hook-finder
Skill1 starsChanged 2 days agoMITView on GitHub
Finds the most clippable moments in a game and turns them into short-form video plans (TikTok, YouTube Shorts, Reels) with a 2-second hook, on-screen text and call to action, plus a script to cut vertical clips from screen recordings. Use when the user wants to make a trailer, short video, TikTok, gameplay clip, GIF, or asks "what is my game's hook" or "quay clip gì cho game".
In this repository5 other files it gives its agents.
- ai-disclosureSkill
- community-launcherSkill
- game-contextSkill
- platform-pickerSkill
- steam-page-wishlistSkill
The file
We split this skill into the rules it gives an agent
4 rules. About 0 are common, 2 are distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Read .agents/game-context.md (run game-context first if missing).
- Distinctive: line 17
Mine moments: Ask the user to describe or record 2–5 minutes of play.
- Distinctive: line 56
Don't promise virality. Viral case studies are survivors.
- Very specific: line 57
If the clip shows AI-generated art, follow the user's disclosure plan (ai-disclosure).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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
Hook Finder
Short video is the cheapest way for an unknown game to reach strangers. The job: find 3–5 moments a stranger would stop scrolling for, then plan clips that test them.
Language
Answer in the user's language. Write on-screen text in the language of the target platform audience (ask which country or language audience the clip targets; one clip per language works better than mixed subtitles).
Workflow
- Read
.agents/game-context.md(rungame-contextfirst if missing). - Mine moments. Ask the user to describe or record 2–5 minutes of play. Score candidate moments with the 6 hook types below. If the user has a recording, ask for timestamps of interesting parts.
- Pick 3–5 moments with the highest score. For each, write a clip plan (template below).
- Cut the clips with
scripts/cut_clips.pyif the user has a recording and ffmpeg. - Set up the test: post variants over 1–2 weeks, record results in the table below, keep the winner's pattern.
- Save the chosen moments to the "Most clippable moments" line in
.agents/game-context.md.
The 6 hook types (score each moment 0–2 per type)
| Type | Stops the scroll because… | Example |
| Satisfying | Repetitive, tactile, completes something | Stripping a cluster of ripe red pepper berries |
| Strange / novel | Viewer has never seen it in a game | A game about a Central Highlands pepper farm |
| Funny / physics | Something goes wrong in a silly way | Ladder slips, pepper sack rolls downhill |
| Atmosphere | Mood is instantly felt | Dawn mist over the plantation, rain on a tin roof |
| Local / cultural | "That's so us!" recognition, meme-able | Phin coffee dripping, specific local food, slang |
| Challenge / question | Viewer wants to answer or try | "Can you harvest 100 kg before the rain?" |
Moments scoring high on 2+ types are the best candidates. Atmosphere and local detail work for cozy games even without action.
Clip plan template
### Clip <n>: <name>
- Hook type(s):
- 0–2s (hook): <first frame must already show gameplay — no logo, no menu fade-in>
- 2–15s: <what happens>
- On-screen text (≤ 8 words):
- Ending / CTA: <"Wishlist on Steam" | "Play free — link in bio" | question to drive comments>
- Length: 10–25s
- Sound: <game audio / trending sound / voiceover>
- Source timestamps: <start–end in recording>Rules for every clip
- Gameplay in frame 1. No studio logos, no slow intros.
- Vertical 9:16, 1080×1920; keep text away from the bottom ~20% (UI overlays).
- One idea per clip. Make 3–5 variants of the strongest moment (different first second, different text) rather than one "perfect" trailer.
- Post consistently; most clips get little reach and that is normal. Judge a pattern after several posts, not one.
- Don't promise virality. Viral case studies are survivors; many devs post regularly with little result.
- If the clip shows AI-generated art, follow the user's disclosure plan (
ai-disclosure).
Results table (keep in the project, e.g. marketing/clips.md)
| Date | Platform | Clip | Variant | Views 48h | Avg watch % | Likes | Comments | Link clicks / wishlists |
Cutting clips
scripts/cut_clips.py cuts vertical clips from a landscape recording using ffmpeg:
python3 scripts/cut_clips.py recording.mp4 --clip 00:01:12 15 harvest --clip 00:03:40 12 dawn --out clips/Each --clip takes START DURATION NAME. Add --no-crop to keep the original framing (letterboxed to 9:16). Requires ffmpeg on PATH.
See references/hook-examples.md for case studies and what to learn from them.
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.

