Claude Code slash commands: reference and custom commands

How the menu works

Type / as the first character of a message and Claude Code lists every command available in your build, filtering as you type. That list is the only authoritative reference, because availability depends on your platform, plan, and environment. Four rules govern the whole system.

• A command is recognised only at the start of a message. Everything after the command name becomes its arguments.
• You can stack several at the start of one message: /write-tests /fix-issue 123 loads both and passes 123 to each.
• A command sent while Claude is responding queues behind the current turn. /status, /tasks, and /usage are exceptions and run immediately.
• Not every command appears for every user. If something in this guide is missing from your menu, that is expected, not broken.

Session and context

These are the ones you will use every day. Context management is the single highest-leverage habit in Claude Code, because cost, rate-limit consumption, and answer quality all degrade as context fills.

Model, effort, cost, and configuration

Switching model mid-session is more useful than it sounds. Plan an approach on a large model, then drop to a smaller one for the mechanical implementation, all inside the same conversation. /effort is the finer dial: low for subagents and simple edits, high or xhigh for anything you would be annoyed to get wrong.

Project setup and repo knowledge

The bundled skills hiding in the menu

A significant share of the / menu is not built-in commands at all. They are bundled skills: prompts Claude can also invoke on its own when relevant. They look identical when you type them and behave differently when you do not.

Writing your own

This is where the real leverage is, and the mechanics changed in 2026: custom commands merged into skills. Both shapes still work and both produce a command you invoke the same way.

A command that only you can fire

---
name: review
description: Review the current diff like a senior engineer
disable-model-invocation: true
argument-hint: "[path]"
---

Review the staged and unstaged changes in this repository.

Focus on, in order:
1. Correctness. Does it do what the commit message claims?
2. Failure modes. What happens on empty input, network failure, concurrency?
3. Tests. Is the new behaviour actually covered?
4. Consistency with the patterns already in this codebase.

Do not comment on formatting; the linter handles that.
Report findings most severe first. If you find nothing real, say so
rather than inventing something to justify the review.

disable-model-invocation: true is the important line. It means only you can trigger this, which is what you want for anything with side effects: /deploy, /commit, /send-slack-message. You do not want Claude deciding to deploy because the code looks ready.

Arguments

Four substitution forms, and the indexed one is 0-based, which trips up anyone who learned the old $1-is-first convention.

---
name: fix-issue
description: Fix a GitHub issue end to end
disable-model-invocation: true
arguments: [issue]
allowed-tools: Bash(gh *)
---

Fix GitHub issue #$issue.

1. Run `gh issue view $issue` to read it.
2. Find the relevant code. Do not guess at file locations.
3. Write a failing test that reproduces the issue first.
4. Fix it. Keep the change minimal.
5. Run the full suite.
6. Commit as `fix: <summary> (#$issue)`.

> /fix-issue 482

Indexed placeholders use shell-style quoting, so /my-skill "hello world" second makes $0 expand to hello world. An indexed placeholder with no matching argument stays in the text unchanged; a named one expands to an empty string. To write a literal $1.00 in prose, escape it as \$1.00.

Dynamic context: run a command before Claude sees anything

This is the feature that turns a prompt template into a real tool. The !`command` syntax runs a shell command and substitutes its output into the prompt before the prompt is sent. Claude never sees the command, only the data.

---
name: pr-summary
description: Summarize changes in a pull request
context: fork
agent: Explore
allowed-tools: Bash(gh *)
---

## Pull request context
- PR diff: !`gh pr diff`
- PR comments: !`gh pr view --comments`
- Changed files: !`gh pr diff --name-only`

## Your task
Summarize this pull request for a reviewer who has not read the code...

• The inline form is recognised only when ! starts a line or follows whitespace. KEY=!`cmd` is left as literal text and does not run.
• Substitution runs once over the original file. Command output is inserted as plain text and never rescanned, so a command cannot emit a placeholder for a later pass.
• For multi-line commands, open a fenced block with ```! instead of using the inline form.
• context: fork runs the skill in its own subagent, so a huge diff never lands in your main conversation. It runs in the background by default; set background: false to wait for the result.

The frontmatter worth knowing

Commands worth writing for your repo

• Release checklist. The eight steps your team always forgets one of. Mark it disable-model-invocation: true.
• New module scaffold. Encodes your directory conventions so the agent stops inventing its own.
• Debug this failure. Points at your actual log locations and test commands, with !`tail -n 200 var/log/app.log` pulling the log in for free.
• Update the docs. Names the files that must change together when an API changes.

Commands, skills, and hooks

Three overlapping extension points. Now that commands and skills are one file format, the question is not "which file do I write" but "who is allowed to fire it".

Questions people ask

Keep reading