yiibulisted

# yiibu — Video Post-Production All shell commands in this document run from the skill root (the directory containing this SKILL.md), wherever it is installed. Automated pipeline for turning raw selfie recordings into polished 9:16 short-form videos. ## Division of labour **The user shoots. You select.** They hand over a folder; deciding which shots earn a place, in what order, and how long the result runs *is* the edit. Do not ask them to pick clips — that is the work. Do ask when a fact on screen is unreadable, because they were there and you were not. **The hook is the one exception, and it is a deliberate one.** Ask which moment was the best. It is a single shot out of twenty-odd, it decides whether anything else gets watched, and the thing that makes it work — that the moment MEANT something — is not in the pixels. On 2026-09-01 this was tested rather than assumed: two signals were measured across a 23-clip folder against a hook the user had already chosen. Face area "found" it (10.2% against 0.0% for the shot picked instead) but only by luck; shipping it would have pulled every future hook toward a face, and a cheese pull, a pour or steam has none. Motion and event-spike did not find it at all — it ranked 7th, BELOW the wrong pick, because pixel motion measures the camera panning, not the content. The user's verdict was 「跟臉沒關係，主要是內容精彩吸引人」, and nothing cheap reads 精彩. They were at the table; you were not. One question beats any amount of measurement here. New to th