Introduction

/rc (Remote Control) — Develop thoroughly from your smartphone

This was the most shocking discovery.

While running a Claude Code session on your PC, you can operate the same session from the Claude app on your smartphone. All the code is executed on your local PC, and your smartphone just acts as a "window."

/rc

With just this command, a QR code is displayed, and once you scan it with your phone, the connection is complete.

Why it's great

It's not just for monitoring. You can issue instructions, have it write code, and perform reviews normally from your smartphone. I can do heavy development on my phone while drinking coffee on the sofa. I don't need to be glued to my PC. I can check the progress, stop it if it's heading in a weird direction, and issue the next set of instructions right away.

Points to note

• The session disconnects if the network is down for about 10 minutes.
• It ends when the terminal is closed.
• There is a bug where the remote connection does not recover after hitting a rate limit (as of March 2026).

It is still a research preview, so it has some rough edges, but the direction is perfect.

/init — Crucial, but not guided

I want to shout this from the rooftops: It's important. It's incredibly important. And yet, it's not prominently guided anywhere. I was caught off guard.

This is an initialization command to help Claude Code understand the project structure when starting a new project. Whether you do this or not makes a huge difference in Claude's level of understanding.

/init

Whenever you start a new session, run /init first. I want you to make this a habit.

I used it for a while without knowing this. I was getting frustrated every time Claude made an off-target proposal, but it was just because I hadn't run /init. I wish I had done it from the start.

CLAUDE.md — Another existence you notice through /init

When you run /init, a file named CLAUDE.md is generated. I suspect many people realize its existence for the first time here, just as I did.

This is an instruction sheet that Claude Code reads every time you start a session if you place it in the project root.

In my case, I write my development policies and code style here. Things like "Apply code changes only after diff suggestion and approval" or "Write comments in Japanese."

Without this, you end up repeating the same instructions every time. I believe the /init -> CLAUDE.md flow is the very first step in using Claude Code effectively.

Plan (Plan Mode) — Don't fix code when I just asked a question

Claude is excellent. He is so excellent, in fact, that he sometimes starts fixing code when I only asked a question.

I just asked, "What's wrong with this error?", and when the response came back, the code had already been rewritten. Being perceptive and human-like isn't necessarily bad, but sometimes I just want to ask.

In such cases, it's better to switch to Plan mode. You can simply toggle the mode from the Claude Code panel in VS Code. Claude will then propose each action, and it will only execute them after you approve.

There seems to be a command (/plan) as well, but I don't use it.

Effort (Depth of thought) — Always set to maximum if you have the capacity

"Effort" is a setting that adjusts the depth of Claude's thinking. You can switch it from the Claude Code panel in VS Code.

Simply put, the higher it is, the better the answers you get. It supposedly increases usage, but if you look at your 5-hour or 1-week limit and have room to spare, I think you should just set it to maximum.

By the way, on the MAX plan, the highest setting is upgraded from "high" to "MAX" lol.

There seems to be a command (/effort) as well, but I don't use it.

/compact — Savior when context grows bloated

When working for a long time, the accumulated conversation causes the context to bloat. If you feel it's getting heavy:

/compact

You can also specify what to keep.

/compact Keep the error handling patterns

As a rule of thumb, run this when context usage exceeds 80%. If you are switching tasks, /clear is better.

/fork and /rewind — Experimentation and rewinding

/fork copies the current conversation into a branch. It allows you to say, "Try experimenting with this policy over here" without dirtying the main line.

/rewind rolls back the conversation along with the code changes. "That's wrong, undo it" becomes a one-command task.

/fork     # When you want to experiment
/rewind   # When it messed something up

Model (Model switching) — Opus is enough if you have MAX

In VS Code, you can switch models by clicking on the part where the model name is displayed in the Claude Code panel. You can also do it with the command /model.

I often see articles about dropping down to Sonnet or Haiku to save costs, but if you are on the MAX plan, you can run everything on Opus without any issues. In normal usage, I almost never hit the limits with Opus. Since you are paying for MAX, there is no need to intentionally lower the performance.

Features I haven't mastered yet

• Skills — A mechanism to define custom skills and have them called automatically. Still experimenting.
• Claude Code Channels — You can operate sessions from Telegram or Discord. I'm interested in it as an alternative to /rc.
• /simplify — Apparently, it lets 3 agents review the code before a PR. This is what I will try next.
• /batch — Process large-scale changes in parallel and automatically create a PR. I want to use this once the scale gets larger.

Summary

If you are just using Claude Code normally, you haven't even touched half of its features. The official documentation is in English, new features are buried in a single line of patch notes, and it's natural that you wouldn't notice them.

For those like me who "pay for MAX and use it heavily but actually don't know many of the features," I will continue to write based on my real experiences.

I have written about my experience actually building apps with Claude Code in the articles OLTranslator and LiveTR. For the story about how Claude Code provided value beyond "writing code," see "Building from Claude and research papers".