WJZ-P/dsh-model-capabilities

🎛️ dsh-model-capability

Choose a model's supported input modalities directly in DeepSeek Harness.

简体中文 · English

---

Features

• Adds an Input modalities selector to native Harness model rows.
• Supports provider defaults, text, image, and text-plus-image declarations.
• Writes the standard model input field used by Harness attachment checks.
• Extends settings.models.model.fields without taking over validation or persistence.
• Follows Harness light/dark theme tokens and English/Chinese locale state.
• Uses the standard DSH bundle and Web client discovery contracts without Tauri APIs.

Example

The model editor below explicitly declares Text + image input, allowing Harness to treat image attachments as supported model input:

Modality mapping

Selection	Stored model value
Inherit provider default	input omitted
Text	["text"]
Text + image	["text", "image"]
Image	["image"]

Install

dsh plugin --profile web add dsh-model-capability
dsh --profile web --dump-config
dsh --profile web

Remove the plugin:

dsh plugin --profile web remove dsh-model-capability

Development

pnpm install
pnpm run build
pnpm test
npm pack --dry-run

The current release targets DeepSeek Harness 0.1.0-rc.5, its standard Web client discovery path, and the settings.models.model.fields extension slot.

Marketplace

The ready-to-copy catalog entry, pull-request checklist, and screenshot naming convention are documented in MARKETPLACE.md. Recommended screenshots show the input-modality field and its expanded option list; the marketplace entry references GitHub-hosted image URLs.

Security

The plugin updates only the current model row's input field through Harness's public settings slot. See SECURITY.md for supported versions, issue scope, and private-reporting guidance.

License

MIT