#652 Neil: Claude Code Mods That Make Your Daily Workflow Easier

0:00Capital One's tech team isn't just talking about multi-agentic AI. They already deployed one. It's called chat-concierge and it's simplifying car shopping using self-reflection and layered reasoning with live API checks. It doesn't just help buyers find a car they love. It helps schedule a test drive, get pre-approved for financing, and estimate trading value. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One. We use advanced AI tools to automate our hardest work. Yet we still manually babysit token limits every single day. Oh yeah. Right. We track our own complex task progress, and we worry about two sessions editing the exact same file. It is a fascinating modern paradox really. It really is. We built a machine to do the heavy lifting, but we still do the administrative chores for the machine. Welcome to the deep dive. It's genuinely great to be here exploring this today. Today we are exploring a solution to that exact paradox. We're looking very closely at Claude

1:03code mods. Right. And we'll thoroughly explore what these tools actually are. Yeah. And then we'll walk through a step-by-step building example together. We're going to build one entirely from scratch. And we're doing this using just plain English. I mean, no coding experiences required for this at all. Exactly. After that, we'll review four specific ready-to-install mods. These tools will fundamentally change your daily workflow. I'm really excited to get into these specific tools. Let's start with the basic foundational concept here. We have this paradox of manual AI management. How do mods actually solve this problem for you? Well, we first need to separate plugins from mods. People confuse these two terms very frequently. Think of plugins as basic add-ons for the system. They give Claude code a few extra backend capabilities. Okay, let's unpack this a little bit further. What's the fundamental difference between the two? Plug-in might just pull data from a server. It doesn't really change how you interact with the software. Right. But mods go much further than that basic functionality.

2:05Mods deeply customize how Claude code actually works. They change what appears on your screen visually. You might want to track your context window constantly. You certainly don't want to keep checking it manually. Exactly. Or maybe you run multiple sessions simultaneously. You want a warning before they edit the same file. Yeah. Mods let you add these features directly to the desktop. And here's where it gets really interesting today. You don't need to know how to write code. That is the core appeal of this entire system. You just describe what you want in plain English. Claude writes the JavaScript or TypeScript for you. It feels like stacking Lego blocks of data. You're the architect giving the high level instructions, the AIs, the builder snapping the pieces together. That's a perfect way to look at it. The AI handles the underlying code architecture completely. You just focus on the ultimate workflow goal. The official guide says mods run through the plugin system. Right. They integrate seamlessly into the official development environment. They add custom panels, interactive buttons, and visual alerts. Wow.

3:07And this all happens right inside the Claude code interface. It allows you to build features fitting your exact workflow. You aren't forced into a generic rigid layout anymore. But if we only use plain English, what's stopping the AI from just guessing the architecture wrong? It relies on the strict structure of the official mods API to keep the generated code grounded. So the API acts as a rigid blueprint for the plain English instructions. Exactly. Okay. It prevents the AI from going totally off script. It keeps the generated code safe and predictable. That makes perfect sense in theory. But I kind of want to see how this holds up in practice. Of course. The source walked through building a specific mod. It was a tool called token weather. This really demystifies the whole building process for you. It shows how accessible this technology has actually become. Definitely. Token weather is a custom mod for tracking usage. Let's define a crucial term before we go any further. Context window. The tools short term memory for your current chat. Perfect definition. So how do we actually build this tracker? If I want to actually build this token weather tracker,

4:10the source suggests starting right in the desktop app. No external tools or complicated terminal commands are needed. You just open the code tab and pick a project folder. The source used a folder called marketing dash brain for this. It feels surprisingly contained and straightforward to set up. You just tell cloud code exactly what you want. You ask it to display context usage above the prompt. Right. And you want it to update after every single turn. But there's a massive pitfall here to avoid. The author of the source made this very clear. Yeah, this is important. You must tell it to use the official mods API. You also have to specifically request API hooks. Let's define what those are really quickly API hooks. Specific event triggers that run code at exact moments. What happens if you don't include that specific prompt requirement? It just makes a very basic custom status line. It won't function as a true visual mod. It'll just sit there doing absolutely nothing useful. Got it. So you have to mention JavaScript, TypeScript, and the plug-in structure. You must tell it not to modify your status line configuration.

