Merge remote-tracking branch 'origin/master' into fix/continuable-subagent-policy-inheritance

# Conflicts: # docs/event-producer-consumer.i18n.yaml # docs/event-producer-consumer.md # docs/event-producer-consumer.zh.md
2026-08-10 14:36:08 +08:00
parent 64e0fbfd6d 71ac961da4
commit 3639b10c83
552 changed files with 24672 additions and 1153 deletions
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/README.md
-README.md: 3cfbb5154713046846a3bfcb2ccea62c0e4cb6c0
+README.md: d3a8943a78238d974d54028e38b773e932429b0e
 README.zh.md: 4b3a1ee57ea61a8e8ba4d01cf7c719bbf8440e30
@@ -2,7 +2,7 @@

 English | [中文](README.zh.md)

-One kind of design doc lives here. An **Agent Note** records a decision or proposal that shapes this codebase — the *why* and *what we gave up*, the parts code and docs can't carry. This file is the front door and contract: where Agent Notes live, when to write one, and [the in-file format](#the-file-format).
+One kind of design doc lives here. An **Agent Note** records a decision or proposal that shapes this codebase — the *why* and *what we gave up*, the parts code and docs can't carry. This file is the entry point and contract: where Agent Notes live, when to write one, and [the in-file format](#the-file-format).

 ## Layout and naming

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-15-llm-model-catalog-and-acp-selection.md
-2026-07-15-llm-model-catalog-and-acp-selection.md: 77f8e379e2b07ecf4e67fa7197752543cfedd6dd
-2026-07-15-llm-model-catalog-and-acp-selection.zh.md: ce4a04a66e345834bc2b16b743be89dc7a0b9424
+2026-07-15-llm-model-catalog-and-acp-selection.md: bfd17c73b01319c10d5dc03333b3c726db5d6f33
+2026-07-15-llm-model-catalog-and-acp-selection.zh.md: aeddada5591bb2da2c0861acc368516eff148172
@@ -26,17 +26,17 @@ Catalog membership is advisory. It drives selectors and diagnostics but never ch

 `dsh-llm-pi-ai` maps the configured provider's installed `getModels(provider)` entries into the neutral catalog. Its existing request-time catalog lookup remains authoritative and still rejects unknown models with `UNKNOWN_MODEL`. `dsh-llm-deepseek` accepts an optional `models` config containing display entries, defaulting to `deepseek-v4-flash` named `DeepSeek-V4-Flash` and `deepseek-v4-pro` named `DeepSeek-V4-Pro`. An explicit list replaces those defaults and an empty list disables discovery. The entries improve selector UX for known public or private models, while every unlisted model id continues to pass through unchanged.

-### Per-session selection in the front door
+### Per-session selection in the front end

-A selection is owned by the front door that offers it (today the TUI `/model` selector), never by `LlmService` or `AgentOptions`: those are deployment-wide or creation-wide objects, and mutating them would couple concurrent sessions. Each opaque choice carries the full provider/model pair, because the same model id may appear under multiple routes.
+A selection is owned by the front end that offers it (today the TUI `/model` selector), never by `LlmService` or `AgentOptions`: those are deployment-wide or creation-wide objects, and mutating them would couple concurrent sessions. Each opaque choice carries the full provider/model pair, because the same model id may appear under multiple routes.

 The ACP automation transport is not a catalog consumer. Its deployment config supplies one optional provider/model target for newly created agents, and it advertises no model selector or configuration-option interface.

 ### Prompt/request consistency and durability

-`installModelSelection` (in `dsh-agent`) installs scoped `system-prompt/assemble` and `agent/request` listeners for a front-door-owned selection. Prompt assembly snapshots the selected pair once per step, overwrites the assembled `provider` and `model` variables after downstream prompt listeners, and the request listener applies that same snapshot after downstream request listeners. A selection during asynchronous assembly therefore starts on the next step rather than splitting prompt text from routing. Other call-config fields remain untouched.
+`installModelSelection` (in `dsh-agent`) installs scoped `system-prompt/assemble` and `agent/request` listeners for a front-end-owned selection. Prompt assembly snapshots the selected pair once per step, overwrites the assembled `provider` and `model` variables after downstream prompt listeners, and the request listener applies that same snapshot after downstream request listeners. A selection during asynchronous assembly therefore starts on the next step rather than splitting prompt text from routing. Other call-config fields remain untouched.

-The request header remains the durable source of truth. When a selection is actually used, the existing full `request/header` snapshot records it, and a front door initializes its selection from the folded last request header before falling back to creation options. A selection that is never used by a request is intentionally in-memory only because it never became model-visible state.
+The request header remains the durable source of truth. When a selection is actually used, the existing full `request/header` snapshot records it, and a front end initializes its selection from the folded last request header before falling back to creation options. A selection that is never used by a request is intentionally in-memory only because it never became model-visible state.

 ## Alternatives considered

@@ -26,17 +26,17 @@ ACP 选择还必须保留提供方维度。同一个模型 ID 可能存在于多

 `dsh-llm-pi-ai` 将已配置提供方的安装目录 `getModels(provider)` 映射为中立目录。其现有请求时目录查询仍是权威依据，未知模型仍以 `UNKNOWN_MODEL` 失败。`dsh-llm-deepseek` 接受可选的 `models` 配置作为展示条目，默认包含名为 `DeepSeek-V4-Flash` 的 `deepseek-v4-flash` 和名为 `DeepSeek-V4-Pro` 的 `deepseek-v4-pro`。显式列表会替换这些默认值，空列表则关闭发现。这些条目改善已知公开或私有模型的选择体验，而所有未列出的模型 ID 仍会原样透传。

-### 前门内的会话级选择
+### 前端内的会话级选择

-选择由提供它的前门拥有（今天是 TUI 的 `/model` 选择器），而不由 `LlmService` 或 `AgentOptions` 拥有：它们是部署级或创建级对象，改动它们会把并发会话耦合在一起。每个不透明选项都携带完整的提供方／模型对，因为同一模型 ID 可能出现在多个路由下。
+选择由提供它的前端拥有（今天是 TUI 的 `/model` 选择器），而不由 `LlmService` 或 `AgentOptions` 拥有：它们是部署级或创建级对象，改动它们会把并发会话耦合在一起。每个不透明选项都携带完整的提供方／模型对，因为同一模型 ID 可能出现在多个路由下。

 ACP 自动化传输层不是目录消费方。它通过部署配置为新创建的 agent 提供一个可选的提供方／模型目标，不展示模型选择器或配置选项接口。

 ### 提示词／请求一致性与持久化

-`installModelSelection`（位于 `dsh-agent`）为前门拥有的选择安装 agent 作用域的 `system-prompt/assemble` 与 `agent/request` 监听器。提示词组装在每个步骤对所选组合做一次快照，在下游提示词监听器之后覆写组装出的 `provider` 与 `model` 变量；请求监听器在下游请求监听器之后应用同一快照。因此，发生在异步组装期间的选择会从下一个步骤生效，而不会让提示词文本与路由分裂。其他调用配置字段保持不变。
+`installModelSelection`（位于 `dsh-agent`）为前端拥有的选择安装 agent 作用域的 `system-prompt/assemble` 与 `agent/request` 监听器。提示词组装在每个步骤对所选组合做一次快照，在下游提示词监听器之后覆写组装出的 `provider` 与 `model` 变量；请求监听器在下游请求监听器之后应用同一快照。因此，发生在异步组装期间的选择会从下一个步骤生效，而不会让提示词文本与路由分裂。其他调用配置字段保持不变。

-请求头仍是持久化的真源。当某个选择真正被使用时，现有的完整 `request/header` 快照会记录它；前门先从折叠后的最后一个请求头初始化其选择，然后才回退到创建选项。从未被请求使用的选择有意只保留在内存中，因为它从未成为模型可见状态。
+请求头仍是持久化的真源。当某个选择真正被使用时，现有的完整 `request/header` 快照会记录它；前端先从折叠后的最后一个请求头初始化其选择，然后才回退到创建选项。从未被请求使用的选择有意只保留在内存中，因为它从未成为模型可见状态。

 ## 考虑过的替代方案

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-19-gui-layering-and-rpc-protocol.md
-2026-07-19-gui-layering-and-rpc-protocol.md: 7997a682c8745f7b3d0a9721acfa9355603f9b2e
-2026-07-19-gui-layering-and-rpc-protocol.zh.md: cb36b5cb725128e1c5067e5e69e52aa851668e52
+2026-07-19-gui-layering-and-rpc-protocol.md: 705b0df5feb5fedaae4d198aed71758b54586e93
+2026-07-19-gui-layering-and-rpc-protocol.zh.md: b28b08b4b9da058e01af62e610d4e226d794151f
@@ -31,7 +31,7 @@ Directories layer as follows:
    - **Fetch-arrival plugin packages** (`ui-layout`, `ui-sidebar`, `ui-conversation`, `ui-trajectory`): dual-entry — the root index is the node half (an empty `apply`, existing so the host Loader governs lifecycle and the web plugin registry discovers the package.json `dshClient` declaration); the implementation lives under `src/client/`, shipped as the `./client` subpath (a tsdown closure-factory bundle). Cross-plugin consumption of `/client` is type-only; value cooperation goes through cordis services.
 - `apps/` holds the externally exported application shapes, assembled from Client / Host mixtures.
    - `apps/web` (`dsh-frontend`) is the vite application: a thin `main.ts` over the shell surface exported by `dsh-client-web`.
-    - `apps/cli` (`@deepseek-ai/dsh`) dispatches shapes: `dsh web` = Host + webserver + the built `dsh-frontend` dist; `dsh run` = [a direct core Agent/Session front door](2026-08-09-headless-direct-core-front-door.md), with zero Host, HTTP, or browser layer.
+    - `apps/cli` (`@deepseek-ai/dsh`) dispatches shapes: `dsh web` = Host + webserver + the built `dsh-frontend` dist; `dsh run` = [a direct core Agent/Session entry point](2026-08-09-headless-direct-core-entry-point.md), with zero Host, HTTP, or browser layer.
    - A future Electron shape reuses the same web client packages over an IPC fetch carrier.

 ```
@@ -79,7 +79,7 @@ Packages under `packages/host/*` and `packages/client/*` **must carry the direct
 2. **Write an assembly module under `apps/`**: `startHost()` + a client subclass + the shape's private signal/print/exit semantics; a mixture never becomes a package — assembly is written in the app.
 3. **Import `dsh-host-webserver` only if you need HTTP carriage**, otherwise zero ports.

-The two existing shapes preserve the boundary: the Web shape mounts Host, carrier, and browser composition, while `dsh run` mounts a direct core runner with zero Host, HTTP, or ports. ACP-class protocol bridges do not follow the client-carrier checklist: they expose core to the external ecosystem, mount via `ctx.plugin(front-door plugin)` directly, and wear no fetch.
+The two existing shapes preserve the boundary: the Web shape mounts Host, carrier, and browser composition, while `dsh run` mounts a direct core runner with zero Host, HTTP, or ports. ACP-class protocol bridges do not follow the client-carrier checklist: they expose core to the external ecosystem, mount via `ctx.plugin(entry-point plugin)` directly, and wear no fetch.

 ## Message protocol

@@ -242,7 +242,7 @@ Every client shape consumes one contract: adding a unary method is a five-step m
 |---|---|
 | Packaging by "product shape" (a web family, an electron family) | What shapes share is host/client capability, not the shape itself; capability-provider layering means a new shape needs zero new packages |
 | A package per mixture (e.g. a standalone headless package) | A mixture has exactly one consumer (its own app); packaging it is ownerless abstraction, while assembly in the app is readable and disposable |
-| Consuming clients connecting to ctx directly (skipping the apiproxy layer) | Client shapes require wire validation, observability, and multi-client consistency. Direct headless is a local front door with no client boundary and uses the public Agent/Session seams rather than a client command plane |
+| Consuming clients connecting to ctx directly (skipping the apiproxy layer) | Client shapes require wire validation, observability, and multi-client consistency. Direct headless is a local entry point with no client boundary and uses the public Agent/Session seams rather than a client command plane |
 | webserver depending on runtime (saving the handler injection) | Structural-typing injection keeps webserver reusable by sidecars/tests with zero workspace deps; a package dependency would drag assembly knowledge into the carrier layer |
 | Package names without the group prefix (continuing dsh-<tail>) | `dsh-runtime`/`dsh-web-ui` lose their belonging in the flat npm namespace; the cost is one explicit paths entry per package |
 | Reusing the in-repo JSON-RPC 2.0 (dsh-jsonrpc) | Numeric error codes degrade to a single fallback code, contracts get aligned by hand in two copies, and naming drifts without a convention |
@@ -29,7 +29,7 @@ Status: implemented
    - **fetch 到达插件包**（`ui-layout`、`ui-sidebar`、`ui-conversation`、`ui-trajectory`）：双入口——根入口是 node 半边（空 `apply`，其存在是为了让 host Loader 管辖生命周期、让 web 插件注册表发现 package.json 的 `dshClient` 声明）；实现住在 `src/client/` 下，经 `./client` 子路径发布（tsdown 闭包工厂 bundle）。跨插件消费 `/client` 只限类型；值层面的协作走 cordis 服务。
 - `apps/` 作为对外导出的应用形态入口，可以由 Client / Host 混合组装。
    - `apps/web`（`dsh-frontend`）是 vite 应用：`dsh-client-web` 导出的壳表面之上的一层薄 `main.ts`。
-    - `apps/cli`（`@deepseek-ai/dsh`）做形态分发：`dsh web` = Host + webserver + 构建出的 `dsh-frontend` dist；`dsh run` = [直接使用核心 Agent／Session 的前门](2026-08-09-headless-direct-core-front-door.md)，不含 Host、HTTP 或浏览器层。
+    - `apps/cli`（`@deepseek-ai/dsh`）做形态分发：`dsh web` = Host + webserver + 构建出的 `dsh-frontend` dist；`dsh run` = [直接使用核心 Agent／Session 的入口](2026-08-09-headless-direct-core-entry-point.md)，不含 Host、HTTP 或浏览器层。
    - 将来的 Electron 形态经由 IPC fetch 载体复用同一套 web client 包。

 ```
@@ -77,7 +77,7 @@ TypeScript 以 solution 根引用的**两个聚合 program** 检查（`tsconfig.
 2. **在 `apps/` 下写拼装模块**：`startHost()` + 客户端子类 + 该形态私有的信号/打印/退出语义；混合体不建包，拼装写在 app 里。
 3. **需要 HTTP 承载才 import `dsh-host-webserver`**，否则零端口。

-现有两种形态保持这一边界：Web 形态挂载 Host、载体与浏览器组合，而 `dsh run` 挂载直接使用核心服务的 runner，不包含 Host、HTTP 或端口。ACP 类协议桥不遵循 client 载体清单：它把 core 暴露给外部生态，直接通过 `ctx.plugin(前门插件)` 挂载，不套 fetch。
+现有两种形态保持这一边界：Web 形态挂载 Host、载体与浏览器组合，而 `dsh run` 挂载直接使用核心服务的 runner，不包含 Host、HTTP 或端口。ACP 类协议桥不遵循 client 载体清单：它把 core 暴露给外部生态，直接通过 `ctx.plugin(入口插件)` 挂载，不套 fetch。

 ## 消息协议

@@ -240,7 +240,7 @@ export type ResponseValue<K> =
 |---|---|
 | 按「产品形态」分包（web 一族、electron 一族） | 形态间共享的是 host/client 两侧能力而非形态本身；能力支持方分层让新形态零新包 |
 | 混合体建包（如 headless 独立包） | 混合体只有一个消费者（它自己的 app），建包是无主抽象；拼装写在 app 里可读可弃 |
-| 消费型 client 直连 ctx（省 apiproxy 一层） | client 形态需要 wire 校验、观测与多 client 一致性。直接 headless 是没有 client 边界的本地前门，使用公开的 Agent／Session seam，而不是 client 命令面 |
+| 消费型 client 直连 ctx（省 apiproxy 一层） | client 形态需要 wire 校验、观测与多 client 一致性。直接 headless 是没有 client 边界的本地入口，使用公开的 Agent／Session seam，而不是 client 命令面 |
 | webserver 依赖 runtime（省 handler 注入） | 结构 typing 注入让 webserver 可被 sidecar/测试复用且零 workspace 依赖；包依赖会把装配知识拖进承载层 |
 | 包名不带组前缀（沿用 dsh-<尾段>） | `dsh-runtime`/`dsh-web-ui` 在扁平 npm 命名空间里失去归属信息；代价只是每包一条显式 paths |
 | 复用仓内 JSON-RPC 2.0（dsh-jsonrpc） | 数字错误码退化成单码兜底、约定双份人肉对齐、命名无 convention 自然漂移 |
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-24-web-config-tree-boot-and-transport-layering.md
-2026-07-24-web-config-tree-boot-and-transport-layering.md: 496499a691dbca012e5e953cbb6eb1d0bf25b635
-2026-07-24-web-config-tree-boot-and-transport-layering.zh.md: c4bed7b730cf37e1d90750f5c93fbccb920ced21
+2026-07-24-web-config-tree-boot-and-transport-layering.md: 92ec665acc745e61f656bd0e57454ad266b722f9
+2026-07-24-web-config-tree-boot-and-transport-layering.zh.md: 7cd96ad9e52c19a005e6bff356dc5159ad3a31bc
@@ -16,23 +16,23 @@ English | [中文](2026-07-24-web-config-tree-boot-and-transport-layering.zh.md)

 **Boot glue is a class pair.** `AppCLIEntry` (apps/cli) and `AppWebEntry` (the shell kernel) hold only what must exist independently of cordis: argv facts, the composed patch set, the parsed boot manifest, the module system instance, loading-page handles — everything else lives in plugins. `AppCLIEntry.run()` is three stages: layered env (ambient > cwd `.env` > `$DSH_HOME/.env`, closing the defect above) → patch composition → Loader include boot plus the activation audit. `AppWebEntry.run()` mirrors it browser-side: parse `window.__DSH_BOOT__` into a `BootManifest` (two views: npm-package rows for the module table, cordis-plugin rows for entry composition; malformed wire throws), build the module system, render the loading page, prefetch the `immediately` tier in parallel with Context/Loader setup, **await the prefetch before creating entries** (materialization is `tree.import`'s synchronous require, unprotected by fiber inject waiting; cross-package require edges such as i18n → runtime/client need every immediately-tier factory registered first — an empirically found 10–25% boot race otherwise), adopt the modules entry, create the graph rows, settle, sweep.

-**Config sources have one declaration place each.** Bundle yml values are engineering defaults, Settings sections are writable user preferences, CLI flags address their owning launcher rows, and env values enter through yml `!!js` expressions. Patches replace a row's config wholesale. The resolved frontend `distIndex` uses that patch channel as an assembly fact. The transport-independent provider/model default belongs to `ctx.agentDefaultModel`; the [direct headless front door](2026-08-09-headless-direct-core-front-door.md) and the Web gateway consume the same state.
+**Config sources have one declaration place each.** Bundle yml values are engineering defaults, Settings sections are writable user preferences, CLI flags address their owning launcher rows, and env values enter through yml `!!js` expressions. Patches replace a row's config wholesale. The resolved frontend `distIndex` uses that patch channel as an assembly fact. The transport-independent provider/model default belongs to `ctx.agentDefaultModel`; the [direct headless entry point](2026-08-09-headless-direct-core-entry-point.md) and the Web gateway consume the same state.

-**The transport splits five ways.** `dsh-host-apiproxy` is the gateway plugin (`api-gateway` row): it default-exports `ApiProxyService`, configures only `{workspaceRoot?}`, consumes the base layer's front-door-neutral `ctx.agentDefaultModel`, provides `ctx.apiProxy`, remains transport-agnostic, and registers no routes. `dsh-host-webserver` is a plain route-registration plugin: `HttpServerService` provides `ctx.httpServer` (`register(route) → disposer` with duplicate-pattern throw, `tapIndex` transforms applied in registration order, `port`), listens on activation, answers per-request failures with 400 and logging, and knows no harness concepts. The connection node half owns the `/api` binding from `ctx.apiProxy` through `toFetchHandler`. The modules node half (`ClientModuleHostService`, providing `ctx.clientModuleHost`) owns incremental package scanning, the bundle route, the index tap, and `onRebuilt`/`onGraphChanged` notification. The hmr node half owns dev reload through `fs.watchFile` membership and the `/plugins/events` SSE route.
+**The transport splits five ways.** `dsh-host-apiproxy` is the gateway plugin (`api-gateway` row): it default-exports `ApiProxyService`, configures only `{workspaceRoot?}`, consumes the base layer's entry-point-neutral `ctx.agentDefaultModel`, provides `ctx.apiProxy`, remains transport-agnostic, and registers no routes. `dsh-host-webserver` is a plain route-registration plugin: `HttpServerService` provides `ctx.httpServer` (`register(route) → disposer` with duplicate-pattern throw, `tapIndex` transforms applied in registration order, `port`), listens on activation, answers per-request failures with 400 and logging, and knows no harness concepts. The connection node half owns the `/api` binding from `ctx.apiProxy` through `toFetchHandler`. The modules node half (`ClientModuleHostService`, providing `ctx.clientModuleHost`) owns incremental package scanning, the bundle route, the index tap, and `onRebuilt`/`onGraphChanged` notification. The hmr node half owns dev reload through `fs.watchFile` membership and the `/plugins/events` SSE route.

 **Package export discipline.** The modules package exposes exactly `.` (node half) and `./client` (the complete browser half: `ClientModuleSystem`, `parseBootManifest`, the adoption plugin face) — no bespoke subpaths; wire types re-export through the root for host-side consumers. The adoption handshake: the kernel writes the constructed instance to `window.__DSH_MODULES__` before cordis exists; the `./client` apply reads the slot (missing = loud throw) and provides `ctx.modules`.

 ## Consequences

 - Recomposing a web deployment is a yml/patch edit; the retired pieces (`mountWebPlugins`, `CLIENT_PACKAGES`, `createHostWebPluginRegistry`, `startWebServer`, the webserver's graph/SSE/api knowledge) are deleted.
- [Headless is a direct core front door](2026-08-09-headless-direct-core-front-door.md): its shipped profile contains the shared base Agent capabilities and omits Host, HTTP, Web, and browser layers. The transport split in this note is the browser surface's contract.
+- [Headless is a direct core entry point](2026-08-09-headless-direct-core-entry-point.md): its shipped profile contains the shared base Agent capabilities and omits Host, HTTP, Web, and browser layers. The transport split in this note is the browser surface's contract.
 - A TypeScript pitfall worth remembering: a `declare module 'cordis'` augmentation in a file with **no cordis import** is demoted to a standalone module declaration and silently shatters the program-wide `Context` merge (`ctx.on`/`ctx.effect` vanish across the program). Anchor with `import type {} from 'cordis'`.

 ## Alternatives considered

 | Rejected | One-line reason |
 |---|---|
-| Dedicated `dsh-host-profile` receiver package | User model state belongs to the Settings-backed `ctx.agentDefaultModel`; an extra Host receiver would duplicate ownership and exclude direct front doors |
+| Dedicated `dsh-host-profile` receiver package | User model state belongs to the Settings-backed `ctx.agentDefaultModel`; an extra Host receiver would duplicate ownership and exclude direct entry points |
 | Runtime `assembly` shim plugin providing an `apiHandler` service | Existed only because `createApiProxy` lived in runtime; moving it into apiproxy made the gateway self-hosting, and `toFetchHandler` is a pure function the binding side calls |
 | Full-rescan + incremental scan coexisting | Two implementations, two semantics; the single per-package path covers the activation pass too |
 | A bespoke `./impl` export on the modules package | Non-uniform export surface; the standard `./client` carries the whole browser half |
@@ -16,23 +16,23 @@ Status: implemented

 **boot 胶水由两个类组成。** `AppCLIEntry`（apps/cli）与 `AppWebEntry`（壳内核）只持有那些必须独立于 cordis、提前存在的东西：argv 事实、合成的 patch 集、解析出的 boot manifest（元数据清单）、模块系统实例、loading 页句柄——其余一律进插件。`AppCLIEntry.run()` 三段：分层 env（ambient > cwd `.env` > `$DSH_HOME/.env`，顺手关掉上述缺陷）→ patch 合成 → Loader include boot 加 activation audit。`AppWebEntry.run()` 在浏览器侧镜像它：把 `window.__DSH_BOOT__` 解析成 `BootManifest`（双视角：npm 包行给模块表、cordis 插件行给 entry 组合；畸形 wire 大声抛）、建模块系统、渲染 loading 页、immediately 层预取与 Context/Loader 准备并行、**create entry 之前等预取齐**（物化是 `tree.import` 的同步 require，不受 fiber inject 等待保护；i18n → runtime/client 这类跨包 require 边要求 immediately 层工厂全部注册完——否则有实测 10–25% 的 boot 竞态）、收编 modules entry、逐一创建图行、settle、sweep。

-**每个配置源有唯一声明位置。** 组合包 yml 值是工程默认，Settings 分节是可写的用户偏好，CLI（命令行界面）flags 面向其归属的启动器配置行，env 值则通过 yml `!!js` 表达式进入。patch 会整体替换一行的 config。解析后的前端 `distIndex` 通过同一条 patch 通道作为组装事实传递。与传输无关的提供方／模型默认值归 `ctx.agentDefaultModel` 所有；[直接 headless 前门](2026-08-09-headless-direct-core-front-door.md)与 Web 网关消费同一份状态。
+**每个配置源有唯一声明位置。** 组合包 yml 值是工程默认，Settings 分节是可写的用户偏好，CLI（命令行界面）flags 面向其归属的启动器配置行，env 值则通过 yml `!!js` 表达式进入。patch 会整体替换一行的 config。解析后的前端 `distIndex` 通过同一条 patch 通道作为组装事实传递。与传输无关的提供方／模型默认值归 `ctx.agentDefaultModel` 所有；[直接 headless 入口](2026-08-09-headless-direct-core-entry-point.md)与 Web 网关消费同一份状态。

-**传输五分。** `dsh-host-apiproxy` 是网关插件（`api-gateway` 行）：默认导出 `ApiProxyService`，只配置 `{workspaceRoot?}`，消费 base 层不偏向特定前门的 `ctx.agentDefaultModel`，provide `ctx.apiProxy`，保持传输无关且不注册路由。`dsh-host-webserver` 是朴素的路由注册插件：`HttpServerService` provide `ctx.httpServer`（`register(route) → disposer`、重复 pattern 即抛、`tapIndex` 按注册序应用、`port`），激活即 listen，单请求失败时答 400 并记日志，且不认识任何 harness 概念。connection node 半拥有从 `ctx.apiProxy` 经 `toFetchHandler` 绑定到 `/api` 的逻辑。modules node 半（`ClientModuleHostService`，provide `ctx.clientModuleHost`）拥有单包增量扫描、bundle 路由、index tap 与 `onRebuilt`/`onGraphChanged` 通知。HMR node 半通过 `fs.watchFile` membership 与 `/plugins/events` SSE 路由拥有开发期重载。
+**传输五分。** `dsh-host-apiproxy` 是网关插件（`api-gateway` 行）：默认导出 `ApiProxyService`，只配置 `{workspaceRoot?}`，消费 base 层不偏向特定入口的 `ctx.agentDefaultModel`，provide `ctx.apiProxy`，保持传输无关且不注册路由。`dsh-host-webserver` 是朴素的路由注册插件：`HttpServerService` provide `ctx.httpServer`（`register(route) → disposer`、重复 pattern 即抛、`tapIndex` 按注册序应用、`port`），激活即 listen，单请求失败时答 400 并记日志，且不认识任何 harness 概念。connection node 半拥有从 `ctx.apiProxy` 经 `toFetchHandler` 绑定到 `/api` 的逻辑。modules node 半（`ClientModuleHostService`，provide `ctx.clientModuleHost`）拥有单包增量扫描、bundle 路由、index tap 与 `onRebuilt`/`onGraphChanged` 通知。HMR node 半通过 `fs.watchFile` membership 与 `/plugins/events` SSE 路由拥有开发期重载。

 **包出口纪律。** modules 包只暴露 `.`（node 半）与 `./client`（完整浏览器半：`ClientModuleSystem`、`parseBootManifest`、收编插件面）——不设专用子路径；wire 类型经根出口 re-export 给 host 侧消费方。收编握手：内核在 cordis 之前把建好的实例写入 `window.__DSH_MODULES__`；`./client` 的 apply 读取该槽位（缺少时显式抛错）并 provide `ctx.modules`。

 ## 后果

 - 重组一个 web 部署 = 改 yml/patch；退役件（`mountWebPlugins`、`CLIENT_PACKAGES`、`createHostWebPluginRegistry`、`startWebServer`、webserver 的图/SSE/api 知识）全部删除。
- [Headless 是直接 core 前门](2026-08-09-headless-direct-core-front-door.md)：其随附 profile 包含共享的 base Agent 能力，并省去 Host、HTTP、Web 与浏览器层。本笔记的传输划分是浏览器 surface 的约定。
+- [Headless 是直接 core 入口](2026-08-09-headless-direct-core-entry-point.md)：其随附 profile 包含共享的 base Agent 能力，并省去 Host、HTTP、Web 与浏览器层。本笔记的传输划分是浏览器 surface 的约定。
 - 一个值得记住的 TypeScript 坑：`declare module 'cordis'` augmentation 所在文件若**没有任何 cordis import**，会被降级成独立 module declaration，无声打散全程序的 `Context` merge（`ctx.on`/`ctx.effect` 全程序消失）。用 `import type {} from 'cordis'` 锚定。

