chr003/dsh-subagent-model-picker0

dsh-subagent-model-policy

Per-session subagent model picker for DeepSeek Harness with immutable child bindings

AI 분석

允许为每个会话独立指定子智能体的模型、服务商及推理等级。适合需要精细控制子任务模型消耗与能力的复杂 Agent 编排任务,需 DSH Web 0.1.2-rc.1+。

패키지
dsh-subagent-model-policy
버전
0.1.0
라이선스
MIT
최근 업데이트
2026. 9. 7.

설치

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:chr003/dsh-subagent-model-picker

Usage

Open Sub: and choose a provider, model and supported reasoning effort. Changes save immediately and affect new subagents only. Inherited keeps DSH's normal selection—usually the parent model, but explicit model/provider options from a spawning tool still apply. An explicit Sub: selection overrides those options.

Children with successfully saved bindings keep their captured model and effort across continuation and restart. Newly created descendants use the owning root conversation's latest explicit policy, or DSH's normal selection when it is Inherited. Children created before installation remain unmanaged.

Ordinary conversation forks have independent settings. Child conversation views are read-only. Save and validation failures are shown rather than silently switching managed children to another model.

Known limitation: If a new child's initial binding save fails, restarting DSH can leave that child unmanaged and using DSH's normal model choice instead. Fix storage and create a new child; see recovery guidance.