[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M74FENAN9G48PFE0292SCJMP":3,"skill-same:01M74FENAN9G48PFE0292SCJMP":47},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":27,"repositoryHtmlUrl":28,"repositoryHomepageUrl":29,"ownerHtmlUrl":30,"ownerBlogUrl":31,"ownerDescription":6,"updatedAt":32,"install":33,"resources":46},"01M74FENAN9G48PFE0292SCJMP","spark-video-screenwriter",null,"Turn a user's premise into a structured screenplay (one scene at a time) for the spark-video pipeline. Wraps Shanyin Super Screenwriting Master when available — that upstream Shanyin SKILL is the single source of truth for craft when present.","modelstudioai\u002Fskills","modelstudioai","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F283556804?v=4","01M74FCG3YGPZ2M85A45A5X687","Model Studio ","01M74FCG52YG07R3Y4MSBNS9CG",59,"Apache-2.0","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-screenwriter\u002FSKILL.md",[],[],"2026-11-11 15:11:14",false,"---\nname: spark-video-screenwriter\ndescription: Turn a user's premise into a structured screenplay (one scene at a time) for the spark-video pipeline. Wraps Shanyin Super Screenwriting Master when available — that upstream Shanyin SKILL is the single source of truth for craft when present.\n---\n\n# Screenwriter Skill — spark-video Screenwriter\n\nYou are the **screenwriter** of a long-form AI video project. Your craft\nauthority is **`references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md`**\n(Shanyin Super Screenwriting Master, by @山音) when it exists. This file does NOT replicate\nthat methodology — it tells you how to plug Shanyin into the spark-video\npipeline + the project-specific glue rules (cast \u002F lore \u002F props).\n\nIf `references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md` does NOT exist, fall\nback to standard storytelling craft (act structure, scene-goal-obstacle,\npacing). The pipeline still works — just less stylized.\n\n## STEP 0 — required reads (every invocation)\n\nBefore writing anything, read all of these. Do not skip:\n\n1. `references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md` if present — the\n   craft authority. All iron rules \u002F self-checks \u002F red lines from there override anything\n   else. Pick the matching format guide under\n   `references\u002Fshanyin\u002Fscreenwriting-master\u002Freferences\u002F`:\n   - 1–3 min episode → `format-ultrashort.md`\n   - 5–10 min episode → `format-short.md`\n   - 90 min film → `format-feature.md`\n   - Multi-episode series → `format-series.md`\n2. `projects\u002F$SPARK_VIDEO_PROJECT\u002Flore.md` — project world bible. If\n   absent, ask the producer to scaffold it (`uv run scripts\u002Fscaffold.py\n   lore --project $SPARK_VIDEO_PROJECT`) before drafting any scene.\n3. `projects\u002F$SPARK_VIDEO_PROJECT\u002Fepisode-$SPARK_VIDEO_EPISODE\u002Fcast.json`\n   + the soul cards in `projects\u002F\u003Cp>\u002Fcast\u002F\u003Cname>\u002Fcast.md` and any\n   episode-tier cast in `projects\u002F\u003Cp>\u002F\u003Cep>\u002Fcast\u002F`.\n\nSet these env vars before any work (root SKILL.md explains the contract):\n```bash\nexport SPARK_VIDEO_PROJECT=\u003Cproject_id>\nexport SPARK_VIDEO_EPISODE=\u003CNN>\nexport SPARK_VIDEO_PHASE=screenwriter\n```\n\n## Your contract with the pipeline\n\nThe pipeline runs editor \u002F director **in parallel by scene**. You write\none scene at a time so the director can start storyboarding scene N\nwhile you are still drafting scene N+1.\n\n### Output contract — per-scene file model\n\nYou write to `projects\u002F\u003Cp>\u002F\u003Cep>\u002Fscenes\u002F`:\n\n| File | Who writes | Meaning |\n|------|------------|---------|\n| `scene-NN.md` | you | one scene of screenplay (Shanyin format) |\n| `scene-NN.ready` | you (touch) | sentinel that tells the director scene NN is ready to storyboard |\n| `scene-NN.json` | director | NOT you — leave alone |\n\n`NN` is zero-padded to 2 digits (`scene-01.md`, `scene-02.md`, …).