 ## 考虑过的替代方案

 | 弃案 | 一行理由 |
 |---|---|
-| 专门的 `dsh-host-profile` 受体包 | 用户模型状态归 Settings 支撑的 `ctx.agentDefaultModel` 所有；额外的 Host 受体会重复归属，并排除直接前门 |
+| 专门的 `dsh-host-profile` 受体包 | 用户模型状态归 Settings 支撑的 `ctx.agentDefaultModel` 所有；额外的 Host 受体会重复归属，并排除直接入口 |
 | 运行时里的 `assembly` 垫层插件（provide `apiHandler`） | 它的存在只因 `createApiProxy` 住运行时；本体迁入 apiproxy 后网关自持插件身份，且 `toFetchHandler` 是绑定方自己调的纯函数 |
 | 全量重扫与增量扫描并存 | 两条实现两份语义；单包路径足以覆盖激活初扫 |
 | modules 包特设 `./impl` 出口 | 出口面不统一；标准 `./client` 承载完整浏览器半 |
@@ -1,6 +1,6 @@
 # Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
-#   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-09-headless-direct-core-front-door.md
-2026-08-09-headless-direct-core-front-door.md: f4604329a9276448a0021bb749b09e8c1b82e3c1
-2026-08-09-headless-direct-core-front-door.zh.md: aaa1289894bf3c69b39aa863493dffdc3437ad01
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-03-per-session-agent-presets.md
+2026-08-03-per-session-agent-presets.md: 6f1643c25008c3363cb10adb7fbff7afeea31cbe
+2026-08-03-per-session-agent-presets.zh.md: 7afe9ade5c98fadb96384a7e0acd47531c370e0c
@@ -0,0 +1,80 @@
+# Agent Note: A session's agent is composed from a preset cordis.yml
+
+Status: implemented
+
+English | [中文](2026-08-03-per-session-agent-presets.zh.md)
+
+## Problem
+
+One `dsh` process serves many sessions, but the composition that decides what an agent *is* — its tools, persona, prompt sections, delegation backends — is fixed for the whole process by the `cordis.yml` the launcher booted. A deployment that wants a benchmark-minimal agent beside a full coding agent has to run two processes, and the shipped workaround (`apps/cli/config/minimal.cordis.yml`, a `--config` overlay that disables tool rows) changes every session at once.
+
+The obvious reading of "let a session pick its composition" is that the loader needs a new tier. It does not. [`dsh-tools`](../../../../packages/core/tools/README.md) and [`dsh-system-prompt`](../../../../packages/core/system-prompt/README.md) already file registrations into the calling context's scope layer, and [the agent is a registration scope](2026-07-08-agent-scope-contexts.md). What was missing is a way to point a whole `cordis.yml` at one agent's scope.
+
+## Decision
+
+A **preset** is a directory holding one `agent.cordis.yml`. The agent factory's `setup(agentCtx)` mounts it as a Cordis `include` subtree plugged into that agent's scope context. Entry contexts chain to the context a subtree was plugged into, so every registration inside the preset lands in that agent's layer and unwinds with the agent. No registry gains a tier, and no session already running is touched.
+
+Composition splits into two planes, decided by what must be shared rather than by what feels agent-related:
+
+| Plane | Instances | Contents |
+|---|---|---|
+| Host | one | The registries themselves (`tools`, `systemPrompt`, `agents`, `agent-loop`, `sessions`), cross-session facilities (persistence, query, projections, storage, settings, credentials, telemetry), and the web host |
+| Agent | one per session | What a single agent contributes to those registries: tool plugins, persona and prompt sections, compaction policy |
+
+Model routing stays out of presets. `installAgentLlmTarget` is already the per-agent seam for provider, model, and reasoning effort, and an LLM adapter mounted inside a preset would never be resolved by `agent-loop`, which lives in the host plane.
+
+The presets the deployment ships are the directories under `apps/cli/config/agent-presets/`; the roster is that listing, not a list restated here.
+
+Mounting is per-session by default. Measured cost for a twelve-row composition is ~3ms and ~600KB per session, so isolation is the cheaper default than any sharing scheme, and a preset authored by a user or by an agent then has the smallest possible blast radius. A preset that genuinely owns an expensive singleton opts into sharing with Cordis's own `isolate` vocabulary: a named realm label is process-global, so two subtrees naming the same label resolve one instance.
+
+Which preset an unnamed session gets is a user setting (`agent-presets.default`) layered over the composition's own `default`, which becomes the `base`. Both layers are needed: the composition value is what a deployment ships and must keep working with no settings provider at all, and the setting is what a person changes without editing a `cordis.yml` they may not own.
+
+## Consequences
+
+**The effective default is read per resolution, never snapshotted.** A cached value would need a `watch` subscription and a reload path to stay honest, and the resolved scope already re-reads a hot-reloaded document. Reading through is also what makes the boundary correct rather than merely cheap: the new value applies to the next session created, and every running session keeps the composition it was built from. That invariant is the same one the session header enforces from the other side — the header records the id a session actually runs, so a resume rebuilds that composition rather than today's default, and the gateway rejects an attempt to adopt a live session under a different one. A snapshot would make the two disagree at exactly the moment the setting changes.
+
+**A directly-plugged subtree is invisible to the boot audit.** It never links itself to an `Entry`, so it is absent from `ctx.loader.entries()` and `assertEntriesActivated` cannot see it. The mount audits its own rows instead, reading the tree through an `Include` subclass that publishes it.
+
+**A preset can only name a group because the app registers one.** Sharing a realm across rows is a `cordis:group` row, and a preset living outside this workspace — the authored ones under the Harness home, which is the point — cannot resolve `@cordisjs/plugin-group` by name: Node's upward `node_modules` walk never reaches the harness from there. `boot()` therefore registers `cordis:group` beside `cordis:include` as a loader builtin, so both load through the ambient module pipeline rather than through the included tree's own specifier resolution. Without it the `isolate` vocabulary above is expressible one row at a time only, and a provider could never be grouped with its consumers.
+
+**A preset may not publish into the root service realm.** Such a service is process-global rather than per-session, so the second session mounting the same preset collides with the first — and the collision surfaces as an unhandled rejection that `setup` never observes, leaving a half-composed agent that looks healthy. The mount rejects it instead, and the package invariant re-checks on every service notification because a row publishing from a timer or an asynchronous continuation would escape a one-shot audit.
+
+**Failure rolls the agent back.** `setup` runs before publication, so a rejected mount fails `ctx.agents.create()` and leaves nothing behind. This is why `setup` is the one supported call site.
+
+**A test that the preset file is never rewritten has to be able to fail.** The first version asserted the file was unchanged after an ordinary mount, and could not have caught anything: the Loader only reaches its write path when it decides the config changed, and nothing in that composition ever self-disposed. The regression plants a row that disposes itself — the shape a real preset hits every time an agent is torn down — and keeps the composition in a temp root rather than under `fixtures/`, because without the override the Loader rewrites the file it read: a committed fixture would be damaged by the very run that proves the bug, and every run after it would compare against the damaged file and pass.
+
+**Fiber membership is object identity, not `uid`.** A `uid` is a per-registry counter, so fibers in two different roots collide on it; comparing by `uid` made one runtime's subtree answer for a service published in another. `ctx.plugin()` returns a thenable `Object.create(fiber)` wrapper that is never identical to the fiber in a parent chain, so the subtree captures its own fiber during construction.
+
+**A preset file is an input, never a persistence target.** `EntryTree.write()` persists a tree whenever the Loader decides the config changed, and a plugin self-disposing is enough — tearing an agent down disposes its whole subtree. Inherited, that rewrites the composition it read, in practice truncating a shipped preset to `[]` the first time a session ends. The subtree overrides `write()` to do nothing.
+
+**A plugin that looks itself up in the global registry breaks inside a preset.** `ctx.tools.register()` files into the CALLING context's scope, so a plugin mounted in a preset registers for one agent and an unscoped `ctx.tools.get(name)` correctly finds nothing. `dsh-tool-skill` did exactly that and threw on every preset mount; it now compares against the definition it registered. Any plugin meant to be preset-mountable must hold its own registration rather than re-read it by name.
+
+**An entry-local `isolate` realm is invisible to the agent's own scope, not only to the host.** Only rows inside that group resolve the service. That is what makes a preset's `skills` registry belong to one agent rather than being shared — and it means a consumer left outside its provider's group silently resolves the host registry and contributes nothing.
+
+**Switching is allowed only while a session is blank.** Once a turn has run, that history was produced under the preset's tools and swapping them would strand logged tool calls, so `agentPreset.select` answers `agent-preset-locked`. A blank switch keeps the agent and the session and replaces only the subtree, because the host discards the `AgentHandle` it creates and there is no delete RPC — and keeping them is the better outcome anyway, since the session id, its workspace attachment, and its projections all stay put. The swap is unmount-then-mount (two compositions would register the same tool names into one layer), so it resolves the new preset before tearing anything down and restores the previous one when the new mount fails.
+
+**Authoring a preset is an RPC, and a privileged one.** A composition is a file, but "edit it on the filesystem" is not a browser affordance, so the roster gained `read`/`write`/`remove` beside `select`. Those three are loopback-pinned: a composition names the plugins a session runs, so reading one is reconnaissance and writing one is arbitrary capability. `list` and `select` deliberately stay ordinary. The roster carries ids and trust only, and a LAN client's picker needs it; and choosing a preset looked like escalation — one of them mounts the toolset that edits the live runtime — but `session.create` already takes an `agentPreset`, so pinning only the switch would have left the same capability one method over. The capability is not the preset's to grant either: the deployment's own default already carries `bash` and the filesystem tools, so any caller that may start a session at all can already run commands as this process. Containment is a property of the id (`[a-z0-9][a-z0-9-]*`), checked before it becomes a directory name rather than by inspecting the joined path afterwards; the text is parsed with the loader's own schema and dialect, so a save cannot leave a file no session could load. Shipped presets are refused for writes and deletes, because the deployment's copy is what a broken local preset is compared against — which also makes "duplicate, then edit" the authoring path rather than an afterthought.
+
+**A service with a consumer outside the agent plane cannot move into a preset.** The aggressive split moved the `subagents` registry and its spawn/fork backends into the delegation group's entry-local realm, and `dsh web` then failed to boot: `dsh-host-apiproxy` is a HOST row that injects `subagents` to answer the browser's cross-session queries (`listChildren`, `followup`), so it waited forever for a service only sessions now provided. A per-session copy is wrong twice over — a provider name registers once, so the second session would have collided anyway. The registry and its backends are host-plane; the preset contributes the delegation TOOLS, which resolve the host registry. `workflows` stays entry-local because nothing outside an agent reads it. Grepping injectors is what should have caught this and did not: the search has to include the host packages, not just the agent-plane ones.
+
+**A real-composition test that disables a host row cannot audit that row.** The web composition test disabled `api-gateway` — the api-proxy itself — as a row with side effects, which is exactly the row whose pending injection would have named the break. It now boots with the api-proxy enabled and the browse directory picker substituted, so the boot audit covers the whole host-plane injection graph; only the port, the asset tree, and the telemetry exporter stay off.
+
+**A preset's package names must resolve from the harness, not from the preset.** `EntryTree.import()` resolves a row against its own tree's `baseUrl`, which `Include` sets to the composition's directory. That is right for a relative specifier and fatal for a package name: a locally authored preset lives under the user's home, where Node's upward `node_modules` walk never reaches the installed harness, so every `@deepseek-ai/dsh-*` row fails to import and the whole preset is unmountable. The shipped presets hid this — they sit inside the install. The mount records the host composition's base before plugging the subtree and sends bare specifiers there, leaving relative paths resolving from the preset so its own files still travel with it. The real-composition test writing a preset into a temp root is what found it.
+
+**The preset id is model-visible and must be logged.** It determines the tool set and prompt, so a resumed session has to restore the same composition; recording it is a session fact, not runtime state. It rides the session header beside `cwd`, and the summary carries it so a picker shows what a session actually runs rather than the deployment's current default.
+
+**A durable header field is not durable until every backend writes it.** `agentPreset` landed on `SessionHeader` with the right rationale and neither persistence backend carried it: the JSONL header line, the SQLite `sessions` row, and the derived query index each map the header column by column, so a resumed session came back with no preset and the surfaces that name it fell silent. `summarizeCold` had the same shape — it hand-built the cold list row instead of reusing the shared projection. A field declared durable needs a test that crosses a real store, not only the type that declares it.
+
+**The choice belongs to the screen where it still works.** The composer seat spent almost its whole life disabled, since the preset is fixed once a turn has run. It moved to the new-session screen beside the workspace picker, where the pick is *staged*: that screen precedes the session it applies to, and the stage lands when a session becomes current and is still blank — covering both the session a workspace connect creates and the blank one it reuses, which riding `sessions.create` would miss. It is spent on first use, matching the workspace picker beside it. What a running session runs is then a read-only label in its header: a control there would promise a switch the host refuses outright.
+
+**A preset multiplies a cost the host was already paying: nothing disposes an agent.** Measured against the shipped compositions with `--expose-gc`, one live agent holds ~0.17 MB on `minimal` and ~1.31 MB on `standard`/`cordis`, mounting in ~38 ms and ~135 ms; the first agent of a process costs ~7 MB more as Node imports the modules, which every later mount then shares. Growth is strictly linear — 10, 30 and 50 agents give the same per-agent delta — and disposal reclaims essentially all of it (50 `standard` agents held 57.8 MB and returned it). So the object graph does not leak; the lifecycle does. `dsh-host-apiproxy` discards the `AgentHandle` it creates, `archiveSession` only edits the workspace registry, `AgentRegistry` has no eviction, and the sole disposal site in the host is the JSON-RPC server's own shutdown. A web host therefore retains every session it has touched, at ~1.3 MB each once presets are composed rather than ~0.2 MB before. Note that pruning the mount registry does not help here: it drops records whose fiber `uid` has cleared, and an agent that never dies never clears one.
+
+- Remaining TODO: idle agent eviction — dispose after the session is persisted and re-mount on resume. It belongs to the host that owns the handle, not to this seam.
+
+## Alternatives considered
+
+**Add a preset tier to the scoped registries.** `ScopedLayers.merge()` combines the global layer with exactly one exact-scope layer. A middle tier would let many sessions share one mounted composition, but it changes `dsh-scope` and every scope-aware registry to save a cost measured in milliseconds, and it gives a preset's registrations a lifetime no agent owns.
+
+**Make the agent's scope key the preset.** Sessions on one preset would share a layer for free, but per-agent registrations — `installAgentLlmTarget`, per-agent tool restrictions — would then collide across sessions.
+
+**Run each preset as a child process.** [`subagent-dsh-sdk`](../../../../packages/subagent/subagent-dsh-sdk/README.md) already proves a full child harness works, and isolation would be absolute. It also means proxying streaming, approvals, and projections per session, which is a transport project rather than a composition one.
@@ -0,0 +1,81 @@
+# Agent Note：会话的 agent 由一份 preset cordis.yml 组装而成
+
+Status: implemented
+
+[English](2026-08-03-per-session-agent-presets.md) | 中文
+
+## 问题
+
+一个 `dsh` 进程服务多个会话，但决定 agent（智能体）究竟是什么的那套组装——它的工具、人设、提示词段落、委派后端——由启动器所引导的 `cordis.yml` 一次性固定给整个进程。若某个部署希望一个 benchmark 精简 agent 与一个完整编码 agent 并存，就必须跑两个进程；而现有的变通方案（`apps/cli/config/minimal.cordis.yml`，一个用来禁用工具行的 `--config` 覆盖层）会一次性改变所有会话。
+
+对"让会话自选组装"最直觉的理解，是 loader 需要新增一层。其实不需要。[`dsh-tools`](../../../../packages/core/tools/README.md) 与 [`dsh-system-prompt`](../../../../packages/core/system-prompt/README.md) 本就按调用方上下文的 scope 分层归档注册，而且 [agent 本身就是一个注册 scope](2026-07-08-agent-scope-contexts.md)。此前缺的只是一种把整份 `cordis.yml` 指向某一个 agent scope 的办法。
+
+## 决策
+
+**preset** 是一个目录，其中放置一份 `agent.cordis.yml`。agent 工厂的 `setup(agentCtx)` 把它作为 Cordis `include` 子树，挂载到该 agent 的 scope 上下文之下。entry 上下文沿原型链连到子树被挂载时所在的上下文，因此 preset 内部的每一次注册都落进该 agent 的分层，并随 agent 一起卸载。没有任何注册表新增分层，也没有任何已在运行的会话被触及。
+
+组装划分为两个平面，依据是什么必须共享，而不是什么感觉上与 agent 有关：
+
+| 平面 | 实例数 | 内容 |
+|---|---|---|
+| 宿主 | 一份 | 注册表本身（`tools`、`systemPrompt`、`agents`、`agent-loop`、`sessions`）、跨会话设施（持久化、查询、投影、存储、设置、凭据、遥测），以及 web 宿主 |
+| agent | 每会话一份 | 单个 agent 对这些注册表的贡献：工具插件、人设与提示词段落、压缩策略 |
+
+模型路由不进 preset。`installAgentLlmTarget` 已经是 provider、model 与 reasoning effort 的按 agent 可替换点；而挂在 preset 内部的 LLM 适配器永远不会被 `agent-loop` 解析到，因为后者位于宿主平面。
+
+部署交付哪些 preset，取决于 `apps/cli/config/agent-presets/` 下有哪些目录；清单是那份目录列表，而不是在此另抄一份。
+
+挂载默认按会话进行。实测一份十二行组装每会话约 3ms、约 600KB，因此隔离比任何共享方案都更划算；而由用户或 agent 写出的 preset 也因此拥有尽可能小的影响面。确实自带昂贵单例的 preset，可以用 Cordis 自身的 `isolate` 词汇显式选择共享：命名 realm 的 label 是进程级全局的，因此两棵子树只要写同一个 label 就解析到同一个实例。
+
+未指名 preset 的会话拿到哪一个，是一项用户设置（`agent-presets.default`），叠在组装自身的 `default` 之上——后者成为 `base`。两层都需要：组装里的值是部署交付的东西，在完全没有 settings 提供方时也必须照常工作；而设置是让人不必去改一份可能并不属于自己的 `cordis.yml` 就能调整的东西。
+
+## 后果
+
+**有效默认值在每次解析时读取，从不快照。** 缓存下来就需要一个 `watch` 订阅和一条重载路径才能保持诚实，而解析后的 scope 本来就会重读热重载过的文档。读穿也不只是省事，它让边界本身是对的：新值作用于**下一个新建的会话**，每个运行中的会话保持它被构建时的那份组装。这条不变量正是 session header 从另一侧执行的同一条——header 记录会话实际运行的 id，因此恢复重建的是那份组装而不是当下的默认值，网关也会拒绝把一个活着的会话收编到另一个 preset 之下。快照会让两者恰好在设置改变的那一刻各说各话。
+
+
+**直接挂载的子树对启动审计不可见。** 它不会把自己关联到 `Entry`，因此不在 `ctx.loader.entries()` 中，`assertEntriesActivated` 也看不到它。改由挂载过程自行校验各行，通过一个会公开自身 tree 的 `Include` 子类读取。
+
+**preset 能写出 group，是因为 app 注册了它。** 跨行共享 realm 就是一个 `cordis:group` 行，而住在本工作区之外的 preset——也就是 Harness home 下由人或 agent 创作的那些，正是这套设计的目的——无法按名字解析 `@cordisjs/plugin-group`：Node 向上查找 `node_modules` 的路径从那里永远走不到 harness。因此 `boot()` 把 `cordis:group` 与 `cordis:include` 并排注册为 loader builtin，两者都经由环境模块管线加载，而不依赖被包含树自身的说明符解析。没有它，上文那套 `isolate` 词汇就只能一行一行地表达，提供方也永远无法与它的消费方归入同一组。
+
+**preset 不得把服务发布进根 realm。** 这类服务是进程级全局而非按会话的，因此第二个挂载同一 preset 的会话会与第一个相撞——而这次相撞表现为 `setup` 永远观察不到的未处理 rejection，留下一个看起来健康、实则组装到一半的 agent。挂载改为直接拒绝它；本包的运行时不变量还会在每次服务通知时复查，因为从定时器或异步续体中发布的行会绕过一次性审计。
+
+**失败会让 agent 回滚。** `setup` 在发布之前运行，因此挂载被拒绝会让 `ctx.agents.create()` 失败且不留残留。这正是 `setup` 是唯一受支持调用点的原因。
+
+**「preset 文件从不被回写」这条断言，必须先有失败的可能。** 最初那版在一次普通挂载之后断言文件未变，其实什么也抓不到：Loader 只在认定 config 变了时才会走到写路径，而那份组装里没有任何一行会自行销毁。回归用例改为植入一个自行销毁的行——真实 preset 在每次 agent 被拆除时都会命中的形状——并把组装放在临时根目录而不是 `fixtures/` 下：没有那个覆写，Loader 会回写它读入的文件，于是提交进仓库的 fixture 会被**恰恰是证明该缺陷的那次运行**改坏，之后每一次运行都拿改坏后的文件作比较从而通过。
+
+**fiber 归属判定用对象同一性，而非 `uid`。** `uid` 是按 registry 计数的序号，因此两个不同根下的 fiber 会在它上面撞号；按 `uid` 比较曾导致一个运行时的子树为另一个运行时中发布的服务背锅。`ctx.plugin()` 返回的是 thenable 的 `Object.create(fiber)` 包装对象，与父链中出现的 fiber 永远不同一，因此子树在构造时捕获自己的 fiber。
+
+**preset 文件是输入，绝不是持久化目标。** 只要 loader 认为配置变了，`EntryTree.write()` 就会回写整棵树，而一个插件自我 dispose 就足以触发——销毁 agent 会 dispose 它的整棵子树。若继承该行为，它会重写自己读入的那份组装，实际后果是第一次会话结束时把随附 preset 截断成 `[]`。子树因此把 `write()` 覆盖为空操作。
+
+**按自身名字回查全局注册表的插件，在 preset 里必然失效。** `ctx.tools.register()` 归档进**调用方**上下文的 scope，因此挂在 preset 里的插件只为一个 agent 注册，而不带 scope 的 `ctx.tools.get(name)` 理所当然查不到。`dsh-tool-skill` 正是这样写的，于是每次 preset 挂载都抛错；现在它与自己注册的那个定义比对。任何希望可被 preset 挂载的插件，都必须持有自己的注册对象，而不是按名字重新读取。
+
+**entry 本地 `isolate` realm 不仅对宿主不可见，对 agent 自身的 scope 同样不可见。** 只有该组内部的行能解析到该服务。这正是让 preset 的 `skills` 注册表归属单个 agent 而非共享的原因——同时也意味着：被留在提供方组之外的消费方会静默解析到宿主注册表，然后什么都不贡献。
+
+**只有空白会话才允许切换。** 一旦跑过任何轮次，那段历史就是在该 preset 的工具下产生的，替换会留下无法执行的已记录 tool call，因此 `agentPreset.select` 返回 `agent-preset-locked`。空白期的切换保留 agent 与 session，只替换子树——因为宿主丢弃了它创建的 `AgentHandle`，也没有 delete RPC；而保留它们本身就是更好的结果，会话 id、workspace 挂接与 projections 都原地不动。该替换是"先卸后装"（两份组装会把同名工具注册进同一分层），因此它在拆除任何东西之前先解析新 preset，并在新组装装载失败时恢复原来的那一份。
+
+**创作 preset 是一次 RPC，而且是特权 RPC。** 组装是一个文件，但“去文件系统里改它”并不是浏览器能提供的操作，因此名单在 `select` 之外新增了 `read`/`write`/`remove`。这三者被固定在环回地址：组装指明了一个会话所运行的插件，因此读取它是侦察，写入它是任意能力。`list` 与 `select` 刻意保持为普通方法。名单只携带 id 与信任级别，而局域网客户端的选择器需要它；至于选择本身，它看起来像提权——其中一个 preset 会挂载可编辑活动运行时的工具集——但 `session.create` 本就接受 `agentPreset`，只固定切换会把同一能力留在隔壁一个方法上。这份能力也不由 preset 授予：部署自带的默认 preset 本就带着 `bash` 与文件系统工具，因此任何被允许开启会话的调用方，早已能以本进程的身份执行命令。约束是 id 自身的性质（`[a-z0-9][a-z0-9-]*`），在它成为目录名之前就检查，而不是事后再去审视拼接出的路径；文本使用 loader 自身的 schema 与方言解析，因此保存不会留下任何会话都无法加载的文件。随部署提供的 preset 拒绝写入与删除，因为部署自带的那一份正是用来对照有问题的本地 preset 的——这也让“先复制、再编辑”成为创作路径本身，而非事后补充。
+
+**在 agent 平面之外还有消费方的服务，不能搬进 preset。** 激进拆分把 `subagents` 注册表连同 spawn/fork 后端一起搬进了 delegation 组的 entry-local realm，于是 `dsh web` 直接起不来：`dsh-host-apiproxy` 是宿主行，它注入 `subagents` 来回答浏览器的跨会话查询（`listChildren`、`followup`），因而永远等待一个此刻只有会话才提供的服务。按会话各一份在两个层面上都是错的——provider 名只能注册一次，第二个会话本来也会相撞。注册表与后端属于宿主平面；preset 贡献的是委派**工具**，它们解析宿主注册表。`workflows` 保持 entry-local，因为 agent 之外没有任何东西读它。本该拦下它的是「检索注入方」这一步，而它没拦住：检索必须覆盖宿主包，而不只是 agent 平面的包。
+
+**真实组装测试若禁用了某个宿主行，就无法审计该行。** web 组装测试把 `api-gateway`——也就是 api-proxy 本身——当作「有外部副作用的行」禁用了，而它恰恰是那个会以 pending 注入点名此次断裂的行。现在它在启用 api-proxy、并替换为 browse 目录选择器的前提下引导，启动审计因此覆盖整个宿主平面的注入图；只有端口、资源目录与遥测导出器仍然关闭。
+
+**preset 的包名必须从 harness 解析，而非从 preset 解析。** `EntryTree.import()` 按行所属树的 `baseUrl` 解析，而 `Include` 把它设为组装文件所在的目录。这对相对标识符是对的，对包名却是致命的：本地创作的 preset 位于用户主目录之下，Node 向上查找 `node_modules` 永远够不到已安装的 harness，因此每一个 `@deepseek-ai/dsh-*` 行都会导入失败，整个 preset 无法挂载。随部署提供的 preset 掩盖了这一点——它们本就在安装目录之内。挂载在插入子树之前先记录宿主组装的基址，并把裸标识符送往那里，同时让相对路径继续从 preset 解析，使它自带的文件仍随它一同迁移。发现它的正是那个把 preset 写入临时根目录的真实组装测试。
+
+**preset id 对模型可见，必须写入日志。** 它决定工具集与提示词，因此被恢复的会话必须还原同一份组装；记录它属于会话事实，而非运行时状态。它与 `cwd` 并列写在会话头部，并由会话摘要携带，使选择器显示的是某个会话实际运行的 preset，而非部署当前的默认值。
+
+**持久化的头部字段，在每个后端都写入之前都算不上持久。** `agentPreset` 带着正确的理由落在了 `SessionHeader` 上，而两个持久化后端都没有携带它：JSONL 头部行、SQLite `sessions` 行、以及派生的查询索引各自逐列映射头部，于是被恢复的会话回来时没有 preset，所有据以命名它的表层随之失声。`summarizeCold` 是同一个形状——它手工拼装冷列表行，而没有复用共享的投影。声明为持久的字段，需要一个跨越真实存储的测试，而不只是声明它的那个类型。
+
+**这个选择属于它仍然可用的那个界面。** composer 座位几乎一生都处于禁用状态，因为一旦跑过一个轮次，preset 即固定。它移到了新建会话界面、工作区选择器旁边，选择在那里是**暂存**的：该界面先于它要应用到的会话存在，暂存值在某个会话成为当前会话且仍为空白时落地——这既覆盖工作区连接新建的会话，也覆盖它复用的那个空白会话，而搭 `sessions.create` 的便车会漏掉后者。它一经使用即被清空，与旁边的工作区选择器一致。至于运行中的会话在跑什么，则是其标题旁的一个只读标签：在那里放控件，等于承诺一次宿主会断然拒绝的切换。
+
+**preset 放大的是宿主本来就在付的代价：没有任何东西会 dispose 一个 agent。** 用 `--expose-gc` 对随附组装实测：一个存活的 agent 在 `minimal` 上约占 0.17 MB、在 `standard`/`cordis` 上约 1.31 MB，挂载耗时分别约 38 ms 与 135 ms；进程里第一个 agent 另需约 7 MB，那是 Node 首次 import 模块的一次性成本，此后每次挂载共享。增长严格线性——10、30、50 个的单个增量一致——且 dispose 后基本全额回收（50 个 `standard` 占住 57.8 MB，释放后全部归还）。所以对象图并不泄漏，缺的是生命周期。`dsh-host-apiproxy` 创建后直接丢弃 `AgentHandle`，`archiveSession` 只改工作区注册表，`AgentRegistry` 没有驱逐机制，而宿主里唯一一处 dispose 是 JSON-RPC 服务器自身的关停。于是一个 web 宿主会留住它接触过的每一个会话，组装 preset 之后每个约 1.3 MB，而在此之前约 0.2 MB。注意：剪枝挂载注册表在这里没有用——它丢弃的是 fiber `uid` 已清空的记录，而永不死亡的 agent 永远不会清空它。
+
+- 遗留 TODO：idle agent 驱逐——会话持久化后 dispose，恢复时重新挂载。它属于持有 handle 的那个宿主，不属于本 seam。
+
+## 考虑过的替代方案
+
+**在 scope 注册表中新增 preset 分层。** `ScopedLayers.merge()` 把全局层与恰好一个精确 scope 层合并。新增中间层可以让多个会话共用一份已挂载的组装，但它要改动 `dsh-scope` 及每个 scope 感知的注册表，换来的只是毫秒级的开销节省，而且会让 preset 的注册获得一个没有任何 agent 拥有的生命周期。
+
+**把 agent 的 scope 键设为 preset。** 同一 preset 上的会话就能免费共享一层，但按 agent 的注册——`installAgentLlmTarget`、按 agent 的工具限制——会跨会话相撞。
+
+**把每个 preset 作为子进程运行。** [`subagent-dsh-sdk`](../../../../packages/subagent/subagent-dsh-sdk/README.md) 已经证明完整的子 harness 可行，隔离性也会是绝对的。但这同时意味着要按会话代理流式输出、审批与投影，那是一个传输层项目，而非组装问题。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-05-profile-plugin-bundles.md
-2026-08-05-profile-plugin-bundles.md: 8b5ab0c99282f6868fc3f70781e618af9a317c09
-2026-08-05-profile-plugin-bundles.zh.md: 2a685d68b3de9210488e26f8e6dd93dfc07f956c
+2026-08-05-profile-plugin-bundles.md: 2924b3cb445064fd47d82bcc94ec8d77ded5721b
+2026-08-05-profile-plugin-bundles.zh.md: b2287034010bcac1048bb385b2266f1bc75921da
@@ -12,7 +12,7 @@ The `dsh` launcher hardcoded its compositions: `base.cordis.yml` + `web.cordis.y

 Everything becomes a **profile**: a directory `$DSH_HOME/profiles/<name>` with a `package.json` (pnpm-managed out-of-tree plugin `dependencies` plus the profile manifest `dsh.profile` with its ordered `bundles` layer list) and a user `cordis.patch.yml`. A **bundle** is an npm package declaring `"dsh": { "bundle": { "patch": "./cordis.patch.yml" } }`; the two manifest kinds live under distinct `dsh.profile` / `dsh.bundle` keys so a package.json states which role it plays. The tree composes over an empty root by applying each bundle's patch in `dsh.profile.bundles` order, then the user layer, then `--patch` overlays, then flag patches — one `applyEntryPatches` call, identical for boot, flag derivation, and `--dump-config`.

-The shipped bundles are `@deepseek-ai/dsh-base` (shared core rows), `@deepseek-ai/dsh-web-app` (browser Host rows and Web runtime glue), and `@deepseek-ai/dsh-headless` (a direct one-shot runner over base, without web-app). `dsh web` is the Web-flag alias for `--profile web`; `dsh run [--profile <name>] "task"` owns one-shot execution and defaults to the headless profile; generic `dsh --profile <name>` boots without a task. Patch overlays use `--patch`. `dsh plugin --profile <name> <args...>` is a thin pnpm forwarder that initializes the profile and reconciles `dsh.profile.bundles` with installed bundle declarations; a package without a bundle declaration remains a plain dependency. [Headless as a direct core front door](2026-08-09-headless-direct-core-front-door.md) owns the headless composition contract.
+The shipped bundles are `@deepseek-ai/dsh-base` (shared core rows), `@deepseek-ai/dsh-web-app` (browser Host rows and Web runtime glue), and `@deepseek-ai/dsh-headless` (a direct one-shot runner over base, without web-app). `dsh web` is the Web-flag alias for `--profile web`; `dsh run [--profile <name>] "task"` owns one-shot execution and defaults to the headless profile; generic `dsh --profile <name>` boots without a task. Patch overlays use `--patch`. `dsh plugin --profile <name> <args...>` is a thin pnpm forwarder that initializes the profile and reconciles `dsh.profile.bundles` with installed bundle declarations; a package without a bundle declaration remains a plain dependency. [Headless as a direct core entry point](2026-08-09-headless-direct-core-entry-point.md) owns the headless composition contract.

