Modern Creator
Kallaway · YouTube

5 Claude Skills To Grow Faster Than 99% of People on Social Media

A creator with a million followers maps out five reusable Claude Cowork workflows for content strategy — channel analysis, audience targeting, outlier tracking, breakout detection, and hook generation — all built on a persistent context-folder system and the Sandcastles MCP.

Posted
yesterday
Duration
Format
Tutorial
educational
Views
5.6K
463 likes
Big Idea

The argument in one line.

Claude Cowork becomes a genuine content-strategy engine once it has persistent memory about you and a live data feed from social platforms, turning five specific, reusable skills into an automated research pipeline instead of one-off prompting.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…
  • You run a personal brand or business content channel and want a repeatable system for deciding what to post next.
  • You already use Claude for writing or research and want to give it durable memory instead of re-explaining context every chat.
  • You're comfortable connecting a paid third-party data tool (an MCP) to Claude to pull real social metrics.
  • You want to track competitors or your own channel for outlier performance without manually scrolling social media.
SKIP IF…
  • You don't post on social media or run channel-level content strategy — this is a research pipeline, not a writing tool.
  • You're not willing to pay for or set up a third-party analytics MCP; several of the five workflows depend on it.
  • You want a five-minute tip, not a folder-structure-plus-tooling system that takes real setup time.
TL;DR

The full version, fast.

Before using Claude for content strategy, set up a persistent context folder (four markdown files describing your voice, work, identity, and instructions) and connect the Sandcastles MCP so Claude can pull real social-platform data. On top of that foundation sit five packaged workflows: deep-analyze any channel's winning topics, hooks, and formats; build a five-ring audience bullseye to plan a 3-2-2 content mix; get an automated daily feed of outlier competitor videos; find the exact video where a creator's growth spiked; and extract a channel's proven hooks into a rubric that grades and rewrites new ones. Several of the workflows don't require paid analytics credits at all, and most can be scheduled to run automatically.

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.

00:0000:39

01 · Cold open

States the promise up front: five reusable Claude workflows for growing faster on social media, framed explicitly as designed skills rather than generic AI output. Sponsor (Anthropic) disclosed.

00:3907:31

02 · Setting up Cowork for success

Three setup steps before any workflow runs: build a persistent context-folder system (a name-Cowork folder with context and work subfolders, plus four markdown files an AI-generated interview skill can fill in for you), optionally add a voice-to-text tool since talking is faster than typing, and install the Sandcastles analytics MCP that pipes real social-platform data into Claude.

07:3111:33

03 · Workflow 1: Deep Channel Analysis

Analyze a channel's videos (your own or a competitor's) inside the analytics tool first, then hand the processed data to Claude to produce a visual breakdown of winning topics, hooks, formats, and non-obvious patterns.

11:3313:54

04 · Workflow 2: Audience Bullseye Builder

A skill that interviews you about your niche and builds a five-ring concentric bullseye from narrow to broad, recommending a 3-2-2 content mix and explaining the audience size at each ring.

13:5417:03

05 · Workflow 3: Outlier Video Pulse

An automated feed that pulls competitor videos above a view-count and outlier-score threshold from a tracked watch list, refreshing on a schedule and optionally posting into a chat tool, workspace app, or standalone webpage — usable without spending analytics credits.

17:0318:54

06 · Workflow 4: Creator Breakout Detector

Scans a creator's upload history (up to two years by default) to pinpoint the exact video where their growth curve spiked, so you can study what specifically changed at that moment.

18:5421:06

07 · Workflow 5: Hook Machine

Extracts a channel's actual written hooks, clusters them into a rubric of storytelling rules, combines it with general hook-writing principles, then uses the rubric to grade and generate new hooks for any topic — including rewriting a hook you already have.

21:0622:11

08 · Other content MCPs

Three additional connectors mentioned but not deeply used: a code-based video editing tool, an AI visual-generation tool, and an email-platform connector for writing and analyzing newsletters from within Claude.

22:1123:14

09 · Summary and sign-off

Recaps that the five workflows are meant to save research time rather than generate slop, teases a follow-up video on ideation and scriptwriting skills, and points to the free skill files linked in the description.

Atomic Insights

Lines worth screenshotting.

