dsh-subagent-model

By default, DSH subagents inherit the main agent's model, so while your flagship model handles core reasoning, every delegated retrieval, cleanup, and audit task also runs on the same expensive model. dsh-subagent-model separates the two concerns: the main model stays under DSH's native management, while subagents get their own provider, model, and reasoning-effort assignment that applies globally without touching the session's primary model.

The plugin revolves around three configuration layers: the subagent model decides who does the work, reasoning effort controls how deep the thinking goes, and the subagent profile applies a persona-plus-tool-filter combination that is functionally equivalent to an Agent preset and covers roughly 90 percent of preset behavior. All three draw from the same shared model directory, the menu mirrors the official UI one-to-one, and every selection persists instantly with no save button. On first launch the plugin also registers two ready-made profiles, Quick Execute and Read-only Research, as real user presets, so renaming, editing, or deleting them in Settings syncs back to the plugin menu automatically.

Built for developers who run a flagship model as the main agent and want to offload independent retrieval, code auditing, or document organizing to a cheaper or more specialised model. Once installed, a Subagent button appears to the left of the official model menu; two clicks are all it takes to switch, and the provider/model/effort/profile tag in the subagent tree lets you verify the actual configuration at a glance.