agentleFS
berkayburakkberkayburakk/tubecrew/skills/make-shorts/SKILL.md

make-shorts

Skill2 starsChanged 28 days agoMITView on GitHub

Turn a finished long video into YouTube Shorts - pick 3-5 self-contained cuts from the script and word timings (re-ranked by retention data when it exists), reframe them to vertical 1080x1920 with re-timed captions, and gate each Short in code (square or vertical, at most 3 minutes, an opening that moves in the first 2 seconds, captions out of YouTube's UI zones) before the render-auditor and a private upload. Use for "make Shorts from this video", "find clips for a Short", "cut a vertical version", "is this Short ready". Not a trailer maker.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈0

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    All of render_check.py's measurements (loudness, silence, black and frozen picture, the motion band, captions against the words) with the …

    line 57

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Make Shorts

A Short cut from a long video is a new video for a stranger who swipes. It carries one idea from its first line to a payoff, and it is not a trailer: "watch the full video to find out" leaves the viewer with nothing. The rules, the platform facts and their sources are in $SKILL/references/shorts-rules.md.

The flow, one gate at a time:

pick → the creator approves the cuts → cut → check → render-auditor → upload-checker → upload-private (private)

$SKILL below is this skill's folder. Start from a finished long video: script.md, the narration's word-timed transcript (qa/voice-check-transcript.json from check-voiceover) and the final render.

1. Pick

python $SKILL/scripts/shorts_pick.py <video> [--retention qa/retention.json] [--min 20 --max 60] [--count 5]

Runs of whole sentences are scored on measurable signals: the pocket rule (the first line works with no context: no "And ...", "So ...", "This is why ...", no intro or "subscribe"; a window that breaks it never passes), a number or a named thing first, a question the window answers, a short last line that loops back, one script section, a length near 25-45 s. Times snap to word boundaries and sentence ends. Exit 1: no self-contained moment exists, so write one rather than force a cut. After 72 hours, pick again with --retention (retention-by-scene's --json report): rewatch bumps rank up, windows opening in a steep drop rank down.

Show the creator the table and let them choose. Cut nothing before they approve: a line is cheap to change, a re-uploaded Short is a new video.

2. Cut

python $SKILL/scripts/shorts_cut.py render/final.mp4 --candidate 1 --words qa/voice-check-transcript.json \
    [--mode fit|crop] [--fill 0.2] [--x 640] [--burn] [--out PATH]
  • fit (default): the whole picture on a blurred copy of itself; --fill trims the sides so it grows. Use it for charts and slides. crop: a 9:16 slice (--center or --x); use it when one face or object carries the shot.
  • Captions are a separate .srt, re-timed to the cut (--srt re-times the long video's captions, --words builds short cues). Read them: transcripts mishear names and numbers. --burn draws them above YouTube's UI zones (needs ffmpeg with libass; without it the script stops and says so).
  • Writes shorts/<name>.mp4 and never touches the source. --out writes straight into the Short's own folder.

3. Check

python $SKILL/scripts/shorts_check.py shorts/<name>.mp4 --srt shorts/<name>.srt --words shorts/<name>.words.json \
    [--burned] [--ocr] [--profile live]

All of render_check.py's measurements (loudness, silence, black and frozen picture, the motion band, captions against the words) with the whole Short as the hook and no end-screen allowance, plus:

CheckFails when
aspectlandscape (YouTube treats it as a regular video)
durationover 180 s (WARN within 2 s of it)
openingthe first 2 s neither move nor show text (OCR, or captions burned in from the first second)
safe_zoneburned captions sit in the bottom 25% or right 15% (OCR finds there only warn)

Resolution under 1080x1920 warns; the loop and music over 60 s are information. Writes qa/shorts-check-<name>.json bound to the video's SHA-256. Exit 1: fix the Short. Exemptions work as in check-render, except for aspect and duration.

4. Review and upload

Each Short is its own video, so it gets its own project folder for the reviews and the upload: render/<name>.mp4 (cut there with --out, or copy it), upload.json with "short": true, qa/. Run shorts_check.py there again.

  1. render-auditor: the same contract (VERDICT: and Video: lines, saved as qa/RENDER-AUDIT-<date>.md). Tell it this is a Short: judge the first 2 seconds muted and with sound, the payoff and the loop, text in the UI zones, and expect no end screen. Give it qa/shorts-check-<name>.json instead of render-check.json.
  2. upload-checker: title and description for a stranger (no "Part 2", no long-video chapters).
  3. upload-private: the dry run checks the Short is square or vertical and at most 180 s and reads the Shorts gate; the thumbnail is not sent (Studio on a computer only). Then --upload: private, as always.

Then tell the creator what is left in Studio: the Short's thumbnail, the related video that links it to the long video (Studio only), and when to publish. Spread Shorts over days, never several on one day, and publish a cut only after its long video is public. Cadence and testing advice: references/shorts-rules.md.

What to report back

The candidates table (or the approved cut), the path of each Short, its check verdict, and the single next action.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.