Big Idea

The argument in one line.

A comment-mining skill that scores competitor posts by their own-channel outlier ratio, reads what audiences asked for but never got, and adapts to your taste over time surfaces angles that generic trend reports never see.

Who This Is For

Read if. Skip if.

READ IF YOU ARE…

You are a creator in a technical or AI niche who burns hours each week just deciding what to make next.
You want a self-hosted free research loop instead of paying monthly for a content research tool.
You already use Claude Code, Cursor, or Codex and want a skill that runs inside your existing workflow.
You have published enough content that recycling past topics and missing audience requests have become real problems.

SKIP IF…

You do not use an AI coding assistant; the skill requires Claude Code, Cursor, or Codex to run.
You need video ideas immediately; the taste-learning feedback loop compounds over weeks, not days.
Your niche has too few active competitors to generate a meaningful feed of trackable posts.

TL;DR

The full version, fast.

Most content research tools show what is popular, but popular is the same data everyone sees. This skill adds two layers: it reads comment sections of high-performing posts to find angles the video never covered, and it scrapes the creator's own channel before every run so it knows what topics are exhausted and what the audience is explicitly requesting. A thumbs-up and thumbs-down feedback system then trains Claude auto-memory to the creator's taste over time. The whole system costs under five dollars per month at a once-a-week publishing cadence.

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 – 01:00

01 · Results

Channel growth from zero to 10k subscribers in three months; three videos hitting 70k, 95k, and 120k views each traced back to the same skill.

01:00 – 02:19

02 · Demo: running /content-ideas live

Live walkthrough of the For You page: competitor posts scored by outlier rating, filtered by platform, and an Ideas tab with up to ten video starting points including hook and format recommendations.

02:19 – 03:13

03 · How it finds angles

The comment-mining mechanic: the skill reads comments on every analyzed post because that is where the question the video did not answer lives. The 120k-view context-limits video traced to a single comment trend.

03:13 – 03:55

04 · Anti-cannibalization

Before every run the skill scrapes the creator's own channel, builds a record of what has been covered, and prioritizes ideas the creator's own audience is explicitly requesting in comments.

03:55 – 05:52

05 · Setup

Install via Claude marketplace plugin or two GitHub commands; first-run setup asks for ScrapeCreators API key and content goals. Works in Claude Code, Claude Chat, Cursor, and Codex. 100 free ScrapeCreators credits; $50 for 25k credits (~160 runs).

05:52 – 07:44

06 · Self-improvement

Thumbs-up and thumbs-down buttons on every card feed Claude auto-memory. The skill reads all past reactions before each new run and builds a taste profile learning format preferences, topic angles, and voice over time.

07:44 – 08:20

07 · Download and CTA

Free download link in description (GitHub repo). Offer: AI strategy calls. Teaser for autopilot scheduling follow-up video.

Atomic Insights

Lines worth screenshotting.

The angle that makes a video break out is usually hiding in the comments of a competitor post, not in the post itself.
Scoring a post by how far it beat that creator's own channel average is more useful than raw view count for spotting breakouts.
A 120k-view video at 700 subscribers was traced to a single repeating comment pattern spotted across dozens of AI-niche posts.
Scraping your own past content before generating ideas prevents the system from suggesting something you have already covered.
Claude auto-memory lets a skill learn content taste from thumbs-up signals rather than requiring you to re-explain preferences every session.
Generic trend tools give every creator the same data; the edge is stacking what the metrics say with what resonates with your voice.
Pay-as-you-go API pricing at $50 for roughly 160 research runs makes a self-hosted loop cheaper than almost any subscription alternative.
A skill that runs on autopilot each morning shifts idea generation from reactive scrolling to a pre-scored inbox.
Feedback notes explain the WHY behind your taste, not just the pattern of what you clicked, making personalization more precise over time.
The creator attributes 10k subscribers to under $5 in total API costs, meaning the bottleneck was research method, not ad spend.

Takeaway

Comment sections hold the angles the video never covered.

WHAT TO LEARN

The information you need to make a video that breaks out is already public, sitting in the replies of videos that performed well but left something unanswered.

Scoring posts by how far they beat a creator's own channel average is more reliable than raw view count for identifying what actually broke out versus what was just popular on a large account.
Comments are audience demand signals: each one represents something a viewer cared enough to type, making them a more reliable indicator of unserved interest than any trend tool topic list.
Tracking your own past content before generating new ideas ensures suggestions stay genuinely fresh rather than recycling territory you have already covered.
A feedback loop that teaches a tool your taste through explicit reactions rather than re-explanation compounds over time so suggestions start to reflect your voice instead of generic AI outputs.
A research process that runs automatically before your workday starts removes decision cost from the creative process and converts a weekly drain into a pre-scored morning inbox.

Glossary

Terms worth knowing.

