The Claude Code Cheat Sheet: Commands + Shortcuts

If you learn nothing else, learn these. They cover the moments where beginners lose the most time - running out of context, wanting to undo a step, giving Claude a file, and getting stuck in a mode they did not mean to enter.

• /clear - wipe the conversation and start fresh. Use it between unrelated tasks so old context stops confusing the model (and stops costing you tokens).
• Esc - interrupt Claude mid-response the instant it goes the wrong way. You do not have to wait for it to finish; stop it, correct course, continue.
• Esc Esc (press twice) - rewind. Jump back to an earlier point in the conversation to edit a previous message and take a different path.
• @ - reference a file by path. Type @ and start typing the filename; Claude reads that exact file instead of guessing.
• Shift+Tab - cycle permission modes (normal, auto-accept edits, plan mode). This is the single most misunderstood shortcut - see the callout below.

Slash commands control the session itself - context, model, memory, login. This is the core set; your exact list can vary by version, so run /help any time to see everything available in your install.

• /help - list every command available in your version. When in doubt, start here.
• /clear - clear the current conversation and its context.
• /compact - summarize the conversation so far to free up context without losing the thread. Use it when a long session starts to feel sluggish.
• /init - scan the project and create a CLAUDE.md file so Claude remembers how your codebase works.
• /memory - open and edit what Claude remembers about you and the project.
• /model - switch models (for example, a faster model for simple edits, a stronger one for hard problems).
• /review - have Claude review recent changes or a pull request.
• /cost - see what the current session has cost so far.
• /config - open settings.
• /login and /logout - manage your account session.
• /resume - pick up a previous conversation where you left off.

• Esc - interrupt the current response.
• Esc Esc - rewind to edit an earlier message and branch from there.
• Shift+Tab - cycle permission modes (normal / auto-accept / plan).
• Up arrow - scroll back through your command history.
• Ctrl+C - cancel the current input, or press again to exit.
• Ctrl+L - clear the screen without clearing the conversation.
• Enter - send. For a multi-line message, use the line-continuation your terminal supports (often the paste of a block, or a trailing backslash) rather than sending each line.

These three characters, typed inside a normal message, are where Claude Code gets fast. They tell Claude exactly what to look at, what to remember, and when to run a shell command.

• @ (at) - reference a specific file or folder. Type @ then the path; Claude reads exactly that instead of searching. Example: 'Refactor @src/auth/login.ts to use the new session helper.'
• # (hash) - add a note to memory on the fly. Start a line with # and the fact gets saved to your CLAUDE.md so Claude remembers it next session. Example: '# We deploy only from the main branch.'
• ! (bang) - run a shell command directly and drop its output into the conversation. Example: '!npm test' runs your tests and lets Claude see the result.

1. 1Open your project and run /init so Claude writes a CLAUDE.md and learns the codebase.
2. 2Add a couple of standing facts with # - your deploy rule, your preferred test command, anything you would otherwise re-explain.
3. 3Start a task and point Claude at the right files with @ instead of describing them.
4. 4When it drifts, hit Esc, correct, and continue - do not let a wrong approach run.
5. 5Between unrelated tasks, run /clear so stale context does not leak into the next job.
6. 6When a session gets long, run /compact to keep it sharp, and check /cost if you want to watch spend.