Skill Generator

You are creating a new skill, command, or agent for the Claude Code agentic system.

Your Task

Guide the user through creating a new enhancement for the agent system. Ask clarifying questions, then generate the appropriate file(s) based on the type selected.

Step 1: Determine Type

If not specified in the command arguments, ask the user which type to create:

1. Skill - Reusable knowledge and patterns (TypeScript patterns, React best practices, etc.)
2. Command - Workflow automation (like /spec, /review, /integrate)
3. Agent - Specialized subagent with specific domain expertise

Ask: "What type would you like to create? (skill/command/agent)"

Step 2: Gather Information

For Skills

Ask the user:

• Name: What should the skill be called? (kebab-case, e.g., "react-hooks")
• Description: Brief description of what knowledge this skill provides
• Content: What patterns, best practices, or knowledge should be included?

Skill Structure:

.claude/skills/[name]/
  SKILL.md

SKILL.md Template:

---
name: "[name]"
description: "[description]. Use when [trigger conditions]."
---

# [Title]

[Comprehensive knowledge content, patterns, code examples, best practices]

## [Section 1]

[Content...]

## [Section 2]

[Content...]

## Best Practices

[Key guidelines and patterns...]

## Common Pitfalls

[What to avoid...]

## Quick Reference

[Cheat sheet or quick lookup...]

For Commands

Ask the user:

• Name: What should the command be called? (kebab-case, e.g., "create-component")
• Argument Hint: What arguments does it take? (e.g., "[component-name]")
• Description: Brief description of what the command does
• Process: What steps should the command execute?

Command Structure:

.claude/commands/[name].md

Command Template:

---
description: [Brief description of what this command does]
argument-hint: "[argument-pattern]"
---

# [Command Title]

You are [role or task description].

## Your Task

[Clear description of what this command should accomplish]

## Process

[Step-by-step instructions for executing the command]

### Step 1: [Name]

[Instructions...]

### Step 2: [Name]

[Instructions...]

## Output Format

[What the command should produce or return]

## Example Usage

User: /[command-name] [args] You: [Example response]

## Remember

- [Key principle 1]
- [Key principle 2]
- [Key principle 3]

[Additional detailed instructions as needed]

For Agents

Ask the user:

• Name: What should the agent be called? (kebab-case, e.g., "testing-specialist")
• Description: Brief description of the agent's domain and purpose
• Tools: What tools should the agent have access to? (usually "*" for all tools)
• Model: Which model to use? (usually "claude-sonnet-4-5")
• Domain: What is the agent's area of expertise?
• Responsibilities: What specific tasks should this agent handle?

Agent Structure:

.claude/agents/[name].md

Agent Template:

---
name: [name]
description: [Brief description of domain expertise]
tools: "*"
model: claude-sonnet-4-5
---

# [Agent Title]

You are the **[Role]** for the agentic knowledge base project. Your role is to [high-level purpose].

## Your Domain

- **[Area 1]:** [Description]
- **[Area 2]:** [Description]
- **[Area 3]:** [Description]

## Decision-Making Authority

You have authority to:
- ✅ [Responsibility 1]
- ✅ [Responsibility 2]
- ✅ [Responsibility 3]

You should defer to:
- **[Other Agent]:** [When to defer]
- **User:** [When to ask for approval]

## Working Approach

### [Primary Process Name]

When [doing primary task]:

1. **[Step 1]:** [Instructions]
2. **[Step 2]:** [Instructions]
3. **[Step 3]:** [Instructions]

## Key Principles

- **[Principle 1]:** [Explanation]
- **[Principle 2]:** [Explanation]
- **[Principle 3]:** [Explanation]

## Context Awareness

Always reference:
- **CLAUDE.md:** [What to check]
- **[relevant-skill]:** [What patterns to use]

## Integration with Other Agents

- **Collaborate with [Agent]:** [How and when]
- **Hand off to [Agent]:** [When to hand off]
- **Request [Agent]:** [When to request]

## Example Workflows

### [Workflow 1 Name]

1. [Step 1]
2. [Step 2]
3. [Step 3]

## Your Success Criteria

- ✅ [Success metric 1]
- ✅ [Success metric 2]
- ✅ [Success metric 3]

---

*[Closing motivational statement about the agent's role]*

Step 3: Create the File(s)

Once you have all the information:

1. Create the directory (if needed for skills):

   • For skills: .claude/skills/[name]/

2. Create the file with appropriate template:

   • Skills: .claude/skills/[name]/SKILL.md
   • Commands: .claude/commands/[name].md
   • Agents: .claude/agents/[name].md

3. Fill in the template with the user's information

4. Customize the content based on the specific use case

Step 4: Verify and Confirm

After creating the file(s):

1. Show the user what was created
2. Provide the file path(s)
3. Explain how to use the new enhancement:
   • Skills: Auto-invoked when relevant or manually with Skill tool
   • Commands: Invoke with /[name] [args]
   • Agents: Invoke with Task tool specifying the agent type
4. Ask if any modifications are needed

Tips for Quality

For Skills

• Include comprehensive examples and code snippets
• Organize by topics with clear headings
• Add "Quick Reference" section for common patterns
• Include "Best Practices" and "Common Pitfalls"
• Make it self-contained and reference-able

For Commands

• Be specific about the process steps
• Include examples of expected input/output
• Define clear success criteria
• Reference relevant agents or skills
• Keep scope focused on one workflow

For Agents

• Define clear domain boundaries
• Specify decision-making authority
• Explain when to defer to other agents
• Include example workflows
• Define success criteria

Example Session

User: /create-skill
You: What type would you like to create? (skill/command/agent)

User: skill
You: Great! Let's create a new skill. What should it be called?

User: css-patterns
You: What's a brief description of what this skill provides?

User: Modern CSS patterns including Grid, Flexbox, and custom properties
You: What specific knowledge should be included? Any particular patterns or techniques you want to document?

User: CSS Grid layouts, Flexbox patterns, CSS custom properties, and responsive design patterns
You: Perfect! I'll create a comprehensive skill for CSS patterns. Creating .claude/skills/css-patterns/SKILL.md...

[Creates the file with comprehensive content]

You: ✅ Created CSS patterns skill at .claude/skills/css-patterns/SKILL.md

This skill will be auto-invoked when working with CSS, or you can manually invoke it with the Skill tool.

Would you like me to add or modify anything?

Remember

• Ask clarifying questions - don't guess
• Use the templates as a starting point, then customize
• Make skills comprehensive with examples
• Keep commands focused on workflows
• Define clear agent boundaries
• Always show the user what was created
• Provide usage instructions

Start by determining what type the user wants to create!