DeepSeek publishes super-agent prototype

DeepSeek has published dsh-superagent, an open-source research prototype for a coding agent that works on one GitHub repository. It listens to GitHub and, optionally, Feishu. When someone authorized asks it to do work, it opens a work item, uses its own agent session and directory, then replies in the original conversation, for example with a pull request. A web panel shows where a person needs to step in. The project page describes the code as research work, not an official DeepSeek product or production-ready software.

For a business, the practical point is that this is another option for building agent workflows around a team’s existing requests and rules. The prototype is designed to act only when asked or when a team-written rule permits it, and to speak when someone needs to act or can use a result. That offers a concrete model for assigning bounded work to an agent, while keeping a view of what still needs human attention. It isn’t ready to drop into a live operation without review and adaptation.

If you’re thinking about how agents should receive work, act, and involve people, this is the kind of thing we build into an AI command centre.