Créer une compétence Claude Code

Orchestre quatre agents spécialisés pour créer des compétences Claude Code prêtes pour la production via un développement basé sur l'élicitation.

Spar Skills Guide Bot
DeveloppementIntermédiaire
0022/07/2026
Claude Code
#claude-code#skill-development#elicitation#agent-orchestration#skill-generation

Recommandé pour

Build Claude Code Skill

Create comprehensive Claude Code Skills through elicitation-driven development

Instructions

This command orchestrates four specialized agents to build production-ready Claude Code Skills from user requirements. Follow this structured workflow:

Phase 1: Requirements Elicitation

Agent: skill-elicitation-agent

  1. Activate Elicitation Agent

    • Launch the skill-elicitation-agent using the Task tool
    • Provide context: "The user wants to create a new Claude Code Skill"
    • Include any initial requirements or ideas the user has shared
  2. Elicitation Questions

    • The agent will ask 3-5 targeted questions to understand:
      • Purpose and scope of the skill
      • Complexity and structure requirements
      • Tool permissions needed
      • Context and references
      • Success criteria
  3. Specification Creation

    • Agent will create a comprehensive skill specification document
    • Includes: metadata, structure, instructions outline, code needs, examples, dependencies
    • Uses progressive disclosure strategy
    • Validates completeness before proceeding
  4. User Approval

    • Present the specification to the user
    • Confirm understanding and agreement
    • Make adjustments if needed
    • Get explicit approval to proceed to generation

Phase 2: Skill Generation

Agent: skill-generator-agent

  1. Activate Generator Agent

    • Launch the skill-generator-agent using the Task tool
    • Pass the approved specification document
    • Specify target location (personal, project, or plugin skill)
  2. Directory Structure Creation

    • Agent creates appropriate directory structure:
      skill-name/
      ├── SKILL.md (required)
      ├── reference.md (if needed)
      ├── examples.md (if needed)
      ├── scripts/ (if needed)
      └── templates/ (if needed)
      
  3. SKILL.md Generation

    • Creates main skill file with proper frontmatter
    • Includes clear instructions and examples
    • Follows progressive disclosure principles
    • Links to supporting files
  4. Supporting Files Generation

    • Creates reference documentation
    • Generates example files
    • Writes scripts with proper error handling
    • Creates reusable templates
  5. Dependency Documentation

    • Documents all required packages
    • Provides installation instructions
    • Notes version requirements
  6. Quality Checks

    • Validates YAML frontmatter
    • Checks file structure
    • Verifies code syntax
    • Tests progressive disclosure

Phase 3: Validation and Testing

Agent: skill-validator-agent

  1. Activate Validator Agent

    • Launch the skill-validator-agent using the Task tool
    • Provide path to generated skill
    • Request comprehensive validation
  2. YAML Validation

    • Check frontmatter syntax
    • Verify required fields
    • Validate optional fields
    • Test YAML parsing
  3. Description Analysis

    • Assess discoverability
    • Check trigger keywords
    • Verify clarity and completeness
    • Compare to best practices
  4. Structure Validation

    • Verify file organization
    • Check all references
    • Test script execution
    • Validate permissions
  5. Code Testing

    • Syntax validation
    • Security checks
    • Dependency verification
    • Error handling tests
  6. Integration Testing

    • Test skill loading
    • Verify triggering
    • Check execution flow
    • Validate outputs
  7. Validation Report

    • Generate comprehensive report
    • Score each category
    • List issues by severity
    • Provide actionable fixes
  8. Issue Resolution

    • If issues found, work with generator agent to fix
    • Re-validate after fixes
    • Iterate until validation passes

Phase 4: Documentation Enhancement

Agent: skill-documenter-agent

  1. Activate Documenter Agent

    • Launch the skill-documenter-agent using the Task tool
    • Provide skill path and specification
    • Request comprehensive documentation
  2. SKILL.md Enhancement

    • Refine instructions for clarity
    • Add comprehensive examples
    • Include best practices
    • Create troubleshooting section
  3. Reference Documentation

    • Create detailed technical reference (if needed)
    • Document API and configuration
    • Provide advanced patterns
    • Include performance tuning tips
  4. Example Collection

    • Generate beginner to advanced examples
    • Include troubleshooting examples
    • Show integration patterns
    • Provide case studies
  5. README Creation

    • Create skill directory README (if distributing)
    • Document installation
    • Provide quick start
    • Link to full documentation
  6. Documentation Quality Check

    • Verify clarity and completeness
    • Test all code examples
    • Check organization
    • Validate accuracy

Phase 5: Final Delivery

  1. Generate Summary

    • List all files created
    • Document location (personal/project/plugin)
    • Provide usage instructions
    • Include test scenarios
  2. Installation Verification

    • Confirm skill is in correct location
    • Verify file permissions
    • Check dependencies documented
    • Test skill loading
  3. Usage Guide

