Preface¶
The core design philosophy of DeepSeek Harness (DSH) is “everything is a plugin.” p-jiangh/dsh-enhance is a toolkit that provides an enhanced experience for DSH, with a positioning similar to oh-my-zsh theme extensions. By encapsulating skills, injecting work rules, providing preset agents, and offering a series of auxiliary tools, it helps users quickly establish a standardized deep-work workflow in DSH.
Core Capabilities¶
After installation, the plugin mainly provides the following features:
- Bundled skills: Built-in skills such as
code-review,git-workflow, andtdd, which can be invoked directly from the/menu. - Work-rules section: A persistent system prompt section that defines common work discipline.
- Doctor tool:
dsh_enhance_doctor, used to diagnose the plugin assembly state and provide recommendations. - Human instructions: Provides two UI entry points,
/doctor(diagnosis) and/boulder(work tracking status). Results are rendered only and do not consume tokens. - Guards + Delivery-gate: Includes guards such as
write-existing,path-traversal, andbash-read, as well as a file manifest-based delivery gate. - Work tracking: Combines boulder and write tokens to provide
start-workand delivery-gate hooks. - Worktree tools: Provides
worktree_create/list/cleanuptools. - Enhance preset: A working-mode preset named
enhance, including the deep-work protocol (enhance-mode) section. - Advisor: A sub-agent that accompanies review runs and provides suggestions such as nit/concern/blocker.
Installation and Enablement¶
This plugin is mounted per profile. Before installing, ensure that the dsh CLI (npm package @deepseek-ai/dsh) has been installed.
- Confirm your profile name. The desktop GUI uses
desktop, Loopback Web usesweb, and headless mode usesheadless. - Run the installation command:
dsh plugin --profile <your-profile> add https://github.com/P-JIANGH/dsh-enhance/releases/download/v0.1.2/p-jiangh-dsh-enhance-0.1.2.tgz
Notes:
* The package name is @p-jiangh/dsh-enhance; do not write the bare name dsh-enhance.
* If the profile name is incorrect, the command will silently create a new profile with the same name, and the desktop GUI may not show the plugin. It is recommended to check existing profiles first: ls $DSH_HOME/profiles.
* Desktop profiles are exclusively managed by the application; the CLI can only manage web, headless, or custom profiles. Install it from the plugin panel in the desktop GUI.
Typical Usage¶
After installation, the plugin is loaded automatically. You can use it as follows:
- Use skills: Type
/in the input box and selectcode-review,git-workflow, ortdd. - Run diagnostics: Type
/in the input box and select/doctor. - Enable Enhance mode: When creating a new session, select Enhance from the preset menu. This preset includes the persistent
enhance-modesection.
Configuration¶
The plugin supports configuration through the UI or YAML files.
- UI configuration: Find
@p-jiangh/dsh-enhancein the “Plugins” panel of the sidebar and click “Configure”. Toggles such asworkRules,doctorTool,guards, andworkTrackingcan be adjusted, andadvisorModelcan be modified. - YAML configuration: Override it in the profile’s
cordis.patch.ymlor$DSH_HOME/cordis.patch.yml:
- id: dsh-enhance
name: '@p-jiangh/dsh-enhance'
config:
workRules: |
# 你的自定义规则
doctorTool: true
Caveats¶
- Fixed preset: The Enhance preset is fixed when a session is created and cannot be switched to in an existing session. To use this preset, you must create a new session.
- Compatibility: The Enhance preset requires
dsh >= 0.2.0and is available only in the Web or desktop GUI. It cannot be selected from the preset menu in pure headless / acp / sdk environments, but the core functionality of the plugin works normally. - Feature status: Advisor’s intended regression feature has not yet been implemented; it currently exists only as an accompanying review runner.
Summary¶
dsh-enhance lowers the barrier to building complex workflows in DSH by providing prebuilt skills, workflow guards, and a deep-work mode. It is suitable for users who want to quickly reuse code review, Git workflows, or structured deep work in conversations.