script-storyboard-writer

name	script-storyboard-writer
description	Use when drafting platform-aware hooks, voiceover, segments, storyboard, and blog copy from grounded claims.

Script Storyboard Writer

Inputs

• research/claims.json
• research/timeline.json
• review/creative_direction.md or ogilvy-creative-director notes when available
• Hook patterns and Hook Lab constraints
• platform_profiles/*.yaml
• Topic and audience intent

Outputs

• script/hooks.json
• storyboard/hook_variants.json
• script/voiceover.md
• script/voice_segments.json
• storyboard/storyboard.json
• blog/blog.md
• PDF outline draft

Allowed Actions

• Write hooks, voiceover, segments, storyboard beats, and blog draft from sourced claims.
• Adapt opening style and pacing to the selected platform profiles.
• Keep one core thesis and make the first scene reference the selected hook.
• Record hook choices, rejected variants, and source-backed rationale.
• Apply the Ogilvy Creative Contract: Big Idea, headline as mini-ad, facts before decoration, visual hero, proof object, and brand consistency.
• Apply the second-layer Ogilvy rules: research before creative, caption as micro-headline, consumer language, numbered facts, news-style layout, image captions, avoid reverse type, and avoid ornate fonts.
• Route weak hooks or unfocused first screens through ogilvy-creative-director after technical review and before short-video-opening-optimizer.
• Make the proof object explicit in the storyboard when a beat depends on a paper figure, formula, code snippet, benchmark, or source-backed diagram.
• Treat each subtitle beat as a caption as micro-headline: it should carry a mechanism, benefit, or curiosity gap instead of repeating filler.
• Use numbered facts when a complex explanation has multiple proof points, and keep each item tied to a claim or proof object.
• Add pronunciation notes for ambiguous Chinese particles and technical terms when drafting voiceover prompts; in particular, 动态地 uses the adverbial particle 地 and must be read as de, not di.
• Prefer low-ambiguity spoken phrasing for TTS-sensitive lines, such as 以动态方式建立关系, when the visible script may contain 动态地建立关系.
• For formula-heavy episodes, draft or request a four-channel contract: source_text, spoken_text, caption_text, and visual_text. The contract must cover high-risk notation, product names, polyphonic Chinese phrases, and formula symbols before TTS or render.
• Avoid EP05-class oversimplifications: do not say a method "returns relative distance" when the reviewed mechanism is a position-related term depending on relative displacement; do not repeatedly recite notation such as n-m when a viewer-language explanation is clearer.
• Each Feynman analogy must map back to an exact mechanism in the same scene or adjacent beat, not remain as a standalone lifestyle metaphor.

Forbidden Actions

• Do not use generic openers in the short-video first scene.
• Do not render video, generate keyframes, or synthesize audio.
• Do not use unsourced exaggeration as clickbait.
• Do not change or invent claim evidence while writing scripts.
• Do not decorate before the claim is clear.
• Do not use a headline as mini-ad if the promised proof object cannot appear in the episode assets.
• Do not use puns, obscure references, long negative constructions, reverse type assumptions, or ornate font assumptions as the basis of the hook.
• Do not put production cues such as 读作, 视觉焦点, 教学边界, Hook, or QA into viewer-facing captions or on-screen text.