 The [`dsh run` command decision](../feature/2026-08-08-dsh-run-headless-command.md) owns the one-shot grammar; this note owns the profile composition it selects.

@@ -12,7 +12,7 @@ Status: implemented

 一切都变成 **profile**：即目录 `$DSH_HOME/profiles/<name>`，其中包含一个 `package.json`（pnpm 管理的树外插件 `dependencies`，加上 profile manifest（元数据清单）`dsh.profile` 及其有序的 `bundles` 层列表）和一份用户 `cordis.patch.yml`。**组合包**（bundle）是声明了 `"dsh": { "bundle": { "patch": "./cordis.patch.yml" } }` 的 npm 包；两种 manifest 分别位于互不相同的 `dsh.profile` / `dsh.bundle` 键下，因此一份 package.json 能说明自己扮演哪种角色。配置树在空的根之上组合：按 `dsh.profile.bundles` 顺序应用每个组合包的 patch，然后是用户层，然后是 `--patch` overlay，最后是 flag patch——全部收敛为一次 `applyEntryPatches` 调用，启动、flag 派生与 `--dump-config` 使用完全相同的路径。

-随附的组合包是 `@deepseek-ai/dsh-base`（共享核心配置行）、`@deepseek-ai/dsh-web-app`（浏览器 Host 配置行与 Web 运行时粘合层）和 `@deepseek-ai/dsh-headless`（直接叠加在 base 上且不含 web-app 的一次性 runner）。`dsh web` 是携带 Web flag 家族的 `--profile web` 别名；`dsh run [--profile <name>] "task"` 负责一次性执行，默认使用 headless profile；通用的 `dsh --profile <name>` 启动 profile 而不携带任务。patch overlay 使用 `--patch`。`dsh plugin --profile <name> <args...>` 是一层薄薄的 pnpm 转发器，负责初始化 profile，并依据已安装包的组合包声明调和 `dsh.profile.bundles`；没有组合包声明的包保持为普通依赖。[Headless 作为直接 core 前门](2026-08-09-headless-direct-core-front-door.md)负责 headless 组合约定。
+随附的组合包是 `@deepseek-ai/dsh-base`（共享核心配置行）、`@deepseek-ai/dsh-web-app`（浏览器 Host 配置行与 Web 运行时粘合层）和 `@deepseek-ai/dsh-headless`（直接叠加在 base 上且不含 web-app 的一次性 runner）。`dsh web` 是携带 Web flag 家族的 `--profile web` 别名；`dsh run [--profile <name>] "task"` 负责一次性执行，默认使用 headless profile；通用的 `dsh --profile <name>` 启动 profile 而不携带任务。patch overlay 使用 `--patch`。`dsh plugin --profile <name> <args...>` 是一层薄薄的 pnpm 转发器，负责初始化 profile，并依据已安装包的组合包声明调和 `dsh.profile.bundles`；没有组合包声明的包保持为普通依赖。[Headless 作为直接 core 入口](2026-08-09-headless-direct-core-entry-point.md)负责 headless 组合约定。

 [`dsh run` 命令决策](../feature/2026-08-08-dsh-run-headless-command.md)负责一次性语法；本 Agent Note 负责该语法所选择的 profile 组合。

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-08-per-preset-standing-mounts.md
+2026-08-08-per-preset-standing-mounts.md: 834d645f5f293a2e137b8faf662e301f1e8bb971
+2026-08-08-per-preset-standing-mounts.zh.md: 45ce0f4e7dec28e5bf807898dc9cdbf32b8e4eb5
@@ -0,0 +1,32 @@
+# Agent Note: Per-preset standing mounts over a scope parent chain
+
+Status: implemented
+
+English | [中文](2026-08-08-per-preset-standing-mounts.zh.md)
+
+## Problem
+
+Per-session preset mounts made the model-facing registry surface per-agent while three independent host readers still assumed it was static: cold `session.history` found no presenters (every card silently degraded to the generic renderer — indistinguishable from "tool has no presenter"), the projections block dropped preset-registered keys (clients treat an omitted key as capability absence and CLEAR the row), and the TypeRT gateway resolved `goals` on the host root (`service-unavailable`). Patching each reader individually traded one silent degradation for another: resuming to reach presenters flipped the projections fold from detached to live and wiped the token counts instead.
+
+## Decision
+
+A preset is one composition per PROCESS, not one per session. The roster mounts it once under a synthetic standing scope; each agent joins by binding its scope key to the mount's (`bindScopeParent(agentKey, standingKey)`). Two `dsh-scope` mechanisms carry everything: registration views walk the parent chain (`agent → preset → global`, nearest shadowing farthest), and scoped dispatch admits listeners tagged with an ancestor of the carrier key — upward only, so a sibling preset's listeners stay deaf.
+
+## Consequences
+
+Standing mounts fix the class, not the instances: the registrations a reader needs exist for the process lifetime, keyed by preset id, no agent required. What made it cheap
+
+- The stateful preset plugins (`plan-mode`, `token-meter`, `compact-basic`, `tasks-local`) already key state by `Session`/`Agent` — they predate presets. Sharing one instance is a return to their design, not a rewrite.
+- Preset ymls are unchanged: one mount per preset = one Entry per preset, whose entry-local realms (`isolate: <name>: true`) keep two presets' same-named services apart exactly as they kept two sessions' apart.
+- A shared realm label was NOT an option: `provide()` throws on a second registration under the same realm symbol, so labels pool the REALM, never the instance — a per-session world sharing a label crashes the second mount.
+
+## Load-bearing details
+
+- **Standing mounts hang off the service's untraced `selfCtx`.** A method invoked through the traceable proxy sees `this.ctx` rebound to the caller with a shadow; reflect resolution for every fiber in a subtree minted from it starts at the shadow's fiber, so entries fail on services their own `inject` declares (`cannot get property "tools" without inject` while the entry's store holds it). The `tasks-local` selfCtx precedent, now with a second consumer.
+- **A settled mount serves until its composition file's stamp changes.** The composition a running session joined must survive its file changing or disappearing; each generation records the file's stamp (mtime + size) and a session that finds it stale starts the next generation, so file edits — the only composition editor once authoring became copy-only — reach later sessions without any authoring call dropping the pointer. Joined sessions keep their generation, and superseded generations are reclaimed only by whole-tree teardown — deliberate, bounded by edit frequency, recorded in the package's Known Limitations.
+- **`peek()` stays chain-blind.** Restrictions and guards address one scope's own contributions; only registration VIEWS inherit. Restrictions along the chain intersect (any scope may mask a global-surface name for everything nested inside it).
+- **Re-linking runs only through the `ScopeParentBinding` the mount's one bind returned** — the roster holds it privately, so the blank-session recompose path is the sole re-link and no other caller can move a composed agent; it stays valid only while nothing produced under the old parent is retained, which the holder must uphold because the relation cannot see session logs.
+
+## Alternatives considered
+
+Resume-on-read (wipes detached projections), a host-plane presenter table plus a block completeness flag (fixes two readers, leaves the class), per-session template mounts (duplicates every instance to serve pure functions). Kept for the record: the gateway-facing `goals` domain stays host-plane regardless — a Remote method whose receiver comes from a generated descriptor resolves on the host, which is the `bash-env` host-plane criterion read from the consuming side.
@@ -0,0 +1,32 @@
+# Agent Note: Per-preset standing mounts over a scope parent chain
+
+Status: implemented
+
+[English](2026-08-08-per-preset-standing-mounts.md) | 中文
+
+## Problem
+
+按会话挂载 preset 让面向模型的注册面变成按 agent 的，而三个独立的宿主读取方仍然假设它是静态的：冷读 `session.history` 找不到 presenter（每张卡都静默退化成通用渲染器——与「工具本无 presenter」无法区分）、投影块丢掉 preset 注册的键（客户端把缺失键当作能力不存在并**清掉**该行）、TypeRT 网关在宿主根上解析 `goals`（`service-unavailable`）。逐个读取方打补丁只是拿一种静默降级换另一种：为拿到 presenter 而 resume，会把投影折叠从 detached 翻到 live，token 计数随之被抹掉。
+
+## Decision
+
+一个 preset 是**每进程**一份组装，而不是每会话一份。roster 在一个合成常驻 scope 下挂载它一次；每个 agent 通过把自己的 scope key 绑定到挂载的 key（`bindScopeParent(agentKey, standingKey)`）加入。两条 `dsh-scope` 机制承载了一切：注册视图沿父链解析（`agent → preset → global`，近者遮蔽远者），带作用域的分发对标签为载体键祖先的监听器放行——只向上，兄弟 preset 的监听器保持失聪。
+
+## Consequences
+
+常驻挂载修的是这一类问题而非其中的个例：读取方需要的注册在进程生命周期内始终存在，按 preset id 索引，不需要任何 agent。让它便宜的原因：
+
+- 有状态的 preset 插件（`plan-mode`、`token-meter`、`compact-basic`、`tasks-local`）本就按 `Session`/`Agent` 分键存状态——它们早于 preset 存在。共享一份实例是回归其设计，不是改写。
+- preset 的 yml 不变：每 preset 挂一次 = 每 preset 一个 Entry，其 entry 本地 realm（`isolate: <name>: true`）让两个 preset 的同名服务互不相干，正如它从前隔开两个会话。
+- 共享 realm label **不是**选项：`provide()` 对同一 realm 符号下的第二次注册直接抛错，label 池化的是 REALM 而非实例——按会话挂载的世界里共享 label 会让第二次挂载崩溃。
+
+## Load-bearing details
+
+- **常驻挂载挂在服务未追踪的 `selfCtx` 上。** 经 traceable 代理调用的方法看到的 `this.ctx` 被重绑到调用方并携带 shadow；从它派生的子树里每个 fiber 的 reflect 解析都从 shadow 的 fiber 起步，entry 会在自己 `inject` 声明的服务上失败（`cannot get property "tools" without inject`，而它的 store 里明明有）。`tasks-local` 的 selfCtx 先例，如今有了第二个消费者。
+- **挂载一旦成功即持续供职，直到组装文件的 stamp 变化。** 运行中会话加入的组装必须在其文件被修改或删除后继续存活；每个代际记录文件 stamp（mtime + 大小），发现过期的会话开启下一个代际，因此文件编辑——创作改为仅复制之后唯一的组装编辑器——无需任何创作调用丢弃指针即可达到后续会话。已加入的会话保持其代际，被替代的代际只由整树卸载回收——刻意为之，上限取决于编辑频率，已记入包的 Known Limitations。
+- **`peek()` 保持不看链。** 限制与守卫定位的是单个作用域**自己**的贡献；只有注册**视图**沿链继承。链上的限制求交（链上任一作用域都可为嵌套其内的一切遮蔽某个全局面名字）。
+- **重新认父只能经由挂载首绑返回的 `ScopeParentBinding`**——roster 私藏该句柄，空白会话 recompose 因此是唯一的重链路径，其他调用方无法挪动已组合的 agent；其合法性仍以旧父之下产出一概不被保留为前提，由持有方保证，因为该关系看不见会话日志。
+
+## Alternatives considered
+
+冷读时 resume（抹掉 detached 投影）、宿主面 presenter 表加投影块完整性标志（修两个读取方、留下这一类）、每会话模板挂载（为了服务纯函数而复制每一份实例）。留档：面向网关的 `goals` 域无论如何留在宿主平面——Remote 方法的接收者来自生成的 descriptor、在宿主上解析，这正是 `bash-env` 宿主平面判据从消费侧读出的样子。
@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-09-headless-direct-core-entry-point.md
+2026-08-09-headless-direct-core-entry-point.md: 49afe2993de7302adbedcdf9e8e2347d6424ee2a
+2026-08-09-headless-direct-core-entry-point.zh.md: 73c1cbe5ac777025f63f46751b1d5ccebbfe9676
@@ -1,22 +1,22 @@
-# Agent Note: headless is a direct core front door
+# Agent Note: headless is a direct core entry point

 Status: implemented

-English | [中文](2026-08-09-headless-direct-core-front-door.zh.md)
+English | [中文](2026-08-09-headless-direct-core-entry-point.zh.md)

 ## Problem

 The `headless` product contract is one local task with final assistant text on stdout, a success-sensitive exit code, empty stderr on success, and no listening port. A composition containing Workspace Host services, ApiProxy, HTTP, the Web runtime, or browser plugins contradicts that contract and makes local completion depend on an unrelated transport tree.

-The direct front door still needs the same deployment model state as Web-created Agents. A separate provider/model default would give one deployment two answers, while deriving completion before the Agent and Session persistence are quiescent permits stdout and the exit code to observe incomplete state.
+The direct entry point still needs the same deployment model state as Web-created Agents. A separate provider/model default would give one deployment two answers, while deriving completion before the Agent and Session persistence are quiescent permits stdout and the exit code to observe incomplete state.

 ## Decision

 The shipped `headless` profile contains `dsh-base` and `dsh-headless`. The headless bundle supplies its persona and tool mode, disables HMR, mounts the Code Mode worker explicitly, and inserts `headless-runner`. Its tree contains no `@deepseek-ai/dsh-host-*` package, ApiProxy, HTTP server, Web runtime, or browser client. Code Mode and Session persistence are one-shot Agent capabilities independent of Web presentation.

-`headless-runner` is a direct core front door. After Loader settlement, it reads `ctx.agentDefaultModel.currentSelection()`, creates a fresh persisted Agent through `ctx.agents.create`, installs that `ModelSelection` in the Agent scope, waits for startup quiescence, anchors the Session sequence, submits one ordinary user message, and waits for quiescence again. It awaits `ctx.sessions.flush`, folds its durable event interval for the last non-empty assistant text and final `turn/end` reason, writes the text plus one newline to stdout, and requests bounded launcher shutdown with exit 0 exactly when the reason is `completed`. A terminal `error` reason writes its durable code and message to stderr; unexpected driver failures also use stderr and exit 1.
+`headless-runner` is a direct core entry point. After Loader settlement, it reads `ctx.agentDefaultModel.currentSelection()`, creates a fresh persisted Agent through `ctx.agents.create`, installs that `ModelSelection` in the Agent scope, waits for startup quiescence, anchors the Session sequence, submits one ordinary user message, and waits for quiescence again. It awaits `ctx.sessions.flush`, folds its durable event interval for the last non-empty assistant text and final `turn/end` reason, writes the text plus one newline to stdout, and requests bounded launcher shutdown with exit 0 exactly when the reason is `completed`. A terminal `error` reason writes its durable code and message to stderr; unexpected driver failures also use stderr and exit 1.

-`@deepseek-ai/dsh-agent-default-model` owns the transport-independent default used for an Agent without a session-local selection. `AgentDefaultModelService` provides `ctx.agentDefaultModel` and registers the `agent-default-model` Settings section. Composition config supplies `{provider, model}`; user settings may also supply `reasoningEffort`. `currentSelection()` returns the live complete selection and `saveSelection()` writes it as a complete section, so a selection without an effort clears any stored effort. `dsh-base` supplies the composition entry. Direct and ApiProxy front doors consume this service; ApiProxy alone owns session-local precedence, model validation, and persistence of accepted Web selections.
+`@deepseek-ai/dsh-agent-default-model` owns the transport-independent default used for an Agent without a session-local selection. `AgentDefaultModelService` provides `ctx.agentDefaultModel` and registers the `agent-default-model` Settings section. Composition config supplies `{provider, model}`; user settings may also supply `reasoningEffort`. `currentSelection()` returns the live complete selection and `saveSelection()` writes it as a complete section, so a selection without an effort clears any stored effort. `dsh-base` supplies the composition entry. Direct and ApiProxy entry points consume this service; ApiProxy alone owns session-local precedence, model validation, and persistence of accepted Web selections.

 `loadProfile` recognizes the exact installation-owned headless tuple (`dsh-base`, `dsh-web-app`, `dsh-headless`) and normalizes it to the shipped headless template while preserving every other manifest field. Extra, missing, or reordered bundle lists are user-owned and remain untouched.

@@ -31,7 +31,7 @@ Package tests use the real Session store and Agent registry around a scripted Ag
 | Alternative | Contract mismatch |
 |---|---|
 | Keep `dsh-web-app` but suppress its observation line | The process still opens a port and carries the Host, Web, and browser trees. |
-| Build a Host-only one-shot bundle around ApiProxy | ApiProxy is a client protocol gateway; a local one-shot front door has no client boundary. |
+| Build a Host-only one-shot bundle around ApiProxy | ApiProxy is a client protocol gateway; a local one-shot entry point has no client boundary. |
 | Use `InProcessApiClient` for product-level protocol coverage | Product execution would depend on an unrelated protocol solely to exercise that protocol. |
 | Give headless a separate provider/model config | Direct and Web creation would have independent defaults and persistence. |
 | Omit Code Mode and Session persistence | Both capabilities belong to one-shot Agent execution rather than Web presentation. |
@@ -2,13 +2,13 @@

 Status: implemented

-[English](2026-08-09-headless-direct-core-front-door.md) | 中文
+[English](2026-08-09-headless-direct-core-entry-point.md) | 中文

 ## 问题

 `headless` 的产品约定是一个本地任务：最终 assistant 文本写入 stdout，退出状态反映成功与否，成功时 stderr 为空，并且不打开监听端口。包含 Workspace Host 服务、ApiProxy、HTTP、Web 运行时或浏览器插件的组合违背这一约定，也使本地完成状态依赖无关的传输树。

-直接前门仍需要与 Web 所创建 Agent 相同的部署模型状态。独立的提供方／模型默认值会让同一部署产生两种答案，而在 Agent 与会话持久化完全停稳之前推导完成状态，会让 stdout 与退出状态观察到不完整状态。
+直接入口仍需要与 Web 所创建 Agent 相同的部署模型状态。独立的提供方／模型默认值会让同一部署产生两种答案，而在 Agent 与会话持久化完全停稳之前推导完成状态，会让 stdout 与退出状态观察到不完整状态。

 ## 决策

@@ -16,7 +16,7 @@ Status: implemented

 `headless-runner` 是直接使用核心服务的入口。Loader 完全加载后，它读取 `ctx.agentDefaultModel.currentSelection()`，通过 `ctx.agents.create` 创建一个新的持久化 Agent，在 Agent 作用域中安装该 `ModelSelection`，等待启动工作完全停稳，锚定会话事件序号，提交一条普通用户消息，再次等待完全停稳。随后，它等待 `ctx.sessions.flush`，折叠自身持有的持久事件区间，以取得最后一条非空 assistant 文本和最终 `turn/end` 结束原因，将文本连同一个换行写入 stdout，并且仅在结束原因为 `completed` 时请求启动器以退出状态 0 有界关闭。结束原因为 `error` 时，其持久化错误码与消息写入 stderr；驱动器的意外失败也写入 stderr 并以 1 退出。

-`@deepseek-ai/dsh-agent-default-model` 拥有与传输无关的默认值，供没有会话级选择的 Agent 使用。`AgentDefaultModelService` 提供 `ctx.agentDefaultModel` 并注册 `agent-default-model` Settings 分节。组合配置提供 `{provider, model}`，用户设置还可以提供 `reasoningEffort`。`currentSelection()` 返回当前的完整选择，`saveSelection()` 则写入完整分节，因此不含强度的选择会清除已存强度。`dsh-base` 提供组合条目。直接前门与 ApiProxy 前门均消费该服务；只有 ApiProxy 负责会话级优先级、模型校验与已接受 Web 选择的持久化。
+`@deepseek-ai/dsh-agent-default-model` 拥有与传输无关的默认值，供没有会话级选择的 Agent 使用。`AgentDefaultModelService` 提供 `ctx.agentDefaultModel` 并注册 `agent-default-model` Settings 分节。组合配置提供 `{provider, model}`，用户设置还可以提供 `reasoningEffort`。`currentSelection()` 返回当前的完整选择，`saveSelection()` 则写入完整分节，因此不含强度的选择会清除已存强度。`dsh-base` 提供组合条目。直接入口与 ApiProxy 入口均消费该服务；只有 ApiProxy 负责会话级优先级、模型校验与已接受 Web 选择的持久化。

 `loadProfile` 识别安装过程拥有的精确 headless 元组（`dsh-base`、`dsh-web-app`、`dsh-headless`），将其规范化为随附的 headless 模板，并保留 manifest（元数据清单）的其他所有字段。带额外项、缺少项或顺序不同的组合包列表归用户所有，保持不变。

@@ -31,7 +31,7 @@ Status: implemented
 | 替代方案 | 约定不匹配之处 |
 |---|---|
 | 保留 `dsh-web-app`，但隐藏观察行 | 进程仍会打开端口并携带 Host、Web 与浏览器插件树。 |
-| 围绕 ApiProxy 构建纯 Host 一次性组合包 | ApiProxy 是客户端协议网关，而本地一次性前门没有客户端边界。 |
+| 围绕 ApiProxy 构建纯 Host 一次性组合包 | ApiProxy 是客户端协议网关，而本地一次性入口没有客户端边界。 |
 | 使用 `InProcessApiClient` 实现产品级协议覆盖 | 产品执行会仅为测试无关协议而依赖该协议。 |
 | 为 headless 单独提供提供方／模型配置 | 直接创建与 Web 创建会拥有彼此独立的默认值和持久化。 |
 | 省略 Code Mode 与会话持久化 | 两项能力都属于一次性 Agent 执行，而不是 Web 呈现。 |
@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-09-layered-skill-registry.md
+2026-08-09-layered-skill-registry.md: 3f092cfb4b722e3dd51fa4dc46c620259eaffa39
+2026-08-09-layered-skill-registry.zh.md: 38b17329c8d46ee9bbd0863f3fae7cf6be39aa75
@@ -0,0 +1,39 @@
+# Agent Note: The skill registry is host-held and layered per scope
+
+Status: implemented
+
+English | [中文](2026-08-09-layered-skill-registry.zh.md)
+
+## Problem
+
+The agent-preset stack moved the whole skill capability — registry, local provider, and the `skill` tool — into each preset's `isolate` realm, because "which skills an agent has" is an agent-plane choice. That framing conflated two different questions: which skills a *deployment* supplies, and whether an *agent* consumes them. A repository plugin's prepared wrapper declares `inject: ['skills']` and mounts its skill root as a host-plane provider; with no host registry composed in the web and headless profiles, that wrapper waited forever and the repository-plugin e2e hung, which was bypassed at the time by dropping the fixture's skill root. A per-preset realm registry also made the gateway's skill listing depend on a live agent — a cold session's `/` popup had no registry to read at all.
+
+The tools registry never had this problem: it is one host singleton layered per scope over `dsh-scope`, so deployment-level tools (MCP servers, plugin entries) register globally while a preset's rows register into that preset's layer.
+
+## Decision
+
+`SkillService` adopts the same shape. It holds `ScopedLayers<SkillLayer>`; `registerProvider()` and `register()` file into the layer of the calling context's scope, so host rows and repository plugins land in the global layer while a preset's `skill-local` — mounted by the standing composition, whose context carries the preset's scope key — lands in that preset's layer. Provider names are unique per layer rather than process-wide, which is what lets every preset mount its own `local` provider.
+
+Reads take the viewing scope through `SkillViewOptions` (the calling agent, which is its own scope key). The registry merges the global layer with the scope's chain: **the nearest layer wins a duplicate name outright, and rank decides duplicates only within one layer** — the tools registry's shadowing rule. Rank-pooling across layers was considered and rejected: ranks were designed to order sources that know about each other, and under a global pool a later-installed repository plugin could silently displace a preset's own same-named skill by registration-order tiebreak, changing a preset's behavior remotely. Nearest-wins keeps a composition's behavior decided by its author.
+
+Discovery caches are keyed by the resolved scope chain plus one revision counter, so a blank-session recompose — which re-parents the agent's scope key without touching the registry — is visible to the next read.
+
+The composition moves with it: the web-app bundle re-enables the base `skill` registry row (only `skill-local` and `tool-skill` stay preset-owned), and preset compositions drop their `isolate: skills` realm for bare rows over the host registry. The gateway's skills domain reads the host registry in the presenter scope — the live agent, else the recorded preset's standing key — so a cold session lists the catalog its composition actually serves instead of failing; the `serviceFor` branch stays for compositions that still realm-mount their own registry.
+
+## Consequences
+
+**A deployment-level skill reaches every preset-composed session that mounts `tool-skill`.** The repository-plugin e2e's skill root and assertions are restored; the shipped-Web e2e proves the badge row (the same host-registration shape) merges into a standard-preset agent's catalog while the host view stays global-only.
+
+**Layer visibility and consumption stay separate choices.** A core-web agent can read the global layer in principle, but composes no `skill` tool — whether an agent has skills at all remains the preset's decision, made by mounting or omitting `tool-skill`.
+
+**Provider options are still the borrowed caller object.** `SkillViewOptions` extends `SkillLookupOptions`; the registry consumes `scope` and providers read only their own contract from the same readonly object, preserving the existing borrow-identity guarantee.
+
+**The TUI profile is unaffected.** With every row at host, there is exactly one (global) layer and the merged view equals the old single-registry view, ranks and all.
+
+**Shadowing across layers is silent.** Within a layer the loser is logged as before; a nearer layer replacing a farther name follows the tools registry's convention and logs nothing. The registry still exposes no API to inspect shadowed definitions.
+
+## Alternatives considered
+
+**Rank pool across all visible layers.** Faithful to the single-registry precedence, but cross-layer ties break on registration order (boot-time providers always beat standing mounts), and a preset's own skill could be displaced by a deployment change it never sees. Rejected for composition stability; see Decision.
+
+**Keep per-preset realm registries and deliver repository skills as directories a preset's provider scans.** Leaves the wrapper's `inject: ['skills']` contract broken (or forks the wrapper per profile), duplicates discovery configuration into every preset, and still gives cold sessions nothing to read. Rejected.
@@ -0,0 +1,39 @@
+# Agent Note：skill 注册表由宿主持有并按 scope 分层
+
+Status: implemented
+
+[English](2026-08-09-layered-skill-registry.md) | 中文
+
+## 问题
+
+agent-preset stack 曾把整个 skill 能力——注册表、本地提供方和 `skill` 工具——搬进每个 preset 的 `isolate` realm，理由是"agent 拥有哪些 skill"属于 agent 平面的选择。这一框架混淆了两个不同的问题：*部署*供给哪些 skill，与*agent*是否消费它们。repository 插件的 prepared wrapper 声明 `inject: ['skills']` 并把它的 skill 根目录挂载为宿主平面的提供方；web 与 headless profile 不再组合宿主注册表后，该 wrapper 永远等待，repository-plugin e2e 因而挂死，当时通过删掉 fixture 的 skill 根目录绕过。按 preset 的 realm 注册表还让网关的 skill 列表依赖存活 agent——冷会话的 `/` 弹窗根本没有注册表可读。
+
+工具注册表从未有过这个问题：它是一个宿主单例，基于 `dsh-scope` 按 scope 分层，因此部署级工具（MCP 服务器、插件 entry）注册进全局层，preset 的行注册进该 preset 的层。
+
+## 决定
+
+`SkillService` 采用同一形态。它持有 `ScopedLayers<SkillLayer>`；`registerProvider()` 与 `register()` 落入调用方上下文 scope 对应的层——宿主行与 repository 插件落入全局层，preset 的 `skill-local`（由常驻组合挂载，其上下文携带该 preset 的 scope key）落入该 preset 的层。提供方名称在每层内唯一而非进程级唯一，这正是让每个 preset 都能挂载自己的 `local` 提供方的前提。
+
+读取通过 `SkillViewOptions` 携带观察 scope（调用中的 agent，agent 本身就是自己的 scope key）。注册表将全局层与该 scope 的链合并：**最近层直接赢得重名，rank 只在单层内裁决重名**——即工具注册表的遮蔽规则。曾考虑跨层 rank 合池并予以否决：rank 的设计前提是各来源彼此知情；在全局池下，后安装的 repository 插件可能凭注册顺序平手规则静默顶掉 preset 自带的同名 skill，远程改变 preset 的行为。最近层优先让组合的行为由其作者决定。
+
+发现缓存以解析后的 scope 链加一个修订计数为键，因此空会话重组——只重设 agent scope key 的父级、不触碰注册表——对下一次读取立即可见。
+
+组合随之调整：web-app bundle 重新启用 base 的 `skill` 注册表行（只有 `skill-local` 与 `tool-skill` 仍归 preset），preset 组合拆掉 `isolate: skills` realm，改为直接落在宿主注册表上的平铺行。网关的 skills 域以 presenter scope 读取宿主注册表——存活 agent，否则记录在案的 preset 的 standing key——冷会话由此列出其组合真正供给的目录而不再报错；`serviceFor` 分支保留，兼容仍以 realm 自挂注册表的组合。
+
+## 影响
+
+**部署级 skill 会到达每个挂载 `tool-skill` 的 preset 会话。**repository-plugin e2e 的 skill 根目录与断言已恢复；shipped-Web e2e 证明 badge 行（同一种宿主注册形态）汇入 standard preset agent 的目录，而宿主视图保持仅全局。
+
+**层可见性与消费仍是两个独立选择。**core-web agent 原则上可读全局层，但不组合 `skill` 工具——agent 是否拥有 skill 依旧由 preset 通过挂载或省略 `tool-skill` 决定。
+
+**提供方选项仍是借用的调用方对象。**`SkillViewOptions` 扩展 `SkillLookupOptions`；注册表消费 `scope`，提供方只从同一个只读对象中读取自己的契约，保持既有的借用恒等保证。
+
+**TUI profile 不受影响。**所有行都在宿主时只有一个（全局）层，合并视图等于旧的单注册表视图，rank 行为不变。
+
+**跨层遮蔽是静默的。**层内败者照旧记录日志；较近层顶替较远层的名称沿用工具注册表的惯例，不记录。注册表仍不提供检查被遮蔽定义的 API。
+
+## 曾考虑的替代方案
+
+**跨全部可见层的 rank 合池。**忠实于单注册表的优先级，但跨层平手按注册顺序裁决（启动期提供方永远赢过常驻挂载），preset 自带 skill 可能被它看不见的部署变更顶掉。因组合稳定性否决；见"决定"。
+
+**保留按 preset 的 realm 注册表，把 repository skill 作为目录交给 preset 的提供方扫描。**wrapper 的 `inject: ['skills']` 契约仍然破损（或者按 profile 分叉 wrapper），发现配置在每个 preset 里重复，冷会话依旧无处可读。否决。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/bug-fix/2026-07-20-error-cause-chain-diagnostics.md
-2026-07-20-error-cause-chain-diagnostics.md: 32716b5a68b3b73bded47633eca95995cdbfc586
-2026-07-20-error-cause-chain-diagnostics.zh.md: 9911c32b6d68c1f5569a1fceadb65e23c6586594
+2026-07-20-error-cause-chain-diagnostics.md: b80dd08d79a57738a6eef2f8638b0336ca80d4bf
+2026-07-20-error-cause-chain-diagnostics.zh.md: 914e983d7ea81eef8bca8fd5aa3791f2f44d4080
@@ -9,7 +9,7 @@ English | [中文](2026-07-20-error-cause-chain-diagnostics.zh.md)
 A TUI run against an unreachable DeepSeek endpoint failed with the single notice `fetch failed` and no further detail. Two independent gaps produced that dead end:

 1. undici's `fetch` wraps every transport failure (DNS, refused connection, TLS, proxy) in a bare `TypeError: fetch failed` whose actionable detail — `ECONNREFUSED`, `bad port`, the Happy Eyeballs AggregateError — lives on `error.cause`. Every diagnostic boundary in the harness rendered only `error.message` (or `String(error)`, which is equivalent for Errors), so the wrapper masked the diagnosis in the TUI notice, the durable `turn/end` reason, and every logger line.
-2. The readline front door (`dsh-stdio`) rendered no failure reason at all: a `turn/end` with `reason.kind === 'error'` printed nothing but the next `> ` prompt, so the same failure in `demo:repl` was pure silence.
+2. The readline entry point (`dsh-stdio`) rendered no failure reason at all: a `turn/end` with `reason.kind === 'error'` printed nothing but the next `> ` prompt, so the same failure in `demo:repl` was pure silence.

