Big Idea

The argument in one line.

Most Claude Code power users are operating at 20% capacity because the CLI's highest-leverage features — hooks, worktrees, teleport, and batch — are invisible until someone shows you where the switch is.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You use Claude Code daily but start every session the same way with no hooks, no agents, and no autopilot.
You want to run parallel Claude instances without the fear of two agents overwriting the same file.
You work across desktop and mobile and want seamless session continuity without saving anything manually.
You have heard terms like hooks, worktrees, or /btw and skipped past them assuming they were too technical.

SKIP IF…

You are new to Claude Code and still learning the basics — this video assumes you are already running sessions regularly.
You are looking for prompt-writing strategy or output quality tips, not CLI feature coverage.

TL;DR

The full version, fast.

Boris Cherny published a 15-feature thread that hit 2M views; this video is the practical companion. Features cluster into three areas: mobility (mobile app, teleport, remote control), automation (hooks, /loop, /schedule), and parallelism (worktrees and /batch for running 3-30 isolated agents simultaneously). The most underused feature is hooks — they inject context deterministically into every session whether Claude wants it or not, unlike CLAUDE.md which Claude can effectively ignore. The highest-leverage combo for power users is worktrees plus /batch, which eliminates file-conflict risk when running parallel agents at scale.

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 →

Chapters

Where the time goes.

00:00 – 00:29

01 · Intro

Context: Boris Cherny's viral tips thread. Promise: 15 features, practical demos.

00:44 – 01:58

02 · Mobile App

Code from iOS/Android via the Claude app. Plan mode on mobile, GitHub sync, push/pull workflow.

01:59 – 02:53

03 · Teleport

Resume unsaved mobile sessions on desktop with claude --teleport. Lists sessions by name.

02:53 – 04:04

04 · Remote Control

Monitor heavy desktop tasks from mobile. Name sessions for multi-task clarity.

04:04 – 05:53

05 · Autopilot: loop and schedule

Autopilot mode. /loop runs while terminal is open; /schedule executes in the cloud on a time interval.

05:53 – 08:17

06 · Hooks

Deterministic event-based rules in settings.json. Auto-inject brand guidelines on every session start.

08:17 – 09:54

07 · Cowork Dispatch

Full computer-use access. Claude can physically click extensions, open apps, reload browsers.

09:54 – 11:22

08 · Claude in Chrome

Visual inspection loop using your real browser. Claude writes code, checks it visually, fixes issues.

11:22 – 12:55

09 · slash config Settings

Toggle thinking mode, remote control, Claude in Chrome, teammate mode from one unified menu.

12:55 – 13:53

10 · Live Preview

Real-time HTML preview in Claude desktop app. No localhost required, changes visible as they happen.

13:53 – 15:35

11 · Session Forking with branch

Fork a session to explore an aggressive/experimental path. Original session ID remains resumable.

15:35 – 16:11

12 · Side Queries with btw

Fire a side query during an active task without flooding main context window.

16:11 – 18:27

13 · Git Worktrees

Three desks, three agents, zero file conflicts. claude -w syntax for named isolated lanes.

18:27 – 20:24

14 · batch Command

Spin up 5-30 parallel workers for breadth tasks. Token-heavy — use with caution on rate-limited accounts.

20:24 – 21:57

15 · bare Flag

Skip all CLAUDE.md and skills injection. Requires API key; subscription login returns not-logged-in error.

21:57 – 22:47

16 · add-dir for Multi-Repo Access

Bring external directories into session context. Useful for templates and brand assets across projects.

22:47 – 24:09

17 · Custom Agents

Name a markdown file in .claude/agents/, start session with personality and tools pre-loaded via --agent flag.

24:09 – 24:39

18 · Outro

Links to community and Claudemaxxing guide.

Atomic Insights

Lines worth screenshotting.

