Transcription
For the past few months, Claude Code has been shipping features non-stop, but most people don't have an idea that they even exist. Either they're buried, undocumented, or that there's so many updates that it's nearly impossible to keep up. Even if you're scrolling through the command menu, you're only seeing the surface cuz a lot of the powerful capabilities aren't as obvious as you think they are. What's wild is that many of these issues people are running into, like the context memory issue of cloud code, already has been fixed with these built-in features, like with /stream. They're just hidden away in the config files, feature flags, and environment variables that nobody really talks about, especially after the recent cloud code leak that went deep and put together a condensed list of the most important feature flags that you should enable right now. These unlock a completely different level of performance and can seriously upgrade your workflow.
Let's first start off with a feature that will potentially fix all of your context memory issues within cloud code. A feature called /dream, also known as autodream. Think of it like REM sleep for your AI agent. It was found within their system prompt on GitHub. And it's a feature that automatically consolidates Claude's memory between sessions. It prones stale or relevant notes versus useful insights and essentially cleans up the memory so it stays accurate and useful over time. So instead of memory getting worse the more you use claude. It actually improves.
To use this new feature, just simply open up your Claude instance and then type in /memory. Once you type this in, you should see this new feature called autodream. Now, obviously, you're going to need to make sure you're on the latest update. But to toggle it on, just simply click enter and you're going to be able to now have it so that it is enabled and that you can use it directly behind your sessions.
Now, this isn't an official command that you can basically use to invoke it. If you are to go back into your session, if you type in /stream, it's not going to work. It will showcase unknown skill. And this is because it's not an official command yet. But if you are to do something like consolidate my memory using dream, it is going to then use and invoke that dream feature so that it could potentially fully clean up the full process of your memory system. It is going to be able to understand and run the full cleanup process using this new autodream feature. And underneath the prompt bar, you'll see that it is stating dreaming, meaning that it has been turned on. And another way to invoke it is just by simply typing in dream in your prompt autodream or just simply do what I did, consolidate my memory, which will trigger it manually by just simply using it without the /dream command, which should be released soon.
If you want the best AI tools, workflows, and drops before everyone else, join my free newsletter with the link in the description below, which is completely free.
One of the biggest problems of cloud code has been the constant flickering. You have a jumpy UI in the terminal and a lot of the components actually get glitched up. But now there is a new flicker mode which is something that you can essentially turn on within your instance today. This is going to directly fix the glitchiness as well as the jump UI by introducing a new experimental renderer that visualizes the viewport instead of relying on limited ANC rendering. You can enable it by simply going into your cloud files and setting the cloud code no flicker equaling to one. Once it's on, everything feels smoother. There's no more flickering, stable performance as chat grows, and even a mouse support. So, you can even click around the terminal now. There's also a trade-off though, cuz there's no more command F and different copy paste behaviors. But overall, it's a huge quality of life improvement.
Next is the new powerup command. One of the biggest problems with cloud code is that there are a lot of features that aren't intuitive or easy to discover. Hence why we made this video. And this is a command that actually helps solve that. When you use the / powerup command, you get interactive lessons built inside the terminal that teach you how to use cloud code as you go. It's a really solid step towards making the tool more accessible, especially with hidden features that are present within your Cloud Code instance. It's still early, but it has a lot of potential for learning directly inside the CLI.
Odd code just leveled up with this external plugin. The open AI team had released a codeex plug-in giving you a full second agent without leaving your workflow, solving the problem of wanting a fresh set of eyes on tricky code where you can now have codecs integrated inside Cloud Code. And this kind of solves a lot of things because you all know that Claude Code has a pretty bad problem with rate limits now, especially since they have cut the rate limits for their pro plans. And now what you can do is essentially use the codeex tier directly within Claude Code. So that you can use all of Claude Code's features with the new rate limits associated with Codeex. This means that once you have installed the plugin, you essentially can use the codeex reviewer tool as a normal readonly check. You can also make codec question assumptions, have it hand off tasks to codeex for a second pass. It's super lightweight, uses your local codec setup, and keeps everything inside cloud code. Perfect for refactors, migrations, odd changes, or any highstake codes where a second opinion can save you headaches.
