ingestlisted

# Recording ingest — data/[channel]/episodes/[topic]/recording/ Turn a screen recording (with voice narration) into **text laid out on a timeline**. The resulting `timeline.md` becomes the primary source that takes the place of the storyboard skill's research (research.md) — what got said and shown in the recording is the material for the scene design. ``` data/<channel>/episodes/<topic slug>/recording/ ├── raw/ # STT, silence, screen-change raw signals (transcribe.sh) ├── timeline.json # machine-readable scene timeline ├── timeline.md # human-readable — scene table + sentence timestamps + screen descriptions ├── keyframes/seg-N.jpg # scene keyframes └── alignment.json # shooting mode only — storyboard scene ↔ recorded cut alignment (§5) ``` ## Procedure ### 0. Recording mode (`record` instead of a file in the argument) If there's no recording yet, **this skill does the recording itself**: ```bash REF=${CLAUDE_PLUGIN_ROOT}/skills/ingest/references bash $REF/record.sh start ~/Movies/social-flow-rec-$(date +%Y%m%d-%H%M%S).mov ``` - Tell the user before starting: demo the screen **while talking into the mic**, and say "recording done" when finished. Add the demo tips too — **pausing for a second or more and then switching the screen** when changing subject makes the scene boundaries land accurately. - **If `storyboard/script.md` exists (shooting mode)**, give them that path before starting and tell them to **put the script on