The argument in one line.
Five specific Claude Code features—slash commands, sub-agents, hooks, think modes, and CLAUDE.md rules—directly speed up coding workflows and improve output quality when implemented through single copy-paste prompts.
Read if. Skip if.
READ IF YOU ARE…
- A developer using Claude Code or Cursor who builds applications regularly and wants to adopt specific workflows like slash commands and sub-agents to reduce manual setup time.
- Someone managing a codebase who struggles with keeping documentation synchronized with code changes and needs a systematic way to maintain CLAUDE.md and README files automatically.
- A programmer new to Claude Code's advanced features like think modes and hooks who learns best from copy-paste prompts and live demonstrations of each feature in action.
SKIP IF…
- You primarily write code in non-Claude environments or don't use Cursor — this breakdown is tightly coupled to Claude Code's specific syntax and integrations.
- You're already comfortable with Claude Code's advanced features and looking for optimization strategies beyond basic feature walkthroughs — this is introductory-level material.
The full version, fast.
Claude Code becomes dramatically more effective when you stop using it as a single chat window and start configuring it as a system. The video walks through five compounding upgrades you implement by asking Claude Code to build them for you: a custom slash command that auto-refreshes README and CLAUDE.md files after every change so the agent always has full context, sub-agents spun up from the official docs to parallelize UI, API, and data work, a hook that auto-commits to GitHub on every successful edit so you can rewind like Cursor, the ultrathink keyword reserved for genuinely complex prompts, and a permanent CLAUDE.md ruleset enforcing to-do decomposition and minimal-impact fixes. Treat the agent as configurable infrastructure, not a chatbot.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:18
01 · Hook
Bold claim: Claude Code is the most powerful AI tool, but viewers are using it wrong. Promise of 5 features that 10x speed overnight.
00:18 – 02:39
02 · Feature 1: Slash Commands
Build a /update-docs slash command via one Claude Code prompt. Auto-regenerates all READMEs and CLAUDE.md files after every successful change. Live demo on a stock portfolio app.
02:39 – 05:34
03 · Feature 2: Sub-Agents
Paste the official Anthropic sub-agents docs into Claude Code and let it decide which agents your app needs, then build them. Demo produces 5 app-specific agents.
05:34 – 07:46
04 · Feature 3: Hooks
Set up a PostToolUse hook that auto-commits to GitHub on every successful code change — recreates Cursor's rewind feature inside Claude Code.
07:46 – 10:48
05 · Feature 4: Think Modes
Three tiers: think / think hard / ultrathink. Side-by-side Three.js FPS demo shows dramatically richer output from ultrathink. Reserve for complex tasks.
10:48 – 12:03
06 · Feature 5: Claude Rules (CLAUDE.md)
Nine CLAUDE.md rules the host uses in every project: plan first in todo.md, think step by step, keep changes minimal, validate all fixes, never be lazy.
12:03 – 12:33
07 · CTA
Like + subscribe + Thursday 4PM ET livestream callout.
Lines worth screenshotting.
- A slash command that auto-updates all README, documentation, and CLAUDE.md files after every successful code change ensures Claude always has full context without any manual maintenance.
- Asking Claude to read the sub-agents documentation and determine the best sub-agents for your specific app produces more relevant agents than any list you would create manually.
- A Claude Code hook that automatically commits code to GitHub after every successful change recreates Cursor's rewind functionality — any broken state can be reversed via git.
- Think modes — extended thinking and ultra-think — give Claude more reasoning time for complex architectural decisions and produce qualitatively different outputs on hard problems.
- Hooks are event-driven triggers that fire automatically on any action in Claude Code — code change, commit, test failure — enabling autonomous workflows that run without user prompting.
- Sub-agents built for tasks, not roles, outperform role-assigned agents — an agent tasked with reviewing UI components in a browser is more reliable than one assigned to 'be a UI designer.'
- The auto-update-docs slash command eliminates the documentation drift problem — the agent always works with an accurate description of the current codebase rather than a stale one.
- Five specialized sub-agents for a stock portfolio app — UI designer, portfolio optimizer, dividend analyst, data importer, API integrator — collectively handle the full complexity of the application.
