Claude Code Custom Slash Commands |

\n\n

Skills (skills) are Claude Code’s extension mechanism, enhancing Claude’s capabilities through custom slash commands (e.g., /deploy, /explain-code).

\n\n

Each custom command is essentially an independent Skill file, capable of encapsulating complex workflows, project specifications, and common operations for quick invocation in any session.

\n\n

Custom commands have been integrated into the Skills system — both the command file at .claude/commands/deploy.md and the skill file at .claude/skills/deploy/SKILL.md create the /deploy command, functioning identically.

\n\n

Essentially, .claude/commands/ is a shorthand for Skills; both directory structures are fully equivalent.

\n\n

Why Use Custom Slash Commands

\n\n

The core value of custom slash commands lies in standardization + reuse:

\n\n

Standardize workflows (Solidify workflows): Encapsulate multi-step operations (e.g., “deploy to production”) into a single command
Ensure team consistency (Ensure team consistency): Embed project standards (e.g., “code review checklist”) into commands to guarantee all team members follow the same criteria
Reduce cognitive load (Reduce cognitive load): No need to memorize complex command sequences — a single phrase triggers the full workflow
Minimize repetitive input (Reduce repetitive input): High-frequency tasks become single invocations instead of multi-step operations

\n\n

\n
Anthropic’s internal team experience: Tasks executed more than twice per day should be turned into Skills. Do it right once, and it works continuously.
\n

\n\n

File Structure

\n\n

Each Skill is a directory containing a required SKILL.md file:

\n\n

my-skill/\n├── SKILL.md        # Main instruction file (required)\n├── template.md     # Template filled by Claude\n├── examples/\n│   └── sample.md   # Example demonstrating expected format\n└── scripts/\n    └── validate.sh # Script executable by Claude\n

\n\n

The file format consists of two parts:

\n\n

YAML frontmatter (between --- delimiters): For configuring metadata
Markdown body: Instructions Claude follows when invoking the skill

\n\n

Create Your First Custom Slash Command

\n\n

1. Create the Skill Directory

\n\n

Create the skill directory at an appropriate location. The location determines the command’s scope:

\n\n

mkdir -p ~/.claude/skills/explain-code\n# Or create within a project:\nmkdir -p .claude/skills/explain-code\n

\n\n

2. Write the `SKILL.md` File

\n\n

Create SKILL.md in the directory, including YAML frontmatter and instructions:

\n\n

Example

\n\n

---\nname: explain-code\n\ndescription: Explain code using visual diagrams and analogies. Use when explaining how code works, walking through a codebase, or answering "how does this work?"\n---\n\nWhen explaining code, always include the following:\n\n1. **Start with an analogy**: Compare the code to something in everyday life\n2. **Draw a diagram**: Use ASCII art to illustrate flow, structure, or relationships\n3. **Step-by-step explanation**: Walk through what happens, line by line\n4. **Point out a pitfall**: What common errors or misconceptions exist?\n

\n\n

3. Use the Custom Command

\n\n

Invoke directly in Claude Code:

\n\n

/explain-code src/auth/login.ts\n

\n\n

Claude will automatically match skills matching the description, or you may invoke explicitly.

\n\n

Configuration Fields Explained

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Field	Description
`name`	Display name; becomes the slash command (format: lowercase letters, digits, hyphens; max 64 chars)
`description`	Skill functionality and when to use it (recommended; Claude’s auto-invocation relies on this)
`argument-hint`	Hint shown during auto-completion (e.g., )
`disable-model-invocation`	Set to `true` to prevent Claude from auto-invoking; only manual invocation works
`user-invocable`	Set to `false` to hide from `/` menu; only Claude can invoke
`allowed-tools`	List of tools Claude may use without asking
`context`	Set to `fork` to run in a forked sub-agent
`agent`	Sub-agent type used when `context: fork`
`paths`	List of glob patterns limiting file scope for skill activation

\n\n

Invocation Control

\n\n

Control who can invoke the skill via frontmatter:

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Configuration	You can invoke	Claude can invoke
(default)	Yes	Yes
`disable-model-invocation: true`	Yes	No
`user-invocable: false`	No	Yes

\n\n

Variable Substitution

\n\n

Use variables in skill content; Claude replaces them automatically:

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Variable	Description
`$ARGUMENTS`	All arguments passed during invocation
`$ARGUMENTS`	Access specific argument by 0-based index
`$N`	Shorthand for `$ARGUMENTS`
`${CLAUDE_SESSION_ID}`	Current session ID
`${CLAUDE_SKILL_DIR}`	Directory containing the skill’s `SKILL.md` file

\n\n

Skill Storage Locations

\n\n

Skills can reside in multiple locations, each with different scopes:

