Claude Code's Creator Uses These Claude Skills Every Single Day

The creator of Claude Code previously revealed his entire workflow and we already made a video on that. We took a lot from it and started applying his principles into our own work. But we wanted to go further. So we started looking for the specific skills and tools that his team actually uses daily. We went through their posts, the open- source repos, the official plug-in marketplace, and pulled out the skills and commands that they actually use. Some of these you can install right now. Others are internal tools that the team never released, but we figured out what they do and how you can build your own. Now, one skill that they use is the front-end designs plug-in for claude code. It was developed to convert designs using simplistic prompts while enhancing them in terms of UI and UX. It is available open source and the repo contains a skill with specialized instructions for design. This is available upfront in the official claude code plug-in repo. Basically, the reason for this is so that AI can avoid general AI aesthetics, which are the usual designs that models tend to converge toward most of the time. The installation process is simple. you can just add it from the claude code plugins official marketplace. But if it is not installed, you can install it using the plugins add marketplace command and then install it from there. After a reload, the plug-in will be available for use. Now, once it's installed, you can invoke the skill using the slash command directly and specify what you want to do with it. Since skills are usually invoked whenever Claude feels the need for their functionality, this skill will also be triggered if Claude determines it is necessary and it can invoke it during execution as well. Now, another skill that the team behind Claude Code uses daily is the batch skill. This skill is integrated directly into the product and is designed to automate tasks that can be easily parallelized, such as migrating code from one library to another or performing similar operations. You can use it directly inside Claude code and give it any instruction after the batch command to execute your task. Once you provide Claude code with that prompt, it first identifies the task and breaks it down into subtasks by entering plan mode. This allows it to explore and plan the implementation thoroughly before executing. Now, you might think that Claude's agent system already breaks down tasks into smaller parts and executes them in parallel. So, why is this necessary? But the difference is that the batch command uses separate work trees, which are isolated copies of your repository, letting each agent work in its own environment. Therefore, this prevents agents from affecting each other's work. The plan generated during planning contains the current state of the app, a division of all the work into units, a list of required additions and instructions for how end-to-end verification should be performed. Once you approve this plan, it will spawn the same number of agents as there are units of work, giving each a dedicated prompt and ensuring that each agent operates in a separate work tree. Each agent reports its progress back to the main agent. And once all of them have completed their tasks, the main agent merges all the isolated work trees into the main branch. It can even manage the pull request if a remote is configured. Now, they also use the code simplifier plugin frequently. This plugin is open- source and can be downloaded for use in your own project. It's available in the official plugins repo and can be installed following the same installation steps. Once installed, you'll see it in your list of installed plugins, and you can also search for and install it directly from the repo. This plug-in refineses code for clarity while preserving functionality. To use it, you provide a prompt and it initializes the agent inside the code simplifier. It then spawns a single agent to simplify the entire codebase by removing duplicates and unnecessary files, making the codebase cleaner and easier to manage. Once it completes its work, it returns a summary of the changes it made. But this is actually similar to the Claude code simplify skill which was released as a product within Claude Code. The main difference is that simplify spawns three agents and rigorously evaluates the code across multiple metrics to make it simpler and more effective. So simplify is a more thorough version of the code simplifier plug-in and both tools are heavily used by the team. But before we move forwards, let's have a word by our sponsor airtop. If you're spending hours every week manually doing repetitive tasks, there's a much better way to handle that. Airtop lets you build AI agents that run 24/7 in real cloud browsers using plain English. No code, no APIs. I set up their competitor spy template, gave it three competitor URLs, and told it to monitor them daily for pricing changes, new features, and positioning updates. Every morning, Airtop spins up a cloud browser, navigates each site, handles captures, and antib-bot measures, compares everything to yesterday's data, and drops a summary straight into my Slack. Set it up once, 10 minutes, and now I get a competitive intelligence briefing every morning before I even open my laptop. Build reliable automations in minutes and sign up for airtop using code AI labs for a free month of their starter plan. Click the link in the pinned comment and build automation with airtop. Now, another skill that the team uses internally is the verify skill. It is not mentioned in their tweets, but from the leaked source code, we can see that it is an internal skill for anthropic team members hidden behind a CLI flag and not available to others. This skill verifies code changes by running the app, testing the changes from different angles, and automatically fixing failures. Since this is used internally, the system prompt is not inside the product's code. It is injected directly into the environment when any one of the team is actually using it. This may be why it hasn't been released publicly as it is highly project specific. But if you are building your own project, you can create your own verify skill using the skill creator or by using the repo as a template. It will scan your codebase and generate a verify skill tailored to your project. Verification can look different depending on the project. Some might use playright MCP verification tests or other methods, but the core idea is that the agent need to have a way to confirm its changes are working as intended, and this skill handles that heavy lifting. The skills instructions include step-by-step guidance on how to handle code changes and report back using the available tools. It also contains examples in CLI.md listing all CLI tools the skill can access, including llinters, exit codes, npm test files, and more. For optimal use, it should be configured with test cases and visual verification using the Claude Chrome extension. Now, another internal skill hidden inside Claude code and available only to anthropic team members is Skillify. This skill captures an entire session and turns it into a reusable skill. Like the verify skill, it is protected by an