  • A four-file context folder (how-i-talk.md, how-you-work.md, who-i-am.md, and an instructions file) gives an AI assistant persistent memory about voice, workflow, and identity across every new chat.
  • Talking is roughly four times faster than typing, which is the practical case for pairing a voice-to-text tool with an AI writing assistant.
  • A recommended content mix across a five-ring audience bullseye is 3-2-2: three videos at the exact niche center, two at the next ring out, two at the ring after that, and zero at the broadest, least-qualified ring.
  • Bulk-analyzing videos inside a dedicated analytics tool first, then handing the already-processed data to an AI assistant for synthesis, is faster than running the analysis through the AI's own data connector.
  • An automated outlier-alert workflow can filter to only videos above a view and outlier-score threshold from a tracked competitor list, refresh on a schedule, and post results into a chat, a workspace tool, or a dedicated webpage.
  • Reverse-engineering a creator's breakout moment means scanning their upload history for the exact video where their performance curve visibly changed, then studying what was different about that one video.
  • A hook-writing system built from real data works in three stages: extract proven hooks from a channel, cluster them into a written rubric of rules, then use that rubric to grade and generate new hooks against a new topic.
  • Several of these research workflows deliberately avoid spending paid analytics credits by pulling only surface-level public metrics, reserving the paid per-video analysis for cases where deeper detail is worth the cost.
  • None of the five workflows generate finished content directly — they compress the time spent manually researching what performs, freeing time for higher-leverage work like actually making videos.
Takeaway

A durable memory system plus live data turns an AI assistant into a content-research engine.

WHAT TO LEARN

The five workflows only work because of what's underneath them: a persistent context folder and a live social-data connector, without which each one collapses back into generic prompting.

02Setting up Cowork for success
  • A four-file context folder (voice, work preferences, personal background, instructions) gives an AI assistant durable memory so it doesn't need to be re-briefed every session.
  • Voice-to-text tools are worth adding because talking is roughly four times faster than typing when feeding an assistant information.
  • A live data connector (an MCP) is what separates real analysis from guesswork — without it, the assistant only has whatever's in the chat.
03Workflow 1: Deep Channel Analysis
  • Bulk-processing data inside a dedicated tool first, then handing it to the assistant for synthesis, is faster than routing everything through the assistant's own connector.
  • Analyzing a full channel — winners and losers both — surfaces which patterns are actually tied to performance, not just present in every video.
04Workflow 2: Audience Bullseye Builder
  • A five-ring targeting model with a fixed content-mix ratio (3-2-2) turns a vague sense of niche vs. broad into a concrete publishing rule.
  • The broadest possible ring is deliberately excluded from the mix because it attracts high volume but low-quality, unqualified attention.
05Workflow 3: Outlier Video Pulse
  • Automating outlier alerts on a schedule replaces manual scrolling with a passive, always-on research feed.
  • The same workflow can run without spending any paid analytics credits by pulling only surface-level public metrics.
06Workflow 4: Creator Breakout Detector
  • Reverse-engineering a creator's breakout means finding the one video where the growth curve visibly changed and studying what was different about it, not the whole channel evenly.
  • Scanning a longer history (up to two years) matters because breakout moments can happen well before an account looks established.
07Workflow 5: Hook Machine
  • A hook-writing system that starts from real, already-proven hooks and turns them into a rubric produces more defensible output than asking an assistant to invent hooks from nothing.
  • The same rubric can grade a hook you already wrote and suggest higher-scoring alternatives, not just generate new ones from scratch.
08Other content MCPs
  • The same connector pattern used for social data extends to other parts of the content stack — video editing, visual generation, and email — each as its own live data or execution layer.
Glossary

Terms worth knowing.

MCP
A connector that lets an AI assistant read live data from an outside tool or platform, instead of the assistant only knowing what's in the chat.
Outlier score
A relative performance metric analytics tools use to flag a video that's dramatically outperforming a channel's or niche's typical view count.
Context folder
A set of reference documents (voice, work preferences, personal background, instructions) an AI assistant reads before every task so it doesn't need to be re-briefed each time.
Deep analyze
A per-video analytics process that extracts topic, hook, format, and pattern data from a video beyond its raw view/like counts, typically at a per-video cost.
Audience bullseye
A model for content targeting using concentric rings from a narrow niche center out to broad, general topics, used to decide what mix of specific-vs-broad content to publish.
Resources

Things they pointed at.

Quotables

Lines you could clip.