Hooks in settings.json are deterministic — Claude cannot skip them the way it can deprioritize a CLAUDE.md instruction.
claude --teleport resumes any unsaved mobile session on desktop; the session never needs to be explicitly saved first.
/loop runs autopilot as long as your terminal is open; /schedule runs in the cloud independent of your machine state.
Worktrees give each agent an isolated workspace — three desks, three agents, zero risk of two processes writing the same file.
/batch can spin up 30 parallel workers but burns tokens at the same rate — it is breadth over depth, not free parallelism.
Claude in Chrome uses your existing authenticated browser session, so there is no re-authentication overhead unlike headless agent browsers.
--bare strips all CLAUDE.md and skills injection but requires an API key; subscription logins return not logged in errors.
--add-dir pulls external directories like brand assets or templates into session context without moving any files.
Custom agents are just markdown files in .claude/agents/ — the filename becomes the agent name, invoked with claude --agent name.
/btw runs a side query during an active task without polluting the main context window.
/branch forks the current conversation to explore a risky path; the original session ID stays alive and resumable anytime.
Naming a remote-control session prevents confusion when monitoring multiple concurrent tasks from mobile.
Hooks can be configured by tagging the claude-code-guide agent which writes correct settings.json syntax so you never touch JSON manually.

Takeaway

Fifteen features, three categories, one upgrade path.

WHAT TO LEARN

The highest-leverage Claude Code features are invisible by default — and knowing which category a feature belongs to tells you exactly when to reach for it.

Hooks are the most underused feature: they inject context deterministically into every session, unlike CLAUDE.md which Claude can deprioritize when the context window fills.
Session mobility features (teleport, remote control, mobile app) remove the constraint that serious Claude Code work requires being at a desktop — the session follows you.
/loop and /schedule create genuine autopilot: Claude re-executes a task on your specified interval, with /loop tied to terminal uptime and /schedule running in the cloud independently.
Worktrees eliminate the file-conflict risk that makes most people hesitant to run parallel agents — each agent gets an isolated workspace with no shared files.
/batch scales breadth tasks to dozens of parallel workers but at full token cost per worker; it is the right tool for update-all-files tasks, not deep single-task reasoning.
/branch lets you explore a risky experimental path without contaminating the main context window — the original session ID stays alive for resumption.
/btw allows mid-task side queries without interrupting or flooding the active session, solving the problem of needing quick context while something longer is running.
--add-dir is the cleanest way to bring shared assets like brand guidelines or templates into any project session without restructuring your file system.
Custom agents are just named markdown files — the filename becomes the CLI invocation, and the file defines personality, tools, and instructions injected at session start.

Glossary

Terms worth knowing.

Teleport: A Claude Code CLI command that lists and resumes existing unsaved sessions from any machine, enabling seamless handoff between mobile and desktop.
Remote Control: A /remote-control command that exposes a running desktop Claude Code session to the mobile app, letting you monitor and interact with a heavy task from your phone.
Hooks: Event-driven rules stored in .claude/settings.json that fire automatically at session start, before/after tool use, or after edits — guaranteed to execute regardless of Claude's context decisions.
Cowork Dispatch: A Claude Code desktop feature that grants full computer-use access to Claude, allowing it to physically click, open apps, and interact with the entire OS rather than just the terminal.
Git Worktrees: Isolated working directories within a single git repo, used here to give each Claude agent its own file sandbox so multiple agents can run in parallel without overwriting each other.
/batch: A Claude Code slash command that spins up multiple parallel worker agents to execute a breadth task like updating pricing across 20 proposal files simultaneously.
--bare flag: A CLI flag that strips all CLAUDE.md, skills, and injected context from a Claude Code session — currently requires API key authentication to function.
--add-dir: A CLI flag that adds one or more external directories to the session context without requiring those files to be inside the current project folder.
/btw: A slash command that fires a side query to Claude during an active running task, returning an answer without interrupting or merging into the main session context.
/branch: A slash command that forks the current Claude Code conversation into a new branch, letting you explore an alternate path while preserving the original session ID for resumption.

Resources

Things they pointed at.

00:00linkBoris Cherny tips thread (X/Twitter)

00:44productClaude Code mobile app (iOS/Android) ↗

09:54toolClaude in Chrome extension

24:09linkMark Kashef early adopters community ↗

24:09productClaudemaxxing Hidden Features Guide ↗

Quotables

Lines you could clip.

