Skill Creator

Guide users through creating high-quality custom skills for claude-code-setup.

Overview

Skills extend Claude's capabilities with specialized knowledge and workflows. This skill helps users create their own custom skills stored in ~/.claude/custom/skills/.

Skill types:

• Command skills: Invoked explicitly with /skill-name
• Context skills: Auto-loaded based on project's Tech Stack

Creation Flow

Follow these steps in order. Use conversational interaction - ask one question at a time.

Step 1: Determine Skill Type

Ask the user:

What type of skill do you want to create?

1. **Command skill** - Invoked explicitly with /skill-name
   Example: /deploy-staging, /generate-report

2. **Context skill** - Auto-loads based on project's tech stack
   Example: Company coding standards, API conventions

Step 2: Gather Concrete Examples

This is the most important step for quality. Ask based on skill type:

For command skills:

Give me 2-3 concrete examples of how you'd use this skill.
Format: "I say X and expect Y"

Example:
- "Deploy to staging" → push code, SSH to server, run deploy script
- "Generate weekly report" → gather metrics, format as markdown, save to docs/

For context skills:

Give me 2-3 concrete examples of how this skill should guide Claude.
Format: "When doing X, Claude should Y"

Example:
- When writing API endpoints, use our standard response format
- When handling errors, use our custom exception classes
- When writing tests, follow our naming convention

Step 3: Analyze and Suggest

Based on the examples, identify:

1. Name suggestion (kebab-case)
2. Patterns/workflows in the examples
3. For context skills: Suggest applies_to values based on mentioned technologies

Present the analysis:

Based on your examples, I see a [pattern description].

Suggested name: [name]
[For context skills only:]
Suggested applies_to: [tech1, tech2, ...]

Want to add more tech stacks? (comma-separated, or continue)

Step 4: Generate or Review Content

Ask the user:

Should I generate the skill content based on your examples,
or do you want to provide the instructions yourself?

1. Generate for me
2. I'll provide it

If generating:

• Create clear, actionable instructions based on the examples
• Structure with headers for different workflows/scenarios
• Include relevant details from the examples

If user provides:

• Ask them to paste or describe the content
• Review their content and suggest improvements:
  • Missing details that would help Claude
  • Unclear instructions
  • Missing examples
• Ask: "Should I incorporate these suggestions?"

Step 5: Review and Refine

Present the complete SKILL.md draft:

---
name: [name]
description: [1-2 sentence description]
type: [command|context]
[applies_to: [tech1, tech2, ...]]  # Only for context skills
[file_extensions: [".ext1", ".ext2"]]  # Only for context skills with file mappings
---

# [Title]

[Content...]

Ask: "Does this look good? Any changes needed?"

Iterate until the user approves.

Step 6: Save the Skill

1. Create directory: ~/.claude/custom/skills/[name]/
2. Write the SKILL.md file
3. Confirm creation with usage instructions:

For command skills:

Created: ~/.claude/custom/skills/[name]/SKILL.md

To use: /[name]
To edit: Modify the SKILL.md file directly

For context skills:

Created: ~/.claude/custom/skills/[name]/SKILL.md

This skill will auto-load when your project's Tech Stack includes:
[tech1], [tech2], or [tech3]

To edit: Modify the SKILL.md file directly

Skill Format Reference

Command Skill

---
name: skill-name
description: Brief description of what this skill does
type: command
---

# Skill Title

Instructions for Claude to follow when this skill is invoked.

Context Skill

---
name: skill-name
description: Brief description of what this skill provides
type: context
applies_to: [python, fastapi, django]
file_extensions: [".py"]
---

# Skill Title

Standards or guidelines that Claude should follow automatically.

Frontmatter Fields

Common applies_to Values

Languages: python, typescript, javascript, rust, go, java Frameworks: fastapi, django, flask, react, nextjs, express Tools: docker, kubernetes, terraform, aws, gcp

Quality Guidelines

When generating or reviewing skill content:

1. Be specific - Vague instructions lead to inconsistent results
2. Include examples - Show what good output looks like
3. Structure clearly - Use headers, lists, tables
4. Stay focused - One skill per domain/workflow
5. Test mentally - Would Claude know what to do with these instructions?

Output Location

All custom skills are saved to:

~/.claude/custom/skills/[skill-name]/
└── SKILL.md

This location is separate from the base installation (~/.claude/skills/) and survives upgrades.