- Asking Claude to determine which sub-agents to create rather than specifying them yourself produces agents you might never have thought of that are specific to your app's architecture.
- Giving Claude extended think time on an architectural decision is worth the token cost because the reasoning process catches tradeoffs that standard-mode responses miss.
- A hook that runs a linter and fixes all errors automatically after every code change maintains code quality as a background process rather than a periodic manual task.
- The five features together — slash commands, sub-agents, hooks, think modes, and CLAUDE.md rules — form a complete autonomous development environment that improves itself over time.
Five Claude Code Features That Change How the Agent Manages Context, History, and Complexity
Claude Code features
Alex Finn's five-feature walkthrough shows that the gap between basic and advanced Claude Code usage is almost entirely about infrastructure — slash commands that keep docs current, sub-agents that parallelize complex work, hooks that auto-commit history, think modes that scale reasoning depth, and CLAUDE.md rules that enforce discipline across every session.
02Feature 1: Slash Commands
- Build the /update-docs slash command with one prompt: Claude generates and saves the command file to the project directory
- The command regenerates all READMEs, documentation, and CLAUDE.md after every successful change — the agent always has accurate context on the next session
03Feature 2: Sub-Agents
- Paste the Anthropic sub-agents documentation into Claude Code and ask it to propose which agents the current app needs — the agent analyzes the codebase before proposing
- The output is five app-specific agents rather than generic role-defined agents — the specificity is what makes sub-agents work in practice
04Feature 3: Hooks
- A PostToolUse hook fires after every successful tool call — wiring it to a git commit creates an automatic checkpoint for every verified edit
- This pattern recreates the rewind functionality without a separate tool — every successful code change is now a recoverable state
05Feature 4: Think Modes
- Three tiers: think, think hard, ultrathink — each adds reasoning depth at additional cost
- Reserve ultrathink for multi-constraint architectural decisions; standard think handles most implementation tasks without the additional overhead
06Feature 5: Claude Rules (CLAUDE.md)
- Nine rules: plan first in todo.md, think step by step, keep changes minimal, validate all fixes, never be lazy, prefer simple solutions, document as you go, test before reporting done, explain tradeoffs
- These rules are not prompts — they are persistent constraints that apply to every session without re-instruction
Terms worth knowing.
- slash command
- A shortcut triggered by typing a forward slash followed by a keyword in Claude Code — executes a predefined multi-step workflow automatically, replacing the need to type out repetitive instructions each time a recurring task needs to run.
- CLAUDE.md
- A markdown file placed in a project's root directory that contains persistent instructions sent to Claude as context with every prompt — used to define coding rules, project conventions, and behavioral guidelines the AI agent follows throughout all sessions.
- sub-agent
- An independent Claude Code agent spawned to handle a specific domain or task in parallel — for example, a UI design agent or API integration agent — allowing multiple specialized agents to work simultaneously rather than one agent handling everything sequentially.
- hook (Claude Code)
- An automated trigger that fires when a specific event occurs in Claude Code — such as a successful code change — and runs a predefined action, like committing code to GitHub, without requiring any manual input.
- think / think hard / ultrathink
- A set of Claude Code reasoning modes activated by including the corresponding keyword in a prompt — each mode allocates progressively more computational effort and extended thinking to the response, with ultrathink producing the most thorough but slowest output.
- Three.js
- A JavaScript library for creating and rendering 3D graphics in a web browser — used here as the rendering engine for a browser-based first-person shooter game generated in a single file by Claude Code.
Things they pointed at.
07:00linkAnthropic Hooks Documentation
Lines you could clip.
00:00
“Claude Code is the most powerful AI coding tool ever made. But what if I told you you're using it all wrong?”
03:57
“Instead of you having to do all the decision making around which sub agents you should create and googling a 100 different things, you just give it to Claude Code.”
06:18
“Setting up this hook basically recreates that functionality so that anytime you wanna rewind, you can.”
10:45
“This last one is mandatory. Everyone should do this.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
analogy
00:00Blogcode is the most powerful AI coding tool ever made. But what if I told you you're using it all wrong? In this video, I'm gonna show you five hidden features that if you start using right now, you'll 10 x your coding speed overnight.
