creating-commands

Slash commands in Claude Code use YAML frontmatter to define metadata (description, arguments, allowed tools, model settings) followed by a prompt block. Structure your command with argument placeholders like `$1`, `$2`, or `$ARGUMENTS`, then use bash execution blocks with backticks to run shell commands and pull dynamic context into your prompt.

Yes. Use backtick syntax (`` `command` ``) within your prompt to execute bash commands directly. This lets you dynamically gather context—like git status or file contents—and pass results into Claude's reasoning, enabling workflows that combine shell execution with AI analysis.

Required YAML frontmatter fields are: description, argument-hint, allowed-tools, model, and disable-model-invocation. These fields control what your command does, which tools Claude can access, and whether to invoke the model. Optional fields include file references and extended-thinking prompts to customize behavior.

Design a slash command that runs git operations (like `git status` or `git log`) using bash blocks, then passes the output to Claude for analysis or summarization. Combine multiple commands in sequence—for example, fetch git status, then ask Claude to summarize changes—turning complex git tasks into a single command invocation.

Yes. Use positional arguments (`$1`, `$2`, `$3`) for ordered inputs or `$ARGUMENTS` to capture all arguments as a string. Define what arguments your command expects in the argument-hint frontmatter field, and reference them in your prompt or bash blocks to make commands flexible and reusable.

Simple commands execute a single task with basic input/output. Workflow commands chain multiple steps—context gathering, validation, generation—using sequential bash execution and Claude reasoning to handle complex development tasks like testing pipelines or code generation sequences in one invocation.