@klarkxy/dsh-model-hub 0.1.7 → 0.2.0

package/README.md

@@ -1,10 +1,10 @@
-# Model Hub
+# Subagents
 
 <!-- release-state-note -->
 
 [简体中文](docs/README.zh-CN.md)
 
-View and change the models your DeepSeek Harness plugins use, control which models subagents may use, and track the model and reasoning effort of each running subagent — all on one page.
+Configure the models and reasoning effort available to new DeepSeek Harness subagents, and see each subagent's last-used model in the conversation's subagent popup.
 
 ## Install
 
@@ -16,20 +16,17 @@ Replace `web` with your profile name and restart the profile.
 
 ## Use
 
-Open **Plugins → Model Hub**.
+Open **Settings → Subagents** to add, replace, or remove models, choose their reasoning effort, and add optional task tendencies. These guide the agent when delegating; the agent makes the final choice.
 
-- **Fields**: every model setting exposed by installed plugins, such as Auto Title, Git Commit, and Safe Auto. Search by plugin, purpose, or current model, then pick a new model. The plugin that owns the setting saves the change.
-- **Subagent Cast**: toggle the subagent model allowlist and add, replace, or remove models. When enabled with at least one model, it guides the agent to discover the models available in the current conversation and pick one explicitly when delegating. The guidance distinguishes provider and model names to help avoid repeating the provider prefix. For independent reviews and alternative approaches it favors a different model when suitable, but allows the same one when it fits better. Per-model task tendencies you add are optional hints; none are supplied by default. Below the list, a live view shows each running subagent with its current model and reasoning effort; an effort marked adapter default comes from the adapter, not an explicit choice.
+In a conversation, open the subagent count or switcher to see each child's last-used model and effort. This is read-only.
 
 ## Limits
 
-- Only plugins that expose model settings appear.
-- Changes made elsewhere update the page automatically. If an edit conflicts with a newer change, refresh and choose again.
-- If the model list cannot load, use **Retry**. The last available list stays visible; warnings name providers that failed while others stay usable.
-- Selection guidance needs host tools that support choosing a subagent model. List changes apply to new conversations; existing conversations and their subagents keep their saved lists. Refreshing the settings page does not change a conversation's list. Omitting an explicit model can still inherit the parent model.
-- Guidance never forces a model choice or adds subtasks just for variety.
-- Remote subagents and finished runs have no readable route; the live list covers only live local subagents.
-- No usage statistics.
+- List changes apply to new conversations; existing conversations keep their saved lists. Effort preferences guide new subagents and do not change running ones.
+- Selection requires host tools that support choosing a subagent model. Omitting a model still inherits the parent model.
+- Model lists use the host's model and effort metadata. If loading fails, use **Retry**; available providers remain usable.
+- The popup uses the host's private tree and style hooks, so host changes may require a plugin update. External subagents and children without a model request have no route to show.
+- Settings changed elsewhere update automatically; refresh to resolve a conflicting edit.
 
 ## License
 

package/docs/README.zh-CN.md

@@ -1,10 +1,10 @@
-# 模型枢纽
+# 子代理
 
 <!-- release-state-note -->
 
 [English](../README.md)
 
-在一个页面查看并修改各 DeepSeek Harness 插件使用的模型，管理子代理可用的模型，并跟踪运行中子代理的模型与思考强度。
+配置新建 DeepSeek Harness 子代理可用的模型与思考强度，并在会话的子代理弹层中查看每个子代理最后使用的模型。
 
 ## 安装
 
@@ -16,20 +16,17 @@ dsh plugin --profile web add @klarkxy/dsh-model-hub
 
 ## 使用
 
-打开「插件 → 模型枢纽」。
+打开「设置 → 子代理」，添加、替换或移除模型，选择思考强度，并填写可选的任务倾向。这些内容供智能体委派时参考，最终选择仍由智能体决定。
 
-- **字段总览**：列出已安装插件提供的模型设置，例如自动标题、Git 提交和安全自动。可按插件、用途或当前模型搜索，再选择新模型；改动由对应插件保存。
-- **子代理阵容**：开关子代理模型白名单，添加、替换或移除模型。开启且至少有一个模型时，会引导智能体发现当前会话可用的模型，并在派发时显式选择。提示会区分提供方与模型名称，帮助避免重复填写提供方前缀。独立复核和方案对比会优先考虑合适的不同模型，同模型更适合时仍可使用。自填的模型「任务倾向」只作可选参考，插件不预置任何倾向。列表下方实时显示运行中的子代理及其当前模型和思考强度；标注「适配器默认」的强度来自适配器，而非显式选择。
+在会话中点开子代理数量或切换器，可查看每个子代理最后使用的模型与思考强度。此处只读。
 
 ## 限制
 
-- 只显示提供模型设置的插件。
-- 别处改动设置时页面自动更新；编辑与较新的改动冲突时，刷新后重新选择。
-- 模型列表读取失败时点「重试」。上次可用的列表保留；部分提供方读取失败会显示提示，其余提供方仍可用。
-- 选模引导需要宿主工具支持选择子代理模型。名单修改对新建会话生效，已有会话及其子代理沿用原名单；刷新设置页面不会更新会话名单。不显式指定模型时仍可继承父代理模型。
-- 引导不强制选模，也不为增加模型种类而新增子任务。
-- 远程子代理和已结束的子代理没有可读取的路由；运行列表只覆盖存活的本地子代理。
-- 不提供用量统计。
+- 名单修改对新建会话生效，已有会话沿用原名单。思考强度用于新建子代理，不修改正在运行的子代理。
+- 选模需要宿主工具支持。省略模型时仍继承父代理模型。
+- 模型与强度选项来自宿主目录。读取失败时点「重试」，其他可用提供方不受影响。
+- 弹层依赖宿主私有的树与样式约定，宿主改动可能需要插件适配。外部子代理和未发起模型请求的子代理没有可显示的路由。
+- 别处改动会自动同步；编辑冲突时刷新后重新选择。
 
 ## 许可证