auto-media-maker

/auto-media-maker

Build and maintain a project's video series — and keep track of which videos the code has since outrun.

The standing law: a project's video series is part of its surface area. When the product changes, the videos that teach it are stale until proven otherwise. A stale tutorial is not cosmetic — someone follows it, it does not match, and that becomes a support ticket or a refund.

The hard part is not making video. It is that six weeks later the login screen moved, the tutorial still shows the old one, and nobody notices until a customer follows it.

The Family

Three skills, one pipeline. This skill owns the shared foundation — the pipeline, the interview, the brand folder, the never_say list, the catalog, the watch/on_change tracking, blessing and versioning — and every video except two. /auto-media-onboarding owns the first-run onboarding video. /auto-media-stinger owns the main stinger commercial. They read the state this skill writes and never fork it. If either runs first in a project, it bootstraps only the part of Steps 1–2 it needs, writes it here, and reports the series as unplanned — so a later run of this skill finds the state and does not re-interview.

Step 1 — Find Or Establish The Pipeline

Ask for the media pipeline repo on first run — the repo holding credentials, project configs and video state, kept separate from the product repo so keys never live beside code. Persist its path in .agent/media.json (gitignored) so later runs skip this.

If a pipeline exists, read its conventions and drive its CLI — do not build a parallel one. A mature pipeline typically provides:

Command	Purpose
doctor	tools + credentials present
init <slug> --repo PATH	register a project
add-video <proj> <slug> --kind …	scaffold script, memory, build
status / check / plan	inventory, what the code outran, the work order
render / build / bless	narration, assemble, mark "matches the code as of now"