dsh-prompt-presets

A local prompt preset plugin for DeepSeek Harness. It stores reusable prompt templates, supports {{variable}} placeholders, and renders a finished prompt when the Agent supplies values.

Preset data stays in the current working directory under .dsh/prompt-presets/presets.json. The plugin does not send templates or variables over the network.

Quick install

Requirements

You need Node.js 22 or newer, pnpm, and a DeepSeek Harness installation that supports JavaScript or TypeScript plugins.

Step 1: Download and build

Copy and run:

git clone https://github.com/fff122/dsh-prompt-presets.git
cd dsh-prompt-presets
pnpm install
pnpm build

Step 2: Create the plugin configuration

Run this command from the repository root:

cat > prompt-presets.patch.yml <<EOF
- insert:
    - id: dsh-prompt-presets
      name: '$PWD/dist/src/index.js'
EOF

The plugin path is absolute because the Harness patch loader requires an absolute path.

Step 3: Start Harness with the plugin

If Harness is installed as a command:

dsh web --patch "$PWD/prompt-presets.patch.yml"

Otherwise, run:

npx @deepseek-ai/dsh web --patch "$PWD/prompt-presets.patch.yml"

Verify the installation

Ask the Agent:

保存一个名为 meeting-summary 的提示词模板：总结 {{topic}}，面向 {{audience}}，语气保持简洁。标签用 work。

Then ask:

应用 meeting-summary，topic 是产品发布会，audience 是工程团队。

If the plugin is loaded, the Agent can use preset_save and preset_apply to save and render the template.

Available tools

Template behavior

Variable names use letters, digits, and underscores, and must start with a letter or underscore. Spaces inside braces are allowed, so {{topic}} and {{ topic }} refer to the same variable. Variables are returned in first-seen order.

If a variable is missing, preset_apply leaves its placeholder in the rendered text and lists the name in missing. This makes it possible to review an incomplete prompt without losing the original placeholder.

Strings are inserted as-is. Numbers and booleans use their text form, while arrays and objects are inserted as compact JSON.

Updating the plugin

From the cloned repository:

git pull
pnpm install
pnpm build

Restart Harness with the same patch command after rebuilding.

Development

Run the complete package quality gate from the repository root:

pnpm run format:check
pnpm run typecheck
pnpm run test
pnpm run build

The implementation is split into small modules:

src/
  index.ts          # Harness entry point and tool registration
  preset-schema.ts  # Preset types, validation, and template rendering
  preset-store.ts   # Local persistence, filtering, updating, and deletion

test/
  index.test.ts
  mount.test.ts
  preset-schema.test.ts
  preset-store.test.ts

For the plugin model and patch overlay format, see the official DeepSeek Harness plugin guide.