GitHub radar
An agent that works on tasks in your GitHub repo on its own
This is a DeepSeek prototype — a proactive agent for one GitHub repository: it takes a request in chat, does the work itself, and answers with a pull request, staying quiet the rest of the time.
It's a service agent that watches issues and discussions in a GitHub repo (and optionally a Feishu chat) and does the work itself when asked.
When it helps
Useful if you want to automate routine code changes without constant back-and-forth with a bot in chat. Skip it if you need something reliable for daily work — this is experimental research code.
Pros
- Acts on its own, no constant chat back-and-forth needed
- Works in an isolated environment and answers with a pull request
- Panel shows every task waiting on a human decision
- The 'stay quiet unless needed' pattern is worth copying regardless
Cons
- Runs on DeepSeek's own models, which lag top models in practice
- It's a long-running service, not a one-off script
- Coding agents wrote most of it — code quality is uneven
How to set it up — step by step
- 1Make sure DeepSeek Harness is already running and you have access to the target GitHub repo (and Feishu, if you want it).
- 2Open your coding agent in your project folder and give it access to the repository.
- 3Send the agent the text from the block below.
- 4Give the agent keys and tokens if it asks for them during setup.
- 5Check the result: have the agent open a work item on request and show it answering with a pull request.
Text for your agent
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy dsh-superagent from https://github.com/deepseek-ai/dsh-superagent-prototype — I have access to the GitHub repository and to DeepSeek Harness (ask me for keys and tokens if you need them along the way), follow SETUP.md, and show me how the agent opens a work item on my request and answers with a pull request.
This isn't something you run on a laptop and forget — it needs an already configured DeepSeek Harness, access to GitHub and optionally Feishu, and per the docs the service itself is meant to run as a persistent deployment, not a one-off script. Setup is involved, but its own SETUP.md walks you through it step by step.
Open on GitHub▌ More finds