 ## Decision

@@ -9,7 +9,7 @@ Status: implemented
 TUI 连接不可达的 DeepSeek 端点时，失败只显示一条 `fetch failed` 通知，没有任何进一步细节。两个独立缺口共同造成了这个死胡同：

 1. undici 的 `fetch` 把所有传输层失败（DNS、连接被拒、TLS、代理）包装成裸的 `TypeError: fetch failed`，可操作的细节——`ECONNREFUSED`、`bad port`、Happy Eyeballs 的 AggregateError——都在 `error.cause` 上。harness 里的每个诊断边界都只渲染 `error.message`（或对 Error 等价的 `String(error)`），于是包装层在 TUI 通知、持久化的 `turn/end` reason 和所有日志行里都掩盖了诊断信息。
-2. readline 前门（`dsh-stdio`）完全不渲染失败原因：`reason.kind === 'error'` 的 `turn/end` 只打印下一个 `> ` 提示符，同样的失败在 `demo:repl` 里就是纯粹的沉默。
+2. readline 入口（`dsh-stdio`）完全不渲染失败原因：`reason.kind === 'error'` 的 `turn/end` 只打印下一个 `> ` 提示符，同样的失败在 `demo:repl` 里就是纯粹的沉默。

 ## 决策

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/bug-fix/2026-08-09-broken-preset-roster-rows.md
+2026-08-09-broken-preset-roster-rows.md: fef6a183b10f98b8ae9d2b42701380c69bc83462
+2026-08-09-broken-preset-roster-rows.zh.md: 196bcf4ef16325a1d7692d2ea13d9fa683d500f4
@@ -0,0 +1,33 @@
+# Agent Note: Broken presets are roster rows, not gaps
+
+Status: implemented
+
+English | [中文](2026-08-09-broken-preset-roster-rows.zh.md)
+
+## Problem
+
+With files as the only composition editor, hand-edit damage had two failure shapes and both were silent until the worst moment. A preset whose `agent.cordis.yml` no longer parsed listed as a perfectly ordinary row — selectable, copyable, settable as the default — and failed only when the next session tried to mount it; set as default, every new session failed to start. A directory whose composition file was deleted outright vanished from the roster while still occupying its id on disk: `copy` refused the name with "delete the existing preset first" and `remove` answered "not found" — two contradictory errors with no way out short of hand-deleting the directory.
+
+## Decision
+
+Discovery owns health, and a damaged directory is a **roster row carrying a `broken` reason**, never a gap. `scanRoot` treats every directory whose name is a usable preset id as a preset slot: composition missing → broken ("still occupies the id; delete it or restore the file"), composition unreadable/unparsable/not-a-list-of-named-rows → broken with the parser's first line. The shape check parses with the loader's own `entryListSchema` (the `!!js` dialect), so health can never call broken what the loader would accept; directories whose names fail `PRESET_ID` are skipped outright, because no copy could ever collide with them. `broken` rides `AgentPreset`, the `agentPreset.list` wire entry, and the UI row. Mounting paths (`mount`/`recompose`/`standingKeyFor`) refuse a broken preset up front via `resolveMountable` with the discovery-reported reason; `resolve` still answers (delete/read/report need the row), and `copy`'s roster check now sees ghosts, which turns the "already exists" refusal actionable — the broken card to delete is on the same page.
+
+Surfaces split by their job: the management section renders broken rows as marked cards (red border, Broken badge, verbatim reason, body and duplicate disabled, location/delete kept on custom rows — the files are the fix, delete is the ghost's way out; shipped broken rows lose the viewer too), while both pickers (General row, new-session chip) drop broken presets entirely via `presetOptions` — they choose the NEXT session's composition, and offering one that cannot compose only defers the failure.
+
+## Consequences
+
+- The ghost dead end is gone end to end: the directory lists broken, its delete clears it, and the freed id is immediately claimable (covered by unit, component, and e2e tests).
+- A default that later breaks still fails the session start loudly — the pickers hide broken rows, but nothing rewrites a stored default; `resolveMountable`'s early refusal is the same message every unloadable shape gets, instead of loader-dependent errors.
+- Health runs on every `list()`: one read+parse per preset per roster read, accepted for the same reason unmemoized discovery was — rosters are small and freshness is the contract.
+- Copying broken is refused in the UI only (disabled with reason); the host keeps `copy` shape-agnostic. A broken source yields an equally broken, equally visible copy — no capability is gained, and the host-side refusal would have needed its own error vocabulary for no journey that survives the disabled button.
+
+## Load-bearing details
+
+- **`PRESET_ID` moved to `types.ts`** so discovery and authoring share one containment vocabulary; authoring re-exports it unchanged.
+- **The reason is one line.** js-yaml appends a multi-line code-frame snippet; the roster card is not a terminal, so `compositionProblem` keeps the first line.
+- **Two mount.spec races were left untouched deliberately**: `ensureStanding` is still reachable with a preset resolved just before deletion (the private-path tests), and its stamp/unstampable semantics are unchanged — the health check happens before, in the public route.
+- **Creator-mode guidance rides the same PR**: the `cordis` preset's persona now forbids editing the shipped install (corrupting `cordis` would disable the mode itself) and points authoring at `${DSH_HOME:-$HOME/.dsh}/.agent-presets/<id>/`; its skill teaches `preset.yml` metadata, the copy-first workflow, the one-escalation sandbox reality (the preset root lies outside the session workspace), and honest verification (the agent cannot start sessions; the settings page's red marking is the user's check). Verified live: asked to edit the shipped `cordis` composition directly, the composed agent refuses citing both rules and offers the copy path; asked for a real preset, it lands it under `$DSH_HOME`, batches writes into one escalation, self-checks with the loader dialect, and hands verification to the user.
+
+## Alternatives considered
+
+Hiding broken presets but refusing the id at copy time with a better message: still no way to clear the ghost from any surface. Validating deep (resolving every row's module at list time): the mount already owns that failure with rollback, and per-row imports on every roster read would be neither cheap nor more actionable. Blocking `settings` writes naming a broken default: the settings domain is generic and the roster is a live directory — a name absent or broken now may be valid by the next session, and the mount's loud failure is the enforcement that owns the moment.
@@ -0,0 +1,33 @@
+# Agent Note：损坏的 preset 是名单行，不是空缺
+
+Status: implemented
+
+[English](2026-08-09-broken-preset-roster-rows.md) | 中文
+
+## 问题
+
+文件成为唯一的组装编辑器之后，手动编辑造成的损坏有两种形态，且都要拖到最糟的时刻才暴露。`agent.cordis.yml` 解析不了的 preset 在名单上是一张完全正常的行——可选择、可复制、可设为默认——直到下一个会话尝试挂载才失败；一旦被设为默认，所有新会话都无法启动。组装文件被整个删掉的目录则从名单上消失，却仍在磁盘上占着它的 id：`copy` 以「先删除既有 preset」拒绝这个名字，`remove` 却回答「找不到」——两条互相矛盾的错误，除了手动删目录别无出路。
+
+## 决定
+
+发现过程负责健康，受损目录是**携带 `broken` 原因的名单行**，绝不是空缺。`scanRoot` 把名字是可用 preset id 的每个目录都当作一个 preset 槽位：组装缺失 → broken（「仍占着该 id；删除目录或恢复文件」），组装不可读/解析失败/不是具名行列表 → broken 并携带解析器的首行。形状检查用加载器自己的 `entryListSchema`（含 `!!js` 的方言）解析，因此健康检查绝不会把加载器接受的组装叫作损坏；名字不符合 `PRESET_ID` 的目录直接跳过，因为复制永远不可能与之相撞。`broken` 依次落在 `AgentPreset`、`agentPreset.list` 的线上条目和 UI 行上。挂载路径（`mount`/`recompose`/`standingKeyFor`）经 `resolveMountable` 用发现时记下的原因在前置拒绝；`resolve` 照样应答（删除/读取/上报都需要这一行），而 `copy` 的名单检查现在看得见幽灵，让「已存在」的拒绝变得可操作——要删的损坏卡片就在同一页上。
+
+界面按职责分开：管理区把损坏行渲染为标记卡片（红边、「已损坏」徽记、原样展示原因、卡片主体与复制禁用，自定义行保留位置与删除——文件正是修复处，删除正是幽灵的出路；损坏的内置行连查看器也不给），而两个选择器（通用设置行、新会话 chip）经 `presetOptions` 完全不列损坏的 preset——它们选的是下一个会话的组装，端出无法组装的选项只会推迟失败。
+
+## 后果
+
+- 幽灵死路端到端消除：目录以损坏行列出，删除即清掉，释放的 id 立刻可用（单测、组件测试与 e2e 各自覆盖）。
+- 事后才损坏的默认值仍会在会话启动处大声失败——选择器隐藏损坏行，但没有任何东西改写已存的默认；`resolveMountable` 的前置拒绝让每种不可加载形态得到同一条消息，而不是依赖加载器内部的报错。
+- 健康检查随每次 `list()` 运行：每次读名单对每个 preset 一次读取加解析，接受的理由与不做缓存的发现相同——名单很小，新鲜是契约。
+- 复制损坏 preset 只在 UI 层拒绝（按钮禁用并给出原因）；宿主的 `copy` 保持形状无关。损坏来源产出同样损坏、同样可见的副本——没有能力增益，而宿主侧拒绝需要为一条被禁用按钮挡住的路径专门发明错误词汇。
+
+## 关键细节
+
+- **`PRESET_ID` 移到 `types.ts`**，让发现与创作共享同一份包含边界词汇；authoring 原样转发导出。
+- **原因只留一行。** js-yaml 会附上多行代码框摘录；名单卡片不是终端，`compositionProblem` 只保留首行。
+- **mount.spec 的两个竞态用例特意不动**：`ensureStanding` 仍可能拿到删除前一刻解析出的 preset（私有路径测试），其 stamp/unstampable 语义不变——健康检查发生在此之前的公开路径上。
+- **创造模式的引导随同一 PR 落地**：`cordis` preset 的 persona 现在禁止编辑随附安装（损坏 `cordis` 会禁用这一模式本身），并把创作指向 `${DSH_HOME:-$HOME/.dsh}/.agent-presets/<id>/`；其技能新教了 `preset.yml` 元信息、先复制再改的流程、一次升级的沙箱现实（preset 根目录在会话工作区之外）与诚实的验证方式（agent 无法自己启动会话；设置页的红色标记是用户的检查项）。已实测：被要求直接改随附 `cordis` 组装时，组装出的 agent 援引两条规则拒绝并给出复制路径；被要求真正创建 preset 时，它落在 `$DSH_HOME` 下、把写入合并为一次升级、用加载器方言自查、并把验证交还用户。
+
+## 曾考虑的替代方案
+
+隐藏损坏 preset 但在复制时用更好的报错拒绝该 id：幽灵仍然无法从任何界面清除。深度校验（读名单时解析每一行的模块）：挂载已经拥有这一失败并带回滚，每次读名单逐行 import 既不便宜也不更可操作。阻止 `settings` 写入指向损坏默认值：settings 领域是通用的，而名单是活目录——此刻缺失或损坏的名字到下一个会话可能已经有效，挂载的响亮失败才是拥有那一刻的强制点。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-24-workspace-context.md
-2026-06-24-workspace-context.md: 53a580aee50752b7c6daeff5caa42ba6409c8885
+2026-06-24-workspace-context.md: e7a3724847b9dc8cfad11e87b2c96a3ef442bcba
 2026-06-24-workspace-context.zh.md: 39c3f52da10b7299301d10bd8b78330cbae8e19c
@@ -78,7 +78,7 @@ There is intentionally no watcher. Detection occurs at the next successful struc

 ## Consequences

-Workspace guidance is isolated per session and shared by the demo front doors, Web Host, and every tool presentation mode. Initial, nested, and changed instructions are durable and replayable. The generic session/agent context contract carries typed source data through inbox-staged and durably entered user messages without flattening entries.
+Workspace guidance is isolated per session and shared by the demo entry points, Web Host, and every tool presentation mode. Initial, nested, and changed instructions are durable and replayable. The generic session/agent context contract carries typed source data through inbox-staged and durably entered user messages without flattening entries.

 Repository text remains untrusted input. Lower-authority user-role framing, explicit precedence language, and delimiter escaping reduce risk but do not eliminate prompt injection. Following a candidate symlink to its target widens that surface to off-tree content, so the permission and sandbox layers that confine `ctx.fs` to trusted roots are the boundary that treats workspace files as data rather than authority (the [instruction-symlink follow note](2026-07-21-follow-instruction-symlinks.md) owns the residual risk).

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-16-harness-level-loop.md
-2026-07-16-harness-level-loop.md: e84a5738a55988d3b968eac829d9daaf10d7e304
-2026-07-16-harness-level-loop.zh.md: d7cda56d66d4456ac5c3c7d55bb238e85fcacbc5
+2026-07-16-harness-level-loop.md: e5dd94fbf76e27f0843666f29622969635ef2fc3
+2026-07-16-harness-level-loop.zh.md: 773b5ff1f4b07206d0d817c29774435ce24407b4
@@ -70,7 +70,7 @@ The human UX follows the compact Codex shape in the [public OpenAI Codex TUI dis

 The model receives only `get_goal`, `create_goal`, and `update_goal`. It may create a goal when a direct human request clearly asks for substantial multi-round work, and it may infer that intent in any language. It must not turn routine one-turn work into a goal. Code requires a direct human message in the current live root-agent turn; semantic interpretation remains model judgment. An autonomous goal round may report `complete` or `blocked` for the exact current goal round but cannot edit, pause, resume, or replace the human objective.

-TUI mounts the shared command registry and complete goal stack by default and exposes `/goal` through one producer. ACP mounts the goal domain, model tools, and same-session driver but deliberately omits the human command plane. Every effective registered command is discoverable and invocable through every composed command adapter; a plugin incompatible with an application omits its command producer from that composition rather than relying on registry-level surface masks. The UI-less agent spine is opt-in so one-shot callers do not silently become multi-round operations. The headless CLI and JSON-RPC front doors do not consume the command plane; ordinary human text can still authorize model goal tools when that stack is composed.
+TUI mounts the shared command registry and complete goal stack by default and exposes `/goal` through one producer. ACP mounts the goal domain, model tools, and same-session driver but deliberately omits the human command plane. Every effective registered command is discoverable and invocable through every composed command adapter; a plugin incompatible with an application omits its command producer from that composition rather than relying on registry-level surface masks. The UI-less agent spine is opt-in so one-shot callers do not silently become multi-round operations. The headless CLI and JSON-RPC entry points do not consume the command plane; ordinary human text can still authorize model goal tools when that stack is composed.

 ### Fresh-agent Ralph execution

@@ -70,7 +70,7 @@ Goal Round 驱动器为每个特定的实时 agent 至多拥有一个待定预

 模型只接收 `get_goal`、`create_goal` 和 `update_goal`。当直接人类请求清楚要求大量多 Round 工作时，模型可以创建目标，并且可以从任何语言推断该意图。它不得把日常单 Turn 工作变成目标。代码要求当前实时根 agent Turn 中有一条人类直接发送的消息；语义解释仍是模型判断。自治目标 Round 可以为确切的当前 Goal Round 报告 `complete` 或 `blocked`，但不能编辑、暂停、恢复或替换人类目标。

-TUI 默认挂载共享命令注册表和完整目标栈，并通过一个生产方暴露 `/goal`。ACP（Agent Client Protocol）挂载目标领域、模型工具和同会话驱动器，但有意省略人类命令平面。每条有效已注册命令都能被每个已组合的命令适配器发现和调用；若插件与某应用不兼容，该应用组合会省略其命令生产方，而不是依赖注册表层面的表面掩码。无 UI 的 agent 主干要求显式选择加入，以免单次调用方静默变成多 Round 操作。无头 CLI（命令行界面）与 JSON-RPC 前端不消费命令平面；挂载目标栈后，普通人类文本仍可授权模型目标工具。
+TUI 默认挂载共享命令注册表和完整目标栈，并通过一个生产方暴露 `/goal`。ACP（Agent Client Protocol）挂载目标领域、模型工具和同会话驱动器，但有意省略人类命令平面。每条有效已注册命令都能被每个已组合的命令适配器发现和调用；若插件与某应用不兼容，该应用组合会省略其命令生产方，而不是依赖注册表层面的表面掩码。无 UI 的 agent 主干要求显式选择加入，以免单次调用方静默变成多 Round 操作。无头 CLI（命令行界面）与 JSON-RPC 运行入口不消费命令平面；挂载目标栈后，普通人类文本仍可授权模型目标工具。

 ### 全新 agent Ralph 执行

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-16-persistent-pty-sessions.md
-2026-07-16-persistent-pty-sessions.md: f27d799574e61d92123c9ca4e31c5c2a9d3229b4
-2026-07-16-persistent-pty-sessions.zh.md: 0c23619faf00794c2c4e7b3f85ef961faddd99e0
+2026-07-16-persistent-pty-sessions.md: fb9cd06bade7bc357baa738f0d9dd03b7f5b7936
+2026-07-16-persistent-pty-sessions.zh.md: 55a5848c1ab1e8c2cd3b29f2d4748ea4abbe088c
@@ -152,7 +152,7 @@ The package ships concise tool guidance explaining persistent state, owner isola

 **Include TUI sequences and BEL handling.** Rejected. The source prototype treats those paths as timing-sensitive and still records unresolved alternate-screen and interaction failures. Line-oriented PTY use proves the core value without making those unverified behaviors foundational.

-**Use an out-of-process daemon immediately.** Rejected for the initial in-process capability because current persistent front doors already keep a Cordis context alive. A daemon becomes justified by cross-process restoration or multi-client attachment, both deferred here.
+**Use an out-of-process daemon immediately.** Rejected for the initial in-process capability because current long-lived entry points already keep a Cordis context alive. A daemon becomes justified by cross-process restoration or multi-client attachment, both deferred here.

 ## Verification

@@ -152,7 +152,7 @@ plugins:

 **包含 TUI sequence 与 BEL 处理。**拒绝。源 prototype 将这些路径视为 timing-sensitive，且仍记录未解决的 alternate-screen 和交互失败。行式 PTY 已能证明核心价值，无需把未经验证的行为放进基础层。

-**立即采用进程外 daemon。**初始的进程内功能不采用，因为当前持久 front door 已能维持 Cordis context。跨进程恢复或多客户端 attach 会让 daemon 变得合理，但两者都已推迟。
+**立即采用进程外 daemon。**初始的进程内功能不采用，因为当前长驻的运行入口已能维持 Cordis context。跨进程恢复或多客户端 attach 会让 daemon 变得合理，但两者都已推迟。

 ## 验证

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-human-goal-command.md
-2026-07-19-human-goal-command.md: 5fdd80f7423b80e84e58f7379130ee59a2e8a723
-2026-07-19-human-goal-command.zh.md: 89d29497abfc758ff8373d272aa8620ca1bcfcc2
+2026-07-19-human-goal-command.md: d68e4025a4d37d07211f15ddfc6069bc7c637124
+2026-07-19-human-goal-command.zh.md: dbda35c731ccdc7aff2d4213a3df8b78070319df
@@ -68,5 +68,5 @@ The producer suite uses the real command registry, goal service, agent registry,
 - The portable command contract has no modal editor or confirmation interaction; inline edit and explicit clear are intentional until a general cross-surface interaction primitive exists.
 - `/goal` does not accept a per-command round cap. Deployment config owns the default, and the authorized model tool can edit a cap after direct human instruction.
 - TUI renders portable plain text rather than a continuously updated goal status widget. Reconnectable command output and adapter-specific status indicators are deferred.
- The ACP automation server, headless CLI, and JSON-RPC front doors do not consume the command registry.
+- The ACP automation server, headless CLI, and JSON-RPC entry points do not consume the command registry.
 - The command observes and mutates state but does not certify completion or blockers. Evaluator-backed certification remains deferred to a separate policy layer with an explicit authority and isolation contract.
@@ -68,5 +68,5 @@ TUI 应用包作出相反的产品选择。它默认让 `goals` 使用所有者
 - 可移植命令约定没有模态编辑器或确认交互；在出现通用跨界面交互原语之前，行内编辑与明确清除是有意选择。
 - `/goal` 不接受逐命令 Round 上限。部署配置拥有默认值；得到直接人类指示后，已授权模型工具可以编辑上限。
 - TUI 渲染可移植纯文本，而不是持续更新的目标状态组件。可重连命令输出和适配器专用状态指示器予以延期。
- ACP 自动化服务器、无头 CLI 与 JSON-RPC 前端不消费命令注册表。
+- ACP 自动化服务器、无头 CLI 与 JSON-RPC 运行入口不消费命令注册表。
 - 该命令观察并改变状态，但不认证完成或阻塞。基于评估器的认证延期到具有明确权限与隔离约定的独立策略层。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-plugin-command-registration.md
-2026-07-19-plugin-command-registration.md: c8f0f2772a41948e9eb257a16f40194518c568f9
-2026-07-19-plugin-command-registration.zh.md: a8f7a3d1e7947aeb41344a72106cee55d249010c
+2026-07-19-plugin-command-registration.md: 76feba84492bd6b245246a6f1fb1e8f68d555684
+2026-07-19-plugin-command-registration.zh.md: e89cb79ba1e15dfefa527fe7bc37c09c3449f5b4
@@ -12,7 +12,7 @@ A shared mechanism must remain a UI concern rather than a model tool or agent-lo

 ## Decision

-`@deepseek-ai/dsh-commands` in `packages/interaction/commands/` is the product command registry. The TUI app bundle mounts it beside its consuming front door; the [automation-only ACP app](../simplification/2026-07-23-acp-automation-only-protocol.md) and the executor-less, UI-less agent spine omit it. TUI injects the service, while command producers depend only on the registry and any domain they operate.
+`@deepseek-ai/dsh-commands` in `packages/interaction/commands/` is the product command registry. The TUI app bundle mounts it beside its consuming front end; the [automation-only ACP app](../simplification/2026-07-23-acp-automation-only-protocol.md) and the executor-less, UI-less agent spine omit it. TUI injects the service, while command producers depend only on the registry and any domain they operate.

 ### Registry contract

@@ -50,7 +50,7 @@ TUI tests exercise all migrated built-ins, live plugin discovery, help/autocompl

 - **Keep adapter-local switches** — rejected because optional plugins cannot contribute discovery and behavior without editing the TUI.
 - **Represent human commands as model tools** — rejected because discovery and direct invocation are human UI behavior; routing through the model adds latency, token cost, and reinterpretation.
- **Put the registry in the core agent spine** — rejected because UI-less front doors do not consume it, while TUI can compose it explicitly.
+- **Put the registry in the core agent spine** — rejected because UI-less entry points do not consume it, while TUI can compose it explicitly.
 - **Make `dsh-agent-loop` inject commands** — rejected because the loop does not execute or discover human commands. Agent-scoped producers declare the UI dependency in a child plugin instead.
 - **Attach adapter masks to each definition** — rejected because support is a composition fact, not command-domain state. Every composed adapter exposes a registered command; an incompatible plugin omits registration in that deployment.
 - **Send unknown slash input to the model** — rejected because typoed or unavailable direct actions must fail predictably rather than change execution planes.
@@ -68,4 +68,4 @@ TUI tests exercise all migrated built-ins, live plugin discovery, help/autocompl
 - Input metadata is limited to an unstructured text hint. Typed forms, argument schemas, and completion providers remain command-owned or require a later registry or consumer extension.
 - Generic command output is live-only and is not reconstructed after TUI restart.
 - Registry cancellation stops awaiting immediately, but external work stops only when a handler cooperates with its signal.
- The ACP automation server, headless CLI, and JSON-RPC SDK front doors do not expose the command plane; only TUI consumes it.
+- The ACP automation server, headless CLI, and JSON-RPC SDK entry points do not expose the command plane; only TUI consumes it.
@@ -12,7 +12,7 @@ TUI 拥有斜杠命令。如果命令名、帮助文本、自动补全、分派

 ## 决策

-位于 `packages/interaction/commands/` 的 `@deepseek-ai/dsh-commands` 是产品命令注册表。TUI 应用 bundle（组合包）把它挂载在消费该服务的入口旁；[仅面向自动化的 ACP（Agent Client Protocol）应用](../simplification/2026-07-23-acp-automation-only-protocol.md)和无执行器、无 UI 的智能体 spine（主干）都省略该服务。TUI 注入该服务，命令生产者只依赖注册表及其操作的领域。
+位于 `packages/interaction/commands/` 的 `@deepseek-ai/dsh-commands` 是产品命令注册表。TUI 应用 bundle（组合包）把它挂载在消费该服务的前端旁；[仅面向自动化的 ACP（Agent Client Protocol）应用](../simplification/2026-07-23-acp-automation-only-protocol.md)和无执行器、无 UI 的智能体 spine（主干）都省略该服务。TUI 注入该服务，命令生产者只依赖注册表及其操作的领域。

 ### 注册表约定

@@ -50,7 +50,7 @@ TUI 测试覆盖全部迁移后的内置命令、实时插件发现、帮助与

 - **保留适配器本地 switch**——不予采纳，因为可选插件无法贡献发现与行为，除非修改 TUI。
 - **把人类命令表示为模型工具**——不予采纳，因为发现与直接调用属于人类 UI 行为；经由模型路由会增加延迟、token 成本和重新解释。
- **把注册表放入核心智能体主干**——不予采纳，因为无 UI 前端不消费它，而 TUI 可以显式组合它。
+- **把注册表放入核心智能体主干**——不予采纳，因为无 UI 运行入口不消费它，而 TUI 可以显式组合它。
 - **让 `dsh-agent-loop` 注入 commands**——不予采纳，因为循环不执行也不发现人类命令。智能体作用域生产者改为在子插件中声明 UI 依赖。
 - **为每个定义附加适配器掩码**——不予采纳，因为支持能力是组合事实，而不是命令领域状态。每个已组合适配器都暴露已注册命令；不兼容插件不会在该部署中注册。
 - **把未知斜杠输入发送给模型**——不予采纳，因为输入错误或不可用的直接操作必须可预测地失败，而不能改变执行平面。
@@ -68,4 +68,4 @@ TUI 测试覆盖全部迁移后的内置命令、实时插件发现、帮助与
 - 输入元数据仅限非结构化文本提示。类型化表单、参数模式和补全提供器仍由命令拥有，或需要后续注册表或消费方扩展。
 - 通用命令输出仅实时存在，TUI 重启后不会重建。
 - 注册表取消会立即停止等待，但外部工作只有在处理器配合信号时才会停止。
- ACP 自动化服务器、无头 CLI 与 JSON-RPC SDK 前端不暴露命令平面；只有 TUI 消费它。
+- ACP 自动化服务器、无头 CLI 与 JSON-RPC SDK 运行入口不暴露命令平面；只有 TUI 消费它。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-21-local-instruction-overlay.md
-2026-07-21-local-instruction-overlay.md: 3c7b2141b0515b5e667be4add6ad765e26c88cd8
+2026-07-21-local-instruction-overlay.md: fb5f916d426595a80bbbaa4192e4a3975b922b8a
 2026-07-21-local-instruction-overlay.zh.md: c97ed04607f497d829da0e904c248836252c73f7
@@ -26,7 +26,7 @@ The base and local candidates in one directory must stay independent across base

 **Keep it opt-in through `instructionFileCandidates`.** Rejected: one directory has a single winner, so a `.local.` name added to that list shadows the base file rather than supplementing it. The packages guidance to keep opt-ins out of shipped defaults is outweighed here by strong prior art and the user-facing expectation that `.local.` files are always read.

-**Default at the product `cordis.yml` level instead of the plugin schema.** Rejected: it would enable `.local.` only for whichever front door remembered to opt in, splitting behavior across TUI/ACP/headless and duplicating a value that belongs beside the existing candidate default.
+**Default at the product `cordis.yml` level instead of the plugin schema.** Rejected: it would enable `.local.` only for whichever entry point remembered to opt in, splitting behavior across TUI/ACP/headless and duplicating a value that belongs beside the existing candidate default.

 **Reuse the bare directory as the scope key for base and local files.** Rejected: base and local files in one directory would collide in every scope-keyed map, so a change to one would suppress or overwrite the other. A distinct scope key per candidate keeps them independent without widening the persisted metadata shape.

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-24-web-session-model-selector.md
-2026-07-24-web-session-model-selector.md: 78017b14a806f8e609a85e340094cd2a349d47e1
-2026-07-24-web-session-model-selector.zh.md: 87a342721f5e47f5bcedfafb578eb6916101d2f5
+2026-07-24-web-session-model-selector.md: e6a96ac62f69a3bd312f61cc920caa259d2dc5b0
+2026-07-24-web-session-model-selector.zh.md: 5a0245359d69ac6e59a20dc3276b9411c4b25e23
@@ -6,7 +6,7 @@ English | [中文](2026-07-24-web-session-model-selector.zh.md)

 ## Problem

-The Web conversation needs a visible, mutable session model selection sourced from the Host. Copying TUI presentation or hardcoding DeepSeek models in the browser would split model discovery and step-boundary semantics across front doors. A switch made while a response is running also needs one atomic boundary: prompt variables and request routing cannot observe different selections.
+The Web conversation needs a visible, mutable session model selection sourced from the Host. Copying TUI presentation or hardcoding DeepSeek models in the browser would split model discovery and step-boundary semantics across front ends. A switch made while a response is running also needs one atomic boundary: prompt variables and request routing cannot observe different selections.

