← 返回 Skills 市场
1775
总下载
3
收藏
11
当前安装
7
版本数
在 OpenClaw 中安装
/install video-download
功能描述
Download videos from 1800+ websites and generate subtitles using Faster Whisper AI. Use when user wants to download videos from YouTube, Bilibili, Twitter, T...
安全使用建议
This skill generally appears to do what it says: it uses yt-dlp to download videos and Faster Whisper to transcribe them. Before installing or running it: 1) Inspect SKILL.md and scripts locally — the pre-scan flagged unicode control characters which can be used to manipulate LLMs; remove or verify them. 2) Do not supply browser cookies or cookiefile paths unless you understand the risk: cookies may include session tokens that grant account access. Prefer using temporary or narrowly scoped credentials. 3) Run pip installs in an isolated environment (virtualenv, container) and verify package sources (yt-dlp, faster-whisper are common but third-party forks exist). 4) Expect large model downloads from HuggingFace (multi-GB) and ensure you have bandwidth/storage; consider running in a sandbox. 5) Be aware of legal/DRM issues for premium services (Netflix, Disney+, etc.); success may require valid authenticated sessions and may still fail due to DRM. 6) If you plan to allow the agent to invoke the skill autonomously, limit exposure: do not store or expose sensitive cookies or credentials to the skill without a clear need. If you want a safer posture, ask the publisher for provenance (source repo, signed releases) or prefer a skill with a known trusted origin.
功能分析
Type: OpenClaw Skill
Name: video-download
Version: 0.1.6
The skill provides a wrapper for `yt-dlp` and `faster-whisper` that includes high-risk capabilities, specifically the ability to extract session cookies directly from local browser profiles (e.g., Chrome, via the `cookiesfrombrowser` parameter) and accept arbitrary cookie files. While these are legitimate features of `yt-dlp` used to bypass anti-bot measures on video platforms, they represent a significant attack surface for credential theft or session hijacking if the AI agent is manipulated via prompt injection. The script `scripts/video_parser.py` also allows specifying an arbitrary output directory, which could lead to unauthorized file writes, although it does implement basic sanitization for video titles to prevent simple path traversal.
能力评估
Purpose & Capability
Name/description (download videos + generate subtitles) aligns with required binaries (python, ffmpeg), the listed pip packages (yt-dlp, ffmpeg-python, faster-whisper), and the included script (scripts/video_parser.py). The supported-sites list is long but consistent with yt-dlp usage; authentication via cookiefile is documented where necessary.
Instruction Scope
Runtime instructions tell the agent to run the included Python script with JSON parameters. The script reads URLs, output paths, and optional cookie cookies/cookiefile inputs, downloads videos, extracts audio, and transcribes locally using Faster Whisper. It does not (in the visible code) reach out to unexpected endpoints beyond package/model downloads, but SKILL.md contains detected unicode control characters (prompt-injection signal) which may be an attempt to manipulate an LLM-based evaluator/agent. Also note the skill accepts browser cookie inputs or cookiefile paths — those can contain sensitive auth data and should only be provided intentionally.
Install Mechanism
The registry record shows no automated install spec, but SKILL.md includes a pip install recommendation for standard public packages (yt-dlp, yt-dlp-ejs, ffmpeg-python, faster-whisper, tqdm). There are no obscure download URLs or extracted archives. Faster-Whisper will download models from HuggingFace on first run (large, multi-GB), which is expected but significant (network usage, storage).
Credentials
The skill declares no required environment variables or external credentials. It does accept cookie/cookiefile/cookiesfrombrowser parameters (user-supplied), which are necessary for authenticated downloads on some sites but are sensitive — the skill does not declare or require unrelated credentials.
Persistence & Privilege
The skill is not always-enabled and uses default autonomous invocation settings. It does not request system-wide modifications or other skills' configs. No elevated persistence or unusual privileges are requested.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install video-download - 安装完成后，直接呼叫该 Skill 的名称或使用
/video-download触发 - 根据 Skill 的参数说明提供必要输入，即可获得结构化输出
版本历史
v0.1.6
Initial release with enhanced authentication and subtitle-only download options.
- Added support for cookie-authenticated downloads using "cookie", "cookiesfrombrowser", and "cookiefile" parameters.
- New "onlysubtitle" option allows downloading subtitles without downloading video files.
- Updated dependencies to require yt-dlp-ejs and use "python" instead of "python3".
- Documentation expanded to include cookie usage instructions and troubleshooting tips for authentication errors.
v0.1.5
- Added references/supportedsites.md, providing a local document listing all supported video sites.
- Added scripts/video_parser.py as the entry point script for downloading videos and generating subtitles.
- These new files make it easier to check supported sites and run the skill from the command line.
v0.1.4
- Removed files: references/supportedsites.md and scripts/video_parser.py have been deleted.
- Documentation fixes: The default Faster Whisper model in prerequisites and parameters changed from "large-v3" (~3GB) to "small" (~75MB).
- The install instructions and troubleshooting sections now reflect "small" as the default model and adjust the model size note accordingly.
- No changes to core logic are described; only documentation and structural adjustments are apparent.
v0.1.3
- Switched transcription backend from OpenAI Whisper to Faster Whisper for improved speed and memory efficiency.
- Updated dependencies: replaced openai-whisper with faster-whisper and added tqdm for progress display.
- Default model for transcription is now "large-v3" from HuggingFace; instructions updated for VPN needs in China.
- Parameter descriptions, examples, and prerequisites revised to reflect switch to Faster Whisper.
- Removed the "verbose" parameter; logging details now handled automatically.
v0.1.2
- Added overwrite_subtitle parameter to allow skipping subtitle generation if the file already exists.
- Updated documentation to include overwrite_subtitle and an example of skipping transcription when subtitles are present.
v0.1.1
- Added a new verbose parameter to enable detailed transcription logs.
- Linked directly to the yt-dlp supported sites list for easier reference.
- Updated documentation to reflect the new verbose option and additional info about supported sites.
v0.1.0
Initial release of video-download: Download videos from 1800+ sites and generate subtitles.
- Supports video downloads from YouTube, Bilibili, Twitter/X, TikTok, Facebook, Vimeo, and more
- Generates subtitles/transcripts from video audio using Whisper AI
- Allows extraction of audio as WAV and configurable subtitle formats (txt, srt, vtt, json)
- Supports batch downloads and custom output directory
- Option to download built-in subtitles from supported videos
- Command-line usage provided with detailed parameter descriptions and examples
元数据
常见问题
video-download 是什么？
Download videos from 1800+ websites and generate subtitles using Faster Whisper AI. Use when user wants to download videos from YouTube, Bilibili, Twitter, T... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 1775 次。
如何安装 video-download？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install video-download」即可一键安装，无需额外配置。
video-download 是免费的吗？
是的，video-download 完全免费（开源免费），可自由下载、安装和使用。
video-download 支持哪些平台？
video-download 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。
谁开发了 video-download？
由 upupc（@upupc）开发并维护，当前版本 v0.1.6。
推荐 Skills