Video Downloader skill

Download media from a URL with yt-dlp, and merge separate video+audio streams into one file with ffmpeg. This is the command-line equivalent of this repo's Chrome extension, for when the browser is not used.

When to use

Use this skill when the user:

• gives a video/audio page URL and asks to download it, or
• asks to grab/save a video or its audio from a site, or
• names a target folder for a download.

Supported sites are everything yt-dlp supports (1000+): Bilibili, X/Twitter, YouTube, and more.

Tool

This skill is self-contained: the downloader script download_media.py lives next to this file in the skill directory. Always invoke it by its absolute path so it resolves no matter where the skill is installed (user level or workspace level) and regardless of the current working directory. When the skill is activated, the reported "Skill directory" is that absolute location — join download_media.py to it:

python "<SKILL_DIR>/download_media.py" "<URL>" [--quality <q>] [--output "<dir>"] [--audio]

The script streams yt-dlp progress to the terminal and prints Done. Saved to: <dir> on success.

Arguments

• URL (required): the media page address. Always wrap it in double quotes — URLs often contain ?, &, and =.
• --quality / -q: one of best (default), 2160, 1440, 1080, 720, 480, 360, audio.
• --audio: shortcut for --quality audio (audio only, no merge).
• --output / -o: target folder. Default is the current user's Downloads folder. The folder is created if it does not exist.

Behaviour you can rely on

• Output filename uses the real media title plus the id, e.g. 月下煮茶 普通话版 [BV1ke5U65ECa].mp4. Non-ASCII titles (Chinese, etc.) are kept; characters illegal on the OS are sanitised.
• best and the height presets merge the best video and audio into one .mp4. Audio-only produces a single audio file with no merge.
• ffmpeg is located automatically (PATH, then ~/ffmpeg/bin and other common spots). Merge quality options fail with a clear message if ffmpeg is absent.

How to act on a request

1. Extract the URL from the user's message. If none is present, ask for it.
2. Pick quality: default to best. Use audio only if the user asks for audio or music.
3. Pick output: default (Downloads) unless the user names a folder; then pass --output "<that folder>".
4. Run the script with execute_pwsh, calling it by the skill directory's absolute path (do NOT assume a workspace-relative .kiro/skills/... path — this skill may be installed at the user level, outside any workspace).
5. Report the saved file path from the script's final Done. Saved to: line. On failure, relay the error (common causes: missing ffmpeg, outdated yt-dlp — suggest yt-dlp -U).

Examples

In each example, <SKILL_DIR> is the absolute skill directory reported when the skill is activated (e.g. C:\Users\<you>\.kiro\skills\video-downloader for a user-level install, or <workspace>\.kiro\skills\video-downloader for a workspace-level one).

Request: "下载 https://www.bilibili.com/video/BV1ke5U65ECa" Run:

python "<SKILL_DIR>/download_media.py" "https://www.bilibili.com/video/BV1ke5U65ECa"

Request: "把这个视频下成 1080p 存到 D:\clips https://x.com/i/status/123" Run:

python "<SKILL_DIR>/download_media.py" "https://x.com/i/status/123" --quality 1080 --output "D:\clips"

Request: "只要这个视频的音频 <URL>" Run:

python "<SKILL_DIR>/download_media.py" "<URL>" --audio

Requirements

• yt-dlp on PATH (yt-dlp --version).
• ffmpeg reachable (on PATH or ~/ffmpeg/bin) for any non-audio quality.

Both are already installed in this environment.