← Hub / 08 — Tool
08, free tool

Content Spy

Open source, MIT Runs in Claude Code Built by Scott, used daily

Read what is actually working in your niche, then write the next post in your own voice.

Content Spy pulls the top short-form posts in your space, breaks down why they land, watches what is trending across the open web, and hands you scripts you can film today. The whole thing runs on your machine.

Three steps to install about five minutes

Run these inside Claude Code.
Open Claude Code in the folder you want it to live in, paste each step, hit enter.
By the third command it is already pulling posts.

Step 01
Clone
Pulls the repo, installs the dependencies, drops a starter env file in place so you have something to fill in.
git clone...
Step 02
Setup
A short interview about your niche, your voice, and the API keys you have. Everything downstream is tuned from your answers here.
/setup
Step 03
Run All
One command runs the full chain. It scrapes, takes a pulse on the wider web, analyses what it found, and writes you a set of scripts.
/run-all
Most creators are quietly guessing

A reel goes off, you copy the format, and the version you film flops. The reason is almost never the part you noticed. The post that worked had a specific hook structure, a pattern interrupt at 0:03, a retention beat at the 8-second mark, and the surface you copied sat above all of it. You were studying the paint, not the frame.

Content Spy automates the research-to-script loop that runs every week for EDGE and becomingscott. It scrapes the top creators in your niche, transcribes the videos, breaks down what is actually doing the work, scans the open web for what is trending right now, and writes you scripts in your own voice that are grounded in what it found.

The whole loop runs on your laptop, sharpens itself every week, and the output is plain markdown that lives in a folder you own.

Three stages, one shared folder

Every stage reads and writes to the same output/ folder, so the next stage picks up exactly where the last one stopped. Run them one at a time when you want to poke at the middle, or chain the whole thing with /run-all and walk away.

Stage 01
Collect and Analyse
Scrapes the top creators and the news sources in your niche, transcribes the reels, and writes a breakdown of each post that names the hook type, the pattern interrupts, the retention structure, and the one technique worth lifting.
Try: Point it at three creators bigger than you, leave it running overnight, then read what the breakdowns have in common in the morning. The recurring hook structure across all three is the format you should be filming next.
Stage 02
Pulse Check
Sweeps the open web, X, Reddit, Hacker News and GitHub, for whatever is moving in your space this week, then cross-checks each trend against your existing library so the angles nobody has touched yet float to the top.
Try: Open the uncovered-angles list every Monday morning. That is where the lowest-competition, highest-relevance scripts of the week are sitting, waiting for someone to film them first.
Stage 03
Script Generation
Takes the validated trends and the proven hooks and writes scripts in your voice, ready to film. There is an insights memory underneath that carries lessons from every previous run, so the writing gets noticeably sharper the longer you use it.
Try: Do not film the first script straight off the page. Rework the hook, drop in one specific reference from your own week, then film it. The system writes the structure, you bring the spike.
Two halves talking through one folder

A small local app handles the mechanical work, an AI prompt layer handles the thinking, and they meet in the output/ folder on your disk. Markdown is the source of truth in the middle, and both sides read and write to it freely.

The App Local

A light Express server with a vanilla-JS dashboard sitting on top. No framework, no database, just JSON indexes derived from the markdown files themselves so everything stays inspectable.

  • Scrapes Instagram through Apify, plus any RSS feeds and websites you point it at
  • Transcribes the reels through Transcript24
  • Dedupes, indexes and shows the stats on the dashboard
  • Dashboard lives on localhost:3000
The Intelligence Layer AI

A set of markdown prompts you run on demand inside Claude Code. This is where the thinking happens, and every prompt is plain text you can open, read and rewrite.

  • Analyse, Pulse Check, Script Gen and Insights, all separate prompts you can chain or run alone
  • Tuned to your niche and your voice during /setup
  • Run the full chain at once with /run-all
  • A compounding insights memory that carries between runs

Local-first on purpose. Everything runs on your machine, your data sits in your own output/ folder, no API keys are baked into a server somewhere, no telemetry pings home, and nothing leaves your laptop unless you decide to send it.

Do not film what it writes, film what it teaches you

The first batch of scripts the tool writes is good enough to film, but the real unlock is the second loop. Take three scripts from one week, rewrite the hooks in three different directions, post them spread across the calendar, then let the dashboard tell you which one actually moved when the numbers refresh.

That signal goes straight back into the insights memory underneath the prompts. By the third cycle the writing is no longer aimed at the niche average, it is aimed at the specific audience you have been training it on through your own posts. That is the point where most people drift away from the tool, and it is also the point where the tool finally starts to earn its keep.

Run /run-all daily, film three from the queue every week, and read the dashboard on a Sunday afternoon. Two cycles in, the scripts will sound more like you than anything you have written by hand in months.

Want this running on your situation?

Book a call and we will work out the shortest path. Sometimes that is me doing the build, sometimes it is advising you through it, sometimes it is handing you to the right person in the network.

Book a call →