DSH Crew plugin
A DeepSeek Harness plugin: Claude Code, Codex, Antigravity and Grok dispatch work to DSH agents, progress in the native task panel
- 212 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- DSH workers write files and run commands in the working folder, and the number of confirmations depends on policy settings
- The installer edits Claude Code, Codex, Antigravity and Grok configuration and permissions
- The standalone mode sends tasks to the DeepSeek API with your key
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add plugins/dsh-crewDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- DSH workers write files and run commands in the working folder, and the number of confirmations depends on policy settings
- The installer edits Claude Code, Codex, Antigravity and Grok configuration and permissions
- The standalone mode sends tasks to the DeepSeek API with your key
This is third-party code. Review the repository files before installing.
What it does
DSH Crew adds a set of MCP tools to DeepSeek Harness through which the main agent hands work to DSH workers. The orchestrator keeps its own model while the task runs on a real DSH agent with its tools. The `flash` tier suits mechanical work, `pro` suits reasoning, and `effort` goes from `off` to `max`. In the host task panel workers show up as regular subagents: dispatch count, current step, tool calls and tokens. Guardrails run before every dispatch: worker nesting is capped at depth 3, cycles are refused and a second worker does not start in a busy working folder. There is a jobs board, a tier policy that retries on `pro` after a `flash` failure and image description through DeepSeek's vision model.
Who it is for. People who already use DeepSeek Harness and want to launch its agents from Claude Code or Codex without losing the familiar subagent interface.
Good fit when
- Routine edits should go to the cheap `flash` tier while hard analysis stays on `pro`
- Several workers must run in parallel and their progress shows in the Claude Code panel
- A failed `flash` run should be retried once automatically on `pro`
- An image must be described while the main model has no vision
Not a fit when
- DeepSeek Harness is not in use: the plugin works as an add-on to it
- Tasks are small and the overhead of a separate worker does not pay off
- You want no third-party keys: the standalone mode needs a DeepSeek API key
Example request
Give the auth module refactoring to a flash-tier worker, run test analysis on pro in parallel, then show the job status.Limitations
The plugin installs into a DSH profile from npm with `dsh plugin --profile web add @zseven-w/dsh-crew@latest`, and the README version 0.1.0-rc.11 is marked as a release candidate. The standalone mode without a running DSH needs a DeepSeek API key in `~/.config/dsh-crew/.env`. The installer connects the hosts: it writes the Claude Code, Codex, Antigravity and Grok configuration and edits permissions. The project was tested with DSH 0.1.1-rc.1. The latest commit in the repository is from November 2026.
How to disable. Run `node src/install/cli.mjs uninstall-claude` for Claude Code, and likewise for `codex`, `agy` and `grok`, then remove the plugin from the DSH profile. The `~/.config/dsh-crew/.env` file with the key is deleted by hand.
FAQ
How do I install DSH Crew?
With `dsh plugin --profile web add @zseven-w/dsh-crew@latest` and then `dsh web`. The Claude Code integration comes from `node src/install/cli.mjs claude` or a button in the plugin settings.
Which MCP tools are there?
Six: `dsh_run_worker` for a blocking run, `dsh_spawn_worker` for an async one, `dsh_worker_status`, `dsh_worker_result`, `dsh_worker_cancel` and `dsh_worker_config`.
What is the difference between flash and pro?
They are capability tiers of a DSH agent: `flash` for mechanical work, `pro` for reasoning. They are backed by DeepSeek V4 Flash and V4 Pro today, and DSH configuration sets the roster.
Is a separate key needed?
Not inside a running DSH, where workers use its settings. Only the standalone mode without DSH reads `DEEPSEEK_API_KEY` from `~/.config/dsh-crew/.env`.
How do I check that it works?
Run `node scripts/smoke.mjs`: within about ten seconds the script dispatches one cheap job and prints `smoke test passed`.
Discussion
No questions or reviews yet.
Related
PluginsA DeepSeek Harness plugin for a standing team of agents with memory, projects, task owners and verification by a second member
dsh-baton hands tasks from Claude Code or Codex to a local DeepSeek Harness and brings the result back with checks.
A DeepSeek Harness plugin that makes the agent change approach after failures and verify results before finishing.
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 plugin opening DeepSeek Harness, Codex, Cursor and Claude Code workspaces from a phone or browser over encrypted P2P.
A CLI that keeps an existing agent on a long task: plan a step, run it, verify the result in the real environment and checkpoint progress
README in short
The README explains why the plugin exists and shows six capability blocks: native progress, tier policy, in-host DSH sessions, vision, guardrails and a jobs board. Installation, verification, the MCP tool list and translations follow.