 ## Decision

@@ -6,13 +6,13 @@ Status: implemented

 ## 问题

-Web 对话需要一项由 Host 提供、可见且可更改的会话模型选择。如果照搬 TUI 的呈现方式，或在浏览器中硬编码 DeepSeek 模型，就会让模型发现逻辑和步骤边界语义分散到不同前门中。响应运行期间发生的切换还需要一个原子边界：提示词变量与请求路由不能观测到不同的选择。
+Web 对话需要一项由 Host 提供、可见且可更改的会话模型选择。如果照搬 TUI 的呈现方式，或在浏览器中硬编码 DeepSeek 模型，就会让模型发现逻辑和步骤边界语义分散到不同前端中。响应运行期间发生的切换还需要一个原子边界：提示词变量与请求路由不能观测到不同的选择。

 ## 决策

 Web Host 为每个新建或恢复的 Agent 安装 `ModelSelection`。如果会话已经使用过模型，提供方／模型／推理（reasoning）选择来自最新的 `request/header`；否则来自 `ctx.agentDefaultModel`。`session.selectModel` 会赋值会话级选择，提示词组装则将它与请求路由一并捕获，因此运行中步骤发生的切换会应用于下一个组装步骤。下一个实际采用的选择通过完整的 `request/header` 快照持久化；尚未进入请求的选择则仅保存在当前进程中。

-会话 RPC 领域公开 `session.models` 模型目录与 `session.selectModel`。该目录从 LLM（大语言模型）注册表动态构建，并按提供方分组；每个已列出模型的精确元数据还会加入由适配器持有的推理强度 ID、名称、说明和可选默认值。各提供方的目录与精确元数据会按提供方并发加载，且彼此独立失败，因此成功加载的分组仍可与可重试的失败记录一同使用。模型是否位于目录仅供参考：`session.models.current` 独立返回，即使不在任何分组中也仍然可以路由，但提供方停止公布该模型后，Host 不会合成未列出行。两个前门对这一状态给出不同回答：TUI 把未列出的当前模型渲染为独立一行，Web 则显示未设置状态的触发器标签并要求选择替代模型。Web 是编辑目录的 surface，因此缺席的目录行代表一项待作出的选择；TUI 只从现有行中选择。显示未设置标签的 Web composer 仍可以使用当前可路由选择发送消息。精确解析决定提供方／模型组合与显式推理强度是否可用。选择操作通过 `resolveCallConfig` 拒绝不支持的推理强度 ID，并在赋值该选择前具体化适配器配置的默认值。
+会话 RPC 领域公开 `session.models` 模型目录与 `session.selectModel`。该目录从 LLM（大语言模型）注册表动态构建，并按提供方分组；每个已列出模型的精确元数据还会加入由适配器持有的推理强度 ID、名称、说明和可选默认值。各提供方的目录与精确元数据会按提供方并发加载，且彼此独立失败，因此成功加载的分组仍可与可重试的失败记录一同使用。模型是否位于目录仅供参考：`session.models.current` 独立返回，即使不在任何分组中也仍然可以路由，但提供方停止公布该模型后，Host 不会合成未列出行。两个前端对这一状态给出不同回答：TUI 把未列出的当前模型渲染为独立一行，Web 则显示未设置状态的触发器标签并要求选择替代模型。Web 是编辑目录所在的前端，因此缺席的目录行代表一项待作出的选择；TUI 只从现有行中选择。显示未设置标签的 Web composer 仍可以使用当前可路由选择发送消息。精确解析决定提供方／模型组合与显式推理强度是否可用。选择操作通过 `resolveCallConfig` 拒绝不支持的推理强度 ID，并在赋值该选择前具体化适配器配置的默认值。

 浏览器中的 `ModelService` 为每个实时会话持有一个 `ModelDirectory`。其快照包含当前完整的 `ModelSelection`、分组目录、提供方失败记录、操作错误，以及 `idle`、`loading`、`ready`、`selecting`、`error` 状态。挂载时会预先填充触发器标签，此后每次打开菜单都会刷新目录。目录与选择调用共用操作代次，防止较早响应覆盖较新结果；连接重置会先丢弃当前进程中的投影，再恢复 Host 选择。失败时保留先前的选择和可用分组。

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-01-pwsh-tool-and-executor.md
-2026-08-01-pwsh-tool-and-executor.md: 7206f8ffe6640f8499f8453c40ab5846b23112c6
-2026-08-01-pwsh-tool-and-executor.zh.md: c59ba8e3e6c68d48d310861325c74dc6cec8e5c3
+2026-08-01-pwsh-tool-and-executor.md: 855d8e5db8a78e7e798061724d89fde31b28e975
+2026-08-01-pwsh-tool-and-executor.zh.md: dcf5c53cd6681be07adafef92d556b400c116736
@@ -6,30 +6,30 @@ English | [中文](2026-08-01-pwsh-tool-and-executor.zh.md)

 ## Problem

-The harness spoke one shell dialect on every platform: `bash`. Windows hosts could run it only through WSL or Git-Bash shims, and the shipped `dsh-bash-local` executor is POSIX-only (`bash` hardcoded, process-group semantics POSIX). The Windows roadmap — defaulting hosts to `pwsh`, later pwsh TUI/GUI rendering — had no execution foundation: there was no PowerShell implementation of the bash executor seam and no model-facing tool that taught the PowerShell dialect. The bash tool itself is also far larger than a Windows-first profile needs: background tasks, sandbox escalation, and the persistent-PTY twin are all bash-shaped surface that a minimal `pwsh` tool should not carry.
+The harness spoke one shell dialect on every platform: `bash`. Windows hosts could run it only through WSL or Git-Bash shims, and the shipped `dsh-bash-local` executor is POSIX-only (`bash` hardcoded, process-group semantics POSIX). The Windows roadmap — defaulting hosts to `pwsh`, later pwsh TUI/GUI rendering — had no execution foundation: there was no PowerShell implementation of the bash executor seam and no model-facing tool that taught the PowerShell dialect. The bash tool is also larger than a Windows-first profile strictly needs — the persistent-PTY twin in particular is bash-shaped surface the `pwsh` tool still does not carry. The original minimal profile also left out background tasks and sandbox escalation: background arrived with the [parity decision](2026-08-02-pwsh-tool-bash-parity.md), and the sandbox surface (denial rendering plus `sandbox_permissions` escalation) arrived with the [Windows ACL sandbox decision](2026-08-08-windows-acl-restricted-token-sandbox.md) — the minimal tool was sized for the danger-full-access Windows posture, and that premise ended when the sandbox PR re-enabled confinement and approval on Windows.

 ## Decision

 Two new packages under `packages/bash/`:

 - **`@deepseek-ai/dsh-pwsh-local`** — a local implementation of the `ctx.bash` executor seam over `ctx.subprocess`, mirroring `dsh-bash-local` call-for-call: `resolve()` defaults and caps from config, `run()` fuses the config-clamped timeout with the caller's signal through one deadline, `start()` returns a consuming background handle whose processes belong to the subprocess service. The command string rides as ONE argv element to `pwsh -NoLogo -NoProfile -NonInteractive -Command`, so PowerShell parses it and no shell-quoting layer exists. Executable resolution (`resolvePwshPath`) is a pure function of `(configured, env, platform)`: explicit config first, then Windows probes PowerShell 7's install, PATH entries (quotes stripped), and Windows PowerShell 5.1, else a bare `pwsh` via PATH.
- **`@deepseek-ai/dsh-tool-pwsh`** — the model-facing tool over `ctx.bash`, PowerShell-dialect by contract, mirroring `dsh-tool-bash` call-for-call minus the sandbox surface: foreground and `run_in_background` execution through the generic task runtime, managed `DSH_*` environment through the shared [`dsh-bash-env`](../feature/2026-08-02-pwsh-tool-bash-parity.md) registry, and the bash marker/truncation rendering story (a clean exit produces no marker). The parity decision supersedes this note's minimal-profile tool description.
+- **`@deepseek-ai/dsh-tool-pwsh`** — the model-facing tool over `ctx.bash`, PowerShell-dialect by contract, mirroring `dsh-tool-bash` call-for-call: foreground and `run_in_background` execution through the generic task runtime, managed `DSH_*` environment through the shared [`dsh-bash-env`](../feature/2026-08-02-pwsh-tool-bash-parity.md) registry, the bash marker/truncation rendering story (a clean exit produces no marker), and — since the Windows ACL sandbox decision — the sandbox denial rendering and `sandbox_permissions` escalation surface, plus the Windows-specific ConstrainedLanguage and named-pipe contracts in the tool description. The parity decision supersedes this note's minimal-profile tool description.

 Windows vitest coverage is deliberately NOT part of this change: the repo's Windows CI lane owns build/static gates, and unit coverage runs on Linux, where both packages' suites run against a real `pwsh` (preinstalled on the GitHub-hosted runners) or self-skip when absent. The vitest `windowsUnsupportedPackages` exclusion narrows from `packages/bash/*` to the bash-requiring packages so the pwsh suites can also run natively on Windows dev machines.

-The roadmap beyond this decision — defaulting Windows hosts to `pwsh` (bash off), and pwsh TUI/GUI rendering — is recorded separately as [a proposal](../../proposed/feature/2026-08-01-windows-pwsh-default.md).
+The roadmap beyond this decision — defaulting Windows hosts to `pwsh` (bash off), and pwsh TUI/GUI rendering — is recorded separately as [the Windows pwsh default decision](2026-08-01-windows-pwsh-default.md).

 ## Alternatives considered

 **Extend `dsh-bash-local` with a pwsh mode.** Rejected: the executor's identity is the shell it spawns; a second dialect inside one package doubles its config surface (`shell` switches) and its test matrix, and the two dialects' quirks (signal facts on Windows, quoting domains) belong to their own packages' documentation.

-**Extend `dsh-tool-bash` with a dialect parameter.** Rejected: the bash tool's background/sandbox surface is bash-shaped; a `pwsh` mode would either hide it (conditional schema churn) or inherit it (surface the minimal profile explicitly rejects). The minimal twin keeps the model contract honest.
+**Extend `dsh-tool-bash` with a dialect parameter.** Rejected: the model-visible contract is the dialect itself (paths, variables, exit facts differ), so a dialect parameter would either churn the schema conditionally or force one tool to teach two dialects; the separate twin keeps the model contract honest — and carries the shared surfaces (background, sandbox, rendering) by mirroring rather than by sharing an implementation.

 **Wire the pwsh tool into the shipped CLI compositions now.** Rejected: mounting `tool-pwsh` + `pwsh-local` in `base.cordis.yml` would change the shipped roster before the Windows-default decision lands; this change ships the capability and its wiring points (`apps/cli` dependencies, tsconfig projects) without switching any default.

 ## Consequences

 - The bash executor seam gains a second, Windows-native implementation with an identical request/spec contract, so model-facing consumers beyond `tool-pwsh` (hooks bridges, in-process plugins) can run PowerShell without dialect shims.
- `tool-pwsh` is the model-visible Windows-first shell tool: behaviorally interchangeable with the bash tool for foreground and background work (minus sandbox), with prompt guidance that states the marker contract precisely.
+- `tool-pwsh` is the model-visible Windows-first shell tool: behaviorally interchangeable with the bash tool for foreground, background, and sandboxed work — including the same-turn `sandbox_permissions` escalation through `ctx.approval` — with prompt guidance that states the marker contract, the sandbox denial/escalation vocabulary, and the ConstrainedLanguage and named-pipe boundaries precisely.
 - Windows semantics differ where the platform differs: forced termination reports exit 1 with no signal (so `signal`/`killed` status facts are POSIX-only), and PowerShell writes CRLF, which tests normalize.
 - The CLI gains two workspace dependencies and two tsconfig projects without mounting either plugin — the composition decision stays with the Windows-default proposal.
@@ -6,30 +6,30 @@ Status: implemented

 ## 问题

-harness 在每个平台只说一种 shell 方言：`bash`。Windows 主机只能通过 WSL 或 Git-Bash 垫片运行它，而交付的 `dsh-bash-local` 执行器仅限 POSIX（硬编码 `bash`，进程组语义是 POSIX 的）。Windows 路线图——让主机默认 `pwsh`，之后再做 pwsh TUI/GUI 渲染——没有执行基础：既没有 bash 执行器 seam 的 PowerShell 实现，也没有教模型 PowerShell 方言的面向模型工具。bash 工具本身也远大于 Windows 优先画像所需：后台任务、沙箱升级与持久 PTY 孪生都是 bash 形状的表面，最小化的 `pwsh` 工具不该背负。
+harness 在每个平台只说一种 shell 方言：`bash`。Windows 主机只能通过 WSL 或 Git-Bash 垫片运行它，而交付的 `dsh-bash-local` 执行器仅限 POSIX（硬编码 `bash`，进程组语义是 POSIX 的）。Windows 路线图——让主机默认 `pwsh`，之后再做 pwsh TUI/GUI 渲染——没有执行基础：既没有 bash 执行器 seam 的 PowerShell 实现，也没有教模型 PowerShell 方言的面向模型工具。bash 工具也大于 Windows 优先画像的严格所需——尤其持久 PTY 孪生是 `pwsh` 工具至今仍不背负的 bash 形状表面。最初的最小画像也没有后台任务与沙箱升级：后台随 [parity 决策](2026-08-02-pwsh-tool-bash-parity.md) 到来，沙箱面（拒绝渲染加 `sandbox_permissions` 升级）随 [Windows ACL sandbox 决策](2026-08-08-windows-acl-restricted-token-sandbox.md) 到来——最小工具当初按 danger-full-access 的 Windows 姿态裁剪，这一前提在 sandbox PR（Pull Request）于 Windows 上重新启用隔离与审批时终结。

 ## 决策

 在 `packages/bash/` 下新增两个包：

 - **`@deepseek-ai/dsh-pwsh-local`** —— `ctx.bash` 执行器 seam 的本地实现，基于 `ctx.subprocess`，逐调用镜像 `dsh-bash-local`：`resolve()` 从配置默认化并设上限，`run()` 通过一个 deadline 融合配置夹取的超时与调用方信号，`start()` 返回消费式后台句柄，其进程归属于 subprocess 服务。命令字符串作为 ONE argv 元素传给 `pwsh -NoLogo -NoProfile -NonInteractive -Command`，由 PowerShell 解析，不存在 shell 引号层。可执行文件解析（`resolvePwshPath`）是 `(configured, env, platform)` 的纯函数：先显式配置，再在 Windows 上探测 PowerShell 7 安装位置、PATH 条目（剥离引号）与 Windows PowerShell 5.1，否则经 PATH 解析裸 `pwsh`。
- **`@deepseek-ai/dsh-tool-pwsh`** —— 基于 `ctx.bash` 的面向模型工具，约定是 PowerShell 方言，逐调用镜像 `dsh-tool-bash`、减去 sandbox 面：经通用任务运行时执行前台与 `run_in_background`，经共享 [`dsh-bash-env`](../feature/2026-08-02-pwsh-tool-bash-parity.md) 注册表管理 `DSH_*` 环境，以及 bash 的 marker/截断渲染故事（干净退出不产生 marker）。parity 决策取代了本 note 的最小画像工具描述。
+- **`@deepseek-ai/dsh-tool-pwsh`** —— 基于 `ctx.bash` 的面向模型工具，约定是 PowerShell 方言，逐调用镜像 `dsh-tool-bash`：经通用任务运行时执行前台与 `run_in_background`，经共享 [`dsh-bash-env`](../feature/2026-08-02-pwsh-tool-bash-parity.md) 注册表管理 `DSH_*` 环境，bash 的 marker/截断渲染故事（干净退出不产生 marker），以及——自 Windows ACL sandbox 决策以来——沙箱拒绝渲染与 `sandbox_permissions` 升级面，外加工具描述中的 Windows 专属 ConstrainedLanguage 与 named-pipe 约定。parity 决策取代了本 note 的最小画像工具描述。

 Windows vitest 覆盖率刻意不属本次改动：仓库的 Windows CI 通道负责构建/静态门禁，单元覆盖在 Linux 上运行，两个包的套件在那里以真实 `pwsh` 运行（GitHub 托管 runner 预装）或缺失时自行跳过。vitest 的 `windowsUnsupportedPackages` 排除从 `packages/bash/*` 收窄为真正需要 bash 的包，使 pwsh 套件也能在 Windows 开发机上原生运行。

-本决策之后的路线图——让 Windows 主机默认 `pwsh`（关闭 bash）与 pwsh TUI/GUI 渲染——另行记录为[提案](../../proposed/feature/2026-08-01-windows-pwsh-default.md)。
+本决策之后的路线图——让 Windows 主机默认 `pwsh`（关闭 bash）与 pwsh TUI/GUI 渲染——已落地为 [Windows 默认 pwsh 决策](2026-08-01-windows-pwsh-default.md)。

 ## 备选方案

 **给 `dsh-bash-local` 增加 pwsh 模式。** 否决：执行器的身份就是它 spawn 的 shell；在一个包内塞第二种方言会翻倍配置面（`shell` 开关）与测试矩阵，且两种方言的怪癖（Windows 上的信号实情、引号域）应各自归入自己包的文档。

-**给 `dsh-tool-bash` 增加方言参数。** 否决：bash 工具的后台/沙箱表面是 bash 形状的；`pwsh` 模式要么隐藏它（条件 schema 翻动），要么继承它（把最小画像明确拒绝的表面带进来）。最小孪生让模型约定保持诚实。
+**给 `dsh-tool-bash` 增加方言参数。** 否决：模型可见约定本身就是方言（路径、变量、退出事实都不同），因此方言参数要么让 schema 按条件翻动，要么逼一个工具教两种方言；独立的孪生让模型约定保持诚实——并以镜像而非共享实现的方式携带共享表面（后台、沙箱、渲染）。

 **现在就接入交付的 CLI 组合。** 否决：在 Windows 默认决策落地前把 `tool-pwsh` + `pwsh-local` 挂进 `base.cordis.yml` 会改变交付清单；本改动交付能力与接线点（`apps/cli` 依赖、tsconfig 工程），不切换任何默认。

 ## 后果

 - bash 执行器 seam 有了第二个、Windows 原生的实现，请求/规范约定一致，因此 `tool-pwsh` 之外的面向模型消费方（hooks 桥、进程内插件）无需方言垫片即可运行 PowerShell。
- `tool-pwsh` 是模型可见的 Windows 优先 shell 工具：在前台与后台工作（减 sandbox）上与 bash 工具行为可互换，提示词指导精确陈述 marker 约定。
+- `tool-pwsh` 是模型可见的 Windows 优先 shell 工具：在前台、后台与沙箱化工作上与 bash 工具行为可互换——包括经 `ctx.approval` 的同轮次 `sandbox_permissions` 升级——提示词指导精确陈述 marker 约定、沙箱拒绝/升级词汇，以及 ConstrainedLanguage 与 named-pipe 边界。
 - Windows 语义在平台差异处不同：强制终止报告退出码 1 且无信号（因此 `signal`/`killed` 状态实情仅限 POSIX），PowerShell 输出 CRLF，测试做归一化。
 - CLI 增加两个 workspace 依赖与两个 tsconfig 工程，但不挂载任一插件——组合决策留给 Windows 默认提案。
@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-01-windows-pwsh-default.md
+2026-08-01-windows-pwsh-default.md: f0da86e52bcdd53a10b60164d7cc12261cfc5c49
+2026-08-01-windows-pwsh-default.zh.md: 41a6429eab8f86a8960ac4aa372aeacfda4661c4
@@ -0,0 +1,44 @@
+# Agent Note: Windows defaults to pwsh
+
+Status: implemented
+
+English | [中文](2026-08-01-windows-pwsh-default.zh.md)
+
+## Problem
+
+The harness's shipped execution profile is bash-first on every platform. Windows hosts must install a bash shim (WSL or Git-Bash) or fall back to the POSIX-only `dsh-bash-local` behavior (hardcoded `bash -c` argv, process-group semantics); the model-facing bash tool teaches the bash dialect. The Windows-native foundation shipped in the [pwsh executor and tool decision](2026-08-01-pwsh-tool-and-executor.md) — a PowerShell implementation of the `ctx.bash` seam and a parity `pwsh` tool — but shipped compositions still mounted the bash stack on Windows, so a Windows host without a shim could not run the shipped shell.
+
+## Decision
+
+Windows hosts booting a shipped profile (`dsh web`, `dsh --profile headless`, one-shot tasks) get the PowerShell stack by default; POSIX hosts are unchanged.
+
+- **The platform layer is a data file, not a roster rewrite.** `@deepseek-ai/dsh-base` ships [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml) alongside its universal `cordis.patch.yml`: it disables `bash-sandbox`/`tool-bash` (the POSIX-only executor and its dialect tool) and inserts `pwsh-local`/`tool-pwsh`. Windows has no OS sandbox runner (landlock/bwrap/seatbelt are POSIX-only), so the layer drops the sandbox stack entirely — `sandbox`, `sandbox-policy`, and `fs-sandbox` are disabled and the unconfined `dsh-fs-local` provides `ctx.fs` — and degrades to danger-full-access: `permission`/`ui-permission` leave the roster (dsh-permission requires a confining executor — presets bundle a sandbox mode the unconfined executor cannot honor; see its constructor guard — and the client knob would advertise a boundary that does not exist), and the `approval` service is disabled — nothing in the Windows roster asks for approval, so the model is never told approval exists or that asks are auto-rejected. Keeping fs-only path rules would be theater: the unconfined shell can bypass them with one command, so the honest Windows posture is full access rather than a boundary only the fs tools pretend to enforce.
+- **The launcher injects the layer by platform.** `apps/cli/src/windows-shell.ts` resolves it from the base bundle layer's `packageDir` between the bundle layers and the user layers on `win32` hosts, in every composition path (boot, config-only HMR recomposition, config dumps). Overriding the shipped default is a composition decision: a Windows host that prefers the bash stack — or confinement — re-enables the bash rows through its profile or home `cordis.patch.yml`. Custom profiles without the base bundle are skipped (they own their shell stack); a base bundle that ships no Windows shell patch fails loud.
+- **Module resolution is restored for cold starts.** The profiles-rework CLI dropped the pwsh packages from `apps/cli`'s dependency closure, so `healProfilesModuleFallback` never linked them into `$DSH_HOME/profiles/node_modules` and a fresh Windows host could not resolve the inserted rows. `apps/cli` and `dsh-base` re-declare `dsh-pwsh-local`/`dsh-tool-pwsh`, and `dsh-base` also declares `dsh-fs-local`; the base bundle lists every row plugin as a dependency by house style.
+
+The pwsh GUI rendering shipped earlier with the [pwsh UI presentation matches bash decision](2026-08-05-pwsh-ui-bash-parity.md); the [pwsh tool bash parity decision](2026-08-02-pwsh-tool-bash-parity.md) ships the tool's surface. Nothing in this decision changes POSIX behavior.
+
+## Alternatives considered
+
+**Default Windows to pwsh inside `dsh-bash-local` (one executor, dialect switch).** Rejected for the same reason the executor decision rejected a mode switch: the executor's identity is the shell it spawns, and platform-gated composition is a deployment choice, not an executor config.
+
+**Ship the platform layer from `apps/cli` code instead of a bundle data file.** Rejected: the patch belongs next to the rows it replaces, in the bundle that owns them, so the shipped roster stays visible as composition data and dumps carry its provenance; the launcher contributes only the win32 gate.
+
+**Keep `permission`/`ui-permission` on Windows.** Rejected: `dsh-permission` hard-requires `ctx.bash.sandboxMode` and fails loud at load over an unconfined executor; making it tolerate an unconfined shell would advertise presets the shell cannot honor.
+
+**Keep fs path-rule confinement on Windows (`sandbox-policy` + `fs-sandbox` without OS runners).** Rejected: the shell is the model's primary tool and unconfined on Windows, so fs-only path rules are trivially bypassable and would overstate the boundary; the honest posture is full degradation to danger-full-access.
+
+**Ship a `DSH_WINDOWS_SHELL` environment escape hatch.** Rejected: decisive behavior changes belong in composition config, which already overrides the platform layer row by id; a second override channel would split the single source of truth for roster decisions.
+
+## Consequences
+
+- A Windows host running a shipped `dsh` surface gets `pwsh` as its shell tool and PowerShell as the `ctx.bash` executor without configuration; `bash` is absent from the model-visible roster there (its tool row is disabled).
+- Windows has no sandbox at all: the fs tools run unconfined (`dsh-fs-local`), the approval service is absent (nothing asks for approval, and the model is never told approval exists), and the permission switcher is gone. The model-visible posture is honest full access rather than a boundary the shell can bypass.
+- POSIX hosts are unchanged: the platform layer never applies, and the bash stack remains the universal `cordis.patch.yml` rows.
+- Windows hosts that prefer the bash stack (e.g. with WSL/Git-Bash on PATH) override the shipped default through their profile or home `cordis.patch.yml` — disabling `pwsh-local`/`tool-pwsh` and re-enabling `bash-sandbox`/`tool-bash` (both executors register the same `bash` service, so an incomplete recipe fails loud at load) — composition config is the one override channel.
+
+## Verification
+
+- Unit: `apps/cli/tests/windows-shell.spec.ts` pins the win32 default, the custom-profile skip, and the missing-patch failure with the platform injected, and composes the REAL shipped bundle layers (dsh-base + dsh-web-app resolved from the app installation) through the boot's patch algorithm to assert the win32 danger-full-access roster and the base-only-profile warning; `packages/bundle/base/tests/base.spec.ts` pins the shipped Windows patch file shape (disables, inserts, and the absent approval service).
+- Keyless: a win32 `dsh --profile <name> --dump-config` shows the pwsh rows with `windows.cordis.patch.yml` provenance and the bash rows disabled; the POSIX dump (CI Linux) is unchanged.
+- The real-composition smoke boots the web profile on win32 with the pwsh stack mounted (the exact roster this note describes).
@@ -0,0 +1,44 @@
+# Agent Note: Windows 默认改用 pwsh
+
+Status: implemented
+
+[English](2026-08-01-windows-pwsh-default.md) | 中文
+
+## 问题
+
+harness 交付的执行画像在每个平台都是 bash 优先。Windows 主机必须安装 bash 垫片（WSL 或 Git-Bash），或退回到仅 POSIX 的 `dsh-bash-local` 行为（硬编码 `bash -c` argv、进程组语义）；面向模型的 bash 工具教的是 bash 方言。Windows 原生基础已随 [pwsh 执行器与工具决策](2026-08-01-pwsh-tool-and-executor.md) 交付——`ctx.bash` seam 的 PowerShell 实现与对等的 `pwsh` 工具——但交付组合在 Windows 上仍然挂载 bash 栈，没有垫片的 Windows 主机跑不了交付的 shell。
+
+## 决策
+
+启动交付 profile（`dsh web`、`dsh --profile headless`、一次性任务）的 Windows 主机默认获得 PowerShell 栈；POSIX 主机不变。
+
+- **平台层是数据文件，不是清单重写。** `@deepseek-ai/dsh-base` 随通用 `cordis.patch.yml` 一起交付 [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml)：它禁用 `bash-sandbox`/`tool-bash`（仅 POSIX 的执行器及其方言工具）并插入 `pwsh-local`/`tool-pwsh`。Windows 上没有 OS 级 sandbox runner（landlock/bwrap/seatbelt 均为 POSIX 专属），因此该层整体移除 sandbox 栈——`sandbox`、`sandbox-policy`、`fs-sandbox` 被禁用，由不限权的 `dsh-fs-local` 提供 `ctx.fs`——并完全退化为 danger-full-access：`permission`/`ui-permission` 离开清单（dsh-permission 要求有限权能力的执行器——preset 捆绑的是无限制执行器无法兑现的 sandbox 模式；见其构造函数守卫——客户端旋钮会宣传一个并不存在的边界），`approval` 服务也被禁用——Windows 清单里没有任何动作需要审批，模型也不会被告知"审批存在"或"请求会被自动拒绝"。保留仅限 fs 的路径规则是摆设：不限权的 shell 一条命令即可绕过，因此诚实的 Windows 姿态是全权访问，而不是一个只有 fs 工具假装执行的边界。
+- **启动器按平台注入该层。** `apps/cli/src/windows-shell.ts` 在 `win32` 主机上从 base bundle 层的 `packageDir` 解析它，置于 bundle 层与用户层之间，覆盖所有组合路径（启动、config-only HMR 重组合、配置转储）。覆盖交付默认是组合决策：偏好 bash 栈（或偏好有限权）的 Windows 主机通过其 profile 或 home 的 `cordis.patch.yml` 重新启用 bash 行。未挂 base bundle 的自定义 profile 被跳过（它们自己拥有 shell 栈）；base bundle 缺 `windows.cordis.patch.yml` 时 fail loud。
+- **冷启动的模块解析已恢复。** profiles 重构把 pwsh 包从 `apps/cli` 的依赖闭包中删掉了，`healProfilesModuleFallback` 因此从未把它们链接进 `$DSH_HOME/profiles/node_modules`，新 Windows 主机解析不到插入的行。`apps/cli` 与 `dsh-base` 重新声明 `dsh-pwsh-local`/`dsh-tool-pwsh`，`dsh-base` 还声明 `dsh-fs-local`；按仓库惯例，base bundle 把每个行插件都列为依赖。
+
+pwsh GUI 渲染已随 [pwsh UI 呈现与 bash 对齐决策](2026-08-05-pwsh-ui-bash-parity.md) 先行交付；[pwsh 工具与 bash 对齐决策](2026-08-02-pwsh-tool-bash-parity.md) 交付了工具表面。本决策不改变任何 POSIX 行为。
+
+## 备选方案
+
+**在 `dsh-bash-local` 内部让 Windows 默认 pwsh（一个执行器，方言开关）。** 否决，理由与执行器决策否决模式开关相同：执行器的身份就是它 spawn 的 shell，而按平台门控的组合是部署选择，不是执行器配置。
+
+**从 `apps/cli` 代码而非 bundle 数据文件交付平台层。** 否决：patch 应放在它替换的行旁边、属于拥有这些行的 bundle，让交付清单作为组合数据保持可见、转储带有出处；启动器只贡献 win32 门控。
+
+**在 Windows 上保留 `permission`/`ui-permission`。** 否决：`dsh-permission` 硬性要求 `ctx.bash.sandboxMode`，在无限制执行器上加载即 fail loud；让它容忍无限制 shell 会宣传 shell 无法兑现的 preset。
+
+**在 Windows 上保留 fs 路径规则限制（无 OS runner 的 `sandbox-policy` + `fs-sandbox`）。** 否决：shell 是模型的主工具且在 Windows 上不限权，仅限 fs 的路径规则一行命令即可绕过，会夸大边界；诚实的姿态是完全退化到 danger-full-access。
+
+**交付 `DSH_WINDOWS_SHELL` 环境变量逃生门。** 否决：决定性的行为变更应集中在组合配置中，而组合配置已能按行 id 覆盖平台层；第二条覆盖通道会分裂清单决策的单一事实来源。
+
+## 后果
+
+- 运行交付版 `dsh` 表面的 Windows 主机无需配置即获得 `pwsh` 作为 shell 工具、PowerShell 作为 `ctx.bash` 执行器；那里的模型可见清单中没有 `bash`（其工具行被禁用）。
+- Windows 上没有任何沙箱：fs 工具不限权运行（`dsh-fs-local`）、`approval` 服务不存在（没有任何动作需要审批，模型也不会被告知审批存在）、权限切换器消失。模型可见的姿态是诚实的全权访问，而不是一个 shell 可以绕过的边界。
+- POSIX 主机不变：平台层永不生效，bash 栈仍是通用 `cordis.patch.yml` 的行。
+- 偏好 bash 栈的 Windows 主机（例如 PATH 上有 WSL/Git-Bash 时）通过其 profile 或 home 的 `cordis.patch.yml` 覆盖交付默认——禁用 `pwsh-local`/`tool-pwsh` 并重新启用 `bash-sandbox`/`tool-bash`（两个执行器注册同一个 `bash` 服务，配方不完整会在加载时 fail loud）——组合配置是唯一的覆盖通道。
+
+## 验证
+
+- 单元：`apps/cli/tests/windows-shell.spec.ts` 以平台注入固定 win32 默认、自定义 profile 跳过与缺文件失败，并通过启动所用的 patch 算法组合真实交付的 bundle 层（从应用安装解析的 dsh-base + dsh-web-app）断言 win32 danger-full-access 清单与 base-only profile 警告；`packages/bundle/base/tests/base.spec.ts` 固定交付的 Windows patch 文件形状（禁用、插入与缺席的 approval 服务）。
+- Keyless：win32 上的 `dsh --profile <name> --dump-config` 显示带 `windows.cordis.patch.yml` 出处的 pwsh 行、被禁用的 bash 行；POSIX 转储（CI Linux）不变。
+- 真实组合冒烟在 win32 上启动 web profile，pwsh 栈挂载成功（即本笔记描述的确切清单）。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-02-pwsh-tool-bash-parity.md
-2026-08-02-pwsh-tool-bash-parity.md: d61dd6f21223121973520014c44e2e5846e7387a
-2026-08-02-pwsh-tool-bash-parity.zh.md: 120f66b1d9c713d5e3b71575186fc2194b0718a5
+2026-08-02-pwsh-tool-bash-parity.md: 79a09cb4c9698660faff18cf9ae016bec0bce227
+2026-08-02-pwsh-tool-bash-parity.zh.md: 3e01fd4447fbf047b7cbb3949e708094a6c06c09
@@ -10,13 +10,13 @@ The first Windows-native foundation shipped `dsh-tool-pwsh` as a deliberately mi

 ## Decision

-`dsh-tool-pwsh` now mirrors `dsh-tool-bash` call-for-call, minus the sandbox surface, and its model-visible text describes exactly that behavior:
+`dsh-tool-pwsh` now mirrors `dsh-tool-bash` call-for-call, and its model-visible text describes exactly that behavior:

 - **Rendering adopts the bash story verbatim**: stdout, a marked `[stderr]` section, truncation notices with spill paths, `(no output)` for an empty body, and exit markers only for non-zero exits — a clean exit produces no marker. The description and the `tool:pwsh` prompt section state this precisely ("Non-zero exits are reported as `[exit code: N]` markers"), deliberately not copying the bash prompt's "every result" phrasing, which its own renderer contradicts.
 - **`run_in_background` is wired through the generic task runtime** exactly like the bash tool: preflight, owner registration, `task_output`/`task_kill` control, and the same outcome mapping. `pwsh-local`'s already-mirrored `start()` handle backs it.
 - **The `DSH_*` environment is shared, not duplicated**: `BashEnvRegistry` moved out of `dsh-tool-bash` into a new tool-independent `@deepseek-ai/dsh-bash-env` package (`ctx.bashEnv` + built-ins + the session-persistence contributor), and both shell tools inject it. Contributors apply to pwsh calls exactly as they do to bash calls; shared environment ownership therefore sits outside either model-facing shell tool.
 - **Windows reality is pinned where bash has no analog**: every command runs under a UTF-8 output preamble so the Windows PowerShell 5.1 fallback cannot garble non-ASCII output through the UTF-8-decoding collector, and the prompts teach that Windows forced termination settles as exit 1 without a signal marker.
- **Out of scope, unchanged**: sandbox escalation (waits for a Windows-confining executor) and persistent PTY shells (backends are Linux/macOS-only; ConPTY is roadmap work). The pwsh-specific terminal card with an exit pill shipped separately in the [pwsh UI presentation matches bash](2026-08-05-pwsh-ui-bash-parity.md) decision.
+- **Out of scope, unchanged**: persistent PTY shells (backends are Linux/macOS-only; ConPTY is roadmap work). Sandbox escalation shipped later with the [Windows ACL sandbox decision](2026-08-08-windows-acl-restricted-token-sandbox.md) — the pwsh tool now carries the sandbox denial rendering and the same-turn `sandbox_permissions` escalation surface, plus the Windows ConstrainedLanguage contract in its description. The pwsh-specific terminal card with an exit pill shipped separately in the [pwsh UI presentation matches bash](2026-08-05-pwsh-ui-bash-parity.md) decision.

 ## Alternatives considered

@@ -28,7 +28,7 @@ The first Windows-native foundation shipped `dsh-tool-pwsh` as a deliberately mi

 ## Consequences

- The bash and pwsh tools are now behaviorally interchangeable for foreground and background shell work (minus sandbox), and the pwsh prompt/description sentences are each backed by the renderer.
+- The bash and pwsh tools are now behaviorally interchangeable for foreground, background, and sandboxed shell work (the sandbox surface arrived with the Windows ACL sandbox decision), and the pwsh prompt/description sentences are each backed by the renderer — the reviewer's grep-against-code check passes.
 - Parity ran BOTH ways once: the pwsh tool's structured foreground abort (`HarnessError('tool call aborted', TOOL_ABORTED)` with name `AbortError`) was backported to the bash tool, replacing its uncoded `Error('command aborted')` — a model-visible/logged change pinned by exact-shape tests on both sides and by the cancel-tool-calls fixture.
 - `@deepseek-ai/dsh-bash-env` is a new shipped package; `dsh-tool-bash`'s `dshHome` config moved there, so compositions mounting the shell tools must also mount `bash-env` (the spine bundles do).
 - Windows-only semantics (CRLF normalization, forced-termination exit-1/signal-null, POSIX-only self-signal) remain pinned by tests as before.
@@ -10,13 +10,13 @@ Status: implemented

 ## 决策

-`dsh-tool-pwsh` 现在逐调用镜像 `dsh-tool-bash`，减去 sandbox 面，其模型可见文本精确描述这一行为：
+`dsh-tool-pwsh` 现在逐调用镜像 `dsh-tool-bash`，其模型可见文本精确描述这一行为：

 - **渲染完全采用 bash 故事**：stdout、带标记的 `[stderr]` 段、带 spill 路径的截断通知、空体渲染 `(no output)`、退出 marker 仅限非零退出——干净退出不产生 marker。描述与 `tool:pwsh` prompt section 精确陈述这一点（"Non-zero exits are reported as `[exit code: N]` markers"），刻意不复制 bash prompt 中与其自身渲染矛盾的 "every result" 措辞。
 - **`run_in_background` 经通用任务运行时接线**，与 bash 工具完全一致：预检、owner 注册、`task_output`/`task_kill` 控制与相同的结果映射。其背后是 `pwsh-local` 早已镜像好的 `start()` 句柄。
 - **`DSH_*` 环境共享而非复制**：`BashEnvRegistry` 从 `dsh-tool-bash` 迁入新的工具无关包 `@deepseek-ai/dsh-bash-env`（`ctx.bashEnv` + 内置事实 + session-persistence contributor），两个 shell 工具都注入它。contributor 对 pwsh 调用与 bash 调用一视同仁；因此，共享环境的所有权不属于任何一个面向模型的 shell 工具。
 - **Windows 现实在 bash 无对应处钉死**：每条命令都在 UTF-8 输出 preamble 下运行，使 Windows PowerShell 5.1 兜底无法经 UTF-8 解码的 collector 破坏非 ASCII 输出；prompt 教授 Windows 强制终止以无 signal 的 exit 1 结算。
- **范围外，不变**：sandbox 升级（等待 Windows-confining 执行器）与持久 PTY shell（后端仅限 Linux/macOS；ConPTY 属路线图）。带退出 pill 的 pwsh 专属 terminal 卡已随 [pwsh UI 呈现与 bash 对齐](2026-08-05-pwsh-ui-bash-parity.md) 决策另行交付。
+- **范围外，不变**：持久 PTY shell（后端仅限 Linux/macOS；ConPTY 属路线图）。sandbox 升级随 [Windows ACL sandbox 决策](2026-08-08-windows-acl-restricted-token-sandbox.md) 稍后交付——pwsh 工具现在携带 sandbox 拒绝渲染与同轮次 `sandbox_permissions` 升级面，外加其描述中的 Windows ConstrainedLanguage 契约。带退出 pill 的 pwsh 专属 terminal 卡已随 [pwsh UI 呈现与 bash 对齐](2026-08-05-pwsh-ui-bash-parity.md) 决策另行交付。

 ## 备选方案

@@ -28,7 +28,7 @@ Status: implemented

 ## 后果

- bash 与 pwsh 工具在前台与后台 shell 工作（减 sandbox）上行为可互换，pwsh 的 prompt/描述句每句都有渲染器背书。
+- bash 与 pwsh 工具在前台、后台与沙箱化 shell 工作上行为可互换（sandbox 面随 Windows ACL sandbox 决策到来），pwsh 的 prompt/描述句每句都有渲染器背书——reviewer 的“拿代码 grep 对证”检查通过。
 - 对齐也反向发生过一次：pwsh 工具的结构化前台中止（`HarnessError('tool call aborted', TOOL_ABORTED)`，name 为 `AbortError`）被回移到 bash 工具，取代其无码的 `Error('command aborted')`——这是模型可见/入日志的变更，由两侧的精确形状测试与 cancel-tool-calls fixture 钉住。
 - `@deepseek-ai/dsh-bash-env` 成为新的交付包；`dsh-tool-bash` 的 `dshHome` 配置迁往那里，因此挂载 shell 工具的组合也必须挂载 `bash-env`（spine bundle 已如此）。
 - Windows 专属语义（CRLF 归一化、强制终止 exit-1/signal-null、仅 POSIX 的自信号）一如既往由测试钉住。
@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-05-per-agent-tool-presentation.md
+2026-08-05-per-agent-tool-presentation.md: 348f7ab0a26e9b39057dbac885304e0d52e0b1fb
+2026-08-05-per-agent-tool-presentation.zh.md: 4920ee6eb061d44934bfc9f5176e244f5aac8553
@@ -0,0 +1,46 @@
+# Agent Note: Per-agent tool presentation, and the `code` preset
+
+Status: implemented
+
+English | [中文](2026-08-05-per-agent-tool-presentation.zh.md)
+
+## Problem
+
+Agent presets compose an agent's tools per session, but not the FORM those tools reach the model in. Code Mode — one `run_code` tool plus a generated TypeScript SDK, replacing a call sequence with one program — was a deployment-wide `mode` field on the host's `dsh-tools` row. A deployment either ran every session in Code Mode or none, so the obvious product shape ("代码模式" beside 标准/极简/创造 in the preset picker) had nothing to hang on.
+
+The naive reading of "move tools down to the agent plane" does not work. `ctx.tools` has host-plane consumers that cannot follow it: `dsh-agent-loop` reads the registry's private scheduler seam, `dsh-apiproxy` reads its presenters to render tool cards, and every tool plugin registers into it. By the stack's own rule — a service moves into a preset only when ALL of its consumers move with it — the registry stays where it is.
+
+## Decision
+
+Split the registry from its projection. The registry stays host-plane; the **presentation** becomes per-agent state inside it, alongside the per-agent restrictions and guards that already live there.
+
+`ToolRegistry.presentAs(mode)` is scoped-only and mirrors `restrict()`: it writes one cell on the calling scope's `ToolLayer` through `ScopedLayers.effect`, so it unwinds with the agent that declared it. `modeFor(scope)` resolves that cell against the config `mode`, which becomes the default for agents declaring nothing rather than a process-wide fact. The three reads that decided presentation — the wire schemas, the `run_code` entry in the visibility view, and the generated SDK section — take the scope's mode instead of the service's.
+
+Two consequences fell out and are load-bearing:
+
+- **`run_code` is appended per scope.** Previously the transport entered every view whenever the transport existed. Per-agent, a native agent must not find `run_code` in its dispatch table because some other agent in the process presents it — so the append is conditional on that scope's own mode, and the transport is built lazily on first need.
+- **The reserved name is now unconditional.** `run_code` was rejected as a registration only while a code mode was configured. Any agent may now select a code mode, so a name that was free to take under a native deployment would become a collision the moment a preset mounted.
+
+The SDK prompt section is registered globally by a code-mode deployment (unchanged) and additionally per agent by `presentAs`, where it shadows by name. Its body renders empty for a native scope, which the prompt renderer drops — that is what keeps an agent opting OUT of a code-mode deployment free of an SDK section.
+
+The preset expresses the choice through one row, `@deepseek-ai/dsh-agent-tool-mode`, whose whole body is a `presentAs` call. A code mode waits for `ctx.codeRuntime` through `ctx.inject` rather than assuming it: the runtime is host-plane, and a pending row is what `dsh-agent-presets` already reports as an unusable mount, naming the row — so a preset selecting Code Mode against a runtime-less deployment fails where an operator can act.
+
+## Alternatives considered
+
+**A second `ToolRegistry` inside the preset's isolate realm.** Rejected: `dsh-agent-loop` resolves the registry once from the host context through a private symbol, so a per-agent registry would be invisible to the scheduler. Making the loop registry-per-agent is a far larger change than making one field scope-aware.
+
+**A top-level key in the preset's own YAML.** Rejected for the reason preset display metadata went to a separate `preset.yml`: the composition is a top-level list of plugin rows and cannot carry sibling keys.
+
+**Naming the package `dsh-tool-mode`.** Rejected by a gate, correctly. `gen-tool-catalog` globs `packages/*/tool-*` and requires every match to publish a model-facing tool schema, because that prefix means "ships a tool" in this repo. This row ships none.
+
+**Registering the SDK section unconditionally from the constructor.** Rejected after trying it: `renderPrompt` drops empty sections but `PromptAssembly.sections` retains them, so every native deployment would carry a `tools:sdk` entry rendering nothing, and two existing assertions on that list would have had to be weakened to accommodate it.
+
+**Sharing `standard`'s composition by include.** Rejected per the stack's own convention: `cordis` already duplicates `standard`, and a preset's value is that its whole composition is readable in one file. The cost — a third copy of ~240 lines that must move together — is real and is the strongest argument for a future include mechanism.
+
+## Consequences
+
+Two sessions in one process can now present differently, so "which tools does the model see" is no longer answerable from the deployment config alone; it requires the agent. Every diagnostic that quotes a mode now quotes the scope's, not the service's.
+
+`ctx.tools.schemas(agent)` remains the agent's CAPABILITY catalog and is unchanged by presentation — only the assembly's tools collapse. Tests asserting what the model receives must read the assembly; `web-agent-presets.spec.ts` asserts both sides of that distinction for the shipped `code` preset.
+
+The shipped roster is four presets (标准/代码/极简/创造), so any golden listing them moves. A deployment that composes no code runtime can compose no code-mode preset; the shipped Web overlay carries one, the base composition does not.
@@ -0,0 +1,46 @@
+# Agent Note: 按 agent 的工具呈现方式，以及 `code` 预设
+
+Status: implemented
+
+[English](2026-08-05-per-agent-tool-presentation.md) | 中文
+
+## Problem
+
+agent preset 已经能按会话组装一个 agent 的工具，却管不了这些工具以何种**形态**抵达模型。Code Mode——一个 `run_code` 工具加一份生成的 TypeScript SDK，用一段程序替代一串调用——此前是宿主 `dsh-tools` 那一行上的部署级 `mode` 字段。一个部署要么所有会话都跑 Code Mode，要么一个都不跑，于是那个显而易见的产品形态（预设选择器里「代码模式」与标准/极简/创造并列）无处安放。
+
+「把 tools 下沉到 agent 平面」这个字面读法行不通。`ctx.tools` 有一批跟不下来的宿主平面消费者：`dsh-agent-loop` 读它私有的调度器 seam，`dsh-apiproxy` 读它的 presenter 来渲染工具卡，每个工具插件都往里注册。按本 stack 自己的规则——只有**所有**消费者一起下沉，服务才能下沉——注册表必须留在原地。
+
+## Decision
+
+把注册表和它的投影拆开。注册表留在宿主平面；**呈现方式**变成它内部按 agent 的状态，与已经住在那里的按 agent 限制和守卫并列。
+
+`ToolRegistry.presentAs(mode)` 只接受 scoped 上下文，形状照抄 `restrict()`：它通过 `ScopedLayers.effect` 在调用方 scope 的 `ToolLayer` 上写一个单元，因此会随声明它的那个 agent 一起卸载。`modeFor(scope)` 将该单元与 config 的 `mode` 一并解析，后者于是成为「未作声明的 agent」的默认值，而不再是进程级事实。原先决定呈现方式的三处读取——wire schema、可见性视图里的 `run_code` 条目、以及生成的 SDK 段——改为读取该 scope 的模式，而非服务的。
+
+有两个随之而来的结果，且都是承重的：
+
+- **`run_code` 按 scope 追加。** 此前只要传输存在，它就进入每一个视图。按 agent 之后，一个 native agent 不能因为进程里别的 agent 呈现了它、就在自己的分发表里看到 `run_code`——因此这次追加以该 scope 自身的模式为条件，传输也改为首次需要时才构建。
+- **保留名现在无条件生效。** `run_code` 此前只在配置了 code 模式时才被拒绝注册。如今任何 agent 都可能选择 code 模式，因此一个在 native 部署下可以随便占用的名字，会在某个 preset 挂载的那一刻变成冲突。
+
+SDK 提示词段由 code 模式的部署全局注册（不变），并由 `presentAs` 额外按 agent 注册一份，后者按名字遮蔽前者。它的正文对 native scope 渲染为空，而提示词渲染器会丢弃空段——正是这一点让「在 code 模式部署下选择退出」的 agent 不带 SDK 段。
+
+preset 用一行来表达这个选择：`@deepseek-ai/dsh-agent-tool-mode`，其全部内容就是一次 `presentAs` 调用。code 类模式通过 `ctx.inject` 等待 `ctx.codeRuntime` 而非假定它存在：运行时在宿主平面，而一个 pending 的行正是 `dsh-agent-presets` 已经会报告的「不可用挂载」并会指名该行——于是在无运行时的部署上选择 Code Mode 的 preset，会在操作者能够动手的地方失败。
+
+## Alternatives considered
+
+**在 preset 的 isolate realm 里再起一个 `ToolRegistry`。** 否决：`dsh-agent-loop` 通过一个私有 symbol 从宿主上下文一次性解析注册表，因此按 agent 的注册表对调度器不可见。把 loop 改成按 agent 解析注册表，远比把一个字段变成 scope 感知的改动大。
+
+**在 preset 自己的 YAML 里加一个顶层键。** 否决，理由与 preset 展示元数据落到独立 `preset.yml` 相同：组装是一个顶层的插件行列表，装不下并列的键。
+
+**把包命名为 `dsh-tool-mode`。** 被一道 gate 否决，而且它是对的。`gen-tool-catalog` 以 `packages/*/tool-*` 通配，并要求每个命中项发布一个面向模型的工具 schema——因为在本仓库里这个前缀就意味着「带工具」。而这一行不带任何工具。
+
+**在构造函数里无条件注册 SDK 段。** 试过之后否决：`renderPrompt` 会丢弃空段，但 `PromptAssembly.sections` 会保留它们，于是每个 native 部署都将携带一个什么也不渲染的 `tools:sdk` 条目，而两处既有断言不得不为此放宽。
+
+**用 include 共享 `standard` 的组装。** 按本 stack 自己的惯例否决：`cordis` 已经复制了一份 `standard`，而 preset 的价值恰在于整份组装能在一个文件里读完。代价——第三份约 240 行、且必须同步演进的副本——是真实的，也正是未来引入 include 机制最有力的论据。
+
+## Consequences
+
+同一进程内的两个会话现在可以有不同的呈现方式，因此「模型看到哪些工具」不再能只凭部署配置回答，必须给出 agent。凡是引用模式的诊断信息，现在引用的都是该 scope 的，而不是服务的。
+
+`ctx.tools.schemas(agent)` 仍然是该 agent 的**能力**清单，不受呈现方式影响——坍缩的只是 assembly 里的工具。断言「模型收到什么」的测试必须读 assembly；`web-agent-presets.spec.ts` 对随附的 `code` 预设同时断言了这个区分的两侧。
+
+随附的名单变成四个预设（标准/代码/极简/创造），因此任何列出它们的 golden 都会变动。未组装 code 运行时的部署无法组装任何 code 模式的 preset；随附的 Web overlay 带了一个，base 组装没有。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-05-pwsh-ui-bash-parity.md
-2026-08-05-pwsh-ui-bash-parity.md: ba92f482e957acd00792a2a8053716047a4da27d
-2026-08-05-pwsh-ui-bash-parity.zh.md: 693b3fc26e632718ea2135282d63daa32e63e1c5
+2026-08-05-pwsh-ui-bash-parity.md: a59ae95ab64ae28babc913958a1a6ba424898210
+2026-08-05-pwsh-ui-bash-parity.zh.md: 8e065640e56bf6d6b92e62cb746fa9dd53f6f2b5
@@ -6,7 +6,7 @@ English | [中文](2026-08-05-pwsh-ui-bash-parity.zh.md)

 ## Problem

-The [pwsh tool bash parity decision](../../implemented/feature/2026-08-02-pwsh-tool-bash-parity.md) made `dsh-tool-pwsh` behaviorally interchangeable with `dsh-tool-bash` for execution, markers, and background tasks, but explicitly deferred the human-visible half: a completed pwsh foreground call presented as a generic `console`-fenced card while the bash tool's completed call presented as a terminal card with a parsed exit-status pill. The roadmap that owned this gap ([Windows defaults to pwsh](../../proposed/feature/2026-08-01-windows-pwsh-default.md)) named "pwsh TUI/GUI rendering" as stage 2, but the TUI package was removed, leaving the Web surface as the only UI the gap affects.
+The [pwsh tool bash parity decision](../../implemented/feature/2026-08-02-pwsh-tool-bash-parity.md) made `dsh-tool-pwsh` behaviorally interchangeable with `dsh-tool-bash` for execution, markers, and background tasks, but explicitly deferred the human-visible half: a completed pwsh foreground call presented as a generic `console`-fenced card while the bash tool's completed call presented as a terminal card with a parsed exit-status pill. The roadmap that owned this gap ([Windows defaults to pwsh](../../implemented/feature/2026-08-01-windows-pwsh-default.md)) named "pwsh TUI/GUI rendering" as stage 2, but the TUI package was removed, leaving the Web surface as the only UI the gap affects.

 ## Decision

@@ -6,7 +6,7 @@ Status: implemented

 ## Problem

-[pwsh 工具与 bash 对齐决策](../../implemented/feature/2026-08-02-pwsh-tool-bash-parity.md) 让 `dsh-tool-pwsh` 在执行、marker 与后台任务上行为可互换，但明确推迟了面向人类的一半：完成的 pwsh 前台调用呈现为通用 `console` 围栏卡片，而 bash 工具的完成调用呈现为带解析退出状态 pill 的 terminal 卡。拥有此缺口的路线图（[Windows 默认改用 pwsh](../../proposed/feature/2026-08-01-windows-pwsh-default.md)）把 "pwsh TUI/GUI 渲染" 列为阶段 2，但 TUI 包已被移除，使 Web 表面成为该缺口唯一影响的 UI。
+[pwsh 工具与 bash 对齐决策](../../implemented/feature/2026-08-02-pwsh-tool-bash-parity.md) 让 `dsh-tool-pwsh` 在执行、marker 与后台任务上行为可互换，但明确推迟了面向人类的一半：完成的 pwsh 前台调用呈现为通用 `console` 围栏卡片，而 bash 工具的完成调用呈现为带解析退出状态 pill 的 terminal 卡。拥有此缺口的路线图（[Windows 默认改用 pwsh](../../implemented/feature/2026-08-01-windows-pwsh-default.md)）把 "pwsh TUI/GUI 渲染" 列为阶段 2，但 TUI 包已被移除，使 Web 表面成为该缺口唯一影响的 UI。

 ## Decision

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-06-web-queue-steer-all-gesture.md
+2026-08-06-web-queue-steer-all-gesture.md: e546f68647dfc9b91ce4699cef4a64694ebc4f76
+2026-08-06-web-queue-steer-all-gesture.zh.md: fb36852f66a86408af12df59001230b2024eccaf
@@ -0,0 +1,33 @@
+# Agent Note: Steer the whole Web queue with an empty-draft Cmd/Ctrl+Enter
+
+Status: implemented
+
+English | [中文](2026-08-06-web-queue-steer-all-gesture.zh.md)
+
+## Problem
+
+While a primary session runs, the Web queue accumulates messages the user typed with plain Enter (or queued while the busy-Enter preference was Queue). Flushing them into the current turn required clicking the per-row 插话发送 button once per message; an empty composer draft had no keyboard gesture at all — the input machine rejects empty drafts, so Enter and Cmd/Ctrl+Enter were both no-ops. With several queued messages, steering them one by one is the obvious multi-click friction, and the empty-draft accelerated chord is the natural slot for "steer everything".
+
+## Decision
+
+Empty-draft Cmd/Ctrl+Enter now steers every still-pending `queued`-placement inbox row into the running turn, in FIFO order, on a primary session that reports running. The gesture decodes in `InputBar.onKeyDown`: accelerated Enter with a trimmed-empty draft, `running`, no subagent address, and at least one `queued` row calls the new `ComposerKeyboard.steerQueue()` verb instead of `submit()`. `SessionInputShell.steerQueue()` delegates to a hub-wired choreography that re-reads the authoritative `session/queue` snapshot, filters `placement: 'queued'` (pending steering rows are already in the turn), and applies the queue dock's exact strict-steer operation — `session.updateQueue(itemId, { kind: 'steer' })` — sequentially, so FIFO ordering is guaranteed at the host. A `steer-unavailable` (turn closed mid-flush) or `queue-item-not-found` (row claimed meanwhile) converges silently; any other failure surfaces one composer notice (`插话发送失败，请重试。`). No wire, on-disk, or agent-loop change: the host already owns the strict-steer boundary.
+
+The gesture is strictly the accelerated chord. Plain Enter with an empty draft stays a no-op even under the busy-Enter Steer preference, draft content outranks the queue (accelerated Enter steers only the draft), and idle or subagent sessions keep the existing empty-draft no-op because steering has no live turn to enter.
+
+The same computed availability gate drives discovery: while the draft is empty, the input is unlocked and not in a transient machine lock, the command menu is closed, an ordinary primary session is running, and at least one row remains `queued`, the textarea placeholder advertises that Cmd/Ctrl+Enter steers all queued messages. An owner-supplied placeholder still takes precedence, and the steer hint deliberately outranks the plan-mode placeholder while available (the gesture genuinely works in that window).
+
+## Consequences
+
+One keyboard gesture now replaces N clicks while keeping a single strict-steer path and a single authority for convergence. The per-row button and the gesture are the same host operation, so races and failure semantics stay identical. The gesture and its placeholder share one presentation-layer gate, while the hub re-checks the snapshot at execution time, so the client gate remains advisory and the host remains authoritative.
+
+## Related
+
+The per-row 插话发送 action and its strict-steer boundary are owned by [Steer a queued Web message into the active turn](../feature/2026-07-30-web-queue-steer-action.md); this note only adds the whole-queue keyboard gesture on top of that decision.
+
+## Alternatives considered
+
+- **Intercepting inside the input machine.** Rejected: the machine is queue-agnostic by design (the wiring layer overlays the queue projection) and cannot distinguish the accelerated chord from plain Enter, which must stay a no-op.
+- **Steering via `session.prompt(mode: 'steer')` per row.** Rejected: that mints new messages instead of transferring the pending occurrences and would split the dock's immutable-message contract; `updateQueue({ kind: 'steer' })` already atomically transfers the exact occurrence.
+- **Firing all row steers concurrently.** Rejected: arrival order at the host is not guaranteed, and steering order is model-visible; sequential awaits preserve FIFO.
+- **A new host RPC for steer-all.** Rejected: the existing per-item operation is idempotent enough — each row is one strict steer, and mid-flush closure converges silently — so a protocol change buys nothing.
+- **A send-button tooltip.** Rejected: the primary button is Stop while an ordinary session is running, which is the only window where the whole-queue gesture is available. The empty-draft placeholder occupies that exact window and can describe the keyboard action directly.
@@ -0,0 +1,33 @@
+# Agent Note: 空输入时 Cmd/Ctrl+Enter 将 Web 排队消息全部插话
+
+Status: implemented
+
+[English](2026-08-06-web-queue-steer-all-gesture.md) | 中文
+
+## Problem
+
+主会话运行时，用户用普通 Enter（或在 busy-Enter 偏好为 Queue 时）输入的消息会在 Web 队列里累积。把它们灌进当前轮次需要逐条点击「插话发送」按钮；而输入框草稿为空时没有任何键盘手势——输入机对空草稿直接拒绝，Enter 与 Cmd/Ctrl+Enter 都是空操作。排队消息一多，逐条插话是明显的多点摩擦，空草稿 + 加速 Enter 正是「全部插话」的自然位置。
+
+## Decision
+
+空草稿的 Cmd/Ctrl+Enter 现在会把仍在排队（`placement: 'queued'`）的 Inbox 行按 FIFO 顺序全部插话进运行中的轮次，仅限报告 running 的主会话。手势在 `InputBar.onKeyDown` 解码：加速 Enter + 去空白后为空草稿 + `running` + 无 subagent 地址 + 至少一条 `queued` 行时，改走新的 `ComposerKeyboard.steerQueue()` 动词而不是 `submit()`。`SessionInputShell.steerQueue()` 委托给 hub 编排的流程：重新读取权威的 `session/queue` 快照，过滤 `placement: 'queued'`（pending steering 行已经在本轮内），并逐条顺序执行 Queue 面板的严格 steer 操作 `session.updateQueue(itemId, { kind: 'steer' })`，从而在 host 侧保证 FIFO 顺序。`steer-unavailable`（flush 中途轮次关闭）或 `queue-item-not-found`（行已被占用）静默收敛；其他失败弹出一条 composer 通知（「插话发送失败，请重试。」）。无 wire、磁盘或 agent-loop 改动：严格 steer 边界本来就在 host 侧。
+
+该手势严格限定为加速组合键。空草稿 + 普通 Enter 仍然无操作（即使 busy-Enter 偏好为 Steer）；草稿内容优先于队列（加速 Enter 只插话当前草稿）；idle 或 subagent 会话保持原有空草稿无操作，因为没有可插入的运行中轮次。
+
+同一套计算得出的可用性门控也负责提示该手势：当草稿为空、输入框未锁定且不处于瞬态机器锁（adjudicating/submitting）、命令菜单未打开、普通主会话正在运行且至少一行仍为 `queued` 时，文本框 placeholder 会提示 Cmd/Ctrl+Enter 将全部排队消息插话发送。owner 提供的 placeholder 仍然优先；可用时 steer 提示会刻意优先于 plan 模式 placeholder（该窗口内手势确实可用）。
+
+## Consequences
+
+一个键盘手势替代 N 次点击，同时保持单一严格 steer 路径与单一收敛权威。逐条按钮与手势是同一个 host 操作，竞态与失败语义完全一致。手势及其 placeholder 共用一个呈现层门控；hub 在执行时会重新读取快照，因此客户端门控仍只是建议性的，host 仍是权威。
+
+## Related
+
+逐条「插话发送」动作及其严格 steer 边界由 [Steer a queued Web message into the active turn](../feature/2026-07-30-web-queue-steer-action.md) 记录；本笔记只在其之上增加整队列键盘手势。
+
+## Alternatives considered
+
+- **在输入机内拦截。** 已拒绝：输入机按设计不感知队列（队列投影由 wiring 层叠加），且无法区分加速 Enter 与必须保持空操作的普通 Enter。
+- **逐条用 `session.prompt(mode: 'steer')` 插话。** 已拒绝：那会铸造新消息而不是转移 pending 行，破坏 dock 的不可变消息契约；`updateQueue({ kind: 'steer' })` 已经原子地转移了确切的那条。
+- **并发触发所有行。** 已拒绝：host 到达顺序无法保证，而插话顺序对模型可见；顺序 await 保证 FIFO。
+- **为 steer-all 新增 host RPC。** 已拒绝：现有逐条操作已足够幂等——每行一次严格 steer，中途关闭静默收敛——协议改动没有收益。
+- **发送按钮 tooltip。** 已拒绝：普通会话运行时，主按钮是 Stop，这也是整队列手势唯一可用的窗口。空草稿时的 placeholder 恰好在该窗口显示，可以直接说明这项键盘操作。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-07-default-model-follows-the-picker.md
-2026-08-07-default-model-follows-the-picker.md: ed7e7a424d2cacadea890506fd9150ffdf7a993c
-2026-08-07-default-model-follows-the-picker.zh.md: 523c6f917dedf21c726ce5631226ba545126c757
+2026-08-07-default-model-follows-the-picker.md: 2a3ada55486345c0f58f0767bed5a93ecba04b88
+2026-08-07-default-model-follows-the-picker.zh.md: 08fecc6ec9b177f6424ada3172c67018ac72baea
@@ -6,13 +6,13 @@ English | [中文](2026-08-07-default-model-follows-the-picker.zh.md)

 ## Problem

-A session model picker and a deployment default are two layers of the same preference. If the picker affects only its addressed session, the next blank session can select a different model with no user-facing way to align the default. If the default lives inside a Host gateway, direct Agent front doors cannot share it without depending on Host or duplicating state.
+A session model picker and a deployment default are two layers of the same preference. If the picker affects only its addressed session, the next blank session can select a different model with no user-facing way to align the default. If the default lives inside a Host gateway, direct Agent entry points cannot share it without depending on Host or duplicating state.