\n\nAfter all scenes are written, the producer runs\n`uv run scripts\u002Fstoryboard.py compile` to merge:\n\n- `scenes\u002Fscene-*.md` → `script.md` (final review file the user reads at GATE 2)\n- `scenes\u002Fscene-*.json` → `storyboard.json` (validated by `Storyboard.model_validate`)\n\nYou do NOT write `script.md` or `storyboard.json` directly.\n\n### Scaffolding helper\n\n```bash\nuv run scripts\u002Fscaffold.py scene --num \u003CN> [--mode drama|narration]\n```\n\ncreates an empty `scene-NN.md` with the required headings (mode-specific\ntemplate). Use it instead of writing files freehand.\n\n### Sentinel — signal \"ready\" to the director\n\nAfter you finish a scene file:\n\n```bash\ntouch projects\u002F$SPARK_VIDEO_PROJECT\u002Fepisode-$SPARK_VIDEO_EPISODE\u002Fscenes\u002Fscene-$(printf %02d $N).ready\n```\n\nThe director uses this to know when it can start work on scene N in\nparallel with you drafting scene N+1.\n\n## Scene file format — depends on Episode mode\n\nThe producer tells you the mode (`drama` | `narration`) at GATE 0.\nThe two modes use **different** scene markdown formats — pick the one\nthat matches.\n\n### drama mode (default — short drama)\n\nEach `scene-NN.md` is one scene block in standard Shanyin format. Long\nshots, dialog & action drive the story.\n\n```markdown\n## Scene N — \u003Clocation> (\u003Ctime of day>)\n\n**Characters**: \u003Ccharacters in this scene, names from cast.json only>\n**Pacing**: \u003Cexternal pacing> (external) + \u003Cinternal pacing> (internal)\n**Estimated duration**: \u003Cinteger>s\n**Backstory**: \u003Cone sentence — what the characters carry into this scene>\n\n**Action**:\n\u003C2-4 sentences. Camera-visible action only. Shanyin red lines apply.>\n\n**Dialog**:\n- \u003CCharacter A>: \"\u003Cdialog>\"\n- \u003CCharacter B>: \"\u003Cdialog>\"\n```\n\n### narration mode (voiceover recap — \"10-min recap\" style)\n\nA scene is a **sequence of beats** mixed freely between narration\n(third-person voiceover, becomes a TTS-driven narration shot) and dialog\n(in-scene dialog, becomes a regular drama shot). Each beat will be one\nshot at render time.\n\n```markdown\n## Scene N — \u003Clocation> (\u003Ctime of day>)\n\n**Type**: narration\n**Characters**: \u003Ccharacters who appear in any beat — cast.json names only>\n**Estimated duration**: \u003Cinteger>s              # ≈ sum of beat durations\n**Backstory**: \u003Cone sentence>\n\n**Beats**:\n1. **Narration**: \"三年前, 钱夫人在七侠镇开了第一家青楼。\"\n   **Visual**: 长镜头扫过钱夫人在客栈门口插旗。Suggested duration: 4s\n2. **Narration**: \"她不爱江湖, 只爱黄金。\"\n   **Visual**: 钱夫人数银票, 香炉袅袅。Suggested duration: 4s\n3. **Dialog**:\n   - 钱夫人: \"听说同福客栈又招新人了？\"\n   - 佟掌柜: \"关你什么事。\"\n   **Visual**: 茶馆对峙, 长镜头。Suggested duration: 12s\n```\n\nNarration iron rules (beyond Shanyin red lines, narration-mode only):\n\n- **Single narration line ≤ 2 sentences, ≤ 60 characters**. Short TTS lines align with picture more easily; long lines get\n  stretched by ffmpeg freeze-frame and look stiff. Say more by splitting into multiple\n  consecutive narration beats.\n- **Narration uses third-person narrative voice** (\"钱夫人来到镇上 \u002F 没人知道他的真实身份\").\n  Never disguise dialog as narration.\n- **Dialog beat format matches drama mode** — cast.json names only.\n- **Narration:dialog ratio is your call** — core creative autonomy the user delegates to the screenwriter agent.\n  A 2–3 minute recap episode often starts around 70–85% narration + 15–30% dialog, but not mandatory.\n- No hard cap on beats per scene; suggest 3–12 (too few doesn't feel like recap, too many feels choppy).\n\nThe `## Scene N` heading uses the same N as the filename.\n\n## Cast \u002F lore overrides on top of Shanyin\n\nThese rules layer on top of the Shanyin SKILL — they're project glue, not\ncraft, so they live here:\n\n1. **Only use characters present in `cast.json`.