Copy an AI Expert’s Claude Content Factory

Claude’s latest models have changed what’s possible for non-coders. Saj Adib proved it by building five custom apps in two weeks for $200. These are his exact prompts for a content factory, an AI second brain, and more time-saving tools.

Create, optimize, and measure content that actually drives leads with Content Hub.
Get started free

The latest Claude models, Fable 5.1 and Opus 5.5, have made a dramatic leap, and Saj Adib, Futurepedia Principal Producer, and founder of Skill Leap, built two apps that prove it. A fully automated YouTube Shorts workflow that edits and posts while you record the next video and a navigable 3D second brain built from your life. These are every prompt he used to build both.

Before you start, you’ll need:

  • Claude Code (Max plan, $200/month)
  • Adobe Premiere or DaVinci Resolve 
  • Blender 
💡 One tip from Saj: Claude will often recommend using a local model for tasks that don’t require its full reasoning power — this saves tokens and keeps costs down. Follow those recommendations when they come.

 

 

Workflow One

YouTube Shorts

After recording a long-form video, Saj records a handful of hook lines after the outro — each one becomes a short. From there, three Claude skills handle everything: cutting the shorts from the long-form footage, writing platform-specific posts for each short, and uploading them as drafts to TikTok, Instagram, Facebook, and LinkedIn.

 

1

How it works in sequence:

  1. Record hook lines after your long-form outro
  2. Skill 1 cuts each hook into a finished short — framed, captioned, with matching B-roll
  3. Skill 2 writes platform-specific posts for each short
  4. Skill 3 uploads them as drafts across platforms
  5. The Command Center app gives you a single place to review, edit, and mark posted

Make a skill that turns my long YouTube recording into vertical Shorts. After the outro I record a few hook lines, each one a 
Short. Find them in the transcript, cut each into a 9:16 video with my face framed, burned-in captions, the screen demo it refers to and matching B-roll, and export them next to the video as S1, S2, S3.

Make a skill called shorts-to-posts. Given one Short, transcribe it with Whisper, then write ready-to-paste posts for LinkedIn (a carousel post with a branded step-by-step PDF, plus a video post with a long caption), X, YouTube, Instagram + Facebook (one caption) and TikTok. Give each a few title and caption 
options and make the thumbnail and cover images. Rules: no dashes, never the word "boost", only claim what the video shows. 
Save it all as "<Short> posts.md" next to the video.

 

Make a skill that uploads one finished Short as a draft on TikTok, Instagram + Facebook (Meta Business Suite) and LinkedIn 
in my signed-in Chrome: the video, the cover and the caption from its posts.md. It must stop before Post every time. Add a 
hook that blocks any click on Post, Share, Publish or Schedule, and never touch YouTube or X.

Add a Shorts page to Command Center. Show every Short grouped by video, with To post / Posted tabs. A Post button opens one 
tab per platform with the cover, each text in an editable box with a big Copy button, "Regenerate this platform's post" (runs 
shorts-to-posts again) and "Mark posted". "Upload as drafts" runs skill 3 in the background through Claude Code and shows 
progress. Nothing in the app can ever post.

Before you test it: have a Short that already has its posts.md file ready before you open the Post window. If the posts.md doesn’t exist yet, it will write one before the window opens — which takes a minute and can feel like something’s broken when it isn’t.

Workflow Two

AI Second Brain

Saj’s second brain is a folder of markdown files about his life — his jobs, travel, books, memories, milestones, and work history, organized across 12 categories. The Life OS Castle turns that folder into a firelit 3D world he can walk through, where every room holds one part of his life.

MATG Picto_Lightbulb

How to build your Life OS folder first:

Before you build the Castle, you need the markdown files. Saj built his by going on 45-minute walks using ChatGPT or Claude voice mode and asking it to interview him on specific topics — all his jobs, places he’s traveled, books he’s read. The voice conversation gets organized into markdown files in a Google Drive folder. He ended up with around 400 files across 12 categories. Start with the categories that matter most to you and add from there.

Build a local Mac app called Life OS Castle (Electron + Three.js + Vite) that turns my Life OS folder, a library of markdown files about my life, into a firelit castle I walk through in 
first person. Read only and local only: never write to the library and block every network request.

