
Slash commands: saved prompts you run with a /shortcut
In short: Slash commands are saved, reusable prompts in coding agents that you run by typing /name, and in Claude Code they are now a kind of skill.
Slash commands are shortcuts in coding agents and chat tools that run a saved instruction when you type /name. Some are built in, such as clearing context or switching models. Custom ones are Markdown files whose contents become the prompt, optionally with arguments. In Claude Code, custom commands have been merged into skills, according to the documentation read in November 2026.
Slash commands in plain words
Think of a speed-dial button. You stop explaining the same multi-step request every time; you press /review and it is sent exactly the way you wrote it once.
How it works
Claude Code loads commands from skill folders such as .claude/skills/<name>/SKILL.md and from older .claude/commands/ files, which the documentation says keep working. Typing / lists what is available. Running one inserts its text as the prompt, substituting arguments such as $ARGUMENTS. Frontmatter can pre-approve tools with allowed-tools or pick a model.
Example
.claude/skills/fix-issue/SKILL.md:
---
description: Fix a GitHub issue by number
disable-model-invocation: true
---
Read GitHub issue #$ARGUMENTS with the gh CLI.
Reproduce it with a failing test, fix it, run the full test suite,
and summarise the change in three bullets.
Run it with /fix-issue 482.
Where you'll meet it
- Claude Code built-in commands such as
/clear,/compactand/model - Project skill and command folders shared through the repository
- Plugins that bundle commands for teams
- Other agents and chat tools with similar shortcut systems
What it changes for you
For developers, commands remove re-typing and make results more consistent because the prompt does not drift. For teams, shared commands are a lightweight way to standardise workflows such as reviews, releases or incident write-ups without building tooling.
Slash commands vs agent skills
In Claude Code the line has blurred. A command file is now treated as a skill, and Claude can invoke it without you typing it unless you set disable-model-invocation: true. What still differs is who triggers the instruction: you, by typing /name, or the model, by matching the description to the task.
Common mistake
My view: stuffing a whole runbook into one command. Long commands are hard to maintain and burn context. Keep the command short and point it at scripts or supporting files for the detail.
Related terms
Agent skills · Claude Code · Claude Code hooks · Headless mode · Prompt engineering
Keep each slash command short and move the detail into scripts or supporting files.


