dsh-agent-teams

When you face a complex goal in DeepSeek Harness that requires multiple perspectives working in parallel—say, reviewing a set of commits from performance, security, and product angles at once—a single session often lacks a structured way to divide and coordinate the work. dsh-agent-teams fills exactly that gap: it turns the current session into a captain that, with one natural-language prompt, assembles a team of durable sub-agents, splits the goal into dependency-aware sub-tasks, lets members proceed in parallel through a shared scheduler, and delivers a single consolidated report.

The coordination protocol is the heart of the plugin. The captain creates the team, assigns roles, and consolidates the final result. Members are continuable DSH sub-agents that exchange durable mailbox messages directly with teammates or the captain—no relay needed. Tasks carry explicit states and dependency constraints; idle members are automatically routed to ready tasks, and reassignment revokes stale attempts before new work begins. A live Web panel shows an interactive task DAG, a collapsible roster, and real-time activity, supporting a plan-first, approve, then-execute workflow as well as optional quality gates spanning requirements, implementation, verification, review, and integration. All state is persisted to disk, so the scheduler can recover stranded work after a cold restart. The plugin ships with 13 native coordination tools and an automatic shared-task scheduler, requiring no external workflow engine.

It is well suited for anyone who needs to orchestrate multiple roles within a single session: multi-perspective code reviews, cross-module refactors, parallel drafting and proofreading, or any task graph where sub-workloads have explicit dependencies. Zero-configuration defaults make it ready to use out of the box, and configuration knobs let you tune member depth, runtime backend routing, and reasoning effort when more control is needed.