Tutorial 02 - Commands: the slash menu

Module 02 · ~15 minutes · hands-on

Everything behind the / key is one of two things: a built-in command (fixed logic shipped with Claude Code: /clear, /permissions, /mcp) or a skill (prompt-based and extensible, including every custom command you will ever write). This tutorial builds fluency with the built-ins and ends with the fact that sets up the rest of the day: custom commands are skills.

Prerequisites

• Tutorial 01 completed (a repo with an AGENTS.md)

Steps

1. Open the menu

In a session, type / and nothing else. Scroll the list. Type a few letters to filter. Note: entries vary by platform and plan, so your list will not exactly match your neighbor’s.

2. The working set

Run each of these once now:

Command	What it does	Why it matters
/context	Colored-grid visualization of the context window	The tool for understanding what your config costs
/permissions	Allow/ask/deny rules for tools	The guardrail surface; previews Plugins/MCP
/mcp	MCP server status + authentication	Module 05 home base
/compact	Summarize the conversation to free space	Use at natural breakpoints; takes optional focus instructions
/clear	Wipe history, start fresh	Cheaper than /compact when you don’t need the thread
/rewind	Roll back conversation and/or code	The undo button most people don’t know exists
/model	Switch model / configure effort	—
/skills	List available skills	Module 03 home base
/doctor	Diagnose installation and settings	First stop when anything misbehaves

3. Read your context bill

Run /context immediately after starting a fresh session in your repo. Identify: system prompt, your AGENTS.md, skill descriptions, MCP tools. This view makes the rest of the day’s trade-offs concrete, because everything you add in Tutorials 03–05 shows up here.

4. Create a custom command (it’s a skill)

$ mkdir -p .claude/skills/standup
$ cat > .claude/skills/standup/SKILL.md << 'EOF'
---
description: Summarize my recent work as a standup update. Use when asked for a standup.
---

Summarize the last day of git history by the current author as a
three-bullet standup update: done, in progress, blocked.

!`git log --author="$(git config user.name)" --since=yesterday --oneline`
EOF

Restart the session (new top-level skill folders need a restart; edits to existing ones don’t), then:

> /standup

Expected: the git log runs before the model sees the prompt (that’s the !`…` dynamic-injection syntax; more in Tutorial 03), and you get a three-bullet update.

5. Know the legacy form

.claude/commands/standup.md (flat file, no frontmatter required) also creates /standup. Commands and skills merged; the skill form wins name conflicts and is the one to write going forward.

✓ Checkpoint

You can answer, without looking: where does /standup come from, what’s the difference between /clear and /compact, and what does /context show that /cost doesn’t.

Exercises

1. Filter intuition. Find three built-ins you’ve never used, run them, and decide whether they earn a spot in your working set.
2. Arguments. Extend /standup to accept a day count: use $ARGUMENTS in the body (--since="$ARGUMENTS days ago"), declare argument-hint: <days>, and run /standup 3.
3. MCP preview. If any MCP server is already configured, type /mcp__ and see whether it exposes prompts as commands.

Troubleshooting

Symptom	Likely cause	Fix
/standup doesn’t appear	New top-level skill dir mid-session	Restart the session
Command exists on a colleague’s machine, not yours	Platform/plan-dependent visibility	Compare claude --version; check the docs for availability notes
! injection didn’t run	Setting disabled or syntax position	Must be at line start; check disableSkillShellExecution
/review missing	Deprecated as a built-in	claude plugin install code-review@claude-plugins-official