DeepSeek open-sources Super-Agent, a coding agent that watches team chats and opens pull requests

DeepSeek has published the code for a team coding agent called Super-Agent on GitHub under an MIT license, The Next Web reported on November 1. The repository, dsh-superagent-prototype, describes a proactive agent for a single GitHub repository, packaged as a bundle for DeepSeek Harness, the open-source agent app the company released in August. The team behind its agent tools described the design in a WeChat post. Bloomberg's Debby Wu reported that the agent is aimed at Anthropic's Claude Code and can watch group chats for coding problems and suggest fixes without being asked.

How it works

The agent listens to its target GitHub repository and, optionally, to Feishu, ByteDance's workplace chat app. When an authorized person asks for work, it opens a work item, does the work in its own agent session and directory, and replies in the original conversation, for example with a pull request. A web panel lists items that need a person.

The README frames the design around limiting interruptions. Its stated principle is to "take as little human attention as possible, and give as much value as possible." It warns that "an agent that can act for a whole team can also waste a whole team's attention," so the agent "acts only when asked or when a rule the team wrote allows it" and "speaks only when someone must act or can use a result." Setup is meant to be done by a coding agent following a SETUP.md file, and the project uses pnpm.

What DeepSeek says it is not

DeepSeek calls the release a research prototype for reference only and "not an official DeepSeek product." The README tells users to "treat it as research code, not production-ready software," says it may contain bugs and inconsistent engineering practices, and asks people to review and adapt it before running it. The published code differs from DeepSeek's internal version, and some ideas in the accompanying article are future directions that are not built yet. The README does not say which models the agent requires.

Why it matters

A typical coding agent session starts with one developer's prompt. This one is designed to sit in a team's channels and decide when to step in, which moves the main cost from tokens to people's time reviewing what it produces. The design choices DeepSeek made, an allowlist of who can assign work and team-written rules for when it can act alone, are a reasonable checklist for any team agent you evaluate. The legal context matters too: The Next Web notes that US civilian federal agencies were told last month to remove DeepSeek products, so organizations with federal exposure should check their own policy before running it.