Skill flagged — suspicious patterns detected

ClawHub Security flagged this skill as suspicious. Review the scan results before using.

Doubao Video Creator

v2.0.0

基于火山引擎豆包视频生成模型,辅助用户配置项目并分场景确认提示词,逐步生成并合成专业短视频。

0· 89·1 current·1 all-time

Install

OpenClaw Prompt Flow

Install with OpenClaw

Best for remote or guided setup. Copy the exact prompt, then paste it into OpenClaw for cindypapa/doubao-video-creator.

Previewing Install & Setup.
Prompt PreviewInstall & Setup
Install the skill "Doubao Video Creator" (cindypapa/doubao-video-creator) from ClawHub.
Skill page: https://clawhub.ai/cindypapa/doubao-video-creator
Keep the work scoped to this skill only.
After install, inspect the skill metadata and help me finish setup.
Use only the metadata you can verify from ClawHub; do not invent missing requirements.
Ask before making any broader environment changes.

Command Line

CLI Commands

Use the direct CLI path if you want to install manually and keep every step visible.

OpenClaw CLI

Bare skill slug

openclaw skills install doubao-video-creator

ClawHub CLI

Package manager switcher

npx clawhub@latest install doubao-video-creator
Security Scan
VirusTotalVirusTotal
Suspicious
View report →
OpenClawOpenClaw
Suspicious
medium confidence
Purpose & Capability
The skill's name, description, and code (prompt_generator, video_project, doubao_video_creator, etc.) are consistent with a Doubao-based video creation assistant. It legitimately asks for a Doubao API key and offers project-level config. However, many publishing-related files and examples include unrelated credentials (GitHub PAT, ClawHub token) and push/publish scripts that are not required for runtime video generation, which is unexpected and increases risk.
!
Instruction Scope
SKILL.md instructs the agent to collect and persist API keys to ~/.openclaw/workspace/doubao-config.json and project.json — reasonable for a project tool — but the runtime docs also reference explicit, hard-coded example keys and other external API keys (通义万相) and an absolute /root path (/root/.openclaw/workspace/wanxiang_generate.py). The instructions and included scripts reference publishing workflows and commands that accept tokens; the presence of hardcoded credentials and instructions to run publish/push scripts broaden the scope beyond the stated creation flow and risk accidental credential exposure or exfiltration.
Install Mechanism
There is no explicit install spec (instruction-only in registry metadata) which reduces installer risk, but the skill package contains many code files and shell scripts (publish.sh, quick_push_github.sh). No external downloads are specified. The build/install approach itself is not high-risk, but included scripts that perform git/publish operations and reference credentials raise concern if executed.
!
Credentials
Registry metadata declares no required env vars, but the skill expects the user to provide Doubao and optionally an image-generation API key. More problematic: multiple files contain hardcoded example secrets (Doubao example key, a 通义万相 API key, a GitHub PAT, and a ClawHub token). Publishing logs and guides include tokens and example credentials which are unrelated to normal skill runtime and are disproportionate to the task of generating videos.
!
Persistence & Privilege
The skill persists API configuration and project state to ~/.openclaw/workspace and per-project project.json (expected for project workflows). That by itself is OK. Concern arises because scripts and docs encourage committing/pushing repos and include hardcoded tokens — a skill with writeable local config plus embedded publish scripts could be used (accidentally or maliciously) to exfiltrate credentials or push data if those scripts are executed.
Scan Findings in Context
[hardcoded_doubao_api_key_in_docs] expected: SKILL.md and examples include a Doubao API Key example string. An example value for the service the skill integrates with is plausible, but hardcoded keys in distributed files are risky — they may be real or encourage unsafe copy/paste.
[hardcoded_wanxiang_api_key_in_docs] expected: The SKILL.md and examples include a 通义万相 (image-generation) API key. If the skill supports image-generation integration this is relevant, but embedding keys in files is inappropriate and risky.
[hardcoded_github_pat_in_publish_guide] unexpected: FINAL_PUBLISH_GUIDE.md and related docs contain a GitHub personal access token (ghp_...) and username/password example used for pushing to GitHub. This is unrelated to video generation and should never be present in published skill files.
[hardcoded_clawhub_token_in_publish_log] unexpected: CLAWHUB_PUBLISH_LOG.md contains a ClawHub auth token (clh_...). This token is unrelated to runtime functionality and its presence indicates sensitive information leaked into the package.
What to consider before installing
This skill largely does what it says (help create Doubao videos) but contains red flags you should address before installing or using it: - Do NOT assume any example API keys/tokens in the files are safe to use — treat them as leaked secrets. If you or your organization used any of those values, rotate the tokens immediately. - Inspect the code (especially publish.sh, quick_push_github.sh, FINAL_PUBLISH_GUIDE.md, CLAWHUB_PUBLISH_LOG.md and any scripts) for hardcoded credentials or commands that push data. Do not run publish/quick-push scripts without reviewing them. - Prefer to run the skill in a restricted/sandboxed environment and avoid running included shell scripts automatically. Verify that the skill asks you interactively for your Doubao API key rather than pulling any environment variables you haven't provided. - Search the package for patterns like 'ghp_', 'clh_', 'API_KEY', 'sk-' and remove or redact any embedded secrets before use or publication. - If you plan to install from a remote source, prefer a verified upstream repository (official GitHub URL) and confirm the repo owner/trustworthiness. If uncertain, ask the author for a cleaned release without embedded secrets. If you want, I can: (1) list exact files/lines that contain credential-looking values, (2) suggest safe config handling code snippets (prompted input + secure storage), or (3) produce a short checklist to sanitize the repo before installing.

