podcast-recap

name	podcast-recap
description	Transcribe and summarize any podcast episode or YouTube video. Given a URL (YouTube, RSS feed, or direct audio), it fetches the transcript, extracts key takeaways with timestamps, and saves a structured markdown note.
version	1.0.0
metadata	{"openclaw":{"emoji":"🎙️","homepage":"https://github.com/openclaw/skills","requires":{"bins":["yt-dlp","ffmpeg"]},"anyBins":["whisper","python3"],"install":[{"kind":"brew","formula":"yt-dlp","bins":["yt-dlp"]},{"kind":"brew","formula":"ffmpeg","bins":["ffmpeg"]}],"envVars":[{"name":"PODCAST_NOTES_DIR","required":false,"description":"Directory where recap notes are saved. Defaults to ~/podcast-notes."}]}}

podcast-recap

Summarize a podcast episode or YouTube video from a URL. Produces a structured markdown note with key takeaways, chapter summaries, and notable quotes — saved locally for future reference.

---

Trigger

The user sends a message containing a URL and a request to recap or summarize it. Examples:

• "recap this podcast: https://..."
• "summarize this episode: https://..."
• "podcast-recap https://..."
• "take notes on https://..."

---

Execution Steps

Step 1 — Detect Input Type

Inspect the URL to determine its type:

• YouTube: URL contains youtube.com/watch, youtu.be/, or youtube.com/shorts/
• RSS feed: URL ends in .xml, .rss, contains /feed, or returns application/rss+xml or application/atom+xml on HEAD request
• Direct audio: URL ends in .mp3, .m4a, .ogg, .wav, .opus
• Podcast episode page: Any other URL — try to find a media file link embedded in the page

---

Step 2 — Get the Transcript

2A: YouTube URL

Use yt-dlp to extract auto-generated or manual captions:

yt-dlp \
  --skip-download \
  --write-auto-sub \
  --write-sub \
  --sub-lang en \
  --sub-format vtt \
  --output "/tmp/podcast-recap/%(title)s.%(ext)s" \
  "<URL>"

If captions are unavailable, fall back to downloading the audio and running Whisper (Step 2C).

Parse the .vtt file into plain text with timestamps:

# Strip VTT metadata and de-duplicate overlapping lines
grep -v '^WEBVTT\|^NOTE\|^[0-9].*-->' /tmp/podcast-recap/*.vtt \
  | sed '/^$/d' \
  | awk '!seen[$0]++' \
  > /tmp/podcast-recap/transcript.txt

Also capture episode metadata (title, channel, duration, upload date):

yt-dlp --dump-json --no-download "<URL>" > /tmp/podcast-recap/meta.json

Extract with jq:

jq -r '{title, uploader, upload_date, duration, description}' /tmp/podcast-recap/meta.json

2B: RSS Feed URL

1. Fetch the feed XML:

   curl -sL "<URL>" -o /tmp/podcast-recap/feed.xml
   

2. Extract the latest episode's enclosure URL (the audio file):

   grep -oP '(?<=url=")[^"]+\.(mp3|m4a|ogg|opus)' /tmp/podcast-recap/feed.xml | head -1
   

   Or, if the user specified an episode number/title, find the matching <item> block.

3. Download the audio and proceed to Step 2C.

2C: Direct Audio or Audio Fallback

Download the audio file:

curl -sL "<AUDIO_URL>" -o /tmp/podcast-recap/episode.mp3

Transcribe with Whisper (if installed):

whisper /tmp/podcast-recap/episode.mp3 \
  --model small \
  --output_format txt \
  --output_dir /tmp/podcast-recap/

If Whisper is not installed, instruct the user:

"Whisper is not installed. Install it with: pip install openai-whisper. I can still summarize from the episode description if you'd like." Then proceed using only the episode title and description.

---

Step 3 — Summarize

Read the full transcript from /tmp/podcast-recap/transcript.txt.

Apply the summarization template in prompts/summarize.md.

Generate the following sections:

1. TL;DR (2–3 sentences): The single most important insight from the episode.
2. Key Takeaways (5–8 bullets): Distilled, actionable or notable points. Use the speaker's own words where impactful.
3. Chapter Summaries: If the episode has chapters (from YouTube or timestamps in the transcript), produce a one-paragraph summary per chapter. Otherwise, divide into ~4 logical sections.
4. Notable Quotes: 2–4 verbatim quotes worth remembering, each with a brief note on why it matters.
5. Topics & Tags: 5–10 keywords or topics covered (for searchability).
6. My Take (optional): If the user has asked for opinion or framing, add a brief editorial note.

---

Step 4 — Resolve Output Path

Determine the output directory:

NOTES_DIR="${PODCAST_NOTES_DIR:-$HOME/podcast-notes}"
mkdir -p "$NOTES_DIR"

Generate a filename from the episode title:

# Sanitize title: lowercase, replace spaces/special chars with hyphens
TITLE_SLUG=$(echo "<EPISODE_TITLE>" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | sed 's/--*/-/g' | sed 's/^-\|-$//g')
DATE=$(date +%Y-%m-%d)
OUTPUT_FILE="$NOTES_DIR/${DATE}-${TITLE_SLUG}.md"

---

Step 5 — Write the Note

Write the structured markdown note to $OUTPUT_FILE using the template below.

# <Episode Title>

> **Source**: <URL>
> **Published**: <Upload date or feed date>
> **Show / Channel**: <Podcast name or YouTube channel>
> **Duration**: <HH:MM:SS>
> **Recap generated**: <today's date>

---

## TL;DR

<2–3 sentence summary>

---

## Key Takeaways

- <takeaway 1>
- <takeaway 2>
- <takeaway 3>
- <takeaway 4>
- <takeaway 5>

---

## Chapter Summaries

### <Chapter 1 Title or "Part 1">
<paragraph>

### <Chapter 2 Title or "Part 2">
<paragraph>

...

---

## Notable Quotes

> "<quote>"
— *<Speaker name or "Host">*, <approximate timestamp>

> "<quote>"
— *<Speaker name>*, <approximate timestamp>

---

## Topics & Tags

`<tag1>` `<tag2>` `<tag3>` `<tag4>` `<tag5>`

---

*Generated by podcast-recap skill*

---

Step 6 — Respond to the User

Send a message confirming the recap was saved:

🎙️ Podcast recap saved!

"" Show: <Podcast / Channel> · Duration:

TL;DR: <the TL;DR text>

Saved to: <OUTPUT_FILE>

Want me to read out the key takeaways or dive into any section?

---

Error Handling

Situation	Response
yt-dlp not installed	Tell the user: "Run brew install yt-dlp to enable YouTube support."
No captions available and Whisper not installed	Summarize from title + description only; flag the limitation
RSS feed has no enclosure URL	Ask the user to provide the direct audio URL
Transcript is very short (<200 words)	Warn the user: "Transcript appears incomplete — summary may be limited."
Network error fetching URL	Report the error and suggest the user check the URL
Output directory not writable	Fall back to ~/Downloads/podcast-notes/ and inform the user

---

Notes

• Temporary files are written to /tmp/podcast-recap/ and can be cleaned up after the note is saved with rm -rf /tmp/podcast-recap/.
• The skill works best with YouTube videos that have auto-captions enabled. For RSS feeds, Whisper provides high-quality transcripts for audio under ~60 minutes using the small model without GPU.
• If the episode is longer than 2 hours, use --model tiny for Whisper to keep transcription time reasonable.
• The PODCAST_NOTES_DIR env var allows routing notes to an Obsidian vault or any synced folder.