Two DeepSeek Harness Plugins Turn Colleagues and Bilibili Into Agent Skills
A plugin ecosystem that can distill a teammate's knowledge or connect an agent to a content platform changes what a coding agent can know and do without custom integration work. The Bilibili-to-GitHub pipeline also offers a distribution model worth watching: video content drives adoption and contribution for open-source agent tooling in a way that pure README-driven projects rarely achieve.
DeepSeek Harness, a plugin-first agent framework, reached 112k GitHub stars within three days of release and already hosts over a thousand community plugins. Two projects topping its trending list show where the ecosystem is heading: colleague-skill (now dot-skill) distills a person's communication patterns from WeChat logs, PDFs, and documents into a reusable AI skill, while OpenBiliClaw builds a cross-platform content discovery agent that learns a user's psychological profile and proactively fetches relevant material from Bilibili, Xiaohongshu, Douyin, YouTube, X, Reddit, and other platforms.
Both projects were built by Bilibili content creators, a pattern that points to a shift in how Chinese open-source projects find early users. Developers explain the problem on video, then route viewers to GitHub for stars, forks, and issues, creating a tight feedback loop between content, code, and real-world usage. The framework's architecture treats models, tools, skills, sessions, and even the UI as swappable plugins, which lowers the barrier for turning a single idea into a community-installable component.
Bilibili is functioning as a launch platform for open-source agent tools, where a video explainer replaces the traditional landing page and creates a direct pipeline to GitHub engagement.
The speed at which DeepSeek Harness accumulated plugins suggests that a plugin-first architecture, where every capability is a swappable component, dramatically lowers the activation energy for contributing to an agent ecosystem.
Distilling a colleague into a skill is a concrete form of organizational memory that goes beyond documentation: the model captures communication patterns and contextual knowledge that would otherwise walk out the door.