Like a lobster shell, security has layers — review code before you run it.

latestvk979ysxbd3vmdq2bqthc3ktn3x842pqf
89downloads
0stars
2versions
Updated 3w ago
v2.0.0
MIT-0

doubao-video-creator - 豆包视频创作助手 v2.0

📋 技能描述

使用火山引擎豆包(Doubao Seedance)视频生成模型,帮助用户创作专业视频。支持项目化配置管理分场景生成方式选择完整状态追踪

🎯 触发条件

用户提到以下关键词时触发:

  • "豆包生成视频"
  • "帮我做视频"
  • "制作短视频"
  • "视频创作"
  • "生成宣传视频"
  • "doubao 视频"
  • "火山引擎视频"

🔄 工作流程(v2.0 增强版)

阶段 0:首次配置(新增!)✅

如果是首次使用,需要询问并保存以下配置:

  1. 豆包 API Key 配置

    您好!我是豆包视频创作助手 🎬
    
    检测到您是首次使用,需要先配置豆包 API 信息:
    
    1️⃣ 请提供您的火山引擎豆包 API Key
       - 获取地址:https://console.volcengine.com/ark
       - 格式示例:65ae8f92-134c-4194-a3af-6e6cb74284e0
    
    2️⃣ 选择文生视频模型版本:
       - doubao-seedance-2-0-pro(推荐,最新版本)
       - doubao-seedance-1-5-pro(稳定版本)
       - doubao-seedance-1-0-pro(基础版本)
    
    3️⃣ 选择图生视频模型版本(如需要):
       - 同上,或保持与文生视频一致
    
    配置将保存在项目文件中,后续使用无需重复输入。
    
  2. 配置保存位置

    • 全局配置:~/.openclaw/workspace/doubao-config.json
    • 项目配置:doubao-video-projects/{project_id}/project.json

阶段 1:需求收集

主动询问并告知用户可提供的资料类型

您好!我是豆包视频创作助手 🎬

为了生成更符合您需求的视频,请告诉我以下信息:

