Merge remote-tracking branch 'origin/master' into worktree/default-model-persistence

# Conflicts: # apps/web/tests/models-settings.e2e.ts # packages/client/ui-models/README.i18n.yaml # packages/client/ui-models/README.md # packages/client/ui-models/README.zh.md # packages/client/ui-models/src/client/ModelsSection.module.css # packages/client/ui-models/src/client/ModelsSection.tsx
2026-08-07 13:54:15 +08:00
parent 9679597204 b767ad02e0
commit d63c70bcf1
@@ -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/2026-07-30-web-config-plane.md
-2026-07-30-web-config-plane.md: 5225460be1d66b85a05ff2fd5ae2826b0e6c41d7
-2026-07-30-web-config-plane.zh.md: 53a21ddf31640d963c413e1793276de694547311
+2026-07-30-web-config-plane.md: 11554077d1848dcdf59b896dd9c29a39fd2f55d4
+2026-07-30-web-config-plane.zh.md: 527c2de8155a56789358b801f9c374e16c81931b
@@ -22,7 +22,7 @@ PR1 made LLM adapter configuration restart-free at the seam, but the only writer

 **A hand-written editor over a schema model layer.** `dsh-client-schema-form` rehydrates the wire's `toJSON()` envelope into live schemastery nodes for validation, path resolution, and immutable draft editing — but no generic rendering: the first cut shipped a full schema-driven form renderer, and the resulting page was an unstyled schema dump (every advanced field flattened onto the card, raw field names as labels, the `retryPolicy` unsupported-fallback in the main flow). The user chose the hand-written direction over adding a hint/grouping system, and a second round removed the reference input entirely: the card's primary field is one **API key** input, a whole-section provider without a configured key opens as its setup card, and the collapsed 自定义设置 fold carries the curated per-family extras (`baseURL` for both families, `reasoningEffort` for deepseek / `reasoning` for pi-ai, plus direct DeepSeek model rows with `id`, `name`, and `contextWindow`). Existing model fields outside that visible set survive array edits; retry policy, timeouts, and other fields remain owned by `settings.yaml`. Validation still runs the rehydrated schema before writing, while adapter-specific checks reject catalog invariants that the serialized schema cannot express. The card's colors resolve through the `--dsw-alias-*` design tokens; it had named `--border`/`--surface`/`--text-*`, which nothing in this app defines, so it rendered their light-mode fallbacks and stayed light under the dark theme. The model catalog takes the row shape the pi-ai provider form introduces: one bordered entry per model, id and display name on the row, and the capacities behind the row's own disclosure, so the two editors read as one design rather than diverging once both land. Every field keeps the indexed `aria-label` that names it. Both capacities are text fields reading a decimal `K`/`M` suffix (`1M` is 1000K, matching how capacities are quoted) and storing the plain count: a field holds the typed text while it has focus, because re-deriving it from the parsed count on every keystroke would rewrite `1000` to `1K` mid-word, and text that does not parse stays on screen so the save-time rejection names a row the user can still see. The shared class names carry this file's token spellings, not that branch's: `--dsw-alias-border-subtle`, `--dsw-alias-text-tertiary`, and `--dsw-alias-text-primary` are undeclared, so they resolve to the light-mode literals in their fallback slots — the defect this section was moved off. A styles test now rejects any `--dsw-*` name the token sheet does not declare, so the next editor to name one fails rather than shipping a light-only surface.

-**The Models page is a three-domain join with seam-shaped apply semantics.** Rows are configured providers; the add card's select is the dormant directory remainder. Route liveness still gates readiness and invalidates the join, but the page does not render it as provider status because configuration presence and runtime availability are distinct. The key path stays reference-shaped without ever showing a reference: a typed key stores **write-only** through `credentials.set` under the profile's `apiKeyEnv`, deriving `<ROUTE>_API_KEY` when none exists (the pi-ai profile records the derivation), so `settings.yaml` never carries a key value. Profile edits and removals land as minimal path-addressed `settings.mutate` operations against the redacted user section, which never names a secret the page did not receive. Removing a user-layer provider first opens a localized model-provider confirmation dialog; cancellation, its close button, and its mask leave the profile untouched, while the destructive confirmation submits the single unset and blocks duplicate submission until it settles. DeepSeek's model list is array-replace configuration: inherited effective rows remain visible until the first edit materializes the complete list in the user layer, and reset unsets the list override.
+**The Models page is a three-domain join with seam-shaped apply semantics.** Rows are configured providers; the add card's select is the dormant directory remainder. Route liveness still gates readiness and invalidates the join, but the page does not render it as provider status because configuration presence and runtime availability are distinct. The key path stays reference-shaped without ever showing a reference: a typed key stores **write-only** through `credentials.set` under the profile's `apiKeyEnv`, deriving `<ROUTE>_API_KEY` when none exists (the pi-ai profile records the derivation only when a key is entered), so `settings.yaml` never carries a key value; a blank pi-ai key materializes a reference-free profile and preserves provider-native authentication. Profile edits and removals land as minimal path-addressed `settings.mutate` operations against the redacted user section, which never names a secret the page did not receive. Removing a user-layer provider first opens a localized confirmation dialog whose row actions, title, description, and final action identify the same provider; confirmation removes an exact configured+writable derived credential before the profile, while custom, environment, and unidentified targets remain untouched. Both stages are idempotent and a partial failure stays in the dialog for retry. DeepSeek's model list is array-replace configuration: inherited effective rows remain visible until the first edit materializes the complete list in the user layer, and reset unsets the list override. The partial-commit and credential-ownership rationale lives in the [provider credential lifecycle note](../bug-fix/2026-08-06-provider-credential-lifecycle.md).