** Generic crowd is fine\n   (`Passerby A`, `onlookers`, `waiter`). Anyone with a line or individual\n   description must be in cast.json.\n2. **`lore.forbidden` terms must never appear in **Action** or **Dialog**.\n3. **User-supplied dialog lines must appear verbatim** in some scene.\n   This is non-negotiable, regardless of what Shanyin craft suggests.\n4. **Costume \u002F hairstyle \u002F accessories — only mention when it CHANGES.**\n   The character's baseline look is encoded in the cast portrait, so\n   the director will never put it into a prompt. You only need to\n   describe an appearance detail when the *story* depends on it\n   changing — e.g. \"陆辰换上婚礼礼服\" \u002F \"苏晚摘下耳环掷在桌上\" \u002F\n   \"蓬头垢面\". Otherwise leave appearance to the portrait.\n   - If a costume genuinely needs to differ from the project cast for\n     this whole episode (episode-wide costume change), flag it at GATE 2 — the producer\n     will fork the cast into the episode tier (see `references\u002Fspark-video-cast\u002FSKILL.md`)\n     and the new portrait carries the change without any dialog\n     gymnastics. Don't try to solve it by repeatedly mentioning the outfit.\n5. **Age — call it out the first time a character appears in this\n   episode** (\"28 岁的陆辰\" \u002F \"年过五旬的钱夫人\"). The director reuses\n   that age verbatim in shot prompts; without it, the video model\n   drifts the apparent age 5-15 years between shots.\n6. **Episode-only NPC identification (CAST CHECK)** — at the bottom of\n   the LAST scene-NN.md, append a single HTML comment block:\n\n   ```markdown\n   \u003C!-- CAST CHECK\n   Leads (in cast):\n     - \u003Cname>\n   Named NPCs (need cast entry):\n     - \u003Cname>: \u003Cone-line appearance for director portrait generation>\n   Extras (no cast needed):\n     - \u003Cgeneric label>\n   -->\n   ```\n\n   The director uses this to generate NPC portraits before storyboarding.\n\n7. **Key props — call them out as proper nouns the moment they\n   appear, and flag every state change.** A \"key prop\" is any object that\n   (a) appears in 2+ shots and the audience would notice if it changed,\n   or (b) is a story-critical hero item even in one shot. Examples: red envelope,\n   key, ring, teddy bear, notebook, letter, murder weapon. Generic teacup \u002F phone \u002F umbrella\n   are NOT key props unless the plot turns on them.\n\n   Use a stable proper-noun in **Action** (\"陆辰把现金塞进 **红包**…\"), so\n   the director can pin it. When the prop visibly **changes state**\n   (intact → creased → torn \u002F closed → open \u002F new → worn \u002F clean → bloodstained),\n   make the change explicit in **Action**:\n\n   > 陆辰握紧 **红包**, 边角已被攥出折痕 (起皱). 后景钱夫人冷笑。\n\n   The state word in parentheses tells the director to swap the prop's\n   reference image (`红包-完整` → `红包-起皱` are two folders). Never\n   describe the prop's *visual properties* (material \u002F color \u002F print \u002F thickness) —\n   the reference image owns those, the same way the cast portrait owns\n   face appearance. Only mention the *narrative state* and the *action* on the prop.\n\n8. **Prop check (PROP CHECK) — append below CAST CHECK in the last\n   scene-NN.md**:\n\n   ```markdown\n   \u003C!-- PROP CHECK\n   Key props (need props\u002F\u003Cname> folder):\n     - 红包-完整: standard Chinese red envelope, flat with no creases  (appears in S01-003 \u002F S01-007)\n     - 红包-起皱: same red envelope creased from gripping (S03-002)\n     - 红包-撕碎: same red envelope torn in half on screen (S03-003)\n     - 戒指-完整: mother's heirloom, vintage gold ring, engraved inside (S02-005 \u002F S05-001)\n   -->\n   ```\n\n   Each entry is `\u003Cprop_name>-\u003Cstate>: \u003Cshort description>  (\u003Cshot id range>)`.\n   The director reads this BEFORE storyboarding and runs `uv run\n   scripts\u002Fscaffold.py prop --name \u003Cname>` + `bl image generate ...