**1. 视频主题**:是什么产品/服务?想表达什么内容?

**2. 目标受众**:视频给谁看的?(如:消费者、客户、内部员工等)

**3. 视频风格**:科技感?温馨?专业?活泼?电影感?

**4. 参考资料**(可选,但强烈建议提供):

📄 **文档资料**:
   - 产品介绍文档
   - 剧本/大纲
   - 品牌指南
   - 其他文字材料

🔗 **网页链接**:
   - 产品官网
   - 参考视频链接
   - 竞品案例

🖼️ **参考图片**(重要!):
   - 产品图片
   - 参考场景
   - 人物形象
   - 风格参考

📝 **文字描述**:
   - 具体需求说明
   - 特殊要求
   - 禁忌内容

---

您可以直接发送文件、图片、链接或文字描述给我~
有参考资料会让视频更精准哦!✨

阶段 2:脚本规划

  1. 分析参考资料
  2. 生成视频场景规划脚本(默认 12 秒)
  3. 发送脚本给用户确认
  4. 根据反馈调整脚本

阶段 2.5:关键元素生成

  1. 提取关键元素:从脚本中识别人物、场景、物品
  2. 生成参考图:用通义万相生成关键元素图片
  3. 用户确认:发送参考图给用户确认
  4. 调整优化:根据反馈重新生成参考图

阶段 3:分场景生成方式确认(新增!)✅

对每个场景,单独确认生成方式

🎬 场景 1 确认(0-4 秒):开场展示

请选择生成方式:
  A) 文生视频 - 直接用文字提示词生成
  B) 图生视频 - 需要先生成场景/人物原型图

当前场景描述:手表特写,镜头从下往上扫过

如果选择 B) 图生视频

  • 自动生成该场景的人物/场景原型图
  • 用户确认原型图
  • 使用原型图 + 提示词生成视频

如果选择 A) 文生视频

  • 直接使用文字提示词生成视频

阶段 4:提示词确认(关键!)✅

⚠️ 重要:必须先确认提示词,才能生成视频!

对每个场景:

  1. 生成视频提示词(基于脚本和参考图)
  2. 发送提示词给用户确认
    • 展示完整提示词
    • 说明将生成的画面内容
    • 显示预计成本
    • 提醒确认后无法撤销
  3. 用户确认提示词后才进入下一步
  4. 如用户要求修改,调整提示词后重新确认

为什么这步很重要?

  • ✅ 避免生成不符合预期的视频
  • ✅ 减少反复生成的成本浪费
  • ✅ 确保每个场景一次成功
  • ✅ 提高整体制作效率

阶段 5:分场景生成(逐个进行!)✅

⚠️ 重要:逐个场景生成,不要批量!

对每个场景按顺序执行:

  1. 检查提示词是否已确认

    • 未确认 → 返回阶段 4
    • 已确认 → 继续
  2. 调用豆包 API 生成视频(4 秒/场景)

    • 文生视频:只用提示词
    • 图生视频:提示词 + 原型图
  3. 等待生成完成(约 1-2 分钟)

  4. 发送视频给用户确认

    • 展示生成的视频
    • 询问是否满意
  5. 用户确认

    • 满意 → 标记为 confirmed,继续下一个场景
    • 不满意 → 记录反馈,修改提示词后重新生成
  6. 完成一个场景后再开始下一个

    • 不要同时生成多个场景
    • 确保每个场景都确认后再继续

为什么逐个生成?

  • ✅ 及时发现问题,避免全部生成后才发现问题
  • ✅ 减少浪费,一个场景不满意只重生成这一个
  • ✅ 用户参与度高,可以随时调整方向
  • ✅ 降低总体成本

阶段 6:视频合成

  1. 将所有确认的场景视频按顺序合成
  2. 添加转场效果
  3. 添加背景音乐(可选)
  4. 发送最终视频给用户

🛠️ 工具调用

1. 火山引擎豆包 API(视频生成)

API_KEY = "65ae8f92-134c-4194-a3af-6e6cb74284e0"
BASE_URL = "https://ark.cn-beijing.volces.com/api/v3"

# 创建视频任务
POST /contents/generations/tasks
{
  "model": "doubao-seedance-1-5-pro-251215",
  "content": [{"type": "text", "text": "提示词"}],
  "parameters": {
    "ratio": "16:9",
    "duration": 4,
    "watermark": False
  }
}

2. 通义万相 API(图片生成)

API_KEY = "sk-d05aba5a2dae4453b97ed07fdb983e5a"

# 生成关键元素参考图
python3 /root/.openclaw/workspace/wanxiang_generate.py \
  "年轻白领男性,商务休闲装扮,微笑,正面照" \
  character_scene1.png

3. 文件处理工具

  • read - 读取用户上传的文档
  • web_fetch - 抓取 URL 内容
  • image - 分析参考图片

4. 视频合成工具

  • ffmpeg - 视频拼接、转场、配乐

📝 对话示例(v2.0 增强版)

