dsh-plugin-model-config

The Settings → Models page in DeepSeek Harness already displays a provider's route, credential, and model catalog, but the remaining configuration — thinking mode, reasoning effort, context-window and token defaults, image budgets, file and quota limits, timeouts — has no editable UI. You end up hand-editing the raw settings document.

This plugin renders those settings as inline rows inside each provider's own card: the setting name on the left, a schema-mapped control on the right (number box with min/max/step, Yes/No switch, select for literal unions, or a text box). Nothing is hardcoded. The namespace list comes from the deployment's llm.listConfigurableProviders() and each setting from its serialized schema, so a brand-new adapter or an added field appears as a row with zero code changes.

Writes target only the user layer via settings.mutate with revision tracking; a reset is an unset, not a write-back of the default; a lost revision race is reported rather than silently overwriting. The plugin registers one keyed entry per configurable namespace into the settings.models.provider-card slot and modifies no file in the DeepSeek Harness repository.

Best for: developers running DeepSeek Harness in a web profile who want form-style editing of advanced provider parameters (especially the DeepSeek adapter's thinking mode and context window), and operators integrating third-party adapters who need to verify new configuration fields quickly without diving into JSON.