DeepSeek Harness's Bilibili Debut Turns Comment Section Into a Plugin Marketplace
Agent tooling competition now hinges on whether users will build and share plugins, not on raw model scores. DeepSeek's Super-Agent bet says the moat is community adoption, which means stability, state management, and low-friction plugin design are the real differentiators.
DeepSeek Harness's official Bilibili debut drew 2.11 million views in its first 24 hours and added 400,000 followers. The video itself demonstrates the framework's 'creation mode' that turns a natural-language request into a working plugin. But the real signal is in the comments: users are publicly showing off their own plugins, discussing implementation details, and competing on plugin counts. One user turned a daughter's drawing into a platformer that got over 6,000 likes.
The author reads this as a shift in the Agent race. It's no longer about which model is smarter, but which ecosystem gets enough people to build on it. DeepSeek's Super-Agent architecture, announced on October 31, tries to remove the human bottleneck of managing multiple sessions. Its two-layer structure has a Router that decides which Worker handles a message, and Workers run with independent contexts. The harder ambition is proactive agents that detect bugs in group chats, reproduce them, and even suggest PRs without being tagged.
The article also warns that plugin ecosystems solve existence, not stability. Version compatibility, state preservation across long multi-day runs, and parallel write conflicts are unsolved problems. IDC research manager Sun Zhenya argues that the competition is shifting from task-based delivery to ecosystem, memory, and concierge services. The author's conclusion: half the arena has changed. Model capability is still the foundation, but the differentiator is whether enough people voluntarily write plugins and tutorials, turning a framework into something the community defines.
The rapid conversion of a comment section into a plugin fair shows that developer enthusiasm for agent frameworks now hinges on personalization and extensibility, not just benchmark scores.
Super-Agent's proactive bug detection shifts the burden of correctness from human oversight to the system's state management and exception handling, which is a bigger engineering lift than making a model call a tool.
The simplicity of DSH's plugin API is the enabling condition for user-generated variety; without a low barrier, the long-tail needs that the official team can't anticipate would remain unmet.
The author's claim that ecosystem moats grow from a thousand small user plugins rather than official additions aligns with how open-source communities build stickiness, but it also exposes the risk that a framework's stability will depend on the patchwork of user code.