Command Creator

Create custom slash commands for Claude Code plugins that users can invoke explicitly.

When to Use This Skill

Use when:

• Creating custom slash commands for plugins
• User requests "create a command" or "add a slash command"
• Building frequently-used prompt shortcuts
• Need user-invoked commands (vs autonomous skills)

Command vs Skill

Slash Commands (User-Invoked):

• Explicitly called by user: /command-name
• Good for frequently-used prompts
• Simple, single-file structure
• Manual invocation

Skills (Model-Invoked):

• Automatically used by Claude when relevant
• Good for complex capabilities
• Multi-file structure with resources
• Autonomous activation

Command File Structure

Commands are markdown files with optional frontmatter:

```yaml
---
description: Brief description of what the command does
argument-hint: [arg1] [arg2]
allowed-tools: Tool1, Tool2
model: sonnet
disable-model-invocation: false
---
```

# Command prompt content

$ARGUMENTS or $1, $2, $3 for argument access

Instructions

Step 1: Determine Command Purpose

Define what the command should do:

• What prompt/instruction does it provide?
• Does it need arguments?
• Should it reference files?
• Does it need to execute bash commands?
• Which tools are required?

Step 2: Create Command File

Filename: <command-name>.md (kebab-case)

Location:

• Plugin commands: <plugin>/commands/
• Project commands: .claude/commands/
• User commands: ~/.claude/commands/

Invocation: /command-name [arguments]

Step 3: Write Frontmatter (Optional)

---
description: Brief description of command
argument-hint: [arg1] [arg2]  # Shown in autocomplete
allowed-tools: Tool1, Tool2  # Restrict tools
model: sonnet  # Specific model
disable-model-invocation: false  # Allow SlashCommand tool
---

All frontmatter fields are optional.

Step 4: Write Command Content

After frontmatter, write the command prompt:

# Task Description

Your command instructions here.

## Arguments

- $ARGUMENTS: All arguments as single string
- $1: First argument
- $2: Second argument
- $3: Third argument, etc.

## File References

Use @file/path to reference files:
Review @src/main.js for errors

## Bash Execution

Use !`command` for bash execution (requires allowed-tools: Bash):
Current status: !`git status`

Step 5: Configure Arguments

All arguments (simple):

Fix issue #$ARGUMENTS

Usage: /fix-issue 123 high-priority Result: "Fix issue #123 high-priority"

Individual arguments (structured):

Review PR #$1 with priority $2 and assign to $3

Usage: /review-pr 456 high alice Result: "Review PR #456 with priority high and assign to alice"

Step 6: Add Bash Commands (Optional)

Include bash output in context:

Frontmatter:

allowed-tools: Bash(git add:*), Bash(git status:*), Bash(git commit:*)

Command content:

Current git status: !`git status`
Recent commits: !`git log --oneline -10`

Based on the above, create a commit message.

Bash commands execute before prompt is sent to Claude.

Step 7: Enable File References (Optional)

Reference files directly in command:

Compare @src/old-version.js with @src/new-version.js

Identify differences and suggest improvements.

User can then use command and file references are loaded.

Frontmatter Reference

description (optional)

Purpose: Shown in /help and autocomplete

Example:

description: Create git commit with conventional format

argument-hint (optional)

Purpose: Shown in autocomplete to guide users

Example:

argument-hint: [pr-number] [priority] [assignee]

Shown when user types /review-pr to indicate expected arguments.

allowed-tools (optional)

Purpose: Restrict which tools command can use

Example:

allowed-tools: Bash(git add:*), Bash(git status:*), Read, Write

model (optional)

Purpose: Use specific model for command execution

Options: Model string like claude-3-5-haiku-20241022 or alias like sonnet

Example:

model: claude-3-5-haiku-20241022

disable-model-invocation (optional)

Purpose: Prevent SlashCommand tool from invoking this command

Default: false (command can be invoked by SlashCommand tool)

Example:

disable-model-invocation: true

Use when command should only be manually invoked.

Example Commands

Simple Command (No Arguments)

File: commands/review.md

```yaml
---
description: Review code for security, performance, and style
---
```

Review this code for:
- Security vulnerabilities
- Performance issues
- Code style violations

Provide specific, actionable feedback.

Usage: /review

Command with Arguments

File: commands/fix-issue.md

```yaml
---
description: Fix GitHub issue by number
argument-hint: [issue-number]
---
```

Fix issue #$ARGUMENTS following our coding standards:

1. Read the issue description
2. Implement the fix
3. Add tests
4. Update documentation

Usage: /fix-issue 123

Command with Bash Execution

File: commands/commit.md

```yaml
---
description: Create git commit with conventional format
allowed-tools: Bash(git add:*), Bash(git status:*), Bash(git commit:*)
---
```

Current git status:
!`git status`

Current git diff:
!`git diff HEAD`

Recent commits:
!`git log --oneline -10`

Based on the above changes, create a conventional commit message and commit the changes.

Usage: /commit

Command with File References and Multiple Arguments

File: commands/compare-files.md

```yaml
---
description: Compare two files and analyze differences
argument-hint: [file1] [file2]
---
```

Compare @$1 with @$2

Analyze:
1. Key differences
2. Which approach is better
3. Recommended improvements

Usage: /compare-files src/old.js src/new.js

Command with Thinking Mode

File: commands/deep-analysis.md

```yaml
---
description: Perform deep analysis with extended thinking
---
```

Perform a comprehensive analysis of this codebase.

Think carefully about architecture, patterns, and potential improvements.

[Use extended thinking to reason through complex architectural decisions]

Extended thinking keywords trigger thinking mode automatically.

Namespacing

Organize commands in subdirectories:

commands/
├── git/
│   ├── commit.md        → /commit (git)
│   └── review-pr.md     → /review-pr (git)
└── deploy/
    ├── staging.md       → /staging (deploy)
    └── production.md    → /production (deploy)

Subdirectories appear in description but don't affect command name.

Plugin Commands

Plugin commands automatically namespace:

Without conflicts:

/command-name

With conflicts (disambiguation):

/plugin-name:command-name

SlashCommand Tool

Commands can be invoked by Claude via SlashCommand tool if:

• Command has description in frontmatter
• disable-model-invocation is not true
• Command appears in SlashCommand tool's character budget

To prevent autonomous invocation:

disable-model-invocation: true

Best Practices

1. Use clear descriptions: Help users discover commands
2. Provide argument hints: Guide users on expected input
3. Limit bash commands: Only necessary operations
4. Restrict tools: Grant only required permissions
5. Keep prompts focused: One clear purpose per command
6. Test with arguments: Verify $1, $2, etc. work correctly
7. Document usage: Include examples in description

Troubleshooting

Command not found:

• Check filename is correct
• Verify location (commands/ directory)
• Confirm no typos in command name
• Run /help to see available commands

Arguments not working:

• Use $ARGUMENTS for all args as string
• Use $1, $2, $3 for individual args
• Test with actual invocation
• Check argument-hint matches usage

Bash commands fail:

• Add allowed-tools with Bash
• Specify allowed commands: Bash(git status:*)
• Verify syntax: !command
• Check command is valid

SlashCommand tool not invoking:

• Add description in frontmatter
• Ensure disable-model-invocation is false or omitted
• Check character budget limit
• Reference command name with / in instructions

Reference

For complete command documentation, see references/command-format.md.