diff --git a/docs/guides/tasks.md b/docs/guides/tasks.md
index beb667fbf..8e3f6c0d6 100644
--- a/docs/guides/tasks.md
+++ b/docs/guides/tasks.md
@@ -64,7 +64,7 @@ Construct, operate and destroy the runtime and group on the same thread. `manual
 
 Each manager owns one library-provided `manual_runtime`. After registering a module, call `manager.create_task_group(module, maximum)` during initialization, then submit work through `manager.get_tasks(module)`. Capacity must be positive; unknown modules, duplicate groups and creation after shutdown begins are rejected. A module pointer can be registered only once. Group references remain stable when other modules register.
 
-Private database bridges use `manager.get_task_runtime()` to share that exact executor. A module's `update()` calls `service.update()` and `bridge.dispatch()` first; `manager.update(timestamp, maximum_handlers = 64)` then polls the runtime once. All groups share the handler budget. Modules must not also poll their group or runtime; a handler budget is not a frame-time limit. Published database plugins have not switched to these private bridges.
+Private database bridges use `manager.get_task_runtime()` to share that exact executor. A module's `update()` calls `service.update()` and `bridge.dispatch()` first; `manager.update(timestamp, maximum_handlers = 64)` then polls the runtime once. All groups share the handler budget. Modules must not also poll their group or runtime; a handler budget is not a frame-time limit. The public MySQL and Redis plugins use these bridges internally.
 
 `manager.stop(timestamp)` stops top-level business-task admission in every managed group and returns `false` without waiting while tasks or an active update/completion stack remain. Accepted coroutines may submit later operations, so keep their services and bridges accepting. Only another `stop()` after task completion and stack unwinding enters shutdown hooks. It then executes `pre_shut()`, `shut()` and unloading synchronously; user hooks may block, so this API does not bound shutdown duration.
 
@@ -81,7 +81,7 @@ Use the manager and its groups on their constructing thread. Do not mutate modul
 
 <Aside type="caution" title="Task completion does not release native IO resources">
 
-This drain check covers only top-level business tasks accepted by managed groups. Modules do not yet define a common phase for stopping incoming requests. External callbacks must return, connection-pool workers must exit and background cleanup must complete before their code can be safely unloaded. A zero task count does not prove those conditions or activate the public native database plugins. Unmanaged work must not reference a module being destroyed.
+This drain check covers only top-level business tasks accepted by managed groups. Modules do not yet define a common phase for stopping incoming requests. External callbacks must return, connection-pool workers must exit and background cleanup must complete before their code can be safely unloaded. A zero task count does not prove those conditions. Unmanaged work must not reference a module being destroyed.
 
 </Aside>
 
@@ -122,6 +122,7 @@ All plugins must be rebuilt against the same updated manager headers; this chang
 
 - [Create a module](modules.md): lifecycle callbacks and borrowed module interfaces.
 - [Create a plugin](plugins.md): plugin registration, build and runtime selection.
-- [Database coroutine tests](../../tests/database/COROUTINES.md): private native-operation bridges, owner update order and explicit test teardown; public database plugins are still separate work.
+- [Database coroutine tests](../../tests/database/COROUTINES.md): private native-operation bridges, owner update order and explicit test teardown.
+- [MySQL](../reference/mysql.md) and [Redis](../reference/redis.md): public plugin configuration, shared business executor and native pool retirement.
 
 Code deriving directly from the scheduler helper base should use `ark::scheduler_task`. `ark::task<T>` denotes an Asio coroutine; `ark::scheduler` scheduling calls keep their existing signatures.
diff --git a/docs/guides/tasks_CN.md b/docs/guides/tasks_CN.md
index 0c2b2f2af..9f93bdaf1 100644
--- a/docs/guides/tasks_CN.md
+++ b/docs/guides/tasks_CN.md
@@ -64,7 +64,7 @@ runtime 和任务组必须在同一线程构造、操作和销毁。`manual_runt
 
 每个管理器拥有一个网络库提供的 `manual_runtime`。在模块注册后的初始化阶段调用 `manager.create_task_group(module, maximum)`，之后用 `manager.get_tasks(module)` 提交业务任务。组上限必须大于零；未注册模块、重复创建或关闭开始后创建都会拒绝。一个模块指针只能注册一次，组引用在其他模块注册时保持稳定。
 
-私有数据库桥接使用 `manager.get_task_runtime()`，从而与业务任务使用同一执行器。模块的 `update()` 先调用 `service.update()` 和 `bridge.dispatch()`；随后 `manager.update(timestamp, maximum_handlers = 64)` 统一轮询一次。多个组共用该 handler 上限。模块不再自行调用组或 runtime 的 `poll()`，handler 上限也不表示单帧耗时上限。公开数据库插件尚未切换到这些私有桥接接口。
+私有数据库桥接使用 `manager.get_task_runtime()`，从而与业务任务使用同一执行器。模块的 `update()` 先调用 `service.update()` 和 `bridge.dispatch()`；随后 `manager.update(timestamp, maximum_handlers = 64)` 统一轮询一次。多个组共用该 handler 上限。模块不再自行调用组或 runtime 的 `poll()`，handler 上限也不表示单帧耗时上限。公开 MySQL 与 Redis 插件在内部使用这些桥接接口。
 
 `manager.stop(timestamp)` 先停止所有模块组受理顶层业务任务，仍有任务或正在执行更新/完成回调时，不等待它们完成，直接返回 `false`。已受理协程可以继续发起后续操作，因此此时不能关闭其 service 或 bridge 的准入。任务完成且外层回调返回后，再次调用 `stop()` 才同步执行 `pre_shut()`、`shut()` 和卸载。用户钩子可能阻塞，因此该接口不保证停机耗时上限。
 
@@ -81,7 +81,7 @@ runtime 和任务组必须在同一线程构造、操作和销毁。`manual_runt
 
 <Aside type="caution" title="任务完成不等于 IO 资源已释放">
 
-此排空检查只覆盖管理器组中受理的顶层业务任务。模块尚未统一定义停止接收新请求的阶段；还需等待外部回调返回、连接池线程退出和后台清理完成，才能安全卸载相关代码。任务数归零不证明这些条件已满足，也不代表本轮已经迁移公开数据库插件。不要让未受组管理的工作引用即将销毁的模块。
+此排空检查只覆盖管理器组中受理的顶层业务任务。模块尚未统一定义停止接收新请求的阶段；还需等待外部回调返回、连接池线程退出和后台清理完成，才能安全卸载相关代码。任务数归零不证明这些条件已满足。不要让未受组管理的工作引用即将销毁的模块。
 
 </Aside>
 
@@ -124,4 +124,5 @@ runtime 和任务组必须在同一线程构造、操作和销毁。`manual_runt
 
 - [创建模块](modules_CN.md)：生命周期回调与借用的模块接口。
 - [创建插件](plugins_CN.md)：插件注册、构建与运行时选择。
-- [数据库协程测试](../../tests/database/COROUTINES_CN.md)：内部原生操作桥接、所属线程更新顺序与显式测试清理；公开数据库插件迁移仍是独立工作。
+- [数据库协程测试](../../tests/database/COROUTINES_CN.md)：内部原生操作桥接、所属线程更新顺序与显式测试清理。
+- [MySQL](../reference/mysql_CN.md) 与 [Redis](../reference/redis_CN.md)：公开插件配置、共享业务执行器与原生连接池退役。