Anthropic Just Dropped 17 New Claude Code Features

16:46

Anthropic Just Dropped 17 New Claude Code Features

Ray Amjad 21.02.2026 15 492 просмотров 433 лайков

Machine-readable: Markdown · JSON API · Site index

Смотреть на YouTube

Поделиться Telegram VK Бот

Транскрипт Скачать .md

Анализ с AI

Описание видео

Stay ahead with my Claude Code Newsletter 👉 https://www.masterclaudecode.com/newsletter?utm_source=youtube&utm_campaign=JTW_sEXLH_o Learn the AI I'm learning with my newsletter 👉 https://newsletter.rayamjad.com/ To avoid bias, I've never accepted a sponsor; my videos are made possible by my own products... —— MY CLASSES —— 🚀 Claude Code Masterclass: https://www.masterclaudecode.com/?utm_source=youtube&utm_campaign=F_frSYyhzOE —— MY APPS —— 🎙️ HyperWhisper, write 5x faster with your voice: https://www.hyperwhisper.com/?utm_source=youtube&utm_campaign=F_frSYyhzOE - Use coupon code YTSAVE for 20% off 📲 Tensor AI: Never Miss the AI News - on iOS: https://apps.apple.com/us/app/ai-news-tensor-ai/id6746403746 - on Android: https://play.google.com/store/apps/details?id=app.tensorai.tensorai - 100% FREE 📹 VidTempla, Manage YouTube Descriptions at Scale: http://vidtempla.com/?utm_source=youtube&utm_campaign=F_frSYyhzOE 💬 AgentStack, AI agents for customer support and sales: https://www.agentstack.build/?utm_source=youtube&utm_campaign=F_frSYyhzOE - Request private beta by emailing r@rayamjad.com ————— CONNECT WITH ME 🐦 X: https://x.com/@theramjad 👥 LinkedIn: https://www.linkedin.com/in/rayamjad/ 📸 Instagram: https://www.instagram.com/theramjad/ 🌍 My website/blog: https://www.rayamjad.com/ ————— Links: - https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md - https://claude.com/solutions/claude-code-security - https://x.com/jarredsumner/status/2022537655339397133 Timestamps: 00:00 - Intro 00:14 - Worktree 00:59 - Subagent Worktrees 01:59 - Resuming Worktrees 02:09 - Exiting Worktrees 02:21 - Tmux Worktrees 03:41 - Advantages of Worktrees 04:57 - from-pr 05:18 - Advantages Continued... 05:40 - Worktree Hooks 06:14 - Security Capabilities 07:03 - Previews for Desktop 08:57 - CI Monitoring 09:18 - /desktop 09:32 - Diffs on Desktop 10:01 - SSH from Desktop 10:52 - Plug-ins for Desktop 11:12 - Share Link 11:19 - /debug 11:42 - /rewind 11:56 - /rewind summarisation 12:13 - AskUserQuestionTool Markdown 13:05 - Memory Improvements 13:13 - Skills and Commands in Claude Code Web 13:35 - Multi-Repo Support 14:08 - Plan Mode in Slack 14:22 - Background Agent Frontmatter 14:43 - Subagent Level Memory 15:58 - Conclusion

Оглавление (29 сегментов)

Intro

Okay, so we had a lot of Claude Code updates to go through from the last 2 weeks. But before getting started, if you want to sign up to my brand new free Claude Code newsletter, then that will be linked down below. You will get access to a bunch of free videos as well about how I'm currently using Claude Code and some of my own workflows. Okay, now first of all, one of the biggest recent

Worktree

changes is they now made it easier to deal with git worktrees in Claude Code. So first of all, you want to make sure you're on the latest version of Claude Code by using claude update. And then you can run claude worktree. Or just run claude -w and call it whatever you want. So I'm going to call it like assistant mode for my application HyperWhisper. And this is my own speech-to-text application. Anyway, as you can see, now I made a brand new worktree and I can have it implement one of my specs in an isolated branch. And then I can do the exact same thing on the right-hand side with another spec that I have. And now I can have two features being implemented at the exact same time without these two sessions of Claude Code interfering with one another. And then I can open up a pull request for this later on. Now you can take this further and have any subagents in Claude Code being able to work in an isolated work tree as well. So a good use case here is that

Subagent Worktrees

for example, with my Claude Code Masterclass, which will be linked down below, it is the most comprehensive Claude Code class online. If I wanted to make like 5 different variations of the landing page, then Claude Code main stores certain packages for one variation that's required, but that's not required for a different variation. Or it might delete some components and move components around. Having all that implemented in the same folder, file structure would get really complicated and these different subagents would be stepping on each other's toes. So if we go to Claude Code and give it a prompt kind of like this and tell each subagent to use its own worktree, then it will go ahead and do that for us. We can add this to the front matter of any agent that we always want to work in its own worktree. So for example, I could go to this subagent over here, just add it right to the top, isolation worktree. And then after calling that subagent, I can go and see in the. claude-worktrees folder for the project, it made agent-a722 blah, blah. I can also see the previous changes that also happened. And come back to any of these worktrees if I were to exit Claude Code and just put in the same name of that worktree.

