2507483326/dsh-eteams

dsh-eteams (branded ETeams) is a multi-agent team orchestration plugin purpose-built for DeepSeek Harness, distributed under the npm package name dsh-eteams and registered with the plugin id eteams. The project ships as a dual-end single bundle covering both the host runtime and the browser-side renderer, with all build-time dependencies inlined into the final artifact so there is no separate runtime install step. Users can install it through the DeepSeek Harness in-app plugin marketplace by searching for ETeams, or mount it from the terminal with dsh plugin --profile desktop add dsh-eteams; source-tree developers can instead clone the repository, run pnpm install && pnpm build, and register the local folder with the same dsh plugin --profile desktop add . invocation. The README badge explicitly tags the project as a DeepSeek Harness-plugin, and there is no mention of Claude Code, Codex, Cursor, Gemini CLI, or any other agent CLI as a supported host.

The intended workflow is structured around three explicit steps. First, the user invokes the /eteam slash command inside an active DeepSeek Harness conversation; an interview-style dialogue then walks the user through generating a role handbook for each team member, which is persisted to a global role library where every character effectively becomes a job card. Second, the user opens the panel's "Team" page, picks candidates from that role library to assemble the roster, and designates exactly one of them as the leader. Third, the user describes the goal in conversation; the leader decomposes it into ordered tasks, the user approves the plan, and the system automatically dispatches each task along the execution chain, while every member works inside its own independent sub-session and can be assigned a different per-role chat model so heterogeneous models can cooperate on the same goal. Progress, failures, and decisions stream back to the panel's kanban, member, team, and task-progress screens in real time, giving the operator full visibility and live re-dispatch control rather than a single opaque model call.

Operationally, the plugin requires Node.js ^22.19 || >=24, declared directly in package.json, and all persistent state is stored locally via the built-in node:sqlite driver against a local SQLite file, so no network calls or telemetry uploads are performed. The package is dual-licensed under MIT, and the only prerequisite for first run is to confirm the harness still recognizes the bundle after install by executing dsh --profile desktop --dump-config and reloading the desktop client so leader and member roles register correctly inside the DeepSeek Harness session environment.