00:12
These are not AI slot machines. These are specifically designed Claude skills that make it way faster to improve your content strategy and make more winning videos.
sets up the whole video's contrarian framing against generic AI contentTikTok hook↗ Tweet quote
05:41
Talking is four times faster than typing, and if you ever typed and kind of tripped over your own fingers, this gets you out of the way.
tight, standalone productivity claimIG reel cold open↗ Tweet quote
13:02
I recommend most people start with a three two two. Three videos in the bull's eye, two videos between level two and three, and two videos at level four.
concrete, actionable ratio a reader can apply immediatelynewsletter pull-quote↗ Tweet quote
16:21
You can brick your phone and not watch social media... you don't need to scroll at all, but you don't miss anything.
punchy resolution to the outlier-tracking workflowTikTok hook↗ Tweet quote
20:07
Take data, build rubric, use rubric to generate and revise hooks that are given.
a three-step system stated as a one-line formulanewsletter 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
00:00In this video, I'm gonna break down five Claude based workflows that'll help you grow significantly faster on social media. And these are not AI slot machines. These are specifically designed Claude skills that make it way faster to improve your content strategy and make more winning videos.
00:16Now if you don't know me, my name is Callaway. I have a million followers, I've done billions of views, and I've built my personal brand across two different spaces, content strategy and AI.
00:25So this is the exact intersection I spend literally all day working on. If you're watching this and you know you should be using AI more, but you haven't quite figured out exactly how to make it work for content, you're gonna love this video. And thank you to Anthropic for sponsoring this one.
00:39Now before we dive into the five workflows, I wanna just start with the setup. If you wanna give Claude superpowers when it comes to making content faster, there are three things I recommend every single person do. The first one is to set up your folder structure and about me context files.
00:54If you've already done this in the past, go ahead and skip this section and go to the next one. But if you haven't done this, this one thing will be a huge game changer for you. The about me files and the folder structure that I'm about to go through will allow Claude to access much deeper context about you that it can automatically reference every time you work with it.
01:10This essentially gives it personalized memory. So here's how to set it up. First, make sure you're on the Claude desktop app.
01:17If you're using something like Google Chrome and you have Claude in the browser, you're not gonna be able to access co work, which is what we're gonna use for the rest of this video. So make sure you're on the desktop app. The link to download it is, of course, free, and it's below in the description.
01:28Now once you open the Claude desktop app, look in the upper left, and you'll see Cowork, the button in the middle of those three toggles. Click on Cowork. When you click on Cowork and you start a new chat, you will see a button that says work in a project or folder underneath the chat window.
01:43And this is where you can point co work to work inside of or to read preexisting context about whatever you want to work on. And having this point to a folder with context about you makes Cowork an absolute machine. So here's the folder structure you wanna set up on your computer one time so that you can always reference this whenever you work on your computer and create a new folder.
02:04You wanna name it your name dash co work. So for me, mine would be Callaway dash co work. Now inside of that folder, you're gonna make two sub folders.
02:12One called context and the other one called work. Inside of the work sub folder, you can make more sub folders mapping to every single project or work stream that you work on. So for me, I might have a folder called YouTube, and then one called short form, and then one called SFA, and on and on.
02:27Within each one of those work stream folders, I like to build five sub folders, and it's the same five in all of the different work streams. Archive, assets, references, skills, and videos.
02:38And this just helps give an organizational context so that Claude can put certain things where they fit in the folders. The skills folder in the YouTube folder for example is where Claude will house skills that it makes for me related to YouTube tasks.
02:51So just get that folder structure set up as I described, and don't worry about putting anything in there right now. Just get the folder scaffolding done. Okay.
02:58Now inside of your context subfolder, you're eventually gonna have four markdown files. And again, don't worry about making these. We will have Claude Cowork make these for us in a second.
03:07Let me just explain what they'll be. The files are gonna be how I talk dot m d, how you work dot m d, who I am dot m d, and then Claude dot m d. How I talk is a file where you can help describe to Claude how you want it to talk to you and how you want it to write for you.
03:23So think of it like a fingerprint for your own voice and writing style, but it's also an SOP to tell it exactly how to communicate with you. For example, I built in mind that I wanted to share different concepts in metaphors because that's just how my brain works. So now every time it talks to me, it kind of shares metaphors.
03:39The how you work file is gonna break out specific preferences for how you actually work within your business. So these could be things like the flows across your organization, the times you work, the people in the org, who owns what, different things like that.
03:52The who I am file, pretty obvious, is giving context and background about you, your work, but also your personal life if you choose to share that. This could be things like all the projects you're working on, your goals, the current status against those goals, what you're struggling with, everything that you want it to have context on about you.
