Component Creation

Overview

This skill helps create properly structured Claude Code components (skills, agents, commands, hooks) following official specifications.

Instructions

Identify Component Type

Component	User Says	Invocation	Purpose
Skill	"add skill", "create capability"	Model-invoked	Procedural knowledge
Agent	"add agent", "create assistant"	Model/user	Specialized delegation
Command	"add command", "create shortcut"	User-invoked	Explicit shortcuts
Hook	"add hook", "add automation"	Event-driven	Automated actions

Creating a Skill

1. Gather information:

   • Skill name (gerund form: analyzing-data)
   • What it does (for description)
   • When to trigger (for description)
   • Key instructions

2. Create structure:

   mkdir -p skills/{{skill-name}}
   

3. Generate SKILL.md from template:

   ---
   name: {{skill-name}}
   description: {{what-it-does}}. Use when {{trigger-conditions}}.
   ---
   
   # {{Skill Title}}
   
   ## Instructions
   {{step-by-step-guidance}}
   
   ## When to Apply
   {{trigger-conditions-detailed}}
   
   ## Output Format
   {{expected-output-structure}}
   

4. Validate: Check naming rules (max 64 chars, lowercase+hyphens only)

Creating an Agent

1. Gather information:

   • Agent name (lowercase-hyphens)
   • Specialization/purpose
   • Tools needed (or inherit all)
   • Model preference (sonnet/opus/haiku/inherit)

2. Generate agent file in agents/:

   ---
   name: {{agent-name}}
   description: {{specialization}}. Use when {{delegation-triggers}}.
   tools: {{tool-list or omit}}
   model: {{model-alias or omit}}
   ---
   
   # {{Agent Title}}
   
   You are a specialized agent for {{purpose}}.
   
   ## Approach
   {{methodology}}
   
   ## Constraints
   {{boundaries}}
   
   ## Output Format
   {{expected-output}}
   

Creating a Command

1. Gather information:

   • Command name (what user types after /)
   • What it does
   • Arguments needed
   • Tools to restrict (optional)

2. Generate command file in commands/:

   ---
   description: {{brief-description-for-help}}
   allowed-tools: {{tool-list or omit}}
   ---
   
   # {{Command}} Instructions
   
   {{what-claude-should-do}}
   
   ## Arguments
   User provided: $ARGUMENTS
   First argument: $1
   Second argument: $2
   

Creating a Hook

1. Gather information:

   • Event to hook (PreToolUse, PostToolUse, etc.)
   • Matcher pattern (which tools/actions)
   • Action type (command, validation, notification)
   • Script or command to run

2. Update hooks/hooks.json:

   {
     "hooks": {
       "{{Event}}": [
         {
           "matcher": "{{pattern}}",
           "hooks": [
             {
               "type": "command",
               "command": "{{script-path}}"
             }
           ]
         }
       ]
     }
   }
   

Validation Checks

After creating any component:

Component	Validation
Skill	Name ≤64 chars, lowercase+hyphens, description ≤1024 chars
Agent	Name unique, tools valid if specified
Command	File in commands/, description present
Hook	Valid event name, script exists and executable

Templates Reference

• templates/skill.template/SKILL.md.template
• templates/agent.template/agent.md.template
• templates/command.template/command.md.template

See Also

• knowledge/components/agent-skills.md — Skill deep dive
• knowledge/components/subagents.md — Agent deep dive
• knowledge/components/custom-commands.md — Command deep dive
• knowledge/components/hooks.md — Hooks deep dive