5:14Once you clarify that, it creates a token weather folder. It builds the module and the plug-in configuration files automatically. It sets up three specific hooks to make it work. The first hook is called session.start. Okay. This checks your context window usage when a session begins. The second hook is called turn.complete. This updates the usage data after each main conversation turn. And the final hook is called UI.render. This actually displays the token weather bar above your prompt. I still wrestle with prompt drift myself. Oh, man. It happens to absolutely everyone who uses these tools. Let's define that term for clarity. Prompt drift. Losing track of your original goal during longchats. It's so easy to lose track. It really is. Having visual token limits on screen is incredibly vital. It keeps you completely grounded in reality. You know exactly what the AI can still remember. So after generating the code, you install the mod locally. You do this via a local marketplace

6:17in the app. You prompted to install the existing token weather mod. You tell it to validate the plug-in before installing. You instruct it to keep the existing files completely unchanged. And you do this entirely inside the desktop application. No GitHub or external terminals required at all. It handles the deployment seamlessly in the background. The AI actually created a local marketplace called My-Mods. It ran the validation and started the installation process. But there was a slight hiccup during this test. Right. Claude initially couldn't find the Claude command in its path. Let's define what that means path. The systems map of where software lives. Because it couldn't find it, it had to pivot. It smartly located the bundled Claude.exe file instead. Oh, wow. Yeah. It finished the installation job completely on its own. That's a highly impressive level of autonomous problem solving. After that, you simply reload your Claude code plug-ins. Right. You just type slash reload dash plug-ins directly into the chat window. In the source test, it loaded 17 different plug-ins. It also loaded 158 skills, 11 agents, and three hooks.

7:23That's a lot. And you can verify the installation by typing forward slash plug-in. Then you get the payoff for all this work. You check your brand new token weather mod. You send a message and look directly above the prompt. And what did it look like? The bar showed 23% context usage immediately. It displayed the token count and the recent usage change. It even showed the estimated session cost right there. Everything is displayed in one small convenient visual bar. Yeah, you never have to open another clunky menu again. It updates automatically after every single conversation turn. It's a massive quality of life improvement for daily users. Why do we need specific event hooks rather than just telling it to update constantly? Well, constant updates would drain your system resources heavily. Whereas hooks tie actions to specific necessary moments like a completed turn. Right. Hooks guarantee it only processes data when something actually changes. Exactly. Made a response or read placeholder. We just walk through building a local mod entirely from scratch. Now let's transition to utilizing existing ones from a directory. Building is fun, but downloading

8:26is much faster. Oh, for sure. There's a whole ecosystem of these tools out there now. We're starting with the most critical workflow issue mentioned earlier. We need to deeply discuss managing your context limits. This brings us directly to the Nateherk Mods Marketplace directory. There are four specific mods we want to cover today. The first one we're looking at is called cash keeper. This is absolutely best mod for long sessions. It solves a massive headache for heavy users. It helps you track your context usage constantly. It shows how long your prompt cash stays active. Let's define what that actually means for you listing warm cash saved memory that makes the next AI response cheaper. And it estimates your API costs in real time. Installation is incredibly simple for this specific tool. Yeah, you just visit the cash keeper page and copy the instructions. You open Claude code desktop and start a local session. You just paste those copied instructions into the chat window. And the AI handles the entire installation process for you. Or you can use the graphical interface instead. Go to plugins, then code, then Nateherk Mods.

9:31You just find cash keeper and click the plus icon. It really couldn't be any easier to do. To use it, you start a new session and type slash cash. The source detailed some very specific test results here. Typing the slash cash showed 121 k tokens used. It showed a 60 minute cash window that was still warm. Usage was at 33% of the 5 hour limit. It was also at 44% of the weekly limit. The estimated session cost was exactly 65 cents. It gives you complete financial transparency for your workload. It even showed a cold send guard enabled at 150 k tokens. It warns you before your warm cash actually expires that saves you from spending unnecessary money on tokens. It prevents those massive unexpected billing surprises. There's also a brilliant slash handoff feature included here. This is incredibly useful for long complex project workflows. But is the slash handoff command just doing a bulk copy pace of the chat history?

