You type a command into an AI, nothing happens. Beat. It's not broken. It's just waiting for you to teach it how to think. We really do treat it like static software. We assume we're just passive consumers. But you can actually rewrite. It's cognitive pathways. Welcome to the deep dive. Today we're exploring five custom slash commands. Yeah, these are built specifically for the Claude Code app. But the catch is you
have to build them yourself. Right. We are your hosts, and we're looking at Plan, Clear, Goal, Rewind, and BW. These commands bridge a really vital gap. They bridge the gap between AI just doing things and AI doing things your way. But before we build them, we need to understand a common frustration. You watched a developer demo on YouTube? Oh, yeah. They attack a slash command. It runs perfectly. Right. You try it on your
machine, and nothing happens. People assume their software is just outdated, or they missed a secret plugin. But let's look at what a slash command actually is. It starts with a forward slash. It triggers an action instantly. Exactly. But the secret here is it's not native software. It's literally just a plain markdown file. We call it a skill. And it lives locally on your machine. Yeah. Claude reads that exact file when you type the command. That is why lists vary
between users. I mean, does this mean we are basically programming the AI's internal habits from scratch? Yes, absolutely. You are dictating its operational boundaries. You write them in plain English. So they were just text files directing the AI's behavior behind the scenes. Exactly, like stacking Lego blocks of data. So let's look at the first command, the plan command. Because Claude has a tendency to rush, right? Yeah, the problem is Claude starts coding instantly. On
big tasks, it picks the wrong approach. It wastes so much time. It's like a driver accelerating before checking the map. That's a great way to put it. The planned skill fixes this. It forces Claude to stop entirely. It forces it to read the task first. Right. It breaks the task into steps, it lists the affected files, it flags risks, and then it waits for your approval. So like building an event signup page. Exactly. Using HTML and CSS, Claude stops and asks question.
It asks if you need email validation. And it does this before writing a single line of code. Yeah. But exception here. You skip this for one line, one file fixes. Right. Why shouldn't we just type out these instructions manually every time? Well, because human discipline fades, you'll forget, and then the AI messes up. Typing it repeatedly wastes your time. This automation saves your sanity. Perfectly said. Once you finish that task, you have a new problem. The AI memory
becomes a liability. We should quickly define context window here. The memory space where the AI reads your current... Chat history. Spot on. The problem is old context slows down responses. Claude tries to connect totally unrelated tasks. Like debugging a login form today, then switching to writing a course product description. Exactly. You do not want code in your marketing copy. The clear skill fixes this. It wipes the slate clean. It tells Claude to treat the prompt as
brand new. Yeah. It ignores past files completely. But there's a crucial detail here. New commands only load when a session starts. So you have to start fresh. Right. You must open a new chat. Then the clear command will appear. Does the AI naturally realize when a topic has completely changed? Uh, no, not at all. It lacks that awareness. It just tries to blend everything. No, it assumes everything connects together unless you explicitly tell it otherwise. Sponsor. Beep. We're back.
So we've started a clean task. But how do we stop the AI from getting lazy? Yeah, this is the huge issue. Instructions usually say what to do. They don't specify when it counts as done. So Claude just stops halfway through. Exactly. And it requires you to constantly reprompt it. The goal skill changes this. It establishes a clear finish line. It gives it a strict condition to meet. Right. Claude works through the steps. It checks the result against your condition.
It does this after every single step. And it only stops when fully met. Yeah. Beat. Whoa. I mean, imagine it iterating through 50 steps overnight, perfectly checking its own work. Like building a landing page. Exactly. Where it done strictly means a hero section, one CTA, and mobile responsive. But wait, how does it actually know it hit the goal without our human eyes on it? It runs internal verification loops against your text. It reads its own output like a human reviewer.
It constantly checks its own output against your strict definition of done. Exactly. But even with clear goals, things go wrong. Sometimes the AI takes a completely wrong turn. We need an undo button that works. Because Claude can make a really bad mid -task edit, and rolling back code by hand takes forever. I still wrestle with prompt drift myself. Sometimes I let it go too far before fixing it. We all do, man. That's why the rewind skill is amazing. It lists
recent session checkpoints. With short little descriptive labels, right? Yeah. You pick one, and it rolls back both the conversation and the code. It takes everything back to that exact state. So if Claude restructures files in a way you didn't ask for? You just rewind to the checkpoint right before it happened. Does rewinding mean we also lose any genuinely good code it wrote after that checkpoint? Uh, yes, unfortunately. Rolling back wipes the slate clean moving forward.
You have to act fast. Yes. Rolling back wipes everything after that point, so you must act quickly. Right. So say the task is flowing nicely now, but suddenly a random thought pops into your head. How do you ask? without derailing the AI. That's the problem. Asking a side question breaks the flow entirely. Claude mixes the answer into the main task's code. It pollutes the context window. Exactly. The B2 skill is the fix. It answers the question directly. It doesn't affect
ongoing files at all. And then it returns focus to the main task. Yeah. Like, say you are coding a landing page. You ask for SEO image naming patterns. It isolates that query. Right. It answers it and acts like the interruption never even happened. What happens if you actually want the side question to influence the code? Oh, well, then just ask it as a normal prompt. Only use this for isolated thoughts. Then just ask it normally. Only use this command to keep tangents
completely isolated. Exactly. So to recap our big idea today, these five tools, Plan, Clear, Goal, Rewind, and B2. They don't exist until you build them. Right. And you definitely don't need all five today. Just pick the one your workflow misses most. Save the markdown file and try it. Yeah, exactly. See how it changes your workflow. Thank you for listening today. We spend so much time marveling at what AI can do. We really do.
But perhaps the real leap forward is realizing we have the power to shape exactly how it thinks. You are no longer just a user. You are the architect of your own AI. Outtio Music.