04:08Lastly, the Claude dot m d file is just a file co work will look at first to get instructions for what to read. So Claude dot m d will tell it read those three things before you start working to get context. Now once you have these four files built, Claude becomes so much more powerful because it has consistent context and memory about why you're doing what you're doing.
04:26And this also improves its writing capabilities quite a bit, especially if you specify how you like and don't like certain things written. So before you do anything else, before you work anymore in Claude, I recommend making these files because it will make things so much better for you. Now to make this process easier to make those files, I built a Claude skill and it's free, where it will interview you and ask you a bunch of questions and use your answers to make those files for you.
04:49So I've linked that skill that will interview you to make these markdown files down below. So using that, just open a new chat and Claude drop that skill in and then just say run process. It will interview you for ten to fifteen minutes and it will spit out those three markdown files as well as claude.md.
05:05You wanna drag those into your context folder, so all four in there, and then we can start working. So that's setup step number one. Very powerful and very useful.
05:13Make sure you do that. Now setup step number two is to download text to voice software if you wanna use it.
05:19This one's optional, but I love this. Personally, I pay for WhisperFlow and I've linked that below, but you could also use other ones like Super Whisper and there are several others. What WhisperFlow allows me to do is just press the function key on my keyboard and then I can just talk to my computer and it will transcribe my voice to text and then I can use that to feed Claude my information.
05:38The reason it's so helpful is because talking is four times faster than typing. And if you ever typed and kind of tripped over your own fingers, this gets you out of the way. Whisper flow is really great because you can ramble and it's smart enough with the AI it has inside it to be able to de duplicate things you say, skip random pauses, and make the words intelligible regardless if you spoke it clearly or not.
06:00Now you could of course use the talk to text in Claude, but I prefer to have one thing that I can use across any application. So for WhisperFlow, I've got that function key across Slack, across Google Docs, across Claude, anything I'd need. Alright.
06:11So now at this point, we have the Claude desktop app downloaded. We built the folder structure. We've got the about me documents, and if you want, you've downloaded something like Whisperflow and got set up.
06:21The last piece in the setup step that I recommend is to download and install the Sandcastles dot AI MCP. If you're using Claude for social media growth, Sandcastles is the number one tool that will give Claude content superpowers. Think of Sandcastles like a bridge that connects all the social platforms directly into Claude.
06:39So you can just ask stuff in Claude, and it will use real social data automatically. The MCP is just a fancy way of saying the plug in that puts Sandcastles inside of your Claude. Now not all of the workflows that I go through in this video today are gonna use the Sandcastles MCP, But many of them do because if you want the latest data and you want insight based on that data, you gotta get the data in somehow.
07:00SandCastle is the best way to do it. If you want the MCP, go to sandcastles.ai, sign up for an account.
07:05As long as you're a pro visionary or Titan user, you can go to settings and connectors, download the Claude MCP, and then install it following the video that I linked below.
07:14Now as we go through these five workflows in this video, I'll be super clear when you need the SandCastle MCP, when you don't. But either way, Claude is the foundation that we're gonna run all of this on top of. And all of the Claude skills and the instructions for how to use those skills that I talk about for each of these workflows will all be linked below for free for you.
07:31Alright. The first Claude workflow that I wanna talk about is how to deep analyze any channel on social media. And I use this specific workflow on my own channel several times per week, but I also use it on competitor channels, and you can do this for any channel across Instagram, TikTok, or YouTube shorts using co work.
07:47First, let me show you what the end state is gonna look like once we have this done. We're gonna end up with a visual breakdown like this, which will take any set of videos that you give it and break out the best topics, hooks, formats, and non obvious patterns, so so you can understand what's really working on that channel.
08:02If there's nothing else that you use AI for when it comes to content, make sure you're analyzing your own channel, so you can figure out what hidden and non obvious patterns are there across the winners, so that you can double down. So this is exactly how we set this up and how we run it live. First, I like to start in the Sandcastle's web app and just make sure whatever channel you wanna run this on is in your watch list inside of the channels tab.
08:25Just go ahead, add that so it's in the watch list, you'll be good to go. Once it's in the watch list, go to the videos tab in the SandCastle's web app, click the channel filter, and select that channel. Now in order to run this analysis, you're gonna need to deep analyze through SandCastle's any of the videos that are in the date range that you care about.
08:42Now you've got two options here. You can either analyze all the videos within a specific range. So you could say last three months, oh, there's 50 videos.
08:49Boom. I'm gonna analyze all of them. Or you can sort by outlier score, still constrain the date range to like three months, but only analyze the top 10 or the top 15 based on performance.
