dsh-agent-team plugin
A DeepSeek Harness plugin for a standing team of agents with memory, projects, task owners and verification by a second member
- 92 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- Managed agent sessions run with danger-full-access in their workspace and act without constant human supervision.
Install
Manual install
dsh plugin --profile web add @contexera/dsh-agent-team@0.2.7Command from the README, run while DSH is stopped. Pin the DSH version too: `npm i -g @deepseek-ai/dsh@0.2.2-rc.1`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Managed agent sessions run with danger-full-access in their workspace and act without constant human supervision.
This is third-party code. Review the repository files before installing.
What it does
dsh-agent-team adds a Team mode to DeepSeek Harness (DSH) where work goes to a team of agents while the human sets direction and holds the gates. Each agent becomes a named Member with its own memory, notes and skills: after a DSH restart or a context swap it stays the same Member and does not need briefing again. Projects have owners and every task has one accountable owner, human or agent. Only someone who did not build a task can accept it: an agent accepts at the review stage and only if it never held a claim on it. Agents pull from their own inbox instead of receiving every message, and a reply to a thread that has moved on is returned with a description of what changed. The Main stream, threads and tasks live in one ledger, and the task board and inbox are views of it. Installation goes into the `web` profile: `dsh plugin --profile web add @contexera/dsh-agent-team@0.2.7`. The plugin is opt-in and does not change the Harness installation.
Who it is for. DeepSeek Harness users who run long parallel lines of work and want agents to keep memory, projects and acceptance without constant human involvement.
Good fit when
- The work splits into lines that move in parallel: development, bug fixing, UI, review
- Work should keep moving while you are away and you want to read the outcome on a task board without asking
- Verification should be done by someone other than the agent who built the result
- Recurring work deserves a standing owner: a project provides one
- Agents need to be briefed once and later rely on their own memory and notes
Not a fit when
- You want to hold the thread yourself: exploring, deep design or a change you want to watch line by line fit an ordinary session better
- Several people need to share one team: only one human and one DSH server are supported now
- You need timer-based runs: agents wake on a message, a mention, a hand-off or a reminder, and a fully quiet team does not wake itself
Example request
Create a project called Release 2.0 in Team mode, make an agent the owner, add a task to prepare the changelog and ask a second member to accept it after review.Limitations
The release is verified against DSH 0.2.2-rc.1, so install the DSH version explicitly instead of following `latest`, and for other DSH lines the author names matching plugin versions. Managed agent sessions run with `danger-full-access` in their workspace, so create agents only in a workspace you trust. DSH is started with the compiled `dsh web` command: running from source gives the error "selected preset is not team-enabled". One human and one machine are supported, and nothing runs while the machine is off. The former package name `@wowyuarm/dsh-agent-team` stays at 0.2.1 and is deprecated.
How to disable. Run `dsh plugin --profile web remove @contexera/dsh-agent-team`: it removes the package from the profile together with its layers. Team mode disappears afterwards and the Harness settings stay as they were.
FAQ
How do I install dsh-agent-team?
Stop DSH and run `dsh plugin --profile web add @contexera/dsh-agent-team@0.2.7`. Pin the version explicitly: without it the package manager may pick the previous release.
How do I check that the plugin is connected?
`dsh --profile web --dump-config` should list `contexera-agent-team-scope` and `contexera-agent-team-client` rows. In the browser a Team mode appears in the DSH navigation.
Who accepts a task?
Only someone who did not build it. An agent may accept a task at the review stage if it never held a claim on it. Close, reopen and transfer stay with the accountable owner and the human.
How does it differ from similarly named plugins?
The others assemble a team around a session, a workflow run or a leader-led group and dissolve it when the job ends. Here the team is standing: members keep memory and project owners keep their projects.
Why do agents get `danger-full-access`?
The Team Member preset intentionally gives managed agent sessions full access in their workspace. Create agents only in a workspace you trust.
Discussion
No questions or reviews yet.
Related
PluginsA DeepSeek Harness plugin: Claude Code, Codex, Antigravity and Grok dispatch work to DSH agents, progress in the native task panel
dsh-baton hands tasks from Claude Code or Codex to a local DeepSeek Harness and brings the result back with checks.
DeepSeek Harness is an open-source AI agent from DeepSeek where every feature is a plugin. It runs in the browser, on macOS and on Windows.
A DeepSeek Harness plugin that makes the agent change approach after failures and verify results before finishing.
A plugin opening DeepSeek Harness, Codex, Cursor and Claude Code workspaces from a phone or browser over encrypted P2P.
A DeepSeek Harness plugin that imports chat history from Claude Code, Codex, Cursor and other agents to continue in DSH.
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| dsh-agent-team (this card) | no | 92 | |
| DeepSeek Harness | yes | 279,346 | DeepSeek Harness itself; dsh-agent-team adds a standing agent team mode to it |
| dsh-auto-review | no | 248 | an automatic review plugin for DSH; this one is a whole team with projects, task owners and independent acceptance |
README in short
The README explains the difference between working in the loop and on the loop, the principles of a standing team, current limits, a quick start with a pinned DSH version and a comparison with three similarly named plugins.