agentleFS
aya-talebaya-taleb/multimedia-skills/skills/youtube-shorts/SKILL.md

youtube-shorts

Skill1 starsChanged 23 days agoMITView on GitHub

Find the Shorts hiding in a long video: reads its timestamped transcript, picks 3 to 8 moments that make sense on their own, and gives exact cut times on sentence edges, the opening line, on-screen text, a title and a vertical reframing note for each, ranked. Can also cut the clips from the user's own file with ffmpeg on request. Use it for "turn my podcast / webinar / video into Shorts", "find clips in this video", "what should I cut for Reels or TikTok", "repurpose my long video", "best moments from this transcript", or after transcribe on a long recording. Works on the user's own files only, never downloads.

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 5 edited versions we read keep these.

  • 1

    Distinctive

    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:

    If there is only a YouTube link, do not download it.

    line 28
  2. Distinctive:

    If the user wants the files, and has the video and ffmpeg, cut each clip from their file:

    line 75

“Common” is measured: kept by at least 6 in 10 of the 5 edited versions of this file we could read. “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

youtube-shorts

A long video often holds a few moments that work alone: a clear tip, a strong opinion, a short story with a point. This skill finds them in the transcript, so the founder does not have to rewatch an hour of footage, and says exactly where to cut.

What this skill reads and writes

  • Reads: a timestamped transcript (media/transcripts/<name>.json or .srt from the transcribe skill, or a YouTube Studio transcript the user exported). The founder-context business file for audience, tone and off-limits topics. Optional: the video file.
  • Writes: media/shorts/<name>-shorts.md, and cut clips in media/shorts/ only if asked.
  • Next: talking-head-cards to add cards to a clip, the linked edit skill for captions and tightening, youtube-seo for the long video itself.

If there is only a YouTube link, do not download it. Ask for their own file or an export, and run transcribe on it.

1. Read the whole transcript first

Skim it end to end before picking anything. Note who the business file says the audience is and what they care about. Clips that serve that buyer beat clips that are merely lively.

2. Pick moments that stand alone

A good Short candidate:

  • Makes one point a stranger understands with no setup. If it needs "as I said earlier", it does not stand alone.
  • Starts strong. The first sentence states the point, a surprise, or a problem. If the best line is in the middle, note a "cold open": start the clip on that line, then the rest.
  • Ends on a finished thought, not a trailing "so yeah".
  • Fits Shorts length: up to 3 minutes for square or vertical video (YouTube's limit since October 2024; see references/youtube-facts.md). Most good clips run 20 to 60 seconds. Shorter is usually better. A Short over one minute with a copyright claim (music in the background, for example) gets blocked, so flag any music.

A worked example, with patterns that do and do not stand alone and how to note a cold open, is in references/picking-clips.md.

Rank them by how clearly they stand alone and how well they serve the audience. Say why for each. Do not predict views: nobody can, and a guess dressed as a number misleads.

3. Write each clip card

For each clip, in media/shorts/<name>-shorts.md:

  • Rank, start and end time (on sentence edges, from the transcript's own timestamps), length.
  • Why it stands alone, in one line.
  • Opening line, exactly as spoken. Cold open note if needed.
  • On-screen text for the first 2 seconds: the point in a few words, from what is said.
  • Title: plain, specific, under 100 characters, no bait.
  • Reframe note for vertical: where the speaker sits in a wide frame, any slide or screen that must stay visible, two-speaker clips that need a split.
  • Anything to check: names, numbers, claims, music, off-limits topics, other people who appear and may need to agree to the clip.

Everything quoted is the speaker's words. Titles and on-screen text frame the clip; they do not add claims, numbers or results the speaker did not say.

4. Cut the clips (only if asked)

If the user wants the files, and has the video and ffmpeg, cut each clip from their file:

ffmpeg -ss 00:12:04.200 -to 00:12:41.900 -i media/raw/episode.mp4 -c:v libx264 -c:a aac media/shorts/episode-clip1.mp4

Re-encoding (not -c copy) keeps the cut on the exact frame. This leaves the clip wide; the vertical reframe is a job for an editor or the linked edit skill. Never overwrite the source.

5. Hand back

Show the ranked list (rank, time, length, opening line, title) and the file path. End with one line naming the next step. The founder uploads the Shorts; this skill never posts.

Rules

  • Their own file or transcript only. No downloading.
  • Quotes are exact. No invented claims, stats or results in titles or text.
  • No view predictions.
  • Respect the business file's off-limits list, and flag other people who appear in a clip.

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.