Each folder becomes a room themed to what it holds: Guide & Rules, Timeline & Years, Books & Learning, Work, Personal Life, 
Fitness. Files appear as real objects (books on shelves, scrolls, ledgers, framed pictures). Press E to read one.

Walk with WASD, mouse look, stairs, jump and sprint. M opens a map, Cmd+K searches every file, L shows a plain list.

In the main hall, hang my life timeline as tapestries: work milestones above the line, life milestones below, read from my timeline files.

A private vault in a tower, behind a password (store only a salted hash). It locks on idle, sleep and quit, and hides the window from screen sharing while open.

Doors lead out to worlds for each part of my life, each with its own sky, light and sound.

A demo mode that loads a made-up person's library instead of mine, so I can show it.

Add a new world behind a door in the Work Lobby: a soundstage for my video career, with the props from my early film years on 
one side and today's YouTube studio on the other. Each prop opens the file it stands for.

Before you load your real data: run it in demo mode first. The demo mode loads a made-up person’s library so you can validate everything works before your own files are read. Add the demo mode step before you point it at your real Life OS folder.

More Applications

Saj built five apps on Marketing Against The Grain. The prompts for Daybook (a screen-monitoring productivity tracker), Analytics Center (a live dashboard pulling from Stripe, QuickBooks, Gmail, Ahrefs, and more), and Usage Center (a Claude token tracker) are also included below for anyone who wants the full suite.

Build me a native macOS menu-bar app called Daybook (Swift + SwiftUI, local SQLite) that shows where my workday actually goes. 
Everything stays on my Mac.

1. Tracking: record the app in front and what I'm doing inside it (window title, browser site, document or project name). Group it into sessions and projects, ignore quick glances, and when I come back from being away, ask me what it was (meeting, call, break).

2. Screenshots: one every 5 minutes, skipped when I'm idle, paused, or nothing changed. Keep 14 days, then archive.

3. Privacy settings with three lists: apps never screenshotted (password managers by default), apps whose titles are never recorded (Messages), and apps ignored entirely. Add a pause hotkey.

4. Screenshot understanding on device: Apple Vision reads the text, then a local open-source vision model through Ollama (Gemma 4) describes each screenshot in one sentence. Run it only on power while I'm idle. Refuse cloud models: images never leave the Mac.

5. Views: Day timeline, Week, Apps, Projects, Screenshots, and PDF reports for a day, week or month.

Start with tracking and the Day view, and let me test on my real data before adding more.

Build a native Mac app called Analytics Center that puts my Claude live dashboards in one window. Read the list from a dashboards.json file: each entry has the artifact id from its 
claude.ai link, a name, a sidebar group (Overview, Money, Inbox & Calendar, Growth), an SF Symbol and the data sources shown 
under the name.

Sidebar of dashboards on the left, the dashboard in a web view on the right, signed in to claude.ai once.

Keep each dashboard loaded after I open it, so switching back is instant and doesn't re-run its queries.

Auto-refresh the one on screen every 5, 15, 30 or 60 minutes, only while the window is visible.

A flag in the JSON to never auto-refresh a dashboard that spends paid API units.

A build.sh that builds the app and installs it in /Applications.

Build a native Mac menu-bar app called Usage Center that shows how many Claude tokens I use. Read the transcripts Claude Code and the Claude app already keep on this Mac (the .jsonl files under ~/.claude/projects and the Claude app's session folders). Nothing leaves the Mac, no API calls.

Count every reply once (resumed sessions repeat them) and group by day, source (Claude Code in the terminal, Claude app, Code, Cowork) and model.

Menu bar: today's output tokens, cache reads and replies, plus the last 7 days.

Window: a daily stacked chart I can break down by token type, source or model, with a toggle to hide cache reads; an output trend; this week vs last week.

Plan limits: chart the 5-hour and weekly percentages the Claude app records, and show each day's peak in a table.

Mark any day with fast mode replies, because fast mode bills as extra usage.

Option to open at login, and a build.sh that installs it in /Applications.