DeepSeek releases open-source coding agent that watches GitHub repos and steps in with fixes

DeepSeek released an open-source coding agent prototype on Saturday that watches a team's GitHub repository and steps in with fixes without waiting for a prompt. The company's Harness team announced the agent, called Super-agent, in a WeChat post and published the code under the MIT license - a move Bloomberg framed as a direct challenge to Anthropic's Claude Code.

The repository, dsh-superagent on GitHub, passed 4,000 stars within a day. DeepSeek labels it a research prototype "not an official DeepSeek product" and warns it is not production-ready. The public version also differs from what the team runs internally.

How the agent handles work without being asked

Super-agent is built as a bundle for DeepSeek Harness, the company's agent scaffolding. It listens to one GitHub repository and, optionally, Feishu group chats. When an approved team member asks for work, the agent opens a work item, handles it in its own session and working directory, and replies in the original conversation - for example, with a pull request. A local web panel lists only items that need a human decision.

The design splits the agent into two layers. A persistent "Router" talks to people and keeps shared context, while separate "Worker" sessions run tasks in parallel. This means the agent can take a new request before the last one finishes. A Worker that owns a pull request follows its comments directly, and a Watchdog process monitors for failures and stalled loops.

The project's stated rule is to "take as little human attention as possible, and give as much value as possible." Proactive work that nobody asked for still needs a rule the team writes, and the agent starts in a mode that only logs what it would have done rather than acting.

Setup requirements and the security warning

Setup needs a dedicated GitHub user account for the agent, Node.js 22.19 or later, GitHub CLI 2.48 or later, and a model provider configured in DeepSeek Harness. Only github.com works - GitHub Enterprise Server is not supported. The agent's sessions run with full access as the operator's OS user, and DeepSeek recommends a separate OS account for isolation.

That warning carries weight after recent supply-chain incidents around AI coding tools, including a tampered Claude Code extension that planted an SSH backdoor on developers' machines.

The broader agent race between Chinese and US labs

Chinese and US labs are now competing on agents, not just chatbots. Anthropic has been tightening how Claude Code can be used, including a new rule against recreating existing games. US labs have also accused Chinese rivals of copying their models - a recent report on GPT-6 Astra distillation named DeepSeek directly.

DeepSeek created its Harness team earlier this year to build agents that can compete with US coding services, and it has raised about 100 billion yuan ($15 billion) ahead of a planned IPO, according to Bloomberg. For now, Super-agent is a blueprint developers can study and test, not a finished product. Teams exploring AI Code Generation Courses will find the architecture instructive, particularly the split Router-Worker design that enables parallel task handling.

Why this matters for developers and product teams

Super-agent signals where coding assistants are headed: from tools you prompt to agents that watch your repos and act on their own. For development teams, the MIT license means the scaffolding is free to inspect, fork, and adapt - even if it is not ready for production today. The two-layer design offers a practical pattern for anyone building internal automation that needs to juggle multiple requests without blocking on each one. Operations and IT leads should note the security recommendation to isolate the agent under its own OS account, a precaution that will only grow more relevant as AI Agent Courses and real-world deployments push these tools deeper into development pipelines.