In a video game, you do not type a whole sentence to open your inventory: you tap a single key. Slash commands are the same idea for Claude Code — short, memorised keystrokes that reach for the control you need without breaking flow. You never memorise all of them; you learn the four or five that pay for themselves every session, and rely on /help for the rest. The point of the toolkit is not encyclopedic knowledge, it is muscle memory on the commands that come back every day.
A slash command starts with / and behaves like a small structured instruction rather than a chat message: it is not sent to the model as free-form text, it directly changes the session state (context, model, permissions, config…). The diagram sketches the five commands that carry most sessions.
The everyday toolkit, in practical terms:
/help — lists the available commands. Always the right answer to "what could I run here?"./clear — wipes the current conversation. Use it between unrelated tasks, so your bug fix does not inherit the context of the previous refactor./compact — condenses a long thread into a summary. Best played before the session feels sluggish, not after./model claude-sonnet-5 — switches the underlying model mid-session. Reach for a faster model on trivial edits, a stronger one when reasoning gets hard./config — opens the settings panel (theme, defaults, editor integration)./permissions — reviews or updates what Claude Code is allowed to do without asking (deep-dived in Module 2)./doctor — checks that your setup is healthy when something feels broken (missing tools, weird auth state)./fast — toggles a faster response mode where the model supports it.In practice, you do not need to memorise the full list on day one. Learn the mental trigger: "typing / gives me the menu, and /help is always there". Everything else you pick up when you meet the situation.
/ on its own to reveal the menu, then run /help and skim the list for 60 seconds./clear and confirm the previous context is gone: ask "what were we talking about?" — Claude should not remember./compact and note how the transcript is summarised into a shorter head./model and switch to a different available model for one message; ask the same question twice and observe the difference in tone or depth./config once just to see where preferences live — you do not have to change anything, you only need to know the door exists./help, /clear, /compact, /model./help is your safety net: it removes the pressure of memorising the full list.Never running /clear between unrelated tasks. When the same session is used all afternoon across three different problems, Claude drags stale context into every answer — you get subtly worse suggestions and blame the model. /clear between tasks is the single cheapest quality boost in a Claude Code session.
So what? You do not need to memorise everything: type /, glance at /help, and grow your toolkit one command at a time as real situations reveal which ones you actually need.