---
type: "post"
id: "deepseek-open-sources-super-agent-a-coding-agent-2c70ef"
slug: "deepseek-open-sources-super-agent-a-coding-agent-2c70ef"
title: "DeepSeek open-sources Super-Agent, a coding agent that reads team chats"
description: "DeepSeek has released dsh-superagent, an MIT-licensed research prototype that can take coding tasks from approved team-chat requests and work against one GitHub repository."
retrieval_nugget: "DeepSeek has released dsh-superagent, an MIT-licensed research prototype that can take coding tasks from approved team-chat requests and work against one GitHub repository. It matters because the design shifts coding agents from individual prompting toward bounded team workflows—but the public code is explicitly experimental and incomplete."
published_at: "2026-11-06"
updated_at: "2026-11-06"
record_date: "2026-11-06"
date_kind: "published_at"
topics: ["github","anthropic","claude"]
entities: ["DeepSeek","dsh-superagent","Anthropic Claude Code","Feishu","GitHub Copilot","CooperBench"]
related_patterns: ["goal-scoped-loops-replace-manual-continuation"]
novelty_label: "same_event"
what_changed: null
prior_refs: ["https://newruntime.com/posts/deepseek-open-sources-dsh-superagent-a-research-prototype-524cb5/"]
source_url: "https://thenextweb.com/news/deepseek-super-agent-open-source-coding-agent"
source_urls: ["https://thenextweb.com/news/deepseek-super-agent-open-source-coding-agent"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:deepseek-open-sources-super-agent-a-coding-agent-2c70ef"
status: "published"
visuals: []
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/deepseek-open-sources-super-agent-a-coding-agent-2c70ef/","markdown":"https://newruntime.com/posts/deepseek-open-sources-super-agent-a-coding-agent-2c70ef.md","json":"https://newruntime.com/posts/deepseek-open-sources-super-agent-a-coding-agent-2c70ef.json"}
markdown_url: "https://newruntime.com/posts/deepseek-open-sources-super-agent-a-coding-agent-2c70ef.md"
---

# DeepSeek open-sources Super-Agent, a coding agent that reads team chats

## Retrieval answer

DeepSeek has released dsh-superagent, an MIT-licensed research prototype that can take coding tasks from approved team-chat requests and work against one GitHub repository. It matters because the design shifts coding agents from individual prompting toward bounded team workflows—but the public code is explicitly experimental and incomplete.

## Summary
DeepSeek has released dsh-superagent, an MIT-licensed research prototype that can take coding tasks from approved team-chat requests and work against one GitHub repository. It matters because the design shifts coding agents from individual prompting toward bounded team workflows—but the public code is explicitly experimental and incomplete.

## What happened
The Next Web reports that DeepSeek released code for Super-Agent on GitHub under the name dsh-superagent and an MIT licence. DeepSeek’s agent-tools team described the design in a WeChat post on Saturday, according to the article. DeepSeek calls the release a research prototype and says it is not an official DeepSeek product. The project is aimed at coding work for a software team; The Next Web says Bloomberg characterized it as targeting Anthropic’s Claude Code.

The prototype works with one GitHub repository and can listen to Feishu, ByteDance’s workplace chat app. When an approved person asks it to do work, it opens a task and handles it in its own session. It then responds in the original thread, with a pull request given as an example. A web panel lists items that need human attention. The stated design goal is to reduce the effort of coordinating multiple AI agents and checking their progress. DeepSeek’s manual summarizes the goal as taking as little human attention as possible while providing as much value as possible.

The project’s philosophy document, as described by The Next Web, limits action to cases where the agent is asked or where a team’s own rule permits it. The article says DeepSeek warns that an agent acting for an entire team could consume the attention of that entire team if it behaves poorly. DeepSeek’s README says the public code differs from its internal version and that some ideas described in the article have not yet been built. The project advises users to review and adapt the code before running it. The source does not give a price; the code is released under MIT terms.