00:14Let's get into it.
00:18The first hidden feature I'm gonna show you is a slash command that made my workflow so much easier. For those that don't know, a slash command in Claude code is a quick command you can give with a slash that automatically runs some sort of process. The one we're about to build will automatically update all our documentation in the app.
00:36Having up to date documentation in Claude code makes your AI agent so much more powerful because it has full context every time it works.
00:44So it is critical that you constantly have your documentation up to date. The issue is it takes a lot of time to do. But with this slash command I'm about to show you, it makes it so much easier.
00:54So I want you to open up ClawdCode inside cursor and put this prompt in. This will build the slash command for you. Build me a slash command in Claude code that automatically updates all my read me's documentation and Claude dot m d files every time a successful change happens.
01:10We're gonna hit enter on that, and what's gonna happen now is Claude will actually build out the slash command for us, save it to our directory, and now all we'll need to do is do slash update docs every time we make a change, and it'll update all our documentation for us. Alright. So Claude built out the slash command for us.
01:28So now all I need to do is do slash update docs, hit enter on that, and it's going to use our new slash command that we just built to update all our documentation. So as you can see, it's in my stock portfolio app I built.
01:42It's going through, and it's creating documentation on every single component built. And the advantage here is next time I put in a prompt to change something in the app, the AI agent will have all the documentation to pour through so it knows exactly how to work on the app. This is gonna make it so the coding AI agent does is 10 times more accurate, and we get way better result.
02:02Alright. Looks like the slash command finished. Let's check out what it did here.
02:05Here's the new readme file it automatically generated. It updated the read me with every single feature, every single technology, how it runs so that next time we build anything out or maybe we even go to a new AI agent, it will have full context on how our app works.
02:22The results you're gonna start getting just by this one prompt I gave you is going to make your app so much better and and building with Claude code so much nicer. So that one feature is gonna speed up your coding, increase the quality of your app, but wait till you see this next one. It's It's even better.
02:39The next hidden feature I'm gonna talk about just released a couple days ago, and that is ClaudeCode subagents. What subagents allow you to do is instead of coding with one agent in ClaudeCode, which can take a tremendous amount of time, you now can create sub agents that can handle multiple tasks at once that will speed up your entire coding workflow.
03:00You can have sub agents for UI design, sub agents for creating content, sub agents for managing your database. There are so many options. But I figured out the best way ever to create sub agents inside a Claude code.
03:13So down below is a link to the documentations for Claude code sub agents. What you wanna do is this.
03:19You wanna copy the entire page of that documentation. Once you've copied the page, we're gonna go back into Claude code, and then you're gonna wanna put in this prompt, which it'll be down below. I link all the prompts down below and all the links you need.
03:30There is a new feature in Claude code called sub agents. The documentation is pasted in down below. Please go through it and determine what would be the best sub agents you can create for my app that would increase the quality of the app and speed up coding time.
03:43Then implement the new sub agents, documentation, and then paste in the documentation we just copied over and hit enter.
03:50Now ClaudeCode's gonna go in, determine what are the best sub agents for your app that will speed up your coding, and then create them for you. Instead of you having to do all the decision making around which sub agents you should create and googling a 100 different things, you just give it to Claude Code, and Claude Code will determine the best sub agents customized for your app.
04:10Then when you give Claude Code any prompts after this, it will automatically spin up the best sub agents for you to speed up your coding time. Alright.
04:18Looks like it completed the agents. Let's see what it built. So, again, as a reminder, this is a stock tracking portfolio app I built.
04:26So it looks like it built me five different sub agents you can spin up at any time, a UI designer that'll be totally in charge of designing the UI and making sure it looks beautiful, build responsive mobile first components, implement interactive data visualizations. I love that. It has a portfolio optimizer that will actually optimize everything in my portfolio to make sure I'm investing in the right stocks.
04:48I like that. A dividend analyst that will analyze all my dividends as I'm building out different parts of my application. A data importer that will handle importing all the data into my app.
04:59An API integrator, this is a big one because my app uses a ton of APIs. It has an entire agent specifically for APIs. That's amazing.
05:09That's the power of using sub agents and having Claude code determine which sub agents to build. It will make sure it builds the best sub agents for your app that you might not have even thought of. Again, this is another feature that will make it so you code so much faster and get way better results from your agents.
05:24So that feature is gonna have a tremendous impact on how you code, but this next one's also gonna have a massive one as well.
05:34Not many people know about Claude code hooks, but they are a massively helpful feature inside Claude code. The way Claude Code hooks work is hooks are called anytime you take an action in your app. So you can set up hooks anytime you make any code changes, anytime you commit code.
05:51Anything you could think of, you can set up hooks that automatically take actions inside code. We are going to set up a hook now that will commit your code to GitHub anytime you make a successful change in your code without errors.
06:05And this hook is so helpful because this is the way you can have check points inside of Claude code. One of the best parts about Cursor is you can automatically rewind anytime the Cursor agent messes up. That doesn't exist in Claude code.
06:19But setting up this hook basically recreates that functionality so that anytime you wanna rewind, you can. By having the Claude code hook automatically commit to GitHub every time you make a code change, you can at any time rewind your Git so that your code goes back to a working state. This is so helpful to do.
06:38So here's the prompt you wanna use. I'll put this down below as well. Feel free to pause anytime.
06:41Just copy paste this and get this done with me. I wanna set up a Claude code hook that automatically commits the code to GitHub anytime we make a successful code change without errors. Please set up this Claude code hook.
06:53I'm gonna hit enter on that. Claude code's gonna set up the hook for us. That's the most beautiful part about using Claude code is you don't need to set anything up yourself.
07:00You just have the AI agent do it for you. So it's going to set up the hook for us now. It's actually going into the Anthropic documentation so we can understand hooks better.
07:09I'm gonna hit enter on that. And now it's gonna go and set that hook up for us. It will automatically commit our code every time we make a change.
07:16So it's so much easier to rewind if we ever make an error. Alright. Looks like it's done.
07:21It triggers after every success or write, edit, or multi edit. So anytime the AI agent makes changes to the code, it will automatically commit our code to GitHub.
07:31So if we mess anything up or we make a mistake, we can easily rewind by rewinding to a past GitHub commit. This is gonna make coding with Claude code so much easier, and we did it again just with one single prompt.
07:46The next hidden feature I wanna talk about is the different think modes. For those who don't know, there are several different think modes in Claude code that will make it work extra hard and give you better code if you use them correctly. For instance, you can tell Claude code to think, think hard, and ultra think.
08:03If you tell Claude Co. Just to think, it will give you pretty good results. But if you tell Claude Co.
08:08To ultra think, it will give you amazing results. So if you use this strategically and you use ultra think for your extra complex tasks, you give yourself a better chance of one shotting your code and getting way better results on your first try. Let me give you an example.
08:23So I'm gonna have Claude Code build me an ultra stylized first person shooter using three JS in a single file. This will just show us kinda baseline the way Claude code works on just normal think mode. I'm gonna hit enter on that.
08:35It's gonna build me a three d first person shooter. We'll see what this looks like on just normal think mode. And then right after, I'll show you what ultra think looks like to see how much better it performs.
08:45Alright. So it finished building on the regular think mode. Let's check this out.
08:48Pretty good app. Right? The enemies are these orange balls going around in this purple.
08:53I I like it. It's solid. I mean, it's kinda plain.
08:57Right? It's got the purple pillars and pink balls. It's fine.
09:01There's not much to it. The bolts are kinda cool, but let's see what this looks like now on ultra think mode. So I'm using the exact same prompt, except this time I'm gonna tell it to ultra think.
09:11So I'm gonna hit enter on this, and we're gonna see what kind of result it comes up with, this time using ultrathink mode, which makes it work even harder on our complex problems. And, again, all you need to do to use ultrathink mode is just tell Claude code ultrathink inside your prompt, and it's gonna give you way better results.
09:28Now this could take a little bit longer, so you don't wanna use UltraThink with every single prompt you give. But on those really strategic and complex tasks you wanna give Claude code, you wanna tell it UltraThink.