Outlier score: A metric measuring how far a post performed above or below that creator's own channel average, used to identify genuine breakout content rather than just popular-by-volume posts.
Anti-cannibalization: A feature that scrapes the user's own published content before every run so the system never suggests a topic the creator has already covered.
Claude auto-memory: A native Claude Code feature that persists information between sessions inside a project folder, allowing a skill to accumulate learned preferences over multiple runs.
ScrapeCreators: A pay-as-you-go third-party API that pulls posts, engagement data, and comments from YouTube, X, Instagram, and TikTok for analysis by the skill.
For You page: The skill's main output: a personalized feed of highest-performing competitor posts scored by outlier rating, plus an Ideas tab with up to ten video starting points derived from what those posts left unanswered.

Resources

Things they pointed at.

00:00tool/content-ideas GitHub repo ↗

04:30toolScrapeCreators API ↗

07:40linkAI Strategy Call booking ↗

Quotables

Lines you could clip.

02:28

“The comments are a gold mine because that's where you find the angle the video itself didn't cover.”

Tight reframe with zero setup needed→ TikTok hook↗ Tweet quote

05:49

“I got 10,000 subscribers for under $5 in API calls, so I'll make that trade.”

Concrete number comparison that deflates the expensive AI tools objection→ IG reel cold open↗ Tweet quote

07:15

“The real edge here is that it knows what you like, your taste, and your angles.”

Names the exact fear creators have about AI content then reframes it→ 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.

00:00So I built a Claude skill that got me over 10,000 subscribers in the last three months, and I've actually got the receipts to prove. Every single one of the videos that got me here started in the same place.

00:10And by the end of this video, I'm going to give you my exact Claude content idea system totally free. But first, let me take you back.

00:17The hardest part for me about content was never filming or editing. It was knowing what to actually make in the first place. For every single video, I'd burn hours on Reels, x, and YouTube just hunting for a seed of an idea worth making.

00:30And I thought, screw this. I can build this myself. So back in March, I handed this whole job to Claude.

00:36I spent months tinkering with it, getting it right, a Claude code system that identifies winning content ideas in my niche. And at first, it was a few 100 extra views here and a booked call there. But then the first one popped, a 120,000 views out of nowhere.

00:50And a couple weeks later, another one hit, 95,000 views, and then again, this time 70 k. And the idea behind every single one of these videos started in the same place, Claude Code, armed with this skill.

01:02I spent months getting this right for myself, and today I'm handing it to you totally free. So let me show you exactly how it works.

01:09So I'm inside of Claude code here on my laptop, and all I have to do is type slash content ideas, and in about two minutes, I get this. It's a fully customized for you page across x, YouTube, Instagram, and TikTok. But instead of having to sift through hundreds of posts, I just get the highest performing post from my competitors.

01:26And that's not all. It actually gives me ideas as well, but I'll get to that in just a second. For each one, I can see what they posted, how it did, and the highest impact comments.

01:36But the bit I actually care about is this, the outlier score. That tells me how that video performed against their channel average. So I'm not just seeing what's popular, I'm seeing what actually broke out for that creator.

01:48And I can filter it by any platform I'm tracking all in one place. Now onto the actual source, and that's the ideas tab where Claude takes everything it just read and hands me up to 10 starting points for videos. Now they're not finished scripts and that's intentional.

02:03They're just shoots for ideas. And for each one, I know where it fits in my funnel, the hook, and the format to shoot it in. So that's the surface, and you're probably wondering how does this actually come up with these ideas.

02:14And the thing that finds the angles that nobody else is covering is hiding somewhere most tools never actually look. This skill doesn't just look for high performing content, it reads the comments on every post it analyzes. And the comments are a gold mine because that's where you find the angle the video itself didn't cover.

02:31Each comment tells you what someone cared enough about to stop and type, and that's actually signal. It might be a question, a floor, or just an observation. And all of these, when analyzed across hundreds of posts, add up and trends start to emerge.

02:44Look at this. My video on Claude context limits came from this single. This is the one that got a 120,000 views when my channel only had 700 subscribers.

02:53This skill picked up on an increasing trend of viewers asking about the best ways to limit their context usage, and that was everywhere across x and YouTube comments. And that's where the video idea was born. Was Claude's version of the idea perfect?

03:04No. But it was enough to give me the idea to make a winning video. I'll come back to this in a second.

03:09But the skill doesn't just read other people's comments, it reads yours too. So before every single run, the Skill scrapes your own content first, and it actually keeps a file on you, what you've posted, how each one did, what's working, and what's not. So when it hands you ideas, it already knows what you've covered, and it won't tell you something you've basically already made or recycle the same idea again and again.

03:32And that's where your own comments come in because it's pulling what your own audience is asking for directly, and it's pushing those ideas to the top of the list. So your next video isn't going to be a guess.

03:42It's going to be based on what your own followers are already telling you that they want. That alone is worth the cost of getting this set up, and the setup process itself is actually nothing. So let me prove it to you right now because you can have all of this running in about a minute.

03:55By the way, this skill actually works in any AI coding assistant, whether you're using Cursor, Codex, or Claude, and the skill works the same in pretty much all of them. Plus, it works in Claude Chat and Claude Cowork as well.

04:07But I still recommend using Claude Code for the best experience as it unlocks self improvement. And this is how Claude learns your content style over time, but I'll get to that soon. But first, let's set this up inside of the Claude desktop app.