首次使用配置

用户: 用豆包帮我生成一个产品宣传视频

助手: 好的!我来帮您创作专业的产品宣传视频。🎬

🔧 首次使用配置

检测到您是首次使用豆包视频创作功能,需要先配置 API 信息:

1️⃣ 请提供您的火山引擎豆包 API Key

2️⃣ 选择文生视频模型版本

  • doubao-seedance-2-0-pro(推荐,最新版本)
  • doubao-seedance-1-5-pro(稳定版本)
  • doubao-seedance-1-0-pro(基础版本)

3️⃣ 选择图生视频模型版本(如需要):

  • 同上,或保持与文生视频一致

配置将保存,后续使用无需重复输入。


用户: 我的 API Key 是 65ae8f92-134c-4194-a3af-6e6cb74284e0,文生视频用 2.0,图生视频用 1.5

助手: ✅ 配置已保存!

  • API Key: 65ae...84e0
  • 文生视频模型:doubao-seedance-2-0-pro
  • 图生视频模型:doubao-seedance-1-5-pro

现在开始创作您的视频吧!请告诉我:

  1. 视频主题:是什么产品/服务?
  2. 目标受众:给谁看的?
  3. 视频风格:科技感?温馨?专业?活泼?
  4. 参考资料(可选):文档、图片、URL 等

用户: [发送产品文档] 这是我们的智能手表产品介绍

助手: 收到!让我分析一下产品文档...

📊 产品分析完成

  • 产品名称:智能手表 X1
  • 核心功能:健康监测、运动追踪、智能通知
  • 目标用户:都市白领、运动爱好者
  • 卖点亮点:7 天续航、血氧监测、50 米防水

现在我来为您生成视频脚本规划(12 秒):

【视频脚本】智能手表 X1 宣传 总时长:12 秒 | 场景数:3 个

场景 1(0-4 秒):开场展示

  • 画面:手表特写,镜头从表盘缓缓上移
  • 文案:"智能生活,从腕间开始"
  • 风格:简约科技感

场景 2(4-8 秒):功能演示

  • 画面:用户运动时手表显示心率、步数
  • 文案:"全天候健康监测"
  • 风格:活力动感

场景 3(8-12 秒):品牌结尾

  • 画面:手表 + 品牌 Logo
  • 文案:"X1,你的智能伙伴"
  • 风格:专业大气

这个脚本可以吗?需要调整吗?😊


用户: 场景 2 改成展示睡眠监测

助手: 好的!已调整脚本:

场景 2(4-8 秒):睡眠监测

  • 画面:夜晚卧室,手表显示睡眠数据
  • 文案:"科学睡眠管理"
  • 风格:温馨安静

其他场景保持不变,确认开始生成关键元素参考图吗?🎨


用户: 确认

助手: 好的!开始生成关键元素参考图...

🎨 生成关键元素 1/3:主角人物

  • 描述:都市白领男性,30 岁左右,运动装扮 ⏳ 正在生成... ✅ 生成完成!

<qqimg>/path/to/character_main.png</qqimg>

这是主角人物形象,符合您的预期吗?


用户: 可以

助手: 🎉 所有参考图已确认!


分场景生成方式选择(新增!)

助手: 🎬 准备生成场景 1/3 (0-4 秒)

请选择生成方式

A) 文生视频 - 直接用文字提示词生成

  • 优点:快速,无需生成参考图
  • 适合:简单场景、抽象画面

B) 图生视频 - 先生成场景原型图,再基于图片生成视频

  • 优点:画面更可控,一致性更好
  • 适合:复杂场景、需要精确控制的画面

当前场景描述:手表特写,镜头从下往上扫过,现代科技感办公室背景

请回复 A 或 B 选择生成方式 📝


用户: B

助手: 好的!将使用图生视频模式生成场景 1。

🎨 生成场景原型图...

  • 描述:现代科技感办公室,白色办公桌,桌上放置智能手表 ⏳ 正在生成... ✅ 生成完成!

<qqimg>/path/to/scene1_prototype.png</qqimg>

这是场景 1 的原型图,确认用于视频生成吗?还是需要调整?


