create
Create Claude Code extensions (prompts, commands, agents, skills, hooks)
mkdir -p ~/.claude/commands && curl -fsSL https://raw.githubusercontent.com/marcusgoll/Spec-Flow/HEAD/.claude/commands/meta/create.md -o ~/.claude/commands/create.mdcreate.md
# /create — Consolidated Creation Command
<context>
**Arguments**: $ARGUMENTS
</context>
<objective>
Unified entry point for creating Claude Code extensions with expert guidance:
- `/create prompt <desc>` → Generate optimized Claude prompts
- `/create command <desc>` → Create new slash command with best practices
- `/create agent <desc>` → Create specialized subagent configuration
- `/create skill <desc>` → Create agent skill with progressive disclosure
- `/create hook` → Create event-driven hook
All creation commands provide expert guidance on structure, best practices, and validation.
</objective>
<process>
## Step 1: Parse Type Argument
**Extract first argument as type:**
```
$type = first word of $ARGUMENTS
$remaining = rest of $ARGUMENTS
```
**If no type provided:**
Use AskUserQuestion to ask:
```json
{
"question": "What would you like to create?",
"header": "Extension Type",
"multiSelect": false,
"options": [
{
"label": "prompt",
"description": "Optimized Claude prompt with XML structure"
},
{
"label": "command",
"description": "New slash command following best practices"
},
{
"label": "agent",
"description": "Specialized subagent configuration"
},
{
"label": "skill",
"description": "Agent skill file with progressive disclosure"
},
{
"label": "hook",
"description": "Event-driven hook for tool/session events"
}
]
}
```
## Step 2: Invoke Appropriate Skill
<when_argument_is value="prompt">
### Create Prompt
**Purpose**: Generate optimized, XML-structured prompts with intelligent depth selection.
**Invoke skill**:
```
Skill: create-meta-prompts
```
Pass remaining arguments to skill for prompt generation.
**What the skill provides**:
- XML-structured prompt templates
- Intelligent depth selection (quick/standard/comprehensive)
- Context engineering best practices
- Progressive disclosure patterns
- Example outputs and validation
</when_argument_is>
<when_argument_is value="command">
### Create Slash Command
**Purpose**: Create new slash command following Claude Code best practices.
**Invoke skill**:
```
Skill: create-slash-commands
```
Pass remaining arguments for slash command creation.
**What the skill provides**:
- YAML frontmatter structure
- Argument parsing patterns
- Tool restrictions and allowed-tools configuration
- Dynamic context integration
- Progressive disclosure via skills
- Validation and testing guidance
</when_argument_is>
<when_argument_is value="agent">
### Create Subagent
**Purpose**: Create specialized subagent configuration with role definition and tool selection.
**Invoke skill**:
```
Skill: create-subagents
```
Pass remaining arguments for subagent creation.
**What the skill provides**:
- Subagent configuration structure
- Role and objective definition
- Tool selection guidance
- Prompt engineering for agent prompts
- XML structure compliance
- Integration with Task tool
</when_argument_is>
<when_argument_is value="skill">
### Create Agent Skill
**Purpose**: Create agent skill file with progressive disclosure and best practices.
**Invoke skill**:
```
Skill: create-agent-skills
```
Pass remaining arguments for skill creation.
**What the skill provides**:
- SKILL.md structure and organization
- Progressive disclosure patterns
- Reference documentation separation
- Example patterns and anti-patterns
- Validation rules
- Testing and quality guidance
</when_argument_is>
<when_argument_is value="hook">
### Create Hook
**Purpose**: Create event-driven hook for tool use, session events, or user prompts.
**Invoke skill**:
```
Skill: create-hooks
```
Pass remaining arguments for hook creation.
**What the skill provides**:
- Hook types (PreToolUse, PostToolUse, Stop, SessionStart, etc.)
- Configuration structure
- Matcher patterns
- Command vs inline hooks
- Error handling
- Testing and debugging guidance
</when_argument_is>
</process>
<success_criteria>
- Type correctly identified from arguments
- Appropriate skill invoked
- All remaining arguments passed through
- Expert guidance provided by skill
- Created extension follows best practices
- Validation rules met
</success_criteria>
---
## Quick Reference
| Command | Purpose | Skill Used |
|---------|---------|------------|
| `/create prompt <desc>` | Generate Claude prompt | create-meta-prompts |
| `/create command <desc>` | Create slash command | create-slash-commands |
| `/create agent <desc>` | Create subagent config | create-subagents |
| `/create skill <desc>` | Create skill file | create-agent-skills |
| `/create hook` | Create hook config | create-hooks |
## Examples
```bash
# Create optimized prompt
/create prompt "analyze code quality and suggest improvements"
# Create new slash command
/create command "deploy to staging environment"
# Create specialized subagent
/create agent "database migration specialist with rollback capability"
# Create agent skill
/create skill "TDD workflow with red-green-refactor cycle"
# Create event hook
/create hook
# (Will prompt for hook type and configuration)
```
## What Each Extension Type Provides
**Prompt** (create-meta-prompts):
- XML-structured templates
- Intelligent depth selection
- Context engineering patterns
- Progressive disclosure
- Research → Plan → Implement workflows
**Command** (create-slash-commands):
- YAML frontmatter configuration
- Argument hint and description
- Tool restrictions (allowed-tools)
- Dynamic context integration
- Skill integration for complex logic
**Agent** (create-subagents):
- Role and objective definition
- Tool selection for agent capabilities
- Prompt engineering for agent instructions
- Integration with Task tool
- XML structure compliance
**Skill** (create-agent-skills):
- Progressive disclosure structure
- Main SKILL.md with essential info
- Reference docs for deep dives
- Example patterns and anti-patterns
- Quality validation rules
**Hook** (create-hooks):
- EventExecute multiple sprints in parallel based on dependency graph from sprint-plan.md
Build and validate locally for projects without remote deployment (prototypes, experiments, local-only dev)
Execute multi-sprint epic workflow from interactive scoping through deployment with parallel sprint execution and self-improvement
Execute feature development workflow from specification through production deployment with automated quality gates
Analyze workflow state and provide context-aware guidance with visual progress indicators and recommended next steps
Initialize project documentation, preferences, or design tokens
Implement small bug fixes and features (<100 LOC) without full workflow. Use for single-file changes, bug fixes, refactors, and minor enhancements that can be completed in under 30 minutes.
Enter deep craftsman mode - question everything, plan like Da Vinci, craft insanely great solutions, then materialize to roadmap