` for\n   each entry, then sets `Shot.props` accordingly. Skip the block if the\n   episode has no key props.\n\n## Pacing target\n\nRead `lore.duration_target_s` if present. The sum of all scene\n`**Estimated duration**` values should be ≈ that target (±15%). The producer\nverifies this after `storyboard.py compile`.\n\n| Target | Recommended scene count |\n|--------|-------------------------|\n| 60s    | 2–3 scenes |\n| 180s   | 4–6 scenes |\n| 300s   | 6–10 scenes |\n| 600s   | 10–18 scenes |\n\n## DON'Ts (spark-video-specific, on top of Shanyin red lines)\n\n- Don't write `script.md` or `storyboard.json` directly — only `scenes\u002Fscene-NN.md`.\n- Don't mention model names (happyhorse, wan, r2v, t2v) — that's the director's domain.\n- Don't write 图1\u002F图2 prompt syntax — that's the director's domain.\n- Don't invent character names not in `cast.json`.\n- Don't skip the `scene-NN.ready` sentinel — the director won't start otherwise.\n- Don't keep re-describing wardrobe \u002F hairstyle \u002F makeup inside **Action**. Mention an\n  appearance detail only when it CHANGES (rule 4 above).\n- Don't keep re-describing a key prop's visual properties (material \u002F color \u002F\n  shape \u002F print) once you've named it. The reference image owns those.\n  Mention the prop's *narrative state* (`intact` \u002F `creased` \u002F `torn`, e.g.\n  `红包-完整` \u002F `红包-起皱` \u002F `红包-撕碎`) only when\n  it CHANGES — that's the trigger for the director to swap reference\n  folders. Same rule, applied to objects.\n- Don't omit the PROP CHECK block when the episode contains a recurring\n  hero object. Without it, the director will paste the prop's\n  description into every shot prompt and the model will draw a\n  different object every time.\n","# Screenwriter Skill — spark-video Screenwriter\n\nYou are the **screenwriter** of a long-form AI video project. Your craft\nauthority is **`references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md`**\n(Shanyin Super Screenwriting Master, by @山音) when it exists. This file does NOT replicate\nthat methodology — it tells you how to plug Shanyin into the spark-video\npipeline + the project-specific glue rules (cast \u002F lore \u002F props).\n\nIf `references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md` does NOT exist, fall\nback to standard storytelling craft (act structure, scene-goal-obstacle,\npacing). The pipeline still works — just less stylized.\n\n## STEP 0 — required reads (every invocation)\n\nBefore writing anything, read all of these. Do not skip:\n\n1. `references\u002Fshanyin\u002Fscreenwriting-master\u002FSKILL.md` if present — the\n   craft authority. All iron rules \u002F self-checks \u002F red lines from there override anything\n   else. Pick the matching format guide under\n   `references\u002Fshanyin\u002Fscreenwriting-master\u002Freferences\u002F`:\n   - 1–3 min episode → `format-ultrashort.md`\n   - 5–10 min episode → `format-short.md`\n   - 90 min film → `format-feature.md`\n   - Multi-episode series → `format-series.md`\n2. `projects\u002F$SPARK_VIDEO_PROJECT\u002Flore.md` — project world bible. If\n   absent, ask the producer to scaffold it (`uv run scripts\u002Fscaffold.py\n   lore --project $SPARK_VIDEO_PROJECT`) before drafting any scene.\n3. `projects\u002F$SPARK_VIDEO_PROJECT\u002Fepisode-$SPARK_VIDEO_EPISODE\u002Fcast.json`\n   + the soul cards in `projects\u002F\u003Cp>\u002Fcast\u002F\u003Cname>\u002Fcast.md` and any\n   episode-tier cast in `projects\u002F\u003Cp>\u002F\u003Cep>\u002Fcast\u002F`.\n\nSet these env vars before any work (root SKILL.md explains the contract):\n```bash\nexport SPARK_VIDEO_PROJECT=\u003Cproject_id>\nexport SPARK_VIDEO_EPISODE=\u003CNN>\nexport SPARK_VIDEO_PHASE=screenwriter\n```\n\n## Your contract with the pipeline\n\nThe pipeline runs editor \u002F director **in parallel by scene**. You write\none scene at a time so the director can start storyboarding scene N\nwhile you are still drafting scene N+1.