09:38So as you can already see here, it build out its to do list for what it's gonna do for this app, and there's, like, way more to do items. So it really is taking every step ultra carefully to make sure it gives us really amazing results. Alright.
09:50Looks like it's done here. Let's check this out to see what we got. As you can see, the world looks a a good a good bit cooler.
09:59You can see there's a lot of different other items inside of the map. Oh, you can see there's, like, walls too. You can see every time I fire, there's, like, sparks that come out of the gun.
10:09And the enemies, instead of just being a ball, have these, like, rotating three d shapes in them. And when I destroy them, they, like, explode, which is really cool. You can see there's an ammo counter in the bottom right.
10:20You can see there's a score in the top left. There's a lot more colors. There's a lot more stylized effects.
10:25This is the advantage of UltraThink. It just spends a lot more time giving you a lot better results. Anytime you do something complex in Claude code, you wanna make sure you tell Claude code to UltraThink about it so you get better code and better results.
10:38So those last four features are amazing and make your coding a lot better, but this last one I'm gonna show you is mandatory.
10:48Everyone should do this. This is so easy to implement, and this is having the correct Claude rules. So Claude rules are rules that are sent to the agent every time you run any prompt.
10:59It's included in the context every single time. So if you create a Claude dot m d file, you can put in whatever rules you want that are sent to the agent every time. These are the nine rules I use in my claw dot m d file with every single app I built, and this gives me so much better results inside a clawed code.
11:18I'm gonna paste these down below, so feel free pause here, grab them, and put them into your own Claude dot m d file. But just going over them real quick, basically, what they cover is making sure Claude thinks in tasks and to do lists. So every time it does anything, it breaks it down into do lists, which makes it so that you get way better results, and it also makes sure that it isn't lazy.
11:39So all fixes need to be validated in the code, and all fixes impact as little code as humanly possible so that you don't run into any of those dreaded error rabbit holes where you just make errors worse and worse and worse. These nine rules I have here made my results from Claude code so much better.
11:56And just by copying and pasting them into your own Claude dot m d file, you'll instantly get better results from Claude. So those five secret features have made my results in Claude code so much better and have made my workflow so much faster.
12:10Go back, rewind, make sure you copy and paste all the prompts from down below, implement them into your own Claude code today. I promise your results will be unbelievable. If you learned anything at all, leave a like.
12:21Make sure to subscribe and turn on notifications. All I do is make incredible videos on building things with AI. Also, I livestream every single Thursday at 4PM eastern, so check out that as well, and I'll see you in the next
The Hook
The bait, then the rug-pull.
Alex Finn opens with the oldest trick in the tutorial playbook — a bold claim followed by an implied insult — and then actually delivers. Five real Claude Code features, five copy-paste prompts, twelve minutes. The threat of leaving speed on the table is enough to keep you watching.
Named ideas worth stealing.
11:00list
Nine CLAUDE.md Rules
- Think through the problem first, read relevant files, write a plan to tasks/todo.md
- Plan should have a checklist of todo items to check off as you complete them
- Check in before starting work so the plan can be verified
- Work on todo items marking them complete as you go
- Give high-level explanation of every change made
- Keep every task and code change as simple as possible
- Add a review section to todo.md with a summary at the end
- Never be lazy — find root cause, never temporary fixes
- Make all fixes as simple as humanly possible — impact as little code as possible
Nine rules for a CLAUDE.md that enforce planning, minimal-footprint changes, and anti-laziness
Steal forJoeFlow CLAUDE.md, any Claude Code project — Joe already follows a Karpathy-derived version of this; this is the simplified YouTuber variant
07:46list
Think Mode Tiers
- think
- think hard
- ultrathink
Three escalating think modes built into Claude Code; ultrathink spends more compute for complex problems
Steal forAny Claude Code prompt engineering guide or JoeFlow tips content
How they asked for the click.
VERBAL ASK
12:03subscribe
“If you learned anything at all, leave a like. Make sure to subscribe and turn on notifications. Also, I livestream every single Thursday at 4PM eastern.”
Standard verbal CTA at end with no visual overlay. Adds livestream invite as a secondary hook.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
