agentleFS
guillempucheguillempuche/shorts-studio/.agents/skills/make-shorts/SKILL.md

make-shorts

Skill2 starsChanged 4 days agoApache-2.0View on GitHub

Run the daily short-video loop with shorts-studio - from audio or a brief to a reviewable draft per post, with the user approving each script and its image prompts before anything is spent on images. Use when asked to make, continue or batch short videos, reels or TikToks, or to prepare today's posts.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 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.

  • 2

    Very specific

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

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

    Write the story to script.md (hook in the first 3 s, the turn, the CTA): concrete subjects, one idea per image, the brand style is added …

    line 23
  2. Very specific:

    Stop: Ask the user to read the script and the prompts, and run script approve --post <slug> in their own terminal.

    line 25
  3. Distinctive:

    After approval: steps 8–9 for each approved post.

    line 38

≈ “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.

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

Make shorts

The order of the steps. The rules behind them are in AGENTS.md; commands and their flags come from shorts-studio --help or the MCP tool list.

Before you start

  1. doctor: everything must be ✓. Stop and tell the user about any ✗.
  2. post list: continue posts already in progress before starting new ones. post show --post <slug> names the next step every time.

Flow A: original audio (a reel, a recording, a track)

  1. post create --post <slug> --title "…"
  2. source import --post <slug> --file <audio-or-video> --origin mine|others|ai --kind speech|music. Use others for someone else's reel: it's a guide track, and the post must be published with the platform's in-app sound.
  3. audio transcribe --post <slug>. It costs a little, counted toward the caps: on a long file, show the --dry-run estimate first. Add --language es only when the auto-detected language is wrong.
  4. beat plan --post <slug>, then direct the beats with the story-direction skill: mark each beat's moment, re-cut with --by-moment, then write one image prompt per beat.
  5. Write the story to script.md (hook in the first 3 s, the turn, the CTA): concrete subjects, one idea per image, the brand style is added for you.
  6. script set --post <slug> --file script.md --hook "3 to 7 words", then beat set --post <slug> --from beats.json.
  7. Stop. Ask the user to read the script and the prompts, and run script approve --post <slug> in their own terminal. Never approve it: it refuses without a terminal anyway. Changing a prompt or the audio afterwards needs approving again.
  8. image generate --post <slug> --dry-run, show the estimate, then run it for real (the CLI needs --yes; over MCP the user's approval prompt is the gate).
  9. draft render --post <slug>, then give the user the path to draft.mp4 for the evening check.
  10. Stop. The user reviews it with post review --post <slug> --decision approve|fix|reject (CLI only). For fix, read post show for their notes, change what they name (beat set, image generate --beat N --reroll, image import), and render again. A new prompt needs the user's script approve before image generate.
  11. Once approved: capcut export --post <slug>; the user polishes in CapCut desktop (steps in capcut/README.txt). Then post finalize --post <slug> --file <final.mp4>, post publish once per platform, and post archive when it's done.

Flow B: AI voice from a brief

Voice generation isn't built yet (openspec/changes/voice-music-animate). Until it is: write the script, have the user approve it, record or generate the voiceover elsewhere, and continue from Flow A step 2 with --origin ai.

Five a day

  • Morning: take each post to step 7 and hand the user all five scripts at once.
  • After approval: steps 8–9 for each approved post; evening: steps 10–11.
  • budget show before generating, so the day's spend stays visible.
  • Re-running any step is safe: finished work is skipped, an interrupted image request resumes without paying again.

When something fails

Every error has a fix: do what it says. over-budget means stop and ask the user; never raise caps yourself. needs-approval means the step is the user's: stop and ask them. needs-confirmation means show the user the estimate, then run the same command with --yes. A failed error that names a setting means stop and tell the user: their caps may not apply until doctor passes.

nextSteps in every result says what comes next as data: run a step whose by is agent (its tool and args over MCP, its cli in a shell), and hand the user each step whose by is user.

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.