# Build safer Claude Code slash commands

Reusable slash commands can be difficult to structure, document, and maintain. This skill provides patterns for Markdown prompts, frontmatter, arguments, interactions, and controlled command context.

## Install

```bash
npx skillstore add anthropics/command-development
```

## Metadata

- Status: approved
- Slug: anthropics-command-development
- Version: 0.2.0
- Author version: 0.2.0
- Skillstore revision: r2
- Version status: valid
- Tree hash: 0bbddf32c86415d3c718464ad5f757aa23f1724e5ef3c99ea4671085af9b327b
- Author: anthropics
- GitHub username: anthropics
- License: MIT
- Repository: https://github.com/anthropics/claude-plugins-official/tree/main/plugins/plugin-dev/skills/command-development
- Ref: 80999bf530a7874d7bedf8ce202001ecb4c4f5e0
- Supported tools: Claude, Codex, Claude Code
- Audit status: complete
- Agent install advisory: confirmation\_required
- Manual install advisory: allowed
- Artifact signature: available
- Audit attestation: unavailable
- Human verification: not\_verified
- Risk factors: external\_commands, filesystem, network
- Quality score: 76
- Public page: https://skillstore.pages.dev/skills/anthropics-command-development
- Manifest: https://skillstore.pages.dev/api/skills/anthropics-command-development/manifest

## Capabilities

- Explains the Markdown structure of project, personal, and plugin slash commands.
- Documents YAML frontmatter fields such as description, allowed-tools, model, and argument-hint.
- Shows patterns for command arguments, file references, and namespaced command organization.
- Provides examples for interactive commands that use AskUserQuestion.
- Describes Bash output insertion and permission-aware command design.
- Includes reference material for testing, documentation, plugin features, and marketplace considerations.

## Use Cases

- Standardize team reviews: Create a project command that guides consistent code reviews with scoped tools and clear output expectations.
- Build personal shortcuts: Design personal commands for repeated development tasks without copying prompts between sessions.
- Document plugin commands: Define plugin command interfaces, arguments, and validation steps for users of a Claude Code plugin.

## Prompt Templates

### Create a basic command

```
Create a Claude Code slash command named /review-notes. It should review the current notes for clarity and return concise improvement suggestions.
```

### Add frontmatter

```
Create a project slash command named /test-summary with YAML frontmatter. Limit tools to read-only Git and test commands, and explain each field.
```

### Handle user input

```
Design an interactive /release-check command. Ask for the target environment, validate the response, and list checks before any action.
```

### Review command safety

```
Review this slash command for unsafe argument handling, excessive tool access, and unclear side effects. Provide a safer revised design.
```

## Limitations

- It provides guidance and examples; it does not create or install commands automatically.
- Shell examples may need adaptation for your operating system, repository, and permissions.
- It cannot guarantee that generated commands are safe for every project or deployment environment.
- Claude Code features and frontmatter support may change over time.

## Best Practices

- Use a short description and an argument hint so users can discover the command.
- Grant the minimum tools needed and validate arguments before shell operations.
- Describe side effects and ask for confirmation before changes, deployments, or network operations.

## Anti Patterns

- Do not grant all tools when a small read-only set is sufficient.
- Do not pass unvalidated user arguments directly to shell commands.
- Do not write commands as user-facing announcements instead of instructions for Claude.

## Security Audit

- Audited at: 2026-07-18T10:07:16.557\+00:00
- Summary: All 400 reviewed static matches occur in Markdown documentation and command-development examples. They describe optional shell, filesystem, network, and diagnostic operations, but the skill includes no executable installer, automatic command, credential collection, or exfiltration behavior. No prompt-injection text or malicious intent was found. Static review was capped at 400/429 representative findings; omitted static matches are unconfirmed, so automatic publishing stays disabled until manual review.

## Stats

- Views: 254
- Downloads: 7
- Favorites: 0
- Popularity score: 0