// the find
JimLiu/baoyu-skills
A collection of agent skills for Claude Code, Codex, and similar AI coding assistants — primarily focused on content creation workflows: generating infographics, cover images, slide decks, comics, and posting to Chinese social platforms (WeChat, Weibo, Xiaohongshu). Most of the value is in the content generation side, not in software engineering tooling. The target audience is Chinese content creators who use AI agents as part of their publishing pipeline.
The style system for visual content is genuinely well thought out — orthogonal dimensions (type × palette × rendering × mood) rather than a flat list of presets means you get combinatorial coverage without exponential maintenance. The baoyu-fetch package doing CDP-based browser automation for WeChat/Weibo is the right call given those platforms' anti-automation stance. The warning about context overhead from bulk-installing all skills is honest and practically useful — most skill repos skip that entirely. The per-skill directory structure that both Claude Code and Codex can scan natively means no custom tooling required to adopt it.
The README is doing enormous work for what is essentially a collection of prompt files with thin JS wrappers — there is almost no code to evaluate, and the actual quality of the skill prompts is invisible from the repo surface. The Chinese platform posting skills (WeChat, Weibo, Xiaohongshu) have hard geographic and credential dependencies that make them useless to the majority of GitHub's audience, yet they dominate the documentation. The baoyu-image-gen skill abstracts over 10+ image generation APIs, which sounds useful but means any change to any provider breaks something and the abstraction leaks everywhere (different providers support different aspect ratios, reference image modes, sizes). Zero tests visible for the actual skill logic — the test files exist for the utility packages but the SKILL.md prompts themselves are untestable.