 Reasoning effort makes the persistence shape significant: a model selection without an effort must clear a stored effort, or the next Agent may apply an effort that its selected model does not accept.

 ## Decision

-`AgentDefaultModelService` provides `ctx.agentDefaultModel` and registers `{provider, model, reasoningEffort?}` as the `agent-default-model` Settings section. Its `{provider, model}` composition entry is the base layer and `settings.yaml` supplies the user layer. The service is front-door-neutral, so direct creation and ApiProxy-backed creation share one default ([headless direct core front door](../architecture/2026-08-09-headless-direct-core-front-door.md)). `workspaceRoot` remains ApiProxy config because it is a Host launcher fact rather than model state.
+`AgentDefaultModelService` provides `ctx.agentDefaultModel` and registers `{provider, model, reasoningEffort?}` as the `agent-default-model` Settings section. Its `{provider, model}` composition entry is the base layer and `settings.yaml` supplies the user layer. The service is entry-point-neutral, so direct creation and ApiProxy-backed creation share one default ([headless direct core entry point](../architecture/2026-08-09-headless-direct-core-entry-point.md)). `workspaceRoot` remains ApiProxy config because it is a Host launcher fact rather than model state.

 `reasoningEffort` belongs to the Settings section but not to the plugin config. Settings layers merge by field, so a configured effort would survive a user selection that omits it. `saveSelection()` instead writes the complete user section; absence therefore clears a stored effort. A deployment-wide effort default belongs to the adapter profile, which resolves it per model.

@@ -6,13 +6,13 @@ Status: implemented

 ## 问题

-会话模型选择器与部署默认值是同一项偏好的两个层次。如果选择器只影响其所在会话，下一个空白会话可能选择不同模型，用户却没有途径使默认值与选择器一致。如果默认值位于 Host 网关内部，直接创建 Agent 的前门只有依赖 Host 或复制状态才能共享它。
+会话模型选择器与部署默认值是同一项偏好的两个层次。如果选择器只影响其所在会话，下一个空白会话可能选择不同模型，用户却没有途径使默认值与选择器一致。如果默认值位于 Host 网关内部，直接创建 Agent 的入口只有依赖 Host 或复制状态才能共享它。

 推理强度使持久化形态成为约定的一部分：不含强度的模型选择必须清除已存强度，否则下一个 Agent 可能会采用所选模型不接受的强度。

 ## 决定

-`AgentDefaultModelService` 提供 `ctx.agentDefaultModel`，并把 `{provider, model, reasoningEffort?}` 注册为 `agent-default-model` Settings 分节。其 `{provider, model}` 组合条目是 base 层，`settings.yaml` 提供用户层。该服务不偏向特定前门，因此直接创建与 ApiProxy 支撑的创建共享同一个默认值（[headless 直接 core 前门](../architecture/2026-08-09-headless-direct-core-front-door.md)）。`workspaceRoot` 仍是 ApiProxy 配置，因为它是 Host 启动器事实，而不是模型状态。
+`AgentDefaultModelService` 提供 `ctx.agentDefaultModel`，并把 `{provider, model, reasoningEffort?}` 注册为 `agent-default-model` Settings 分节。其 `{provider, model}` 组合条目是 base 层，`settings.yaml` 提供用户层。该服务不偏向特定入口，因此直接创建与 ApiProxy 支撑的创建共享同一个默认值（[headless 直接 core 入口](../architecture/2026-08-09-headless-direct-core-entry-point.md)）。`workspaceRoot` 仍是 ApiProxy 配置，因为它是 Host 启动器事实，而不是模型状态。

 `reasoningEffort` 属于 Settings 分节，但不属于插件配置。Settings 层按字段合并，因此已配置的强度会在用户选择省略它时继续存在。`saveSelection()` 写入完整的用户分节；缺席值由此清除已存强度。部署级强度默认值属于适配器 profile，并由它按模型解析。

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-08-dsh-run-headless-command.md
-2026-08-08-dsh-run-headless-command.md: 566eeb5b2a09a0d07d72a68e4a5f449d2822e708
-2026-08-08-dsh-run-headless-command.zh.md: 177410e783a37026940829da5f81343ddc61cb29
+2026-08-08-dsh-run-headless-command.md: ed095f4077a23e51bffb647d24eed19ba09e11ed
+2026-08-08-dsh-run-headless-command.zh.md: 89d54e35573f14786e05d648f2b42891ca27a043
@@ -22,7 +22,7 @@ dsh run [--profile <name>] [--patch <path>...] <task...>

 `RunInvocation` is a distinct `DshInvocation` member. The generic profile invocation carries no task state and accepts no positional arguments. Both dispatch paths use `runProfile`: profile boot omits `task`, while `run` supplies it. A one-shot profile without `headless-runner` fails through the composed-row check, and profile boot containing that row without a task points to `dsh run --profile <name> "<task>"`.

-The [profile plugin bundle decision](../architecture/2026-08-05-profile-plugin-bundles.md) owns composition. [Headless is a direct core front door](../architecture/2026-08-09-headless-direct-core-front-door.md) owns the execution contract: one fresh persisted Session, final assistant text on stdout, completed/non-completed exit mapping, empty stderr on success, no listening port, and bounded signal shutdown after Agent quiescence and Session flush.
+The [profile plugin bundle decision](../architecture/2026-08-05-profile-plugin-bundles.md) owns composition. [Headless is a direct core entry point](../architecture/2026-08-09-headless-direct-core-entry-point.md) owns the execution contract: one fresh persisted Session, final assistant text on stdout, completed/non-completed exit mapping, empty stderr on success, no listening port, and bounded signal shutdown after Agent quiescence and Session flush.

 The `run` verb belongs only to one-shot task execution. Application-file launch requires a distinct command name.

@@ -22,7 +22,7 @@ dsh run [--profile <name>] [--patch <path>...] <task...>

 `RunInvocation` 是单独的 `DshInvocation` 成员。通用 profile 调用不携带任务状态，也不接受位置参数。两条分派路径都使用 `runProfile`：profile 启动省略 `task`，而 `run` 提供该字段。缺少 `headless-runner` 的一次性 profile 会触发组合行检查；如果启动的 profile 包含该行却未提供任务，错误会指向 `dsh run --profile <name> "<task>"`。

-[profile 插件组合包决策](../architecture/2026-08-05-profile-plugin-bundles.md)负责组合。[Headless 是直接 core 前门](../architecture/2026-08-09-headless-direct-core-front-door.md)负责执行约定：一个新的持久化会话、stdout 上的最终 assistant 文本、completed／非 completed 的退出状态映射、成功时为空的 stderr、无监听端口，以及 Agent 完全停稳且会话 flush 后的有界信号关闭。
+[profile 插件组合包决策](../architecture/2026-08-05-profile-plugin-bundles.md)负责组合。[Headless 是直接 core 入口](../architecture/2026-08-09-headless-direct-core-entry-point.md)负责执行约定：一个新的持久化会话、stdout 上的最终 assistant 文本、completed／非 completed 的退出状态映射、成功时为空的 stderr、无监听端口，以及 Agent 完全停稳且会话 flush 后的有界信号关闭。

 `run` 动词只负责一次性任务执行。应用文件启动需要不同的命令名。

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-08-user-explicit-skill-invocation.md
-2026-08-08-user-explicit-skill-invocation.md: 74d9f01f191005db6d3d283a3c56f5ee664447f8
-2026-08-08-user-explicit-skill-invocation.zh.md: 0f7c9e1261dda796d988c17ddf7199b74be4133e
+2026-08-08-user-explicit-skill-invocation.md: a7c2c15703af318cb4112f2d3dfda698bc5e3bc2
+2026-08-08-user-explicit-skill-invocation.zh.md: a8d685e5eb12766bebddebb7f5993579cf08531e
@@ -10,14 +10,14 @@ A `disable-model-invocation: true` skill is user-only by design: it never enters

 ## Decision

-User-explicit invocation is a host-side pre-step injection, uniform for every user-invocable skill and every front end:
+User-explicit invocation is a host-side pre-step injection, uniform for every user-invocable skill and every entry point:

 - `dsh-tool-skill` registers a second `agent/pre-step` listener (beside its catalog listener, the same seam `workspace-instructions` and the runtime-context snapshot ride): it scans the step's claimed messages for whitespace-bounded `/name` tokens — anywhere in the text, the same word-boundary shape the transcript chip decoration uses — collects first-seen-deduplicated names, loads each through `ctx.skills.get`, checks `isUserInvocable` on the loaded definition (the single lookup that produces what is injected), renders it with the shared `renderSkillContent`, and appends the injections after every other injection of the step: background first (workspace rules, runtime policy, catalog), the material the model must act on last, closest to its answer. Registration order pins the placement — the gesture listener registers before the catalog listener, so the waterfall hands it the catalog-bearing list to extend.
 - Precision is closed-set matching, exactly like slash commands: `/goal` resolves against the command registry, `/name` against the workspace's user-invocable skill directory; a miss stays ordinary prose, so nothing is ever guessed. Only `source.kind === 'user'` messages are scanned — external text cannot forge a gesture. Paths (`/usr/bin`), fractions (`5/8`), and prefixed tokens (`foo/name`) all break the boundary.
 - The client keeps the [plain-text-reference decision](../architecture/2026-07-25-web-input-machine-and-slash-pipeline.md): a menu pick lands the literal `/name ` and the prompt ships it verbatim; ui-skill implements no adjudication hooks and no reference codec. `skill.list` (now the domain's only RPC) serves every user-invocable skill with `modelInvocable` so menus mark user-only entries. A name shared with a host command resolves to the command — adjudication claims the line client-side before it becomes a prompt.
 - The injection is a `user`-role message carrying the `skill-invocation` source (`{ name, form: 'instructions' }`), so `user/message` logging, the context-injection transcript row (labelled with the skill name), and replay all come free; `renderSkillContent` lives in the `dsh-skill` seam, shared verbatim with the `skill` tool result, and the catalog's closing sentence tells the model to follow an injected block instead of re-loading it.

-Peer-product survey (Pi, OpenCode, Claude Code, Kimi Code, Codex, DeepSeek-Reasonix — local checkouts) was unanimous that user-explicit triggering is programmatic injection with zero model participation; the final shape is closest to Codex's core-side `$name` mention scanning, which likewise frees every front end from implementing recognition.
+Peer-product survey (Pi, OpenCode, Claude Code, Kimi Code, Codex, DeepSeek-Reasonix — local checkouts) was unanimous that user-explicit triggering is programmatic injection with zero model participation; the final shape is closest to Codex's core-side `$name` mention scanning, which likewise frees every entry point from implementing recognition.

 ## Alternatives considered

@@ -10,14 +10,14 @@ Status: implemented

 ## 决策

-用户显式调用是一次宿主侧的 pre-step 注入，对每一个用户可调用的 skill 和每一种前端一致：
+用户显式调用是一次宿主侧的 pre-step 注入，对每一个用户可调用的 skill 和每一种运行入口一致：

 - `dsh-tool-skill` 注册第二个 `agent/pre-step` 监听器（与其目录监听器并列，也是 `workspace-instructions` 与运行时上下文快照搭乘的同一 seam）：它在该步骤已认领的消息中扫描以空白为界的 `/name` token——文本中任意位置均可，与 transcript（文本记录）chip 装饰所用的词边界形状相同——收集按首见去重的名称，逐个经 `ctx.skills.get` 加载，在已加载定义上检查 `isUserInvocable`（产生注入内容的正是这同一次查找），用共享的 `renderSkillContent` 渲染，并把注入追加在该步骤所有其他注入之后：背景在前（工作区规则、运行时策略、目录），模型必须着手处理的材料在最后、最贴近它的回答。注册顺序钉住了这一位置——手势监听器先于目录监听器注册，因此 waterfall（瀑布式事件）会把携带目录的列表交给它来扩展。
 - 精确性来自封闭集合匹配，与斜杠命令完全一致：`/goal` 对照命令注册表解析，`/name` 对照工作区的用户可调用 skill 目录解析；未命中即保持为普通行文，因此绝不猜测。只扫描 `source.kind === 'user'` 的消息——外部文本无法伪造手势。路径（`/usr/bin`）、分数（`5/8`）与带前缀的 token（`foo/name`）都会破坏该边界。
 - 客户端沿用[纯文本引用决策](../architecture/2026-07-25-web-input-machine-and-slash-pipeline.md)：菜单 pick 落下字面文本 `/name `，该文本随提示词原样提交；ui-skill 不实现任何裁决钩子，也没有引用 codec。`skill.list`（现在是该领域唯一的 RPC）提供每一个用户可调用的 skill 并携带 `modelInvocable`，供菜单标出仅限用户的条目。与宿主命令同名的名称解析为命令——客户端会在该行成为提示词之前完成裁决并将其认领。
 - 注入是一条携带 `skill-invocation` 来源（`{ name, form: 'instructions' }`）的 `user` 角色消息，因此 `user/message` 落账、上下文注入的 transcript 行（以 skill 名称标注）与回放全部免费获得；`renderSkillContent` 位于 `dsh-skill` seam，由注入和 `skill` 工具结果共用，二者内容逐字相同，目录的结尾一句会告诉模型遵循注入块而不是重新加载。

-同类产品调研（Pi、OpenCode、Claude Code、Kimi Code、Codex、DeepSeek-Reasonix——本地检出）一致表明：用户显式触发都是模型零参与的程序化注入；最终形态最接近 Codex 核心侧的 `$name` mention 扫描——它同样让每一种前端免于自行实现识别。
+同类产品调研（Pi、OpenCode、Claude Code、Kimi Code、Codex、DeepSeek-Reasonix——本地检出）一致表明：用户显式触发都是模型零参与的程序化注入；最终形态最接近 Codex 核心侧的 `$name` mention 扫描——它同样让每一种运行入口免于自行实现识别。

 ## 考虑过的替代方案

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-08-windows-acl-restricted-token-sandbox.md
+2026-08-08-windows-acl-restricted-token-sandbox.md: 7e8f229269233d9ac9baa65241ca02a4cf4c3f7c
+2026-08-08-windows-acl-restricted-token-sandbox.zh.md: eeb346b228b3559f487448e5d4ec525b7bb89525
@@ -0,0 +1,43 @@
+# Agent Note: Windows sandbox rung: raw ACL restricted tokens over mxc and AppContainer
+
+Status: implemented
+
+English | [中文](2026-08-08-windows-acl-restricted-token-sandbox.zh.md)
+
+## Problem
+
+The [sandbox decision](2026-07-06-sandbox.md) leaves `PLATFORM_CHAINS.win32` empty, so shipped Windows profiles degrade to danger-full-access because no confining executor exists. The win32 rung must confine the two file-effect modes the sandbox vocabulary promises — `read-only` (zero writes) and `workspace-write` (writes under the workspace root plus a backend-defined temp area) — while leaving reads, network, and process visibility alone, because every mode permits reading.
+
+## Decision
+
+Implement the rung directly on the raw ACL mechanism: duplicate the caller's token into a `WRITE_RESTRICTED` token (`CreateRestrictedToken` with `WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`) whose restricting SIDs include a write SID (`S-1-4-x-y`); the write SID's Write ACEs on the workspace and temp roots are the entire write allowlist, because `WRITE_RESTRICTED` intersects write accesses only and reads keep the caller's full ambient access. The mechanism is the one huoyaoyuan/windows-acl-restrict-poc (`10e4dfb`) demonstrates; this port checks every API call and fails closed (the POC fail-opened on every ignored return value). The write SID is the PER-WORKSPACE identity, derived deterministically from the canonical workspace path (`workspaceWriteSid` — sha256 → `S-1-4-x-y`) and stored NOWHERE: the workspace-root ACE therefore materializes once per workspace per machine — the standing ACE is the cross-session reuse cache, and every later provision hits the exact-ACE skip (idempotent re-grant skips the eager full-tree re-propagation — no garbage collection) — instead of once per session, which is what the earlier per-session random SID paid a full tree propagation per session for. The seam derives the session's PRIVATE temp subdirectory from the session id + workspace (sha256, 16 hex — stored nowhere, so no tamper surface exists) and creates it exclusively; it is removed on provider dispose, and a crash leaves it as `%TEMP%` litter whose next resume fails loudly at the exclusive creation until temp hygiene reclaims it. The seam materializes the workspace ACE STANDING (never revoked — the cache) and the temp ACE REVOCABLY (revoked on provider dispose, so an inheritable ACE never outlives its session's temp directory on the ambient temp root). The token's restricting list is the keep-alive group plus the write SID only under workspace-write: read-only = [logon SID, Everyone] and workspace-write = [logon SID, Everyone, write SID]. The keep-alive invariants are logon SID + Everyone (early DLL init dies with 0xC0000142 and CNG crashes pwsh with 0xE0434352 without them). Read-only carries no write SID: a standing grant ACE from an earlier workspace-write period stays INERT (the pass-2 check grants only what the list carries, so read-only remains strictly zero-grant across a `/permission` downgrade or a crash-resumed session, while the standing ACE keeps the re-upgrade free). Authenticated Users is absent from BOTH lists — the WMI namespace security check fails (0x80041003), so CIM is unavailable in every confined mode, and the C:\-root tree-creation escape (standing `AU:(AD)` + `AU:(OI)(CI)(IO)(M)` ACEs) is closed in both; INTERACTIVE/LOCAL are likewise absent from both (the Public tree writes are denied — pinned by the runner's Public-probe regression). Workspace-write children see a PRIVATE per-session temp subdirectory (`<temp>\dsh-<16 hex>` derived from the session id + workspace — created exclusively, reparse points rejected, removed on provider dispose — TMP/TEMP rewritten by the runner — bwrap `--tmpfs /tmp` semantics). The restricted token's DEFAULT DACL is extended with a full-access write-SID ACE (`SetTokenInformation(TokenDefaultDacl)`): new objects created without an explicit security descriptor (anonymous pipes — CreatePipe, sync objects) then carry a restricting-SID ACE and pass the write pass-2 check at creation; NAMED pipes are exempt — their default security descriptor is the Win32 layer's user-mode default SD template (built by KernelBase — owner/SYSTEM/Admins full, Everyone/ANONYMOUS read-only), which the token cannot influence, so piped stdio capture stays denied for confined grandchildren (the POC-documented boundary, pinned by the runner suite). It ships as [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md) (backend plus the `./runner` argv-prefix entry), the `win32` chain rung of [`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md), and [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) as the confining executor; the Windows platform layer re-enables the full permission surface (sandbox/sandbox-policy/permission/approval/fs-sandbox) over the confined pwsh stack.
+
+## How the restriction works (why no new identity)
+
+The identity routes restrict by *who* runs the child; this rung restricts by *token derivation*. An identity route (landstrip's restricted-user, AppContainer) runs the child under a fresh account or container SID that starts with zero ACEs on the host's files — everything, reads included, defaults to denied, and every path the child may touch must then be opened back up by writing ACEs for that identity: the wholesale DACL mutation that disqualified both alternatives. The restricted token keeps the caller's own SID and logon session: [`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) derives a token that adds the restricting SIDs and the `WRITE_RESTRICTED` flag, so Windows performs the access check twice — once against the normal SIDs, once against the restricting SIDs — and grants write-class access only where both checks pass. Reads pass on the normal check alone (the caller's SIDs already carry read access everywhere the caller can read), which is why this rung needs no read grants and no new account; writes must additionally clear the orphan-SID check, which only the workspace and temp ACEs satisfy. `DISABLE_MAX_PRIVILEGE | LUA_TOKEN` synthesize the limited-user effect of a fresh account token-side, so even an elevated caller derives a filtered token. The same primitive could restrict reads (`SidsToDisable` turning SIDs deny-only), but a read-restricted token would need per-path read grants — reintroducing exactly the cost the identity routes pay — and the sandbox vocabulary never requires read confinement.
+
+## Alternatives considered
+
+### Why not mxc (Microsoft xContainer)?
+
+Two disqualifiers. First, the OS floor is too new: the [mxc OS-version policy](https://github.com/microsoft/mxc/blob/main/docs/process-container/os-version-support.md) sets the product floor at Windows 11 24H2 (build 26100), and the BaseContainer tier (T1, `Experimental_CreateProcessInSandbox`) exists only on 25H2+ (build 26600+) with the OS feature enabled — on every supported release at or below 25H2 the filesystem policy falls back to T3, AppContainer plus host-side DACL ACE augmentation. Second, supporting arbitrary-path reads under either tier means granting read access by writing ACLs over every path the child may read: a model that reads the whole workspace and arbitrary files would require wholesale host DACL mutation — a standing side effect and a cost a write-only restriction does not need.
+
+### Why not AppContainer?
+
+An AppContainer token carries no ambient read access: every readable path must be pre-granted through capabilities or explicit ACEs, so arbitrary-path reads — the harness's read model — are unsupported without the same wholesale grants. The restricted token needs no read grants at all: it intersects write access only.
+
+### Why not landstrip?
+
+The [landstrip evaluation](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md) was rejected before implementation (not battle-tested; the in-house launcher plan won), and its Windows backend is AppContainer-shaped, inheriting the same arbitrary-read problem.
+
+## Consequences
+
+Bought: write-only confinement with no new OS floor (`CreateRestrictedToken` predates the mxc releases by two decades), reads/network/process visibility untouched exactly as the mode vocabulary requires, and fail-closed errors carrying the API name and the exact Win32 code. Cost: no read-side or network isolation; console isolation unavailable (hidden-console children die with `STATUS_DLL_INIT_FAILED`; children share the host console); standing ACE mutations on the granted roots (caller-owned directories; workspace ACEs stand forever by design — the reuse cache, invisible residue when a workspace is renamed — temp ACEs revoked by provider dispose together with the derived private temp directory — a crash leaves both behind and the next resume fails loudly at the exclusive creation until temp hygiene reclaims the directory); grant materialization is an EAGER full-tree propagation (`SetNamedSecurityInfoW` walks every descendant immediately — tens of seconds on large workspaces), paid once per workspace per machine by the per-workspace identity; CIM is unavailable in BOTH confined modes (AuthUsers dropped from both lists — the WMI namespace security check fails, and `Get-ComputerInfo` silently returns incomplete results) as the price of closing the C:\-root tree-creation escape in both; FAT-class (non-ACL) targets outside the granted roots remain writable under both modes (no security descriptors to intersect — a legacy residue treated as unsupported, warn-only, documented in the README); NULL-DACL directories are not identity-preserving under a grant+revoke round-trip (documented edge, the POC shares it); `whoami` and token-inspection cmdlets fail under the restricted token (diagnostic noise, documented); and BOTH confined modes run `pwsh` in ConstrainedLanguage mode — the restricted token trips PowerShell's lockdown detection, so `Add-Type`, non-core .NET statics (`[System.IO.*]::`, `[math]::`), COM objects, and reflection fail with "only core types" errors while `-f` formatting, property access, and core cmdlets/types keep working, and the language mode cannot be lifted back to FullLanguage from inside — taught to the model in the pwsh tool description and documented in the package README's Known Limitations; BOTH confined modes also deny named-pipe opens — libuv's piped-stdio spawns fail with EPERM (the POC-documented "no output redirection" boundary; inherited/ignored stdio and anonymous pipes work) — documented in the package README's Known Limitations and taught to the model in the pwsh tool description.
+
+## Testing
+
+The product-visible Windows roster flip is win32-only, so the keyless snapshot fixtures — which must replay on macOS/Linux — cannot cover it; the bundle composition specs ([`base.spec.ts`](../../../../packages/bundle/base/tests/base.spec.ts), [`windows-shell.spec.ts`](../../../../apps/cli/tests/windows-shell.spec.ts)) plus the win32 real-runner suites (`packages/sandbox/sandbox-windows-acl/tests/`, `packages/bash/pwsh-sandbox/tests/`) are the substitute evidence, and the CI Windows lane owns the assembled signal. The grant machinery is pinned cross-platform by `packages/sandbox/sandbox-local/tests/acl-grants.spec.ts` (the derived private-temp identity — deterministic per session + workspace, distinct across sessions — one-shot materialization, exclusive temp creation with reparse-point rejection and self-cleanup on failure, clean-restart re-grant of the same derived directory, the standing-vs-revocable lifecycle across dispose and the mode-switch cycle, and the derived-SID argv contract — with the Win32 surface mocked) and on win32 by `workspace-sid.spec.ts` (derivation determinism/shape/distinctness), `grant.spec.ts` (real-DACL materialization: revocable paths revoke on dispose, standing paths survive it), the `acl.spec.ts` idempotent-grant fast-path and standing-ACE-after-dispose contract, the `failure-paths.spec.ts` suspension-orphan regression (AssignProcessToJobObject failure terminates the child), and the `runner.spec.ts` `--write-sid` contract (caller-owned grants, private temp subdir through TMP/TEMP, both-mode CIM-denial probes, the mode-downgrade regression — a standing grant ACE is inert under read-only and effective again on re-upgrade — the ambient-writable Public-probe regression (a C:\Users\Public subdirectory write is denied under both modes), and the ConstrainedLanguage pins in both modes, plus the grandchild-stdio matrix pins — inherited/ignored stdio spawns succeed while piped capture is DENIED in both modes). The runner-failure classification is exit-gated on 127 (a confined command that merely prints the `windows-acl-run:` signature on a non-127 exit is never misclassified as "the command did not run" — pinned in the pwsh-sandbox helper suite).
+
+## Related
+
+The [pwsh executor decision](2026-08-01-pwsh-tool-and-executor.md) owns the pwsh-sandbox/tool-pwsh dialect split this rung consumes.
@@ -0,0 +1,43 @@
+# Agent Note: Windows sandbox rung: raw ACL restricted tokens over mxc and AppContainer
+
+Status: implemented
+
+[English](2026-08-08-windows-acl-restricted-token-sandbox.md) | 中文
+
+## Problem
+
+[沙盒决策](2026-07-06-sandbox.md)把 `PLATFORM_CHAINS.win32` 留空，交付的 Windows profile 因为没有可用的隔离执行器而退化为 danger-full-access。win32 档必须实现沙盒词汇表承诺的两个文件效果模式——`read-only`（零写入）与 `workspace-write`（仅工作区根目录加后端定义的临时区域可写）——同时保持读、网络与进程可见性不受影响，因为所有模式都允许读取。
+
+## Decision
+
+直接基于原始 ACL 机制实现该档：把调用者令牌复制为 `WRITE_RESTRICTED` 受限令牌（`CreateRestrictedToken`，`WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`），其 restricting SIDs 中包含写入 SID（`S-1-4-x-y`）；工作区与临时目录上写入 SID 的 Write ACE 就是全部写入白名单，因为 `WRITE_RESTRICTED` 只对写访问做交集检查，读保持调用者的完整环境访问。该机制来自 huoyaoyuan/windows-acl-restrict-poc（`10e4dfb`）的演示；本移植检查每一个 API 调用并 fail-closed（POC 因忽略返回值而 fail-open）。写入 SID 是**按工作区**的身份，由规范工作区路径确定性派生（`workspaceWriteSid`——sha256 → `S-1-4-x-y`），且**任何地方都不存储**：工作区根目录 ACE 因此每台机器每个工作区只物化一次——常驻 ACE 就是跨会话复用缓存，此后每次供给都命中精确 ACE 跳过（幂等重授权跳过急切的全树重传播——不做垃圾回收）——而不是每会话一次，这正是先前每会话随机 SID 每个会话都要付一次全树传播的代价。seam 从会话 id + 工作区派生会话的**私有**临时子目录（sha256、16 位 hex——任何地方都不存储，因此不存在篡改面）并独占创建；它在提供方 dispose 时移除，崩溃则把它留作 `%TEMP%` 垃圾，其下一次恢复会在独占创建处大声失败，直到临时目录卫生机制将其回收。seam 把工作区 ACE **常驻**物化（绝不撤销——就是缓存），把临时 ACE **可回收**物化（提供方 dispose（资源释放）时撤销，因此可继承 ACE 不会在环境临时根目录上比其会话的临时目录活得更久）。令牌的 restricting list 是保活组加上仅 workspace-write 下的写入 SID：read-only = [登录 SID、Everyone]，workspace-write = [登录 SID、Everyone、写入 SID]。保活不变式是登录 SID + Everyone（没有它们，早期 DLL init 会以 0xC0000142 死亡，CNG 会让 pwsh 以 0xE0434352 崩溃）。Read-only 不含写入 SID：先前 workspace-write 时期留下的常驻授权 ACE 保持**失效**（pass-2 检查只授予列表所携带的内容，因此 read-only 在 `/permission` 降级或崩溃后恢复的会话中始终保持严格零授权，而常驻 ACE 让重新升级保持零成本）。Authenticated Users 在**两种**列表中都缺席——WMI namespace 安全校验失败（0x80041003），因此 CIM 在每一种受限模式下都不可用，且 C:\-root 建树逃逸（驻留的 `AU:(AD)` + `AU:(OI)(CI)(IO)(M)` ACE）在两种模式下都被关闭；INTERACTIVE/LOCAL 同样在两种列表中都缺席（Public 树的写入被拒绝——由 runner 的 Public-probe 回归钉住）。Workspace-write 子进程看到的是私有的每会话临时子目录（`<temp>\dsh-<16 hex>`——由会话 id + 工作区派生、独占创建、拒绝 reparse point、提供方 dispose 时移除——TMP/TEMP 由 runner 重写——bwrap `--tmpfs /tmp` 语义）。受限令牌的**默认 DACL** 被扩展一条写入 SID 全权 ACE（`SetTokenInformation(TokenDefaultDacl)`）：此后不带显式安全描述符创建的新对象（匿名管道——CreatePipe、同步对象）自带 restricting SID ACE，创建时的写 pass-2 检查通过；**named pipe 例外**——其默认安全描述符是 Win32 层在用户态安装的默认 SD 模板（由 KernelBase 构建——owner/SYSTEM/Admins 全权、Everyone/ANONYMOUS 只读），令牌无法影响，因此受限孙进程的管道 stdio 捕获保持拒绝（POC 记载的边界，由 runner 套件钉住）。它以 [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md)（后端加 `./runner` argv 前缀入口）、[`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md) 的 `win32` 链档、以及作为隔离执行器的 [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) 交付；Windows 平台层在受限 pwsh 栈之上重新启用完整权限面（sandbox/sandbox-policy/permission/approval/fs-sandbox）。
+
+## How the restriction works (why no new identity)
+
+身份路线靠"**谁**在跑子进程"来限制，本档靠"令牌派生"来限制。身份路线（landstrip 的 restricted-user、AppContainer）用全新账户或容器 SID 运行子进程，该身份在宿主的文件上从零条 ACE 开始——一切访问（包括读）默认拒绝，子进程要碰的每条路径都必须事后为那个身份补写 ACE 才能放行：这正是让两个备选方案出局的全盘 DACL 改造。受限令牌保留调用者自己的 SID 与 logon session：[`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) 派生一个加入 restricting SIDs 与 `WRITE_RESTRICTED` 标志的令牌，于是 Windows 做两次访问检查——一次按正常 SID，一次按 restricting SIDs——只有两次都放行，写类访问才被授予。读只凭正常检查即可通过（调用者的 SID 在其可读范围内本来就携带读权限），所以本档不需要任何读授权、也不需要新账户；写还必须额外通过孤儿 SID 检查，而只有工作区与临时目录的 ACE 能满足它。`DISABLE_MAX_PRIVILEGE | LUA_TOKEN` 在令牌侧合成了新账户的受限用户效果，即使提升过的调用者派生的也是过滤令牌。同一原语其实也能限制读（`SidsToDisable` 把 SID 变为 deny-only），但受限读的令牌需要逐路径的读授权——恰好重新引入身份路线付出的代价——而沙盒词汇表从不要求读隔离。
+
+## Alternatives considered
+
+### 为什么不选 mxc（Microsoft xContainer）？
+
+两个否决理由。其一，OS 版本要求太新：[mxc 的 OS 版本支持文档](https://github.com/microsoft/mxc/blob/main/docs/process-container/os-version-support.md)把产品下限设在 Windows 11 24H2（build 26100），而 BaseContainer 档（T1，`Experimental_CreateProcessInSandbox`）只在 25H2+（build 26600+）且启用 OS feature 时存在——在 25H2 及以下的所有受支持版本上，文件系统策略都会回退到 T3，即 AppContainer 加宿主侧 DACL ACE 改造。其二，在任一档下支持任意路径读都意味着要为子进程可读的每个路径写 ACL 授予读权限：模型要读整个工作区和任意文件，就需要全盘改写宿主 DACL——对只做写限制的需求而言，这是不必要的驻留副作用与代价。
+
+### 为什么不选 AppContainer？
+
+AppContainer 令牌没有环境读访问：每个可读路径都必须预先通过 capability 或显式 ACE 授予，因此任意路径读——harness 的读模型——在不做同样的全盘授予时无法支持。受限令牌完全不需要读授予：它只对写访问做交集。
+
+### 为什么不选 landstrip？
+
+[landstrip 评估](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md)在实现前已被否决（未经实战检验；自建 launcher 方案胜出），且其 Windows 后端是 AppContainer 形态，继承同样的任意路径读问题。
+
+## Consequences
+
+所得：仅写隔离、不引入新的 OS 版本下限（`CreateRestrictedToken` 比 mxc 的版本早二十年）、读/网络/进程可见性完全不受影响（与模式词汇表一致）、fail-closed 错误携带 API 名与精确 Win32 错误码。所失：无读侧或网络隔离；控制台隔离不可用（隐藏控制台子进程以 `STATUS_DLL_INIT_FAILED` 死亡；子进程共享宿主控制台）；被授权根目录上有驻留 ACE 改动（目录须为调用者所有；工作区 ACE 按设计永久常驻——复用缓存，工作区改名时成为不可见残留——临时 ACE 由提供方 dispose 连同派生的私有临时目录一起回收——崩溃会把两者都留下，下一次恢复会在独占创建处大声失败，直到临时目录卫生回收该目录）；授权物化是急切的全树传播（`SetNamedSecurityInfoW` 立即遍历每个后代——在大型工作区上耗时数十秒），因按工作区身份，每台机器每个工作区只付一次；CIM 在**两种**受限模式下都不可用（AuthUsers 从两种列表中被移除——WMI namespace 安全校验失败，`Get-ComputerInfo` 静默返回不完整结果），作为关闭两种模式下 C:\-root 建树逃逸的代价；位于被授权根目录之外的 FAT 类（无 ACL）目标在两种模式下仍可写（没有可做交集的安全描述符——作为历史残留处理：不支持、仅警告、已在 README 中记录）；NULL DACL 目录在 grant+revoke 往返下不保持身份（记录在案的边角，POC 亦有此行为）；`whoami` 与令牌检查 cmdlet 在受限令牌下失败（诊断噪音，已记录）；且**两种**受限模式都以 ConstrainedLanguage 模式运行 `pwsh`——受限令牌触发 PowerShell 的锁定检测，因此 `Add-Type`、非核心 .NET 静态调用（`[System.IO.*]::`、`[math]::`）、COM 对象与反射都会以“only core types”错误失败，而 `-f` 格式化、属性访问与核心 cmdlet/类型继续工作，语言模式也无法从内部提升回 FullLanguage——已在 pwsh 工具描述中教给模型，并记录在包 README 的 Known Limitations 中；**两种**受限模式同样拒绝 named-pipe 打开——libuv 的管道 stdio spawn 以 EPERM 失败（POC 记载的“无法重定向输出”边界；继承/忽略的 stdio 与匿名管道可用）——记录在包 README 的 Known Limitations 中，并在 pwsh 工具描述中教给模型。
+
+## Testing
+
+产品可见的 Windows 阵容切换仅存在于 win32，而 keyless 快照夹具必须在 macOS/Linux 上可重放，因此无法覆盖它；替代证据是 bundle 组合 spec（[`base.spec.ts`](../../../../packages/bundle/base/tests/base.spec.ts)、[`windows-shell.spec.ts`](../../../../apps/cli/tests/windows-shell.spec.ts)）加上 win32 真实 runner 套件（`packages/sandbox/sandbox-windows-acl/tests/`、`packages/bash/pwsh-sandbox/tests/`），组装态信号由 CI 的 Windows lane 负责。授权机制在跨平台侧由 `packages/sandbox/sandbox-local/tests/acl-grants.spec.ts` 钉住（派生的私有临时身份——按会话 + 工作区确定性、跨会话相异——一次性物化、独占临时目录创建并拒绝 reparse point、失败时自我清理、干净重启时对同一派生目录的重新授权、dispose 与模式切换循环中的常驻/可回收生命周期，以及派生 SID 的 argv 契约——mock 掉 Win32 表面），win32 侧由 `workspace-sid.spec.ts`（派生的确定性/形态/相异性）、`grant.spec.ts`（真实 DACL 物化：可回收路径在 dispose 时撤销、常驻路径存活）、`acl.spec.ts` 的幂等授权快速路径与 dispose 后常驻 ACE 契约、`failure-paths.spec.ts` 的 suspension-orphan 回归（AssignProcessToJobObject 失败会终止子进程）与 `runner.spec.ts` 的 `--write-sid` 契约（调用者所有目录的授权、经 TMP/TEMP 的私有临时子目录、两种模式下的 CIM 拒绝探针、模式降级回归——驻留授权 ACE 在 read-only 下失效并在重新升级后再度生效——环境可写 Public-probe 回归（对 C:\Users\Public 子目录的写入在两种模式下都会被拒绝），以及两种模式下对 ConstrainedLanguage 的钉定，加上孙进程 stdio 矩阵钉定——继承/忽略的 stdio spawn 成功，而管道捕获在两种模式下都被**拒绝**）钉住。runner 失败分类以 127 退出码为门槛（受限命令仅仅在非 127 退出时打印 `windows-acl-run:` 签名，也绝不会被误分类为"命令未运行"——由 pwsh-sandbox helper 套件钉住）。
+
+## Related
+
+[pwsh 执行器决策](2026-08-01-pwsh-tool-and-executor.md)拥有本档所消费的 pwsh-sandbox/tool-pwsh 方言划分。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/process/2026-07-05-uniform-agent-note-format.md
-2026-07-05-uniform-agent-note-format.md: 06082251c1b96c90ed470d84224662e00e29791b
+2026-07-05-uniform-agent-note-format.md: c05d81c700b81f0d619173a261f405b1cc039df1
 2026-07-05-uniform-agent-note-format.zh.md: 3daa686b64b31ee2638b25dd4c42e5d8172f1d97
@@ -23,7 +23,7 @@ The whole corpus was normalized in the same change that defined the format — t
 - **A bare `# <title>` H1** — rejected: the `Agent Note: ` prefix self-describes the genre when a file is read outside its tree, and the format gate prevents it from drifting.
 - **`## What we give up` as the implemented closer** (the README's own phrase for what an Agent Note records) — rejected: it names only costs, and an honest consequences section records what the trade-off bought as well.
 - **Convention without a gate** (write the contract down, enforce by review) — rejected: the slop checklist already outlawed spec-speak in `implemented/` by convention, and nineteen files show what convention alone achieves here.
- **A standalone `FORMAT.md` contract file** — rejected because one front door carrying layout, classification, and format is easier to discover and maintain than two contract files.
+- **A standalone `FORMAT.md` contract file** — rejected because one entry point carrying layout, classification, and format is easier to discover and maintain than two contract files.

