.md with YAML frontmatter
6. ✅ Validate structure (frontmatter, fields, format)
7. ✅ Restart Claude Code
8. ✅ Test with trigger question

Remember: Restart Claude Code after creating or modifying skills!