08:58You can pick what you wanna do. Personally, I like to analyze my entire channel, it has both the winners and losers to differentiate the insights from the winners, but you can do whichever you want. Thirty to sixty days is what I recommend for your own channel.
09:09Now once you pick the date range and set whatever sort condition you want, in the header, you'll see how many videos map to that filter criteria. Now when you wanna analyze these, you have two options. You can go to Claude via the MCP and say, hey, for this channel, I want you to analyze this many videos.
09:25Or you can first analyze them in Sandcastles, which is much faster. And then once they're analyzed, you go to Cowork and have it do all the heavy analysis.
09:34That second option is what I recommend. So in Sandcastles, either hover over every individual video and press analyze button, or go up to bulk analyze and type in whatever number you see on the header if you wanna analyze all of them. And just note, this will cost you one credit in Sandcastles for every video you analyze.
09:49I'm just calling that out. Okay. So once all the videos are analyzed and I see the green check marks in Sandcastles, I then go to Cowork and I start this deep analysis process.
09:57And the reason I like to analyze in Sandcastles is it's just faster in the web app than going through the MCP for this specific piece. Alright. So in Cowork, I start a new chat.
10:05I make sure my Callaway dash Cowork folder is selected, and then I drop in this Sandcastle's channel analysis skill, which I've got linked below. Now when you run this skill, it'll first ask you how many videos you want analyzed across which date range.
10:19If you already analyze them in Sandcastle, it'll be ready to go, but if you haven't, you can then instruct it to analyze how many you want via the MCP. And that's it. You wait a few minutes, and it will run the entire analysis based on what we've coded into the skill for you.
10:31And what's cool is if you open this artifact that it gives you in Google Chrome, you'll see that this is a dynamic visual web page. So you can see all the thumbnails are loading in.
10:41You can click on any of the videos and it will take you right back to sandcastles where that video came from. So this really is as dynamic of a content strategy breakdown as you could have. Now when I run this channel analysis on my own channel, the insights are extremely valuable.
10:55All I've done from month one to month two on my Calendly marketing is just run this analysis and do what it said, and my growth has really picked up. Now of course, once you do this deep analysis, you've got all that data piped into CoWork. So you're not done here if you don't want to.
11:09You've got all the transcripts and all the detailed information. You can ask for anything else. There's dozens and dozens of other queries you could run on top of this data.
11:16You could also set up a recurring task to have co worker refresh this every two weeks and just ping you when the report is done. And if you wanted to do that for you, ask it to set up that recurring task and it will do it automatically. So that's the first workflow content strategy analysis for your channel and your competitors.
11:33Alright. Workflow number two does not require the Sandcastles MCP, and this one is called the audience bullseye builder. One of the most common problems I get asked when I work with business owners on content is how they can get top content performance while still having a super narrow niche.
11:47And when I explain it, I like to talk about it as a dart board or concentric rings around a bull's eye. Let's say the center of the bull's eye is your exact niche. High ticket b to b tech sales for example.
11:58And then what you do is you draw rings outside of that getting broader and broader ring by ring. So one ring outside might just be b to b tech sales. One ring outside that might just be b to b sales.
12:09One ring outside that might just be sales. And then the final ring might just be business. And this cascading set of rings creates what I call an audience bull's eye.
12:18Now the power of this is trying to figure out what should your ideal content mix be. For every seven videos you make, should you make them at level five, four, three, two, or one in the bull's eye? How do you stack it up?
12:28Now every niche is gonna be different and it requires experimentation to dial it in, but I recommend most people start with a three two two. Three videos in the bull's eye, two videos between level two and three, and two videos at level four. You don't wanna make any videos at level one, the broadest business in this case, because it's gonna bring you a lot of noise and people that are not qualified based on your niche targeting.
12:49Now the reason I'm talking about this here is because the clawed bullseye builder is a skill that allows you to build out these concentric rings, and it will work through what your niche positioning is for you. It interviews you and helps you make this five tiered bull's eye visual, and this will help you set your content strategy.
13:04And again, the skill file to do this is free for you below. It doesn't require any cost or anything to run. So if we run it, for example, here's what happens.
13:11The first thing it asked for is for us to describe our niche. So I might say something like, I help business owners with social media growth, and my specific goal is to help them turn views into dollars so they can both build authority, but also monetize their audience. And this skill will ask you a few questions to help you narrow down and really define that exact ICP, that single person that you could have in a chair that's the perfect viewer for you.
13:32Now once it's defined that, it will help you build out those five rings and it will explain the total TAM size of each ring and the way you should think about targeting. And I find this to be a super helpful tool when trying to hone in what your content mix should be. This one visual will really help add clarity for people with narrow niches.
