القائمة
视频制作策划。四阶段交互式策划流程：Creative Memo → Resources → Script → Storyboard。包含完整的影视标准（剪辑七律、景别、机位、灯光）、盖曼讲故事法则（叙事引擎、冰山理论、叙事声音、灵感堆肥、人物推进、幽默感、对白法则）、视听语言系统（情绪曲线、景别/运镜/节奏/张力/声画关系/转场）、文档模板和审批流程。
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Remotion Shot 组件实现。将 storyboard 转化为 React 组件。包含关键的帧计算规则（Math.round 强制）、interpolate 安全规则（最小 1 帧）、1080p 视觉标准、动画编排模式、音视频同步架构、组件复用模式。这是技术密度最高的 skill。
Remotion 资产管道。调用 project_migrate tool 将 raws/ 素材迁移到 template-project/public/ 并生成 resourceMap.ts、constants.ts、timing.ts。Agent 只需 edit constants.ts 中的 COLORS 和 FONTS。
音频素材生成与获取。批量 Edge TTS 旁白生成（支持 storyboard pacing 字段驱动语速）、BGM/SFX 检索（BGM 节奏匹配 BPM 规则）、音频时长提取。包含 Edge voice 配置、速度调整规则、durations.json 格式规范（含 audio_visual_relation 说明）和关键的音频时序规则。
视觉素材获取与生成。根据 resources.yaml 定义，通过库存图片/视频搜索（Pexels/Pixabay）或 AI 生图（Caro LLM）获取视觉素材。包含搜索策略、视觉隐喻搜索（counterpoint 声画关系）、构图感知搜索（framing 字段）、prompt 工程指南、素材质量评估标准。
旁白脚本口语化处理与情绪节奏标注。
所有 agent 共享的协议定义。包含资源类型注册表、storyboard schema、dispatch context schema、amendment action 类型。Planner 写入时遵守、Media 处理时校验、Builder 实现时参考、Orchestrator 调度时构造。
| name | carocut-planner-planning |
| description | 视频制作策划。四阶段交互式策划流程：Creative Memo → Resources → Script → Storyboard。包含完整的影视标准（剪辑七律、景别、机位、灯光）、盖曼讲故事法则（叙事引擎、冰山理论、叙事声音、灵感堆肥、人物推进、幽默感、对白法则）、视听语言系统（情绪曲线、景别/运镜/节奏/张力/声画关系/转场）、文档模板和审批流程。 |
Interactive video production planning that generates four core documents through guided user interaction. Each stage requires explicit user approval before proceeding.
This skill guides users through professional filmmaking workflows, generating production documents that follow industry standards. The output drives all subsequent implementation phases.
你必须将文件写入以下精确路径。禁止修改文件名、禁止创建新目录、禁止使用项目名或自定义名称。
| Document | EXACT Path (不可修改) | Purpose |
|---|---|---|
| Creative Memo | manifests/memo.md | Vision, tone, technical specs |
| Resources | manifests/resources.yaml | Asset and character definitions |
| Script | manifests/script.md | Voiceover lines with [VO_XXX] markers |
| Storyboard | manifests/storyboard.yaml | Shot-by-shot breakdown |
违规示例（绝对禁止）：
memo/killer-story.mdstoryboard/my-storyboard.yamlmanifests/killer-memo.md如果 dispatch_context 中包含 output_rules 字段，以其中的绝对路径为准。
digraph planning_workflow {
"Analyze Materials" [shape=doublecircle];
"Stage 1: Memo" [shape=box];
"User Review" [shape=diamond];
"Stage 2: Resources" [shape=box];
"Stage 3: Script" [shape=box];
"Stage 4: Storyboard" [shape=box];
"Complete" [shape=doublecircle];
"Analyze Materials" -> "Stage 1: Memo";
"Stage 1: Memo" -> "User Review";
"User Review" -> "Stage 1: Memo" [label="revise"];
"User Review" -> "Stage 2: Resources" [label="approved"];
"Stage 2: Resources" -> "User Review";
"User Review" -> "Stage 3: Script" [label="approved"];
"Stage 3: Script" -> "User Review";
"User Review" -> "Stage 4: Storyboard" [label="approved"];
"Stage 4: Storyboard" -> "User Review";
"User Review" -> "Complete" [label="approved"];
}
Each stage: Ask questions -> Generate document -> Show to user -> Accept feedback -> Iterate -> Confirm approval -> Next stage
All outputs must implicitly follow these principles:
Shot Sizes: ECU (Extreme Close-Up), CU, MCU, MS, MLS, LS, ELS, XLS
Camera Movements: Static, Push In, Pull Out, Pan, Tilt, Dolly, Track, Crane, Handheld
Angles: Eye Level, High Angle, Low Angle, Dutch Angle, Overhead, Worm's Eye
Lighting References: Key/Fill/Back, Hard/Soft, Natural/Motivated
Forbidden Language: "amazing", "stunning", "breathtaking", "beautiful" - use technical precision instead
Based on Neil Gaiman's MasterClass "The Art of Storytelling". These principles must be internalized and applied throughout all four planning stages. They are not optional add-ons — they are the foundation of every creative decision.
