feitangyuan/oh-my-cover-design

Oh My Cover Design is a small AI skill that helps non designers turn a written article into a ready to use prompt for generating a social media cover image. The README is in Chinese, and the project is aimed at people who post on Xiaohongshu, the Chinese lifestyle and shopping app where each post needs an eye catching cover. You hand the agent the text of your article, the skill activates automatically, and the agent asks you a short series of questions before writing the final image prompt. The questions cover the basic creative choices: which layout style you want, whether there is a face reference photo to include, what facial expression to use, whether to drop in a product screenshot, the background color palette, and the font choice. After the answers come in, the skill emits a single prompt block that you paste into your image model of choice. The README notes that the title text is suggested by the agent and that the layout reasoning is built into the skill, so you do not have to know anything about design or write your own prompt from scratch. The skill ships with ten named layout styles, each labeled with the kind of content it suits. The list includes a deep gradient style with large text behind a centered character, a flat color style on a clean background, a product hero style when a UI screenshot carries the cover, a before and after comparison style, a minimalist whitespace style, a poster collage style with stacked layers, a side placed character style that frees one side for the title, a back view style for inspiring posts, a partial framing style that shows only a hand or half a face, and a direct eye contact style for emotional posts. It supports Claude Code, Codex, and any AI agent that allows custom skills. To install the full skill you clone the repository into the local Claude skills directory at ~/.claude/skills/cover-design-open. If you only want the prompt logic itself, the README also shows a curl command that downloads just SKILL.md into the same folder. Reference images are optional. The README explains that the first slot is your own face photo, which the skill uses to keep facial features consistent across covers, while later slots can be any supporting material such as product photos, UI screenshots, or brand assets. The license is MIT.