# Youtube Transcript

> Use when: YouTube/podcast transcript via yt-dlp for cite or corpus. Transcript-only — NOT video analysis.

- Skill: `markusstrasser/youtube-transcript` (Agent Skill)
- Install (CLI): `npx skillmds@latest add markusstrasser/youtube-transcript`
- Raw SKILL.md: https://api.skillmd.com/api/skills/markusstrasser/youtube-transcript/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: markusstrasser (https://skillmd.com/u/markusstrasser)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/markusstrasser/youtube-transcript

---


# YouTube Transcript

Lightweight wrapper around `yt-dlp` for transcript-only fetching.

## Install check

```bash
command -v yt-dlp || uvx yt-dlp --version  # uvx works without install
```

## Fetch transcript (no video)

```bash
# Replace VID with the 11-char YouTube ID or full URL
VID="Pxyl2O_AxPk"
OUT_DIR="sources/<topic>/transcripts"
mkdir -p "$OUT_DIR"

uvx yt-dlp \
  --skip-download \
  --write-auto-sub --write-sub \
  --sub-lang "en.*" \
  --sub-format "vtt/srv1/best" \
  --convert-subs vtt \
  -o "${OUT_DIR}/%(id)s-%(title)s.%(ext)s" \
  "https://www.youtube.com/watch?v=${VID}"
```

## VTT → plain text

VTT has timestamps and cue tags that make grep noisy. Strip them:

```bash
# Simple: drop all lines that look like timestamps, cues, or metadata
awk '
  /^WEBVTT/ { next }
  /^NOTE/ { next }
  /^[0-9]+$/ { next }
  /-->/ { next }
  /^$/ { next }
  { gsub(/<[^>]*>/, ""); print }
' "${OUT_DIR}/${VID}-*.en.vtt" | awk '!seen[$0]++' > "${OUT_DIR}/${VID}.txt"
```

The `!seen[$0]++` dedupes the rolling-captions pattern (same line repeated with growing cue).

## Citation convention

After staging, cite as:
```
[SOURCE: youtube.com/watch?v=<VID>, transcript at sources/<topic>/transcripts/<VID>.txt]
```

Include approximate timestamp in-quote if the claim is time-specific:
```
At ~01:23:45, [speaker] claims X [SOURCE: <VID> transcript, line N]
```

## Known failures

- **Auto-captions unavailable** — some channels disable them. yt-dlp errors out; fall back to manual transcript request or Whisper (not covered here).
- **Age-restricted / members-only** — requires `--cookies-from-browser chrome`.
- **Live streams** — transcript not finalized until stream ends.
- **Very long videos (>6h)** — .vtt can be 2-5MB. Strip before committing to git.

## Evidence

- `sources/jre-2460-rachel-wilson-transcript.txt` + `.en.vtt` — one-off, no reusable pattern.
- Future: any podcast/video citation in research memos.