Every video needs a narrative engine — the invisible force that compels viewers to keep watching. Before writing any script, identify:
让别人从你的视角看世界，比直接陈述经历更重要。找到驱动观众看下去的引擎是一切的核心。
Do not rush from material analysis to script. Let ideas ferment:
Show only the tip. The power of a story comes from what is implied but not stated:
Every video must have a consistent narrative voice — the personality behind the words:
Even in non-fiction, explainer, or data-driven videos, apply character-driven storytelling:
Gaiman insists stories need humor, even serious ones:
When writing voiceover or character dialogue:
Apply these throughout the planning process:
# Video Production Memo
**Project:** [Project Name]
**Date:** [Creation Date]
**Version:** 1.0
## Creative Direction
### Core Theme
[What is this video fundamentally about? One sentence that captures the essence.]
### Narrative Engine
[What question/tension drives viewers to keep watching? What is at stake? What promise does the opening make?]
### Narrative Voice
[Define the narrator's personality: tone, attitude, relationship with audience. This voice must remain consistent throughout.]
### Narrative Strategy
[Three-act structure / Problem-solution / Chronological / Comparative analysis — with specific turning points identified]
### Emotional Arc
[Describe how viewer emotions should evolve throughout. Reference specific chapter transitions. Where are the surprises, the tension peaks, the release moments?]
### Audience
- **For:** [Target audience with specific traits]
- **Not For:** [Who this video deliberately excludes — sharpens creative decisions]
### Humor Strategy
[Where does natural humor arise from the material? If no humor fits, state explicitly: "Material does not support humor — maintain earnest tone throughout."]
### Visual References
[Describe style influences using specific references, avoiding generic terms]
## 情绪曲线
[规划每个时间段的情绪走向，确保整体张力呈波浪形、节奏有变化。此表直接指导 Stage-4 的 visual_tension 和 pacing 字段。]
| 时间段 | 情绪 | 张力(0-1) | 节奏 | 对应章节 |
|--------|------|-----------|------|---------|
| 0:00-0:30 | 好奇 | 0.3→0.5 | medium | 开场悬念 |
| 0:30-2:00 | 沉浸 | 0.5→0.6 | slow | 现象展示 |
| 2:00-2:15 | 震惊 | 0.6→0.9 | fast | 反转/冲突 |
| 2:15-3:30 | 理解 | 0.9→0.4 | slow→medium | 深入解析 |
| 3:30-4:00 | 共鸣 | 0.4→0.7 | slow | 结尾升华 |
## Technical Specifications
- **Duration:** [XX] seconds / [X.X] minutes
- **Resolution:** [1920x1080 / 3840x2160]
- **Aspect Ratio:** [16:9 / 9:16 / 1:1]
- **Frame Rate:** [24 / 30 / 60] fps
- **Delivery Format:** [MP4 H.264 / ProRes]
After gathering requirements, analyze user's materials to extract structured data:
重要： 资源类型定义见 carocut-shared-schema skill。写入 resources.yaml 前必须先加载该 skill，确保类型和 source 值合法。
e.g.:
resources:
visual:
- id: vis_001
type: image # image | video | sprite（禁止使用 animation）
description: "Precise description of visual content and purpose"
source: existing | retrieve | generate
- id: vis_002
type: video
description: "Another visual asset"
duration_ms: 5000
source: existing | retrieve # video 不支持 generate
audio:
- id: VO_001
type: voiceover
description: "Male voice, mid-tempo, explanatory tone, warm and authoritative"
- id: aud_001
type: music
description: "Upbeat electronic background, 120 BPM, minimal melody to support voice"
data:
# data 类型由 Builder 直接消费（AnimatedChart/DataTable），Media 不处理
- id: data_001
type: table
description: "Comparison table showing three product features"
structure:
headers: ["Feature", "Product A", "Product B", "Product C"]
rows:
- ["Price", "$99", "$149", "$199"]
- ["Storage", "128GB", "256GB", "512GB"]
- id: data_002
type: chart
description: "Line chart showing growth over 5 years"
structure:
chart_type: line
x_axis: ["2020", "2021", "2022", "2023", "2024"]
y_axis: [120, 145, 178, 210, 265]
x_label: "Year"
y_label: "Revenue (Million USD)"
- id: data_003
type: statistics
description: "Key metrics displayed as callout numbers"
structure:
metrics:
- label: "Active Users"
value: "2.5M"
- label: "Satisfaction Rate"
value: "98%"
components:
# motion-graphics 类型由 Builder 直接消费，生成 src/components/*.tsx
# 只描述画面意图和数据内容，不写 CSS/React/动画细节
- id: comp_001
type: motion-graphics
description: "三级上报流程图：普通告警→严重告警→灾难级，每级标注响应时间和处理人"
used_in: [shot_003, shot_008]
- id: comp_002
type: motion-graphics
description: "三列关键指标卡片：处理量(2.5M)、响应时间(< 200ms)、可用率(99.99%)"
used_in: [shot_005, shot_012]
characters:
- id: char_001
name: "Narrator"
description: "Professional voice, clear pronunciation, measured pace. Conveys expertise without condescension."
