← Plugin catalog
Creativity

Buzzy

Buzzy AI, Inc. v0.1.0

Create images, videos, and ads

Create a free account to save this plugin.

From the developer

Declared capabilities

  • Generate images and videos
  • Make product and brand ads
  • Create cinematic and 3D clips

Not specified in the available declarations: limitations, pricing & account requirements.

Onboarding example

Generate an image from this brief.

View onboarding source →

Sources & dates

Developer declarations; not independently tested. Listing last checked Nov 7, 2026. Package observed Nov 7, 2026.

View package source →

Description

From the marketplace listing

Create images, videos, branded ads, and short films with Buzzy in ChatGPT. Start with an idea, a product link, or a reference image, choose a creative workflow, and refine the result through conversation. With Buzzy, you can: • Generate images and videos from text or a reference image, video, or audio file. Choose a model, then use a resolution, aspect ratio, and duration that model supports. • Upload a local image, video, or audio file and use the returned link as a reference for the next generation. • Make a UGC-style product ad from a product page, product image, or product description, including a showcase, unboxing, or creator-style review. • Produce a TVC or brand commercial from a product link, image, or description, with a storyboard, character sheets, clips, and music. • Create a single cinematic live-action shot, or a short 3D rendered clip, from a brief and an optional reference. • Build a longer live-action micro-film or animated story, with a script, character sheets, scene previews, clips, and music. • Pick a style, background, music track, or character image when a workflow asks you to choose one. Describe what you want to make, review the results in chat, and continue from the same project. Connect your Buzzy account to get started.

Language: English · Automatically detected from descriptions.

Matches for “reference”

Exact text from the indicated source. A mention alone does not establish support for your task.

Publisher description

Create images, videos, product ads, and short films with Buzzy. Start from an idea, a product link, or a reference image, then refine the result in conversation.

Publisher full description

Create images, videos, branded ads, and short films with Buzzy in ChatGPT. Start with an idea, a product link, or a reference image, choose a creative workflow, and refine the result through conversation. With Buzzy, you can: • Generate images and videos from text or a reference image, video, or audio file. Choose a model, then use a resolution, aspect ratio, and duration that model supports. • Upload a local image, video, or audio file and use the returned link as a reference for the next generation. • Make a UGC-style product ad from a product page, product image, or product description, including a showcase, unboxing, or creator-style review. • Produce a TVC or brand commercial from a product link, image, or description, with a storyboard, character sheets, clips, and music. • Create a single cinematic live-action shot, or a short 3D rendered clip, from a brief and an optional reference. • Build a longer live-action micro-film or animated story, with a script, character sheets, scene previews, clips, and music. • Pick a style, background, music track, or character image when a workflow asks you to choose one. Describe what you want to make, review the results in chat, and continue from the same project. Connect your Buzzy account to get started.

What changed

Keywords

Files & skills

File archives

Plugin package71 files · 231 KBBrowse files →
Skill instructions
3d-cgi28.7 KB

View saved version →

---
name: 3d-cgi
description: "Always use this skill to make a single-shot 2–10s 3D CGI / rendered video clip. Must trigger for /3d-cgi, 3d-cgi, 3D animation, CGI, Pixar-style, Unreal Engine, Blender, octane render, ray tracing, photorealistic 3D, computer-generated, digitally rendered, volumetric, subsurface scattering, PBR, or when a 3d-cgi run is already in progress (intent answers, specs, CGI prompt, clip, BGM merge) or the user skips a step, supplies a reference, jumps ahead, or revises an earlier step without restarting. Even a vague \"make it look 3D\" or \"help me make a 3D render\" must start this skill. After the first instruction, ask 2 professional direction questions grounded in that brief before the formal pipeline. Preset durations 2s / 4s / 6s / 8s / 10s (default 4s)."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - generate_video
  - get_task
  - merge_videos

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# 3D CGI Clip Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/3d-cgi/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。  
本文件是编排器。成片词由 agent **按本链 reference 直接写出**,禁止发明 `prompt_key`,禁止调 `run_prompt`。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。`merge_videos` 传 `projectId`;`generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。Host tools 开场问询 / 规格提问那一拍仍禁止同拍调 `create_project`。问询完成且即将调**需要项目 id** 的 Buzzy tool 时若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此。

**HARD — 宿主提问**:等人选择或确认时,必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`,且**该题全部可选项必须传齐**(禁止只塞一档)。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框;**规格题的编号列表必须每行列出该字段全部档位**。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用、编号单值、或缺档都算没发出提问。  
**例外 — 成片词 `content`**:结构化 CGI prompt **必须原文贴进对话**(禁止 code fence),不受「正文最多一句」限制。禁止说 on the canvas。贴完后仍发对应 Host tools 等待提问(FastForward / AutoRun 除外)。

| 何时读 | 文件 |
|--------|------|
| 开场组题(默认 2 题) | `intent-questions.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_video` 参数 | 本文件 §运行时填参 |
| 写成本片词时 | `prompt-write.md`(必读)+ 按需 `hook-framework.md` / `render-styles.md` / `materials.md` / `lighting-library.md` / `camera-language.md` / `particles.md` / `mistakes.md` / `examples.md` |
| 跑到成片 / 合成时 | `clip-generate.md` |

禁止短路径、读其它 skill、把 YAML 当自由 prompt。禁止发明 `prompt_key`。禁止 Higgsfield / `@material[name]`。禁止把 4K / 60fps / codec / 比特率写成平台承诺。

用户可见文案 **固定 English**。禁止出现 `on the canvas`。`SessionScriptLanguage` / `user_lang` 只约束对白与可读场景描述。  
参考素材**一律走 URL 字段**(`image_url` / `image_urls` / `video_url` / `video_urls`,以各 tool 声明为准)。**禁止传 `nodeIds`**。用户上传的参考图 / 视频有才加;禁止编造 URL。

**HARD — generate_image / generate_video 阻塞轮询**:`generate_video` 只返回 `task_id`,**不算完成**。同一 continuation 必须立刻反复调用 `get_task(task_id)`,直到终态:`TASK_STATE_PENDING` / `TASK_STATE_RUNNING` → 继续 `get_task`(禁止停住、禁止等用户下一句、禁止只写「生成中」就结束、禁止发下一阶段提问);`TASK_STATE_SUCCEEDED` → 记下返回里的**素材 URL**,本段才算完成;失败态 → 说明失败,禁止当成功、禁止再叠一次 generate_image / generate_video。Apps widget 只展示进度,**不能**代替 `get_task`。后续 `generate_video` / `merge_videos` 都用这些 URL,**禁止**把 `node_id` 当素材入参。

**HARD — 本链无画布操作**:禁止 `PushStyleList` / `PushConfiguration` / `PushDirectionTool` / `PushSu

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 12 — file references are metadata, not package contents.

ai-animation44.1 KB

View saved version →

---
name: ai-animation
description: "Always use this skill to make a flexible long-form animation / animated micro-film / multi-chapter storyboard-to-video. Must trigger for /ai-animation, ai-animation, long animation, animated feature, animated micro-film, or when an ai-animation run is already in progress (intent answers, style, specs, script, character sheets, preview images, video prompts, clips, BGM merge) or the user skips a step, supplies an asset, jumps ahead, or revises an earlier step without restarting. Even a vague \"make a long animation\" or \"help me make an animated feature\" must start this skill. After the first instruction, ask 2 professional direction questions grounded in that brief before the formal pipeline. Preset durations 15s / 15-30s / 30-1min (default 15s) or a typed total such as 180s / 300s."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - run_prompt
  - generate_image
  - generate_video
  - get_task
  - merge_videos
  - choose_skill_asset

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# AI Animation Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/ai-animation/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。  
本文件是编排器,不是 `run_prompt` 的填空稿,不要改写成 `user_lang`,不要压成短 brief。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。workflow tool(`run_prompt` / `merge_videos`)传 `projectId`;`generate_image` / `generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。`choose_skill_asset` **不需要** `projectId` / `project_id`。Host tools 开场问询 / 规格提问那一拍、以及风格卡那一拍,仍禁止同拍调 `create_project`。风格卡只调 `choose_skill_asset`。问询完成且即将调**需要项目 id** 的 Buzzy tool 时若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此:还没有 id 就先创建,再调带项目字段的 tool。

**HARD — 宿主提问**:等人选择或确认时,必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`,且**该题全部可选项必须传齐**(禁止只塞一档)。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框;**规格题的编号列表必须每行列出该字段全部档位**。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用、编号单值、或缺档都算没发出提问。  
**例外 — 风格库**:`choose_skill_asset` 打开风格单选控件。必传 `style_template: "animation"` + `title: "Styles"`。返回值(`items`)只表示控件已打开,**不是**确认。有控件 → **停住**等 `Confirmed Styles info`;禁止同拍 Host tools。无控件宿主才把 `items[].name` 塞进 Host tools。确认前禁止进 `settings`。  
**例外 — run_prompt `content`**:剧本 Markdown 与各段 Video Prompt **必须原文贴进对话**(禁止 code fence),不受「正文最多一句」限制。禁止说 on the canvas。贴完后仍发对应 Host tools 等待提问(FastForward / AutoRun 除外)。

| 何时读 | 文件 |
|--------|------|
| 开场组题(默认 2 题) | `intent-questions.md` |
| 风格卡 / choose_skill_asset | `style-prompt.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_image` / `generate_video` 参数 | 本文件 §运行时填参 |
| 跑到该步且未跳过时 | `script-prompt.md` / `assets-prompt.md` / `frames-prompt.md` / `video-prompts.md` / `clips-merge.md` |

已注册文案模板(只许用这些 key,禁止发明,禁止借用 ugc-ads / tvc-ads):  
`long_video_step1_script` / `long_video_step3_video_prompts`。  
`long_video_step2_frames` 主路径不用。禁止用 STEP2 写成片词,禁止为出预览图去补 STEP2。  
`long_video_step3_video_prompts` 对外仍是一次调用。Agent **必须**传 `instruction.script_markdown` + 规格(`duration_sec` / `ratio` / `model`)。**禁止**传 `frames[]`。Host 内部跑 STEP2,细镜全文零删减进 `records[]`,与这份剧本并行喂给 Gemini。写词规则是**细镜主参考、剧本辅参考**:画面先写细镜;剧本补缺口;同一事实两边否定才选边(台词→剧本,

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 8 — file references are metadata, not package contents.

ai-story44.1 KB

View saved version →

---
name: ai-story
description: "Always use this skill to make a flexible long-form live-action cinematic micro-film / multi-chapter storyboard-to-video. Must trigger for /ai-story, ai-story, long video, micro-film, storyboard-to-finished-film, script-to-clip, live-action long video, or when an ai-story run is already in progress (intent answers, style, specs, script, character sheets, preview images, video prompts, clips, BGM merge) or the user skips a step, supplies an asset, jumps ahead, or revises an earlier step without restarting. Even a vague \"make a long video\" or \"help me make a micro-film\" must start this skill. After the first instruction, ask 2 professional direction questions grounded in that brief before the formal pipeline. Preset durations 15s / 15-30s / 30-1min (default 15s) or a typed total such as 180s / 300s."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - run_prompt
  - generate_image
  - generate_video
  - get_task
  - merge_videos
  - choose_skill_asset

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# AI Story Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/ai-story/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。  
本文件是编排器,不是 `run_prompt` 的填空稿,不要改写成 `user_lang`,不要压成短 brief。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。workflow tool(`run_prompt` / `merge_videos`)传 `projectId`;`generate_image` / `generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。`choose_skill_asset` **不需要** `projectId` / `project_id`。Host tools 开场问询 / 规格提问那一拍、以及风格卡那一拍,仍禁止同拍调 `create_project`。风格卡只调 `choose_skill_asset`。问询完成且即将调**需要项目 id** 的 Buzzy tool 时若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此:还没有 id 就先创建,再调带项目字段的 tool。

**HARD — 宿主提问**:等人选择或确认时,必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`,且**该题全部可选项必须传齐**(禁止只塞一档)。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框;**规格题的编号列表必须每行列出该字段全部档位**。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用、编号单值、或缺档都算没发出提问。  
**例外 — 风格库**:`choose_skill_asset` 打开风格单选控件。必传 `style_template: "ai_film"` + `title: "Styles"`。返回值(`items`)只表示控件已打开,**不是**确认。有控件 → **停住**等 `Confirmed Styles info`;禁止同拍 Host tools。无控件宿主才把 `items[].name` 塞进 Host tools。确认前禁止进 `settings`。  
**例外 — run_prompt `content`**:剧本 Markdown 与各段 Video Prompt **必须原文贴进对话**(禁止 code fence),不受「正文最多一句」限制。禁止说 on the canvas。贴完后仍发对应 Host tools 等待提问(FastForward / AutoRun 除外)。

| 何时读 | 文件 |
|--------|------|
| 开场组题(默认 2 题) | `intent-questions.md` |
| 风格卡 / choose_skill_asset | `style-prompt.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_image` / `generate_video` 参数 | 本文件 §运行时填参 |
| 跑到该步且未跳过时 | `script-prompt.md` / `assets-prompt.md` / `frames-prompt.md` / `video-prompts.md` / `clips-merge.md` |

已注册文案模板(只许用这些 key,禁止发明,禁止借用 ugc-ads / tvc-ads):  
`long_video_step1_script` / `long_video_step3_video_prompts`。  
`long_video_step2_frames` 主路径不用。禁止用 STEP2 写成片词,禁止为出预览图去补 STEP2。  
`long_video_step3_video_prompts` 对外仍是一次调用。Agent **必须**传 `instruction.script_markdown` + 规格(`duration_sec` / `ratio` / `model`)。**禁止**传 `frames[]`。Host 内部跑 STEP2,细镜全文零删减进 `records[]`,与这份剧本并行喂给 Gemini。写词规则是**细镜主参考、剧本辅参考**:画面先写细镜;剧本补缺口;同一事实两边否定才选边(台词→剧

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 8 — file references are metadata, not package contents.

cinematic28.4 KB

View saved version →

---
name: cinematic
description: "Always use this skill to make a single-shot 4–15s live-action cinematic / film-look / Hollywood-style video clip. Must trigger for /cinematic, cinematic, film look, movie scene, movie-quality, dramatic lighting, depth of field, lens flare, anamorphic, letterbox, noir, epic, Steadicam, dolly, crane shot, or when a cinematic run is already in progress (intent answers, specs, cinematic prompt, clip, BGM merge) or the user skips a step, supplies a reference, jumps ahead, or revises an earlier step without restarting. Even a vague \"make it cinematic\" or \"help me make a cinematic shot\" must start this skill. After the first instruction, ask 2 professional direction questions grounded in that brief before the formal pipeline. Preset durations 4s / 8s / 10s / 15s (default 8s)."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - generate_video
  - get_task
  - merge_videos

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# Cinematic Clip Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/cinematic/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。  
本文件是编排器。成片词由 agent **按本链 reference 直接写出**,禁止发明 `prompt_key`,禁止调 `run_prompt`。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。`merge_videos` 传 `projectId`;`generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。Host tools 开场问询 / 规格提问那一拍仍禁止同拍调 `create_project`。问询完成且即将调**需要项目 id** 的 Buzzy tool 时若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此。

**HARD — 宿主提问**:等人选择或确认时,必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`,且**该题全部可选项必须传齐**(禁止只塞一档)。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框;**规格题的编号列表必须每行列出该字段全部档位**。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用、编号单值、或缺档都算没发出提问。  
**例外 — 成片词 `content`**:结构化 cinematic prompt **必须原文贴进对话**(禁止 code fence),不受「正文最多一句」限制。禁止说 on the canvas。贴完后仍发对应 Host tools 等待提问(FastForward / AutoRun 除外)。

| 何时读 | 文件 |
|--------|------|
| 开场组题(默认 2 题) | `intent-questions.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_video` 参数 | 本文件 §运行时填参 |
| 写成本片词时 | `prompt-write.md`(必读)+ 按需 `hook-framework.md` / `timeline.md` / `camera-language.md` / `lighting-library.md` / `color-grading.md` / `sound-design.md` / `platform-optimization.md` / `mistakes.md` / `examples.md` |
| 跑到成片 / 合成时 | `clip-generate.md` |

禁止短路径、读其它 skill、把 YAML 当自由 prompt。禁止发明 `prompt_key`。禁止 Higgsfield / `@material[name]`。

用户可见文案 **固定 English**。禁止出现 `on the canvas`。`SessionScriptLanguage` / `user_lang` 只约束对白与可读场景描述。  
参考素材**一律走 URL 字段**(`image_url` / `image_urls` / `video_url` / `video_urls`,以各 tool 声明为准)。**禁止传 `nodeIds`**。用户上传的参考图 / 视频有才加;禁止编造 URL。

**HARD — generate_image / generate_video 阻塞轮询**:`generate_video` 只返回 `task_id`,**不算完成**。同一 continuation 必须立刻反复调用 `get_task(task_id)`,直到终态:`TASK_STATE_PENDING` / `TASK_STATE_RUNNING` → 继续 `get_task`(禁止停住、禁止等用户下一句、禁止只写「生成中」就结束、禁止发下一阶段提问);`TASK_STATE_SUCCEEDED` → 记下返回里的**素材 URL**,本段才算完成;失败态 → 说明失败,禁止当成功、禁止再叠一次 generate_image / generate_video。Apps widget 只展示进度,**不能**代替 `get_task`。后续 `generate_video` / `merge_videos` 都用这些 URL,**禁止**把 `node_id` 当素材入参。

**HARD — 本链无画布操作**:禁止 `PushStyleList` / `PushConfiguration` 

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 13 — file references are metadata, not package contents.

setup1.42 KB

View saved version →

---
name: setup
description: Set up the Buzzy plugin after install. Use when the user just installed Buzzy, asks to connect Buzzy, or needs to confirm the Buzzy MCP server is signed in.
---

# Buzzy setup

Confirm the bundled Buzzy MCP server is connected, then stop. Do not ask for an API key. This package does not store credentials.

1. The host signs the user in with Buzzy OAuth using the ChatGPT client `https://chatgpt.com/oauth/client.json` and the `mcp` scope. No client secret is stored in this package. If the connection is missing or unauthorized, ask the user to finish that sign-in and wait.
2. Call `list_model` once. A successful response means the server is ready. If it fails because the user is not signed in, send them back to the host sign-in and do not retry with a made-up token.
3. Tell the user Buzzy is ready, in one short sentence. Do not start a generation during setup.

After setup, use the connected server directly:

- A single image or video: call `list_model`, copy `resolution`, `aspect_ratio`, and `duration` from the chosen model, then call `generate_image` or `generate_video`. Poll `get_task` until it succeeds. Do not submit the same generation again while it is still running.
- A narrated, ad, UGC, or story video: call `load_workflow` with no arguments, then load the matching workflow by name. Call `create_project` before any tool that needs `projectId` or `project_id`, and reuse that id for the rest of the run.
tvc-ads62.4 KB

View saved version →

---
name: tvc-ads
description: "Always use this skill to make a TVC / brand commercial from a product link, product image, or product description. Must trigger for /tvc-ads, tvc-ads, TVC ad, brand commercial, cinematic product film, brand ad, film commercial, or when the user drops a product page and asks for a TVC or brand film. Also use when a tvc-ads run is already in progress (Confirmed Product Info, selling-point confirm, specs, storyboard cards, character sheets, preview / first frames, clips, BGM merge) or the user skips a step, supplies an asset, jumps ahead, or revises an earlier step without restarting. Even a vague \"make a TVC\" or \"help me make a brand ad\" must start this skill. First-round with-questions (default) asks 1 content-direction question; no-questions starts the pipeline."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - parse_product_info
  - display_product_info
  - run_prompt
  - generate_image
  - generate_video
  - get_task
  - merge_videos

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# TVC Ads Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/tvc-ads/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。workflow tool(`parse_product_info` / `display_product_info` / `run_prompt` / `merge_videos`)传 `projectId`;`generate_image` / `generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。Host tools 开场问询那一拍仍禁止同拍调 `create_project`;问询完成(或 `IntentQuestions=false`)后、进 `product` 的第一拍若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此:还没有 id 就先创建,再调带项目字段的 tool。

