CreateVideo - Podcast to Video

视频生成工具。当用户说"CreateVideo"、"创建视频"、"生成视频"或提供文案要求制作播客视频时触发。支持双人播客音频生成（通过 ListenHub MCP）、模版视频裁剪合并、内容分析输出。依赖 ffmpeg 和 ListenHub MCP Server。

Security Scan

The skill's behavior generally matches a podcast-to-video generator, but the instructions request external installs (npx) and an API key and assume ffmpeg/ffprobe and writable /root paths while the declared metadata lists no required env vars or binaries — these mismatches are concerning.

Purpose & Capability

The skill claims to generate videos from podcast text and to use ListenHub MCP + ffmpeg, which is coherent with its purpose. However the registry metadata declares no required environment variables or binaries, while the SKILL.md explicitly depends on LISTENHUB_API_KEY and ffmpeg/ffprobe and on template files under /root/.openclaw/workspace/video-projects. The undeclared dependencies and hardcoded root workspace path are inconsistent with the stated requirements.

Instruction Scope

Runtime instructions tell the agent to create directories and copy template files under a hardcoded /root path, call ffprobe/ffmpeg, and run an external MCP workflow (create_podcast) that downloads audio. The workflow also instructs running a command that installs an npm package via npx. These steps stay within the skill's stated purpose, but they perform filesystem writes in /root and execute external install commands — operations that exceed a simple, read-only instruction and should have been declared.

Install Mechanism

There is no formal install spec in the registry, yet SKILL.md instructs running: claude mcp add listenhub --env LISTENHUB_API_KEY=... -- npx -y @marswave/listenhub-mcp-server. That invokes npx to fetch and run code from npm at runtime. Using npx to execute a package is effectively downloading/executing remote code and should be declared explicitly; the lack of an install spec is a mismatch and increases risk.

Credentials

The documentation and references require a LISTENHUB_API_KEY and speaker IDs (YOUR_SPEAKER_ID_1/2) but the skill metadata lists no required env vars or primary credential. Requesting an external API key to use ListenHub is proportional to the feature, but failing to declare it is an important omission and a transparency problem. No other unrelated credentials are requested.

Persistence & Privilege

The skill is not marked always:true and does not claim elevated platform privileges. It will create files under /root/.openclaw/workspace/video-projects and instruct installing an MCP server (which may persist components). This is expected for a local video generation workflow but the hardcoded /root path and implicit installation mean it may require elevated write access; consider whether the path should be configurable to a non-root location.

What to consider before installing

This skill appears to do what it says (generate podcast-style audio via ListenHub and combine it with templates using ffmpeg), but there are several mismatches you should resolve before installing: 1) The SKILL.md requires LISTENHUB_API_KEY and speaker IDs but the skill metadata does not declare any env vars — expect to provide that API key. 2) The instructions call ffmpeg/ffprobe and execute npx to run @marswave/listenhub-mcp-server; these actions download and run remote code and need your explicit approval. 3) The default BASE_DIR is hardcoded to /root/.openclaw/... which requires write access and may be inappropriate; change it to a safe, user-writable directory. Recommendations: verify the ListenHub package and its npm publisher (@marswave), obtain and store LISTENHUB_API_KEY securely, ensure ffmpeg/ffprobe are installed from trusted sources, and run the skill in an isolated or non-root workspace. If you cannot confirm the npm package or don't want remote installs/executables, decline or request a version that supplies an explicit install spec and properly declares required env vars and binaries.

CreateVideo Skill

工作流程

1. 接收输入

用户提供：

• 必须：视频文案内容
• 可选：模版选择（默认根据文案内容自动选择）
  • 书房 / 夜景 → $BASE_DIR/v_template.mp4（上海夜景书房，29.6s）
  • 能量 / 能量人 → $BASE_DIR/v_template_energy.mov（能量人，14.7s）

2. 创建任务目录

DATE=$(date +%Y%m%d)
BASE_DIR="/root/.openclaw/workspace/video-projects"
TASK_DIR="$BASE_DIR/${DATE}_001"

COUNTER=1
while [ -d "$TASK_DIR" ]; do
  COUNTER=$((COUNTER + 1))
  TASK_DIR="$BASE_DIR/${DATE}_$(printf "%03d" $COUNTER)"
done

mkdir -p "$TASK_DIR/temp"

3. 复制模板视频

根据文案内容或用户指定选择模版：

# 自动选择逻辑：能量/潜意识/吸引力类 → 能量人模版，其他 → 书房夜景模版
# 用户也可以直接指定："用能量人模版" / "用书房模版"
cp "$BASE_DIR/v_template.mp4" "$TASK_DIR/template.mp4"        # 书房夜景（默认）
# 或
cp "$BASE_DIR/v_template_energy.mov" "$TASK_DIR/template.mp4"  # 能量人

