reader-youtube

当用户提供了 YouTube 的 URL（youtube.com、youtu.be、Shorts、嵌入式视频）或请求获取 YouTube 视频的字幕、中文字幕或英文字幕时，可使用此功能。此外，当出现以下情况时也应触发该功能：(1) 用户分享 YouTube 链接并想了解视频内容；(2) 用户说“看看这个视频”“这个视频里有什么”“把字幕提取出来”；(3) 需要对 YouTube 视频进行分析或摘要总结。系统会提取视频字幕文本，并同步获取视频元数据（标题、频道）。

SKILL.md

--- frontmatter

name: reader-youtube
description: "Use when the user provides a YouTube URL (youtube.com, youtu.be, shorts, embed) or asks to get a transcript, subtitles, or captions from a YouTube video. Also trigger when: (1) user shares a YouTube link and wants to know what's in it, (2) user says 'посмотри видео', 'что в этом видео', 'достань транскрипт', (3) a YouTube video needs to be analyzed or summarized. Extracts transcript text and video metadata (title, channel)."

YouTube Transcript

Extract transcript and metadata from YouTube videos using youtube-transcript-api.

Usage

bash

uv run scripts/get_transcript.py "VIDEO_URL_OR_ID"
uv run scripts/get_transcript.py "VIDEO_URL_OR_ID" --timestamps
uv run scripts/get_transcript.py "VIDEO_URL_OR_ID" --lang en,ru

Parameters

Flag	Default	What it does
`--timestamps` / `-t`	off	Add `[MM:SS]` before each line
`--lang` / `-l`	`ru,en`	Language priority (first available wins)

Output

Script prints to stdout: metadata header (title, channel, URL, video ID), separator, then transcript text.

The Iron Law

NEVER modify the transcript text. Output it exactly as received. Reformat into paragraphs (merge short lines into logical blocks) but NEVER change, summarize, or rephrase any words.

After Getting Transcript

•Save raw transcript as {video_id}-transcript.txt (or user-specified file)
•Then — and only then — analyze, summarize, or process if the user asked for it

Limitations

•Requires video to have subtitles (manual or auto-generated). If none — inform user, do not attempt workarounds
•Subject to YouTube rate limiting on heavy use