DSH AgentTeams
An open-source AI agent coordination plugin for DeepSeek Harness, with local team records, desktop and web views, and support for mixed-model teams.
DSH AgentTeams adds coordinated AI agent teams to DeepSeek Harness for coding, research, and review work. A lead agent divides a goal among specialist agents and combines their results. It works within Harness's desktop, CLI, and web interfaces, and stores team state in your workspace. The plugin is open source under the MIT license.
You review the plan before agents start work. The proposed team and task dependencies remain editable until approval, so you can change roles or model choices before execution. Members can use different providers and models available through Harness, or inherit the lead agent's model settings.
Tasks wait for their prerequisites to finish, and idle agents automatically pick up ready work. Members keep their sessions for follow-up tasks and exchange persistent messages directly with teammates. Interrupted work can resume, while recovery after a process restart retries stranded tasks. Reassignment stops the previous attempt before its replacement begins.
The team view shows progress, task ownership, and an interactive dependency graph. You can open member conversations, and completed teams retain their task and member history. Optional quality gates connect implementation with verification and review, including repair and repeat review when checks fail.
One lead agent manages one active team at a time. Team records are local, while model routing uses the selected Harness provider. Separate Harness processes don't coordinate edits to the same team state.
