@aswless_854771076/ai_short_studio_cli 0.1.98 → 0.1.100

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-drama-art-style/prompts/art_style_decision.zh.txt

@@ -0,0 +1,18 @@
+# Codex 执行合同：art_style_decision
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目与用户确认组装的具名输入槽。
+
+项目当前画风：
+{current_art_style}
+
+用户画风要求与确认：
+{user_style_decision}
+
+只返回严格 JSON：{"mode":"confirmed-default | custom","style":{},"decision_reason":"...","approved":true}。
+
+规则：
+1. 用户确认沿用当前默认画风时，mode 为 confirmed-default，style 原样复制项目当前完整画风。
+2. 用户确认自定义画风时，mode 为 custom，style 必须是已用 config_generation 合同生成并确认的完整对象，包含 id、name、description、prompt、renderingContract、visualBible。系统画风也必须保留其 renderingContract，不得只复制名称或 prompt。
+3. 未获得用户确认时 approved 必须为 false；不得替用户确认画风。
+4. decision_reason 记录确认依据，不写 API Key、远程任务 ID 或无关剧情。
+5. 只输出 JSON，不要 Markdown 或解释。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-drama-art-style/prompts/config_generation.en.txt

@@ -0,0 +1,67 @@
+# Codex execution contract: config_generation
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Execution role: professional film art director and AI image prompt designer. Generate a complete art-style configuration that can be saved directly.
+
+Style name: {name}
+User description: {description}
+
+Quality requirements:
+1. Write prompt in English and faithfully represent the requested style. Cover medium/rendering method, material detail, color relationships, composition tendencies, and motivated lighting. Do not add trendy motifs unrelated to the name or description.
+2. Write description in English, briefly stating the core visual traits, suitable subjects, and likely misuse.
+3. renderingContract is the pre-production medium and rendering contract. It explicitly defines medium, rendering_method, surface_language, image_formation, and forbidden medium substitutions. Never default to live action and never replace the medium with vague words such as cinematic or premium. visualBible is independent from prompt and renderingContract and defines cross-shot continuity. Do not copy the prompt into it.
+4. All four palette keys must exist, but only primary and neutral must be non-empty; secondary or accent may be empty for monochrome, low-chroma, or source-limited designs. Prefer explicit hexadecimal colors and use at most five values per group. Use accent only for a defined story object, character, or information beat, never evenly across the whole film.
+5. narrative_arc covers the opening, development, turn/climax, and resolution phases actually needed by the project. Change color only for an evidenced shift in story, time-space, or viewpoint; keep the baseline when no motivated change exists instead of inventing a grade change to fill a phase.
+6. Complete every lighting and motion-language field. Each rule must serve this specific style instead of relying on vague terms such as cinematic or premium. Lighting states an explainable environmental source, direction, quality, and continuity; never default to arbitrary rim light, neon, or colored smoke.
+7. Provide 2 to 12 negative_constraints, prioritizing cheap-looking effects, excessive spectacle, loss of material realism, and cross-shot drift.
+8. Return valid JSON only. Do not use Markdown fences, explanations, or omitted fields.
+
+Return shape:
+{
+  "prompt": "English art style prompt",
+  "description": "English description",
+  "renderingContract": {
+    "medium": "explicit medium such as live-action-photography / 2d-animation / 3d-animation / stop-motion / user-defined-medium",
+    "rendering_method": "specific image-making or rendering method",
+    "surface_language": "shared surface and material language for characters, wardrobe, environments, and props",
+    "image_formation": "how lens, linework, shaders, brushwork, or compositing forms the final image",
+    "forbidden_substitutions": ["other media or cheap substitutes that must not replace the selection"]
+  },
+  "visualBible": {
+    "version": 1,
+    "palette": {
+      "primary": ["#RRGGBB"],
+      "secondary": ["#RRGGBB"],
+      "accent": ["#RRGGBB"],
+      "neutral": ["#RRGGBB"],
+      "scheme": "monochromatic | analogous | complementary | split-complementary | triadic | duotone | custom"
+    },
+    "baseline": {
+      "saturation": "muted | balanced | vivid",
+      "brightness": "low-key | balanced | high-key",
+      "contrast": "soft | balanced | high",
+      "temperature": "warm | neutral | cool"
+    },
+    "lighting": {
+      "key_direction": "specific direction rule",
+      "key_quality": "specific quality rule",
+      "continuity": "cross-shot continuity rule"
+    },
+    "narrative_arc": [{
+      "name": "stage name",
+      "trigger": "story trigger",
+      "color_shift": "color change",
+      "saturation": "saturation change",
+      "brightness": "brightness change"
+    }],
+    "motion_language": {
+      "camera": "camera movement rule",
+      "stillness": "stillness rule",
+      "impact": "impact language",
+      "motion_blur": "motion blur rule",
+      "slow_motion": "slow-motion boundary"
+    },
+    "negative_constraints": ["forbidden visual trait"]
+  }
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-drama-art-style/prompts/config_generation.zh.txt