08:08

“This is deterministic. This will get shoved into that context window whether Claude Code wants it or not.”

Crisp contrast between hooks (guaranteed) and CLAUDE.md (optional). Standalone, no context needed.→ TikTok hook↗ Tweet quote

10:18

“Would you trust a chef who can't taste their own food?”

Strong analogy for Claude visually inspecting its own output. Memorable, repeatable.→ IG reel cold open↗ Tweet quote

20:10

“If you are rich in tokens, feel free to go for this. But if you are erring on the side of caution, you're being frugal, I would avoid this — for now.”

Honest, useful caveat on /batch. Practical financial framing.→ newsletter pull-quote↗ Tweet quote

The Script

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.

metaphoranalogy

00:00So Boris Journey, the guy who actually built Claude Code, just dropped his latest tips and tricks on how to get the most out of it. And given that it's gotten more than a million views in the last twenty four hours, I went through each and every feature that he mentioned. So in this video, I'm gonna walk you through 15 different features that you may or may not know how to use or not use to the full potential.

00:19But I'm not just going to conceptually walk you through these concepts, but actually show you how you can practically apply them. So if you watch this video till the end, you'll be one step closer to mastering Cloud Code. Let's get into it.

00:29So from these 15 core features and concepts, I'm gonna breeze past the ones that are pretty straightforward and double click into the ones that are a little bit more nuanced.

00:44So many of you might know you can access Claude code on your phone, but most people don't know that they're constantly trying to achieve feature parity. So eventually, you can run the majority of Cloud Code using a terminal through the Cloud Code app.

00:57And despite building my own personal assistant I called Cloud Claw that I showed on this channel a few weeks ago, I still use the Cloud Code app on my mobile because it's helpful to see things like bash, the actual code executions. If you're using something like an OpenClaw or something derivative of that, if you can't see what's happening and you're doing very code intensive tasks, even if you're not the one coding, then you're missing out on some of the transparency.

01:21A couple key things to note is that to make the use of this app, then you'd want to create a GitHub repository that you log into so that all your code lives in one place. So you can push it from one end and then pull it from the other.

01:32And when it comes to clicking on this code tab at the bottom left hand side of the mobile, you can actually go and click on plan mode. So you can do a lot of the planning on the go and maybe push that planning to your GitHub repo, then pull that from an existing session on your desktop and continue executing from there.

01:54So tip number two is one thing that I have underused greatly, which is teleport. And the whole point of teleport is that you can have an existing unsaved session on your cloud code on your app that you can just resume from a desktop computer. So let's say I executed this prompt from my mobile device and I just said read the following website and make the changes to the hero section, make sure you have a better call to action, etcetera.

02:17Let's say I am resuming this session and I'm getting home and I wanna hop onto my laptop, all I have to do is nothing from the app itself. I just have to go and write claud dash dash teleport and it'll be able to look at my existing sessions in the cloud.

02:31And you can see right here, two minutes ago, update hero section for spring promotion. And if we do enter right here, we'll be able to go through and this was the exact conversation that we had on the mobile device.

02:48Now tip three is very much related to teleport, but instead of going from your mobile device to your desktop, you can go from your desktop to your mobile device. So how you would do that is if you go to your terminal and you do slash remote control, this is another feature that I discounted.

03:03But now that I understand where and when to use different types of tools, like when to use dispatch, when to use remote control, when to use my own configuration on my telegram, now I understand that when it comes to monitoring a task that's a little heavier, you wanna be able to navigate and see what are all the logs and the details happening right now.

03:21Remote control is an awesome way to take an existing terminal where all you wanna do is narrowly focus on that task and have it on the go. But one additional nugget from Boris is beyond remote control on its own, you can actually put in quotes the name of the session because usually if I pull up my mobile phone right here, you can see that some of these are basically the name of the task or the first thing that's being asked of that chat.

03:43So if I move my phone to the right for a second, you'll see that right here, I've called this session website redesign, and we have the exact same thing popping up on mobile. Otherwise, we'd get confused.

