The DeepSeek Harness team pushed dsh-superagent-prototype to GitHub on October 31 under the MIT license, and it had 11,198 stars when we checked this morning. The README is blunt about what it is: a research prototype, 'not an official DeepSeek product', with much of the engineering done by coding agents, and not identical to DeepSeek's internal version.
It's a proactive coding agent for one GitHub repository, packaged as a bundle for DeepSeek Harness (DSH). It listens to GitHub and, if you want, Feishu. When an allowed person mentions the agent account with a request, it opens a ticket, works on it in its own session and directory, and answers in the same thread, for example with a pull request. A web panel lists only what needs a person.
The design rule is in docs/000-philosophy.md: take as little human attention as possible. Work nobody asked for needs a rule the team wrote, and starts in a mode that only records what the agent would have done (the default accept mode is observe). SETUP.md recommends a dedicated GitHub account for the agent, never the operator's own, and no auto-merge until its pull requests have been right for a while.
Read one line in docs/003-execution.md before you run it: executor sessions run commands without approval as the operator's OS user.
We cloned the repo and ran pnpm check on a Linux box (Node 24, pnpm 11.7.0). 1,196 of 1,197 tests passed; one Feishu source test expected an error state and got ready. We didn't run the agent itself, which needs DSH and a model provider.