For my first month in Claude Code I treated it like a misbehaving router: when a session got confused, I quit and relaunched. Fresh start, clean context, off we go. I called it starting over. It was really just throwing away everything Claude had learned in the last twenty minutes because I did not know there was a steering wheel.
The slash commands are that wheel. Not a reference to memorize, a set of controls that decide whether you are driving the tool or fighting it. Here is the handful I actually use, what each is for, and how to turn your own repeated prompt into a command you type in one word.
Slash commands are the steering wheel
A slash command is something you type starting with a slash, inside Claude Code, to control the session rather than ask Claude to do work. /clear resets the conversation. /model switches which model you are on. /rewind undoes a change. They are the difference between an editor who knows their keyboard shortcuts and one who reaches for the menu every time.
You do not need all of them. The full list runs long and most of it you will meet once and never think about again. What matters is the six or so you will reach for every day, and knowing which command fits the situation in front of you. Type a single slash at any point and Claude Code shows you the list in context, so this is a map, not a thing to cram.
The payoff is muscle memory. Once /clear and /compact are reflexes, you stop babysitting the session and start conducting it: trim here, undo there, switch engines, all without breaking your train of thought. That fluency is worth more than knowing every command, because the six you actually use are the ones that keep a session productive for an hour instead of ten minutes before it collapses into confusion.
The ones you'll actually type every day
Three commands manage the single most important resource in a session, the context. /clear wipes the conversation and starts fresh, which you want when you switch to an unrelated task and do not want the last one bleeding in. /compact keeps the thread but summarizes it down, which you want when a long session is getting heavy but you still need what came before. /rewind steps back to an earlier checkpoint when a change went wrong.
The instinct most people get wrong is reaching for /clear when they meant /compact. Clearing throws away the account context you spent twenty minutes building. Compacting keeps the gist and buys you room. Reserve /clear for a genuine change of subject; use /compact when the same job is just getting long.
- /clear: wipe the conversation and start clean. Use it when you switch to an unrelated task.
- /compact: keep the thread, shrink it to a summary. Use it when the current job is getting long.
- /rewind: step back to an earlier state when a change went wrong.
- /resume: pick up a previous session instead of starting from zero.
TipIf you find yourself quitting Claude Code to fix a confused session, that is the reboot reflex. /clear or /compact does the same thing without losing the setup.
Watch what's eating the window
/context shows you what is currently filling the context window, which sounds like a developer nicety until you notice it is also your bill and your quality. A window stuffed with three finished tasks is one that reasons worse and costs more on every single turn. Seeing what is actually in there is what tells you whether to /compact or /clear, instead of guessing.
This is the command that turns the reboot reflex into a decision. Instead of a vague feeling that the session has gotten heavy, you look: forty percent of the window is a research task you wrapped an hour ago and no longer need. Now the move is obvious, compact it out, and the session gets sharp and cheap again. You are managing a resource, not superstitiously restarting when things feel off.
TipGlance at /context when a session starts feeling sluggish or expensive. It usually shows one finished task hogging the window that a single /compact will clear out.
Context and how it behaves
The next tier sets up what Claude can see and how it acts. /init writes a starter CLAUDE.md for the folder, the standing brief it reads every session. /memory opens what it remembers across sessions so you can steer it. /model switches between Haiku, Sonnet, and Opus, which you will do more than you expect, cheap model for the grunt work, stronger one for the hard reasoning. /permissions controls what it is allowed to do on its own.
These are the commands that decide whether Claude Code behaves like a sharp new hire who read the onboarding doc or a temp who did not. /init and /memory are where you invest the context; /model and /permissions are where you tune the run to the task and your trust in it.
TipSwitch to a cheaper model with /model for mechanical work like reformatting a file, then back up for the analysis. Same session, right engine for each stretch.
Tools and reusable plays
The last group extends what Claude Code can reach and how you package your own work. /mcp manages the MCP servers that connect it to your tools, your CRM, your warehouse, an enrichment service. /agents lets you define and manage subagents, workers that take a piece of a big job in their own context. /help is the one that shows you everything else when you are stuck.
This is also where custom commands live, which is the part that pays off most for a team, so it gets its own section.
Worth knowing: these tool commands are also how you audit what Claude Code can reach. Running /mcp shows every connected server and /agents shows every subagent type defined. When you are reasoning about what the agent can touch, and for GTM work that means what can touch your CRM, these are the commands that give you the honest picture in one line.
Custom slash commands: your play as one word
The built-in commands are the same for everyone. The ones that change how your team works are the ones you write. A custom slash command is a Markdown file in your project's .claude/commands folder; the file name becomes the command, and the file's contents become the prompt Claude runs when you type it.
This is the same idea as a skill, aimed at a slightly different job. A skill is a procedure Claude loads on its own when the task matches. A custom command is a play you invoke on purpose, by name, when you want exactly that thing to happen. For GTM, that is your pre-call prep, your lead-scoring pass, your weekly pipeline hygiene, each turned from a paragraph you retype into a word you type.
