skillMaker

Skill Maker - Claude Code Skill Generator

Create production-ready Claude Code skills following established patterns and best practices.

Overview

This skill guides you through creating well-structured, effective Claude Code skills by:

1. Understanding the skill's purpose and trigger conditions
2. Gathering requirements through targeted questions
3. Generating appropriate file structure and content
4. Following optimization patterns and conventions

When This Skill Activates

Use this skill when the user wants to:

• Create a new Claude Code skill from scratch
• Understand skill structure and patterns
• Generate skill templates with best practices
• Convert manual workflows into reusable skills

Skill Creation Workflow

Phase 1: Requirements Gathering

Ask the user these key questions (use AskUserQuestion for efficiency):

1. Purpose & Trigger

   • What should this skill do?
   • When should Claude invoke it automatically?
   • What phrases or keywords would users say to trigger it?

2. Invocation Control

   • Who can invoke it?
     • Both user and Claude (default)
     • User-only (for side-effect tasks: deploy, commit, delete)
     • Claude-only (for background knowledge/reference)

3. Argument Pattern

   • Does it need arguments? (e.g., /migrate-component SearchBar React Vue)
   • If yes, what are they and how should they be used?

4. Tool Requirements

   • What tools will it need? (Read, Write, Edit, Bash, Grep, Glob, Agent, etc.)
   • Should tool access be restricted for safety?

5. Complexity Level

   • Simple reference skill (conventions, patterns)
   • Task workflow skill (multi-step process)
   • Dynamic context skill (uses shell commands)
   • Bundled skill (includes templates, scripts, examples)

6. Execution Context

   • Inline in main conversation (default)
   • Isolated subagent (context: fork)
   • Specific agent type (agent: Explore)

Phase 2: Structure Decision

Based on requirements, determine file structure:

Simple skill (reference/conventions):

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

Bundled skill (with supporting files):

.claude/skills/skill-name/
├── SKILL.md              # Main instructions
├── templates/            # Templates for generation
│   └── template.md
├── examples/             # Example outputs
│   └── example.md
├── references/           # Detailed reference docs
│   └── patterns.md
└── scripts/              # Helper scripts
    └── helper.sh

Phase 3: Content Generation

Generate skill content following these patterns:

A. Frontmatter Configuration

---
name: skill-name                    # Required: kebab-case identifier
description: |                      # Required: Clear trigger conditions
  [Action] when user asks to "[phrase]",
  "[another phrase]", mentions "[keyword]",
  or discusses [topic].
version: 1.0.0                      # Optional: Semantic version
disable-model-invocation: true      # Optional: User-only invocation
user-invocable: false               # Optional: Claude-only invocation
allowed-tools: Read, Grep, Glob     # Optional: Tool restrictions
model: claude-3-5-haiku             # Optional: Model override
context: fork                       # Optional: Isolated execution
agent: Explore                      # Optional: Specific agent type
---

Description Writing Formula:

• Start with verb: "Create", "Analyze", "Generate", "Review"
• Include specific trigger phrases users would say
• List relevant keywords
• Mention the domain/topic area

Examples:

# Good - Specific and actionable
description: Generate API documentation from code. Use when user asks to "document the API", "create API docs", mentions "OpenAPI" or "Swagger", or discusses API documentation needs.

# Bad - Too vague
description: Helps with documentation tasks.

B. Main Content Structure

Follow this template for SKILL.md body:

# [Skill Name] - [One-line Purpose]

[2-3 sentence overview of what this skill does and its value]

## Overview

[Detailed explanation of the skill's capabilities and when to use it]

## When This Skill Activates

[List specific scenarios that trigger this skill]

## Workflow / Process

[Step-by-step instructions if it's a task skill]

### Phase 1: [First Phase Name]
[Instructions for first phase]

### Phase 2: [Second Phase Name]
[Instructions for second phase]

## Guidelines / Patterns

[Best practices, conventions, or patterns to follow]

## Examples

[Include 2-3 concrete examples if relevant]

## Supporting Files

[Reference any templates, examples, or scripts]
- For [use case], see [file-name.md](file-name.md)
- Script for [action]: `${CLAUDE_SKILL_DIR}/scripts/helper.sh`

## Common Patterns

[Include reusable patterns or code snippets]

## Optimization Tips

[Performance or quality improvements]

C. Supporting Files

When to create supporting files:

1. templates/ - Create when:

   • Skill generates structured output (API docs, test files, configs)
   • Users need consistent format across invocations
   • Content follows a repeatable pattern

2. examples/ - Create when:

   • Complex output that benefits from reference
   • Multiple valid approaches to show
   • Helping users understand expected results

3. references/ - Create when:

   • Detailed reference material (>500 lines)
   • Technical specifications or API docs
   • Comprehensive pattern libraries
   • Keep main SKILL.md focused, reference these for details

4. scripts/ - Create when:

   • Preprocessing data before Claude sees it
   • Complex calculations or transformations
   • External tool integration
   • File system operations

Reference linking pattern:

## Additional Resources

For complete details, see:
- API patterns: [references/api-patterns.md](references/api-patterns.md)
- Code examples: [examples/sample-output.md](examples/sample-output.md)
- Generation template: [templates/component-template.md](templates/component-template.md)
- Preprocessing script: `${CLAUDE_SKILL_DIR}/scripts/analyze.sh`

D. Dynamic Context Injection

Use shell command substitution for live data:

## Current Project Context

- Repository: !`git remote get-url origin 2>/dev/null || echo "No git remote"`
- Current branch: !`git branch --show-current 2>/dev/null || echo "Not a git repo"`
- Recent commits: !`git log --oneline -3 2>/dev/null || echo "No commits"`
- Staged files: !`git diff --cached --name-only 2>/dev/null || echo "No staged files"`

## Dependencies

- Node version: !`node --version 2>/dev/null || echo "Node not installed"`
- Package manager: !`[ -f "package-lock.json" ] && echo "npm" || [ -f "yarn.lock" ] && echo "yarn" || [ -f "pnpm-lock.yaml" ] && echo "pnpm" || echo "Unknown"`

Command patterns for common needs:

Phase 4: Optimization

Apply these optimizations:

A. Context Efficiency

• Keep SKILL.md under 500 lines
• Move detailed references to separate files
• Use dynamic injection for live data
• Reference supporting files explicitly

B. Tool Restrictions

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

# Safe modification skill
allowed-tools: Read, Write, Edit

# Full automation
allowed-tools: Read, Write, Edit, Bash(npm *), Bash(git *)

C. Invocation Control

# Default: Both can invoke (most common)
# Omit both flags

# User-only: Prevents accidental auto-invocation
disable-model-invocation: true

# Claude-only: Background knowledge/reference
user-invocable: false

D. Execution Context

# Default: Inline execution (most efficient)
# Omit context flag

# Isolated subagent: For research/exploration
context: fork
agent: Explore

# Isolated subagent: For code work
context: fork

Phase 5: Testing & Validation

After generating the skill, validate:

1. Frontmatter syntax - Valid YAML
2. Description clarity - Specific trigger conditions
3. File references - All referenced files exist
4. Shell commands - Valid syntax in dynamic injection blocks
5. Tool alignment - allowed-tools matches actual needs
6. Invocation logic - Flags match intended behavior

Test invocation:

# Test manual invocation
/skill-name arg1 arg2

# Test in conversation
# Use trigger phrases from description

Skill Templates by Category

Template 1: Reference/Conventions Skill

---
name: coding-standards
description: Apply coding standards and conventions. Use when writing code, reviewing code, or when user asks about "code style", "conventions", or "standards".
user-invocable: false  # Claude-only background knowledge
---

# Coding Standards

Apply these conventions when writing code:

## Naming Conventions
- Variables: camelCase
- Constants: UPPER_SNAKE_CASE
- Classes: PascalCase

## File Structure
[patterns...]

## Best Practices
[guidelines...]

Template 2: Task Workflow Skill

---
name: create-feature
description: Create a new feature following project standards. Use when user asks to "add a feature", "create new feature", "implement feature", or discusses adding functionality.
disable-model-invocation: true  # User-only
allowed-tools: Read, Write, Edit, Bash
---

# Feature Creator

Create a complete feature following project patterns.

