Create Skill Command

This command interactively creates standardized skill documentation through conversational interaction with the user.

Purpose

Create high-quality Claude Code skills through guided conversation. Supports:

• Conversation Analysis: Auto-generate skill metadata from session content
• Interactive Collection: Step-by-step information gathering
• Progressive Disclosure: Automatic file structure optimization
• Multiple Scopes: Project-specific, global, or plugin skills

💡 Evaluation: To evaluate existing skills, use /evaluate-skill.

---

Core Workflow Overview

The skill creation process follows 4 phases (with optional Phase 0):

Phase 0: Conversation Analysis (conditional)
    ↓
Phase 1: Type & Basic Info Collection (5 questions)
    ↓
Phase 2: Content Collection (6 questions)
    ↓
Phase 3: File Generation (automated)
    ↓
Phase 4: Validation & Completion

---

Extended Thinking

Role: Skill creation expert providing guided, interactive skill development.

📋 Phase Checklist:

• Phase 0: Conversation analysis (conditional)
• Phase 1: Type & basic info (type, name, location, description, scenarios)
• Phase 2: Content collection (concepts, examples, practices, pitfalls, formatting, tools)
• Phase 3: File generation (Progressive Disclosure + templates)
• Phase 4: Validation & completion

🔄 Progress Tracking:

✅ Phase X complete → 🔄 Starting Phase Y: [description]
📊 Overall progress: X/4 Phases complete (YY%)

Within phases (Phase 1-2):

📊 Phase progress: X/Y questions complete

Core Actions:

1. Analyze conversation → generate recommendations
2. Collect information → type-specific questions
3. Generate files → Progressive Disclosure + templates
4. Validate → quality check + usage guidance

Execution Steps

📖 Detailed Instructions: For complete phase details, see @creator-suite/references/create-skill/phase-details.md

Phase 0: Conversation Analysis (Conditional)

Skip if:

• Conversation < 5 messages
• No technical content
• Only simple greetings

Execute if sufficient conversation:

1. Extract main topics, keywords, code examples, patterns
2. Generate 2-3 skill name suggestions (kebab-case, gerund form)
3. Auto-generate description, scenarios, concepts, best practices, pitfalls
4. Present analysis results:

   📊 Conversation analysis complete!
   🔍 Discovered: topics, examples, patterns, concepts
   💡 Proceeding with recommendations
   

If insufficient:

ℹ️ Insufficient conversation content
💡 Proceeding in manual input mode

Result: Set mode to "recommendation" or "manual" for Phase 1-2.

📖 See: phase-details.md → Phase 0 for complete conversation analysis procedures.

Phase 1: Type & Basic Information

Collect 5 pieces of information via AskUserQuestion:

1. Skill Type (use AskUserQuestion):

   • Quick Workflow (200-400 words): Simple tools
   • Comprehensive Guide (600-1,500 words): Standard workflows [Default]
   • Technical Reference (1,500-5,000+ words): Complex toolkits
   • Philosophy-Driven (1,000-3,000 words): Conceptual frameworks

2. Skill Name:

   • Format: kebab-case, ≤64 chars
   • Style: Gerund form (processing-pdfs, managing-containers)
   • Validate: ^[a-z0-9]+(-[a-z0-9]+)*$

3. File Location (use AskUserQuestion):

   • Project (.claude/skills/) [Recommended, Default]
   • Global (~/.claude/skills/)
   • Current Directory (./{skill-name}/)

   Store path: .claude/skills or ~/.claude/skills or .

4. Description:

   • Format: Third person, ≤1024 chars
   • Include: Specific trigger phrases ("create X", "build Y")
   • Example: "This skill should be used when the user asks to "create hook", "add PreToolUse hook"..."

5. Usage Scenarios:

   • Format: Multi-line list
   • Examples: "When designing new API", "During code review"

Question Mode:

• Recommendation mode: Present extracted content → Modify/Use/Custom
• Manual mode: Guidance → Example → Request input

📖 See:

• phase-details.md → Phase 1 for detailed question flows
• skill-types.md for complete type descriptions

Phase 2: Content Collection

Collect 6 pieces of content information:

6. Core Concepts:

   • Format: Free-form text, multiple paragraphs
   • Include: Main concepts, principles, theories
   • Recommendation: Present auto-organized content

7. Patterns/Examples (optional):

   • Ask: Include examples? Yes/No
   • If yes, quality level:
     • Basic: Syntax only
     • Standard: Working code (recommended)
     • Production: Error handling + edge cases
   • Format: Pattern name + language + code block

