reels-scripting.

Short-form video rewards format mimicry more than originality. The hook patterns, pacing, and rhythm of an outlier Reel are reusable in a way that the topic is not. reels-scripting is the skill that automates the reverse-engineering pass, scraping the reference, analysing the video, and writing a fresh script that applies the same structural moves to the writer's own newsletter content.

The premise behind this skill is that outlier short-form videos are outliers because of structure, not because of topic. The hook lands in the first two seconds because of how it is constructed; the script holds attention because of how the language is paced; the close converts because of where the call sits in the runtime. Apply the same structure to a different idea and the result is a video that performs against the same shape, even though the content is unrelated to the reference.

§01What it does

The skill takes a Reel URL or a Notion link to one (it will follow the Notion page and extract the embedded Reel URL). Then it asks for the newsletter topic the writer wants to repurpose, and reads newsletter-voice.md, voice.md, and about-me.md from the project so the new script lands in the writer's voice rather than the reference creator's.

The scrape pass writes a small Node script that calls Apify's apify/instagram-reel-scraper with the URL, downloads the resulting .mp4 to ~/Desktop/Reels/downloads/, and saves the metadata (views, likes, comments, caption first 200 chars) for context. If the primary actor returns nothing, it falls back to two other Instagram scrapers in order.

Then it sends the video to Gemini 2.5 Flash with a structured analysis prompt: full transcript with timestamps, exact hook wording with word count, language patterns including average sentence length and you-versus-I ratio, pacing markers, and the structural arc. With that analysis in hand, Claude writes a new script that applies the same hook shape, pacing, and structure to the writer's newsletter topic.

§02The reference is the form, not the substance

The interesting choice this skill encodes is its strict separation between what the reference contributes (structure) and what the writer contributes (substance). Most "make a video like this one" prompts let the reference's topic leak through into the rewrite, which produces scripts that read like topical knock-offs of the original. By sourcing the substance entirely from the writer's newsletter and using the reference only for structural cues, the skill produces videos that look familiar in shape but carry the writer's actual ideas.

This is also the only skill in the bundle that is an honest two-tool composition. Apify does the scraping job. Gemini 2.5 Flash does the video-understanding job (which Claude cannot do natively at usable cost). Claude does the writing job. Each tool is doing the part it is best at; the skill is the choreography.

§03Setup

# Required env vars:
export APIFY_API_TOKEN=your_token
export GOOGLE_AI_API_KEY=your_key

# Required Node packages:
npm install apify-client @google/generative-ai

# Trigger phrases:
#   "script a reel"
#   "reels scripting"
#   "turn this into a reel"
#   (or paste an Instagram Reel URL)

The Node script the skill writes lives at ~/Desktop/Reels/analyse-reel.js. Inspect it on first run; the skill is happy to explain any of it on request.

§04Caveats

Apify scrapes cost money in aggregate. The skill is built to scrape one reference at a time, which keeps the unit cost predictable. Resist the urge to bulk-scrape "all my outliers"; the per-script analysis is what makes the output good.

Gemini 2.5 Flash is fast and cheap relative to alternatives, but video analysis still has a latency floor. Expect a minute or so between submitting the video and receiving the structural analysis. The skill writes that wait into the workflow rather than trying to hide it.

A reference Reel that is structurally weak will produce a script that is structurally weak. The reference selection is the most important decision in this skill. Pick outliers; do not pick whatever just happened to scroll past.