03:53Let's say you had five or six different remote control sessions. Now you can name each one something you can always remember, especially if you're busy on the go. Now the next tip is definitely more powerful, and this is how Boris and his team runs Cloud Code on what's called autopilot mode, where they take advantage of two newer features.

04:11One of them is called slash loop and the other is called slash schedule. The core difference is slash loop works as long as your terminal session and your computer is on, whereas schedule can be written in the cloud and executed in the cloud. But the core idea is this, you can execute some form of request like check my inbox, and then every five, ten, thirty minutes, you just allocate what is the time interval to re execute said task.

04:36So Boris is using it to babysit his code in his own words every five, ten minutes. So if we pop into the terminal, let's say we have a series of hypothetical files here. So we have brand assets, we have client documents, we have proposals, we have reports, we have templates.

04:51All of them kind of feed into this demo. And let's say our first prompt was the following. Read through all the client folders in clients and give me a quick summary where each project stands.

05:00So this will walk through, and then we can do the following. For this one, we could do slash loop, and right after slash loop, the syntax is to put the time interval as to when it should execute. Check my email inbox.

05:12Draft professional responses for anything that needs a reply. Save drafts, but don't send anything. And if we execute this, now this will naturally try to invoke a series of skills on my computer, but so long as this terminal session and my computer are open, it should be able to go on this autopilot mode.

05:29And you can see right here, it invokes something called cron create. This will go and execute this every thirty minutes like we mentioned. This is the job ID.

05:37This is the interval. This is the task, and this is when it auto expires unless you cancel it sooner.

05:48Now next up are one of my favorite features of Cloud Code and the ones that I mainly teach in all the workshops we do at different companies, which are hooks. And if you still don't understand the concept of hooks, you can think of it as a series of rules that automatically fire based on an event.

06:03If you want it layman's terms, then imagine you have your phone. Let's say you have an iPhone and before you go to bed, it automatically goes into sleep mode or do not disturb. Technically, that's a hook.

06:13That's a hook based on a time. Let's say 10PM is your bedtime or 11:30 then that would fire and then execute your phone to not have notifications during that sleeping time. And then in the morning when you wake up and it sees that you're actually using your phone, you're sending messages, it asks you do you want me to remove this mode and go into the normal mode where we turn off your alarm and everything else is just normal.

06:34Hooks work in exactly the same way. So you can attach a hook to a session start before any particular action or tool use or after a tool is used after any or every edit, or after a session even ends. So you have so many checkpoints where you can attach a hook.

06:50So one example of a practical use case of a hook is most people rely on something like a Claude MD file to auto inject all of their preferences. And you can use it to say always learn and understand our brand guidelines before we create any new asset. But another way you can do this is to attach a hook.

07:07And the way you can attach it is say, set up a hook so that every time I start a new session in this project, you automatically read the brand guidelines. So you're forcing the tool calls of Claude Code to read this and have it in context each and every time. And then you can also do the following where you tag the Claude code guide agent.

07:26This is my favorite thing to do from a syntax standpoint because at the end of the day, a hook is just a series of code in what's called a JSON file. It's called a settings dot JSON file. In case you don't wanna get your hands dirty, you're nontechnical, then you can tag this specific agent and it will help you do it with the best syntax.

07:44And you could see it writes this hook to your settings file directly and then the next time you have a session start, then it should be able to fire said hook and make sure that auto injects the same document every single time. And the core difference between this versus relying on Claude code with CloudMD is that this is deterministic.

08:02This will get shoved into that context window whether Claude code wants it or not.

08:12Now the next one is very straightforward and it's something you probably have heard of because there was so much coverage about it which is co work dispatch. And it's essentially the quote unquote open claw killer wink wink nudge nudge more so for enterprise and SMBs where they don't want to get their hands dirty and all they want is an experience of having remote control access to your desktop.

08:32So the way you set it up is unbelievably straight straightforward. All you have to do is download the Claude code desktop app, and then once you navigate to dispatch right here, you'll be able to whitelist a series of access requests. And once you whitelist everything that's needed, you can always toggle on things like keep awake and enable computer use.