    • Explain how to trigger the skill
    • Provide example prompts
    • Show expected behavior
    • Link to documentation
  4. Next Steps

    • Suggest testing approach
    • Recommend improvements
    • Explain maintenance
    • Note future enhancements

Agent Coordination

Sequential Flow

User Request
    ↓
skill-elicitation-agent (Requirements)
    ↓
User Approval
    ↓
skill-generator-agent (Creation)
    ↓
skill-validator-agent (Testing)
    ↓
Fixes if needed (loop back to generator)
    ↓
skill-documenter-agent (Enhancement)
    ↓
Final Delivery

Agent Communication

Between Agents:

  • Elicitation → Generator: Pass specification document
  • Generator → Validator: Pass skill location and files
  • Validator → Generator: Pass validation issues (if any)
  • Generator → Documenter: Pass skill for enhancement
  • Documenter → User: Final documentation

With User:

  • Get approval after elicitation
  • Confirm location preference (personal/project)
  • Review validation results
  • Approve final deliverable

Best Practices

  1. Always Start with Elicitation

    • Don't skip the question phase
    • Understand deeply before building
    • Get user approval on specification
  2. Use All Four Agents

    • Each has specialized expertise
    • Complete flow ensures quality
    • Don't shortcut the process
  3. Iterate Based on Validation

    • Fix issues immediately
    • Re-validate after changes
    • Don't proceed with failures
  4. Test Comprehensively

    • Manual trigger testing
    • Script execution testing
    • Integration testing
    • Real-world scenario testing
  5. Document Thoroughly

    • Clear instructions
    • Comprehensive examples
    • Troubleshooting guides
    • Best practices

Location Options

Personal Skills (~/.claude/skills/)

Use for:

  • Individual workflows
  • Experimental skills
  • Personal preferences
  • Private tools

Project Skills (.claude/skills/)

Use for:

  • Team-shared workflows
  • Project-specific expertise
  • Version-controlled skills
  • Collaborative tools

Plugin Skills (Plugin directory structure)

Use for:

  • Distributable skills
  • Marketplace deployment
  • Public sharing
  • Bundled capabilities

Output Format

Provide the user with a comprehensive summary:

🎉 Skill Created Successfully!

📋 Skill: [Skill Name]
📁 Location: [path]
📝 Type: [simple/multi-file/tool-restricted/code-execution]

📄 Files Created:
✅ SKILL.md - Main skill instructions
✅ reference.md - Technical reference (if created)
✅ examples.md - Comprehensive examples (if created)
✅ scripts/[name].py - Helper scripts (if created)
✅ README.md - Installation guide (if created)

🧪 Validation: ✅ PASSED (Score: X/10)

📦 Dependencies:
[List if any, or "None"]

🎯 Usage:
Trigger this skill by saying:
- "[Example trigger 1]"
- "[Example trigger 2]"

Or explicitly: "Use [skill-name] to [task]"

✨ Test with:
[Specific test scenario]

📚 Documentation:
- See SKILL.md for instructions
- See examples.md for comprehensive examples
- See reference.md for technical details

🚀 Next Steps:
1. Test the skill with provided scenarios
2. Refine based on usage
3. Share with team (if project skill)
4. Consider adding more examples over time

Troubleshooting

Agent Not Found: Ensure .claude/agents/skill-builder/ exists with all agent files

Permission Errors: Check file permissions with chmod +x scripts/*.py

YAML Errors: Validator will catch and report these

Agent Confusion: Use explicit agent names: "Launch skill-elicitation-agent"

Examples

Example 1: Simple Instruction-Only Skill

Request: "Create a skill for writing conventional commit messages"

Flow:

  1. Elicitation asks about commit style, projects, examples
  2. Generates simple single-file SKILL.md
  3. Validates structure and description
  4. Documents with examples and best practices

Example 2: Multi-File Skill with Scripts

Request: "Create a skill for PDF form filling"

Flow:

  1. Elicitation asks about PDF types, operations, dependencies
  2. Generates SKILL.md + scripts/fill_form.py + FORMS.md
  3. Validates code execution and file structure
  4. Documents with comprehensive examples and API reference

Example 3: Tool-Restricted Read-Only Skill

Request: "Create a skill for security code analysis"

Flow:

  1. Elicitation determines read-only requirement
  2. Generates SKILL.md with allowed-tools: Read, Grep, Glob
  3. Validates tool restrictions
  4. Documents security patterns and analysis techniques

Remember

  • Quality over speed - Take time for thorough elicitation
  • User involvement - Get approval at key stages
  • Comprehensive validation - Test everything
  • Excellent documentation - Make skills easy to use
  • Iterative improvement - Skills can evolve over time

This command ensures every skill is production-ready, well-documented, and follows Claude Code best practices!

Skills similaires