Commands

Manually triggered instructions with the / prefix

What is a Command?

A command is a Markdown file that contains instructions for Claude. It is explicitly triggered by the user with the / prefix.

# Invocation examples
/work:work-explore "Understand the architecture"
/dev:dev-tdd "Implement the user service"
/qa:qa-security

File structure

Commands are organized by domain in .claude/commands/:

.claude/commands/
├── work/                    # Main workflow
│   ├── work-explore.md
│   ├── work-plan.md
│   ├── work-commit.md
│   └── work-pr.md
├── dev/                     # Development
│   ├── dev-tdd.md
│   ├── dev-api.md
│   └── dev-component.md
├── qa/                      # Quality
│   ├── qa-security.md
│   ├── qa-perf.md
│   └── wcag-audit.md
├── ops/                     # Operations
├── doc/                     # Documentation
├── biz/                     # Business
├── growth/                  # Growth
├── data/                    # Data
└── legal/                   # Legal

Anatomy of a command

Minimal structure

# Command title

Description of what the command does.

## Instructions

1. Step 1
2. Step 2
3. Step 3

## Expected output

Description of the expected result.

Advanced structure with frontmatter

---
description: Short description for the help
allowed-tools:
  - Read
  - Grep
  - Glob
  - Edit
---

# My Command

## Context
$ARGUMENTS

## Instructions
...

Special variable $ARGUMENTS

The $ARGUMENTS variable is replaced by the arguments passed to the command:

/dev:dev-tdd "Implement JWT authentication"

In the dev-tdd.md file:

## Context
$ARGUMENTS
# Becomes: "Implement JWT authentication"

Characteristics

Property	Value
Trigger	Manual (/name)
Context	Shared with the conversation
Tools	All available (unless restricted)
Visibility	The user sees the invocation

Difference with Skills and Agents

Aspect	Command	Skill	Agent
Trigger	Manual	Auto (keywords)	Auto (delegation)
Context	Shared	Fork	Isolated
Control	Total	Partial	Delegated
Use case	Explicit actions	Recurring behaviors	Autonomous tasks

Best practices

1. Consistent naming

domain-action

Examples:

• work-explore (workflow + explore)
• dev-tdd (development + TDD)
• qa-security (quality + security)

2. Clear instructions

## Instructions

IMPORTANT: Always explore the code before modifying.

YOU MUST follow the existing pattern.

NEVER modify configuration files without validation.

3. Output structure

## Expected output

### Format
- Section 1: Summary
- Section 2: Details
- Section 3: Recommendations

### Example
\`\`\`markdown
## Summary
...
\`\`\`

Create a new command

1. Create the file

# In the right domain
touch .claude/commands/dev/dev-my-command.md

2. Write the content

# My New Command

## Context
$ARGUMENTS

## Instructions

1. Analyze the request
2. Execute the action
3. Verify the result

## Output

Provide a structured report.

3. Test

/dev:dev-my-command "Command test"

Command examples

Simple command (exploration)

# Work Explore

Explore and understand an existing codebase.

## Context
$ARGUMENTS

## Instructions

1. Identify the main files (package.json, README, etc.)
2. Analyze the folder structure
3. Spot the patterns and conventions
4. Document the key dependencies

## Output

Provide a structured overview of the project.

Complex command (workflow)

# Work Flow Feature

Complete workflow to implement a new feature.

## Context
$ARGUMENTS

## Steps

### 1. Exploration
Use /work:work-explore to understand the context.

### 2. Specification
Use /work:work-specify to define the User Stories.

### 3. Planning
Use /work:work-plan to create the implementation plan.

### 4. Development
Use /dev:dev-tdd to implement with tests.

### 5. Review
Use /qa:qa-review to check the quality.

### 6. Delivery
Use /work:work-pr to create the Pull Request.

List the available commands

In Claude Code, use /help to see the available commands, or explore the .claude/commands/ folder directly.

---

See also

• Skills - Automatic behaviors
• Agents - Autonomous sub-agents
• Commands catalog - All claude-base commands