Presenter-led explainer
The human is the presenter. Your job is everything around them: shape the idea, write a script they can read or listen to, build a deck they can talk to, get the recording running in the browser, and edit the take afterwards. You never synthesize their voice.
Workflow
- Discuss the idea. Settle three things with the user before writing anything: the question the video answers, the one core idea, and the audience. Keep the video to one idea. Ask only what you cannot decide yourself.
- Scaffold. From the user's project root run
node <skill>/scripts/new.mjs <slug> [--title "..."] [--lang sv]. This createsvideos/NNN-slug/withscript.md,deck.json,scenes/,runtime/andtakes/. - Write
script.md. One section per slide, in the language the user will speak. It is a listening script: the user reads or hears it to prepare, then talks freely on camera. They do not read it aloud. Short sentences, concrete examples. - Build the deck. One slide = one entry in
deck.json(id,title, 2-4points) + one filescenes/<id>.html. Readreference/deck-format.mdfor the page contract andstyle/STYLE.mdfor the look. The points are the presenter's reminders: each arrow-key press reveals the next point, and the scene shows one handwritten caption per point. Runnode <skill>/scripts/build.mjsafter changingdeck.json. - Review with the user. Run
node <skill>/scripts/open.mjs. The user steps through the deck (arrow keys, T = talking points). Iterate until they approve the script and the slides. Do not record before they approve. - Record.
node <skill>/scripts/record.mjsserves the deck on localhost and opens it in Chrome or Edge. In the page: F11 for full screen, R starts (3-2-1) and R stops, arrow keys step through the points, N opens the talking points in a separate window that is not recorded. The webcam appears as a bubble in the corner. Takes land invideos/NNN-slug/takes/. The user presents a few takes. Details:reference/recording.md. - Edit.
node <skill>/scripts/transcribe.mjs takes/<file>gives a transcript. Then followreference/editing.md: pick the best take, cut mistakes and dead air, add chapters, write the title and description. Keep the original takes.
Rules
- Recordings (
*.mkv,*.mp4, audio) stay local and gitignored. Never commit them or upload them anywhere without asking. - Never ask for, print or store API keys. Nothing in this skill needs one.
- Follow
style/STYLE.mdunless the user gives another direction. Hand-authored SVG only, no raster images in decks. - Keep each slide simple. If a scene needs more than a few objects and one motion, split the slide.
- Do not commit or publish unless the user asks.