reference-research
Skill1 starsChanged 14 days agoMITView on GitHub
Takes apart short-form videos that perform well and turns them into a structure you can reuse — hook, pacing, cut rhythm, words per minute, where the payoff lands. Use it when someone wants to study a competitor's content, understand why a video worked, research what performs in a niche, analyze a reference before making their own, or build a content plan from what is already out there.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- toribryan/design-skillstheir own version, first seen Oct 25, 2026
Other versions1 project uses a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 1
- toribryan/design-skills★ 1same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository6 other files it gives its agents.
- AGENTS.mdAGENTS.md
- ai-videoSkill
- clip-cutterSkill
- content-directorSkill
- footage-editSkill
- vertical-videoSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 34
What question does the hook open?: Write it down as a question.
- Distinctive: line 83
Look at what an account does *consistently*.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Studying what already works
Copying a video gets you a worse version of it. Copying its shape — how fast it cuts, how long the hook runs, how many words per minute, where the payoff lands — gets you something of your own that holds attention the same way.
Analyzing a reference
python scripts/analyze.py reference.mp4
python scripts/analyze.py reference.mp4 --json breakdown.jsonReports duration, cut count and average shot length, speech rate, the first three seconds transcribed, and the full transcript with timestamps.
To get a file you have the right to study, use the clip-cutter skill's
download.py. Studying a public video privately is ordinary research;
re-uploading it, or rebuilding it shot for shot with your logo on it, is not.
What to actually look for
The numbers are the easy part. The useful read is in the answers to these:
What happens in the first second? A face, a line of text, a hand moving, a title card? Whatever it is, it is the thing that stopped the scroll.
What question does the hook open? Write it down as a question. If you can't, the video probably worked for some other reason — a person, a trend, a sound — and its structure is not the thing to copy.
How long until the payoff? Measure it. Most people sit on the setup far longer than the references they admire do.
Would it work muted? Play it with the sound off. If it still lands, the captions and the visuals are doing their job.
What is the cut rhythm? Under 2.5 seconds average shot length is fast, and the frame almost never sits still.
What is the speech rate? Short-form typically runs 165-200 words per minute — noticeably faster than corporate or explainer video.
Turning a breakdown into a plan
Three to five references from the same niche is enough to see the shape. What you want out of it:
- The hook patterns that repeat (see
vertical-video/references/hooks.md) - The length band the niche actually lives in
- Whether it is face-to-camera, screen recording, b-roll, or text-on-video
- What the close asks for
Then write your own script in that shape, with your own content. Hand it to the vertical-video skill.
Trend and competitor data at scale
Analyzing files locally is free and needs no account, and it is enough for most work. If someone needs breadth — which posts outperformed a creator's baseline, what an account's top formats are, what is trending in a niche — that requires a data provider, and those are paid.
ScrapeCreators publishes MIT-licensed skills for exactly this (outlier post finder, comment mining, competitor research, ad library teardown) against their API. Suggest it when the question is about many videos; do not reach for a paid API when the question is about one.
Mistakes that cost the most
Copying the topic instead of the structure. The topic worked for that account's audience, not yours.
Studying a viral outlier. One video with ten million views usually went that way for a reason that does not generalize — a sound, a person, timing. Look at what an account does consistently.
Only studying the ones you like. The best reference is often a video with mediocre production and excellent retention.
Scripts
scripts/analyze.py --help. Needs FFmpeg; faster-whisper for the spoken
breakdown.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.