The release arrives amid efforts by software companies to run agents across repositories and sessions. The article says GitHub showed a Copilot app at GitHub Universe that runs agent work across several repositories in parallel sessions. It also cites CooperBench research finding that coding agents in teams had average success rates 30% lower than agents working alone; that study did not test DeepSeek’s agent. DeepSeek’s August preview of a coding-agent rival is also mentioned, but the released prototype described here is a separate codebase and workflow.

## What the archive adds
New Runtime’s earlier coverage described dsh-superagent as an MIT-licensed research prototype of a proactive coding agent for DeepSeek Harness ([October 31, 2026 note](https://newruntime.com/posts/deepseek-open-sources-dsh-superagent-a-research-prototype-524cb5/)). The current report adds operational detail about its repository scope, Feishu input, task sessions and human-review panel; it is a new fact on a known story, not a separate launch.

## Trend signal
This supports the assumption that scheduled, resumable and subagent-based runs will take a growing share of coding-agent use beyond interactive single-chat sessions. DeepSeek’s team-scoped task handling is a concrete instance, though the prototype does not establish adoption or production reliability ([Goal-scoped agent loops](https://newruntime.com/shifts/goal-scoped-agent-loops/)).

## Why now
The Next Web places the release alongside competing efforts to coordinate work across agents and repositories, including GitHub’s Copilot app announcement at GitHub Universe. The timing reflects an active product question: how to delegate multiple coding tasks without making humans spend their time supervising every agent.

## Connections to other events
New Runtime’s earlier coverage described DS Bot as a third-party multi-agent plugin for DeepSeek Harness, with long-lived bots, bot-to-bot messaging and group chats ([November 1, 2026 note](https://newruntime.com/posts/ds-bot-is-a-multi-agent-plugin-for-fc4e19/)). Both stories explore team-scale interaction around agents, but they concern distinct projects; the current release is DeepSeek’s own research prototype.

## Why it matters
For a team evaluating coding agents, Super-Agent makes the unit of delegation more concrete: a request can arrive in a work chat, be handled in a repository-specific session, and return to the thread where the work began. That can reduce context switching if the agent’s permissions, task boundaries and review path match the team’s existing practices. The human-attention framing is also a useful test for any agent workflow: automation that creates noisy notifications or unclear ownership may cost more attention than it saves.

The prototype does not yet demonstrate that this workflow is reliable enough for production. Teams should treat it as an architecture to inspect, not a ready-made deployment. Its one-repository scope and explicit human-review surface are useful constraints to assess before considering broader delegation. Updated coverage: [DeepSeek open-sources dsh-superagent, a research prototype of a proactive coding agent for DeepSeek Harness](https://newruntime.com/posts/deepseek-open-sources-dsh-superagent-a-research-prototype-524cb5/).

## Limits and uncertainty
The release is explicitly a research prototype, not an official DeepSeek product, and DeepSeek says its public code differs from its internal version. Some ideas discussed around the project are not implemented. The article does not establish how reliably the agent completes tasks, how its permissions are enforced in deployment, or how much human attention it saves in real teams. The cited 30% team-performance drop comes from CooperBench and did not test this agent, so it is context rather than a result about Super-Agent.

## Reader takeaway
Super-Agent is worth examining as a bounded team workflow: chat request in, repository task in a separate session, result back in the original thread, with humans surfaced for review. But it remains experimental. The practical question is not whether teams can add another agent, but whether the workflow delegates useful work without creating a new layer of supervision.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Github](https://newruntime.com/topics/github.md): Explore the github topic hub.
- topic hub: [Anthropic](https://newruntime.com/topics/anthropic.md): Explore the anthropic topic hub.
- topic hub: [Claude](https://newruntime.com/topics/claude.md): Explore the claude topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Sources](https://newruntime.com/sources/): See the original sources this archive cites.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)