One of the biggest game changers in cloud code right now is the computer use through the MCP. It's a new feature that just launched last week and it kind of solves a lot of problems. Normally testing and interacting with apps is quite a hassle where you have to leave the CLI. But now with the new feature enabled through the MCP code can open your apps, click through UIs, find bugs, fix them, and even verifies the fix all from a single prompt. It's currently in research preview from Pro and Max plans, but only available on Mac OS. It works with anything you can open on the Mac with Swift UI apps, local electron builds, or GUI tools without a CLI. What you can do is simply enable it using the /mcp command. And this is where you can then rightclick on the computer use MCP to enable it to then interact with cloud code using computer use. Don't worry, I have a substitute for users on Windows as well as Linux. This is a plugin that is not from the official Enthropic uh provider, but this is still something that does exactly the same thing as what Computer Use does natively on the Cloud Code instance. And I actually made a video showcasing what you can do with it. So, I'll leave that as well in the description showcasing how you can set it up.
Next up is the teleport command where you can move cloud sessions between devices. Just simply run the /t teleport command to continue a cloud session on another machine. This works great if you started a session on the mobile or web and you want to pick it up in your terminal or desktop. You also have the remote control feature which recently dropped and it is something that lets you control a locally running session from another device simply from your phone, tablet or even browser. You can simply use the run/ remote control command to connect and it essentially keeps everything local on your machine while letting other devices act as a window into that session. Personally, you can enable the remote control for all session command in your config file so that it is going to be ran locally in any session and it is instantly accessible from any device. So teleport equals cloud session hopping and remote control equals local session control from anywhere.
Next is the output style feature and this is something that changes how cloud code interacts with you. Output styles is something that you can run within your instance using the /output-style command and it essentially modifies the system prompt to adapt to the agents behavior. You have three different types. You have the default output style which is for efficient software engineering. the explanatory which adds educational insights while coding and then the learning one which is Claude guiding you to write key parts yourself. If you're learning a new codebase, you can try learning mode. You'll actually understand what Claude is building instead of just accepting diffs.
Now many of us know that cloud code supports hooks but letting it run logic at specific points in your agents life cycle is going to be a gamecher because it solves the problem of repetitive tasks or manual monitoring by automating them. For example, Boris had stated that you can use hooks to essentially have it work dynamically to load context each time clot starts where you can set a hook for session start. You can also set a hook for pre-tool use where you can log every bash command. the model uses. You also have the ability to set a permission request where it roots permissions and different prompts to WhatsApp for approval or denial. Another one is to use the stop hook that pokes Claude to keep going whenever it stops. Hooks is going to essentially give you the deterministic control over Claude's behavior and workflow. And they have a full documentation showcasing how you can set this up.
Next is the ability to fork your sessions within cloud code, which is perfect if you want to experiment without affecting your main workflow. You can do this by simply opening up your session and running the /branch command or from the CLI, you can go into cloud code and type in the /resume feature with the session ID and then specify the fork session. This is going to essentially let you create a separate copy of your sessions so that you can try new ideas safely.
If you didn't know, Cloud Code has a deep support for Git work trees, which is perfect for running multiple parallel sessions in the same repository, which is going to solve the problem of juggling dozens of workflows at once. If you are to start a session in a worksheet with the cloud workree command, you can also check the workree box within cloud desktop. And this is essentially not going to use git. It's going to use the get work tree create hook that lets you add custom logic for creating work trees in other VC systems. This makes managing multiple sessions and branches much smoother.
This is something that a lot of people don't know about cuz my buddy was just asking me about this how you can change the effort level of various sorts of sessions. And that is something that you can do cuz within claude code you can set the effort level in skills or slash commands solving a problem or wanting more control over the model's reasoning time. This is where it is going to let you now determine how long Claude thinks before answering. So you can override your session default, letting you fine-tune responses for speed or depth on the fly.
Next is if your skill relies on dynamic content, Claude Code lets you embed shell outputs directly into your prompts using this specialized command. When the skill runs, Claude executes the command, replaces the placeholder inline, and the model only sees the results, not the command itself. This is going to make your skills dynamic and contextaware without exposing internal shell details.
If you like this video and would love to support the channel, you can consider donating to my channel through the super thanks option below. Or you can consider joining our private Discord where you can access multiple subscriptions to different AI tools for free on a monthly basis, plus daily AI news and exclusive content, plus a lot more.
Now, as you saw from the video, Claude Code has came a long way. It's packed with so many features, and honestly, this unlocks a lot of serious productivity and control within your instance. I'll leave all the links that I use in today's video in the description below. But with that thought, guys, thank you guys so much for watching. Make sure you go ahead and take a look at the second channel, join the newsletter, join the Discord, follow me on Twitter, and lastly, make sure you guys subscribe, turn on notification bell, like this video, and please take a look at our previous videos so that you can stay up to date with the latest AI news. But with that thought, guys, have an amazing day. Spread positivity, and I'll see you guys really shortly.