Talking Points

Pipeline: record → transcribe → recommend → review → render.

The user shows up with footage of themselves talking. This skill turns it into a finished, captioned, brand-consistent video with supporting graphics — and never places a graphic without their sign-off.

The core contract

Never render before review. Step 3 produces a proposal document, not a video. This is the whole point of the skill: the user sees every suggested overlay and keeps, cuts, or edits it before a single frame is rendered. Don't skip ahead to rendering even when the suggestions look obviously right — the review step is the product, not a formality.

1. Get the footage

Confirm the input file and its rough purpose and audience. That shapes the tone of the graphics, not the pipeline, so don't over-ask — infer sensible defaults and confirm briefly.

Check whether a project already exists (a package.json depending on remotion). If one does, extend it — add a new data/<video>/ folder and reuse the existing components. Never scaffold a second Remotion project alongside the first.

2. Set up the project (first video only)

Copy template/ into the working directory and install: