DeepSeek's dsh-superagent: a research prototype of a proactive coding agent

What it says

AllAbstractFactsIdeasInsightsPointsQuestionsQuotesSurprisesTensionsThesis

1. dsh-superagent is a research prototype from the DeepSeek team: a proactive coding agent for one GitHub repository, packaged for DeepSeek Harness. It listens to GitHub and optionally Feishu, opens a work item when an allowed person asks, works in its own session and directory, and answers in the original conversation, for example with a pull request. Its guiding principle is to take as little human attention as possible while giving as much value as possible. The page says the code is not production-ready, and it describes setup done with a coding agent following SETUP.md, the numbered documentation, and the pnpm check command.
2. When an allowed person asks for work, dsh-superagent opens a work item, works on it in its own agent session and directory, and answers in the original conversation, for example with a pull request. u4 The pull request is given as one example of an answer, not the only form.
3. The documentation consists of nine numbered documents, from 000 Philosophy to 008 Cookbook, which the reader should read in order because each builds on the ones before it. u10–19 The list runs from 000 to 008, which gives nine documents.
4. The command pnpm check runs the typecheck, the tests and the build. u20–21
5. Set up dsh-superagent by asking one's own coding agent to read SETUP.md in the repository and do the setup together with the user. u6–8 Addressed to operators who will install the tool.
6. The agent acts only on request or under a team rule, because an agent that can act for a whole team can also waste a whole team's attention. u5 This explains why restraint is built into the agent's design, not just its settings.
7. dsh-superagent is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness. It listens to GitHub and optionally Feishu, and it acts only when asked or when a team rule allows. u4–5
8. Setup is meant to be done with a coding agent: the agent asks one short question at a time, shows the full plan once, and then installs and configures everything by following SETUP.md. u6–9
9. How can a proactive coding agent serve a whole team while taking as little human attention as possible and giving as much value as possible? u5 The source states the principle but does not fully answer this question in the page itself; it points to docs/000-philosophy.md.
10. The principle the agent is built around: take as little human attention as possible and give as much value as possible. u5 It states the project's guiding principle in one line.
11. Much of the prototype's engineering was done by coding agents, not only by people. u1–2 A reader might expect a team's own research code to be written by its engineers alone.
12. The code is a research prototype that may still contain bugs and has inconsistent style, and it is not identical to the internal version or the article's implementation, so some designs in the article are future directions only. u1–3 The source warns readers to review and adapt the code before using it in their own environment.
13. dsh-superagent is a proactive coding agent for one GitHub repository that acts only when asked or when a team rule allows, and it speaks only when someone must act or can use a result. How can a proactive coding agent serve a whole team while taking as little human attention as possible and giving as much value as possible?