NanmiCoder/dsh-agent-teams
AgentTeams multi-agent teams.
Install
via npm
Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.
via GitHub · npm package
Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.
What happened when we ran it
Installed cleanly when we ran it
Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 11d ago.
Show it in your README
For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.
Due diligence
What it is — Converts a DeepSeek Harness session into a captain-led multi-agent team.
Who it is for — When you need to break a goal into dependency-aware tasks and coordinate sub-agents through direct messages in a DeepSeek Harness session, if your work only involves a single agent without team coordination.
Watch out — Sandbox test passed: installed in a fresh profile and registered by harness. Static check shows it will execute shell commands. No obvious pitfalls found.
The verdict — I would install it because it provides all coordination tools and live UI in one session, but only if using a supported Harness version.
Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.
dsh-agent-teams is a DeepSeek Harness plugin that turns the current session into a captain for a persistent team of coordinated sub-agents. The captain creates role-specific members, divides a goal into tasks with dependencies, and consolidates their reports. Members are continuable DSH sub-agents that can be woken for follow-up work, while direct messages are stored in durable mailboxes.
The plugin uses a shared scheduler and persisted team state. It tracks running, idle, and ready member states, atomically assigns one ready task to each idle member, and retries stranded open attempts after interruption or restart. Members update work with an attempt_id; reassignment revokes stale attempts before a new one begins. Team data is written under <workspace>/.agent-teams/, and the Web UI reads that state alongside live sub-agent activity. By default, member creation snapshots the captain's active provider, model, and reasoning effort. An explicit heterogeneous request can instead provide a member-specific provider and model.
It suits investigations, reviews, and other goals that benefit from parallel roles and a consolidated result. It is a poor fit when one agent can complete the work directly, or when multiple DSH processes must edit the same team: file state is serialised only within one DSH process. The plugin also has a terminal surface. That matters because its sub-agents perform assigned work within the current harness environment and available agent permissions, rather than in an isolated workflow service.
The plugin works with its defaults, or accepts configuration in a trusted profile. The stateDir value is relative to the workspace unless changed by the surrounding profile.
| Key | Example value | Meaning |
|---|---|---|
stateDir |
.agent-teams |
Directory for team state |
memberProvider |
spawn |
Sub-agent runtime backend; spawn or fork |
memberModel |
deepseek-v4 |
Default model for all members |
memberMaxDepth |
1 |
Member depth setting |
maxMembers |
8 |
Maximum member count |
memberProvider is not an LLM provider. Cross-provider routing uses the optional provider and model fields accepted by agent_teams_add_member; memberModel remains a shared default.
After a source build or profile change, dsh --profile web --dump-config validates the composed profile. Start the Web UI with dsh web. Source development uses pnpm build and pnpm verify; pnpm build must be run again after source changes when using a linked local installation.
The current session becomes one captain and can lead one active team. Members claim ready tasks automatically when idle. A reassignment or captain takeover revokes the previous attempt_id and waits for the old worker to quiesce. Messages that cannot be delivered live remain durable and are retried at a later status boundary. The activity panel exposes roles, current work, unread messages, dependencies, and archived history from the persisted state.
An existing DeepSeek Harness installation is required. Team state is file-backed and shared through <workspace>/.agent-teams/; concurrent processes editing the same team are not coordinated. The panel reports persisted state as-is, so a model can finish work without making the expected task-state update. Defaults snapshot the captain's active provider, model, and reasoning effort; per-member model or reasoning prompts are not otherwise provided.
Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub
Same category
Q00
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
loopx-project
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
huangruiteng
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
chuspeeism
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
tong-io
TongFlow film-crew studio for image, voice, music and video production: the agent writes per-asset TongFlow workflow files (.tongflow.json) that run through TongFlow plugins, with an embedded workflow canvas, a shot/character/take project layout and a manga-drama template; sessions starting with @tongflow open the Studio view.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
EthanYoQ
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
No one-line install — this plugin lives inside a larger repository and publishes no npm package.