10:32No, it actively prepares and distills the important context, rather than just dumping raw text into a new session. It saves your core project memory without dragging the dead weight along. Exactly. It makes your next session fast and highly efficient. It strips out the conversational filler and keeps the code. Managing the AI's memory is beautifully solved by cashkeeper. Now we must move to managing human level operational issues. We're talking about privacy, project scale, and file management. These are the things that usually trip to velvet for a resupp daily. The second mod to look at is recording mode. Privacy is a massive hurdle when sharing work spaces. Oh, absolutely. If I'm recording a tutorial, my API keys are usually visible. Right. And that's exactly what recording mode is designed to fix. It's for people who record tutorials or share screens frequently. It keeps sensitive information completely out of public view. It hides email addresses, API keys, and financial details seamlessly. It places visual placeholders directly on your screen, but, Cloud Code can still use the original information behind the scenes.

11:36You're activated by simply typing slash rec. It instantly redacts everything you want hidden from the recording. You can also use slash rec strict for stronger protection. When you finish recording, you just type slash rec off. Always test this with sample data first, though. It might occasionally miss some highly specific sensitive details. Good point. You don't want to accidentally leak a production database key. You should always check what's visible before sharing publicly. The third mod on our list is called goal meter. This is a progress dashboard designed specifically for large projects. It's easy to lose track of what Cloud has finished, especially if you're running several concurrent sessions at once. Gole meter shows the current task plan very clearly. It shows completed tasks and the overall completion percentage. The source tested this with a multi-step workflow task. They asked Cloud to build a simple landing page. Right. It was specifically for an AI newsletter. They created a plan, implemented the page, and verified it. Gole meter recognized the workflow and tracked the interface progress.

12:39It gave a visual representation of the entire build process. It makes the most sense for those longer complex projects. It doesn't add much value for quick questions or simple edits. Yeah, that makes sense. The fourth and final mod today is collision guard. This solves a major problem for complex development workflows. It really does. Multiple sessions might accidentally edit the exact same file. They can easily override each other's hard work. You could lose hours of prompting in a single second. Collision guard checks a 30 minute window for any recent edits. If it detects a possible conflict, it warns you immediately. You activate it by using the slash guard command. The source tested this on that exact newsletter landing page. They asked Cloud to create an HTML index file. Cloud immediately responded that collision guard was currently turned on. It actively monitored plan.md and index.html. It successfully verified no conflicts occurred in that 30 minute window. Well, imagine scaling to dozens of concurrent sessions without the AI

13:42ever stepping on its own code. That completely changes how we manage automated file building. It acts as an incredible safety net for developers everywhere. You can confidently run parallel tasks without any fear. So going back to privacy for a second, does recording mode actually delete your private data from the backend logs? No, it only masks the data visually on your screen for screen sharing purposes. The data remains in the actual workflow. Got it. It's a visual shield for your screen, not a data eraser. Right. It just keeps your secrets off the video recording itself. The AI still uses the real data to do the work. Let's step back and look at the big idea here. Mods solve workflow friction points. You might not even realize you have. You absolutely don't need to change your entire workflow. You don't have to overhaul your whole development process today. You just pick one single tool to start with today. Cash Keeper is fantastic for beginners managing those tricky token limits. Let plain English do the heavy lifting of customizing your environment. You're literally talking your custom tools into existence.

14:45It's a fundamentally new way to interact with software. It bridges the gap between idea and execution perfectly. It leaves me with a very deep reflective question. We're already using AI to generate these custom tools. The tools monitor the AI's own limitations like tokens. They proactively correct the AI's own workflow collisions. How long until the system starts proactively coding and installing its own mods to optimize our workflow before we even ask? That's a staggering concept to think about. A self-optimizing workspace that evolves alongside your habits. Evaluate your own daily pain points with AI today. Try installing just one of these mods to fix it. Thank you for joining us on the deep dive. Capital One's tech team isn't just talking about multi-agentic AI. They already deployed one. It's called chat-concierge and it's simplifying car shopping. Using self-reflection and layered reasoning with live API checks, it doesn't just help buyers find a car they love.

15:46It helps schedule a test drive, get pre-approved for financing, and estimate trading value. Advanced, intuitive, and deployed. That's how they stack. That's technology at Capital One.