One 25-minute recording becomes eight approved clips across four platforms, with captions in your voice — while you're in a Tuesday meeting. Nine modules. About four hours to build.
The full architecture, the module reference, and all four caption prompts are on this page. No email needed for any of it.
Build it one module at a time and run the scenario after each addition. Never assemble nine and press play — when it fails you'll have nine suspects and no information.
Google Drive · Watch Files
Point at /raw-video/. Set it hourly, not every 15 minutes — for a weekly recording that saves ~2,100 operations a month for identical results.
OpusClip · create clips
Finds the hooks, reframes to 9:16, burns captions, scores each clip 0–100. Processing is async — handle it with a webhook back into Make, not by making the scenario wait. It'll time out.
Iterator
Splits the returned array so each clip flows through separately. If your chain works but only ever handles the first clip, this is why. Everyone forgets it once.
Router + filter · virality score > 70
Two modules of config, and the difference between a content machine and a spam machine. It also protects module 5 — eight clips a week is a real review, twenty-four is a rubber stamp.
Telegram · approve / skip
Thumbnail, score, first line of transcript, two inline buttons. Your entire weekly involvement: eight taps in a queue somewhere. Never remove this, however well it's been running.
Claude · platform-specific captions
Four separate prompts, one per platform — never one "repurpose this." The platforms are different rooms. Prompts below.
Buffer · queue the post
Pre-set your schedule once; the chain only adds to the queue. Add a file-size check before this — Buffer will reject an oversized video without a useful error and the post silently never happens.
Airtable · log it
Clip name, source video, platforms, score, performance. The module people skip and the one that makes month three intelligent — at 90 days you can sort and find which topics actually produce buyers.
Error route → Automation Log
Low severity: nobody outside the business notices a missed clip, so it logs and waits for Sunday. One exception — promote OpusClip quota errors to Telegram. They're predictable and entirely preventable.
Copy these into your Claude Project so they inherit your Business Brief, writing samples, and buyer vocabulary. Everything above the platform block is identical across all four.
WEAK CLIP is in thereSome clips clear a 70 score and are still fragments that don't say anything complete. Without an escape hatch the model will dutifully manufacture a point the transcript doesn't support — in your voice, on your account. Giving it permission to refuse costs one line and prevents the most common form of caption embarrassment. Filter those out with a router before Buffer.
All three are cheap to prevent and expensive to diagnose live.
Credits run out mid-month, silently
150 credits sounds generous until you upload a 55-minute video in week one. The chain doesn't announce it — the module errors, the error route logs it, and everything downstream just doesn't happen. Cap raw uploads at 30 minutes and set a usage alert at 70%.
Buffer rejects oversized files without telling you
The clip appears to schedule. It never posts. Add the file-size check before module 7 and route oversized files to Telegram. Once a month, spot-check that something you approved actually appeared publicly — looking is the only reliable detection.
You edit it at 11pm
The most common failure in the whole book and the one nobody writes down. You change something small, don't test it because it's late, and it's broken for nine days. Clone before you edit — two clicks in Make — and don't touch live scenarios after 10pm.
| Tool | Monthly |
|---|---|
| OpusClip Starter (150 credits) | $15 |
| Buffer, 3 channels | $18 |
| Make.com Core (10,000 ops) | $12 |
| Chain total | $45 |
Prices as of press date. The maintained version of every price in the book is in the stack directory.
Everything above is enough to build this from scratch in about four hours. The blueprint turns that into about twenty minutes.
It's a JSON file and a base template, so I can't paste them into this page — tell me where to send them.
One email a week about building things that run without you. Unsubscribe link at the bottom of every one, and it works.
Test after every single step. Step seven is not optional — every builder who skipped it has a story.
© 2026 Jeff Sheikowitz. All rights reserved.