Slash Commands

Claude Code Part 2 AI Docs Explore in Brain

Slash Commands

Slash Commands are one of Claude Code’s most powerful features — and one of the biggest reasons it has become such a productive coding tool. This guide covers everything you need to know, from built-in commands to sessions to model management.


🤔 What are Slash Commands?

Slash Commands are shortcuts you type inside a Claude Code session, starting with /, that trigger a specific predefined action or workflow instantly — without writing a full prompt.

When developers work on software, they repeat the same types of tasks constantly: reviewing sessions, switching models, checking usage, exporting conversations. Writing a full prompt each time is tedious.

Claude Code’s developers asked a smart question:

What if we convert these reusable patterns into single-word commands?

The result is Slash Commands — you type one word, and an entire workflow executes.


🗂️ Types of Slash Commands

There are two categories:

Type Who Creates It When Available
Built-in Commands Anthropic Immediately after installing Claude Code
Custom Commands You (the developer) When you create a .claude/commands/ file in your project

Custom commands are perfect for project-specific repetitive workflows — for example, “run tests → lint → commit” as a single /ship command.


💡 You’ve Already Used One

Even if this is your first time hearing the term, you used a Slash Command in the previous video.

When closing the Claude session, we typed:

/exit

That’s a Slash Command. Instead of prompting “Please end this session”, Claude Code gives you a one-word shortcut that does it reliably every time.


📋 Sessions — The Foundation

Before diving deeper into commands, you need to understand sessions — because most Slash Commands revolve around them.

What is a Session?

A session is one complete conversation with Claude Code.

Start:   claude          ← session begins
  ...    (your work)
End:     /exit           ← session ends

Every session contains:

  • A unique session ID
  • Your complete prompt history
  • Claude’s replies
  • All tool calls (file reads, Bash commands, web searches)
  • Bash commands you executed in Bash mode

Sessions are automatically saved to disk and can be resumed at any time, even days later.


▶️ Resuming Old Sessions

claude -r

This opens a list of all your previous sessions. Select one to resume it exactly where you left off — Claude remembers everything.

🔀 Switching Mid-Session

If you’re already inside a session and want to jump to a different one:

/resume

Claude shows your session list and lets you switch without exiting.


✅ Best Practices for Sessions

One session = One task. This is the golden rule. Mixing multiple features in one session pollutes the context and degrades Claude’s focus.

Here’s how to structure sessions on a real project:

Feature Session
Login System login-feature
Registration registration-feature
Dashboard dashboard-feature
Payment Integration payments-feature

Close each session when the feature is done. Start fresh for the next one.


🏷️ Rename Sessions Immediately

When Claude starts a new session, it auto-names it based on your first prompt — which is usually ugly and unhelpful.

Rename it right away:

/rename login-feature

Now when you run claude -r, you’ll see a clean, meaningful list instead of auto-generated titles.


💾 Commit at Milestones

Every time you complete a meaningful step inside a session, create a Git commit. This gives you checkpoints to roll back to if something goes wrong.

We’ll commit after every feature throughout this playlist. Treat commits as save points in a video game.


📎 Command Reference

🔤 /btw — Ask a Side Question

This is one of the most underrated commands. Suppose Claude is building your Login feature and you suddenly wonder:

“What is Flask anyway?”

That’s not relevant to the login task — adding it to the conversation would pollute the context.

Instead:

/btw What is Flask in Python?

Claude answers the question outside the session context. After reading it, press Space — the answer disappears and the main conversation is untouched.

Use /btw for conceptual questions, quick lookups, or anything that’s not directly part of the current task. It keeps your context window clean.


📤 /export — Save a Conversation

Before making large refactoring changes, export your conversation as a backup:

/export session-backup.md

Claude creates a Markdown file inside your project directory with the complete conversation history. You can feed it back later as context:

Read session-backup.md to understand the login feature we built.

🔁 /logout and /login

/logout    ← Signs out of Claude Code
/login     ← Starts the login flow again

Useful when:

  • You have a personal account and a work account
  • You need to switch between API keys
  • You’re setting up Claude Code on a new machine

After /logout, restarting Claude walks you through theme selection → login method → browser authorization.


🤖 Claude Models — Choosing the Right One

Claude Code uses Anthropic’s three models, each designed for different use cases:

Model Comparison

Model Speed Quality Cost Best For
Opus Slow ⭐⭐⭐⭐⭐ Highest 💰💰💰 Most expensive Planning, architecture, complex reasoning
Sonnet Fast ⭐⭐⭐⭐ Great 💰💰 Balanced Everyday coding — default & recommended
Haiku ⚡ Fastest ⭐⭐⭐ Good 💰 Cheapest Simple, repetitive tasks

Use Opus for thinking, Sonnet for building.