environment flag. But unlike verify, the system prompt for Skillify is included in the source code. The purpose of Skillify is to record a session where you brainstorm a workflow with Claude, refine the process through multiple conversations, and then save it as a reusable skill. It identifies repeatable processes discussed, all tools and permissions mentioned, the agents involved, and everything needed to create a skill. It then interacts with the user to clarify intentions, breaks down the steps, finalizes the skill.md file, and generates a detailed guide containing all necessary instructions. But you don't need access to the internal version. You can create your own Skillify skill using the source code as a guide. Once invoked, the Skillify command thoroughly analyzes the session and confirms with you whether its deductions are correct. It asks follow-up questions to clarify the workflow, then generates a skill.md file and provides it for your review. After confirming that it matches your needs, you can save it to your project and use it as a reusable skill. Now, another tool they use is the Claude Code Security Scan command, which scans the codebase for vulnerabilities and suggests patches based on various factors. It has been used internally by the team to identify multiple security issues and was later released as a security scan feature inside cloud code. The system prompt can be accessed from the leaked source and the tool is configured to perform security scans from multiple angles including input validation, authentication issues, secret management, injection and code execution, endpoint exposure, and more. It also includes an in-depth analysis methodology and incorporates multiple standards directly into the tool. This is a built-in feature, so you can access it directly to perform a security review of all files. It follows its instructions and reports all findings. You can then prompt Claude to fix the identified issues, and it will implement the changes needed to make the app functional and secure. This is especially helpful for large-scale applications where security issues can be difficult to spot and might go unnoticed in production. With AI generated code producing large volumes of changes, vulnerabilities could accidentally make it into production. And this tool helps prevent that. Also, if you are enjoying our content, consider pressing the hype button because it helps us create more content like this and reach out to more people. The creator of Claude Code says that their team uses SL commands for every inner loop workflow that involves repetitive tasks. For example, they use a commit pushpr/comand that pushes code to the repo and opens a pull request. This command is open source and available on the Claude Code plugins official marketplace. You can access the commit command which generates commits based on staged and unstaged changes by following a specific series of steps. It contains the exact command the creator mentioned in his expost and includes the workflow for it as well. To install it, go to the plugins marketplace, search for the commit command and install it. Once installed, you can access all the commands available in the repo. When working with it, you can use these commands directly in your workflows to streamline repetitive tasks. In the workflow that the creator of Claude Code revealed, he actually mentioned that their team uses a tech debt skill. They advised running it at the end of every session to find duplicated code. This is not actually open- source and just used by the team internally since tech debt varies by the project it is being used in. The same went for them. But you can use it for your own purposes and tailor this skill to your own project. You can create your own for your own project. name it the same as he did and add in your instructions for the skill using the skill creator skill which is actually open source. The instructions for the tech debt skill should contain a step-by-step workflow on how to identify changes, how it should actually detect duplicates and how it should identify the tech debt that was collected during the session. It should also define how it handles the file structure relative to the project in detail and it should end with verifications like running npm test and the llinter. Now once all of this is in place, if you run the tech debt skill, it will go through your entire existing codebase first using multiple agents to analyze it and will report back the duplications in detail that are actually redundant and not needed. It then creates a shared library where all components can use the same feature and updates the components to access the same things from different places. Therefore, your session ends with a proper cleanup and a much more organized codebase with minimal tech debt. Now, this skill that we created can be found in AIABS Pro for this video and for all our previous videos from where you can download and use it for your own projects. If you found value in what we do and want to support the channel, this is the best way to do it. The links in the description. Now, the creators of Claude Code also use another skill called DDUP. This skill is used internally by their team and is not available in any open-source form or inside Claude Code. But from the results of the issue they linked in their tweet, we can reverse engineer what the skill possibly does. Even though we cannot access the exact specialized system prompt, from what we observed, this skill checks whether a GitHub issue is a duplicate of an existing issue, and if it finds a duplicate, comments on it. It receives issues, parses the input, and searches for similar issues using the GitHub CLI. It contains criteria for determining what qualifies as similar or not. in case it doesn't wrongly qualify an issue is duplicate or not. There is a set threshold which must be met before the issue is flagged as a duplicate. Once it finds a duplicate with around 70% certainty, it comments on that issue. The comment it makes on the duplicate issues includes information about why it matches and an explanation related to it. It still requires human verification though. So in case you are working on a project with multiple people and someone creates an issue that has already been dealt with, you can use this skill to handle the repeated ones, allowing you to focus on new issues instead of wasting time clarifying that the issue has already been resolved. Now another skill that the team at Anthropic uses frequently is the remotion skill inside Claude Code. This skill allows you to create motion graphics and videos with AI using simple prompts. You can install it and use it directly inside Claude code specifying how you want your animations to look. Now, did you know that this skill is actually used across all product announcement videos and other marketing videos produced by the Claude marketing team? It's one of the most important skills in their workflow, and it powers many of the team's non-development tasks as well. That brings us to the end of this video. If you'd like to support the channel and help us keep making videos like this, you can do so by using the super thanks button below. As always, thank you for watching and I'll see you in the next one.