Volcano Engine Podcast

生成火山引擎豆包语音播客（PodcastTTS）。输入主题文本，自动生成双人对话式播客音频。触发关键词：豆包语音播客、生成播客、语音播客。

Security Scan

The skill's code and instructions match its stated purpose (generating PodcastTTS audio via a Volcano/Bytedance websocket API); the footprint is coherent and self-contained, with only minor configuration-name mismatches to be aware of.

Purpose & Capability

Name/description (generate double-person PodcastTTS audio) align with bundled Python scripts, which implement a WebSocket client to a Volcano/Bytedance PodcastTTS endpoint and helpers to save/copy MP3s. Required binary (python3) and websockets package are appropriate.

Instruction Scope

SKILL.md explicitly limits flow (ask for reference material, generate audio, copy final MP3 to a messaging send directory). It references reading ~/.openclaw/config.json and copying to /root/.openclaw/media/qqbot/downloads — these are specific paths the skill will read/write and are necessary for the stated behavior, but they are filesystem locations outside the skill directory that the agent must be permitted to access.

Install Mechanism

No install spec; this is instruction+script-only. It requires the websockets package (declared in metadata). No remote downloads or archive extraction are used.

Credentials

Declared requirements list no environment variables, but code will read ~/.openclaw/config.json for volc_podcast.appid and volc_podcast.access_key and falls back to VOLC_APPID / VOLC_ACCESS_TOKEN environment variables. This is proportionate (API keys needed to call the service) but the metadata/instructions should declare these env vars and the exact config keys. Also the config JSON example includes secret_key which the code does not use; there is a hardcoded app_key value in code (likely a client id), which is unusual but not obviously malicious.

Persistence & Privilege

Skill is not always-enabled and does not request elevated platform privileges. It writes files to specified local paths for delivery (its own output and copying into a send directory) but does not modify other skills or global agent configuration.

Assessment

This skill appears to do what it says: it connects to a Volcano/Bytedance PodcastTTS websocket endpoint to generate MP3s and then copies the file into /root/.openclaw/media/qqbot/downloads so the agent can send it. Before installing, confirm: (1) you trust the skill source (GitHub homepage) and the openspeech.bytedance.com endpoint, (2) you are comfortable storing your Volcano API keys in ~/.openclaw/config.json or in VOLC_APPID / VOLC_ACCESS_TOKEN env vars, and (3) the agent has permission to write to /root/.openclaw/media/qqbot/downloads. Note the minor mismatches: the skill metadata declares no required env vars but the code does read env vars/config; the config example contains a secret_key field the code doesn't use, and there is a hardcoded app_key in the code. These are quality/documentation issues rather than clear red flags, but you may want the author to clarify the exact config/env var names and remove/justify the hardcoded app_key before deploying in a sensitive environment.

火山引擎豆包语音播客生成

基于火山引擎 PodcastTTS API，输入主题文本，AI 自动生成双人对话播客音频（含片头音乐、多轮对话、片尾结束）。

---

🎯 触发条件

用户提到以下关键词时触发：

• "用豆包语音播客帮我生成..."
• "生成语音播客"
• "豆包播客"
• "PodcastTTS"

---

🔄 收到请求后流程

⚠️ 重要：收到用户请求后，必须先询问是否有参考资料！

步骤 1：询问参考资料

用户：用豆包语音播客帮我生成XX主题的语音播客

助手回复：
收到！请选择资料来源：

1️⃣ **有参考资料**
   - 请发送相关文档、链接或已有内容
   - 我会基于资料生成播客

2️⃣ **没有参考资料**
   - 直接按你提供的主题生成播客
   - AI 会自动扩展内容

请回复 "1" 或 "2"，或直接发送资料（文档/链接）

步骤 2：生成播客

• 有资料：先读取资料内容，整合后生成播客
• 无资料：直接使用主题文本生成播客

步骤 3：发送音频

生成完成后，必须把 MP3 文件发送给用户：

<qqvoice>/root/.openclaw/media/qqbot/downloads/播客名称.mp3</qqvoice>

---

📋 配置

火山引擎播客 API 配置：

// ~/.openclaw/config.json
{
  "volc_podcast": {
    "appid": "3398567544",
    "access_key": "your_access_key",
    "secret_key": "your_secret_key"
  }
}

---

🔧 调用方式

快速调用（推荐）

import asyncio
import sys
sys.path.insert(0, "/root/.openclaw/workspace/skills/volcano-engine-podcast/scripts")
from generate_podcast import PodcastGenerator

async def generate_podcast(text, output_name="podcast"):
    gen = PodcastGenerator(
        appid="3398567544",
        access_token="your_token",
    )
    
    result = await gen.generate(
        text=text,
        output_dir=f"/tmp/{output_name}",
        encoding="mp3",
        use_head_music=True,
    )
    
    if result["success"]:
        # 复制到发送目录
        import shutil
        src = result["final_files"][0]
        dst = f"/root/.openclaw/media/qqbot/downloads/{output_name}.mp3"
        shutil.copy(src, dst)
        return dst
    
    return None

# 使用
audio_path = await generate_podcast("今天来聊聊AI编程助手", "AI编程助手播客")

---

📊 返回结果

{
    "success": True,
    "duration": 135.99,           # 播客时长（秒）
    "final_files": ["...mp3"],    # 最终音频文件
    "texts": [...],               # 对话文本列表
    "usage": {                    # Token 消耗
        "input_text_tokens": 1245,
        "output_audio_tokens": 3217,
        "total_tokens": 4462
    }
}

---

⚙️ 参数说明

参数	默认	说明
text	必填	输入主题文本
encoding	mp3	音频格式
use_head_music	True	片头音乐
use_tail_music	False	片尾音乐

---

📝 注意事项

1. 每次调用消耗 audio token
2. 对话角色由 AI 自动分配（男女双人对话）
3. 音频采样率 24kHz
4. 生成后必须发送 MP3 给用户