Why it matters: DeepSeek is publishing its own design for an agent that takes requests from team conversations and works on them in the background, while labeling the code research only and not production-ready.
AI developer DeepSeek has published the code for an experimental coding agent, dsh-superagent, on its official GitHub account. GitHub's records show the repository was created on Oct 31 with a single commit, and the code is released under the MIT license.
What it is: the project's README describes dsh-superagent as a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness, the company's agent software, which DeepSeek's product page says is in public preview worldwide and open source. The agent listens to GitHub and, optionally, to chats in the workplace messaging app Feishu. When an allowed person asks it for work, it opens a work item, handles the task in its own agent session and directory, and replies in the original conversation, for example with a pull request. A web panel shows the items that need a person, according to the README.
How it decides to act: DeepSeek's design notes say the agent acts only when asked or when a rule the team wrote allows it, and that work nobody asked for starts in a mode that only records what the agent would have done. The authorization document tells teams to start in that observe mode and switch to automatic acceptance only after reviewing the recorded decisions. The setup guide recommends giving the agent a dedicated GitHub account and never the operator's own, because everything it writes appears under that name.
What it is not: DeepSeek labels the code a research prototype for reference only and says it is not an official DeepSeek product or production-ready software. The README says much of the engineering was done by coding agents, that the published code is not identical to DeepSeek's internal version, and that some designs in a companion article, "Exploring Super-agent Design: Towards a More Autonomous Agent System," are future directions that the prototype does not include.
The descriptions of how the agent works are DeepSeek's own. Funded Tech has not tested the software.
Sources
- DeepSeek on GitHub: dsh-superagent-prototype README, Oct 31, 2026
- DeepSeek on GitHub: dsh-superagent-prototype, "000 Philosophy"
- DeepSeek on GitHub: dsh-superagent-prototype, "002 Authorization"
- DeepSeek on GitHub: dsh-superagent-prototype, SETUP
- DeepSeek: DeepSeek Harness product page
How we checked this: We read the dsh-superagent-prototype repository on DeepSeek's official GitHub account (README, philosophy, authorization and setup documents, license, and GitHub's repository, commit and language data) and DeepSeek's Harness product page, in copies saved on Nov 1, 2026 at about 8:12 AM ET, and checked every date, feature and label against them.
See something wrong? Tell us and we'll fix it. Every fix is dated and listed on our Corrections page.