@@ -0,0 +1,67 @@
+# Codex 执行合同：config_generation
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+执行职责：专业的影视美术指导和 AI 图像提示词设计师。根据输入生成一份可以直接保存的完整画风配置。
+
+画风名称：{name}
+用户描述：{description}
+
+质量要求：
+1. prompt 必须使用英文，准确体现用户指定的画风；包含媒介/渲染方式、材质与细节、色彩关系、构图倾向和有动机的光线。不要擅自混入与名称或描述无关的流行元素。
+2. description 使用中文，简洁说明核心视觉特征、适用题材和应避免的误用。
+3. renderingContract 是前期媒介与渲染合同，必须明确 medium、rendering_method、surface_language、image_formation 和禁止替换的媒介；不得默认真人写实，也不能只写“电影感”“高级感”。visualBible 独立于 prompt 和 renderingContract，是跨镜头连续性规范，不得把 prompt 原文重复粘贴进去。
+4. palette 四组字段都必须存在，但只有 primary 与 neutral 必须非空；secondary 或 accent 不适合单色、低彩或来源受限方案时可为空。优先使用明确十六进制色值，每组最多 5 项。accent 只在存在明确剧情强调对象时使用，并绑定具体道具、人物或信息节点，不能平均铺满全片。
+5. narrative_arc 覆盖项目实际需要的开场、发展、转折/高潮和收束阶段；色彩只有在剧情、时空或人物视点发生可证变化时才改变，没有动机的阶段保持基线，不为填满阶段强造调色变化。
+6. 光线与运动语言的每个字段都必须完整，且服务于该画风，而不是套用空泛的“电影感、高级感”。光线规则应说明可解释的环境光源、方向、质感和连续性，不默认添加轮廓光、霓虹或彩色烟雾。
+7. negative_constraints 至少 2 项、最多 12 项，优先约束廉价感、过度特效、失去材质真实性和跨镜头漂移。
+8. 只返回合法 JSON；不要使用 Markdown 代码块，不要解释，不要省略字段。
+
+返回结构：
+{
+  "prompt": "英文画风提示词",
+  "description": "中文画风说明",
+  "renderingContract": {
+    "medium": "明确媒介，例如 live-action-photography / 2d-animation / 3d-animation / stop-motion / user-defined-medium",
+    "rendering_method": "该媒介的具体成像或渲染方法",
+    "surface_language": "人物、服装、场景与道具共享的表面和材质语言",
+    "image_formation": "镜头、线条、着色器、笔触或合成如何形成最终图像",
+    "forbidden_substitutions": ["不得偷换成的其他媒介或廉价替代"]
+  },
+  "visualBible": {
+    "version": 1,
+    "palette": {
+      "primary": ["#RRGGBB"],
+      "secondary": ["#RRGGBB"],
+      "accent": ["#RRGGBB"],
+      "neutral": ["#RRGGBB"],
+      "scheme": "monochromatic | analogous | complementary | split-complementary | triadic | duotone | custom"
+    },
+    "baseline": {
+      "saturation": "muted | balanced | vivid",
+      "brightness": "low-key | balanced | high-key",
+      "contrast": "soft | balanced | high",
+      "temperature": "warm | neutral | cool"
+    },
+    "lighting": {
+      "key_direction": "中文说明",
+      "key_quality": "中文说明",
+      "continuity": "中文说明"
+    },
+    "narrative_arc": [{
+      "name": "阶段名",
+      "trigger": "剧情触发条件",
+      "color_shift": "色彩变化",
+      "saturation": "饱和度变化",
+      "brightness": "亮度变化"
+    }],
+    "motion_language": {
+      "camera": "运镜原则",
+      "stillness": "静止原则",
+      "impact": "冲击语言",
+      "motion_blur": "动态模糊原则",
+      "slow_motion": "慢动作边界"
+    },
+    "negative_constraints": ["禁止项"]
+  }
+}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/prop_description_update.en.txt

@@ -0,0 +1,14 @@
+# Codex execution contract: prop_description_update
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Update the same prop's complete static description from visible reference facts and the change request.
+
+Prop: {prop_name}
+Original: {original_description}
+Change request: {modify_instruction}
+Reference observations: {image_context}
+
+Change only explicitly requested material, color, shape, structure, quantity, or decoration. Preserve untouched distinctive marks and proportions. The reference does not prove use, origin, or plot. Exclude people, hands, action, scene, camera, and background.
+
+Return strict JSON only: {"prompt":"updated complete prop description"}. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/prop_description_update.zh.txt

@@ -0,0 +1,14 @@
+# Codex 执行合同：prop_description_update
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+结合参考图可见事实与修改指令，更新同一道具的完整静态描述。
+
+道具：{prop_name}
+原描述：{original_description}
+修改指令：{modify_instruction}
+参考图观察：{image_context}
+
+只修改明确要求的材质、颜色、形状、结构、数量或装饰；保持未涉及的独特标志和比例。参考图不证明用途、来源或剧情，不写人物、手部、动作、场景、机位或背景。
+
+只返回严格 JSON：{"prompt":"更新后的完整道具描述"}。使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/prop_generate.en.txt

