Skip to content

dsh-superagent

A research prototype of a coding agent that watches one GitHub repo and acts when asked.deepseek-ai/dsh-superagent-prototype

Stars
11k
Forks
1.1k
Licence
MIT
Language
TypeScript
Last update
6 days ago
Type
AI model / agent

What it is

dsh-superagent is a proactive coding agent built for one GitHub repository, packaged as a bundle for DeepSeek Harness. It listens to GitHub and optionally Feishu; when an allowed person asks for work, it opens a work item, works on it in its own session and directory, and replies in the original conversation, for example with a pull request. A web panel shows what needs a person's attention. The README describes it as a research prototype, not an official DeepSeek product, and not production-ready.

  • Have an allowed team member ask the agent in GitHub to work on an issue and receive a pull request
  • Route work requests from Feishu conversations to the agent (the README says Feishu is optional)
  • Review pending items that need a person through the web panel
  • Study the documented design of authorization, execution and prompts in the docs folder

Great for

  • Engineering teams exploring autonomous coding agents on a single GitHub repository
  • Developers studying how agent work items, authorization and a review panel can fit together
  • Researchers reading the accompanying design article alongside working code
  • Operators who want an agent to act only when a team rule or a person allows it

Not ideal for

  • Teams needing production-ready software, since the README warns it may contain bugs and inconsistent practices
  • Anyone wanting a simple install in minutes, since the README says setup has many steps and recommends an agent-guided setup
  • Users who need features described in the article, since the README says some designs are not implemented in this prototype
Step-by-step guide: how to use dsh-superagentInstall, first run, examples, tips and FAQs

How to get it

  1. 1

    Clone and build from source

    Clone the repository

    git clone https://github.com/deepseek-ai/dsh-superagent-prototype

    Run the project checks (typecheck, tests, build)

    pnpm check
  2. 2

    Run it

    • Read the README and the docs in order, starting with docs/000-philosophy.md
    • Follow SETUP.md, ideally with a coding agent that walks you through the steps
    • Configure the GitHub source and optional Feishu source per docs/006-deployment.md
    • Open the web panel to see items that need a person
  3. 3

    Get the most out of it

    • Read docs/006-deployment.md for manual setup, configuration and troubleshooting
    • Use docs/008-cookbook.md to learn how to add a new Source
    • Review docs/002-authorization.md before granting anyone permission to ask for work
    • Read docs/007-architecture.md to understand how the pieces fit before modifying code

Downloads

Strengths

  • Clear stated philosophy of minimizing human attention, documented in docs/000-philosophy.md
  • Ordered documentation set covering domain model, authorization, execution, prompts and architecture
  • Uses the MIT license, which permits reuse with few conditions

Trade-offs

  • The README explicitly labels it research code that should be reviewed and adapted before use
  • The README says it is not identical to the internal version or the article's implementation
  • Setup has many steps and the full install details are not included in the truncated README

How it compares

ProjectStarsTypeLevelUpdatedBest for
dsh-superagent-prototype (this)11kAI model / agentadvanced6 days agoEngineering teams exploring autonomous coding agents on a single GitHub repository
OpenHands92.5kAppintermediate4 days agoDevelopers who want coding agents running on a server while their laptop is off
SWE-agent20.7kAI model / agentintermediate10 days agoResearchers studying autonomous coding agents or running SWE-bench experiments

Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-11-02 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.