{"version":1,"type":"story","url":"https://digestai.news/story/deepseek-harness-team-releases-open-source-super-agent-prototype","json":"https://digestai.news/story/deepseek-harness-team-releases-open-source-super-agent-prototype.json","markdown":"https://digestai.news/story/deepseek-harness-team-releases-open-source-super-agent-prototype.md","slug":"deepseek-harness-team-releases-open-source-super-agent-prototype","headline":"DeepSeek Harness team releases open-source Super-agent prototype","summary":"The DeepSeek Harness team has published the underlying architecture for their Super-agent as an open-source prototype on GitHub. The release is explicitly labeled as a research prototype for reference only, not an official DeepSeek product. The code originated from the team's internal development work, with much of the engineering performed by coding agents. The team warns that the software is not production-ready, may contain bugs, and has inconsistent engineering practices. Users are advised to review and adapt the code before using it in their own environments.\n\nThe prototype, named dsh-superagent, functions as a proactive coding agent for a single GitHub repository. It operates as a bundle for DeepSeek Harness and listens to GitHub and, optionally, Feishu. When an authorized user requests work, the agent opens a work item, processes it in its own session, and responds in the original conversation, often with a pull request. A web panel displays items requiring human attention. The design philosophy aims to minimize human attention while maximizing value, ensuring the agent acts only when asked or when permitted by team-written rules.\n\nThe published code is adapted for public release and is not identical to the internal version or the implementation described in the paper \"Exploring Super-agent Design: Towards a More Autonomous Agent System.\" Some designs mentioned in the paper are future directions and are not implemented in this prototype. Setup is complex, so the repository includes a SETUP.md file designed to guide a coding agent through the installation and configuration process step-by-step.","keyPoints":["DeepSeek Harness team released dsh-superagent as an open-source research prototype on GitHub.","The agent listens to GitHub and Feishu, handling work items and responding with pull requests.","The code is not production-ready and differs from the internal version described in their research paper."],"whyItMatters":"This release provides insight into DeepSeek's approach to autonomous coding agents, offering a concrete implementation of proactive, low-attention agent workflows for developers to study and adapt.","category":{"slug":"agents","name":"Agents & Tools","url":"https://digestai.news/category/agents"},"entities":{"companies":["DeepSeek"],"models":[],"people":[]},"firstPublishedAt":"2026-10-31T23:32:03Z","updatedAt":"2026-10-31T23:32:03Z","sourceCount":1,"hasPrimarySource":true,"sources":[{"outlet":"github.com","title":"DeepSeek Harness team published the underlying architecture for their Super-agent, an always-on autonomous agent, as an open-source prototype on GitHub.","url":"https://github.com/deepseek-ai/dsh-superagent-prototype","publishedAt":"2026-10-31T23:32:03Z","type":"primary","primary":true,"lead":true}],"sourceNotes":null,"discussions":[{"site":"Reddit","url":"https://www.reddit.com/r/singularity/comments/1xkoi9g/deepseek_harness_team_published_the_underlying/","points":null}],"thread":null,"cite":{"text":"Digest AI, \"DeepSeek Harness team releases open-source Super-agent prototype\", 31 October 2026, https://digestai.news/story/deepseek-harness-team-releases-open-source-super-agent-prototype","publisher":"Digest AI","title":"DeepSeek Harness team releases open-source Super-agent prototype","datePublished":"2026-10-31T23:32:03Z","url":"https://digestai.news/story/deepseek-harness-team-releases-open-source-super-agent-prototype"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}