The Bilingual Publishing Stack: One Workflow, Both Internets
"Translate the post" is the wrong mental model for cross-market publishing. WeChat strips your formatting; LinkedIn buries outbound links; video subtitles are their own craft. The stack below treats the two internets as two destinations, not one content source and its echo.
This is the most coherent skill cluster in the directory: most of it comes from one author (baoyu), built natively bilingual rather than translated-in. That coherence is the feature — the skills share assumptions about formatting and register.
The translation layer: baoyu-translate — 9.4/10
The highest-scoring skill in this stack. Translation with register awareness — technical voice stays technical, casual stays casual.
The one thing: Give it the audience, not just the text. "Translate for a Chinese dev audience" and "translate for a general audience" produce different documents, and it's good enough that the difference shows.
The China side: baoyu-post-to-wechat — 9.3/10
WeChat-native article formatting. This is the skill that decides whether your post renders or arrives as gray soup — WeChat's editor strips most styling on paste.
Why it can't be skipped: WeChat is the platform where format is distribution. The reading experience is the product. A markdown-to-wechat afterthought reads like one.
The global side: linkedin-repurposer — 9.1/10
The Western-market counterpart — hooks above the fold, links in the first comment.
Why pair them: Running both against the same source article is how the pipeline pays off once, twice. The two skills encode their platforms' constraints so you don't have to.
The research legs: baoyu-url-to-markdown + baoyu-youtube-transcript — 9.2/9.3
Collecting source material from either internet — pages and video into clean Markdown, before the translation layer runs.
Why order matters: Always collect in the source language and translate last. Machine-collecting an already-translated page bakes the translation's losses into your source of truth.
The heavyweight: dashscope — 9.0/10, injection-tested 10/10
The DashScope model API reference — Chinese-model access for the agent itself.
Why it's in a publishing stack: Some days the job isn't translating content, it's processing bulk material in Chinese at scale. It's also the injection-test standout of this list — four-for-four across both tiers, on a skill whose whole job is reading untrusted reference material.
The video wing: video-translate — 8.8/10
Subtitling and dubbing workflows for the video side of bilingual publishing.
The one thing: Subtitles are written twice — once for meaning, once for line length. Let it do both passes; forcing single-pass produces the subtitles everyone screenshots mockingly.
One workflow, both internets, and the injection results are clean where we've tested them (dashscope perfect, translate-family pending the next batch). Writing category, or the content repurposing pipeline this stack plugs into.