What problem does it solve? AI agents handed a brief like "make a 5s launch film" default to producing a slideshow of disconnected beats (hook card, feature card, logo card) that passes automated gates but nobody watches twice. This Skill replaces that failure mode with a planning grammar where one object stays on screen from first frame to last and every cut is that object changing state. ## Core Features & Use Cases - Object-spine planning: A seven-step method that names one object, writes its state at every beat, and treats every cut as a transform (no cut, match cut, or hard cut with a carried element), never a jump. - Measured time budget: Concrete per-second costs read off a frame-measured reference film (33 chars/sec for read copy, 0.4s for a morph, 0.6s minimum final hold) that enforce three beats and roughly 20 words at 5 seconds. - Pipeline-compatible storyboard output: Emits a STORYBOARD.md with frontmatter (object, object_t0, object_states, object_last) and per-beat type/onscreen/why/becomes fields that make storyboard-check and make intent already consume, plus seven anti-slop tests (reorder, object-death, pixel-jump, who-moved-it, logo end-card, unbacked-claim, payoff). - Use Case: Given a brief for a 5-second launch film for a voice-notes app, produce a storyboard where the record pill swells, flattens into a list row via match cut, and unfolds into a structured note card, then validate it with the repo's gates before writing any scene JSON. ## Quick Start Ask the agent to plan a 5-second launch film for your product as one continuous action using the vawe-continuous-action skill, naming the single object and its state at every beat before writing any JSON.