用户: 可以

助手: ✅ 场景 1 原型图已确认!

现在确认视频生成提示词:

【场景 1 提示词】(图生视频模式)

参考上传图片的场景,现代科技感办公室,白色办公桌,
桌上放置智能手表 X1,镜头从下往上缓缓移动,
聚焦手表表盘,自然光线从窗户射入,简约商务风格,
高清写实,电影级画质,4K 细节

这个提示词会生成:开场展示手表特写的 4 秒视频 确认开始生成吗?还是需要调整?📝


用户: 提示词可以,开始生成

助手: 好的!开始生成场景 1...

🎬 生成场景 1/3 (0-4 秒) - 图生视频模式 ⏳ 正在创建视频任务... ✅ 任务创建成功!Task ID: cgt-xxx ⏳ 等待视频生成完成...(约 1-2 分钟) ✅ 场景 1 生成完成!

<qqvideo>/path/to/scene1.mp4</qqvideo>

这是场景 1 的视频,效果满意吗?需要调整吗?


⚙️ 配置参数

首次使用配置(v2.0 新增)

全局配置文件: ~/.openclaw/workspace/doubao-config.json

{
  "default_api_key": "YOUR_API_KEY",
  "default_text_to_video_model": "doubao-seedance-2-0-pro",
  "default_image_to_video_model": "doubao-seedance-1-5-pro",
  "last_updated": "2026-03-31T12:00:00",
  "is_configured": true
}

配置项说明:

  • default_api_key: 火山引擎豆包 API Key
  • default_text_to_video_model: 文生视频默认模型
  • default_image_to_video_model: 图生视频默认模型

可用模型列表:

模型 ID版本推荐说明
doubao-seedance-2-0-pro2.0最新版本,效果最佳
doubao-seedance-1-5-pro1.5稳定版本,广泛使用
doubao-seedance-1-0-pro1.0基础版本,兼容性好

图片生成参数(通义万相)

  • 分辨率: 1024x1024
  • 风格: 写实/插画/3D(根据视频风格)
  • 数量: 每个元素生成 2-3 个版本供选择

视频生成参数

  • 时长: 每个场景 2-12 秒(推荐 4 秒)
  • 分辨率: 480p / 720p / 1080p(推荐 720p)
  • 比例: 16:9 / 9:16 / 1:1(推荐 16:9)
  • 帧率: 24 fps
  • 格式: MP4

生成方式选择

  • 文生视频: 直接使用文字提示词生成
  • 图生视频: 使用原型图 + 提示词生成(画面更可控)

📁 文件管理

项目目录结构

/root/.openclaw/workspace/doubao-video-projects/
└── video_20260331_120000/
    ├── project.json            # 项目配置(含 API 配置、模型选择、生成方式)
    ├── config.json             # 豆包 API 配置(首次使用保存)
    ├── references/             # 参考资料
    │   ├── doc_product.pdf
    │   └── url_content.txt
    ├── characters/             # 人物参考图
    │   ├── character_main_v1.png
    │   └── character_main_v2.png
    ├── scenes/                 # 场景参考图
    │   ├── scene_gym_v1.png
    │   └── scene_gym_bright.png
    ├── objects/                # 物品参考图
    │   └── watch_product.png
    ├── videos/                 # 场景视频
    │   ├── scene_1.mp4
    │   ├── scene_2.mp4
    │   └── scene_3.mp4
    └── final_video.mp4         # 最终合成视频

project.json 结构(完整项目记录)

{
  "project_id": "video_20260331_120000",
  "theme": "智能手表 X1 宣传",
  "created_at": "2026-03-31T12:00:00",
  "status": "collecting|planning|generating_elements|generating|confirming|completed",
  
  "api_config": {
    "api_key": "65ae8f92-134c-4194-a3af-6e6cb74284e0",
    "text_to_video_model": "doubao-seedance-2-0-pro",
    "image_to_video_model": "doubao-seedance-1-5-pro",
    "configured_at": "2026-03-31T12:00:00"
  },
  
  "references": {
    "documents": [],
    "urls": [],
    "images": [],
    "notes": ""
  },
  
