DeepSeek Harness Officially Handpicks Four Plugins; Some Worry Review Posts Will Become the Standard Answer
Officially selected four out of 35,000 plugins and named them for recommendation, with data showing 1.54 million DAU and 92 trillion tokens consumed daily. Two voices in the comments: quality has a baseline, but diversity is gone.
On November 5, DeepSeek Harness's official WeChat account published an article, selecting four from 35,000 third-party plugins and naming them as recommendations.
First, some background. Harness v0.1 launched on August 13, and the desktop version was released on September 29. As of November 4, among users of the official API, daily active users were 1.54 million and daily token consumption was about 92 trillion. There were about 35,000 third-party plugins, and about 60% of active users had created or installed at least one third-party plugin.
The four plugins officially promoted this time each cover a different functional direction.
① Self-Media Creator Workbench dsh-media-desk
By woniuniuniu, whose Xiaohongshu account is Snail's Tech Notes. It packs idea capture, script writing, cover design, and storyboard design into one workbench, continuing Harness's minimalist style. In the author's own words: wake up, open DSH, instantly open the workbench, and get going.
② Design Plugin dsh-guizang-design
By op7418, WeChat account Guizang's AI Toolbox. It combines commonly used design and copy-polishing Skills into a plugin; an article can be turned into a webpage, an image-text card, or a cover, with the finished work stored in 'My Works' and supporting 'Create Similar' reuse. According to the author, the plugin system can offer more than Skills: works, columns, and tone can all accumulate, making content generation more consistent.
③ Context Insight Plugin dsh-context
By bowenliang123, Bilibili account Bowen Bowen. From the composition and evolution of Context to the details of multi-Agent team collaboration, users can see what information an Agent used while working and how that information changes with the conversation. The author believes that the open plugin mechanism puts each Agent's context and Harness back in users' hands.
④ Paper Reading Plugin agentero-dsh
By Phil-Fan, Xiaohongshu PhilFan. It connects Agentero into Harness, so reading papers, highlighting, translating, and note-taking are all done in one place, and Agents can also help organize ideas. The idea came from the team's daily research work; the goal is a literature tool that is comfortable for both people and Agents to use.
Four demos from the source material:
The official team also gave plugin authors reassurance: improving the stability and compatibility of the plugin API, refining plugin security mechanisms, and building an official plugin marketplace to make uploading easier for authors and discovery easier for users. It is also preparing a support program for authors, including Token incentives, technical exchanges, and work recommendations. Specific rules will be announced when the program is officially launched; for now, the sign-up form is only collecting expressions of interest.
Official endorsement is a double-edged sword.
One netizen (@青青草原我为王) put it bluntly: the upside of official recommendation is that quality has a baseline; the downside is that these four will soon become the default setup in every Harness review post, and diversity will have to come from the next batch of wild plugins.
Another netizen (@Jane) ran into a more concrete problem: a month ago she customized the workbench for her own project, and now she can't sync official features or use two apps at the same time. The official article did not mention this kind of customization cost.
Someone else (@Hale) took a longer view: the official team personally stepping in to endorse community plugins sends a clear signal—the model itself is only a ticket to entry; the ecosystem plugins that grow around workflows are the real moat.
Out of 35,000 plugins, the official team named only four. These four will very likely become standard equipment; the remaining long tail is what this ecosystem really looks like.
发布时间: 2026-11-06 12:33