大模型网摘DeepSeek Open-Sources dsh-superagent: A Proactive Coding Agent Prototype Largely Written by Agents
On October 31, the deepseek-ai organization published the dsh-superagent prototype repository on GitHub (dsh-superagent-prototype) under the MIT license. Two things stand out. First, it is a "proactive" coding agent that, once deployed, keeps watch over a single repository and picks up work. Second, per the project's own README, much of the engineering was done by coding agents: "This code grew out of our team's day-to-day development work, with much of the engineering done by coding agents." Two Hacker News threads about the project appeared the same day.
What it is: a proactive agent that watches one repository
dsh-superagent is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness. It listens to GitHub events and, optionally, Feishu. When an allowed user asks it for work, it opens a work item, works on it in its own agent session and directory, and then answers in the original conversation — for example, with a pull request. A web panel shows which items need a human.
Design philosophy: spend as little human attention as possible
The README also frames the project as "a way of working with an agent": take as little human attention as possible, and give as much value as possible. An agent that can act for a whole team can also waste a whole team's attention, so it acts only in two cases — when explicitly asked, or when a rule the team wrote allows it. Likewise, it speaks only when someone must act or can use a result. docs/000-philosophy.md in the repository explains the principle and what it asks of the operator.
Setup: an agent installs it with you
Setup has many steps, and the recommended path is to hand it to your coding agent: ask it to read SETUP.md in the repository and set up dsh-superagent with you. It asks one short question at a time, shows the full plan once, and then installs and configures everything. The documentation is organized as nine numbered pieces (from 000 Philosophy to 008 Cookbook, the latter covering how to add a new source), each building only on the ones before it. On the development side, a single pnpm check runs the typecheck, the tests and the build.
Prototype caveats: not identical to the paper
The repository states plainly that this is a research prototype for reference only, not an official DeepSeek product; it may still contain bugs, and its coding style and engineering practices may be inconsistent, so review and adapt it before use. It has also been adapted for publication and is not identical to the internal version or to the implementation described in the paper "Exploring Super-agent Design: Towards a More Autonomous Agent System" — some designs discussed in the paper are future directions and are not implemented in this prototype.
Why it matters
This is DeepSeek's second notable open-source move under its official GitHub organization in recent weeks, after the DeepInfer inference framework, and the direction has shifted from the inference layer to the agent-workflow layer. For developers, it is a deployable sample of a team-level proactive agent; for observers, an "agent framework largely written by agents" is itself a signal worth tracking. For the broader consolidation trend among coding agents, see our earlier roundup of nine coding agents integrated with the DeepSeek API and the DeepSeek Harness v0.2.1 tools overview; for the inference side, read our DeepInfer open-source report.