← Back to Skills Marketplace
363
Downloads
0
Stars
3
Active Installs
10
Versions
Install in OpenClaw
/install youtube-whisper
Description
YouTube影片一鍵轉文字！自動下載影片並用AI轉成中文/英文字幕，沒有字幕的影片也能用。
Usage Guidance
This skill appears to do what it says: it downloads YouTube audio (via yt-dlp) and transcribes locally with Whisper. Before installing/using: (1) ensure you trust the author and review the included script (it is plain shell, not obfuscated); (2) install yt-dlp, ffmpeg, and openai-whisper from their official sources (brew/pip) and be aware pip packages can execute code at install time; (3) the script downloads content from YouTube to /tmp and will remove temp files—be mindful of disk and network usage and of any copyright/legal constraints for downloading content; (4) the script prompts for interactive confirmation when memory is low and may hang in non-interactive agent runs—use the --force flag and a small model for automated contexts; (5) no credentials are requested, and there is no evidence of data exfiltration to third-party endpoints beyond fetching YouTube content.
Capability Analysis
Type: OpenClaw Skill
Name: youtube-whisper
Version: 1.4.1
The youtube-whisper skill is a legitimate tool for transcribing YouTube videos using yt-dlp and the OpenAI Whisper model. The core logic in scripts/youtube-whisper.sh is well-implemented, featuring system resource checks (RAM/CPU), support for both subtitle extraction and audio transcription, and proper cleanup of temporary files in /tmp. The script uses proper shell quoting for variables like URL and output paths, mitigating common injection risks, and contains no evidence of data exfiltration or malicious intent.
Capability Assessment
Purpose & Capability
Name/description claim: download YouTube videos and transcribe with Whisper. Manifest and SKILL.md list yt-dlp, ffmpeg, and openai-whisper (pip) which are exactly what's needed. The included shell script implements subtitle detection, yt-dlp downloads, and local Whisper invocation—consistent with purpose.
Instruction Scope
SKILL.md instructs installing yt-dlp/ffmpeg and pip openai-whisper and running the provided script; the script only accesses the given YouTube URL, /tmp for temporary files, and local system resource info. It prompts interactively when memory is low (read -p) which can block non-interactive agents—this is a runtime usability note rather than malicious behavior. No instructions to read unrelated config files or transmit data to external endpoints beyond YouTube/yt-dlp.
Install Mechanism
No install spec provided (instruction-only + bundled script). Dependencies are installed via standard package managers (brew, pip) per SKILL.md; nothing is downloaded from an unusual URL or extracted by the skill installer.
Credentials
Skill declares no required environment variables or credentials and the script does not access secrets or external tokens. All environment/system accesses (free, sysctl, top) are reasonable for performing resource checks.
Persistence & Privilege
always:false and default autonomous invocation settings; the skill does not attempt to modify other skills or system-wide configs. It runs transient shell commands and cleans up its temp directory on exit.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install youtube-whisper - After installation, invoke the skill by name or use
/youtube-whisper - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.4.1
youtube-whisper v1.4.1
- Minor update to scripts/youtube-whisper.sh.
- No changes to user-facing documentation or features.
v1.4.0
youtube-whisper v1.4.0
- 新增 --force 參數，允許在記憶體不足時強制執行（跳過記憶體檢查）
- SKILL.md 增加可用模型表格與各模型記憶體需求說明
- 說明與使用範例同步更新支援 --force 參數的操作方式
- 新增 _meta.json 檔案
v1.3.1
- Major SKILL.md rewrite: clarified usage, streamlined instructions, and improved feature descriptions in Chinese.
- Updated description for broader, simpler appeal and clearer functionality.
- Added language auto-detection info and expanded usage scenarios.
- Standardized usage syntax and requirements.
- Cleaned up duplicate or redundant author credits.
v1.3.0
youtube-whisper 1.3.0 introduces a structured response format with additional output details:
- Adds a standardized reply format showing video title, processing time, transcription source (subtitles/Whisper), and transcript content.
- Updates documentation to reflect the enhanced output, including details on time spent and the method used (subtitles or Whisper transcription).
v1.2.2
- Bumped version to 1.2.2.
- Documentation updates in README.md and SKILL.md.
- No functional script changes.
v1.2.1
- Updated version to 1.2.1.
- Documentation refreshed in SKILL.md and README.md; no functional script changes.
- No user-facing feature or behavior changes.
v1.2.0
- Version bump to 1.2.0.
- Documentation updates only; no functional changes.
- README.md and SKILL.md refreshed for clarity and accuracy.
v1.1.0
No changes detected in this version.
v0.1.1
- Bumped skill version to 1.1.0.
- Added version field to SKILL.md.
v0.1.0
Initial release of youtube-whisper:
- Download YouTube videos and transcribe audio locally with Whisper.
- Designed for cases where videos have no subtitles or youtube-watcher fails.
- Supports specifying output filename and Whisper model.
- Default transcription language is Traditional Chinese.
- Provides hardware requirements, tested environment, and basic setup instructions.
Metadata
Frequently Asked Questions
What is Youtube Whisper?
YouTube影片一鍵轉文字！自動下載影片並用AI轉成中文/英文字幕，沒有字幕的影片也能用。 It is an AI Agent Skill for Claude Code / OpenClaw, with 363 downloads so far.
How do I install Youtube Whisper?
Run "/install youtube-whisper" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is Youtube Whisper free?
Yes, Youtube Whisper is completely free, licensed under MIT-0. You can download, install and use it at no cost.
Which platforms does Youtube Whisper support?
Youtube Whisper is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created Youtube Whisper?
It is built and maintained by ken (@dolphins1123); the current version is v1.4.1.
More Skills