 ## Alternatives considered

@@ -36,4 +36,4 @@ PR1 made LLM adapter configuration restart-free at the seam, but the only writer

 ## Consequences

-The whole loop is pinned keyless in the browser lane (`apps/web/tests/models-settings.e2e.ts`): the add card offers the dormant pi-ai catalog, adding `minimax-cn` with a typed key writes the reference-only profile into `settings.yaml`, stores the value into the harness home's `.env` under the derived `MINIMAX_CN_API_KEY`, registers the route live on the topology frame, and the customized fold merges `reasoning` beside the reference — zero model calls, ARIA goldens for the add-card, configured, and delete-confirmation states, plus a scaffold `harnessHome` so tests never touch a real `~/.dsh` (the provider under test is one whose derived reference cannot collide with a developer's exported keys). The settings-shell scenario intercepts the pathless native intent; seam, provider, wire, React, and native-opener tests separately pin provider absence, custom-path resolution, absent-file materialization, owner-only permissions, hidden remote/unavailable states, duplicate-click collapse, localized failure, macOS text-editor dispatch, and Linux/Windows desktop dispatch. The removal scenario proves cancellation leaves the profile intact, confirmation removes it, and the intentionally retained credential survives. The DeepSeek onboarding fixture edits the default catalog into a user-owned list, persists an arbitrary model id/name/context window, removes the active row, and observes the model selector's empty-selection fallback. The rename touched 239 files (fixtures, goldens, docs, python) in one commit with no compatibility alias. The renderer replacement cost one commit and no wire change: apply semantics, redaction, and the directory join were renderer-agnostic all along. Deferred: a per-row models preview (the picker already lists models), a page address for live routes that never declared configurability, and explicit removal of a provider's retained credential.
+The whole loop is pinned keyless in the browser lane (`apps/web/tests/models-settings.e2e.ts`): the add card offers the dormant pi-ai catalog, adding `minimax-cn` with a typed key writes the reference-only profile into `settings.yaml`, stores the value into the harness home's `.env` under the derived `MINIMAX_CN_API_KEY`, registers the route live on the topology frame, and the customized fold merges `reasoning` beside the reference — zero model calls, ARIA goldens for the add-card, configured, and identified delete-confirmation states, plus a scaffold `harnessHome` so tests never touch a real `~/.dsh` (the provider under test is one whose derived reference cannot collide with a developer's exported keys). The settings-shell scenario intercepts the pathless native intent; seam, provider, wire, React, and native-opener tests separately pin provider absence, custom-path resolution, absent-file materialization, owner-only permissions, hidden remote/unavailable states, duplicate-click collapse, localized failure, macOS text-editor dispatch, and Linux/Windows desktop dispatch. The removal scenario proves cancellation leaves both profile and key intact, then confirmation removes both the profile and its identified managed credential. The DeepSeek onboarding fixture edits the default catalog into a user-owned list, persists an arbitrary model id/name/context window, removes the active row, and observes the model selector's empty-selection fallback. The rename touched 239 files (fixtures, goldens, docs, python) in one commit with no compatibility alias. The renderer replacement cost one commit and no wire change: apply semantics, redaction, and the directory join were renderer-agnostic all along. Deferred: a per-row models preview (the picker already lists models) and a page address for live routes that never declared configurability.
@@ -22,7 +22,7 @@ PR1 让 LLM（大语言模型）适配器配置在 seam 层面免重启，但唯

 **架在 schema 模型层之上的手写编辑器。**`dsh-client-schema-form` 把 wire 的 `toJSON()` 信封还原（rehydrate）为活的 schemastery 节点，用于校验、路径解析与不可变草稿编辑——但不做通用渲染：第一版交付了完整的 schema 驱动表单渲染器，得到的却是一个未加样式、把 schema 原样倾倒出来的页面（每个进阶字段都平铺到卡片上、原始字段名直接充当标签、`retryPolicy` 的「不支持」回退落在主流程里）。用户没有再加一套提示／分组系统，而是选择了手写方向，第二轮又把引用输入框整个移除：卡片的主字段是一个 **API 密钥**输入框，未配置密钥的整分节提供方会以其设置卡片的形式打开，收起的「自定义设置」折叠区承载按家族精选的额外字段（两个家族都有 `baseURL`，deepseek 有 `reasoningEffort`／pi-ai 有 `reasoning`，另有直接 DeepSeek 模型行的 `id`、`name` 和 `contextWindow`）。现有模型字段中不在可见集合内的部分会在数组编辑后保留；重试策略、超时及其他字段仍归 `settings.yaml` 所有。校验仍会在写入前运行还原出的 schema，适配器特有的检查则会拒绝序列化 schema 无法表达的目录不变量。卡片的颜色经 `--dsw-alias-*` 设计 token 解析；它此前引用的 `--border`／`--surface`／`--text-*` 在本应用中无人定义，于是渲染出的是它们的亮色模式回退值，在暗色主题下依旧保持亮色。模型目录采用 pi-ai 提供方表单引入的行形态：每个模型一个带边框的条目，ID 与显示名称落在行上，容量则收在该行自己的折叠区里，使两个编辑器呈现为同一套设计，而不是在双方都落地后各自分岔。每个字段都保留那个为其命名的带序号 `aria-label`。两项容量都是文本输入框，读取十进制的 `K`／`M` 后缀（`1M` 即 1000K，与容量的通行标注方式一致）并存储纯数值：字段持有焦点期间保留键入的文本，因为若每次按键都从解析出的数值重新推导该文本，`1000` 会在尚未输完时就被改写成 `1K`；无法解析的文本也会留在屏幕上，因此保存时的拒绝点名的是用户仍能看见的那一行。共用的类名承载的是本文件的 token 写法，而非那个分支的：`--dsw-alias-border-subtle`、`--dsw-alias-text-tertiary` 和 `--dsw-alias-text-primary` 均未声明，于是它们解析为各自回退槽位中的亮色模式字面值——正是本节此前迁离的那个缺陷。现在有一个样式测试会拒绝 token 表未声明的任何 `--dsw-*` 名称，因此下一个写出这类名称的编辑者会当场失败，而不是交付一个只有亮色的界面。