08:49And beyond that initially, I was confused where to derive the most value out of this. And yesterday, I was battle testing a Chrome extension that I built for my community and I realized that if you something like Claude in Chrome, let's say we actually go into Chrome and this is the Chrome extension I put together, it couldn't actually physically click on the extension and execute requests.

09:09I had to basically click on it myself or have it go and access things through the network tab because Claude in Chrome can access everything within the browser right here. But using Claude dispatch, now technically, it could take control of my computer in full, then open the browser, then physically click on this.

09:29So if I ever wanted it to actually reload the extension and keep testing itself, I would need to use something like a dispatch to physically take control of every single part of my browser. Now there are probably even more use cases than this, but this is an example of one that I found actually useful.

09:49Now the next one is something that I personally use every single day when it comes to validating anything that needs a visual inspection layer. So instead of just checking whatever Claude code builds yourself on, let's say, local host or an HTML for website, you can just tell it recursively. Go and use the MCP Claude in Chrome to go and review your settings and review anything that you built and make sure that it lines up with or our specifications.

10:16So the analogy here is would you trust a chef who can't taste their own food? So the goal is you have Claude code right. It will then visually inspect its own creation, and then once it checks it through the Claude in Chrome extension, all you have to do is download this and then log in through your Claude code account, then you have a very systematic process where Claude code writes the code then repeats until it thinks it matches the requirements, then it can actually go and check it and see the result, fix any issues, and go into this recursive loop where typically it's not endless.

10:48Cloud code is somewhat arrogant where sometimes they will think that it's done, but it's done the 80 of the eighty twenty. So you might pop in and say, you tested eight out of 10 things, but you didn't actually test it deeply. I want you to take this specific user journey or assume that someone's clicking into each and every nook and cranny of our app, go and battle test that yourself.

11:08And one extra bonus nugget here if you wanna make sure you're always enabling cloud in Chrome as well as a series of other things, you could do slash config. Once you do that, you will see all of these settings and whether or not they are set to true, whether you wanna do auto compaction in your sessions, whether you wanna be able to go into thinking mode by default, everything.

11:28And if you go to the very bottom here, you will see things at the very, very bottom, like enable control for all sessions as well as cloud and chrome enabled by default. So you'll see here I have it set to true.

11:40So whenever it makes logical sense for it to invoke and use said tool, then it always will. Whereas with something like enable remote control, right now I have this on default. So if I set it to true, then every single session technically that I spin up will be something that I'll be able to port into my Claude mobile phone.

11:58Now one expert tip on this from yours truly is that if you have multiple skills or MCP servers that know how to use browsers in different ways and there could be some overlap on which one fires when you ask for a task that involves this, you can call it out specifically. So you can say use Cloud and Chrome for a very particular task and then it will know 100% of the time to use it.

12:20And I find it personally useful for logging into pre authenticated apps or services. So if I wanted to log into my Gmail and you use, let's say, agent browser skill from Vercel, you'd have to physically go into the new browser it spins up, authenticate yourself every single time, and then that adds some more burden. So Cloud and Chrome is great for quick and dirty because it's using your actual browser, just creating a separate group within it.

12:50So the next feature is also very straightforward and it's also back in the Claude desktop app. And instead of Claude Cowork or Claude Code, this one works primarily in Claude chat where you have a chat on the left hand side and instead of just giving you some form of HTML file for you to render on your own, you can actually get a preview.

13:08So if we pop into the Claude desktop app, you'll see here I've asked for a brand new landing page. Instead of just cranking out a file, then we can actually click on this sidebar right here. When we click on the sidebar, we can click on this file.

13:22We'll be able to get a live preview of the site. And the cool thing is given the name, live preview, as you're making changes, you can actually see it changing in real time. So if you want the light version of seeing everything and you don't feel like opening up Cloud Code, asking it to spin up a local host on your computer, then to preview it, and you just wanna draft something, then this is an amazing way with the path of least resistance.

13:50Next up is a feature that I've also reviewed on this channel called slash branch, and it comes very handy when you have a little fork in the road where you wanna decide do I go path a or path b. But you don't wanna risk path a in case that is the right path, but you feel like physically side questing to investigate an additional idea or a different take.