Resuming Worktrees

And bear in mind that now if you do /resume, then it will firstly show all the sessions in our current worktree. But then we can press Ctrl+W to show all worktrees, including the default main

Exiting Worktrees

branch. And then if you try and exit out of that session of Claude Code, it will tell you, you have three uncommitted files on that worktree. And you have the choice of either keeping the worktree or deleting it. So I'm going to keep the worktree for now so I can come back to it later. Now you

Tmux Worktrees

can also combine worktrees with tmux as well. So if you run claude tmux normally, then it won't work. But if you run claude worktrees, I'm gonna call it like research-tmux, then it will launch Claude Code within tmux inside of the folder that you're currently in. So you can see this is our tmux session name. And the advantage here is that I now have many of the benefits of tmux. So if you've seen my video from 2 weeks ago about Claude Code Teams, if you have that inside of tmux, every single team member running in its own split pane. But we can also use the benefits of things run in the background even if you exit Claude Code. So if I give it a research task, so for example, my Claude Code Masterclass, if I copy the URL, paste into Claude Code and say, "Can you tell me why this is the best Claude Code Masterclass on the internet right now? " And then press enter. If I press Ctrl+B and then press colon, you can see at the very bottom over here, it's like in this orange section. If I type detach, press enter, then it will detach from that session of Claude Code that was just running. And then I can reattach to it later. And see that Claude Code has continued working on it. So if I run the exact same command again on the exact same worktree with tmux, and now reattaching to session, I can then see that Claude Code went ahead and completed the research task in the background. Now, of course, you could have done these things before Claude Code added this functionality, but these commands basically just make the whole process of using it much easier. Because if you use it with tmux, then you can have long-running background agents

Advantages of Worktrees

for example. If you want to do some testing, then you can basically do that in a worktree and not worry about it affecting any other tasks that are currently running. And if it goes sideways or like it doesn't work properly, then you can basically destroy that worktree by exiting out of the session and using that command. You can have different features being implemented in parallel. And the key benefit here is that firstly, for example, if you use plan mode with Claude Code, then Claude Code establishes, okay, we gotta find this file in this part of the codebase, other part of the codebase. But if you had 3 sessions of Claude Code, all editing the same file structure without worktrees, then one could, make a change or move a file around. And then the other two sessions would be confused being like, okay, wait, this like suddenly changed. But by having it in separate work trees, the file structure stays constant until that thing has been implemented. And then you can commit that as a PR to merge into a main branch. You can also do like random investigations or testing with Claude Code in a work tree and throw that away later. That includes like running a full test suite, for example. You can prepare a brand new PR without stashing and switching. So for example, if you're working on a, like a big task and a colleague asks you to like make a quick change, you can just spin that up in a new session of Claude Code inside a worktree and then update that PR, for example. And speaking of which, there is a brand new command

from-pr

which is called --from-pr. So you can put in a PR ID. So let me make a new PR. I can then say I now have a conversation linked to this PR. Going to from-pr, typing in sim-1, I can then load in that previous conversation that just happened. That was linked to that PR. So that's like an easier way of basically finding previous sessions from Claude Code. You can do the thing that I showed before

Advantages Continued...

where you can have different subagents implement different solutions without interfering with each other, installing different packages, removing different packages and so forth, and then make a brand new PR from whichever subagent solution that you like the most. So that makes experimentation much easier. Like I showed with the early example of having different landing pages being created in different worktrees. And finally, you can now hook into worktree events

Worktree Hooks

with worktree create and remove, which basically allows you to do things like installing packages, uh, which may not be installed in some cases, injecting any config or like adding anything to environment variable files so the worktree could actually run locally if you require environment variables for it to run locally. And with worktree remove, once a worktree is deleted, you can stop any Docker instances that were created for that worktree, and delete any temporary databases, caches, or files. If you do wanna learn more about hooks, and I do cover that in my Claude Code Masterclass in a whole new section about hooks, it will be linked down below. Okay

