dsh-submodel-change

When an AI session spawns a subagent in DeepSeek Harness, that child is locked to the parent model route. If your workflow needs a reasoning model for deep planning and a lighter model for simple lookups, dsh-submodel-change steps in at exactly the right moment: the instant the AI decides to launch a subagent, a model-selection card appears so you can pick any registered provider and model, or simply follow the parent default, with no manual configuration required.

After your first choice, the plugin keeps things quiet. Subsequent subagents in the same parent conversation reuse that decision automatically, and nested subagents inherit the route through the built-in resolveChildAgentOptions path, so you never see a cascade of prompts. The implementation is dependency-free, written in plain ESM JavaScript, and hooks into the official agent/request waterfall LlmCallConfig replacement point, touching nothing outside in-process spawn subagents and introducing no build step.

It is built for advanced Harness Desktop users who routinely orchestrate multi-model subagents and want per-task routing without config files. Let the reasoner handle complex planning, let a lightweight model grind through simple queries, and keep the UI unobtrusive: install, pick once per parent, and the route holds for the entire chain.