@@ -0,0 +1,16 @@
+Generate a professional prop visual reference sheet, not a single product image. The sheet is a layout for structural evidence, not an instruction to default to concept art, illustration, or any other medium.
+
+[PROP]
+{prop_name}
+
+[APPROVED STATIC DESCRIPTION]
+{prop_description}
+
+[PROJECT ART STYLE]
+{art_style}
+
+First read the selected style's medium/rendering method, material language, color, lighting, and negative constraints, then use that medium for every view. Never default to concept art, product photography, or a 3D product render: live-action/photographic, 2D animation/comic, 3D animation, and other styles each retain their own material and image-making logic. White background and columns control layout only and must not override the selected medium. When an example conflicts with the project style, the project style wins.
+
+Use one 16:9 landscape canvas with a white background and clear columns. The left third is a main-view close-up. The right two thirds arrange front, exact side, and back views of the same prop at identical height and scale. Use small supporting studies for only the explicitly supported opening structure, patterns, material joints, edge wear, and key details. Omit missing information rather than inventing function, branding, or attachments.
+
+Keep material, color, shape, construction, quantity, proportion, and decoration identical across every view. Exclude people, hands, use action, tabletop staging, environment, extra objects, plot imagery, poster composition, watermarks, and garbled text. Never replace the selected style with a mismatched concept-art, photographic, painted-2D, or 3D product-render treatment. The project art style controls medium, linework/shaders, material, color treatment, lighting, and image response, and must not redesign the prop.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/prop_generate.zh.txt

@@ -0,0 +1,16 @@
+生成专业道具视觉设定板，不是单张商品图。设定板只是组织结构证据的版式，不代表“原画”“插画”或任何默认媒介。
+
+【道具】
+{prop_name}
+
+【已确认静态描述】
+{prop_description}
+
+【项目画风】
+{art_style}
+
+先读取项目画风指定的媒介/渲染方式、材质语言、色彩、光线和负面约束，再以该媒介完成全部视图。不得默认做成概念原画、商品摄影或三维产品渲染：真人/实拍、二维动画/漫画、三维动画及其他画风分别保持各自的材质与成像逻辑。白底与分栏只控制排版，不得覆盖项目媒介。项目画风与模板示例冲突时，以项目画风为准。
+
+使用一整张 16:9 横向画布、纯白背景和清晰分栏。左侧约三分之一为道具主体主视图特写；右侧约三分之二依次排列同一道具的正面、严格侧面和背面视图，三视图等高、同比例。以小型辅助图展示输入已经明确的结构开合、纹样、材质连接、边角磨损和关键细节；缺失信息直接省略，不虚构功能、品牌或附属物。
+
+所有视图严格统一材质、颜色、形状、结构、数量、比例和装饰。不得出现人物、手部、使用动作、桌面陈设、环境背景、额外物件、剧情画面、海报构图、水印或乱码。禁止把项目画风替换成不相符的概念图、写实摄影、二维厚涂或三维产品渲染。项目画风控制媒介、线条/着色器、材质、上色、光线和成像质感，但不重设计道具。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/select_prop.en.txt

@@ -0,0 +1,17 @@
+# Codex execution contract: select_prop
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Extract key props worth independent generation and cross-shot reuse.
+
+[TEXT]
+{input}
+
+[EXISTING PROP LIBRARY]
+{props_lib_name}
+
+Include only movable physical objects with explicit narrative function, distinctive visual identity, and evidence of being carried, transferred, reused, or treated as key evidence. Exclude fixed facilities, ordinary furniture or tableware, one-off replaceable items, costume and makeup, abstractions, and objects without recurrence evidence. Omit uncertain items; normally return no more than three. Do not duplicate an existing stable prop identity.
+
+description may contain only materials, colors, shape, structure, quantities, and decoration, with no people, hands, action, use, plot, camera, or background. summary identifies the object class only.
+
+Return strict JSON only: {"props":[{"name":"stable name","summary":"object class","description":"static visual description"}]}. Return {"props":[]} if none qualify. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-prop-assets/prompts/select_prop.zh.txt

@@ -0,0 +1,17 @@
+# Codex 执行合同：select_prop
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+从文本中提取值得独立生成并跨镜复用的关键道具。
+
+【文本】
+{input}
+
+【已有道具库】
+{props_lib_name}
+
+只有同时具备明确剧情功能、独特视觉身份，并被携带、传递、反复使用或作为关键证据的可移动实体才入库。场景固定设施、普通家具餐具、一次性可替换物件、服装妆容、抽象概念和无后续证据的物件不提取；不确定时不报，通常不超过三个。已有同一稳定身份的道具不得重复。
+
+description 只写材质、颜色、形状、结构、数量关系和装饰细节，不写人物、手部、动作、用途、剧情、镜头或背景。summary 只说明物件类别，不复述剧情。
+
+只返回严格 JSON：{"props":[{"name":"稳定名称","summary":"物件类别","description":"静态视觉描述"}]}。无合格道具返回 {"props":[]}；使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_create.en.txt

@@ -0,0 +1,11 @@
+# Codex execution contract: location_create
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Turn the user request into one reusable static scene description.
+
+{user_input}
+
+Specify period, spatial layout, architecture and materials, fixed furnishings, depth layers, key light direction, and stable anchors. If input is generic, add only enough neutral detail for a controllable layout. Default to an empty set; allow anonymous background crowds only when public activity is explicitly requested. Exclude named people, plot action, camera movement, captions, watermarks, and quality tags.
+
+Return strict JSON only: {"prompt":"complete scene description"}. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_create.zh.txt

