video-downloader（视频下载器）

Video Downloader

Download single videos from user-provided URLs with predictable quality, format, and output paths.

Setup

On first use, read setup.md for integration guidelines.

When to Use

Use this skill when the user asks to download a video or extract audio from a video URL. It is optimized for one-off downloads with explicit quality and format requirements.

Architecture

Memory lives in ~/video-downloader/. See memory-template.md for structure.

~/video-downloader/
├── memory.md             # Status + user preferences
├── downloads-log.md      # Optional history of completed downloads
└── failed-downloads.md   # Optional retries and failure reasons

Quick Reference

Load only what you need to keep context small during execution.

Core Rules

1. Confirm Rights and Target First

• Ask for the exact URL and intended use when unclear.
• If the request implies unauthorized copying, refuse and suggest legal alternatives.

2. Inspect Metadata Before Downloading

• Run metadata check first to confirm title, duration, and available formats.
• If metadata fetch fails, stop and report the exact error instead of retrying blindly.

3. Match Quality to User Intent

• Use best when user says "highest quality".
• Use capped quality (1080p, 720p, etc.) for smaller files or device limits.
• Use audio-only mode only when they explicitly want audio extraction.

4. Use Deterministic Output Names

• Save files as %(title)s [%(id)s].%(ext)s to reduce collisions.
• Keep downloads in a user-approved directory and never write outside it.

5. Prefer the Local Wrapper Script

• Use python3 download_video.py "<url>" ... for consistent behavior.
• Fall back to raw yt-dlp commands only if the user asks for custom flags not covered by the script.

6. Verify Output Before Declaring Success

• Confirm file exists, extension matches request, and size is non-zero.
• For audio-only downloads, confirm output is .mp3.

Common Traps

• Downloading playlists accidentally -> use --no-playlist by default.
• Choosing best for limited storage -> oversized files and slow transfers.
• Re-trying blocked URLs repeatedly -> temporary ban risk and no progress.
• Saving with title only -> filename collisions across similar uploads.
• Skipping metadata check -> wrong media downloaded from redirected links.

External Endpoints

The downloader only contacts domains implied by the user-provided URL.

No other data is sent externally.

Security & Privacy

Data that leaves your machine:

• Only the target media URL and standard downloader request headers sent by yt-dlp.

Data that stays local:

• Downloaded files in the selected output folder.
• Optional memory notes under ~/video-downloader/.

This skill does NOT:

• Store credentials in plain text.
• Access files outside user-approved output paths.
• Download playlists unless the user explicitly asks.
• Make undeclared network requests outside the target media host.

Trust

By using this skill, requests are sent to the video host domains behind the provided URL. Only install if you trust those services with your request metadata.

Related Skills

Install with clawhub install <slug> if user confirms:

• video — video processing workflows
• ffmpeg — codec conversion tasks
• audio — audio cleanup workflows
• youtube-video-transcript — transcript extraction

Feedback

• If useful: clawhub star video-downloader
• Stay updated: clawhub sync