**HARD — 宿主提问**:等人选择或确认时(**卖点分析、Apps 商品控件确认除外**),必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用或编号列表都算没发出提问。  
**例外 — 商品确认**:`parse_product_info` 打开编辑控件。有产品页 URL 则传 `productLink`;没有则**不要编造链接、不要传 `productLink`**,打开空表。返回值(即使带 Title / Description / Images)只表示控件已打开,**不是**确认。然后**停住**,等用户确认。确认信号(任一即可):消息含 `Confirmed Product Info`;或用户明确确认/继续该商品(`Confirm` / `Continue` / `looks good` / 确认 / 继续)。在此之前禁止 `display_product_info`,禁止 Host tools 问要不要继续。无 iframe 宿主才用 Host tools 收集标题/描述/图。  
**例外 — 卖点分析**:不走 Host tools。run_prompt 返回后按 `selling-analysis-prompt.md` 贴三段:固定引导句 `Here are the product selling points:` + `content` 原文 markdown + 固定确认句 `Confirm to continue, or tell me what you'd like to change.`,然后停住等用户下一条自由回复。禁止只贴正文、不说明这是商品卖点。用户回复 `ok` / `continue` / `looks good` / `confirm` / 确认 / 继续 = **已确认**,必须进入 `info_card`。  
**例外 — run_prompt `content`**:卖点正文必须原文贴进对话(禁止 code fence),不受「正文最多一句」限制。分镜 `content` 是 GFM 管道表;Chat **禁止**贴 `|` 管道表,必须按 `script-table-format.md` 转成分镜卡片。卡片贴完后仍发 Host tools Continue/Revise。

| 何时读 | 文件 |
|--------|------|
| 开场组题(`with-questions` 时恰好 1 题) | `intent-questions.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_image` / `generate_video` 参数 | 本文件 §运行时填参 |
| 商品确认后的卖点分析 | `selling-analysis-prompt.md` |
| 用户自带参考视频时 | 本文件 §`ref_video`(无单独 reference;禁止搜片) |
| 分镜 / 角色图 / 对话卡片 | `storyboard-prompts.md` + `script-table-format.md` |
| 脚本计划 / Master 提示词 / 开场视觉 / 首帧 | `plan-and-opening.md` |
| 声音身

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 8 — file references are metadata, not package contents.

ugc-ads46.4 KB

View saved version →

---
name: ugc-ads
description: "Always use this skill to make a UGC product ad from a product link, product image, or product description. Must trigger for /ugc-ads, ugc-ads, UGC ad, UGC ads, product ad, TikTok ad, Reels ad, shoppable short, creator review, unboxing ad, product seeding video, sales video, Amazon/Shopify/product URL, or when the user drops a product page and says make a video or shoot a clip. Also use when a ugc-ads run is already in progress (Confirmed Product Info, selling-point confirm, specs, storyboard cards, first frames, clips, merge) or the user skips a step, supplies an asset, jumps ahead, or revises an earlier step without restarting. Even a vague \"make a UGC\" or \"help me make a UGC\" must start this skill. First-round with-questions (default) asks 1 content-direction question; no-questions starts the pipeline."
source: "buzzy"
risk: safe
scope: mcp
allowed-tools: |
  Buzzy MCP tools
  - create_project
  - parse_product_info
  - display_product_info
  - run_prompt
  - generate_image
  - generate_video
  - get_task
  - merge_videos

  Host tools — not part of this MCP server
  - AskQuestion
  - ask_question
  - AskUserQuestion
---

# UGC Ads Orchestrator

本链自包含。查阅本 skill 文件只用完整前缀 `skills/mcp/ugc-ads/reference/`。  
**怎么调平台能力、填哪些参数,对照各 tool 自己的声明**;本文件只写目的、产物和硬底线。

**HARD — allowed-tools 分类**:`Buzzy MCP tools` 是本 MCP 服务注册的平台 tool,经 Buzzy MCP `tools/call`。`Host tools — not part of this MCP server` 是宿主内置提问 tool,**不是** Buzzy MCP tool,也不走 `load_workflow` / ExecuteTools。Cursor 用 `AskQuestion`,Claude 用 `AskUserQuestion` / `ask_question`,Codex 用同类。选择题必须调 Host tools 里当前可用的那个,禁止因为「不在 Buzzy MCP tools 列表里」就改成编号纯文本。  
**HARD — create_project**:进入任何需要 `projectId` / `project_id` 的 Buzzy MCP tool 之前,必须先调 `create_project` 拿到真实 id。本会话只创建一次,后续全程复用。workflow tool(`parse_product_info` / `display_product_info` / `run_prompt` / `merge_videos`)传 `projectId`;`generate_image` / `generate_video` 传 `project_id`。禁止编造 id,禁止省略,禁止因为 generate_image / generate_video 写了「可自动建项目」就跳过。`create_project` 的入参是 `name`(可读项目名),不要把 catalog 的 `name` 或 `workflow` 抄过来。Host tools 开场问询那一拍仍禁止同拍调 `create_project`;问询完成(或 `IntentQuestions=false`)后、进 `product` 的第一拍若还没有 id,必须先 `create_project`。Jump / FastForward 到更后阶段同此:还没有 id 就先创建,再调带项目字段的 tool。