Security Capabilities

now in a limited research preview, Claude Code added the ability to have it automatically scan your codebase and find security vulnerabilities. So they said that using Claude 3. 5 Sonnet, they found over 500 vulnerabilities in open source codebases. Bugs that had gone undetected for decades despite years of expert review. Now, unfortunately I did not get early access to it, but this is kind of what it looks like behind the scenes. And I imagine they're using a different model in this version of Claude Code compared to the one like on your computer, probably using a model that is less restricted when it comes to like identifying security vulnerabilities, which means I can probably identify more of them compared to you just having a security tester subagent inside of Claude Code itself. Because this kind of stuff is a double-edged sword, like the better the model gets at hacking, the better it gets at— identifying security vulnerabilities. Now they did add a bunch more features to Claude Code desktop application

Previews for Desktop

the biggest of which is previews where Claude Code can run a website locally and then click through it to make sure that everything works as intended and take screenshots along the way. So let's quickly test it out. So clicking over here, you want to switch to local and you can see they added the work tree functionality to desktop application too. And then choose your folder and I'm going to do— auto accept edits, and then say, can you go to a landing page of the Next. js application after running the server locally and then test it to make sure all the buttons work as intended. And then if you click the button on the top right where it says preview, press setup, then it will start setting up a development server. And you can see we kind of have a browser-like functionality over here. And it also passed a command telling it how to set up a development server automatically. And this launch. json file will be saved for the future if you commit it to GitHub. So it will be faster spinning up the environment in the future. Anyways, we can now see which server we need set up. So we're gonna choose Next. js only. And then you can see it navigates to that particular website server, except it's kind of misplaced over here. And then I can see the logs at the very bottom for the server. And you can see it's still clicking around to make sure buttons work. I can switch it to a mobile device. So I can ask it to make sure that everything works well on the mobile as well. And then I can select a particular element. So let me just stop it from doing what it's doing. Select this element, for example, over here. And then it kind of like attach the image with the class ID as well and say, change this to a green button, press enter. And now this button is green. So this can be pretty handy for design-related work and also getting Claude Code to click around and behave like a user for you to make sure things work as intended. So I can also do persist sessions as well so that any cookies and local storage will be saved across server restarts, which can be pretty handy to prevent it logging in every time. And then I can do the reverse where I clear the session data so it will log in again. And now if I allow Claude

CI Monitoring

Code to make a PR, in addition to viewing the PR, I can see any CI monitoring and I can set up like auto fix, which basically means that any commands that have failed with the CI monitoring, it will automatically identify them and then fix them. And then doing auto merge, like once all the CI commands have passed successfully, then it will do the merge for me. Now if you did make a

/desktop

change in the Claude Code CLI and you wanted to move over to desktop app, to take advantage of the newer features like the element picker, then you can do /desktop in the CLI. It will then load that session into desktop app and then you can continue previewing it and stuff. They now also

Diffs on Desktop

made it easier to view diffs on the desktop app. So you can press the view diff button over here, the like line count, and then you can see all the changes that were made and anything that was removed. You can click the plus icon on any line and then enter in a comment and then submit that comment over to Claude Code. So we'll make those changes. So this can mean that you no longer need to use VS Code or Cursor just to view diffs because you can do that in the Claude Code desktop app by doing /desktop to move the session over and then you can view the diffs on there. The desktop

SSH from Desktop

app now supports SSH connections. So you can go here, add SSH connection. And let me add one of my Hetzner servers. So if I fill in all the details over here, add SSH connection, connect. Then I can see it's now connected. So I can go over here and choose any remote folders from the server. So I have this folder over here and then say, can you make it a light theme and then redeploy the application? And then this will basically connect Claude Code to server and have it running on there. And this can save you a bunch of hassle in doing very quick commands on servers with Claude Code. So often I just have very tiny side projects hosted for me and my friends online, uh, that I use Claude Code to help manage. And if the projects are small enough, then I just get Claude Code to make it directly on the server. And don't bother using Git or anything else. So now claims to have made all the changes. So let me refresh application. And yeah, it seems to be working

Plug-ins for Desktop

just fine. We also now have support for plugins on the desktop application. So go into plugins. I can see that I already have one installed, which is a CodeSimplifier one. I can then press add to add a plugin. So I'm going to choose the frontend design plugin, which is pretty popular and install just for the project. And of course, when looking through plugins, I can see exactly what they contain. They now made it easy to share Claude Code sessions via the desktop application

Share Link

and the website. So you can make a public link and then share it to any colleagues. They added

/debug

a default skill called /debug, and you can put in any issues that you observed with Claude Code, and it'll basically figure out why that issue is happening using the logs that are stored in your. claude folder in your root directory. So for example, if you're just coding with Claude Code and you noticed a hook that you had set up did not trigger properly, you can just say like, Why did this recent hook not trigger for like blocking migrations? And then after you identified

/rewind

the problem and you fixed it, you can then rewind the session to before you gave this particular command by doing /rewind. And just before that command, it was commit push over here. Restore conversation. Likewise, if you do /rewind, you can now select a point and do a summarize from here.

