DeepSeek opens a super-agent prototype for one GitHub repo: it works on request and answers in the original thread
Synopsis
DeepSeek has published dsh-superagent, a research prototype of a proactive coding agent scoped to a single GitHub repository and packaged as a bundle for DeepSeek Harness. It listens to GitHub and optionally Feishu, opens a work item when an authorized person asks for work, handles it in its own agent session and directory, and answers in the original conversation, for example with a pull request. The release states it is research code, not an official product, and may differ from the internal version and the implementation described in the accompanying article.
Interpretation
A coding agent now appears as a standing helper shared by a whole team rather than a one-off chat tool: it watches one GitHub repository, starts a work item when someone asks for work, and returns the result to the original conversation. The release frames this as a way of working that takes as little human attention as possible and gives as much value as possible, while noting that an agent acting for a whole team can also waste a whole team's attention. To constrain that, the prototype acts only when asked or when a rule the team wrote allows it, and speaks only when someone must act or can use a result; a web panel shows what needs a person.
In operation, each request becomes a separate work item executed in a dedicated agent session and directory, with the reply sent back to the conversation that started it. The flow is defined layer by layer in numbered documents covering philosophy, domain model, authorization, execution, prompts, panel, deployment, architecture and how to add a Source. Setup has many steps, so a coding agent performs it with the user by following SETUP.md: one short question at a time, the full plan shown once, then installation and configuration; pnpm check runs the typecheck, tests and build.
Perspective
The release limits the scope to a research prototype of a coding agent for one GitHub repository, taking GitHub and optionally Feishu as input and answering in the original conversation, for example with a pull request; it is positioned as research code rather than production-ready software and must be reviewed and adapted before use.
The release says the prototype may differ from the internal version and from the implementation described in Exploring Super-agent Design, and that some designs discussed in the article are future directions not implemented in the prototype; the code may still contain bugs and its coding style and engineering practices may be inconsistent.