SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsread-podcast-workflow

read-podcast-workflow

Writing & research

View repository

Updated 10-16 08:37

About this skill

Podcast processing workflow

A one-stop podcast processing flow from YouTube to Feishu publishing.

Two entry points

Entry A: Fetch podcast updates (recommended)

When the user says "fetch podcast updates" or "any new podcasts":

获取更新 → 展示列表 → 用户选择 → 字幕提取 → [询问] → Content-Digest → 飞书

Entry B: Directly process a link

When the user provides a YouTube link:

YouTube 链接 → 字幕提取 → [询问] → Content-Digest → 飞书

Entry A: Fetch podcast updates flow

Step A1: Fetch the update list

Run the command:

python3 /Users/ugreen/.claude/skills/read-youtube-feed/scripts/get_updates.py --days 2

Output format:

📺 最近 2 天共有 N 个新播客更新：

1. 【Lenny's Podcast】
   📌 视频标题
   🕐 2026-01-30 15:00
   🔗 https://www.youtube.com/watch?v=xxx
   📝 简要描述...

2. 【No Priors】
   ...

Step A2: User selection

After displaying the list, ask:

请选择要处理的播客（输入序号，如 1 或 1,3），或输入 "跳过"：

After waiting for the user's selection, proceed to Step 1.

Entry B: Direct processing flow

Step 1: Extract YouTube subtitles

Run the command:

python3 /Users/ugreen/.claude/skills/read-youtube-transcript/scripts/get_transcript.py "YOUTUBE_URL"

Output: Subtitle text in Markdown format

Step 2: Ask whether to process

After subtitles are extracted successfully,you must ask the user：

字幕提取成功！是否使用 read-content-digest 处理成精华内容？
- 处理后会生成：标题 + 核心观点列表 + 精华片段
- 处理完成后自动保存到飞书「每日播客推荐」

Wait for the user's confirmation before continuing。

Step 3: Content-Digest processing

Refer to /Users/ugreen/.claude/skills/read-content-digest/SKILL.md for the complete flow.

Key format requirements (v2 — updated 2026-05-12):

Each podcast summary uses the following unified format,all fields must be filled in：

文章标题：MMDD-文章中文一句话主题（如 0512-如何搭建个人 AI OS）
嘉宾：[嘉宾名]（Twitter @handle）
频道：[频道名]
日期：YYYY-MM-DD
主题：[一句话主题描述，30 字内]
播客地址：[YouTube 完整 URL]

# 核心观点

1、[观点]。[完整的逻辑阐述，包含推理过程，2-4 句话]

2、[观点]。[完整的逻辑阐述，包含推理过程，2-4 句话]

...（10-15 条）

# 精彩片段

[使用 Style B 对话式访谈格式，挑选 3-5 个最精彩的片段，保留嘉宾原话翻译]

Field descriptions:

  • 文章标题: Used as the Feishu document title and local filename prefix, format MMDD-中文一句话主题
  • 嘉宾: If a Twitter account can be found, include it; if not, just write the name
  • 日期: The video's publication date on YouTube, not the processing date
  • 主题: A one-sentence description that lets people decide whether to watch

Step 4: Save locally

Save location: /Users/ugreen/Documents/obsidian/每日播客/

File naming: MMDD-嘉宾名-主题关键词.md

Example:0130-Peter-Steinberger-AI编程革命.md

Step 5: Automatically save to Feishu

Run the command:

python3 /Users/ugreen/.claude/skills/feishu-wiki/scripts/save_to_wiki.py \
  --file "/Users/ugreen/Documents/obsidian/每日播客/MMDD-xxx.md" \
  --title "MMDD：嘉宾名 X 栏目名：一句话观点" \
  --parent "TOSJwKzxTiFdiRk0aducHNBFntg"

Parent node Token: TOSJwKzxTiFdiRk0aducHNBFntg(Daily podcast recommendations)

Step 6: Return the result and ask about images

After processing is complete, show the user:

✅ 播客处理完成！

📝 本地文件：/Users/ugreen/Documents/obsidian/每日播客/MMDD-xxx.md
📤 飞书文档：https://my.feishu.cn/wiki/xxx

核心观点数：N 条
精华片段数：N 个

---

是否要将核心观点转为图片海报？（适合小红书/朋友圈分享）

Step 7: Convert to an image (optional)

If the user confirms they want to convert to an image, call markdown-to-image Skill：

  1. Extract a short-text version: Keep only the title + list of core points (excluding highlight clips)
  2. Optimize the format：
    • Add emoji numbering (1️⃣ 2️⃣ 3️⃣)
    • Condense each point to 1-2 sentences
    • Keep the total word count within 500 words
  3. Use the browser MCP to operate Madopic：
    • Open https://madopic.thus.chat
    • Paste the optimized Markdown
    • Choose a mode (Xiaohongshu 3:4 or Moments long image)
    • Export PNG
  4. Tell the user the image has been downloaded to the downloads directory

Quick usage example

The user says:

Process this podcast https://www.youtube.com/watch?v=xxxxx

Agent executes:

  1. Extract subtitles ✓
  2. Ask the user → user confirms
  3. Content-Digest processing ✓
  4. Save locally ✓
  5. Save to Feishu ✓
  6. Return result

Error handling

Error Handling method
Subtitle extraction failed Inform the user that the video may not have subtitles and ask whether to provide the text manually
Feishu write failed Check permissions and prompt the user to add the app as a knowledge base collaborator
Network timeout Retry once; if it still fails, inform the user

Followed YouTube channels

Data source:Zara's AI Learning Library

Category Channel
AI education Andrej Karpathy, Anthropic, Lex Fridman
AI products Lenny's Podcast, Peter Yang, The MAD Podcast, Every
VC investment Y Combinator, Latent Space, South Park Commons, No Priors, a16z
Big tech research Google DeepMind, Google for Developers, Stanford GSB
Vibe Coding Mckay Wrigley, Tiago Forte, The Pragmatic Engineer
AI news The AI Daily Brief, TBPN, Brett Malinowski

21 channels in total

Add a new channel: edit /Users/ugreen/.claude/skills/read-youtube-feed/scripts/get_updates.py

Dependent Skills

  • read-youtube-feed: Get creator updates
  • read-youtube-transcript: Subtitle extraction
  • read-content-digest: Content processing
  • feishu-wiki: Feishu write
  • markdown-to-image: Convert to image poster (optional)

Configuration

Configuration item Value
Local save path /Users/ugreen/Documents/obsidian/每日播客/
Feishu parent node TOSJwKzxTiFdiRk0aducHNBFntg(Daily podcast recommendations)
Number of viewpoints 10-15 items
Update check days 2 days