04:19First, you wanna go to the link in the description and grab the repo. In there, you'll find the marketplace link. You'll copy that and grab come into Claude here to the customize tab.

04:27Once you're on the customize tab, add the plugin and then add the skill. In Claude code, it's easier. You literally just copy these two commands from GitHub and then add them in.

04:36And once it's installed, stays it up to date automatically. Every time I push an improvement, it'll come down to you. On cursor codex or anything else, there's one command in the GitHub here, and you just copy that in and you'll use Vercel skills to install it really easy.

04:50The The first time you run it, Claude's gonna run you through a full setup walkthrough. It asks you for the API key for a service called Scrape Creators, and that's the thing that pulls all of the posts for you super easily. Now Scrape Creators isn't free, but don't freak out.

05:03It's pay as you go. So there's no subscription involved, and it's dead cheap. This video isn't sponsored either.

05:08It's just a good tool that makes my life easier. And they actually give you a 100 free credits to try it, so you don't have to pay to try this tool out. If you do wanna go further, it's $50 for 25,000 credits, and that should last you for about a 160 runs.

05:21So for me, I make a video a week. That's a long time. I got 10,000 subscribers for under $5 in API calls, so I'll make that trade.

05:28Once you do make an account, you can jump into scrape creators, and you'll be able to grab the API key from up the top here. Like I said just before, when you start the setup, Claude actually interviews you about your content. It's who you wanna track and what your goal for the content is.

05:41That way, the content strategy that Claude comes up with all aligns with your actual business goals, and it's not gonna be some generic report trend. But here's the thing that makes this different to every other content research tool you've ever tried. It actually learns your taste over time.

05:55You might have missed it earlier, but on every single card here, there's actually up and down vote buttons. And this is the part that nobody else is doing. And that's the reason why I'm still running this exact skill every single week.

06:06Because every week, every post and idea that comes into the feed, I can thumbs up or thumbs down it just like on YouTube, and you can leave a note on what you did or didn't like so Claude knows exactly why you gave that feedback. And this is the real power of the plugin because it takes advantage of a native feature in Claude, auto memory, which lets Claude learn about you and your context of the current project.

06:26Now this works best in Claude code because the feedback is saved a 100% automatically. But it still works in co work. You just need to hit the download feedback button and save the feedback in the research run folder.

06:36Anyways, once you do leave some feedback on a run, the next time you actually run the skill and before it pulls any new post, Claude goes back and reads everything you reacted to last time and saves all of that information into its memory. But it's not just remembering that one post you didn't like, it's looking for the pattern and it's learning your taste over time.

06:54So say you thumbs up three different listicles over the course of a couple of runs. It doesn't know he didn't like three posts. It works out that listicles maybe just don't land for you, and it stops surfacing them.

07:05Thumbs up a few contrarian takes, and it might figure out that you lean contrarian for a lot of the content. So it starts bringing you more of those like an algorithm. Plus the notes feature, you can be even more explicit.

07:14You can explain exactly what does or doesn't work about this content idea for you, and here's why that's the whole game. Sure. This skill knows what the metrics say, and it can tell you what performed well and got a lot of likes and saves, but so does every other tool and everyone else.

07:28Half the story and the real edge here is that it knows what you like, your taste, and your angles. That's the stuff that makes content sound like you and not like Claude. And those are two completely different signals that when you stack them on top of each other, you start getting ideas that feel like yours.

07:44So that's my Claude content idea system. It's free. The link is in to install it is in the description below, so go grab it.

07:49Once it is running, here's your next move. If you want help actually building this kind of AI into your business and not just for content, but across the board, I do AI strategy calls. And there's a link in the description below to go book one.

08:01Spots are limited, and I keep them that way on purpose. So if that's you, go and book now. Subscribe if this was useful, and once you've got the skill running, the real unlock here is putting it on autopilot.

08:11So a fresh feed is waiting for you every single morning without you lifting a finger. I show you exactly how to do that in this video right here. Thanks for watching.

The Hook

The bait, then the rug-pull.

A creator with 700 subscribers published a video about Claude context limits and watched it hit 120,000 views. The idea did not come from scrolling trends or hiring a strategist. It came from a single repeating pattern in the comment sections of competitor posts, surfaced automatically by a Claude Code skill he built himself.

Frameworks

Named ideas worth stealing.

01:30model

Outlier scoring

Rank posts by how far they exceeded the creator's own channel average rather than by raw view count. This surfaces genuine breakouts, not just posts from large channels.

Steal forAny competitive research workflow tracking multiple channels with different baseline sizes

02:30concept

Comment-to-angle pipeline

Every comment is a signal of something the viewer cared about but the original video did not fully address. Aggregating comments across hundreds of posts lets patterns emerge as unserved topic angles.

Steal forContent ideation in any educational or tutorial niche

06:20model

Taste-learning feedback loop

Thumbs-up/down plus optional notes on each content card trains an AI memory that learns format preferences and topic angles over multiple sessions without manual re-configuration.

Steal forAny AI recommendation layer where personalization needs to compound over time

CTA Breakdown