@@ -0,0 +1,11 @@
+# Codex 执行合同：location_create
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+把用户要求转成一个可复用的静态场景描述。
+
+{user_input}
+
+明确时代、空间结构、建筑与材质、固定陈设、前中后景、主光方向和稳定锚点。输入过泛时只补足可控布局。默认无人空镜；仅当用户明确要求公共活动时允许无名背景人群。不得写具名人物、剧情动作、机位运动、字幕、水印或质量词。
+
+只返回严格 JSON：{"prompt":"完整场景描述"}。使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_description_update.en.txt

@@ -0,0 +1,14 @@
+# Codex execution contract: location_description_update
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Update the complete scene description from visible reference facts and the change request.
+
+Scene: {location_name}
+Original: {original_description}
+Change request: {modify_instruction}
+Reference observations: {image_context}
+
+The reference proves visible space only; do not infer location history or plot. Preserve untouched structure, entrances, anchors, and light sources. Keep the original for hidden areas. Add no named characters or camera action.
+
+Return strict JSON only: {"prompt":"updated complete scene description"}. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_description_update.zh.txt

@@ -0,0 +1,14 @@
+# Codex 执行合同：location_description_update
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+结合参考图可见事实与修改指令，更新场景完整描述。
+
+场景：{location_name}
+原描述：{original_description}
+修改指令：{modify_instruction}
+参考图观察：{image_context}
+
+参考图只证明可见空间，不推断地点历史或剧情。保留未要求改变的空间结构、出入口、锚点和光源；不可见区域沿用原描述。不得新增具名人物或镜头动作。
+
+只返回严格 JSON：{"prompt":"更新后的完整场景描述"}。使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_modify.en.txt

@@ -0,0 +1,13 @@
+# Codex execution contract: location_modify
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Update the same scene without rebuilding untouched spatial structure.
+
+Scene: {location_name}
+Current description: {location_input}
+Change request: {user_input}
+
+Preserve period, architectural shell, entrances, fixed anchors, and lighting continuity. Change only explicitly requested materials, furnishings, time state, or public background crowd. Add no named characters, plot action, or unrelated spaces.
+
+Return strict JSON only: {"prompt":"updated complete scene description"}. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_modify.zh.txt

@@ -0,0 +1,13 @@
+# Codex 执行合同：location_modify
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+按指令更新同一场景，不重建未要求改变的空间结构。
+
+场景：{location_name}
+当前描述：{location_input}
+修改指令：{user_input}
+
+保留时代、建筑骨架、出入口、固定锚点和光线连续性；只修改明确指定的材质、陈设、时间状态或公共背景人群。不得新增具名人物、剧情动作或不相关空间。
+
+只返回严格 JSON：{"prompt":"更新后的完整场景描述"}。使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_regenerate.en.txt

@@ -0,0 +1,12 @@
+# Codex execution contract: location_regenerate
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Regenerate three candidates for the same scene.
+
+Scene: {location_name}
+Current descriptions: {current_descriptions}
+
+Vary composition and nonessential decoration only. Preserve period, architectural shell, entrances, fixed anchors, material language, and light direction. Default to an empty set and add no plot, people, or unsupported facilities.
+
+Return strict JSON only: {"descriptions":["candidate one","candidate two","candidate three"]}. Use standard JSON escaping.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/location_regenerate.zh.txt

@@ -0,0 +1,12 @@
+# Codex 执行合同：location_regenerate
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+为同一场景重做三个候选描述。
+
+场景：{location_name}
+当前描述：{current_descriptions}
+
+三条候选只改变构图和非关键装饰；时代、建筑骨架、出入口、固定锚点、材质主线和光源方向保持一致。默认无人空镜，不新增剧情、人物或无证据设施。
+
+只返回严格 JSON：{"descriptions":["候选一","候选二","候选三"]}。使用标准 JSON 转义。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/scene_asset_sheet.en.txt

@@ -0,0 +1,19 @@
+Generate a professional environment visual reference sheet, not a single empty shot or a basic three-view layout. The sheet is a layout for spatial evidence, not an instruction to default to concept art, illustration, or any other medium.
+
+[APPROVED SCENE DESCRIPTION]
+{scene_description}
+
+[FIXED CHARACTER SLOTS]
+{available_slots}
+
+[BACKGROUND CROWD POLICY]
+{crowd_policy}
+
+[PROJECT ART STYLE]
+{art_style}
+
+First read the selected style's medium/rendering method, material language, color, lighting, and negative constraints, then use that same medium for every view. Never default to environment concept art, a 3D game scene, or live-action photography: live-action/photographic, 2D animation/comic, 3D animation, and other styles each retain their own image-making logic. The grid, floor plan, and axonometric view organize information only and must not alter the project style of the principal views. When an example conflicts with the project style, the project style wins.
+
+Use one 16:9 landscape multi-view grid. Every panel shows the same scene, time, and design. Include: 01 master wide, 02 left view, 03 right view, 04 reverse view, 05 bird's-eye view, 06 key spatial anchor close-up, 07 exterior or establishing view, 08 main entrance view, and 09 circulation/depth perspective. If no independent exterior exists, use an interior/exterior entrance connection for 07 instead of inventing a facade. Also include a concise floor plan, axonometric view, palette swatches, and 3–5 key material or fixed-object studies.
+
+Keep floor plan, architecture, doors, windows, circulation, object count and scale, materials, palette, light direction, and period identical across views. Every available-slot anchor must be visible with usable open space. Follow crowd_policy for an empty set or anonymous background crowd only. Exclude named characters, plot action, speech bubbles, poster composition, random furniture changes, and contradictory layouts. Never replace the selected project style with a more common but mismatched concept-art, photographic, painted-2D, or 3D-rendered treatment.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/scene_asset_sheet.zh.txt

