Original video in Chinese.
Key Takeaway
- Claude Code is considered the most powerful AI coding tool on earth because it is deeply adapted to and optimized for the Claude model, and it does not care about context length consumption.
- The
/initcommand is used to initialize a project and create aCLAUDE.mdfile to record project background information, improving efficiency and context consistency. - Plan Mode allows users to have Claude Code make a plan before executing complex features, and it is especially useful when you are not sure how to proceed.
- The
/idecommand supports integration between Claude Code and IDEs such as VS Code and Cursor, providing a better code editing and prompting experience. - Users can customize commands, turning common operations such as explaining code into shortcut commands to improve work efficiency.
- The
/costcommand is used to view API consumption, and you can switch models with the/modelcommand to save costs. - The author emphasizes that when model capability is the same, AI with more tools is more productive, and encourages users to actively use and explore MCP.
Full Content
The best AI coding tool on earth is definitely Claude Code. And it will only keep getting better from here.
Why is it the strongest? I posted this in the community:
If you think of the Claude model as an engine, then although every company can buy that engine and assemble it into a race car, only Anthropic can unleash its ultimate power.
Most likely, when Anthropic trained Claude 4, it had already used the more than ten tools built into Claude Code on it and carried out targeted reinforcement training.
In other words, this engine was already adapted and optimized with the chassis and other systems during the development stage. They are the best combination. How could other vendors possibly keep up?
And for the sake of performance, Anthropic is being especially generous now and can afford not to care about context length consumption. By comparison, Cursor and the others have to count every penny, which sometimes leads to a very poor user experience.
Claude Code is so powerful, but there isn’t much introduction to it in China. The reason is simple: Chinese self-media is too weak. They only know how to shout “shocking” and “pulling out a big move.” When it comes to this kind of complex tool, and especially one that is command-line based and also requires configuring a clean residential IP, they just don’t know where to start.
No problem. I’ll be making a series of videos to go deep into this tool. Today, let’s start with five very practical tips to help everyone use Claude Code well.
/init
When you start a new project, or when you bring Claude Code into a project midway through, you must first run the /init command, meaning initialize.
The main purpose of this command is to create a CLAUDE.md file. This file will contain all the background information about the project, including the project’s core overview and goals; important coding conventions and style guidelines; a list of key files and utility functions, and so on.
With this file, every time you start Claude Code, it will load automatically, so it understands the whole project and you don’t need to repeat yourself. Also, whenever your project has any progress in development, you can have Claude write that progress into this document.
So this simple command, and the CLAUDE.md file it generates, are very important for improving efficiency and maintaining context consistency. Remember to use them.
Plan Mode
Most of the time, we let Claude Code run on “autopilot.” But when there is some fairly complex functionality we want to implement, or when we ourselves are not sure exactly how to do it, we can switch to Plan Mode and let Claude Code help us plan first, and then execute.
Switching to Plan Mode is very simple: just press the shortcut Shift plus Tab. For example, I want to improve the system’s ability to handle high concurrency and server disconnects. I’ve thought of some solutions myself, such as setting up a backup pool of API keys. I tell Claude Code all of this and let it help me make a complete plan.
After receiving the request, it will go through all the code related to the requirement and then give a very detailed plan. If it looks good, you can let it start executing according to that plan.
Honestly, this mode is pretty useful. When you’re not sure, remember to let Claude help you plan.
/ide
Although Claude Code mainly interacts through the command-line interface, or CLI, running in the terminal, it also supports integration with VS Code and Cursor, allowing users to get a better experience in the IDE environment.
For example, after integration, you can see code changes in files, just like you can in Cursor. Also, when you select a few lines of code, Claude Code will show prompts on its side as well.
To do this, you need to do two things, and they’re very simple:
First, install the Claude Code plugin. Just search for it, then click install.
Second, run the /ide command and select the corresponding IDE, such as Cursor in my case. Then press Enter and you’re done.
Custom Command
Claude Code has many built-in commands you can use. Beyond that, you can actually customize commands according to your own needs.
For example, I often ask Claude to explain code for me. But typing out the prompt every time is pretty annoying. At this point, I can turn this need into a custom command. It’s very simple and only takes two steps.
The first step is to run this command, which creates a commands folder inside the claude folder.
The second step is to go into that folder and create a Markdown file. The file name is the name of the command you want, for example I set it to explain to me here. The content of the file is the command itself, meaning the prompt.
After saving, exit Claude Code, reopen it, type a slash, and you’ll be able to see the command you just set.
Using this method, we can even get Claude Code to complete some complex tasks. It has a lot of development potential, and everyone can explore it on their own.
/cost
If you are using the API access method, remember to use the /cost command to check how much has been consumed.
This command doesn’t work for subscribers. But for API users, it is very important, because Claude Code is too expensive. If you’re not careful, several thousand yuan can disappear.
If you want to save some money, you can also use the /model command to set the model. For example, for some small changes, there is no need to use Opus; Sonnet is enough to get the job done.
If you’re just starting to use Claude Code, the five tips I introduced earlier will definitely help you. This AI coding tool has many more advanced uses, and I’ll introduce them one by one later.
OK, that’s it for this episode. If you want to discuss AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!