Flash

Flash·Apps & Agents·2026-11-02 08:04

DeepSeek Open-Sources Super-Agent Prototype: It Listens to Repos and Chats, Acts Only When Allowed

DeepSeek's Harness team published dsh-superagent-prototype on GitHub on Oct 31 under the MIT license: a proactive coding-agent prototype for a single repository that listens to GitHub and Feishu and answers with a pull request once allowed.

A laptop showing code and a team chat in a sunny studio
DeepSeek super-agent prototype (AI-generated illustration)

What happened

The prototype has two layers: a resident router takes conversations and dispatches work, while isolated workers execute in their own sessions and directories and send results back to the original conversation, for example as a pull request. A companion web panel shows work items that need a person, and the design principle is to take as little human attention as possible: it acts only on an allowed request or under a rule the team wrote, and speaks only when someone must act or can use a result.

The repository statement draws the line up front: this is a research prototype for reference only, not an official DeepSeek product and not production-ready software, it may still contain bugs and its engineering style may be inconsistent, so users should review and adapt it before use. The published code was adapted for release and is not identical to the team's internal version or to the implementation in the design article 'Exploring Super-agent Design: Towards a More Autonomous Agent System'; some designs in the article are future directions not implemented in the prototype.

In a companion article, the team said its internal super-agent already reproduces and fixes bugs, responds to code reviews and helps build DeepSeek Harness. In the team's own simulation, scaled at about 300 messages and 20 work requests, the median completion time was 12.4 minutes against 13.8 to 20.4 minutes for a single-session baseline, but each run consumed about 23.2 million cached tokens against 6.1 to 10.8 million for the baseline; the figures come from the team's own test, not an independent evaluation.

Key facts

  1. Publisher and date: DeepSeek's Harness team, published on GitHub on Oct 31, 2026
  2. Repository: deepseek-ai/dsh-superagent-prototype, created Oct 31, 2026 at 09:32 UTC, MIT license
  3. Positioning: a proactive coding agent for a single GitHub repository, packaged as a bundle for DeepSeek Harness
  4. How it works: listens to GitHub and optionally Feishu; on an allowed request or under a team rule it opens a work item, works in its own session and directory, and answers in the original conversation, for example with a pull request
  5. Architecture: a resident router dispatches work, isolated workers execute, and a web panel shows items needing a person
  6. Limits: a research prototype for reference only, not an official product and not production-ready; the published code differs from the internal version and the design article's implementation

Why it matters

coding agents move from waiting for instructions to picking up work proactively, and the router-worker split becomes a reusable open-source reference, but the prototype is explicitly not a finished product, so team rules and human review remain gates before deployment.
Useful Tap if this story helped you

SourcesGitHub (Oct 31, 2026); MadRobot (Nov 1, 2026). This article is compiled from public information and does not constitute investment advice.

From HIMIAO Horizon · himiao.ai

Follow HIMIAO Horizon

No sign-up · Unsubscribe any time

Email

Notifications

A ping on your phone or computer when the brief is out. No email needed.

Comments

  1. Loading comments…
Ask the cat