@@ -0,0 +1,19 @@
+生成专业场景视觉参考设定板，不是单张空镜或简单三视图。设定板只是组织空间证据的版式，不代表“原画”“插画”或任何默认媒介。
+
+【已确认场景描述】
+{scene_description}
+
+【固定可站人位置】
+{available_slots}
+
+【背景人群策略】
+{crowd_policy}
+
+【项目画风】
+{art_style}
+
+先读取项目画风指定的媒介/渲染方式、材质语言、色彩、光线和负面约束，再以同一媒介完成全部视角。不得默认做成概念原画或三维游戏场景，也不得默认转为真人实拍：真人/实拍、二维动画/漫画、三维动画或其他画风分别保留各自的成像逻辑。网格、平面图与轴测图只控制信息组织，不得改变主视角的项目画风。项目画风与模板示例冲突时，以项目画风为准。
+
+使用一整张 16:9 横向画布的多镜头网格，所有画格表现同一场景、同一时间、同一设计方案。固定包含：01 主全景、02 左侧视角、03 右侧视角、04 反向视角、05 鸟瞰、06 最重要空间锚点特写、07 外观或建立镜头、08 主入口视角、09 动线与纵深透视；无独立外观时，07 改为入口内外衔接，不虚构外立面。另附简洁平面布局、轴测透视、配色色卡和 3–5 个关键材质/固定物件样本。
+
+各视角必须统一平面布局、建筑结构、门窗通道、物件数量尺度、材质、配色、光源方向和时代；所有 available_slots 的锚物清晰可见且留有可用空间。按 crowd_policy 决定空镜或仅保留无名背景群众，禁止具名人物、剧情动作、对白框、海报构图、随机增减家具和相互矛盾的空间。禁止把项目画风替换成更常见但不相符的概念图、写实摄影、二维厚涂或三维渲染。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/select_location.en.txt

@@ -0,0 +1,18 @@
+# Codex execution contract: select_location
+
+The owning Skill reads this file to constrain Codex inputs, decisions, and output. It is not a system prompt sent to an external text model. Braced fields are named input slots assembled from selected local project assets.
+
+Extract reusable scene assets from the source. A scene is a stable physical space, not camera action or character performance.
+
+[SOURCE]
+{input}
+
+Rules:
+1. Extract only spaces where action occurs or which characters explicitly enter. Add rooms or zones as sub_locations only when a real containment relationship exists.
+2. Give each scene three compositions with identical spatial structure. Preserve period, architecture, materials, fixed furnishings, depth layers, light direction, and at least three stable anchors.
+3. Provide 2–6 available_slots located against anchors that appear in the descriptions; do not describe character action.
+4. Default to an empty set. Only when the source explicitly requires public activity may has_crowd be true, using anonymous non-subject background crowds and never named characters.
+5. Make only minimal neutral completions for missing dimensions, floors, weather, facilities, and decoration. Make generic locations controllable without inventing plot facts.
+
+Return strict JSON only: {"locations":[{"name":"stable_scene_name_time","summary":"spatial purpose","has_crowd":false,"crowd_description":"","available_slots":["anchor-specific position"],"descriptions":["static description one","static description two","static description three"],"sub_locations":[]}]}.
+Return {"locations":[]} when none qualify. Use standard JSON escaping and no extra fields.

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-scene-assets/prompts/select_location.zh.txt

@@ -0,0 +1,17 @@
+# Codex 执行合同：select_location
+
+本文件由所属 Skill 读取，用于约束 Codex 的输入、判断与输出；不是发送给外部文本模型的 system prompt。花括号字段是从当前项目已选资产组装的具名输入槽。
+
+从原文提取需要跨镜复用的场景资产。场景是可稳定复现的空间，不是镜头动作或人物表演。
+
+【原文】
+{input}
+
+规则：
+1. 只提取真实发生戏份或明确将进入的空间；同地点的房间/区域仅在存在明确包含关系时放入 sub_locations。
+2. 每个场景给出三条构图不同但结构一致的静态描述；保留时代、建筑、材质、固定陈设、前中后景、光源方向和至少三个稳定锚点。
+3. 每个场景给 2–6 个 available_slots，必须以描述中真实锚物定位，不写人物动作。
+4. 默认生成人物空镜。只有原文明示且场景语义离不开公共人群时，才设 has_crowd=true，并只写无名、非主体背景人群；不得出现具名角色。
+5. 原文缺失的尺寸、层数、天气、设施和装饰只做最小中性补全，不制造事实。过泛地点要具体化到可控布局，但不得借机扩写剧情。
+
+只返回严格 JSON：{"locations":[{"name":"稳定场景名_时间","summary":"空间用途","has_crowd":false,"crowd_description":"","available_slots":["锚物与方位明确的位置"],"descriptions":["静态描述一","静态描述二","静态描述三"],"sub_locations":[]}]}。无场景返回 {"locations":[]}；使用标准 JSON 转义，不要额外字段。

