MIT-licensed dsh agent-preset plugin: 1 coordinator + 17 experts with smart delegation, cross review, a message bus, a taskboard and quality gates — activated by mounting the preset into DSH’s preset discovery directory.
DSH integration
Native runtime
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-10
License
MIT
01What can it help you accomplish?
Turn a DeepSeek Harness workspace into a Chief Coordinator mode where tasks are routed to the right specialist expert
One coordinator plus 17 experts — full-stack, security/data, business — each exposed as a tool (`expert_frontend_dev`, `expert_security`, `expert_product_manager`…) that the coordinator can call, including in parallel
DeepSeek Harness users who want a multi-agent team without writing custom prompts or maintaining several configs
Coordinate several experts on a risky task without them drifting or duplicating work
Quality gates (5-stage pipeline with independent review and a 2-round rework limit), a file-based message bus, a filesystem taskboard with dependency DAG and crash recovery, and an experience pool
Teams running complex, high-risk changes through DeepSeek Harness where a single agent is not enough
02How to install into DeepSeek Harness
Prerequisites
- DeepSeek Harness installed (the preset is selected inside a DSH workspace)
- A writable `~/.dsh/.agent-presets/` directory — DSH’s preset discovery directory, one subdirectory per preset
Installation steps
- 01
Install the package: `dsh plugin add dsh-expert-mode`, or `npm install dsh-expert-mode`
$ dsh plugin add dsh-expert-mode
- 02
Create the preset directory and copy the preset files: `mkdir -p ~/.dsh/.agent-presets/expert-mode` then `cp -r node_modules/dsh-expert-mode/agent.cordis.yml node_modules/dsh-expert-mode/preset.yml node_modules/dsh-expert-mode/cordis.patch.yml ~/.dsh/.agent-presets/expert-mode/`
- 03
Optionally copy the methodology docs too: `cp -r node_modules/dsh-expert-mode/.expert-mode ~/.dsh/.agent-presets/expert-mode/`
- 04
Restart `dsh web`, then select “专家模式” in the workspace preset selector
$ dsh web
Verify the integration
- The preset appears in the workspace preset selector after the restart
- Asking a question routes it to an expert — the README example shows the coordinator identifying “后端开发 + 安全” and delegating to Backend Dev and Security
03DSH integration and capability boundaries
Native dsh agent-preset plugin (not a Cordis service plugin): the npm package ships `agent.cordis.yml`, `preset.yml` and `cordis.patch.yml`, which are copied into DSH’s preset discovery directory `~/.dsh/.agent-presets/` and then selected as “专家模式” in the workspace preset selector.
Smart delegation and fast track
an incoming task→domain identification plus routing to the best expert; complex tasks dispatch several experts in parallel, simple ones are handled by the coordinator directly
experts stay online after a task completes so follow-up modifications do not restart the whole flowInter-expert message bus
messages between experts (`comm/bus.py`)→direct async delivery into `.expert-mode/comm/mailboxes/<expert>/*.msg` with an audit log at `comm/logs/bus.log`, zero coordinator relay
every message is persisted as a `.msg` fileQuality gates and fault recovery
a high-risk task→requirement clarity → implementation → verification → review → integration, with independent-expert review and a 2-round rework limit; auto-retry on timeout and strategy switch on failure
progressive disclosure injects methodology on demand, which the README reports as 28% token savingsExperience pool and cross review
lessons from a finished task; high-risk tasks→lessons saved and injected next time; high-risk tasks get multi-expert independent review
04Who is it for? When not to use it?
Good for
- DeepSeek Harness users who want a multi-agent team without writing custom prompts or maintaining several configs
- Teams running complex, high-risk changes through DeepSeek Harness where a single agent is not enough
Not for
- Installing the npm package is not enough: the preset only activates once its files are mounted into DSH’s preset discovery directory `~/.dsh/.agent-presets/`, and the preset name comes from `preset.yml`’s `name` field.
- Do not add the host-plane `@deepseek-ai/dsh-memory-connect` plugin to this preset’s `agent.cordis.yml`: it injects `sessions` + `systemPrompt`, registering it inside a preset throws `service has been registered at <cross-session-memory>` and the preset fails to mount, falling back to the default preset.
05Compatibility, maintenance and safety notes
- Installing the npm package is not enough: the preset only activates once its files are mounted into DSH’s preset discovery directory `~/.dsh/.agent-presets/`, and the preset name comes from `preset.yml`’s `name` field.
- Do not add the host-plane `@deepseek-ai/dsh-memory-connect` plugin to this preset’s `agent.cordis.yml`: it injects `sessions` + `systemPrompt`, registering it inside a preset throws `service has been registered at <cross-session-memory>` and the preset fails to mount, falling back to the default preset.
- Cross-session memory is not included: the preset deliberately does not register the memory service, and `dsh-memory-connect` has to be installed into the host composition separately.
- The methodology docs are optional: copying only the three YAML files activates the preset, while copying `.expert-mode/` brings the methods, experts and `comm/` bus material along.
MIT · actively maintained (latest release v0.9.2, 2026-08-27) · featured in Awesome DSH Plugin
06Frequently asked questions
Is this a normal Cordis plugin?
No. It is an agent-preset plugin: installing the npm package pulls the files into `node_modules`, but the preset only activates once `agent.cordis.yml`, `preset.yml` and `cordis.patch.yml` are copied into `~/.dsh/.agent-presets/expert-mode/`.
How do I activate it after installing?
Create `~/.dsh/.agent-presets/expert-mode`, copy the three YAML files (and optionally `.expert-mode/` for the methodology docs), restart `dsh web`, then pick “专家模式” in the workspace preset selector.
Does it come with cross-session memory?
No. The preset deliberately does not register the memory service because it is a host-plane plugin; registering it inside a preset throws `service has been registered at <cross-session-memory>` and the preset fails to mount. Install `dsh-memory-connect` into the host composition separately.
How do experts talk to each other?
Through a file-based message bus (`comm/bus.py`): messages land as `.msg` files in `.expert-mode/comm/mailboxes/<expert>/` with an audit log at `comm/logs/bus.log`, so content flows between experts without going through the coordinator’s context.
What stops experts from drifting on a risky task?
A five-anchor constraint with per-turn self-check, cross review by multiple independent experts for high-risk tasks, and a 5-stage quality gate pipeline (requirement clarity → implementation → verification → review → integration) with a 2-round rework limit.
07Related DSH workflows
memos
by memtensor
Self-evolving memory OS for LLM & AI Agents: ultra-persistent memory, hybrid-retrieval, and cross-task skill reuse, with 35.24% token savings and DeepSeek Harness support.
dsh-anchored-standard
by xiaobright
Two-phase DeepSeek Harness preset: Minimal-aligned bootstrap, then full Standard tools (Project2 98/99)
mem9
by mem9-ai
Unlimited memory for OpenClaw
dsh-mnemon
by omdsh-dev
Composable, view-based context for DeepSeek Harness. Pluggable sources and strategies, with layered memory and self-evolving skills out of the box.
08Data and sources
this is an **agent-preset plugin**, not a Cordis service plugin.
`~/.dsh/.agent-presets/` is DSH's preset discovery directory.
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-10. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.