13:49Again, the skill file to run this, super simple, linked below. Open a new chat, drop it in, press run, and you're good. Alright.
13:55Claude workflow number three is called the outlier video pulse. Here's how it works. My goal to try to preserve what's left of my sanity is to try not to scroll on social media to do research.
14:06But in order to figure out what's working in the niche, it is objectively helpful to look at what your top competitors are talking about just to have a sense or a pulse for what's going on. So using Claude, I built an outlier pulse skill, and it's very simple, but this is super helpful when you set it up. What it does is automatically give me an ordered list of all the videos posted by my competitors, and it refreshes automatically every morning at 7AM.
14:28And right now, I have it serving up in Claude itself, but if you wanted, you could connect Slack or Notion into Claude as a connector and have it drop it directly in a Slack channel or directly in a Notion page. Now all I have mind doing is just adding the outlier score, sorting it by the outlier score, and just giving me the raw link so I can open and watch the video.
14:47But of course, on the way in, you could also deep analyze every one of those videos and have it give you a bunch more data as well. So here's how you set this up if you wanna build one of these outlier pulse detection systems. The first thing I do is go into the Sandcastle's web app and I set my watch list with all the competitors that I want to track.
15:02Just way easier to do this in the web app, so go to the channels tab and then build your watch list. I've linked a tutorial below if you want help with that. Now you've got two options to run this skill and I've listed both of the Claude skills below.
15:13The one that I'm running right now is the outlier detector. It'll only pull videos that are above 25,000 views and one x outlier score from channels in that list within its height date range. We're going last seven days.
15:25If you wanna pull every video regardless of performance, you can run the all recent video, the all recent skill that's linked below as well. So in this example, let's run the outlier one, and again, super simple. Add a new chat, make sure you got your folder, take the skill, drop it in, and just say run.
15:40Now the good part about this workflow is that it doesn't actually cost any credits on Sandcastle. Still uses the MCP to pull all the metrics, but you don't have to deep analyze the videos in order to get them put in this table. So you can pretty much run this infinitely at infinite scale, and you'll be good.
15:53Again, just to zoom out, the goal of these workflows is to speed up the time you typically spend doing something that's low leverage, or do it for you to free up your time to do things that are higher leverage, two sides of the same coin. This one does both.
16:07Now you don't need to scroll at all, but you don't miss anything because you can just go to every morning, watch, watch, watch, watch, and then you're done. You can brick your phone and not watch social media. This type of thing, if you schedule it to run with scheduled tasks in Cowork, especially if you connect it to Slack or Notion, it is magical once you have it set up.
16:22Now one other option, you don't wanna use Slack or Notion, you can actually have it set to a mini live leaderboard that is just a web page that Claude automatically updates and populates.
16:32You just go and hit the URL and refresh it every morning. You could do any of those things, whichever works for your workflow, just describe it and CoWork will update the skill. Also, you can of course change the parameters for how often this runs to like six x or 12 x per day and get this more real time.
16:46You could also say every time one passes these thresholds, I want you to ping me in Slack and I want like a real time alert system. I'm doing it more like one time a day that I can check on my terms. You could also bring it in as an outlier proactive detector one time per day.
17:00So you have full flexibility. You can just customize it however you want. Alright.
17:04Workflow number four is a slight variation on top of the channel analysis that we talked about in workflow number one. This one's called the creator breakout detector. And this feels cool because it allows you to identify the exact moment or the exact video where a specific creator broke out, where they were kind of going along like this, and then boom, they had a spike in performance.
17:23This will identify when that happened, pull the video for you, and then you can deep analyze around it to figure out what specifically caused that change. I find this to be extremely helpful when you come across someone new, especially if they're smaller, but have recently grown and to reverse engineer their success. So here's how this one works.
17:39Like always, new chat, set the folder system, drop in the creator breakout detector, and press run. Now again, with this skill, you don't have to spend any sandcastles credits if you don't want to. This is pulling all the surface level metrics to figure out where the spike happened.
17:52If you wanna deep analyze from there, you can, but you don't have to. And so for this example, we will run it on my channel in co work, and let's do it. Now by default, this skill is designed to look back over two years, so seven hundred thirty days.
18:02And we're doing that because many creators have had a long arc. This will obviously take tens of minutes to go through all that video data. So if you want to be quicker, you can just reduce down the scope to forty five, sixty, or a hundred days.
18:13So for mine, for example, I've only been posting like seventy five days on this channel, so we'll just set that and we'll go. Now once it runs, again, this will give you a super helpful visual report that chronologically maps the progress and then shows you where those spikes occurred. If you can't already tell, the whole goal here on the content research and ideation is to figure out what are the outlier performers, and then let's go all the way down the rabbit hole to figure out what happened.