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Location	Path	Scope
Enterprise-wide	Determined by hosting settings	All users in the organization
Personal	`~/.claude/skills/<skill-name>/SKILL.md`	All projects
Project-level	`.claude/skills/<skill-name>/SKILL.md`	Current project only
Plugin	`<plugin>/skills/<skill-name>/SKILL.md`	Scope of plugin activation

\n\n

\n
Files in .claude/commands/ remain valid and support the same frontmatter. If a skill and command share a name, the skill takes precedence.
\n

\n\n

Advanced Examples

\n\n

Example 1: Deployment Command (Manual Invocation Only)

\n\n

Example

\n\n

---\nname: deploy\n\ndescription: Deploy the application to production\n\ndisable-model-invocation: true\n---\n\nDeploy $ARGUMENTS to production:\n\n1. Run the test suite\n2. Build the application\n3. Push to the deployment target\n

\n\n

Example 2: Component Migration Command with Arguments

\n\n

Example

\n\n

---\nname: migrate-component\n\ndescription: Migrate a component from one framework to another\n---\n\nMigrate the $0 component from $1 to $2.\n\nPreserve all existing behavior and tests.\n\nUsage: /migrate-component SearchBar React Vue\n

\n\n

Example 3: PR Summary Command with Dynamic Context

\n\n

Example

\n\n

---\nname: pr-summary\n\ndescription: Summarize changes in a Pull Request\n\ncontext: fork\n\nagent: Explore\n\nallowed-tools: Bash(gh *)\n---\n\n## Pull Request Context\n\n - PR diff: !`gh pr diff`\n - PR comments: !`gh pr view --comments`\n - Changed files: !`gh pr diff --name-only`\n\n## Your Task\n\nSummarize this Pull Request...\n

\n\n

Built-in Slash Commands

\n\n

Claude Code provides several built-in commands out of the box:

\n\n\n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n \n

Command	Function
`/batch <instruction>`	Orchestrate large-scale changes across the codebase
`/claude-api`	Load API reference materials
`/debug`	Enable debug logging
`/loop <prompt>`	Repeat prompt execution at intervals
`/simplify`	Review code quality issues

\n\n

Best Practices

\n\n

1. How to Write `description`

\n\n

The description is the basis for Claude’s auto-invocation — make it clear:

\n\n

Specify trigger conditions: “Invoke when user requests code review / analysis / quality check”
State prerequisites: “Use only after spec document is confirmed”
Define boundaries: Clarify when NOT to invoke

\n\n

2. Single Responsibility Principle

\n\n

Each skill should do one thing, with clear input/output:

\n\n

/deploy: Deployment only
/test: Testing only
/review: Code review only

\n\n

3. Set Invocation Permissions as Needed

\n\n

Dangerous operations (deploy, delete): Set disable-model-invocation: true
Claude-assisted tasks: Set user-invocable: false
General workflows: Use default configuration

\n\n

4. Leverage `allowed-tools` for Permission Control

\n\n

Assign minimal permissions for security-critical operations:

\n\n

Example

\n\n

---\nname: read-only-analyzer\n\ndescription: Read-only code analysis for review and understanding\n\nallowed-tools: Read, Grep, Glob, Bash(gh *) # Only these tools allowed\n---\n\nPerform read-only code analysis...\n

\n\n

Team Sharing Strategy

\n\n

Commit project-level skills to Git for team sharing:

\n\n

my-project/\n├── .claude/\n│   ├── skills/\n│   │   ├── deploy/\n│   │   │   └── SKILL.md\n│   │   ├── test/\n│   │   │   └── SKILL.md\n│   │   └── review/\n│   │       └── SKILL.md\n│   └── commands/ # Also supports commands/ directory\n└── CLAUDE.md\n

\n\n

Reference and describe project-specific skills in CLAUDE.md:

\n\n

Custom Slash Commands

\n\n

This project provides the following custom slash commands:

\n\n

/deploy — Deploy to specified environment (dev/staging/prod)
/test — Run test suite or specified module
/review — Perform standard code review

\n\n

For detailed usage instructions, refer to @.claude/skills/README.md.

\n\n

Frequently Asked Questions

\n\n

Q: What’s the difference between skills and commands directories?

\n\n

There is no essential difference. .claude/commands/ is shorthand for .claude/skills/; both work identically. If a skill and command share a name, the skill takes precedence.

\n\n

Q: How to pass multiple arguments during invocation?

\n\n

Wrap multiple arguments in quotes: /migrate SearchBar "from:React to:Vue", then use $ARGUMENTS in the skill to retrieve the full string.

\n\n

Q: Can skills be used in sub-agents?

\n\n

Yes. By setting context: fork and the agent field, the skill runs in an independent context within a sub-agent.

\n\n

Q: How to make Claude remember commonly used skills?

\n\n

List all available custom commands and their use cases in CLAUDE.md; Claude will automatically match appropriate skills based on descriptions.

YouTip