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-storyboard-images/prompts/panel_grid_enhance.en.txt

@@ -0,0 +1,29 @@
+Restore split grid cell {cell_index}/{panel_grid_size} as a faithful high-resolution version.
+
+Task type:
+- This is faithful restoration / upscaling / denoising / sharpening of the source cell. It is not image regeneration, not recomposition, and not redrawing from the text description.
+- The final image must look like a high-resolution version of reference image #1. Only clarity and detail quality may improve.
+
+Reference image rules:
+- The first reference image is the original split cell and must be preserved strictly: composition, character placement, action direction, shot size, camera angle, lighting, and subject relationships must follow it.
+- The number of characters, character identity, placement order, face silhouette, hairstyle, costume main colors, expression, action, and occlusion relationships in reference image #1 must not change.
+- Additional reference images may be character turnarounds, character references, location/background images, composition previsualizations, or sketches. Use them only to correct identity, costume, hairstyle, accessories, scene structure, and the project's visual style.
+- Do not paste additional reference images directly into the frame; use them only as consistency constraints.
+
+Enhancement goals:
+- Keep the original composition, character placement, identity, costume, expression, action direction, scene structure, lighting, and project style consistent.
+- Improve sharpness, edge quality, material detail, facial detail, and background detail. Remove compression artifacts, blur, jagged edges, and low-resolution artifacts from grid cropping.
+- Output one complete single storyboard image. Do not recreate a grid or split-screen layout.
+
+Forbidden:
+- Do not redesign characters or rearrange placement based on the cell information below. The cell information is only for understanding and must not override reference image #1.
+- Do not change shot size or camera angle.
+- Do not add, replace, or redesign characters. Do not change age, face shape, hairstyle, costume main colors, or signature accessories.
+- Do not add text, subtitles, watermarks, numbers, borders, split screens, or UI elements.
+- Avoid over-sharpening, excessive HDR, plastic skin, waxy faces, distorted features, extra fingers, and smeared low-resolution details.
+
+Cell information:
+Image prompt: {image_prompt}
+Video prompt: {video_prompt}
+Description: {description}
+Location: {location}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-storyboard-images/prompts/panel_grid_enhance.zh.txt

@@ -0,0 +1,29 @@
+请对九宫格拆分后的第 {cell_index}/{panel_grid_size} 格进行保真高清化修复。
+
+【任务类型】
+- 这是“原图保真修复/超分/去噪/锐化”，不是重新生图，不是重新构图，不是按文字描述重画。
+- 最终画面必须看起来像第 1 张参考图的高清版本，只允许提升清晰度和细节质量。
+
+【参考图规则 - 必须严格遵守】
+- 第 1 张参考图是必须严格保持的原始拆分单格，构图、人物站位、动作方向、镜头景别、机位、光线和主体关系都必须以它为准。
+- 第 1 张参考图中的人物数量、人物身份、站位顺序、脸型轮廓、发型、服装主色、表情、动作和遮挡关系不得改变。
+- 后续参考图可以是角色三视图、角色形象参考、场景背景图、构图预演图或草图，只能用于纠偏角色身份、服装、发型、配饰、场景结构和项目整体画风。
+- 不允许把后续参考图直接贴入画面；只能用于约束角色/场景/风格一致性。
+
+【高清化目标】
+- 保持原图构图、人物站位、角色身份、服装、表情、动作方向、场景结构、光线和项目画面风格一致。
+- 提升清晰度、边缘质量、材质细节、面部细节和背景细节，消除压缩感、模糊、锯齿和宫格裁切带来的低清问题。
+- 画面必须是单张完整分镜图，不要恢复成九宫格，不要生成分屏。
+
+【禁止事项】
+- 不要根据下面的画面信息重新设计人物或重新安排站位；画面信息只用于理解，不得覆盖第 1 张参考图。
+- 不要改变镜头景别和机位。
+- 不要新增角色，不要替换角色，不要改变角色年龄、脸型、发型、服装主色或标志性配饰。
+- 不要添加文字、字幕、水印、编号、边框、分屏或 UI 元素。
+- 不要出现过度锐化、过度 HDR、塑料皮肤、蜡像感、畸形五官、多余手指、低清涂抹感。
+
+【该格画面信息】
+画面提示词：{image_prompt}
+视频提示词：{video_prompt}
+画面描述：{description}
+场景：{location}

package/skills/using-vvicat-ai-short-studio-cli/references/short-drama/assets/modules/generate-storyboard-images/prompts/panel_grid_image.en.txt

