video-short
Skill0 starsChanged 17 days agoMITView on GitHub
Short-form game clips (8-60 s) for X, Reddit, YouTube Shorts, TikTok and Reels. Picks a proven format, puts the hook in frame 0 and the payoff by 3-5 s, sets length and framing per channel (native 9:16), handles captions and sound for muted feeds, cuts per-channel versions from one moment, and reads the numbers. Use with video-craft whenever a marketing video is a clip or a Short, or when a clip feels slow, unreadable or generic.
In this repository21 other files it gives its agents.
- art-craftSkill
- blender-animSkill
- char-3d-sourceSkill
- char-animSkill
- design-craftSkill
- design-principlesSkill
- elevenlabsSkill
- enemy-designSkill
- fx-craftSkill
- godot-devSkill
- hero-designSkill
- marketing-playbookSkill
- motion-graphicsSkill
- post-voiceSkill
- save-the-catSkill
- skill-designSkill
- ui-craftSkill
- video-craftSkill
- video-longformSkill
- video-trailerSkill
- video-youtubeSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
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 60
Run each format 8-10 times before judging it.
≈ “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
Short-form clips
Load video-craft first (pipeline, tools, QA gate). This skill decides what the clip is and how it is cut. Beat sheets for every format: references/formats.md.
What a short is for
- Until the Steam page exists, a clip earns follows and tells you which hooks hold. Short-form is a booster and a test lab; the demo, festivals and streamers sell games (
video-craftchannels.md "Expectations"). - Find the levers that travel for your game. Fear reads in a second, and so does a striking image. Deep strategy doesn't read in a feed, so a clip shows a feeling and one visible consequence, never the rules.
The hook
- Frame 0 already shows the thing: a strike mid-swing, a monster's eye opening, a number racing up. No intro, title card, dialogue window, logo or slow establishing shot. Every documented miss opened with one of those.
- 3 s retention is similar across clips; what separates hits is the 3→5 s drop. The second beat must pay off or escalate what frame 0 promised.
- Hook types: an impossible image; a number in motion; a threat line ("It can't die."); a question over the picture ("Which one is the hero?"); the result first, then how.
- Any series tag lives in the post text or title, never in the first frame.
Length and framing
| Channel | Cut | Frame | Notes |
| X | 15-45 s | 16:9 or 4:5 | autoplays muted; the post text carries the words |
| 15-45 s | 4:5 (16:9 for desktop-heavy subs) | title read first; mobile crops 9:16 to about 4:5 | |
| YouTube Shorts | 15-30 s | 9:16 native | sound on; the first second is the thumbnail |
| TikTok / Reels | 9-15 s | 9:16 | sound on |
- Make one master per moment and cut each channel's version from it: same footage, own framing and own text.
- Vertical comes from a native 9:16 camera where possible; cropping 16:9 keeps 31.6% of the frame. Compose tall: threat in the top third, the clash in the middle, the player's side above y 1380. Fallbacks: a tracked crop on one subject, or a stacked split (boss above, team below) (tools.md). Blur-fill only for wide shots, and it reads like a repost.
- Loops: if the last frame can meet the first, make it seamless. Replays count, and a hidden detail earns rewatches.
Cutting a short
- One idea, one moment. Cut on events; hold the payoff. 1.5-3 s shots in a hype cut, 3-6 s for one readable mechanic.
- For card games: a card choice and its consequence in one shot or two adjacent shots, card art big and held ≥ 1.5 s.
- For horror: normal → wrongness → hold → sting → the response. One sting per clip, after silence, with a visual twin.
- Sound on, survives muted: SFX on each key action (about one every 2-3 s), licensed audio only, −14 LUFS.
- On-screen text: ≤ 6 words a line, one line at a time during action, in the post's language, inside the safe box. On X the words mostly go in the post, and the video stays pure picture.
- End: cut on the high point. An end card lasts 0.7 s at most; the follow or link ask goes in the post or a reply.
Choosing the format
Pick from formats.md by the material you have and the feeling it serves:
| Material | Formats |
| A fight with a big hit or a combo | F1 cold-open climax · F2 number goes up · F3 synergy reveal · F6 chaos escalation |
| Unit or card picks and a fight | F4 building the team |
| A discovery or hidden-information mechanic | F5 hidden in plain sight |
| A loss or a last stand | F7 the wipe · F8 clutch finish |
| A run that shows the loop | F9 "would you play this?" · F10 the game in 15 s |
| One character in a real setting | F13 tone clash · F14 micro loop |
| A new thing in today's build | F15 "I just added…" |
| A comment asking something the game does | F12 comment reply |
Run each format 8-10 times before judging it.
Checklist (on top of the video-craft QA gate)
- Frame 0 is the strongest moment, and 3-5 s pays it off.
- One idea; the choice and its consequence are both on screen.
- Each channel's file is cut for that channel (frame, length, text language).
- Muted at phone size, the sentence "it's the game where…" comes to mind.
- At most one sting; the clip ends on a payoff.
- Not the same footage as a recent or queued post.
Reading the result
Read at 24 h and 7 d: X views and % watched, Reddit upvotes in the first 2 h and comment quality, Shorts stayed-to-watch and average % viewed. Compare with your own median, never with one viral case. A curve that cliffs early means the hook failed: recut the hook, don't repost the same file.
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.

