wtf is video-publish-skill?

Video Publish Skill is a local-first tool built for Codex that helps someone prepare a video for publishing without actually publishing it for them. Starting from a video file or its subtitles, it generates promotional copy such as a title, summary, and description, plus AI-generated cover images in four different aspect ratios. Once the user explicitly says yes, the skill can stage the video, cover, and metadata onto a content platform's creation page. The README is clear that the final publish button is always checked and clicked by a human, not the tool. The skill can pull a full transcript from MP4, MOV, MKV, or WebM video files, or work directly from an existing subtitle file, then turn that transcript into a title, summary, and cover text. Covers are produced in 16:9, 4:3, 3:4, and 9:16 ratios through Codex's image generation feature, and support reference images, keeping a consistent look for a person across covers, and redoing just one ratio while keeping the others. After two separate confirmations, it can open the creation page for platforms including Xiaohongshu, Douyin, Bilibili, and WeChat Channels in Chrome or Chromium, and optionally use a separate tool called Ego Lite to fill in the platform's form fields, though it still stops short of clicking publish. Installation happens through the skills.sh installer or a standalone install script for macOS, Linux, or Windows, placing the skill into the Codex skills folder. Extra tools like FFmpeg, Python, or Ego Lite are only checked for and required when a specific feature actually needs them, not installed automatically. By default, session data is written to a local, git-ignored folder, and video files, subtitles, transcripts, and covers are not uploaded anywhere unless the user picks a reference image to send to Codex's image generator. The project is released under the Apache License 2.0 and includes some vendored code from a related project under the MIT license.