git:20260702.ecf453b to git:20260705.380563c

---

name: video-transcript

description: "Extract video transcripts: yt-dlp subtitles to clean paragraphs."

user_invocable: false # default -- router-dispatched, not user-typed

agent: python-general-engineer

allowed-tools:

- Bash

- Read

routing:

triggers:

- "video transcript"

- "youtube transcript"

- "extract transcript"

- "download subtitles"

- "what does this video say"

category: research

pairs_with:

- research-pipeline

---

# Video Transcript

Pull a video's transcript as readable paragraphs. Two paths, in order:

**Path 1 — uploader subtitles** (accurate, prefer when present):

```bash

yt-dlp --skip-download --write-subs --sub-langs en --sub-format vtt \

-o '<work-dir>/%(id)s' '<URL>'

```

**Path 2 — auto-generated captions** (fallback when path 1 writes no file):

```bash

yt-dlp --skip-download --write-auto-subs --sub-langs en --sub-format vtt \

-o '<work-dir>/%(id)s' '<URL>'

```

Then clean the VTT into paragraphs:

```bash

python3 skills/research/video-transcript/scripts/vtt_to_paragraph.py <work-dir>/<id>.en.vtt

```

Default output is plain paragraph text with `[Music]`-style cues stripped and the rolling duplicates of auto-captions deduplicated. Use `--timestamps` for `[mm:ss]` markers, `--keep-brackets` to keep cue tags, `-o FILE` to write to a file.

For other languages, change `--sub-langs` (e.g. `de`, `en.*`). List what a video offers with `yt-dlp --list-subs '<URL>'`.

## Error handling

### Both paths write no .vtt file

Cause: video has no subtitles or captions in the requested language.

Solution: run `yt-dlp --list-subs '<URL>'` and pick an available language; if none exist, report that and offer audio transcription via the `markdown-converter` skill on a downloaded audio file.

### HTTP 429 / "Sign in to confirm"

Cause: platform rate-limiting the host.

Solution: wait and retry with `--sleep-requests 2`; keep request volume low.

### Cleaner output repeats lines

Cause: VTT came from a third path (e.g. translated captions) with cue formats the dedupe misses.

Solution: rerun the cleaner; if repeats remain, file the sample VTT alongside a fix to `vtt_to_paragraph.py`.