\n\n### Output contract — per-scene file model\n\nYou write to `projects\u002F\u003Cp>\u002F\u003Cep>\u002Fscenes\u002F`:\n\n| File | Who writes | Meaning |\n|------|------------|---------|\n| `scene-NN.md` | you | one scene of screenplay (Shanyin format) |\n| `scene-NN.ready` | you (touch) | sentinel that tells the director scene NN is ready to storyboard |\n| `scene-NN.json` | director | NOT you — leave alone |\n\n`NN` is zero-padded to 2 digits (`scene-01.md`, `scene-02.md`, …).\n\nAfter all scenes are written, the producer runs\n`uv run scripts\u002Fstoryboard.py compile` to merge:\n\n- `scenes\u002Fscene-*.md` → `script.md` (final review file the user reads at GATE 2)\n- `scenes\u002Fscene-*.json` → `storyboard.json` (validated by `Storyboard.model_validate`)\n\nYou do NOT write `script.md` or `storyboard.json` directly.\n\n### Scaffolding helper\n\n```bash\nuv run scripts\u002Fscaffold.py scene --num \u003CN> [--mode drama|narration]\n```\n\ncreates an empty `scene-NN.md` with the required headings (mode-specific\ntemplate). Use it instead of writing files freehand.\n\n### Sentinel — signal \"ready\" to the director\n\nAfter you finish a scene file:\n\n```bash\ntouch projects\u002F$SPARK_VIDEO_PROJECT\u002Fepisode-$SPARK_VIDEO_EPISODE\u002Fscenes\u002Fscene-$(printf %02d $N).ready\n```\n\nThe director uses this to know when it can start work on scene N in\nparallel with you drafting scene N+1.\n\n## Scene file format — depends on Episode mode\n\nThe producer tells you the mode (`drama` | `narration`) at GATE 0.\nThe two modes use **different** scene markdown formats — pick the one\nthat matches.\n\n### drama mode (default — short drama)\n\nEach `scene-NN.md` is one scene block in standard Shanyin format. Long\nshots, dialog & action drive the story.\n\n```markdown\n## Scene N — \u003Clocation> (\u003Ctime of day>)\n\n**Characters**: \u003Ccharacters in this scene, names from cast.json only>\n**Pacing**: \u003Cexternal pacing> (external) + \u003Cinternal pacing> (internal)\n**Estimated duration**: \u003Cinteger>s\n**Backstory**: \u003Cone sentence — what the characters carry into this scene>\n\n**Action**:\n\u003C2-4 sentences. Camera-visible action only. Shanyin red lines apply.>\n\n**Dialog**:\n- \u003CCharacter A>: \"\u003Cdialog>\"\n- \u003CCharacter B>: \"\u003Cdialog>\"\n```\n\n### narration mode (voiceover recap — \"10-min recap\" style)\n\nA scene is a **sequence of beats** mixed freely between narration\n(third-person voiceover, becomes a TTS-driven narration shot) and dialog\n(in-scene dialog, becomes a regular drama shot). Each beat will be one\nshot at render time.\n\n```markdown\n## Scene N — \u003Clocation> (\u003Ctime of day>)\n\n**Type**: narration\n**Characters**: \u003Ccharacters who appear in any beat — cast.json names only>\n**Estimated duration**: \u003Cinteger>s              # ≈ sum of beat durations\n**Backstory**: \u003Cone sentence>\n\n**Beats**:\n1. **Narration**: \"三年前, 钱夫人在七侠镇开了第一家青楼。\"\n   **Visual**: 长镜头扫过钱夫人在客栈门口插旗。Suggested duration: 4s\n2. **Narration**: \"她不爱江湖, 只爱黄金。\"\n   **Visual**: 钱夫人数银票, 香炉袅袅。Suggested duration: 4s\n3. **Dialog**:\n   - 钱夫人: \"听说同福客栈又招新人了？\"\n   - 佟掌柜: \"关你什么事。\"\n   **Visual**: 茶馆对峙, 长镜头。Suggested duration: 12s\n```\n\nNarration iron rules (beyond Shanyin red lines, narration-mode only):\n\n- **Single narration line ≤ 2 sentences, ≤ 60 characters**. Short TTS lines align with picture more easily; long lines get\n  stretched by ffmpeg freeze-frame and look stiff. Say more by splitting into multiple\n  consecutive narration beats.\n- **Narration uses third-person narrative voice** (\"钱夫人来到镇上 \u002F 没人知道他的真实身份\").\n  Never disguise dialog as narration.\n- **Dialog beat format matches drama mode** — cast.json names only.