-**Models 页是一次三领域联接，应用语义与 seam 同形。**每一行是一个已配置的提供方；「新增」卡片的选择框是可配置提供方目录中剩余的休眠条目。路由存活状态仍用于就绪判定，并会使该联接失效，但页面不将其渲染为提供方状态，因为配置存在与运行时可用性是两个不同概念。密钥通道保持引用形态，却从不展示任何引用：键入的密钥经 `credentials.set` **只写**存入 profile 的 `apiKeyEnv` 之下，引用不存在时便派生 `<ROUTE>_API_KEY`（pi-ai profile 会记录该派生），因此 `settings.yaml` 从不携带密钥值。profile 的编辑和删除会针对脱敏后的用户分节，以按路径寻址的最小 `settings.mutate` 操作落地，绝不会点名页面未收到的机密。删除用户层提供方时，会先打开本地化的模型提供方确认对话框；取消操作、关闭按钮和遮罩均不会改动 profile，而破坏性确认会提交唯一一条 unset，并在其完成前阻止重复提交。DeepSeek 的模型列表是数组替换配置：继承而来的生效模型行会一直显示，直到第一次编辑将完整列表具化到用户层；重置则会取消设置该列表覆盖。
+**Models 页是一次三领域联接，应用语义与 seam 同形。**每一行是一个已配置的提供方；「新增」卡片的选择框是可配置提供方目录中剩余的休眠条目。路由存活状态仍用于就绪判定，并会使该联接失效，但页面不将其渲染为提供方状态，因为配置存在与运行时可用性是两个不同概念。密钥通道保持引用形态，却从不展示任何引用：键入的密钥经 `credentials.set` **只写**存入 profile 的 `apiKeyEnv` 之下，引用不存在时便派生 `<ROUTE>_API_KEY`（仅在输入密钥时，pi-ai profile 才会记录该派生），因此 `settings.yaml` 从不携带密钥值；留空 pi-ai 密钥会具化一个不带引用的 profile，并保留提供方原生认证。profile 的编辑和删除会针对脱敏后的用户分节，以按路径寻址的最小 `settings.mutate` 操作落地，绝不会点名页面未收到的机密。删除用户层提供方时，会先打开本地化确认对话框，其行操作、标题、说明和最终操作都会点名同一个提供方；确认后会先清除与派生目标精确匹配且已配置、可写的凭据，再删除 profile，自定义目标、环境目标和无法识别的目标则保持不变。两个阶段都具备幂等性，部分失败会留在对话框中供重试。DeepSeek 的模型列表是数组替换配置：继承而来的生效模型行会一直显示，直到第一次编辑将完整列表具化到用户层；重置则会取消设置该列表覆盖。部分提交与凭据所有权的理由记录在[提供方凭据生命周期 note](../bug-fix/2026-08-06-provider-credential-lifecycle.md)中。