18:36What was different about the topics? What was different about the hooks? What was different about the formats?
18:39What is non obvious? That is where all the juice and the alpha is if you do this content research. So these skills are meant to like speed up that process, so you're not searching for needles and needle stacks.
18:50Immediately just hand you the needle proverbially with the metaphor, and you know exactly what to look for. Alright. Workflow number five, the last one is the hook machine.
18:58And some of you may have already tried this one before, but I'll link a video below if you haven't where I'm going through this in more detail. For this one, you are gonna wanna use the Sandcastles MCP with credits because you're gonna deep analyze as many videos that are relevant, extract out the exact hooks, and then use those hooks to build a hook machine for you to generate new hooks with.
19:16Now this one is probably the most advanced Claude skill workflow that I built if you look under the hood, but it is extremely powerful for writing net new hooks that still maintain creativity but use data. So here's how it works. The hook machine first starts by taking in one or several channels that you specify.
19:33I recommend doing it with one, like your own channel to start if you have winners or a single competitor if you don't. Again, to make things easier, make sure whatever channel you're gonna give it's already in a watch list in Sandcastles, and ideally, you've already deep analyzed those videos in the Sandcastles videos tab before you run the hook machine in Claude.
19:50Now once you run this, what happens is pretty magical. First, it will take whatever channel and whatever videos you've specified, and it will extract out the actual written hooks. It then clusters those and tries to find a pattern.
20:01Storytelling principles that you used commonly across the winners that it can kind of pull out and build a rubric around. It takes those principles from whatever channel you gave it and combines it with the fundamental storytelling pillars that I've hard coded in, combines that into one rubric. So think of it like 15 or 20 rules for writing winning hooks, mostly based on the channel you gave it, but also based on what I know to work as well.
20:25When you feed in a new topic, that rubric is used to write new hooks, and it grades them accordingly. The coolest part in my opinion is that you can feed it your own hook, and it will grade it, and then give you alternatives that are better grades based on that rubric.
20:38So the whole thing is kind of like this macro take data, build rubric, use rubric to generate and revise hooks that are given. The whole reason I built this is because I was trying to figure out how we could use Claude and Cowork to write winning hooks, but still leverage the data and not have it sound like AI slop.
20:54I wanted it to feel a lot more organic and creative. So this is the best and closest solution that I found. Again, if you wanna run the hook machine skill, it is linked free below.
21:03You're gonna want the sandcastles MCP and you wanna deep analyze videos to run it, of course. Now before I end this video, I wanna just call out a few other content based MCPs that are working like sandcastles, tackle different pieces of the content stack.
21:16Now transparently, these three MCPs I have not used extensively. I've tried some of them, but I haven't gone down the rabbit hole. If So you want to make a video deep dive in the future about any of these, make sure you drop that in the comments.
21:27The first one is the Claude Remotion skill. Remotion is the world's best AI editing skill inside of Claude right now. So using this, you can create anything from simple animations all the way to pretty complex edits, all programmatically from Claude.
21:41The second one is the Higgs Field MCP. This is one of the leading tools right now for generating top quality AI visuals directly from the Claude interface, either Claude code or Cowork. And lastly, personal favorite is the Beehive MCP.
21:54Beehive is the first email platform ever to launch an MCP, and this allows you to use Claude literally to write emails and analyze your email subscriber list through Beehive, but via Claude. All three of these are extremely powerful, I've linked them below if you wanna play with them. Again, I haven't used them extensively, but I will test them if you guys want me to in the comments.
22:12Alright, guys. That is all I've got for this video. As promised, these are five of the most helpful Claude content based workflows.
22:19Using co work and using all the skills below, they don't produce slop and they help you improve your content and content strategy much faster. Now I've got several more of these Claude skills in the chamber, specifically around ideation, script writing, and more. So if you want me to make a part two to this video where I do another batch of five, comment the chef like usual and let me know.
22:38Say, give me the MCP tool chef, and I'll know to make a second video. As always guys, I'm trying my absolute best to give you the non obvious stuff that I just don't see covered all the time in this space. If you're a serious creator or business owner, you can probably already tell this is gonna be one of the best channels you could possibly watch.
22:53So make sure to like, make sure to subscribe, and drop in the comments what you wanna see me cover next. And lastly guys, make sure you download all the free stuff in the description. I spent hours and hours making all those skills, organizing them for you, they're right there for you.
23:05So take those and it'll make it way easier to use co work to grow your channels. Thanks again to Anthropic for sponsoring this video. We will see you guys on the next one.
23:13Peace.
The Hook