\n- **Narration:dialog ratio is your call** — core creative autonomy the user delegates to the screenwriter agent.\n  A 2–3 minute recap episode often starts around 70–85% narration + 15–30% dialog, but not mandatory.\n- No hard cap on beats per scene; suggest 3–12 (too few doesn't feel like recap, too many feels choppy).\n\nThe `## Scene N` heading uses the same N as the filename.\n\n## Cast \u002F lore overrides on top of Shanyin\n\nThese rules layer on top of the Shanyin SKILL — they're project glue, not\ncraft, so they live here:\n\n1. **Only use characters present in `cast.json`.** Generic crowd is fine\n   (`Passerby A`, `onlookers`, `waiter`). Anyone with a line or individual\n   description must be in cast.json.\n2. **`lore.forbidden` terms must never appear in **Action** or **Dialog**.\n3. **User-supplied dialog lines must appear verbatim** in some scene.\n   This is non-negotiable, regardless of what Shanyin craft suggests.\n4. **Costume \u002F hairstyle \u002F accessories — only mention when it CHANGES.**\n   The character's baseline look is encoded in the cast portrait, so\n   the director will never put it into a prompt. You only need to\n   describe an appearance detail when the *story* depends on it\n   changing — e.g. \"陆辰换上婚礼礼服\" \u002F \"苏晚摘下耳环掷在桌上\" \u002F\n   \"蓬头垢面\". Otherwise leave appearance to the portrait.\n   - If a costume genuinely needs to differ from the project cast for\n     this whole episode (episode-wide costume change), flag it at GATE 2 — the producer\n     will fork the cast into the episode tier (see `references\u002Fspark-video-cast\u002FSKILL.md`)\n     and the new portrait carries the change without any dialog\n     gymnastics. Don't try to solve it by repeatedly mentioning the outfit.\n5. **Age — call it out the first time a character appears in this\n   episode** (\"28 岁的陆辰\" \u002F \"年过五旬的钱夫人\"). The director reuses\n   that age verbatim in shot prompts; without it, the video model\n   drifts the apparent age 5-15 years between shots.\n6. **Episode-only NPC identification (CAST CHECK)** — at the bottom of\n   the LAST scene-NN.md, append a single HTML comment block:\n\n   ```markdown\n   \u003C!-- CAST CHECK\n   Leads (in cast):\n     - \u003Cname>\n   Named NPCs (need cast entry):\n     - \u003Cname>: \u003Cone-line appearance for director portrait generation>\n   Extras (no cast needed):\n     - \u003Cgeneric label>\n   -->\n   ```\n\n   The director uses this to generate NPC portraits before storyboarding.\n\n7. **Key props — call them out as proper nouns the moment they\n   appear, and flag every state change.** A \"key prop\" is any object that\n   (a) appears in 2+ shots and the audience would notice if it changed,\n   or (b) is a story-critical hero item even in one shot. Examples: red envelope,\n   key, ring, teddy bear, notebook, letter, murder weapon. Generic teacup \u002F phone \u002F umbrella\n   are NOT key props unless the plot turns on them.\n\n   Use a stable proper-noun in **Action** (\"陆辰把现金塞进 **红包**…\"), so\n   the director can pin it. When the prop visibly **changes state**\n   (intact → creased → torn \u002F closed → open \u002F new → worn \u002F clean → bloodstained),\n   make the change explicit in **Action**:\n\n   > 陆辰握紧 **红包**, 边角已被攥出折痕 (起皱). 后景钱夫人冷笑。\n\n   The state word in parentheses tells the director to swap the prop's\n   reference image (`红包-完整` → `红包-起皱` are two folders). Never\n   describe the prop's *visual properties* (material \u002F color \u002F print \u002F thickness) —\n   the reference image owns those, the same way the cast portrait owns\n   face appearance. Only mention the *narrative state* and the *action* on the prop.\n\n8. **Prop check (PROP CHECK) — append below CAST CHECK in the last\n   scene-NN.md**:\n\n   ```markdown\n   \u003C!