 ## 曾考虑的替代方案

@@ -36,4 +36,4 @@ PR1 让 LLM（大语言模型）适配器配置在 seam 层面免重启，但唯

 ## 后果

-整条闭环以无密钥方式固定在浏览器测试通道（`apps/web/tests/models-settings.e2e.ts`）：「新增」卡片提供休眠的 pi-ai catalog，携键入的密钥添加 `minimax-cn` 会把只含引用的 profile 写入 `settings.yaml`、把密钥值存入 harness 家目录 `.env` 中派生的 `MINIMAX_CN_API_KEY` 之下、路由随拓扑帧注册为存活，「自定义设置」折叠区则把 `reasoning` 合并到引用旁边——全程零模型调用，「新增」卡片态、已配置态与删除确认态各有 ARIA golden，另有脚手架式的 `harnessHome`，测试绝不触碰真实的 `~/.dsh`（受测提供方是派生引用不可能与开发者已导出密钥相撞的那一个）。设置外壳场景会截获无路径参数的原生意图；seam、提供方、wire、React 与原生打开器测试分别固定了提供方缺失、自定义路径解析、缺失文件创建、仅属主权限、远程／不可用时隐藏、重复点击合并、本地化失败、macOS 文本编辑器分发，以及 Linux/Windows 桌面分发。删除场景证明：取消后 profile 保持原样，确认后会将其删除，而刻意保留的凭据依然存在。DeepSeek 首次使用 fixture 会把默认目录编辑为用户自有列表、持久化任意模型的 ID／名称／上下文窗口、移除活动模型行，并观察模型选择器的空选择回退。这次重命名在一次提交中触及 239 个文件（fixture（测试前置数据）、golden、文档、python），未保留兼容别名。替换渲染器只花了一次提交，且没有任何 wire 变更：应用语义、脱敏与目录联接从一开始就与渲染器无关。延后事项：每行的模型预览（选择器已能列出模型）、为从未声明可配置性的存活路由提供页面地址，以及显式删除提供方所保留的凭据。
+整条闭环以无密钥方式固定在浏览器测试通道（`apps/web/tests/models-settings.e2e.ts`）：「新增」卡片提供休眠的 pi-ai catalog，携键入的密钥添加 `minimax-cn` 会把只含引用的 profile 写入 `settings.yaml`、把密钥值存入 harness 家目录 `.env` 中派生的 `MINIMAX_CN_API_KEY` 之下、路由随拓扑帧注册为存活，「自定义设置」折叠区则把 `reasoning` 合并到引用旁边——全程零模型调用，「新增」卡片态、已配置态与已点名目标的删除确认态各有 ARIA golden，另有脚手架式的 `harnessHome`，测试绝不触碰真实的 `~/.dsh`（受测提供方是派生引用不可能与开发者已导出密钥相撞的那一个）。设置外壳场景会截获无路径参数的原生意图；seam、提供方、wire、React 与原生打开器测试分别固定了提供方缺失、自定义路径解析、缺失文件创建、仅属主权限、远程／不可用时隐藏、重复点击合并、本地化失败、macOS 文本编辑器分发，以及 Linux/Windows 桌面分发。删除场景证明，取消会保留 profile 和密钥，随后的确认会同时删除 profile 及其已识别的受管凭据。DeepSeek 首次使用 fixture 会把默认目录编辑为用户自有列表、持久化任意模型的 ID／名称／上下文窗口、移除活动模型行，并观察模型选择器的空选择回退。这次重命名在一次提交中触及 239 个文件（fixture（测试前置数据）、golden、文档、python），未保留兼容别名。替换渲染器只花了一次提交，且没有任何 wire 变更：应用语义、脱敏与目录联接从一开始就与渲染器无关。延后事项：每行的模型预览（选择器已能列出模型）和为从未声明可配置性的存活路由提供页面地址。
@@ -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/2026-07-31-code-runtime-portable-identifier-seam.md
+2026-07-31-code-runtime-portable-identifier-seam.md: 9e9df50235b3505458e3645e2c6ff6e9bd439183
+2026-07-31-code-runtime-portable-identifier-seam.zh.md: 31d2410ee4809d0693f2e7897e61c50163bb0758
@@ -0,0 +1,44 @@
+# Agent Note: the code-runtime seam owns portable-identifier exclusions
+
+Status: implemented
+
+English | [中文](2026-07-31-code-runtime-portable-identifier-seam.zh.md)
+
+## Problem
+
+The code-runtime seam promises that a binding-namespace list valid on one backend is valid on every backend, so a Code Mode consumer can hand the same bindings to any registered runtime without knowing its language. The first backend, `dsh-code-runtime-worker`, privately owned the identifier rules that enforce part of that promise: an `IDENTIFIER` regex that allowed the JS-only `$`, a `RESERVED_WORDS` set holding only ECMAScript keywords, and a `RESERVED_ERROR_PROPERTIES` set of three JS `Error` slots. Those rules described the worker's own language, not the seam's portability contract.
+
+A second backend written against a different language (CPython, arriving in a later PR of this stack) would either re-declare its own rules — letting `lambda` pass the worker and fail Python, or `$tools` pass the worker and fail every non-JS backend — or import the worker's, inverting the dependency so the seam's implementation reached into a sibling implementation. Neither keeps the portability promise real: it would hold only for the backend a caller happened to test against.
+
+## Decision
+
+The seam package (`@deepseek-ai/dsh-code-runtime`) exports the portable-identifier exclusion contract as four named constants, and every backend imports them rather than re-declaring:
+
+- `PORTABLE_RESERVED_WORDS` — the union of ECMAScript and Python reserved words. A namespace global or error-class name matching any is refused on all backends, so `lambda` is refused even though it is a legal JS parameter name. Adding a language widens this union, which is a deliberate breaking review of existing binding names.