8. Best Practices:

   • Format: List
   • Examples: "Use clear naming", "Consistent error handling"
   • Recommendation: Present collected practices

9. Common Pitfalls:

   • Format: List
   • Examples: "Avoid over-abstraction", "Consider performance"
   • Recommendation: Present identified pitfalls

10. Formatting Style (optional):

    • Minimal (default): Clean, simple
    • Enhanced: Emoji markers (🚀, 📋, ⚠️), Visual indicators (✅/❌)
    • Default: Minimal

11. Allowed Tools (optional):

    • Default: No restriction (all tools available)
    • Restrict when: Specific workflows needed, prevent side effects
    • Tools: Read, Write, Edit, Grep, Glob, Bash, WebSearch, WebFetch, Task, SlashCommand
    • Recommendation mode: Analyze tool usage → Suggest restriction
    • Manual mode: Ask if restriction needed → Multi-select tools

📖 See: phase-details.md → Phase 2 for detailed content collection procedures

Phase 3: File Generation

Automated file creation based on collected information.

Process:

1. Calculate total word count
2. Determine Progressive Disclosure level (1-4)
3. Create directory structure with mkdir -p
4. Apply type-specific template
5. Generate files with Write tool

Level	Word Count	Files	Structure
1	< 400	SKILL.md only	All content in one file
2	400-1,500	+ examples/	Separate example files
3	1,500-3,500	+ REFERENCE.md	Detailed API docs separated
4	> 3,500	+ EXAMPLES.md + FORMS.md	Full documentation suite

Frontmatter (all skills):

---
name: {skill-name}
description: "{skill-description}"
allowed-tools: {tools-list}  # Only if restricted
---

Required body sections:

• # {Skill Title} (H1)
• ## When to Use This Skill
• ## Core Concepts

Recommended sections (type-specific):

• Quick Start / Overview / Philosophy (entry point)
• Patterns / Examples
• Best Practices
• Common Pitfalls
• Additional Resources (with references for Level 3+)

📖 See:

• progressive-disclosure.md for complete level specifications
• templates.md for all type-specific templates

Phase 4: Validation & Completion

Validation (against @shared/skill/validation-criteria.md):

1. Structure: File existence, Progressive Disclosure level
2. Frontmatter: name, description format, allowed-tools
3. Required sections: When to Use, Core Concepts
4. Content quality: Code blocks, hierarchy, links

Validation Report:

📊 Generation complete - Validation result

## Basic quality check
✅ File structure: {status}
✅ Frontmatter: {status}
✅ Required sections: {status}
⚠️ Improvable items: {if_any}

💡 For detailed evaluation: /evaluate-skill

User Guidance:

✅ Skill created!

📂 {location_path}/{skill-name}/SKILL.md
🎯 {scope_description}

Usage: @{skill-name}/SKILL.md
Evaluate: /evaluate-skill

Scope:

• Project: Available in this project only
• Global: Available in all projects
• Current: Move to .claude/skills/ or ~/.claude/skills/ to activate

Error Handling

Common Errors:

• Invalid skill name: Re-ask with kebab-case format guidance
• Skill exists: Offer rename / overwrite / cancel
• Missing required info: Re-ask for missing fields

---

Implementation Notes

Critical Requirements:

1. Progress tracking: Show phase transitions and overall progress
2. Conversation analysis priority: Phase 0 before Phase 1 when possible
3. Question modes: Recommendation (with extracted content) vs Manual (with examples)
4. Tool usage:
   • AskUserQuestion: All information collection
   • Write: All file creation
   • Bash: Directory creation (mkdir -p)
5. Validation: Always validate after generation
6. User feedback: Clear progress at each step

Phase 0 Behavior:

• Always attempt conversation analysis first
• Present discovered content to user
• Fall back to manual mode gracefully if insufficient

---

Additional Resources

Detailed Documentation:

• @creator-suite/references/create-skill/phase-details.md - Complete phase-by-phase procedures
• @creator-suite/references/create-skill/skill-types.md - Detailed type descriptions and selection guide
• @creator-suite/references/create-skill/templates.md - All template structures and examples
• @creator-suite/references/create-skill/progressive-disclosure.md - Complete Progressive Disclosure specifications

Validation:

• @shared/skill/validation-criteria.md - Quality standards and validation procedures

---

Ready to create skill. Starting with conversation analysis, then proceeding through guided phases.