recoup-content-make-videolisted
Install: claude install-skill recoupable/skills
# Recoup Content — Make Video
Every motion job, one skill. Read the bundled references first:
`references/workspace-context.md`, `references/account-resolver.md`,
`references/research-context.md`, `references/content-api.md` (the video modes +
async create→poll), `references/song-sourcing.md` (real audio, no placeholders),
`references/analyze-gate.md` (verify before claiming done).
## Backbone
Resolve the artist (use the **`account_id`**, not row `id` — wrong one 404s), read
context first, layer live signals, generate, **analyze-gate** (you can't see motion —
analyze and regenerate on failure), persist + commit. Real audio only.
## Modes
- **short** — finished 9:16 of the artist + song. The look is a **template**
(bedroom/stage/outside/album-record-store; default `artist-caption-bedroom`; list
via `/content/templates`). Fire `POST /content/create`, poll `/tasks/runs?runId=`
every ~10s until terminal, read `output.{videoSourceUrl,captionText}`. The async
path is agent-safe (sync `/content/video` times out).
- **lyric** — transcribe for word-level timings (`POST /content/transcribe`,
`audio_urls` array → `segments:[{start,end,text}]`), generate/use an on-brand
background, **burn in synced lyrics**, mux full-length. Don't reproduce third-party
lyrics.
- **visualizer** — seamless loop, **no text/logos** (Spotify rejects them), 9:16,
3–8s; generate with identical `image_url` and `end_image_url` frames, then trim
if needed; analyze-gate the seam.
- **reformat