@@ -0,0 +1,77 @@
+You are a storyboard visual director. Given a single-shot panel JSON and source text, render a single grid image containing {panel_grid_size} sub-cells.
+
+[⚠️ REFERENCE IMAGE RULES — HIGHEST PRIORITY, MUST FOLLOW STRICTLY]
+Character/location reference images are attached to this request. EVERY sub-cell MUST be drawn strictly from these references — never invent characters or scenes:
+- Character references: each character's appearance, facial features, hairstyle, costume, accessories, and body type in every sub-cell MUST match the character reference exactly. Do not alter or redesign them.
+- Location references: use them for the environment's composition, architectural structure, style, and mood. Lock the same camera position, axis, shot scale, and background relationship by default. The scene's visual features MUST come from the reference — do not paste the whole reference flat into the background.
+- The text labels above the reference images mark character/location names; match them to the characters/locations in the panel JSON. Labels are for identification only — never paint them into the image.
+- Art style, line work, color, texture, and lighting MUST stay consistent with the references; no cell may deviate from the reference style.
+- Sub-cells show only temporal progression and action breakdown within one continuous shot. The visual identity of characters and scene MUST stay locked to the references: the SAME character, SAME costume, SAME location across ALL sub-cells — no cell may break away with a different design.
+
+[⚠️⚠️⚠️ GLOBAL ART-STYLE CONSISTENCY — NON-NEGOTIABLE]
+The ENTIRE grid (main characters, supporting characters, passersby, crowd, backgrounds, props, environmental elements in every sub-cell) MUST be rendered in ONE unified art style:
+- Passersby / NPCs / crowd who have no dedicated reference image MUST still be painted in the SAME style as the main character references — same outline weight, same shading method, same rendering texture. NEVER improvise a different style for them.
+- ALL sub-cells share ONE style: if main characters are 2D anime, then every person and every background in every cell must be 2D; if they are photoreal, keep everything photoreal.
+- FORBIDDEN: any 3D, CG, CGI, Blender, Pixar/Disney 3D-cartoon look, Unreal-Engine render, or photoreal texture leaking into a 2D frame; NO 2D characters pasted onto a 3D background.
+- FORBIDDEN: hand-drawn characters next to photo-textured props or backgrounds; multiple rendering textures within one frame.
+- Line thickness, coloring method (flat vs gradient), shadow hardness, and highlight treatment MUST be uniform across all sub-cells.
+- Anti-AI look: avoid over-sharpening, overdone HDR, oversaturation, plastic/waxy skin, greasy textures, extra fingers, and distorted faces.
+- Lighting direction and intensity must be coherent within each sub-cell.
+
+Requirements:
+1. The output is a single grid image, not separate panels.
+2. Arrange sub-cells as {grid_layout}, total {panel_grid_size} cells, reading order left-to-right then top-to-bottom.
+3. The {panel_grid_size} sub-cells are temporal keyframes of one continuous shot. Keep the same camera position, axis, shot scale, focal length, and background relationship by default; advance only action, expression, lip movement, blocking, and secondary environmental motion. Only `panel.camera_move` or `composition_reference` may continuously change composition along its explicit path. Never invent a close-up, insert shot, cut, or alternate viewpoint.
+4. Maintain one unified style across the whole image, governed by the selected style contract and references: "{style}". Every cell, character, and background uses the same medium, rendering_method, surface_language, and image_formation. If a reference belongs to an old style or conflicts with the current contract, stop and return it for asset reselection instead of silently following the stale image.
+5. Overall aspect ratio: {aspect_ratio} (each sub-cell is compressed to fit the row/column size).
+6. No unrelated Chinese/English text, watermarks, numbering, dialogue bubbles, or UI elements. When the panel explicitly requires system text or interface content, preserve it verbatim and reveal it in `grid_plan` order without breaking brackets, quotation marks, or numbers.
+7. Clear borders between sub-cells but the border should not overpower the subjects.
+8. Character appearance, costume, location, and props must remain consistent across sub-cells, and must follow the reference images.
+9. `context.neighbor_panel_continuity` is only for cinematographic continuity — never paint adjacent-shot story content into this grid image.
+10. If the "Source text" contains dialogue content, render character lip movements and expression changes in corresponding cells based on the grid's temporal order:
+    - Start of speech: mouth slightly open, initial expression state
+    - Mid-speech: mouth opening/closing variations, expression progressing with emotion
+    - End of speech: mouth closed or holding final expression
+    Ensure lip movements match the emotion and rhythm of the dialogue content.
+11. When `panel` or the source specifies N total seats/supports with X occupied and Y empty, every cell must show exactly N: the specified X characters' pelvises and legs visibly rest on their own supports, with no empty duplicate support drawn in front of them; only the remaining Y supports may be unoccupied. Any mismatch in people, support count, occupied count, or empty count is a failure.
+12. Enforce each character reference's `performance_constraints`. Distribute psychological change across key cells as stimulus/information → gaze shift → social mask → brief leak → decision/action → residue. Gaze lands on a named person, prop, or spatial target with correct screen direction and never enters the lens without explicit POV/direct-address evidence. A close shot changes only one or two facial regions subtly; at wider scales use head, breathing posture, pause, distance, and blocking. A listener keeps independent attention and reaction rather than copying the speaker. Use a habitual_actions item only when its recorded trigger occurs and always avoid forbidden_performance.
+
+[Composition reference — HIGHEST PRIORITY]
+- If storyboard data contains `composition_reference`, follow its camera framing and character blocking precisely.
+- `composition_reference.primary_camera` is the primary shot; obey its FOV, camera position/target, and character positions/facing.
+- `composition_reference.alternate_views` lists other camera references for the same shot (close-ups, reverse angles, etc.). Reference images follow that array order. Preserve identity, lighting, and style; compose the frame to match the primary camera.
