Screen tutorial

Preserve the sequence someone needs to reproduce the result: every step, warning and prerequisite, in order.

Workflow

1. Understand. Transcribe narration; inspect the recording (agentWorkspace action source_inspect) to find the actual UI actions and where they happen.
2. Structure. Remove confirmed waiting, loading and failed attempts only. Keep warnings, prerequisites and cause and effect.
3. Point at the action. Identify the actual control from visual evidence before zooming or pointing; never infer coordinates from speech alone. Zoom to the control being used (punchIn with a hold, or scale keyframes), hold long enough to read it, then pull back.
4. Label steps with short on-screen text (a verb and an object) that matches what is shown: addTitle role callout or lower_third.
5. Captions from real speech timing when there is narration (captions skill).
6. Sound. Music, if any, stays low under explanation (musicBed).
7. End on the result: show the finished outcome.

Gotchas

• UI text must stay readable: hold a zoom long enough to read the control, and never zoom so far the pixels break up.
• Use only chapter and navigation commands that exist in the schema (setChapters); do not invent chapter ops.
• Never invent product facts, prices or claims. Ask for what only the user knows.