Run this task with `hai` (https://github.com/braincore/hai-cli).
Install `hai` from your terminal: curl -LsSf https://hai.superego.ai/hai-installer.sh | sh
Run this task from your terminal: hai task ken/youtube

---

name = "ken/youtube"
version = "1.1.0"

description = "Extracts YouTube transcripts (subtitles) given a YouTube video URL."

steps = [
    """/pin This task extracts YouTube transcripts (subtitles) for a given video URL using yt-dlp.

Usage: yt-dlp [OPTIONS] URL [URL...]

Always use:

--skip-download — This disables downloading the video since it's unnecessary.
--write-auto-subs — Can use auto-generated subs
--sub-format srt — Always use SRT
--sub-lang {LANG} — The language to use
-o {VIDEO_ID} — Specify output file

When the user requests a transcript with the !hai tool, return the following 3 subcommands:

1. Run yt-dlp using uv:
  - /exec uv run --with yt-dlp==2025.06.09 yt-dlp ...
2. /f0 "<output_file>"
  - Must be quoted since the input is a Python object declaration!
3. Remove the output file.

If uv doesn't exist, warn user and refuse commands.
""",
    """!fn-py(cache=true) Define a function that takes a file path to an SRT.
- Read the SRT file
- Print lines in transcript to stdout but OMIT sequence numbers and blank lines.
  - If a line begins with a capital letter, assume it's a new sentence and
    add a blank line before it to help break up large blocks.
  - If a line doesn't begin with a capital letter, assume it's part of the previous
    line and separate the two with just a space rather than a newline!
  - Only show timecodes every minute.
    - Surround timecode with double brackets [[TIME CODE]].
    - Add two newlines before each timecode.
- Return nothing
""",
    "/ask-human(cache=true) What language (`en` is default)?",
    "/pin uv is required, checking...",
    "/exec which uv",
    "/pin Which video?",
    "!hai",
]