DeepSeek has released dsh-superagent, an MIT-licensed research prototype designed to function as a proactive coding agent for GitHub repositories. The new tool uses a router-worker architecture to turn allowed requests into parallel work items, with each task running in its own isolated agent session.
The system operates by having a long-running router handle interactions and dispatch tasks, while isolated worker sessions perform the actual coding work. According to DeepSeek's simulations under noisy message streams, the super-agent achieved a p95 task completion time of 15.7 minutes, compared to 29.5 minutes for a single-session baseline. The release also shipped alongside a design document exploring autonomous agent systems and human attention bottlenecks.
This open-source release matters because it addresses human attention bottlenecks in AI coding workflows by shifting from single-session limits to automated parallel task execution.