In the first week you began working with your agent. This video introduces a few Claude Code commands and keyboard shortcuts that make that work easier. There are many more options than we can show here. Try the ones you find useful, and explore the menus as you go. We use the command line version of Claude Code throughout the course. Some of these controls also work in the web version and the IDE plugins, but the interfaces differ. We recommend trying all of them, and then settling on whichever tool you prefer.
Let's start with three keys you met in the first week. Escape interrupts the current response or tool call. The work already done remains, so you can give a correction without waiting for the whole task to finish.
Shift and Tab together cycle through the permission modes, which decide how much the agent is allowed to do before it stops and asks you.
Press Escape twice on an empty prompt to open the rewind menu. It lets you return to an earlier point in the conversation, restore captured file edits, or do both. The prompt must be empty. If you have typed a draft, double Escape clears it instead. Rewind does not reverse every effect of a shell command, so check what you are restoring.
These input shortcuts help you give the agent the material it needs. The first is the at sign, followed by a path. It points the agent directly at a specific file or folder in your project. For example, the prompt here asks the agent to explain the auth file at the path shown. Naming the file removes the need to describe where to find it.
The exclamation mark, followed by a command, lets you run a shell command yourself, without leaving the session. That's useful for a quick check while staying in the same place as the agent. The agent also gets to see the command and its result.
You can also paste or drag in an image, such as a screenshot or a design mock-up. This gives the model the picture you want it to examine.
When something breaks, paste the error text itself. It may include a file name, a line number, and a stack trace, which give the agent much more to work with than a description of the problem alone.
These shortcuts all help with the same task: giving the agent the right context. Show it the file, image, or error you mean. That gives it the details directly and saves you describing them from memory.
A few everyday keys also make editing a prompt easier. The up and down arrows scroll through your previous prompts, so you can reuse one, or edit it slightly, instead of typing it again from scratch.
To add a line without sending the prompt, type a backslash and press Enter. Shift and Enter also works in supported terminals, though some need setup first.
You also don't have to wait for the agent to finish before you type. If you type while it's still working, your messages simply queue up, and the agent takes them in turn.
Control and S together stash whatever you are in the middle of typing, and clear the prompt. Pressing it again on an empty prompt brings your text back, exactly as you left it. That is handy when you are halfway through a long instruction and you want to ask something small first.
Control-C cancels whatever you're currently typing, and Control-D quits the session.
You don't need to memorize these keys. Press the question mark on an empty prompt to see the shortcuts for your setup.
The command menu gives you another way to find a control when you need it. When you type a forward slash at the prompt, a menu of commands appears above it, ready for you to choose from.
As you keep typing, the list narrows down to the commands that match, and you pick the one you want with the arrow keys. On screen you can see what is left after typing just slash, C, O.
You can look through the matching commands instead of remembering an exact name. The built-in slash commands control Claude Code itself. The menu can also contain commands that send instructions to the model. Other agent tools have similar menus with their own commands.
The slash help command opens the help panel shown here.
One of its tabs lists every command that is available to you, each with a one-line description of what it does.
The panel on screen also lists the keys, including several of the ones we just went through, and a few we did not. When you can't recall a command or shortcut, start with the tool's own help. It shows what is available in the version you are using.
Let's look at three commands that are worth knowing early. The slash model command switches the model for the session. You can try a more capable model for a difficult task, or a faster one for a simpler job.
The slash config command opens the settings, where you can change things like the colour theme and the editor mode. It's worth opening once, just to see what is adjustable.
The slash B-T-W command asks a side question about the current work without adding the exchange to the main conversation. So you can ask what a file was called, or what some flag does, without pushing the agent off the task it is in the middle of.
Some commands are missing here on purpose, because they make much more sense once we have explained the ideas behind them. The commands for memory and for the context window come later this week, in the video on context. Permissions come back in Week 3. Writing commands of your own comes later still.
Reasoning models have a setting for how much effort they spend. You don't need to ask for step-by-step reasoning before every task. The setup shown here uses high effort. The starting level depends on the model and your settings.
The slash effort command shows the available levels and lets you choose one. High is selected in this screenshot, with lower levels to the left and higher ones to the right. The last option is "ultracode". It is extra-high effort plus workflows. A workflow lets Claude Code split one large task across many agents.
The word ultrathink asks for deeper reasoning on one turn. Claude Code recognizes it and adds an instruction for that request. It does not change the session's effort setting.
Start with the current setting. If a difficult problem needs more work, try a higher effort level and compare the result. More effort can mean a longer wait.
You now have a few ways to give the agent useful context and control the session. Open the menu and try them on your own work. You can look up the rest when you need them.
Later this week we look at what the agent keeps in view while it works, and what it remembers between sessions.