+- `RESERVED_BINDING_GLOBALS` — globals some backend owns in the program's namespace: `console` (the worker's log capture), `__dsh_main__`/`__builtins__`/`__name__` (the Python bootstrap's wrapper and seeded module globals), and `__debug__` (not a seeded slot but a CPython compile-time constant that rejects assignment, so an injected global under that name is unreachable — the same portability split by a different mechanism). Refused everywhere so a namespace list cannot pick a name that works on one backend and collides on another.
+- `RESERVED_ERROR_MEMBERS` — error-member names every backend refuses: the JS `Error` slots (`name`, `message`, `stack`) and Python's exception-protocol members (`args`, `with_traceback`, `add_note`).
+- `DUNDER_MEMBER` — the dunder-form regex (`__x__`, non-empty middle), refused as an error member wholesale because several are constrained CPython descriptors whose exact set is an interpreter-version detail.
+
+The seam also narrows the portable identifier subset to `[A-Za-z_][A-Za-z0-9_]*` (documented on `CodeBindingNamespace.global` and `CodeBindingErrorClass`), dropping the JS-only `$`. The worker consumes the shared constants directly by their seam names — `PORTABLE_RESERVED_WORDS` for both binding-global and error-class names, `RESERVED_BINDING_GLOBALS` for backend-owned slots, `RESERVED_ERROR_MEMBERS` plus `DUNDER_MEMBER` for error members — with no local re-alias; its `IDENTIFIER` regex loses `$`.
+
+The constants live at the seam even though only one backend ships in this PR: the whole point is that the contract is language-agnostic and owned above any single language. A backend that violated it would be the bug, and the shared set is where a reviewer looks to see what "portable" means.
+
+## Scope
+
+This PR delivers only the seam extension and the worker's adoption of it. No Python backend, `py-types` renderer, or Code Mode language dispatch ships here — they are later PRs in the stack that depend on these exports. The seam README's worker-only wording is left unchanged for the same reason: linking to a `dsh-code-runtime-python` README that does not yet exist would break the dead-link gate.
+
+`RESERVED_BINDING_GLOBALS` currently encodes the not-yet-merged Python bootstrap's concrete design: it seeds exactly `__builtins__`/`__name__` and wraps the program under `__dsh_main__`. The Python-backend PR that seeds any additional module global (`__doc__`, `__loader__`, `__spec__`, `__file__`, `__package__`, …) MUST widen this set in the same change, exactly as adding a language widens `PORTABLE_RESERVED_WORDS` — a name the bootstrap seeds but the set omits is the portability split this contract exists to prevent.
+
+## Alternatives considered
+
+**Each backend declares its own exclusions.** Rejected: it makes the portability promise per-backend. A binding list the caller tested on the worker could be refused by Python, which is exactly the split the seam exists to prevent.
+
+**The Python backend imports the worker's constants.** Rejected: it inverts the dependency — the seam's implementations would reach into a sibling implementation for a contract neither owns. The contract belongs above both, at the seam.
+
+**Keep `$` in the portable identifier subset.** Rejected: `$` is JS-only spelling. Allowing it would let `$tools` pass the worker and fail every non-JS backend, breaking portability for a purely cosmetic gain.
+
+## Consequences
+
+Bought: one place — the seam package — defines what a portable binding name is, and every backend enforces the same contract by import. A namespace list valid on one backend is valid on all, verifiably, not by coincidence of which backend the caller tested.
+
+Cost: existing worker callers using a `$`-containing global now fail identifier validation. Under the pre-release stance this is a corrected foundation, not a compatibility break to shim. The worker's seam-misuse tests gain cases for `$tools`, Python exception members (`args`), dunders (`__dict__`), and a Python-owned global (`__dsh_main__`), proving the shared set is enforced from the worker side.