The real power is arguments and files. A command can take input, so /prep-call Acme Corp passes the account straight through, and it can reference files in the repo, so the same command always reads your current positioning doc rather than a copy that quietly went stale in someone's prompt. Write it once against your live files and it stays correct as they change, which is the difference between a play that ages well and one that starts lying to you a month later.
TipPut custom commands in the repo, not just your machine. Then the play travels: your best pre-call prompt becomes the whole team's /prep-call, run the same way every time.
The commands worth writing first
If custom commands are where the real payoff is, the question becomes which ones to write. Start with the plays you run identically and often, because those are where the retyping tax is highest and the drift risk is worst. For most GTM teams that is a short, obvious list, and writing four or five of them changes the texture of the week more than any single built-in feature.
The pattern that works: pick a task you catch yourself pasting the same prompt for, turn that prompt into a command, commit it, and tell the team it exists. Do that a handful of times and you have encoded your best plays into words anyone can type. The commands below are the ones I would write on day one.
- /prep-call: build the pre-call brief from the account's site and last transcript.
- /score-leads: rank a list against the ICP with a one-line reason for each.
- /pipeline-hygiene: flag stale opps, missing owners, and deals with no next step.
- /follow-up: draft the post-call follow-up in your voice, under 120 words.
- /account-brief: a one-pager on any account, refreshed from live sources.
Where people go wrong
The most common mistake is the one I opened with: never learning /clear and /compact, so every confused session ends in a relaunch. The second is the opposite, clearing so aggressively that Claude keeps losing the thread and you keep re-pasting the same context. Both come from not feeling the difference between wiping and shrinking.
The third is hoarding custom commands nobody else has. A brilliant /prep-call that lives only on your laptop helps exactly one rep. The value shows up when it is in the repo and the whole team types the same word and gets your version of the play.
- Reaching for /clear when /compact was the move, and losing the account context you built.
- Over-clearing, so Claude keeps forgetting and you keep re-pasting.
- Custom commands stranded on one machine instead of shared in the repo.
- Memorizing the full list instead of learning the six you use and typing a slash for the rest.
A GTM day, in commands
Here is what a real morning looks like once the commands are muscle memory. You open Claude Code, type /prep-call to build the brief for your ten o'clock, then /clear when it is done because the next job is unrelated. You run /score-leads on the overnight inbound, drop to a cheaper model with /model for the mechanical pass, and bump back up for the judgment calls. A change goes sideways, so /rewind, and you are back to good in one command.
None of that is exotic. It is the difference between a tool you wrestle and one you drive, and the whole gap is a handful of commands plus two or three you wrote yourself.
How to set it up
Create the commands folder
Custom commands live in a .claude/commands folder in your project. Make it once, and every Markdown file you drop in becomes a command.
Write the play inside the file
The file's contents are the prompt Claude runs when you type the command. Write your best pre-call prep as that prompt, with your standards baked in:
"Prep me for an upcoming call. Read the account's website and our last transcript with them, then give me: their likely priorities, two risks in the deal, three questions to ask, and one specific thing to reference. No hype words, keep it to one screen."
TipReference files and arguments in the command so it works for any account, not just one you hard-coded.
Invoke it by name
Type the file name as a slash command and Claude runs the play. The paragraph you used to retype every morning is now one word:
Share it in the repo
Commit the .claude/commands folder so the command travels with the project. Now the whole team types /prep-call and gets your version of the prep, not fifteen different from-scratch prompts.
Frequently asked questions
How do I see every available slash command?
Type a single slash inside Claude Code and it lists them in context, with a short description of each. That is faster and more current than any reference, including this one.
What is the difference between /clear and /compact?
/clear wipes the conversation and starts fresh. /compact keeps the thread but summarizes it down to save room. Clear when you change subjects; compact when the same job is getting long.
Do slash commands work in every Claude Code surface?
The core ones work across the CLI, IDE, and web. A few are surface-specific. When in doubt, type a slash and see what is offered where you are.
How do I write my own slash command?
Put a Markdown file in your project's .claude/commands folder. The file name becomes the command and the contents become the prompt. Commit it to share it with your team.
Is a custom command the same as a skill?
They overlap. A skill is a procedure Claude loads on its own when the task matches. A custom command is a play you invoke on purpose by name. Use a command when you want exactly that thing to run when you ask for it.
Can a command take arguments, like an account name?
Yes. You reference the input inside the command file, so /prep-call Acme Corp passes the account through to the prompt. That is what makes one command work for every account.
Will /clear delete my files?
No. It clears the conversation, not your files. Your work on disk, and any commits, are untouched. It only resets what Claude is currently holding in the chat.
What should a GTM team turn into commands first?
The plays you run identically every week: pre-call prep, lead scoring, pipeline hygiene, a weekly account summary. Anything you catch yourself pasting the same prompt for is a command waiting to be written.
Sources & further reading
Claude ships fast. This page was last reviewed Aug 22, 2026; verify time-sensitive details against the official docs above before relying on them.