14:09So to put this into action, let's say I have the following prompt, and I say something like, I need your help to refine a proposal for a specific client, and these are the addresses of the files. I break down what the situation is to give it context. I tell exactly what I want the proposal to look like for what denomination and a series of criteria here on how it should rewrite it.

14:30So let's assume it executed that, and I ran this before this video just so that you wouldn't have to watch me watch Cloud Code crank it out. And let's assume we get some version, and this goes through the current state of things. But let's say I wanna experiment with really aggressive pricing, but I don't want to flood that in my context window and have the language model start getting con confused between our current rate versus this very aggressive different possible reality.

14:55So in order for us to side quest, we could do something like branch very aggressive pricing versions.

15:05So now we're in a separate branch where I can ask for the exact same proposals with very different structures of pricing. And the good part of this is if I review this and it looks like this is the way to go, I can now purely go down this path and leave our old conversation in a way where as you can see here, we can always resume it by entering Claude dash r with this ID in any terminal.

15:31The next one is the simple, useful, and very applicable slash by the way, and this one doesn't need too much information. All we have to do is enter some form of prompt. And while said prompt is executing, we can just do slash by the way, what are our numbers for q one?

15:49And then this allows us to multitask in a way where where we're not flooding the context window of the main session, and we can just get all the information as we go. We can click on space and keep the conversation flowing as we need it to.

16:07Now this next tip is typically one that nontechnical people run away from because it sounds very intimidating, but conceptually it's actually unbelievably simple, which are using work trees and how to actually invoke them. Now work trees, if the analogy of a tree isn't helpful, you can think of it as having three separate desks with three separate agents with three separate types of work and documents.

16:30The goal is each one can actually work mutually exclusively without having a risk that multiple agents are working on the exact same file or trying to edit or write the same file at the same time. So if we think of it as different lanes with different desks with different agents and different work, then lane one could be writing the client proposal, lane two could be website redesign, and lane three could be financial reporting where all of these files shouldn't have any form of overlap.

16:58Now one thing to talk about it and another thing to show it. So the syntax is the following. You would write Claude.

17:04You don't need this part. I'm just running this in YOLO mode, and then you do dash w. This stands for work tree.

17:10This tells Claude that you are executing a separate desk with a separate set of work. And in this case, we're just basically assigning it what this work tree is about. So in this case, it's about proposal for Acme company.

17:22And beyond that, there's really not much to it. You just prompt as you normally would. So in this case, I would just say write a revised proposal for Acme Corp based on their project notes in this folder.

17:32They're now in phase two. Propose an upsell to enterprise tier for phase three. And again, exact same concept for another task.

17:39So you'd enter the exact same syntax right here. You do the dash w for a brand new desk, work tree, whatever concept makes sense, and now you have a separate lane called website update. And this will be designated to adding testimonials to your website at the very bottom using real client names from our client folder.

17:57And then while this is running, you can also run this one, is dash w quarterly report. Read the reports from this folder and create a professional q one board report that looks at x y and z and then execute it. So if you've historically been intimidated by this concept, be intimidated no more, really easy to execute and very useful.

18:23So this next step should be taken with a gigantic grain of salt and a lot of caution, especially if you're very token conscious of your rate limits and everything that's now happening with all these different usage capacities. So if work trees can spin up three clouds at once at different desks, technically batch can spin up five, ten, fifteen, thirty at once.

18:43So slash batch has a higher emphasis on breadth of execution versus depth and thoughtfulness of how to derive the most of each singular agent. So the way it works is if we pop into the terminal, you can open a session and do the following.

18:57Do slash batch, then after batch, you have to enter some form of request. You could say update the hourly rate from one fifty to one seventy five across all proposal files in the proposals folder, keep existing formatting, recalculate all line items totals, and the grand total in each proposal.

19:15So if we send this over, I will show you what happens. So now it enters plan mode, and in plan mode, it researches and sees basically what is the scope of this task.

19:25So now it's going through each and every possible proposal that we have in our workspace. Once it finishes its plan, it goes through and it breaks down all the context, all the formatting conventions that need to change, what's gonna change in each and every proposal, so it has a draft of that.

