Given that everyone's been talking about Claude Code lately (even people who once dismissed AI coding tools entirely), your feed is likely flooded with posts about the ideal setup, hidden tips and tricks, and underrated features you're supposedly missing out on. We at XDA are no exception. We absolutely love Claude, and we're always writing about ways to get the most out of your subscription.

But at the end of the day, we can only tell you so much. We can dig into what the community is doing by heading to X, Reddit, and other forums (which we already do), but the real golden advice comes straight from the people who built it. Claude Code was created by Boris Cherny as a side project, and he's always sharing his setup and tips on X. Anthropic dropped Opus 4.7 a few weeks back, and with the launch, Cherny dropped a thread sharing his top tips for getting the most out of the new model.

He suggests enabling Auto Mode

Stop babysitting Claude with every click

Claude Code asks for your go-ahead by pausing mid-task and waiting for you to approve the action before moving on. Every file edit, every terminal command it wants to run, every dependency it wants to install, every time it wants to search the web, and so on. It's a smart safety measure, and it's there to ensure you're kept in the loop and nothing happens without your say-so. That said, it can seriously slow you down when you're locked in.

Anthropic was well aware of this, which is why it launched permission modes that control how often Claude pauses to ask. There are several to choose from: Default, Accept Edits, Plan, Don't Ask, and Bypass Permissions. Each mode pre-approves certain actions and commands. The newest addition to the lineup is Auto mode, and it's the one Cherny swears by. In his X post, Cherny shared that Opus 4.7 "loves doing complex, long-running tasks." He shared that it iterates until it hits a performance benchmark, and while that's great when you're building, it also means you need to manually give it the thumbs up dozens of times throughout the process.

You could use the Bypass Permissions mode here, but the word "dangerously" is in its official name for a reason. Auto mode is the sweet spot. It lets Claude run without constant permission prompts (which is the advantage you get with Bypass Permissions), but a model-based classifier decides whether each action is safe before it runs. Routine stuff like local file edits and dependency installs go through without a hitch, while anything genuinely risky gets flagged and blocked.

Cherny explains that this also means you can run even more Claude Code sessions in parallel (which was one of his tips in his earlier setup thread). Unfortunately, Auto mode is only available on the Max, Team, and Enterprise plans. So if you're on the Pro tier, you'll need to stick to one of the other permission modes.

Use /fewer-permission-prompts to clean up your allowlist

You've been rubber-stamping permissions for nothing

There are likely some permissions that you've been approving on autopilot without a second thought. However, despite you always approving them, Claude Code doesn't really know that and it'll keep asking you every single time. Well, Anthropic quietly launched a /fewer-permission-prompts skill that fixes this and it's the next tip Cherny had in his thread.

This skill scans your session history, and spots the safe commands you keep approving. It finds both bash and MCP commands. It then gives you a list of commands it recommends adding to your permissions allowlist. One run and you've cut out a bunch of prompts you never needed in the first place!

Use recaps to pick up where you left off

Helps you skip the wait, where was I moment

If you're anything like me, you keep multiple Claude Code sessions running at once and constantly switch between them. The problem is, by the time you come back to one, you've completely forgotten where you left off. Anthropic launched a recap feature that solves exactly this. When you return to a session after stepping away, Claude Code shows you a one-line recap of what happened in the session so far.

This recap only generates when at least three minutes have passed since the last completed turn and the terminal is "unfocused," which essentially means you've switched to another window or stepped away from the terminal completely. It won't interrupt you mid-flow. Instead, it quietly prepares in the background and is ready the moment you switch back. While I spotted the recap feature while using Claude Code a few days ago, Cherny's thread was how I realized that you can also run the /recap slash command at any time if you want a summary on demand.

Use /focus to see only what matters

Just show me the results

With how fast Claude Code is shipping nowadays, there are always some features that manage to slip through the cracks. Focus mode is one of them, and I didn't even realize it had dropped until I saw Cherny's thread!

This mode strips your session down to just three things: your prompt, a summary of the tools Claude used, and the final response.

All the intermediate stuff, like the files it read, the commands it ran, the step-by-step reasoning, etc, gets hidden. You just see what you asked and what you got back. You can toggle the mode on and off using the /focus command.

Configure your effort level for smarter thinking

Tell Claude how hard to think (yes, really)

If you've been using Opus 4.7 lately, you might've noticed something called Adaptive Thinking. This mode basically allows Claude to assess how deeply to think about each step based on how complex the task is. Simple stuff gets a quick response, whereas harder problems get more reasoning.

While Adaptive Thinking automatically handles the depth of reasoning, you thankfully still get to influence the overall intensity through effort levels. Opus 4.7 offers five: low, medium, high, xhigh, and max. Cherny says he uses xhigh for most tasks and only bumps it to max for the hardest ones. Xhigh is now the default in Claude Code, and you can change it anytime using the /effort slash command.

He also recommends giving Claude a way to check its own work

Let Claude see what it built

His final tip is something he says has "always been a way to 2–3x what you get out of Claude," and he explains that it's more important than ever with Opus 4.7. The tip is fairly simple: give Claude a way to verify its work. He explained that there are a few ways to do this, and it differs depending on the task. For instance, if you're doing some frontend work, you can use the Claude in Chrome extension to get Claude to control your browser. If it's an application, you can use the Computer Use feature to get Claude to verify its output.

This is something I've been doing a fair bit myself, and it's changed how I work. I vibe-coded a Chrome extension using Claude Code a few days back, and it just wouldn't work. The extension required some manipulation of Instagram's DOM, and no matter how many times I described the issue, Claude just couldn't get it right. The moment I gave it access to my browser through the Chrome extension though, it could actually see what was going wrong and fix it on its own. At the end of the day, there's only so much you can explain in a text prompt. Sometimes Claude just needs to see it for itself, and the Chromium extension, as I like to describe it, gives Claude eyes.

A few small changes go a longer way than you'd thinkIts

It's easy to scroll past tips like these and think they're far too basic to matter. But after spending a few days using Claude Code with the tips above in mind, I can confidently say they make a difference. And if the tips come straight from the guy who built the thing, they're probably worth trying.