The bait, then the rug-pull.

Five packaged Claude Cowork workflows, built on a persistent context-folder system and a live social-data connector, turn scattered outlier-hunting into an automated content-research pipeline.

Frameworks

Named ideas worth stealing.

02:10model

Cowork Context Folder

  1. A name-Cowork folder as the root
  2. A context subfolder holding four markdown files
  3. A work subfolder with one folder per project or workstream
  4. Five standard subfolders inside each workstream: archive, assets, references, skills, videos
  5. Four context files: how-I-talk.md, how-you-work.md, who-I-am.md, and an instructions file the assistant reads first

A one-time folder and file structure that gives the assistant durable, personalized context about voice, business operations, and identity across every new session.

Steal forany AI writing or research workflow that currently requires re-explaining context every chat
11:33model

Audience Bullseye

  1. Center ring: exact niche
  2. Ring 2: one level broader than the exact niche
  3. Ring 3: broader still
  4. Ring 4: broader still
  5. Outer ring: broadest possible category

A concentric-rings model for content targeting, used to decide how narrow vs. broad each piece of content should be.

Steal forplanning a content calendar mix for a niche brand
13:00acronym

3-2-2 Content Mix

  1. 3 videos at the bullseye center (exact niche)
  2. 2 videos between the second and third rings
  3. 2 videos at the fourth ring
  4. 0 videos at the outermost, broadest ring

A recommended default split of seven videos across the audience-bullseye rings, deliberately excluding the broadest ring as too unqualified.

Steal forany content calendar that currently has no rule for how specific vs. broad to go
18:54model

Hook Machine pipeline

  1. Pull a channel's already-analyzed videos
  2. Extract the literal hooks used
  3. Cluster them to find recurring storytelling patterns
  4. Combine channel-specific patterns with general hook-writing principles into one rubric
  5. Use the rubric to grade and generate or revise new hooks against any topic

A four-stage system that turns a channel's proven hooks into a reusable, gradeable rubric instead of copying individual lines.

Steal forany repeatable hook-writing process that wants to stay grounded in real data instead of guessing
CTA Breakdown

How they asked for the click.

VERBAL ASK
22:38subscribe
make sure to like, make sure to subscribe, and drop in the comments what you wanna see me cover next... make sure you download all the free stuff in the description

Closes with a standard like/subscribe/comment ask paired with a lead-magnet push (the free skill files in the description), plus a second Anthropic sponsor thank-you.

Storyboard

Visual structure at a glance.

cold open
hookcold open00:00
context folder setup
valuecontext folder setup03:39
deep channel analysis
valuedeep channel analysis09:25
3-2-2 content mix
value3-2-2 content mix12:34
outlier video pulse
valueoutlier video pulse17:03
creator breakout detector
valuecreator breakout detector18:55
hook machine
valuehook machine19:27
sign-off
ctasign-off23:05
Frame Gallery

Visual moments.

Watch next

More from this channel + related breakdowns.

Chat about this