Slash Commands

CC 2.1.3: Custom “commands” now live in .claude/skills/ with disable-model-invocation: true. The /name syntax and behavior are identical.

Principle

A slash command is a Markdown file that defines a workflow. Claude executes it as if it had received that text as a prompt, with the ability to pass dynamic arguments.

File locations

.claude/skills/          # Project commands (team)
├── commit.md              → /commit
├── review-pr.md           → /review-pr
└── tech/
    └── deploy.md          → /tech:deploy

~/.claude/skills/        # Global commands (personal)
├── release.md             → /release
└── sync.md                → /sync

Global commands are available in all sessions, regardless of the project.

Invocation and arguments

/commit                    # No argument
/tech:deploy production    # Positional argument
/release minor             # Version bump

In the Markdown file, arguments are accessible via variables:

Deploy to environment: $ARGUMENTS[0]
With strategy: $ARGUMENTS[1]

If no argument provided: use "staging" by default.

The $0, $1 syntax is equivalent to $ARGUMENTS[0], $ARGUMENTS[1]. The old dot notation ($ARGUMENTS.0) is deprecated since v2.1.19.

Recommended structure

# Command Name

## Objective
What this command does in one sentence.

## Process
1. **Step 1** — detailed instructions
2. **Step 2** — detailed instructions
3. **Step 3** — detailed instructions

## Arguments
- $0: target environment (default: staging)

## Output format
Description of the expected result.

Recent native commands

Useful commands to create

Command vs Skill vs Agent

A command cannot embed additional reference files (use a skill for that). A command has no persistent memory (use an agent for that).

Practical tips

One command, one responsibility. As soon as a command does two distinct things, split it. Name with an action verb first (commit, release, sync) so that the intent is immediately readable in the list of available commands.