  "script": {
    "total_duration": 12,
    "scene_count": 3,
    "scenes": [
      {
        "id": 1,
        "time": "0-4 秒",
        "title": "开场展示",
        "description": "手表特写,镜头从下往上扫过",
        "copy": "创新科技,改变生活",
        "style": "现代科技感",
        "generation_mode": "text_to_video|image_to_video",  // 生成方式
        "prototype_images": [],  // 图生视频时的原型图路径
        "prompt": "",
        "prompt_confirmed": false,
        "video_path": null,
        "status": "pending|generating|pending_confirmation|confirmed|rejected"
      }
    ]
  },
  
  "key_elements": {
    "characters": [
      {
        "id": 1,
        "description": "都市白领男性,30 岁左右",
        "image_path": "/path/to/character.png",
        "status": "pending|confirmed|rejected",
        "feedback": ""
      }
    ],
    "scenes": [],
    "objects": []
  },
  
  "final_video": null
}

全局配置文件

// ~/.openclaw/workspace/doubao-config.json
{
  "default_api_key": "65ae8f92-134c-4194-a3af-6e6cb74284e0",
  "default_text_to_video_model": "doubao-seedance-2-0-pro",
  "default_image_to_video_model": "doubao-seedance-1-5-pro",
  "last_updated": "2026-03-31T12:00:00"
}

⚠️ 注意事项

1. 首次配置

  • ✅ API Key 和模型选择只需配置一次
  • ✅ 配置保存在 doubao-config.json
  • ✅ 后续项目可覆盖全局配置(项目级配置)

2. 生成方式选择

  • ✅ 每个场景独立选择生成方式
  • ✅ 文生视频:快速,适合简单场景
  • ✅ 图生视频:可控,适合复杂场景
  • ✅ 选择结果保存在 project.json

3. 项目记录

  • ✅ 所有配置、状态、日志保存在 project.json
  • ✅ 支持中断后恢复
  • ✅ 完整追溯生成历史

4. 参考图的重要性

  • ✅ 确保人物形象一致性
  • ✅ 确保场景风格统一
  • ✅ 避免视频生成偏差
  • ✅ 减少重新生成成本

5. 时长限制

  • 豆包模型支持 2-12 秒
  • 建议每个场景 4 秒
  • 总场景数建议 3 个

6. 生成时间

  • 图片生成:约 30 秒/张
  • 视频生成:约 1-2 分钟/场景
  • 完整流程:约 10-15 分钟

7. 成本控制

  • 图片生成:约 ¥0.05-0.10/张
  • 视频生成:约 ¥0.15-0.50/场景
  • 先确认参考图再生成视频(避免浪费)

8. QQ Bot 限制

  • ⚠️ 媒体消息发送有 bug
  • ✅ 使用邮件发送大文件
  • ✅ 使用文件链接下载

🚀 扩展功能

未来可添加

  • 自动添加字幕
  • AI 配音
  • 多种分辨率选择
  • 批量生成不同版本
  • 视频模板库
  • 品牌风格预设
  • 角色一致性检查
  • 场景连续性优化

Skill 版本: v2.0
创建时间: 2026-03-31
更新时间: 2026-03-31(v2.0:新增首次配置、生成方式选择、完整项目记录)
作者: 卡妹 🌸


🔄 更新日志

v2.0 (2026-03-31)

新增功能:

  • ✅ 首次使用配置流程(API Key + 模型选择)
  • ✅ 分场景生成方式选择(文生视频/图生视频)
  • ✅ 完整项目记录文件(project.json)
  • ✅ 配置管理器(config_manager.py)
  • ✅ 项目状态追踪
  • ✅ 生成日志记录

改进:

  • ✅ 支持项目级 API 配置覆盖
  • ✅ 支持中断后恢复
  • ✅ 完整追溯生成历史
  • ✅ 优化对话流程

文件变更:

  • 新增:config_manager.py - 配置管理模块
  • 新增:project_template.json - 项目配置模板
  • 新增:FLOWCHART_v2.md - v2.0 完整流程图
  • 更新:doubao_video_creator.py - 支持图生视频模式
  • 更新:SKILL.md - 更新文档和示例

Comments

Loading comments...