- id: char_002
name: "Animated Character"
description: "Stylized 3D mascot with simple expressions. Movement should be smooth, never cartoonish."
| Source | When to Use | Consumer |
|---|---|---|
existing | User-provided in raws/ | Media (skip) → Builder |
retrieve | Stock image/video from Pexels/Pixabay | Media STEP-4 → Builder |
generate | AI-generated custom image/sprite (use ONLY if images_generate tool available) | Media STEP-4 → Builder |
| 无 source 字段 | data/components 类型，Builder 代码实现 | Builder STEP-7 直接消费 |
脚本的关键不是结构，是那个场景或那句话——整个故事存在的理由。
Apply these principles when writing each chapter:
Voice Consistency — Every line must sound like it comes from the same narrator established in the memo. Read aloud: if a line could belong to a different video, rewrite it. 必须保持统一的叙事口吻。
Iceberg Principle — Do not state everything. For each fact in the script, let supporting context live in the visuals rather than the voiceover. The script should feel like the visible tip of deep knowledge. 写出来的只是冰山一角。
Show, Don't Summarize — Prefer specific examples, numbers, and moments over abstract statements. "三年前处理10万条消息，现在500万条" is stronger than "消息处理能力显著提升".
Dialogue Over Declaration — When quoting or paraphrasing, make it sound overheard, not composed. Natural speech has rhythm and imperfection. 对白要像偷听到的，不是灌输的。
Humor Where Natural — If the material supports a wry observation, an unexpected comparison, or a moment of levity, include it. Do not force humor into material that does not support it.
Chapter Turns — Each chapter must contain at least one moment that shifts the viewer's understanding or emotion. A chapter with no turn is just information — not story. Plot的核心必须被激起或改变。
Cut Ruthlessly — Every VO line must advance the narrative, reveal something about the subject, or provide essential information. Lines that do none of these must be deleted, no matter how well-written. 编辑要狠。
# Video Shooting Script
## Chapter 1: [Chapter Title]
### Purpose
[What this section accomplishes in the overall narrative]
### Narrative Turn
[What shift in understanding or emotion occurs in this chapter? Every chapter must have at least one turn.]
### Visual Approach
[General description of how this section will look - not shot-by-shot, but overall visual strategy. Apply iceberg principle: what do visuals convey that words do not?]
### Voiceover
- **[VO_001]** "Opening sentence that hooks viewer attention — plant the question or tension."
- **[VO_002]** "Second sentence developing the hook into main theme."
- **[VO_003]** "Transition to key concept — use specific detail, not abstract statement."
---
## Chapter 2: [Chapter Title]
### Purpose
[Narrative function of this section]
### Narrative Turn
[The specific moment that shifts the viewer's understanding or emotion]
### Visual Approach
[Visual strategy — what does this chapter show that the words leave unsaid?]
### Voiceover
- **[VO_004]** "Content continues with clear numbered references."
- **[VO_005]** "Each line has unique ID for storyboard linking."
---
[Continue for all chapters]
[VO_XXX] where XXX is zero-padded (VO_001, VO_002, ..., VO_099, VO_100)Shots do not need to align 1:1 with voiceover lines. One shot may span multiple VO segments, or one VO may be illustrated by multiple shots. The storyboard should be driven by visual storytelling logic, not script structure.
