// Agents and skills //

Data · runs in Claude Code

**What goes in:** the word-by-word transcript (from video-transcricao), the project's art direction (`da/da.json`, from video-direcao-de-arte — both come inside the zip), and a list of scenes where you don't write times — you write **the line the person says**. **What comes out:** the table of resolved anchors (each scene with the second it comes in, its duration, and what it matched in the speech), a preview PNG per scene for you to look at before spending render time, the scene as an MP4 — or a MOV with alpha, for badges and inserts — and `som/eventos_motor.json` with the points where the entrance sound should hit. There are 13 ready-made scenes: photo, typewritten text with highlight, counting number, headline, quote with stamp, board of people connected by strings, timeline, document, infographic, map, insert, title card, and badge. No literal colors or fonts — everything comes from the art direction tokens. **Who it's for:** people who narrate on video and want the screens to stick to the speech, with the same look from start to finish.

• Data
• Runs in Claude Code
• Intermediate