Is DeepSeek's Super-Agent an official product?

No. The repository's own README calls it a research prototype for reference only, not an official DeepSeek product, adapted for publication and not identical to DeepSeek's internal version. Some designs described in DeepSeek's accompanying article are future directions not implemented in the prototype.

Does dsh-superagent act without being asked?

Only within rules the team writes. Its authorization model lets an allowed person who mentions or messages the agent make an explicit request; bots on an allowed list can only produce inferred candidates, never explicit requests; and messages, comments and quoted documents cannot change permissions. It is designed to speak only when someone must act or can use a result.

Is dsh-superagent safe to run?

Treat it as research code. Its documentation says executor sessions run commands without approval as the operator's operating-system user. Run it in an isolated machine or container with a scoped GitHub token, restrict the allowed-person list, and review every pull request before merging.

Quick Answer

What Is DeepSeek dsh-superagent? Open-Source Coding Agent

Q: What is DeepSeek dsh-superagent?

dsh-superagent is an open-source research prototype from DeepSeek's Harness team, published on GitHub as deepseek-ai/dsh-superagent-prototype on October 31, 2026 under the MIT licence. It is a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness: it listens to GitHub and optionally Feishu, opens a work item when an allowed person asks for work, does the work in its own agent session and directory, and replies in the original conversation, for example with a pull request.

Published: November 4, 2026

The short answer

dsh-superagent is DeepSeek’s open-source “super-agent” prototype: a coding agent that sits on one GitHub repository, takes requests from allowed teammates in GitHub or Feishu, and answers with work such as pull requests. DeepSeek published it on October 31, 2026 at deepseek-ai/dsh-superagent-prototype under the MIT licence; written in TypeScript, it had about 17,800 GitHub stars by November 4, 2026. DeepSeek labels it a research prototype, not an official product.

It runs as a bundle for DeepSeek Harness, DeepSeek’s agent runtime.

What it does

Listens to a GitHub repository and, optionally, Feishu (Lark) chats.
Opens a work item (a “Ticket”) when an allowed person asks the agent for work by mention or direct message.
Works in its own session and directory per Ticket, using an Executor session named worker-<ticket id>.
Replies where it was asked, for example with a pull request.
Shows a web panel listing what needs a human: answers, reviews, blocked or failed work.

Its design rule, in the README’s words: take as little human attention as possible, and give as much value as possible. It acts only when asked or when a team-written rule allows, and speaks only when someone must act or can use a result.

How it decides what to do

The authorization document splits incoming messages in two:

Eval (starting work): an allowed person addressing the agent makes an explicit request. Bots on an allowed list can produce only inferred candidates, never explicit requests; other bots are ignored.
Subscriptions (continuing work): every message on a subscribed thread — including CI and bot messages — reaches the Ticket as context but grants no new authority. Each Ticket stays limited by its “Mandate”.

Crucially, messages, comments and quoted documents cannot change permissions — a defence against prompt injection through issue text.

The risk to understand first

The execution document states that Executor sessions run commands without approval as the operator’s OS user, and long shell commands move to background jobs after 5 seconds. Anyone you allow to request work can therefore cause code to run on that machine. Run it in an isolated VM or container, with a GitHub token scoped to the one repository, and require human review before merge.

How it compares

	dsh-superagent	Claude Code / Codex CLI	GitHub-native coding agents
Trigger	Teammate request in GitHub or Feishu; team rules	Developer at a terminal	Assigning an issue or mention
Scope	One repository, many requesters	Your local checkout	Repositories in an org
Licence	MIT, self-hosted	Proprietary CLI (Claude Code); open-source CLI (Codex)	Hosted service
Status	Research prototype	Supported products	Supported products
Best for	Studying team-level proactive agents	Daily individual coding	Teams wanting a supported, hosted agent

For production team agents today, compare supported options in best Claude Code alternatives and DeepSeek Harness vs OpenCode vs Claude Code.

How to try it

The README’s setup is agent-driven: you ask your own coding agent to “Read SETUP.md in this repository and set up dsh-superagent with me.” It asks one question at a time, shows the full plan, then installs and configures. The numbered docs (000 Philosophy through 008 Cookbook) explain the domain model, authorization, execution, prompts, panel, deployment and architecture; pnpm check runs typecheck, tests and build.

Who should care

Platform teams designing how an agent shares one repository with many people — the authorization model is the most reusable part.
DeepSeek Harness users who want a team-facing bundle.
Not teams looking for a supported product: there is no SLA, and the code may be inconsistent by DeepSeek’s own description.

Last verified: November 4, 2026.