 ## Consequences

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/process/2026-07-19-remove-generated-agent-note-index.md
-2026-07-19-remove-generated-agent-note-index.md: ee85ec0757d5924f5784c43a50003eb96e0a9531
+2026-07-19-remove-generated-agent-note-index.md: 652ac72afe69284240667e61af1f3fbfcb182ddb
 2026-07-19-remove-generated-agent-note-index.zh.md: 6e1967fbde0c6ac59bbc3a184dad68c989efadbc
@@ -12,7 +12,7 @@ The centralized chronological list adds little discovery value beyond browsing t

 ## Decision

-The lifecycle/class filesystem tree is the Agent Note inventory. [README.md](../../README.md) remains the curated front door and contract, while ordinary tree navigation and repository search provide discovery.
+The lifecycle/class filesystem tree is the Agent Note inventory. [README.md](../../README.md) remains the curated entry point and contract, while ordinary tree navigation and repository search provide discovery.

 `scripts/agent-note-tree.ts` owns the closed lifecycle/class sets and structural walker. `verify-agent-note-classification` validates that tree and rejects the legacy homes and a root `INDEX.md`; it does not render or freshness-check a centralized list.

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/process/2026-07-22-product-first-root-readme.md
-2026-07-22-product-first-root-readme.md: 32542a45019d64ed1826d4eb21e68c67c3c3d52e
+2026-07-22-product-first-root-readme.md: 00f6084da9e83135c881abfef21b0b249cd4e30a
 2026-07-22-product-first-root-readme.zh.md: 8ef6f4b99ca2c935183a225b6357d2d128edb3b0
@@ -6,7 +6,7 @@ English | [中文](2026-07-22-product-first-root-readme.zh.md)

 ## Problem

-The root README is the repository's product front door. Its product-first structure and established voice remain useful, but concrete entry points and capability claims drift as the runtime grows. Rewriting sections whose facts remain correct increases the review surface and discards language that already works.
+The root README is the repository's product entry point. Its product-first structure and established voice remain useful, but concrete entry points and capability claims drift as the runtime grows. Rewriting sections whose facts remain correct increases the review surface and discards language that already works.

 ## Decision

@@ -26,7 +26,7 @@ Detailed package and service inventories remain at their owning documentation. T

 **Use a long marketing page with screenshots, badges, and duplicated tutorials.** Rich media can demonstrate a stable product journey, but it ages separately from commands and source contracts. The root stays compact and links to runnable examples and owned guides.

-**Project the root README as the documentation website home page.** A single landing page avoids two narratives, but the website's user guide and the repository's product/developer front door have different navigation and maintenance needs.
+**Project the root README as the documentation website home page.** A single landing page avoids two narratives, but the website's user guide and the repository's product/developer entry point have different navigation and maintenance needs.

 ## Consequences

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/process/2026-08-03-package-anchored-subsystem-pages.md
-2026-08-03-package-anchored-subsystem-pages.md: f429f3d41c1f152e83faeb12c379d221627e767f
-2026-08-03-package-anchored-subsystem-pages.zh.md: 3a56fa39357591197606a28b49cf0eac8f58963e
+2026-08-03-package-anchored-subsystem-pages.md: 2a47f35e9d3755286bed7d42f59fd21e30f9148d
+2026-08-03-package-anchored-subsystem-pages.zh.md: 53216a430b979f0612f256aaec9774b88fe14bbd
@@ -14,7 +14,7 @@ Every `docs/subsystems/` page anchors to the package or package group that decla

 Every type a generated signature references must resolve somewhere in the folder: the agent ownership vocabulary moved from the generator's `TYPE_LINK_EXEMPTIONS` into `LINK_MAP → core.md`, so exemptions are reserved for genuinely service-local or vendored shapes. Each pasted declaration has one home (`SessionEvent` lives on [session.md](../../../../docs/subsystems/session.md); core.md summarizes and links).

-Every `packages/<group>/README.md` pair is a thin front door in one shape: a why-first intro paragraph, a package table (Package / Role / ctx key), and a closing pointer to the owning subsystems page. Load-bearing prose that outgrows that shape relocates to the owning subsystems page rather than being deleted.
+Every `packages/<group>/README.md` pair is a thin entry point in one shape: a why-first intro paragraph, a package table (Package / Role / ctx key), and a closing pointer to the owning subsystems page. Load-bearing prose that outgrows that shape relocates to the owning subsystems page rather than being deleted.

 The [subsystems README](../../../../docs/subsystems/README.md) indexes every page in the folder on both language sides; `scripts/project-doc-site.spec.ts` enforces one table row per page, so a page added by a later PR (or absorbed in a merge) cannot silently miss the index.

@@ -14,7 +14,7 @@ Status: implemented

 生成签名引用的每个类型都必须能在目录中某处解析：agent 所有权词汇从生成器的 `TYPE_LINK_EXEMPTIONS` 移入 `LINK_MAP → core.md`，因此豁免只留给真正服务本地或 vendored 的形状。每个粘贴的声明只有一个家（`SessionEvent` 位于 [session.md](../../../../docs/subsystems/session.md)；core.md 概括并链接）。

-每个 `packages/<group>/README.md` 配对都是统一形状的轻薄门面：一段以「为什么」开头的介绍、一张包表格（包 / 角色 / ctx 键）、一个指向拥有方子系统页面的收尾指针。超出该形状的承重散文迁移到拥有方子系统页面，而非删除。
+每个 `packages/<group>/README.md` 配对都是统一形状的精简入口：一段以「为什么」开头的介绍、一张包表格（包 / 角色 / ctx 键）、一个指向拥有方子系统页面的收尾指针。超出该形状的承重散文迁移到拥有方子系统页面，而非删除。

 [子系统 README](../../../../docs/subsystems/README.md) 在两个语言侧索引目录中的每一页；`scripts/project-doc-site.spec.ts` 强制每页一行表格，因此后续 PR 新增（或合并吸收）的页面无法悄悄缺席索引。

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-07-23-acp-automation-only-protocol.md
-2026-07-23-acp-automation-only-protocol.md: 29da3025251f9826c5780493bb2725b114a40601
+2026-07-23-acp-automation-only-protocol.md: 56c433acf59d3a0c4b5c8b6605e422d3c4efede2
 2026-07-23-acp-automation-only-protocol.zh.md: bf326ce9f6fea8c55b842bc61e3d8521032ea4d1
@@ -46,7 +46,7 @@ Protocol and lifecycle tests pin stop-reason and prompt codecs, version negotiat

 ## Consequences

-ACP has a narrow contract suitable for agents and automation, while TUI and Web own human interaction and presentation. The package has fewer injected services, dependencies, protocol branches, and lifecycle states, and it no longer claims compatibility as a general editor front door.
+ACP has a narrow contract suitable for agents and automation, while TUI and Web own human interaction and presentation. The package has fewer injected services, dependencies, protocol branches, and lifecycle states, and it no longer claims compatibility as a general editor entry point.

 Automation clients receive complete committed text rather than token deltas or structured tool UI. They inspect durable logs or another API when they need reasoning, tool traces, titles, or richer state. Fresh-session-only operation also means callers that need durable browsing or resume use a host API rather than ACP.

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-08-04-remove-tui-package.md
-2026-08-04-remove-tui-package.md: f048a04db02b582d038ebf549999c0fda50e30d3
+2026-08-04-remove-tui-package.md: 19cc7d1a89a55bb57a69b9fce301f48f89384acd
 2026-08-04-remove-tui-package.zh.md: be18cbd33cd4a2bb592de4e7986b2786da272d0f
@@ -8,7 +8,7 @@ English | [中文](2026-08-04-remove-tui-package.zh.md)

 Removing the implicit `dsh` terminal application left `@deepseek-ai/dsh-tui` without a shipped composition. The package still carried a terminal renderer, interactive command and question adapters, extension overlays, snapshot fixtures, a patched `pi-tui` dependency, and SDK scaffolding that advertised TUI as a supported application interface. Keeping that surface required maintaining a product-sized frontend whose only remaining consumer was the project generator itself.

-The package also made the repository's supported application inventory misleading. Current runnable products use Web, ACP, JSON-RPC, or one-shot CLI front doors, while the SDK continued to offer a terminal choice that no example or product command exercised.
+The package also made the repository's supported application inventory misleading. Current runnable products use Web, ACP, JSON-RPC, or one-shot CLI entry points, while the SDK continued to offer a terminal choice that no example or product command exercised.

 ## Decision

@@ -34,6 +34,6 @@ Repository searches and generated catalogs contain no TUI package, dependency pa

 ## Consequences

-DeepSeek Harness has no terminal UI package or generated TUI application. Existing imports, `cordis.yml` rows, SDK `--interface=tui` requests, and projects that depend on the package fail instead of being translated. Web remains the shipped interactive surface; ACP, JSON-RPC, and one-shot CLI remain the non-Web front doors.
+DeepSeek Harness has no terminal UI package or generated TUI application. Existing imports, `cordis.yml` rows, SDK `--interface=tui` requests, and projects that depend on the package fail instead of being translated. Web remains the shipped interactive surface; ACP, JSON-RPC, and one-shot CLI remain the non-Web entry points.

 The provider-neutral command, user-interaction, approval, tool-presentation, PTY, and session-projection capabilities remain available to other hosts. Reintroducing a terminal frontend requires a named product or deployment, an explicit package boundary, a concrete interaction provider, and assembled lifecycle and transcript acceptance for that frontend.
@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-08-08-copy-only-preset-authoring.md
+2026-08-08-copy-only-preset-authoring.md: c16518b087c7acedbee3d89ce5cc8dbcaa0a0cde
+2026-08-08-copy-only-preset-authoring.zh.md: dc2d7924cb0fd9363efa9387bc8ba68bf11af5d7
@@ -0,0 +1,30 @@
+# Agent Note: Copy-only preset authoring, and the way into a preset's files
+
+Status: implemented
+
+English | [中文](2026-08-08-copy-only-preset-authoring.zh.md)
+
+## Problem
+
+The agent-preset settings page carried a web YAML editor: `agentPreset.write` accepted arbitrary composition text, the page held a textarea with no completion, highlighting, or diff, and the shape check leaned on the Loader's own `entryListSchema` — whose dialect includes `!!js`, so "shape-checked text" was still arbitrary code on the next mount. Weak as an editor, wide as a capability, and the source of the editor-vs-roster races the section had to defend against.
+
+## Decision
+
+Authoring is a host-side copy, and files are the editor. `agentPreset.write` became `agentPreset.copy { from, agentPreset, name? }`: two ids the host resolves against its own roots plus an optional display name, whole-directory `cp` (symlinks dereferenced, modes re-tightened to owner-only with owner-execute kept), metadata rewritten to keep the source's description but never its name or `order`. The page becomes: read-only viewer over shipped compositions, copy dialog as the only create entry (no blank "new preset" — writing YAML from nothing is not a thing people do), delete for custom rows, and a location action that leads to the files — `agentPreset.openDocument { agentPreset }` resolves the directory host-side and opens it natively, or answers `{ opened: false, path }` for the row to show as text where the deployment has no desktop (`hasDocument` on `list`, pinned by the gateway's `nativeOpen` config where `canOpenNativePath` platform detection would mislead, e.g. e2e and containers).
+
+## Consequences
+
+- No composition text and no path crosses the browser wire in either authoring direction; the `entryListSchema`/`!!js` concern dissolves with `assertComposition` itself (deleted). The privileged set is now `read`/`copy`/`openDocument`/`remove` — none accepts a filesystem target.
+- With the editor gone, hand-editing `agent.cordis.yml` is the ONLY composition edit, so the standing-mount layer grew stamp-keyed generations: `ensureStanding` compares the file's mtime+size and starts the next generation for later sessions ([standing-mounts note](../architecture/2026-08-08-per-preset-standing-mounts.md), updated in place). Without this, an edited file would serve stale compositions until process restart.
+- A copy is a full snapshot that drifts from an upgraded shipped source — accepted; the preset layer has no patch semantics (that is the bundle layer's `cordis.patch.yml`), and the shipped set itself pays the same cost (`cordis`/`code` are full copies of `standard`) for one-file readability.
+- `read` dropped `writable` (no editor to gate) and builtin directories are never opened (`openDocument` refuses non-`user` trust like `remove`): the install is overwritten by upgrades, and pointing an editor into it invites edits an upgrade silently discards.
+
+## Load-bearing details
+
+- **Copy target refusal is two checks on purpose.** The roster check refuses any id a root supplies — a user directory named like a shipped preset would be shadowed, so "create" would land a file nothing ever lists; the disk check (`PresetExistsError` before `cp` with `errorOnExist` as the race backstop) refuses a directory occupying the name without being a preset, which discovery cannot see.
+- **The revealed path is response-direction disclosure, loopback-pinned.** The invariant "no browser payload can select an arbitrary filesystem target" is about the request direction; showing the resolved directory to the loopback user is the fallback the plan requires. It never rides the unprivileged `list`.
+- **The e2e lane pins `nativeOpen: false`** (`agent-preset-authoring.overlay.yml`) — both so goldens render the same branch on macOS dev and headless Linux CI, and so test runs never pop a real file manager. The revealed directory is tokenized as `{{presetRoot}}` by the lane itself, since `normalizeAria` only knows the workspace cwd.
+
+## Alternatives considered
+
+Keeping write with a better editor (CodeMirror etc.): still arbitrary capability over the wire, still the race source, and still a worse editor than the user's own. Patch-semantics copies ("standard plus this diff"): no such layer exists below the bundle plane, and the repo's own shipped presets chose full copies deliberately. Browser-side `host.openPath` with a returned path: breaks the README's no-arbitrary-target invariant the moment the path is a request parameter.
@@ -0,0 +1,30 @@
+# Agent Note: 仅复制的 preset 创作，与通往 preset 文件的入口
+
+Status: implemented
+
+[English](2026-08-08-copy-only-preset-authoring.md) | 中文
+
+## Problem
+
+agent-preset 设置页带着一个网页 YAML 编辑器：`agentPreset.write` 接收任意组装文本，页面是一个没有补全、高亮或 diff 的文本域，形状检查依赖 Loader 自己的 `entryListSchema`——其方言含 `!!js`，所以「过了形状检查的文本」在下一次挂载时仍是任意代码。作为编辑器很弱，作为能力很宽，还是该分区不得不防御的「编辑器 vs 名单」竞态的来源。
+
+## Decision
+
+创作改为宿主端复制，文件就是编辑器。`agentPreset.write` 变为 `agentPreset.copy { from, agentPreset, name? }`：两个由宿主对照自身根目录解析的 id 加一个可选显示名，整目录 `cp`（符号链接解引用，权限收紧为仅属主并保留属主执行位），元数据重写为保留来源描述、但绝不保留其名称与 `order`。页面变为：随附组装的只读查看器、作为唯一创建入口的复制对话框（不再有空白「新建预设」——从零手写 YAML 不是人会做的事）、自定义行的删除，以及通向文件的位置操作——`agentPreset.openDocument { agentPreset }` 在宿主端解析目录并原生打开，部署没有桌面时回答 `{ opened: false, path }` 供卡片以文本展示（`list` 上的 `hasDocument`；在 `canOpenNativePath` 平台探测会失真处由网关的 `nativeOpen` 配置钉死，例如 e2e 与容器）。
+
+## Consequences
+
+- 创作两个方向都不再有组装文本或路径跨越浏览器传输层；`entryListSchema`/`!!js` 的顾虑随 `assertComposition` 本身（已删除）一并消解。特权集现为 `read`/`copy`/`openDocument`/`remove`——没有一个接收文件系统目标。
+- 编辑器移除后，手改 `agent.cordis.yml` 成为**唯一**的组装编辑方式，因此常驻挂载层增加了以 stamp 为键的代际：`ensureStanding` 比对文件的 mtime+大小，为后续会话开启下一代际（[常驻挂载 note](../architecture/2026-08-08-per-preset-standing-mounts.md)，已就地更新）。没有它，改过的文件要等进程重启才生效。
+- 副本是完整快照，会随随附来源升级而漂移——接受；preset 层没有 patch 语义（那是 bundle 层 `cordis.patch.yml` 的能力），随附集合自己也为「一个文件读完整份组装」付了同样的代价（`cordis`/`code` 就是 `standard` 的完整副本）。
+- `read` 去掉了 `writable`（没有编辑器可门控），内置目录绝不被打开（`openDocument` 与 `remove` 一样拒绝非 `user` 信任）：安装目录会被升级覆盖，把编辑器指向它等于招揽会被升级悄悄丢弃的编辑。
+
+## Load-bearing details
+
+- **复制目标的拒绝刻意分两道检查。** roster 检查拒绝任一根目录提供的 id——与随附 preset 同名的用户目录会被遮蔽，「创建」只会落下一个永远不被列出的文件；磁盘检查（`cp` 之前的 `PresetExistsError`，`errorOnExist` 作竞态兜底）拒绝占着名字却不是 preset 的目录，那是 discovery 看不见的。
+- **展示的路径是响应方向的披露，且钉在环回。**「没有任何浏览器载荷能选中任意文件系统目标」这条不变量说的是请求方向；把解析出的目录展示给环回用户正是方案要求的降级。它绝不搭乘非特权的 `list`。
+- **e2e lane 钉死 `nativeOpen: false`**（`agent-preset-authoring.overlay.yml`）——既让 golden 在 macOS 开发机与无头 Linux CI 上渲染同一分支，也让测试运行永不弹出真实文件管理器。揭示的目录由 lane 自己 token 化为 `{{presetRoot}}`，因为 `normalizeAria` 只认识 workspace cwd。
+
+## Alternatives considered
+
+保留 write 换个更好的编辑器（CodeMirror 等）：传输层上仍是任意能力，仍是竞态来源，而且仍不如用户自己的编辑器。带 patch 语义的副本（「standard 加这点 diff」）：bundle 面之下没有这样的层，仓库自己的随附 preset 也刻意选了完整副本。浏览器端拿返回路径调 `host.openPath`：路径一旦成为请求参数，就打破了 README 的「不可选中任意目标」不变量。
@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-08-08-remove-cli-demo.md
-2026-08-08-remove-cli-demo.md: c1153f5e9fcc89585e926f8f088829c849d1f6c1
+2026-08-08-remove-cli-demo.md: 403e01f94c976d2d17eb391830721b31675cd6a9
 2026-08-08-remove-cli-demo.zh.md: 7f11e0c17a15454b99b32d14ea6eda177f4b01f6
@@ -6,7 +6,7 @@ English | [中文](2026-08-08-remove-cli-demo.zh.md)

 ## Problem

-After [`dsh run`](../feature/2026-08-08-dsh-run-headless-command.md) became the product one-shot command, `@deepseek-ai/dsh-cli-demo` remained a second application package for the same job. It carried another executable, argument grammar, app composition, cancellation lifecycle, text/JSON/stream-JSON output contract, built artifact, documentation surface, and test suite. The two front doors also assembled different trees, so a successful demo did not prove the shipped `headless` profile and users had to choose between overlapping commands.
+After [`dsh run`](../feature/2026-08-08-dsh-run-headless-command.md) became the product one-shot command, `@deepseek-ai/dsh-cli-demo` remained a second application package for the same job. It carried another executable, argument grammar, app composition, cancellation lifecycle, text/JSON/stream-JSON output contract, built artifact, documentation surface, and test suite. The two entry points also assembled different trees, so a successful demo did not prove the shipped `headless` profile and users had to choose between overlapping commands.

 The replay suites still need canonical session events to pin assembled backend behavior. That testing need does not require a published command or compatibility contract.

@@ -1,6 +0,0 @@
-# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
-# side as of the last confirmed-consistent state. Both languages carry equal authority;
-# after editing either side, bring the other along and re-record with:
-#   pnpm run verify-translation-pairing --write .agents/notes/proposed/feature/2026-08-01-windows-pwsh-default.md
-2026-08-01-windows-pwsh-default.md: 64d713aecda60d8747199aff5334a80bf1867d92
-2026-08-01-windows-pwsh-default.zh.md: 74ba6c43dea984384a9b7dd2675b681ee025d6ca
--- a/Show More
+++ b/Show More