19:41And then at the very bottom, it walks through the worker instructions template. Each worker that it will spin up receives the specific file, the exact old to new values for each line item, and the new grand total.

19:53So now if you go on to YOLO mode and execute it, it will employ as many agents as it wants as if it's venture capital funded and it can hire whoever, whenever to execute the task. So if you are rich in tokens, feel free to go for this. But if you are erring on the side of caution, you're being frugal, I would avoid this comma for now.

20:20Now the next recommendation is one that I barely use, and I don't really see many use cases at least for myself to use. But just in case it's useful for you, it's basically adding what's called dash dash bear. And dash dash bear will allow you to run a command without having Claude always inject the Claude MD, the skills, all of the additional burden and bloat that comes into each and every session.

20:42When you add dash Dash bear, it will circumvent all of that and just execute the request you have in seconds. So just in case you thought I meant bear the animal, you would actually write it Claude DashDashBear then execute some form of command.

20:56Now one thing I noticed in trying to use this myself is if you're using your normal max account or your subscription, then running this won't work because it says not logged in. So it appears, correct me if I'm wrong, if you know otherwise in the comments, but you have to use this via an API request.

21:12And then once you have the API, then you should be able to run it. So I tried multiple different ways to log in, exit my own terminal, but the core idea is you would send something like this.

21:23So this is a shortcut that does work, that's dash p, and you ask the question and you get the answer immediately. But if you use this dash p alone without the dash dash bear, you're not getting technically the benefits of not having any form of bloat or burden in the conversation. So you might have a use case on your end where this makes sense, but for myself, I'm okay with having the clot MD loaded in for a basic question versus having to wrangle this myself.

21:53For the penultimate tip, this one's also very straightforward. This is purely a syntax play. So instead of just giving Claude code access to one singular project, you can also give it one basic command, I'll show you right now, that will allow it to also shop different projects and different resources in those projects.

22:10So if we pop into the terminal and you write something like Claude and then you say dash dash add dir, in this case, directory, and then templates, and then I do another one which is adding brand assets, I'm technically adding multiple different directories within my session.

22:26So that's where all of these come from, where I added the templates folder right here as well as the brand assets. These were external for my project, but now with this command, I'm bringing them into context.

22:43And the last core tip was being able to call an agent by name at the very beginning of your Claude code session. So instead of starting a Claude code session, then asking Claude to go and remember how to use a specific agent, you start the session with the agent already injected in context.

22:59So if that doesn't make sense, you basically start a session with your agent already auto injected with all the personalities, the quirks, the rules, the regulations, and the tools that it has access to.

23:10So if you watch my last video, I spoke at length at the power of this dot Claude folder and every permutation and everything inside that makes it tick. One of the things that makes it tick is this agents folder where you can write a markdown file and whatever the name of the markdown file is technically the name of the agent.

23:29So if I take a click on this, this is the name which is called proposal writer. This is the description, and these are the tools that this agent can invoke.

23:37So if we go down, these are the structure of the commands, and you can start a session by saying Claude dash dash agent proposal writer.

23:45So you can have it by name, and when it loads it up and you say something like write a proposal for tech corp, it will execute this entire structure using all the rules and the paradigms and a description with this specific agent.

23:58Now I know that was a lot, but I did my best to be as concise as I possibly could to show you the concept as well as the application. And don't worry if you wanna be able to access all the different prompts that I showed you, the demos, the actual shortcut commands, I'll make those all available to you in the second link in the description below.

24:15And if you found this video helpful and you learned something that I'd super appreciate if you left a like, a comment, and a sub on the channel. It really means a lot to me, and it really helps you grow the channel. And to anyone else that wants to go much deeper into Claude Code and be in a community with tons of folks, tons of coaches, my own wisdom, and exclusive content, then check out the first link in the description below, and maybe I'll see you in my early adopters community.

24:38I'll see you in the

The Hook

The bait, then the rug-pull.

When the person who built the tool tells you you are leaving 80 percent of it on the table, you pay attention. Boris Cherny's 15-feature tips thread hit two million views in 24 hours — and Mark Kashef spent the next day turning every one into a hands-on demo.

Frameworks