-- PROP CHECK\n   Key props (need props\u002F\u003Cname> folder):\n     - 红包-完整: standard Chinese red envelope, flat with no creases  (appears in S01-003 \u002F S01-007)\n     - 红包-起皱: same red envelope creased from gripping (S03-002)\n     - 红包-撕碎: same red envelope torn in half on screen (S03-003)\n     - 戒指-完整: mother's heirloom, vintage gold ring, engraved inside (S02-005 \u002F S05-001)\n   -->\n   ```\n\n   Each entry is `\u003Cprop_name>-\u003Cstate>: \u003Cshort description>  (\u003Cshot id range>)`.\n   The director reads this BEFORE storyboarding and runs `uv run\n   scripts\u002Fscaffold.py prop --name \u003Cname>` + `bl image generate ...` for\n   each entry, then sets `Shot.props` accordingly. Skip the block if the\n   episode has no key props.\n\n## Pacing target\n\nRead `lore.duration_target_s` if present. The sum of all scene\n`**Estimated duration**` values should be ≈ that target (±15%). The producer\nverifies this after `storyboard.py compile`.\n\n| Target | Recommended scene count |\n|--------|-------------------------|\n| 60s    | 2–3 scenes |\n| 180s   | 4–6 scenes |\n| 300s   | 6–10 scenes |\n| 600s   | 10–18 scenes |\n\n## DON'Ts (spark-video-specific, on top of Shanyin red lines)\n\n- Don't write `script.md` or `storyboard.json` directly — only `scenes\u002Fscene-NN.md`.\n- Don't mention model names (happyhorse, wan, r2v, t2v) — that's the director's domain.\n- Don't write 图1\u002F图2 prompt syntax — that's the director's domain.\n- Don't invent character names not in `cast.json`.\n- Don't skip the `scene-NN.ready` sentinel — the director won't start otherwise.\n- Don't keep re-describing wardrobe \u002F hairstyle \u002F makeup inside **Action**. Mention an\n  appearance detail only when it CHANGES (rule 4 above).\n- Don't keep re-describing a key prop's visual properties (material \u002F color \u002F\n  shape \u002F print) once you've named it. The reference image owns those.\n  Mention the prop's *narrative state* (`intact` \u002F `creased` \u002F `torn`, e.g.\n  `红包-完整` \u002F `红包-起皱` \u002F `红包-撕碎`) only when\n  it CHANGES — that's the trigger for the director to swap reference\n  folders. Same rule, applied to objects.\n- Don't omit the PROP CHECK block when the episode contains a recurring\n  hero object. Without it, the director will paste the prop's\n  description into every shot prompt and the model will draw a\n  different object every time.\n",{"name":5,"description":7},"main","e396986d3c7d4b668808fa7e7d1b862e8e9e3de1","skills\u002Fspark-video\u002Freferences\u002Fspark-video-screenwriter",9,"https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills","https:\u002F\u002Fmodelstudioai.github.io\u002Fskills\u002F","https:\u002F\u002Fgithub.com\u002Fmodelstudioai","https:\u002F\u002Fbailian.console.aliyun.com\u002F","2026-11-14 04:33:35",{"source":8,"skillName":5,"commands":34,"prompt":44,"zipAvailable":45},[35,38,41],{"tool":36,"command":37},"npx","npx skills add modelstudioai\u002Fskills --skill spark-video-screenwriter",{"tool":39,"command":40},"bunx","bunx skills add modelstudioai\u002Fskills --skill spark-video-screenwriter",{"tool":42,"command":43},"pnpm","pnpm dlx skills add modelstudioai\u002Fskills --skill spark-video-screenwriter","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-screenwriter\u002FSKILL.md",true,[],[48,55,62,69,76,83],{"skillNo":49,"name":50,"nameZh":6,"description":51,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":52,"categories":53,"keywords":54,"publishedAt":19},"01M74FEYG88ZVEVSQC85Y97PF7","vox-video-director","Turn ONE topic into a finished Vox-style paper-collage explainer \u002F ad video, end to end with Aliyun Bailian CLI + local ffmpeg — script, collage keyframes, motion, voice-over, music, captions, all automated. Use this whenever the user wants a \"Vox style\" video, a paper\u002Ftorn-paper collage animation, a \"motion collage\", a narrated explainer or short ad built from AI-generated collage