@@ -0,0 +1,44 @@
+# Agent Note：code-runtime seam 拥有可移植标识符排除集
+
+Status: implemented
+
+[English](2026-07-31-code-runtime-portable-identifier-seam.md) | 中文
+
+## Problem
+
+code-runtime seam 承诺：在一个后端上有效的绑定命名空间列表，在每个后端上都有效，因此 Code Mode 消费方可以把同一组绑定交给任何已注册的运行时，而不必知道它的语言。首个后端 `dsh-code-runtime-worker` 私自拥有了执行这项承诺一部分的标识符规则：一个允许 JS 专有 `$` 的 `IDENTIFIER` 正则、一个只含 ECMAScript 关键字的 `RESERVED_WORDS` 集合，以及一个含三个 JS `Error` 槽位的 `RESERVED_ERROR_PROPERTIES` 集合。这些规则描述的是 worker 自身的语言，而非 seam 的可移植性契约。
+
+一个针对不同语言（CPython，将在本 stack 后续 PR 中到来）编写的第二后端，要么重新声明自己的规则——让 `lambda` 通过 worker 却在 Python 上失败，或让 `$tools` 通过 worker 却在每个非 JS 后端上失败——要么导入 worker 的规则，从而反转依赖，使 seam 的一个实现伸手进入另一个兄弟实现。二者都无法让可移植承诺成真：它只对调用方恰好测试过的那个后端成立。
+
+## Decision
+
+seam 包（`@deepseek-ai/dsh-code-runtime`）以四个具名常量导出可移植标识符排除契约，每个后端导入它们而非重新声明：
+
+- `PORTABLE_RESERVED_WORDS`——ECMAScript 与 Python 保留字的联集。任何命名空间 global 或 error-class 名称匹配其中之一，都在所有后端上被拒绝，因此 `lambda` 即便是合法的 JS 参数名也被拒绝。新增一门语言即扩宽此联集，这是对现有绑定名称的一次有意的破坏性复审。
+- `RESERVED_BINDING_GLOBALS`——某个后端在程序命名空间中拥有的 global：`console`（worker 的日志捕获）、`__dsh_main__`/`__builtins__`/`__name__`（Python bootstrap 的包装器与预置模块 global），以及 `__debug__`（不是 seed 的槽位，而是 CPython 编译期常量，赋值会被拒，故以该名注入的 global 不可达——同一种可移植性分裂，只是机制不同）。在所有后端上被拒绝，使命名空间列表无法选到一个在某后端能用、在另一后端冲突的名称。
+- `RESERVED_ERROR_MEMBERS`——每个后端都拒绝的 error-member 名称：JS `Error` 槽位（`name`、`message`、`stack`）与 Python 异常协议成员（`args`、`with_traceback`、`add_note`）。
+- `DUNDER_MEMBER`——dunder 形式正则（`__x__`，非空中缀），作为 error member 被整体拒绝，因为其中若干是受约束的 CPython 描述符，其确切集合是解释器版本细节。
+
+seam 同时把可移植标识符子集收窄为 `[A-Za-z_][A-Za-z0-9_]*`（记录在 `CodeBindingNamespace.global` 与 `CodeBindingErrorClass` 上），去掉 JS 专有的 `$`。worker 直接以 seam 名消费这些共享常量——binding-global 与 error-class 名称用 `PORTABLE_RESERVED_WORDS`、后端拥有槽位用 `RESERVED_BINDING_GLOBALS`、error member 用 `RESERVED_ERROR_MEMBERS` 加 `DUNDER_MEMBER`——不再本地起别名；其 `IDENTIFIER` 正则去掉 `$`。
+
+尽管本 PR 只交付一个后端，这些常量仍置于 seam：要点正是该契约与语言无关，且拥有权在任何单一语言之上。违反它的后端才是 bug，而共享集合正是复审者查看"可移植"含义的地方。
+
+## Scope
+
+本 PR 只交付 seam 扩展与 worker 对它的采用。这里不交付任何 Python 后端、`py-types` 渲染器或 Code Mode 的语言分发——它们是本 stack 中依赖这些导出的后续 PR。seam README 中仅描述 worker 的措辞保持不变，理由相同：链接到一个尚不存在的 `dsh-code-runtime-python` README 会破坏死链 gate。
+
+`RESERVED_BINDING_GLOBALS` 当前编码了尚未合并的 Python bootstrap 的具体设计：它恰好 seed `__builtins__`/`__name__`，并把程序包装在 `__dsh_main__` 之下。任何 seed 额外模块 global（`__doc__`、`__loader__`、`__spec__`、`__file__`、`__package__` 等）的 Python 后端 PR 必须在同一改动中扩宽此集合，正如新增一门语言即扩宽 `PORTABLE_RESERVED_WORDS`——bootstrap 会 seed 却不在集合中的名称，正是本契约要防止的可移植性分裂。
+
+## Alternatives considered
+
+**每个后端声明自己的排除集。** 拒绝：这让可移植承诺变成逐后端成立。调用方在 worker 上测过的绑定列表可能被 Python 拒绝，而这正是 seam 存在要防止的分裂。
+
+**Python 后端导入 worker 的常量。** 拒绝：这反转依赖——seam 的实现会为一个二者都不拥有的契约伸手进入兄弟实现。契约属于二者之上，即 seam。
+
+**在可移植标识符子集中保留 `$`。** 拒绝：`$` 是 JS 专有拼写。允许它会让 `$tools` 通过 worker 却在每个非 JS 后端上失败，为纯粹表面的好处破坏可移植性。
+
+## Consequences
+
+获得：一个地方——seam 包——定义什么是可移植绑定名称，每个后端通过导入执行同一契约。在一个后端上有效的命名空间列表在所有后端上都有效，这是可验证的，而非取决于调用方测试了哪个后端的巧合。
+
+代价：现有使用含 `$` global 的 worker 调用方现在会在标识符校验时失败。在预发布立场下这是一次被纠正的地基，而非需要 shim 的兼容性破坏。worker 的 seam-misuse 测试新增了 `$tools`、Python 异常成员（`args`）、dunder（`__dict__`）与一个 Python 拥有的 global（`__dsh_main__`）等用例，从 worker 侧证明共享集合被执行。
@@ -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/bug-fix/2026-07-29-sticky-composer-conversation-scroll.md
-2026-07-29-sticky-composer-conversation-scroll.md: d3fed7a9d0b1f39f9551fbd85e0f83515b1a2690
-2026-07-29-sticky-composer-conversation-scroll.zh.md: 2beee34d3bb68832d14b7607b43aa11e1425d53d
+2026-07-29-sticky-composer-conversation-scroll.md: 8657080078e475d6e90f17a6c0f8cb5b2fb1555d
+2026-07-29-sticky-composer-conversation-scroll.zh.md: a588903adf51c6fcdd01f851d4f5e6570f695242
@@ -14,7 +14,7 @@ The active conversation column split scrolling: the chat (and trajectory) view o

