dsh-smart-subagent

Adds a SEPARATE smart_subagent tool with a model-facing model parameter: the model picks the subagent's provider/model per call. No automatic routing. The built-in subagent / subagent_fork tools are untouched.

• smart_subagent gains:
  • model (optional, enum of all currently available models, values like deepseek-official/deepseek-v4-pro); omit to inherit this conversation's model.
  • reasoning_effort (optional: off/minimal/low/medium/high/xhigh/max; applies only when the chosen model supports it).
• Foreground and background (continuable) children both run on the chosen model; follow-ups via send_message keep it.
• The model enum mirrors the composer picker (registered providers' models) and refreshes on llm/adapters-updated.
• A short system-prompt section explains the parameter; every new conversation sees it automatically.

Install

dsh plugin --profile web add "link:<space-free-path>/dsh-smart-subagent"

Restart the web profile afterwards.