+- Coordinate system: meters, y up, -z toward camera, x right.
+
+[Style and location conflict handling]
+- "{style}" is a visual style directive for art style, line work, color, lighting, and texture. If it contains concrete scene/location terms that conflict with the panel location, follow `panel.description` and `panel.location`.
+- Do not change the panel into an unrelated location just because the style directive mentions another place.
+- `visual_bible` independently controls palette, tone, lighting continuity, and motion language; never turn it into a new location, character, or rendering medium.
+
+Compact panel JSON:
+{storyboard_text_json_input}
+
+[Grid keyframe execution]
+- `grid_plan.cells` in the panel JSON gives each cell's temporal order and keyframe intent. Draw them strictly in `cellIndex` order.
+- Cell 1 is the starting moment, the final cell is the ending moment, and middle cells bridge action, expression, lip movement, blocking, and camera motion.
+- Every cell must serve the same continuous shot; do not turn the grid into unrelated storyboard shots.
+
+[SPATIAL LEDGER — BUILD INTERNALLY BEFORE RENDERING; DO NOT WRITE IT INTO THE IMAGE]
+- Lock each named character to one identity, screen left/center/right, foreground/midground/background, body facing, face direction, eyeline target, foot position, and entry/exit edge. Only movers explicitly named by `grid_plan` may change position; everyone else remains anchored across cells.
+- Lock each key prop to one count, initial position, owner, contacting hand, before/contact/after state, and tail position. Never duplicate, vanish, swap hands, float, or draw an empty duplicate support in front of its occupant.
+- In group shots lock exact headcount and left-to-right order first. For seating, also lock total supports, occupants, empty-seat identities, and physical sitting contact. Never duplicate one character, merge two people, swap identities, or add an unnamed person in any cell.
+- Screen direction means final-image direction, not raw world-coordinate direction. Map blocking through the `composition_reference` camera before locking axis, eyelines, and travel. If the inputs cannot determine these relationships uniquely, stop and return the missing facts as unresolved instead of guessing.
+- A story effect appears only when source text or a selected prop/location asset explicitly supports it, with a locked source, path, target, count, color, and end state. Never interpret borders, arrows, circles, color marks, or direction notes as portal light, energy, smoke, or a physical prop.
+
+[⚠️ COMPOSITION CONTRACT OVER TIME — GEOMETRY CHANGES BETWEEN CELLS]
+- If the panel JSON carries composition_contract in its photography rules, or a professional storyboard nests a composition contract under photographyPlan.composition / visual_plan.composition, it is the confirmed contract for this shot. If only the thin plan-time visual_plan.composition exists (primary/secondary/reason/evidence), do not demand the 11 geometry fields: derive the missing geometry from the panel and source text, and where it cannot be derived treat the shot as a legacy fallback; never invent field values. The primary/secondary composition codes and pattern names must never appear as text in any cell, and bare abstractions ("cinematic", "oppressive", "lonely") are not instructions. Every cell translates the contract into visible geometry progressing over time:
+  - Cell 1 is the opening composition: the read_order first-glance landing, every character's subject_placement (left/right, depth layer, camera height), eyeline_room gaze and negative space, depth_layers contents and blur, and light_and_color_geometry key-light direction with its in-world source are all established at once;
+  - Middle cells evolve continuously along action_vector: approach/recede, entry/exit edges, movement between depth layers, and when occlusion appears and yields the subject must all read on screen; for action, distribute anticipation—contact—result—reaction across adjacent cells, with clear cells for the contact point and the force result, while the first-read information stays readable throughout;
+  - The final cell lands on tail_state: position, facing, and residual tension briefly stable;
+  - Light direction, eyeline axis, screen direction, and character left-right order stay identical across all cells; when action_vector is null, only secondary motion allowed by the contract and panel is permitted — never invent push-ins, orbits, drift, or extra motion energy.
+- The evidence gate is OR and applies per cell: every level in evidence (composition-only/set_dressing/weather_atmosphere/action_blocking) must be either explicitly described in the source text or present in the attached selected character/location/prop assets — either one suffices (they need not both be present, aligned with director-composition.md §0.2); the action_blocking level may additionally be supported by confirmed blocking facts, which cannot replace the source-text or asset evidence required for weather/set dressing; each component holds when compound evidence is joined with +. Rain, snow, fog, smoke/dust, neon, mirrors, doors/windows/columns, and weapons at the weather_atmosphere/set_dressing level may enter a cell only with source-text or selected-asset support, while action_blocking elements such as crowds may additionally rely on confirmed blocking facts; the existing reference-asset requirements for identity consistency and set structure remain unaffected. A single unsupported cell must never be patched with imagination: stop generation and return the shot to the storyboard stage as unresolved with the missing evidence named, so evidence can be added or the pattern changed. The model may never add rain, snow, fog, neon, mirrors, smoke, doors/windows/columns, weapons, crowds, or extra people.
+- Guard each ai_risks code visibly in every cell: foreground occluders cover at most one third of one side or only a thin top strip and never cover the face or key reactions, negative space stays empty, contact points read clearly, keep a 10% margin at motion edges, and screen direction stays consistent between adjacent cells.
+- Vertical 9:16: relationship cells use front/back depth, over-shoulder framing, or vertical staging instead of squeezing side by side; put key faces, evidence props, and contact points in each cell's upper-middle safe area with both eyes readable; split crowd action across cells as crowd establishment — protagonist action — reaction/result rather than lining up many figures horizontally.
+- Legacy panels without the contract follow the panel and grid_plan text; never invent codes or pattern names.
+
+Source text:
+{source_text}