deepseek-ai/dsh-superagent-prototype

dsh-superagent is a research prototype of a coding agent that watches a single GitHub repository (and optionally Feishu), turns allowed requests into work items, handles each in its own agent session and directory, and answers back where the request was made — for example with a pull request. A web panel shows what needs a person. It is packaged as a bundle for DeepSeek Harness and is aimed at teams curious about agent-driven repository maintenance, not at production deployment.

How a request flows

Someone authorized asks the agent for work in GitHub or, if enabled, Feishu. The agent opens a work item, works in an isolated agent session and directory, and posts its answer in the original conversation. The project's stated operating idea is to take as little human attention as possible while still giving value: it acts only when asked or when a rule the team wrote allows it, and reports back only when a person must act or can use the result. The design rationale is in the philosophy document 000 Philosophy, and the domain model, authorization and execution are documented in sequence domain model authorization execution.

A concrete use case: a maintainer mentions the agent in an issue; it picks up the task, produces a branch and pull request in the repository, and replies in the issue thread. The web panel panel collects the items that need a human decision, so routine work does not have to be polled from GitHub notifications.

Setup and requirements

This is not a drop-in install. Setup has many steps and is meant to be done conversationally: you ask a coding agent to read SETUP.md and configure the system with you, one short question at a time, showing the full plan before installing SETUP.md. Because it ships as a bundle for DeepSeek Harness, using it means adopting that harness and the associated deployment and configuration steps described in the deployment document deployment.

Constraints and caveats

The README is explicit that this is a research prototype for reference only, not an official DeepSeek product. It grew out of the team's daily work with heavy assistance from coding agents, and the authors warn it may contain bugs with inconsistent coding style and engineering practices; they ask you to review and adapt it before using it in your own environment README. The published version is not identical to the internal implementation or to the design described in the accompanying article, and some designs in that article are future directions not implemented in the prototype README.

The code is MIT-licensed LICENSE, so adaptation is permitted, but the project is best treated as an experimental reference for exploring an autonomous repository agent rather than a supported tool. Development checks run via pnpm check (typecheck, tests, build), which indicates a TypeScript/pnpm toolchain, but the source material does not specify supported Node or pnpm versions or platform requirements beyond the harness dependency.