  1. Start with Opus to plan, architect, and write specs
  2. Switch to Sonnet to implement and generate code

This workflow gives you the best results at the lowest token cost. We’ll follow this pattern throughout the playlist.


🔀 /model — Switch Models

/model

A selection menu appears. Choose:

  • claude-sonnet-4-6 (default, recommended)
  • claude-opus-4
  • claude-haiku-4

📊 Usage & Stats Commands

/usage — Token Consumption

/usage

Shows your token usage in two dimensions:

Metric What it Tracks
Current session Tokens used in this conversation
Weekly total Cumulative tokens across all sessions this week

Example:

Session:  16% of limit
Weekly:    5% of limit

If you use Opus heavily, you’ll burn through tokens much faster. Switch to Sonnet for routine coding to extend your weekly budget.


/extra-usage — Buy More Tokens

/extra-usage

Hit your weekly limit? Instead of waiting for the reset, top up your account ($5 or $10) for additional token capacity.

This probably won’t be needed for our playlist, but large enterprise projects often require it.


/stats — Your Coding Stats

/stats

Displays cumulative statistics:

  • Total tokens consumed
  • Models used and their breakdown
  • Number of sessions started
  • Active days
  • Longest session
  • Current streak

This dashboard becomes more interesting the longer you use Claude Code. It gamifies your productivity.


/insights — AI-Generated Usage Report

/insights

Generates a detailed HTML report analyzing:

  • How you’ve been using Claude Code
  • Your prompting patterns
  • Good habits you’ve built
  • Areas for improvement
  • Specific recommendations

Run /insights after completing 10–15 sessions to get a meaningful report. Early on, there’s not enough data to make it useful.


⚙️ Configuration Commands

/config — Settings Panel

/config

Opens Claude Code’s settings. Key options include:

Setting Description
Thinking mode Show Claude’s internal reasoning
Verbose mode Display detailed tool call logs
Terminal progress bar Visual indicator for long tasks
Language Interface language preference
Auto-commit Automatically commit at milestones

🔐 /permissions — Tool Access Control

Claude Code is an AI agent backed by tools:

  • File reading
  • File writing
  • Bash execution
  • Web search
  • MCP tools (if configured)

By default, Claude asks permission before using each tool. This gets annoying fast.

/permissions

For each tool, you can set:

Permission Level Behavior
Allowed Claude uses it without asking
Always ask Claude asks every time (default)
Denied Claude never uses this tool

Example: Allow Web Search permanently so Claude can look things up without prompting you each time.

Permissions are stored in your project:

.claude/settings.local.json

Be thoughtful about what you permanently allow — especially Bash commands. A careless rm -rf allowed without confirmation could be destructive.

You can scope permissions to:

  • Local Project — applies only in this repository
  • Global — applies across all projects for your account
  • User — applies to your machine only

🎨 /theme — Appearance

/theme

Switch between:

  • dark (recommended)
  • light
  • system (follows your OS preference)

🎙️ /voice — Voice Input Mode

/voice

Enables Voice Mode. Instead of typing, hold Space and speak your prompt:

“Explain the project structure to me.”

Claude converts your speech to text and processes it normally. Useful when typing feels slow or your hands are busy.

Run /voice again to disable it.


📋 Quick Reference — All Commands at a Glance

Command Category What It Does
/exit Session End the current session
/resume Session Switch to a different session
/rename <name> Session Rename the current session
/export <file> Session Export conversation to Markdown
/btw <question> Context Ask a side question without polluting context
/model Models Switch between Opus / Sonnet / Haiku
/usage Usage Show token consumption
/extra-usage Usage Purchase additional token quota
/stats Analytics View coding statistics
/insights Analytics Generate an HTML usage report
/config Settings Open configuration panel
/permissions Security Manage tool access permissions
/theme UI Change terminal appearance
/voice Input Toggle voice input mode
/login Auth Start the login flow
/logout Auth Sign out of Claude Code

Discover more: Type / inside any Claude session and scroll through the full list using the arrow keys. Each command shows its description inline.


🔥 One-Line Summary

Slash Commands turn repeated, manual workflows into single keystrokes — keeping your context clean, your sessions organized, and your coding momentum unbroken.

Jatin Sharma
DevOps Engineer & Platform Architect. Passionate about cloud-native systems, automation, observability, and GitOps.