## Workflow

### Phase 1: Planning
1. Read existing similar features
2. Identify files to modify/create
3. Confirm approach with user

### Phase 2: Implementation
1. Create feature files
2. Add tests
3. Update documentation

### Phase 3: Validation
1. Run tests
2. Check formatting
3. Create commit

## Arguments

$0 - Feature name (kebab-case)
$1 - Feature type (component|page|api)

Template 3: Dynamic Context Skill

---
name: pr-summarizer
description: Summarize pull request changes. Use when user asks to "summarize PR", "explain changes", mentions "pull request summary", or discusses PR content.
allowed-tools: Bash(gh *)
---

# Pull Request Summarizer

## Current PR Context

- Title: !`gh pr view --json title -q .title 2>/dev/null || echo "Not in a PR context"`
- Author: !`gh pr view --json author -q .author.login 2>/dev/null || echo "N/A"`
- Status: !`gh pr view --json state -q .state 2>/dev/null || echo "Unknown"`

## Changes

!`gh pr diff 2>/dev/null || echo "No PR diff available"`

## Summary Instructions

Analyze the PR and provide:
1. High-level summary (2-3 sentences)
2. Key changes by file
3. Potential impacts
4. Testing recommendations

Template 4: Bundled Skill with Templates

---
name: api-documenter
description: Generate API documentation from code. Use when user asks to "document API", "create API docs", mentions "OpenAPI" or "Swagger", or discusses API documentation.
allowed-tools: Read, Write, Bash
---

# API Documenter

Generate comprehensive API documentation.

## Templates

Use these templates for structured output:
- OpenAPI spec: [templates/openapi.yaml](templates/openapi.yaml)
- Endpoint docs: [templates/endpoint.md](templates/endpoint.md)

## Process

1. Scan for API routes
2. Extract endpoint details
3. Generate documentation using templates
4. Validate with: `${CLAUDE_SKILL_DIR}/scripts/validate-openapi.sh`

## Examples

See [examples/sample-api-docs.md](examples/sample-api-docs.md) for expected format.

Generation Checklist

Before finalizing, verify:

• name matches directory name (kebab-case)
• description includes specific trigger phrases
• Invocation control flags align with intended use
• allowed-tools covers actual tool needs
• Main SKILL.md is under 500 lines
• Supporting files are referenced in main content
• Shell commands use proper dynamic injection syntax
• Examples demonstrate key patterns
• File structure matches complexity needs

Installation & Usage

After creating the skill:

1. Save to correct location:

   • Project: .claude/skills/skill-name/SKILL.md
   • Personal: ~/.claude/skills/skill-name/SKILL.md

2. Test invocation:

   # Manual test
   /skill-name [arguments]
   
   # Automatic test
   # Use trigger phrases in conversation
   

3. Verify activation:

   • Check skill appears in / autocomplete
   • Verify description triggers work
   • Test with and without arguments

Best Practices Recap

DO:

• Write specific, trigger-rich descriptions
• Keep SKILL.md focused and under 500 lines
• Use supporting files for complex content
• Restrict tools when possible for safety
• Use dynamic context for live data
• Include concrete examples
• Test trigger phrases

DON'T:

• Write vague descriptions ("helps with tasks")
• Put everything in one huge SKILL.md
• Forget to reference supporting files
• Over-restrict tools unnecessarily
• Hardcode data that can be dynamic
• Create skills with overlapping triggers
• Skip testing invocation patterns

Common Patterns Library

Pattern: Multi-Argument Skill

Fix $0 by $1 in $2 files.
# $0 = bug description
# $1 = approach
# $2 = file pattern

Pattern: Optional Arguments

Generate tests for ${1:-all components}.
# $1 = optional component name, defaults to "all components"

Pattern: Conditional Logic

!`[ -f "package.json" ] && cat package.json || echo "{}"`
# Returns package.json if exists, else empty object

Pattern: Error Handling

!`git status 2>/dev/null || echo "Not a git repository"`
# Suppresses errors and provides fallback

Skill Quality Metrics

Evaluate skill quality:

Troubleshooting

Skill doesn't auto-trigger:

• Check description has specific trigger phrases
• Verify no disable-model-invocation: true flag
• Test with exact phrases from description

Skill not in autocomplete:

• Verify SKILL.md in correct location
• Check frontmatter YAML is valid
• Ensure no user-invocable: false flag

Shell commands fail:

• Verify dynamic injection syntax (backticks with exclamation prefix)
• Add error handling with || echo "fallback"
• Check commands available in user's environment

Related Tools

• Hooks: For automatic actions on tool events
• Subagents: For parallel execution and isolation
• Plugins: For bundling multiple related skills
• MCP Servers: For external tool integration

Supporting Files

For detailed skill patterns and examples:

• Skill structure examples: examples/skill-structures.md
• Template library: templates/
• Reference patterns: references/skill-patterns.md