Slash Commands (now Skills)

Short answer: slash commands still exist, but they are now just skills with user-invocable: true. If you’re writing a new one, read Skills Authoring. If you have legacy commands/*.md files from before Claude Code 2.1.88, read the migration section below.

---

Historical Note

Before CC 2.1.88, user-invocable slash commands lived in two directories:

~/.claude/commands/       # Global
.claude/commands/         # Project-scoped

Each was a single Markdown file with YAML frontmatter. In 2.1.88, Anthropic merged commands into skills — the commands/ directories were deprecated and any /<name> invocation that used to be a command is now a skill with user-invocable: true.

There is no longer a separate command type. Skills are the one primitive.

---

Invocation Patterns

Plain skill invocation

/skill-name
/skill-name argument text here

The first form runs the skill with empty arguments. The second passes everything after the name as $ARGUMENTS (see Skills Authoring → Arguments).

Plugin-namespaced invocation

Skills installed via a plugin are addressed with the plugin prefix:

/plugin-name:skill-name

Example: the superpowers plugin exposes /superpowers:brainstorming, /superpowers:writing-plans, etc.

Built-in slashes (2.1.108+)

Several built-in commands are now rewired through the Skill tool:

Slash	What it does
/init	Initialize a new CLAUDE.md for the project
/review	Review a pull request
/security-review	Security review of pending changes

Plus the classic built-ins (not implemented as skills): /help, /clear, /cost, /config, /doctor, /model, /resume, /compact, /release-notes, and others. These are hard-wired into the CLI.

---

Authoring a User-Invocable Skill

The short form — full details in Skills Authoring:

---
name: my-command
description: "Short action-verb description. Use when user says 'my command' or types /my-command."
user-invocable: true
argument-hint: "[optional argument description]"
---

# My Command

$ARGUMENTS

## Workflow

Steps to execute using the argument text above.

Key fields:

• user-invocable: true — required to show in the / menu
• argument-hint — text displayed in the UI when the user types /my-command<space>
• $ARGUMENTS — expanded to whatever the user typed after the command name

To prevent accidental automatic invocation (e.g., for deploy commands that must be explicit):

disable-model-invocation: true

This keeps the skill user-invocable via slash but prevents Claude from auto-triggering it based on conversation context.

---

Migrating Legacy commands/ Files

If you have files in ~/.claude/commands/ or .claude/commands/ from before 2.1.88, migrate them to the skills layout:

Before (legacy command file)

~/.claude/commands/
  deploy.md
  ship-it.md
  weekly-review.md

Each file had frontmatter and a body. The filename (minus .md) was the slash name.

After (skill directory)

~/.claude/skills/
  deploy/
    SKILL.md
  ship-it/
    SKILL.md
  weekly-review/
    SKILL.md

The directory name becomes the slash name. SKILL.md replaces the old command file.

Migration steps

# For each legacy command file:
OLD="deploy"
mkdir -p ~/.claude/skills/$OLD
mv ~/.claude/commands/$OLD.md ~/.claude/skills/$OLD/SKILL.md

Then edit SKILL.md and ensure the frontmatter has:

---
name: deploy                         # optional; directory name is used if omitted
description: "Deploy to production. Use when user says /deploy or 'deploy to prod'."
user-invocable: true
---

Important frontmatter hygiene during migration:

• Rename allowed_tools → allowed-tools (hyphenated, not underscored — the old form is silently ignored)
• Convert JSON-array tool lists to comma-separated strings: ["Read", "Bash"] → Read, Bash
• Add user-invocable: true if the old command was invocable (most were, by default)
• Update description to include “Use when…” trigger text

Then remove the empty commands/ directory (if nothing else remains in it):

rmdir ~/.claude/commands 2>/dev/null || true

---

Common Migration Issues

Problem	Cause	Fix
Slash doesn’t appear in / menu	Missing user-invocable: true	Add the field
Tool restrictions ignored	Used allowed_tools (underscore)	Rename to allowed-tools
Arguments not expanded	Legacy command used ``	Replace with $ARGUMENTS
Tool list parsed as empty	Used JSON array format	Use comma-separated string
Slash fires when unwanted	Model auto-invokes from context	Add disable-model-invocation: true

---

See Also

• Skills Authoring — frontmatter fields, body structure, arguments, testing
• Skills Maintenance — keeping user-invocable skills healthy
• CLI Flags and Env Vars — built-in slashes and CLI options