4. 生成双人播客音频（ListenHub MCP）

通过 ListenHub MCP 生成双人播客音频：

1. 调用 create_podcast：
   • query: 用户文案内容
   • speakerIds: ["YOUR_SPEAKER_ID_1", "YOUR_SPEAKER_ID_2"]（主播A先说，主播B后说）
   • language: "zh"
   • mode: "quick"
2. 等待生成完成，下载音频到 $TASK_DIR/podcast_audio.mp3

详细 MCP 参数参见 references/listenhub-guide.md

5. 无缝循环模版视频（stream_loop）

使用 ffmpeg 的 stream_loop 实现无缝循环，一步完成循环+音频合并+输出：

⚠️ 不要用 concat demuxer + 正放↔倒放方案，会在接缝处冻帧（2026-03-22 踩坑验证）

DURATION=$(ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$TASK_DIR/podcast_audio.mp3")
TEMPLATE_DUR=$(ffprobe -v error -show_entries format=duration -of default=noprint_wrappers=1:nokey=1 "$TASK_DIR/template.mp4")
LOOPS=$(python3 -c "import math; print(math.ceil($DURATION / $TEMPLATE_DUR))")

TITLE="根据文案内容生成的标题"

ffmpeg -y -stream_loop $LOOPS -i "$TASK_DIR/template.mp4" \
  -i "$TASK_DIR/podcast_audio.mp3" \
  -t $DURATION \
  -c:v libx264 -preset fast -b:v 1400k \
  -c:a aac -map 0:v:0 -map 1:a:0 \
  "$TASK_DIR/${TITLE}.mp4"

说明：

• stream_loop 直接循环原始模版，无接缝问题
• -b:v 1400k 控制输出大小（720×1280 竖屏约 40MB/4min）
• 一条命令同时完成：循环视频 + 合并音频 + 裁剪时长 + 输出最终文件
• 不再需要中间文件（trimmed.mp4、merged.mp4、template_reversed.mp4）

6. 生成内容分析

基于原始文案和播客脚本，生成并保存到 $TASK_DIR/内容分析.txt：

1. 核心金句 — 最具传播力的一句话
2. 洞见 — 深层洞察和价值点
3. 内容描述 — 50字概括视频核心内容
4. 朋友圈文案 — 200字推广文案

7. 输出结果格式

**视频创建完成！**

**任务目录**: {路径}
**最终视频**: {文件名}

**播客信息**:
- 标题: {标题}
- 音色: 主播A + 主播B
- 时长: {时长}

---

**内容分析**:

核心金句：...
洞见：...
内容描述（50字）：...
朋友圈文案（200字）：...

任务目录结构

video-projects/20260322_001/
├── template.mp4          # 复制的模版视频
├── temp/                 # 临时中间文件（可清理）
├── podcast_audio.mp3     # 播客音频
├── {标题}.mp4            # 最终视频
└── 内容分析.txt          # 内容分析

默认配置

配置项	默认值
基础目录	/root/.openclaw/workspace/video-projects
模版视频A	v_template.mp4（720×1280，29.6s，上海夜景书房）— 认知/读书/财富类
模版视频B	v_template_energy.mov（720×1280，14.7s，能量人）— 能量/潜意识/吸引力类
播客音色1（先说）	主播A（YOUR_SPEAKER_ID_1，在 ListenHub 获取）
播客音色2（后说）	主播B（YOUR_SPEAKER_ID_2，在 ListenHub 获取）
播客模式	quick（3-5分钟）
播客语言	zh（中文）

依赖

• ffmpeg / ffprobe — 视频处理
• ListenHub MCP Server — @marswave/listenhub-mcp-server（需要 LISTENHUB_API_KEY）

安装配置

1. 安装 ListenHub MCP

claude mcp add listenhub --env LISTENHUB_API_KEY=<你的API Key> -- npx -y @marswave/listenhub-mcp-server

API Key 在 ListenHub 官网 注册后获取。

2. 配置音色 ID

在使用前，先调用 get_speakers 查询可用音色，将你选择的音色 ID 填入 Skill 配置：

• YOUR_SPEAKER_ID_1 — 主播A（先说）
• YOUR_SPEAKER_ID_2 — 主播B（后说）

3. 准备模版视频

将你的背景视频放入 video-projects/ 目录：

• v_template.mp4 — 默认模版（720×1280 竖屏）
• v_template_energy.mov — 备选模版（可选）

注意事项

1. 播客音频生成需要几分钟，耐心等待
2. 每次任务自动创建日期序号目录（如 20260321_001）
3. 中间文件在 temp/ 子目录，可手动清理
4. 模板文件复制到任务目录，不影响原始文件