**HARD — 宿主提问**:等人选择或确认时(**卖点分析、Apps 商品控件确认除外**),必须对 Host tools 里当前可用名字真调用一次。选项只进提问 tool 的 `options`,且**该题全部可选项必须传齐**(禁止只塞一档)。禁止用 `1.` `2.` `3.` 编号列表、bullet、`Recommended:` 前缀段落作为用户可见选项——那样写 = 本步失败,必须立刻改成真调用。assistant 正文最多一句问句或状态句,**不要复述选项**。不要自己造第 4 个 `Other` 选项:Cursor `AskQuestion` 自带 Other 自由输入。仅当 `tools/list` 里 `AskQuestion` / `ask_question` / `AskUserQuestion` 三个名字都没有时,才允许编号列表,并写明本宿主没有选择框;**规格题的编号列表必须每行列出该字段全部档位**(`Language: English / Spanish / Arabic / Custom`),禁止写成单值(`Language: English`)。Cursor 有 `AskQuestion`,禁止走这条。媒体产物就绪只用 assistant 正文状态句。伪调用、编号单值、或缺档都算没发出提问。  
**例外 — 商品确认**:`parse_product_info` 打开编辑控件。有产品页 URL 则传 `productLink`;没有则**不要编造链接、不要传 `productLink`**,打开空表。返回值(即使带 Title / Description / Images)只表示控件已打开,**不是**确认。然后**停住**,等用户确认。确认信号(任一即可):消息含 `Confirmed Product Info`;或用户明确确认/继续该商品(`Confirm` / `Continue` / `looks good` / 确认 / 继续)。在此之前禁止 `display_product_info`,禁止 Host tools 问要不要继续。无 iframe 宿主才用 Host tools 收集标题/描述/图。  
**例外 — 卖点分析**:不走 Host tools。run_prompt 返回后按 `selling-analysis-prompt.md` 贴三段:固定引导句 `Here are the product selling points:` + `content` 原文 markdown + 固定确认句 `Confirm to continue, or tell me what you'd like to change.`,然后停住等用户下一条自由回复。禁止只贴正文、不说明这是商品卖点。用户回复 `ok` / `continue` / `looks good` / `confirm` / 确认 / 继续 = **已确认**,必须进入 `info_card`。  
**例外 — run_prompt `content`**:卖点正文必须原文贴进对话(禁止 code fence),不受「正文最多一句」限制。分镜 `content` 是 9 列管道表;Chat **禁止**贴 `|` 管道表,必须按 `script-table-format.md` 转成分镜卡片。卡片贴完后仍发 Host tools Continue/Revise。

| 何时读 | 文件 |
|--------|------|
| 开场组题(`with-questions` 时恰好 1 题) | `intent-questions.md` |
| 下游缺输入时怎么组槽 | `slot-adaptation.md` |
| 任意状态句 / 提问引导 | 本文件 §双通道文案 |
| 填 `generate_image` / `generate_video` 参数 | 本文件 §运行时填参 |
| 商品确认后的卖点分析 | `selling-analysis-prompt.md` |
| 用户自带参考视频时 | 本

Preview of the first 4,000 characters. Read full instructions →

Referenced files: 11 — file references are metadata, not package contents.

Package details

Publisher declarations from the archived package. These are separate from our research and the live service's terms.

Package author
Buzzy AI, Inc.
Keywords
See keywords
Getting started skill
./skills/setup/SKILL.md

Declared capabilities

  • Generate images and videos
  • Make product and brand ads
  • Create cinematic and 3D clips
  • Produce stories

Package observed Nov 7, 2026.

Technical details
First seen
Oct 24, 2026 · 12:40 UTC
Last seen
Nov 7, 2026 · 12:00 UTC
Collection status
Collected

plugin_asdk_app_6ad8acf81ccc819193bc600c3a69f612

Download plugin data (JSON)

Plugin details: served from cache. Prepared . Refreshed when plugin data changes or after 6 hours. Your account and Follow status are live.