 Session stats live on `'conversation.composer.dock'` (above `'conversation.input.dock'`). The InputBar textarea, when inside the host, chains `wheel` with `{ passive: false }`: while the capped textarea can still scroll in that direction it keeps the native gesture; only at its own edge does it `preventDefault` and apply `deltaY` to the host.

-Chat history prepend follows reader intent through stable rendered node/call identities rather than whole-scrollport height deltas. `ChatView` records the first visible `data-chat-anchor-key` and its top relative to the scrollport when paging starts, reselects the currently visible stable anchor after every reader scroll while the request is in flight, and compensates by that row's post-prepend rectangle delta. Reaching the bottom or appending the reader's own message cancels the paging anchor, so a late page cannot pull the view away from the newest content. Bottom follow is stored state rather than raw scroll geometry. A passive wheel listener takes its pre-input baseline from the last main-thread-delivered or programmatically written `scrollTop`, because Chromium may advance compositor geometry before delivering the event; the current non-negative floor excludes a concurrent layout clamp from reader movement. A scroll without matching wheel movement re-pins while following and only refreshes the semantic position while reading. ChatView's single `ResizeObserver` follows streaming, tool disclosure, and draft resize only while bottom ownership remains pinned, without a second per-chunk scroll write.
+Chat history prepend follows reader intent through stable rendered node/call identities rather than whole-scrollport height deltas. `ChatView` records the first visible `data-chat-anchor-key` and its top relative to the scrollport when paging starts, reselects the currently visible stable anchor after every reader scroll while the request is in flight, and compensates by that row's post-prepend rectangle delta. Reaching the bottom or appending the reader's own message cancels the paging anchor, so a late page cannot pull the view away from the newest content. Bottom follow is stored state rather than raw scroll geometry; how reader input is recognized — device-agnostic deviation from the observed-top ledger of the last delivered or written `scrollTop` — is owned by the [reader-scroll-attribution note](2026-08-06-reader-scroll-attribution-observed-top-ledger.md). ChatView's single `ResizeObserver` follows streaming, tool disclosure, and draft resize only while bottom ownership remains pinned, without a second per-chunk scroll write.

 ## Alternatives considered

@@ -26,8 +26,8 @@ Chat history prepend follows reader intent through stable rendered node/call ide

 **Keep StatsLine inside ChatView below the message column.** Rejected: outside the sticky composer it would scroll away while the input stayed pinned.

-**Model every browser scroll input source.** Rejected for this narrow fix: the reproduced desktop path uses wheel/trackpad input. Pointer/touch scrolling, native-scrollbar dragging, keyboard scrolling, focus navigation, and nested overflow ownership remain outside the provenance model instead of adding a general input state machine.
+**Model every browser scroll input source.** Rejected for this narrow fix: the reproduced desktop path uses wheel/trackpad input. Pointer/touch scrolling, native-scrollbar dragging, keyboard scrolling, focus navigation, and nested overflow ownership were left outside the provenance model instead of adding a general input state machine. The [reader-scroll-attribution note](2026-08-06-reader-scroll-attribution-observed-top-ledger.md) later closed this deferral by generalizing attribution through the observed-top ledger, still without an input state machine.

 ## Consequences

-Wheel over the footer scrolls the transcript; the visible layout is a fixed header, scrolling transcript, and sticky bottom composer. Stats appear on every active view tab. Nested view scrollers under the host are suppressed so sticky Turn headers in Trajectory stick to the column host. Concurrent history, streaming, tool expansion, and composer reflow preserve wheel/trackpad scroll decisions, including Chromium's compositor-first delivery and stream-finalization clamp/regrow. Other browser scroll inputs do not change follow ownership under this narrow provenance rule. No session → blank Hero and Hero → active both keep the same textarea DOM node and InputHub draft.
+Wheel over the footer scrolls the transcript; the visible layout is a fixed header, scrolling transcript, and sticky bottom composer. Stats appear on every active view tab. Nested view scrollers under the host are suppressed so sticky Turn headers in Trajectory stick to the column host. Concurrent history, streaming, tool expansion, and composer reflow preserve reader scroll decisions, including Chromium's compositor-first delivery and stream-finalization shrink clamp. Follow ownership extends to every reader input under the [reader-scroll-attribution note](2026-08-06-reader-scroll-attribution-observed-top-ledger.md). No session → blank Hero and Hero → active both keep the same textarea DOM node and InputHub draft.
@@ -14,7 +14,7 @@ Status: implemented

 会话统计挂在 `'conversation.composer.dock'`（位于 `'conversation.input.dock'` 之上）。InputBar 的 textarea 在宿主内以 `{ passive: false }` 链式处理 `wheel`：在限高 textarea 仍能沿该方向滚动时保留原生手势；仅在自身边缘才 `preventDefault` 并将 `deltaY` 施加到宿主。

