Image: GitHub
Why it mattersAn agent that acts for a whole team can also waste a whole team's attention. The design rule here is to speak only when someone must act, or can use a result.
Most coding agents sit beside the keyboard and wait for a prompt. DeepSeek's new one sits inside GitHub and watches for a mention. It opens the issue and works on it in its own session. It answers in the thread, often with a pull request.
DeepSeek released dsh-superagent on October 31. It is a research prototype packaged as a bundle for DeepSeek Harness. The GitHub repository was created on October 31 and holds 882 stars and 63 forks today. The repo is ranked 19th on Trendshift's daily list on October 31. The license is MIT. The code is TypeScript. DeepSeek says the prototype is research code, not production software.
What it does
dsh-superagent is a proactive coding agent for one GitHub repository. It listens to GitHub and, optionally, to Feishu, a Chinese team chat tool. An allowed user asks the agent for work in a comment or an issue. The agent opens a work item. It runs in its own agent session and its own directory. It answers in the original conversation, for example with a pull request. A web panel shows what needs a person.
The repo is tied to DeepSeek's paper, "Exploring Super-agent Design: Towards a More Autonomous Agent System." DeepSeek says the public code is adapted and does not match the internal version. Some ideas in the paper are future work and are not in the prototype.
Setup has many steps. The recommended way to install it is to point another coding agent at the included SETUP.md. The install agent asks one short question at a time, shows the full plan once, and then configures the environment.
The design rule
The README states one design rule as the point of the whole project. In its words: take as little human attention as possible, and give as much value as possible. The team argues that an agent for a whole team can also waste a whole team's attention. So dsh-superagent acts only when asked, or when a rule the team wrote allows it. It speaks only when someone must act or can use a result.
The repo includes numbered documentation that reads top to bottom: philosophy, domain model, authorization, execution, prompts, panel, deployment, architecture, and a cookbook for adding a new source.
What a reader does with it
For a team already running a coding harness, dsh-superagent shows one shape for giving an agent one repository. The authorization layer and the "act only on allowed asks" rule are the parts worth reading. For a team using Feishu, it is also a working bridge between GitHub and Feishu threads.
For a team not using DeepSeek Harness, this prototype is tied to that harness. The patterns are portable. The runtime is not.
DeepSeek says review the code and adapt it before using it anywhere real. "It may still contain bugs, and its coding style and engineering practices may be inconsistent" is in the top-of-README warning. That is the right posture for research code with one commit and a day on GitHub.
Source
deepseek-ai/dsh-superagent-prototype on GitHub, under MIT license.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
