NanmiCoder/dsh-agent-teams
AgentTeams plugin for DeepSeek Harness
dsh-agent-teams turns the current DeepSeek Harness session into a captain that can assemble a durable multi-agent team — all from natural language. The plugin provides ten coordination tools, persistent state, an automatic shared-task scheduler, and a live Web UI inside DSH. One prompt: 'Use AgentTeams to review the commits after v0.5.3 from performance, security, and product perspectives. Return one consolidated report.' The captain creates a team, splits the goal into dependency-aware tasks, assigns roles to sub-agents (continuable DSH sub-sessions), coordinates via direct messages, and returns one consolidated result. The scheduler uses real running/idle/ready state — tasks cannot be claimed before dependencies finish; stranded tasks are automatically retried. The live activity panel shows segmented progress, a collapsible roster, and an interactive task DAG. Completed teams are archived with full member and task history.
Install
dsh plugin --profile web add @nanmicoder/dsh-agent-teamsRequires an existing DSH installation. After install: dsh --profile web --dump-config (to validate), then dsh web. Restart DSH and refresh the Web UI. From source: git clone, pnpm install, pnpm build, then dsh plugin --profile web add . (local checkout).
Compatibility
Works with DSH web profile. Requires an existing DeepSeek Harness installation. State stored under <workspace>/.agent-teams/.
Details
- Repo: NanmiCoder/dsh-agent-teams
- Category: Coding & Development
- Stars: 686
- Version: see npm @nanmicoder/dsh-agent-teams for latest
- Last push: 2026-09-17
- First seen: 2026-08-12
Recent updates
Release notes at github.com/NanmiCoder/dsh-agent-teams/releases/latest. Cold recovery retries open attempts after process restart. Direct mailbox messaging between members (no relay).
FAQ
- How do I install dsh-agent-teams?
- Run: dsh plugin --profile web add @nanmicoder/dsh-agent-teams — then validate the profile (dsh --profile web --dump-config), restart DSH, and refresh the Web UI.
- How do I start a multi-agent team?
- Just ask in natural language: 'Use AgentTeams to [your goal] from [perspective A], [perspective B], and [perspective C]. Return one consolidated report.' The captain creates the team, assigns tasks, and coordinates without a separate workflow engine.
- What happens if DSH restarts mid-task?
- Team state is persisted under <workspace>/.agent-teams/. The shared scheduler auto-retries open tasks (cold recovery): any stranded attempt is re-queued and a fresh attempt starts from the last consistent state.
Alternatives
yjh051108/dsh-routing-suite · foryourhealth111-pixel/Vibe-Skills · csyangwen/dsh-memory-evolve