Your DeepSeek chat just became a project manager

What it does

dsh-agent-teams promotes your current DeepSeek Harness session to a team captain and spawns durable sub-agents to work in parallel. You describe a goal in natural language; the plugin splits it into tasks, wires up dependencies, and runs a scheduler that wakes idle members to claim ready work. A live Web UI tracks the roster, an interactive task DAG, and progress, while state persists to disk so cold restarts can retry stranded attempts.

The interesting bit

Rather than immediately burning tokens, the system stages an unspawned roster and task graph first, letting you edit member routes and dependencies in the UI before approving execution. Members communicate through durable direct mailbox messages instead of routing everything through the captain, and the scheduler safely reassigns work by revoking stale attempts before new ones start.

Key highlights

• Captain-led delegation with continuable sub-agents that can be woken for focused follow-up turns
• Dependency-aware task scheduling that prevents claims until prerequisites finish
• Automatic member reuse and cold-recovery retry for stranded work after process restarts
• Live activity panel with an interactive task DAG, collapsible roster, and segmented progress
• Optional quality gates with explicit requirements → implementation → verification → review → integration contracts

Caveats

• One captain can lead only one active team at a time
• Team state is file-backed and serialized within a single DSH process, so concurrent processes editing the same team are not coordinated
• Models may occasionally finish work without updating the expected task state, leaving the activity panel out of sync

Verdict

Useful if you already run DeepSeek Harness and want to parallelize complex work across specialized sub-agents without a separate workflow engine. If you are not using DSH, this is an elaborate plugin you cannot use.