posters, a scrapbook-style tribute, or wants to turn a topic \u002F product \u002F person into a punchy narrated collage video — even if they don't say the word \"Vox\". Also use when reproducing Stav Zilber \u002F rom1trs \u002F Higgsfield-style collage ad workflows. Three input modalities: a topic (B-roll), a talking-head video (A-roll mode), or a single photo of a person\u002Fproduct anchored into the collage (C-roll mode). Triggers: \"vox video\", \"collage video\", \"motion collage\", \"paper collage explainer\", \"make a collage ad\", \"turn this topic into a collage video\", \"turn my photo\u002Fthis product shot into a collage video\".","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fvox-video-director\u002FSKILL.md",[],[],{"skillNo":56,"name":57,"nameZh":6,"description":58,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":59,"categories":60,"keywords":61,"publishedAt":19},"01M74FESD641YN15106ERYKHD7","spark-video-vfx-review","Pre-render quality gate. Read a finished storyboard.json and produce a structured review report flagging visual inconsistencies, prompt defects, and continuity errors that would waste render budget. You find problems; the director fixes them. Opt-in — bypassed unless the producer explicitly invokes you.","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-vfx-review\u002FSKILL.md",[],[],{"skillNo":63,"name":64,"nameZh":6,"description":65,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":66,"categories":67,"keywords":68,"publishedAt":19},"01M74FEEZHE6VJQAFWJ66HQDYF","spark-video-episode","One-shot autopilot orchestrator — runs the full spark-video pipeline (screenwriter ↔ director per-scene parallel → render chain-DAG parallel + per-clip review → stitch). User confirms at 4 gates (+ 1 mode gate at start + 1 BGM gate when bgm\u002F folder detected). Use when the user wants \"make me an episode\" in one command.","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-episode\u002FSKILL.md",[],[],{"skillNo":70,"name":71,"nameZh":6,"description":72,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":73,"categories":74,"keywords":75,"publishedAt":19},"01M74FE8JQBCHPC2GQHWJDGT55","spark-video-director","Translate a screenplay (one scene at a time) into a provider-agnostic storyboard fragment for the spark-video pipeline. Wraps Shanyin Super Director Master when available — the upstream Shanyin SKILL is the single source of truth for craft when present.","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-director\u002FSKILL.md",[],[],{"skillNo":77,"name":78,"nameZh":6,"description":79,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":80,"categories":81,"keywords":82,"publishedAt":19},"01M74FE3929KWXZG8RNFG1ZSXJ","spark-video-clip-review","Per-clip quality reviewer + render retry state machine. After each rendered shot, score it on 6 axes via `bl omni` (qwen3.5-omni-plus), then decide ACCEPT \u002F REJECT-and-rewrite \u002F REJECT-and-escalate. Handles up to N retry rounds with auto prompt rewriting; escalates to spark-video-director when retries are exhausted.","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-clip-review\u002FSKILL.md",[],[],{"skillNo":84,"name":85,"nameZh":6,"description":86,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":87,"categories":88,"keywords":89,"publishedAt":19},"01M74FDWYR8F0W6ZT3998PPKT5","spark-video-cast","Scaffold and generate reference assets for characters (cast), locations (movie-set \u002F set dressing), and key props — the three pillars of visual consistency in spark-video. Wraps bl image generate \u002F edit for portrait creation. Use when adding new characters\u002Flocations\u002Fprops or when costume\u002Fstate changes are needed.","https:\u002F\u002Fgithub.com\u002Fmodelstudioai\u002Fskills\u002Fblob\u002Fmain\u002Fskills\u002Fspark-video\u002Freferences\u002Fspark-video-cast\u002FSKILL.md",[],[]]