完整字段定义、视觉语言参考、visual description 规范见 carocut-shared-schema skill 的 references/storyboard-schema.md。
voiceover_refs links shot to corresponding VO line(s) from script.md[VO_001] or [VO_001, VO_002, VO_003] for multiple lines[] for shots with no voiceover (B-roll, transitions)resource_refs lists ALL resource IDs used in this shotbreathing: true 的 shot（2-3秒，无旁白）。breathing: true 和 pacing: "pause" 必须同时出现——breathing: true 是规范标识，pacing: "pause" 确保下游正确处理时长和 TTS 跳过。sync（声画同步）lead-visual（画面先出）lead-audiocounterpointAfter all four documents approved:
manifests/[阶段名称] 文档生成完成。
文件位置: manifests/[filename]
请审阅并提供反馈。确认无误后回复"继续"进入下一阶段。
生产规划完成。
概要:
总时长: 4分23秒 (263秒)
分镜数: 38
资源数: 25 (15 视觉, 8 音频, 2 数据)
角色数: 2
文档目录: manifests/
- memo.md
- resources.yaml
- script.md
- storyboard.yaml
准备进入 STEP-3 文稿润色阶段。
| Mistake | Correction |
|---|---|
| Moving to next stage without approval | Always wait for explicit user confirmation |
| Using marketing language in visuals | Use technical cinematography terms only |
| Restarting VO numbering per chapter | Sequential numbering across entire script |
| Vague shot descriptions | Must include shot size, angle, movement, lighting |
| Missing voiceover_refs in storyboard | Every shot must include voiceover_refs (even if empty) |
| Missing resource_refs in storyboard | Every shot should reference its used resources |
| Invalid resource IDs | Only reference IDs that exist in resources.yaml |
Using animation type | Removed — Media has no animation tool. Use sprite for character animation |
Missing components section | motion-graphics elements must be explicitly defined, not left for Builder to guess |
| Writing implementation details in components | components description should only contain visual intent, not CSS/React code |
| Missing required cinematic fields | Every shot must have ALL fields: framing, camera_movement, pacing, visual_tension, audio_visual_relation, transition_in (struct), breathing |
| Invalid enum values | framing must be one of: ECU/CU/MCU/MS/MLS/LS/ELS; camera_movement must be one of the defined enum values; pacing must be slow/medium/fast/pause |
| visual_tension out of range | Must be numeric 0.0-1.0 |
| Forcing 1:1 shot-to-VO mapping | Use semantic grouping based on visual logic |
| No narrative engine defined | Memo must identify what drives viewers to keep watching |
| Script reads like a summary | Apply iceberg theory: state less, imply more |
| Chapters without turns | Every chapter must shift understanding or emotion |
| Generic ending ("bright future") | End with resonance — a specific image or thought that lingers |
| Inconsistent narrator voice | All VO lines must sound like the same person speaking |
| Trying to please everyone | Define who video is NOT for; commit to specific audience |
| Voiceover explains what visuals already show | Let visuals and words work in counterpoint, not redundancy |
| All shots same duration/pacing | Use 快-慢-快-停顿 rhythm; vary pacing across shots |
| Missing breathing shots between chapters | Every chapter transition needs a breathing: true shot |
| Flat visual_tension curve | Tension should be wave-shaped, matching memo 情绪曲线 |
| Same transition_in for 3+ consecutive shots | Vary transition types; no 3 identical in a row |
| Same camera_movement for 3+ consecutive shots | Vary camera movement; no 3 identical in a row |
| 100% sync audio_visual_relation | At least 20% of shots should use non-sync relations |
| Missing 情绪曲线 in memo | Memo must include emotion curve table linking to storyboard fields |
| transition_in as plain string | transition_in is a struct with type and duration_ms |
[]，不占用 VO 编号。编号始终为 3 位数字零填充（VO_001），超过 999 则自动扩展（VO_1000）。memo.md、resources.yaml、script.md、storyboard.yamlmanifests/ 目录，绝对不创建 memo/、script/、storyboard/、resources/ 等新目录When user signals completion with "looks good", "approved", "continue", or "next", proceed to following stage without re-asking.
增量模式用于在已完成的策划基础上进行局部修改，不重新生成整套策划文档。
manifests/storyboard.yaml，了解当前 shot 列表和编号manifests/script.md，了解当前 VO 编号范围manifests/resources.yaml 追加资源条目manifests/script.md