Claude Code Commands - Meta Reference

This skill provides the definitive reference for creating Claude Code slash commands. Use this when building new commands or improving existing command patterns.

2026 note: Custom slash commands have been merged into skills. Prefer

.claude/skills/<name>/SKILL.md

.claude/commands/<name>.md

When to Use This Skill

Use this skill when you need to:

• Create a new slash command for repeated workflows
• Add

  $ARGUMENTS

  handling to commands
• Invoke agents from commands
• Include file context or bash output in commands
• Organize commands for team sharing

Quick Reference

review.md

/review

$ARGUMENTS

/review auth.js

$ARGUMENTS = "auth.js"

$1

$2

/compare a.js b.js

$1 = "a.js"

${CLAUDE_SESSION_ID}

logs/${CLAUDE_SESSION_ID}.log

@file

@CLAUDE.md

!command

!git status

Command Locations

Claude Code supports both legacy command files and skill-based commands. Skills are the recommended format because they support frontmatter controls and bundled supporting files.

.claude/skills/<name>/SKILL.md

/name

~/.claude/skills/<name>/SKILL.md

/name

.claude/commands/<name>.md

/name

~/.claude/commands/<name>.md

/name

packages/*/.claude/skills/<name>/SKILL.md

/name

Nested discovery: Claude automatically discovers

.claude/skills/

.claude/commands/

Command Structure

Skill-based (recommended):

.claude/skills/
|-- review/SKILL.md           # /review
|-- test/SKILL.md             # /test
`-- deploy/SKILL.md           # /deploy

Legacy commands:

.claude/commands/
|-- review.md                 # /review
|-- test.md                   # /test
`-- deploy.md                 # /deploy

Command Template (Skill-Based)

---
name: command-name
description: Brief description for invocation and auto-loading
argument-hint: [path|#pr] [options]
allowed-tools: Read, Grep, Bash(git:*)
disable-model-invocation: false
---

# Command Title

[Clear instructions for what this command does]

User request: $ARGUMENTS

## Steps

1. [First action Claude should take]
2. [Second action]
3. [Third action]

## Output Format

[Specify expected output structure]

Frontmatter Fields (Skills)

name

description

argument-hint

allowed-tools

disable-model-invocation

true

user-invocable

false

/

model

context

fork

agent

hooks

Legacy

.claude/commands/*.md

allowed-tools Syntax

allowed-tools: Read, Grep, Bash(git:*)

Tool

Tool(prefix:*)

Bash(git:*)

Bash(npm test:*)

$ARGUMENTS Usage

Single Argument

# Code Review

Review the following file or code for quality, security, and best practices:

$ARGUMENTS

Focus on:
- Code quality issues
- Security vulnerabilities
- Performance concerns
- Best practice violations

Usage:

/review src/auth.js

Multiple Arguments

# Compare Files

Compare these two files and explain the differences:

$ARGUMENTS

Provide:
- Line-by-line diff
- Semantic changes
- Impact analysis

Usage:

/compare old.js new.js

Optional Arguments

# Run Tests

Run tests for the specified scope.

Scope: $ARGUMENTS

If no scope specified, run all tests.
If scope is a file, run tests for that file.
If scope is a directory, run tests in that directory.

Usage:

/test

/test auth/

/test login.test.ts

Positional Arguments

Use

$1

$2

# Compare Files

Compare $1 with $2.

Show:
- Line differences
- Semantic changes
- Which version is preferred

Usage:

/compare old.js new.js

$1 = "old.js"

$2 = "new.js"

File References (@ Prefix)

Include file contents directly in the command with

@

# Review with Context

Review this code following our standards.

Project standards:
@CLAUDE.md

Code to review:
$ARGUMENTS

Usage:

/review-context src/auth.js

Bash Execution (! Prefix)

Execute bash commands and include output with

!

# Smart Commit

Current status:
!git status --short

Recent commits:
!git log --oneline -5

Staged changes:
!git diff --cached

Generate a commit message for the staged changes.

Usage:

/smart-commit

Important: The

!command

Backtick Syntax

For inline execution, use backticks:

PR diff: !`gh pr diff`
Changed files: !`gh pr diff --name-only`

Command Patterns

Agent Invocation

# Security Audit

Perform a comprehensive security audit.

Target: $ARGUMENTS

Use the **security-auditor** agent to:
1. Scan for OWASP Top 10 vulnerabilities
2. Check authentication patterns
3. Review data validation
4. Analyze dependencies

Provide a severity-rated findings report.

Multi-Agent Orchestration

# Fullstack Feature

Build a complete fullstack feature.

Feature: $ARGUMENTS

Workflow:
1. Use **prd-architect** to clarify requirements
2. Use **system-architect** to design approach
3. Use **backend-engineer** for API implementation
4. Use **frontend-engineer** for UI implementation
5. Use **test-architect** for test coverage

Coordinate between agents and ensure integration.

Validation Command

# Pre-Commit Check

Validate changes before commit.

Files: $ARGUMENTS (or all staged files if not specified)

Checklist:
- [ ] All tests pass
- [ ] No linting errors
- [ ] No type errors
- [ ] No console.log statements
- [ ] No TODO comments
- [ ] No hardcoded secrets

Return READY or BLOCKED with details.

Command Categories

Development Commands

/review

/test

/debug

/refactor

Architecture Commands

/design

/architecture-review

/tech-spec

Security Commands

/security-scan

/secrets-check

/dependency-audit

Operations Commands

/deploy

/rollback

/incident

Naming Conventions

{action}

/review

{action}-{target}

/security-scan

{domain}-{action}

/pm-strategy

{tool}-{action}

/git-commit

Command vs Agent vs Skill

/command

Flow: User -> Command -> Agent -> Skill

Invocation Control

Control who can invoke commands using frontmatter:

/name

disable-model-invocation: true

/name

user-invocable: false

Example: User-Only Command

---
description: Deploy to production
disable-model-invocation: true
allowed-tools: Bash(kubectl:*), Bash(docker:*)
---

# Deploy

Deploy $ARGUMENTS to production cluster.

Claude cannot auto-invoke this - user must explicitly type

/deploy

Context Budget

Default skill/command description budget: 15,000 characters.

If many commands are excluded from context:

export SLASH_COMMAND_TOOL_CHAR_BUDGET=20000

Check with

/context

Best Practices

DO

# Good Command

Clear, specific instructions.

Target: $ARGUMENTS

1. First, analyze the target
2. Then, perform action X
3. Finally, output result Y

Expected output:
- Summary of findings
- Actionable recommendations

DON'T

# Bad Command

Do stuff with $ARGUMENTS.

Make it good.

Advanced Patterns

Conditional Logic

# Smart Review

Review target: $ARGUMENTS

If target is a PR number (e.g., #123):
  - Fetch PR details with `gh pr view`
  - Review all changed files

If target is a file path:
  - Review that specific file

If target is a directory:
  - Review all files in directory

Template with Options

# Generate Tests

Generate tests for: $ARGUMENTS

Options (parsed from arguments):
- `--unit` - Unit tests only
- `--e2e` - E2E tests only
- `--coverage` - Include coverage report

Default: Generate both unit and E2E tests.

Navigation

Resources

• references/command-patterns.md - Common patterns
• references/command-examples.md - Full examples
• data/sources.json - Documentation links

Related Skills

• ../claude-code-agents/SKILL.md - Agent creation
• ../claude-code-skills/SKILL.md - Skill creation
• ../claude-code-hooks/SKILL.md - Hook automation