/rewind summarisation

So Claude Code will summarize a conversation, kind of like the /compact command up until that point and then start a brand new conversation. So just pressing enter, it will compact it and then restart a brand new chat. The ask user question tool in Claude Code can now display markdown.

AskUserQuestionTool Markdown

So if I say something like, can you suggest 4 different orders for the FAQ and how it should be displayed on the landing page? Show me a markdown of how it works, use the ask user question tool. So often that will happen automatically, but I'm just trying to force it to happen for the sake of demonstration. So it kind of looks like this. If you remember previously, the ask user question tool did not have this section over here, the right-hand section. But now if I switch between it, I can kind of see a few different options for how it should look. Unfortunately, it's kind of cut off at the bottom over here. So if I zoom out, then I can see like more of how this would be. So if you have a bigger monitor, that would be helpful. So if— let's say I wanted to go for this two-column approach, I would press N and then I can add in any notes. So say swap the columns and then press enter and then it will start doing that. So you can ask for like tiny mockups, uh, in the ask user question tool for how things should be laid out. The recently acquired Bun team

Memory Improvements

have been making a bunch of memory improvements to Claude Code. So they say that the Bash tool runs much faster and uses less memory. So on the desktop version of Claude Code, you can use your

Skills and Commands in Claude Code Web

slash commands and skills. So I can use my spec developer slash command. And now you can do the same thing on the web version of Claude Code. So typing it out, I can see all the ones that I have over here. Let's do spec developer again. And that will load the skill that I have in this particular project. So this can be slightly better for interfacing with your skills if you don't like using the CLI. On the web version of Claude Code, there is now multi-repo support. So I

Multi-Repo Support

can add another repository over here. And then I can do like cross-repo work. So for example, I can say something like, copy the design of the FAQ part of the master Claude Code landing page to the HyperWhisper application. And pressing Enter, I can see that I now have two repos that are going to be cloned. And it's gonna like do the cross-repo work. So I usually find this most handy when it comes to comparing two different repos to see how an implementation differs, or sometimes I find a better implementation for something and I want it quickly copied over from one project to another one. They now added support for using plan mode from the Claude Slack application. So

Plan Mode in Slack

you can @Claude and then say, "Can you plan some color scheme changes for the landing page? " And then you can basically collaborate with your team on coming up with a plan for how something should

Background Agent Frontmatter

be implemented. And finally, they added two more things to frontmatter of Subagents. So if you do background: true, then that subagent will always run in the background. So you can see I just triggered the subagent. It's now running in the background. And I can actually kill all background agents by doing Ctrl+F twice. So Ctrl+F twice, all background agents killed. So that's also a new command. And subagents now also have memory. So that can be on the user level, project level

Subagent Level Memory

or local level. So that means that if you want a subagent to remember things from the previous time it had ran, Then you can do like memory. So user, project, or local. So let me do project in this particular case. Ctrl+Save. And you may want to update the prompt for the subagent to tell it to make use of its memory or just tell it directly when you're triggering it. So I can say, can you tell the subagent to basically write a poem into its memory? So in this particular case, the subagent declined the request for some reason. So I'll tell it to write something else. And then once the subagent has finished writing something into its memory, You can see there's a new agent memory folder in the. claude folder for the project with the name of the agent. So macOS Log Analyzer, and this is its own memory. md file, which is injected into the subagent every time it runs. So this can solve the issue of having subagents remember something previously they did the last time they ran. But you may have to change your prompt for the subagent to tell it to make better use of this. For whatever functionality you've given the subagent. So combining it with background true, you could have like self-improving, self-iterating, uh, background research subagents, for example. Anyways, that's everything for the video. If you

Conclusion

are interested in my Claude Code newsletter, then that will be linked down below. And you can also check out my Claude Code masterclass that goes through pretty much every single feature of Claude Code. Many people have taken the class and have gone on to be the best Claude Code users at their company and have even gotten promoted because of their newly learned Claude Code skills. Many of the biggest companies in the world are already using Claude Code and expecting proficiency in Claude Code to be a given like nowadays. I also recently added a class where I go through my own workflows or what I'm doing day to day when it comes to using Claude Code. And I will be updating over time on a month by month basis so you get a sense of what I'm currently doing and how I'm making use of many of the features that I have described to you. There will be a link down below if you are interested. And if you want to buy the class for your team as well, there will be another page for that. And you can also book a call as well to like discuss any requirements.

Другие видео автора — Ray Amjad

Ctrl+V

Экстракт Знаний в Telegram

Экстракты и дистилляты из лучших YouTube-каналов — сразу после публикации.

Подписаться

Лучшие методички за неделю — каждый понедельник