-Chat 历史前插通过稳定的已渲染 node／call 身份跟随读者意图，而不是使用整个滚动容器的高度差。分页开始时，`ChatView` 记录第一个可见的 `data-chat-anchor-key` 及其相对滚动容器的顶部位置；请求在途期间，每次读者滚动都会重新选择当前可见的稳定锚点；页面到达后则按该行矩形的前后差值补偿。到达底部或追加读者自己的消息会取消分页锚点，因此迟到的页面不能把视图从最新内容拉走。贴底跟随采用存储状态，而不是原始滚动几何状态。passive wheel 监听器以最近一次由主线程交付或由程序写入的 `scrollTop` 作为输入前基线，因为 Chromium 可能先推进合成器几何状态，之后才交付事件；当前使用的非负下限不会将并发的布局钳制计入读者移动。没有对应滚轮／触控板输入位移的滚动，在跟随状态下会重新贴底，在阅读状态下则只刷新语义位置。`ChatView` 的单个 `ResizeObserver` 只会在贴底所有权仍保持时跟随流式输出、工具展开与草稿尺寸变化，且每个 chunk 不会触发第二次滚动写入。
+Chat 历史前插通过稳定的已渲染 node／call 身份跟随读者意图，而不是使用整个滚动容器的高度差。分页开始时，`ChatView` 记录第一个可见的 `data-chat-anchor-key` 及其相对滚动容器的顶部位置；请求在途期间，每次读者滚动都会重新选择当前可见的稳定锚点；页面到达后则按该行矩形的前后差值补偿。到达底部或追加读者自己的消息会取消分页锚点，因此迟到的页面不能把视图从最新内容拉走。贴底跟随采用存储状态，而不是原始滚动几何状态；读者输入如何被识别——即以与设备无关的方式偏离由最近一次交付或写入的 `scrollTop` 构成的 observed-top ledger——由[读者滚动归因笔记](2026-08-06-reader-scroll-attribution-observed-top-ledger.md)负责。`ChatView` 的单个 `ResizeObserver` 只会在贴底所有权仍保持时跟随流式输出、工具展开与草稿尺寸变化，且每个 chunk 不会触发第二次滚动写入。

 ## Alternatives considered

@@ -26,8 +26,8 @@ Chat 历史前插通过稳定的已渲染 node／call 身份跟随读者意图

 **把 StatsLine 留在 ChatView 消息列下方。** 否决：落在 sticky 编辑器之外会随内容滚走，而输入区仍钉在底部。

-**为每一种浏览器滚动输入来源建模。** 此次窄范围修复不采用：已复现的桌面端路径使用滚轮／触控板输入。指针／触控滚动、拖动原生滚动条、键盘滚动、焦点导航与嵌套 overflow 所有权仍不纳入输入来源模型，也不为此新增通用输入状态机。
+**为每一种浏览器滚动输入来源建模。** 此次窄范围修复不采用：已复现的桌面端路径使用滚轮／触控板输入。指针／触控滚动、拖动原生滚动条、键盘滚动、焦点导航与嵌套 overflow 所有权当时被留在输入来源模型之外，也未为此新增通用输入状态机。[读者滚动归因笔记](2026-08-06-reader-scroll-attribution-observed-top-ledger.md)后来通过 observed-top ledger 泛化了归因，补上了这一延后事项，且仍未引入输入状态机。

 ## Consequences

-在页脚上滚轮会滚动 transcript；可见布局是固定标题栏、可滚动 transcript 与 sticky 底部编辑器。统计出现在每一个活跃视图标签上。宿主下的嵌套视图 scroller 被抑制，因而 Trajectory 的 sticky Turn 标题贴在列宿主上。并发历史加载、流式输出、工具展开与编辑器重排会保留滚轮／触控板的滚动决定，包括 Chromium 先推进合成器几何状态再交付事件，以及流收尾阶段滚动位置受钳制后滚动容器重新增长的情况。在这条窄范围的输入来源规则下，其他浏览器滚动输入不会改变贴底跟随所有权。无 session → blank Hero 与 Hero → active 都保持同一 textarea DOM 节点以及 InputHub 草稿。
+在页脚上滚轮会滚动 transcript；可见布局是固定标题栏、可滚动 transcript 与 sticky 底部编辑器。统计出现在每一个活跃视图标签上。宿主下的嵌套视图 scroller 被抑制，因而 Trajectory 的 sticky Turn 标题贴在列宿主上。并发历史加载、流式输出、工具展开与编辑器重排会保留读者的滚动决定，包括 Chromium 先推进合成器几何状态再交付事件，以及流收尾阶段的收缩钳制。贴底跟随所有权依据[读者滚动归因笔记](2026-08-06-reader-scroll-attribution-observed-top-ledger.md)扩展到每一种读者输入。无 session → blank Hero 与 Hero → active 都保持同一 textarea DOM 节点以及 InputHub 草稿。
@@ -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/2026-08-06-provider-credential-lifecycle.md
+2026-08-06-provider-credential-lifecycle.md: ce45207e7ac7224f44e34945e36ba85db0971f09
+2026-08-06-provider-credential-lifecycle.zh.md: c476417517b8ed72036344a13720a8ba378775e6
@@ -0,0 +1,27 @@
+# Agent Note: Recoverable provider credential lifecycle
+
+Status: implemented
+
+English | [中文](2026-08-06-provider-credential-lifecycle.zh.md)
+
+## Problem
+
+The Models editor spans independent settings and credential RPC domains. It previously committed provider settings before storing the API key but kept the revision and original subtree from when the card opened. If the credential write failed, retry replayed the already-committed settings mutation with a stale revision and produced a conflict, leaving the user unable to complete the second stage from the same card. A blank pi-ai key also wrote the derived `apiKeyEnv` without a credential, which prevented pi-ai from using provider-native discovery. At deletion, the inverse leak remained: the profile disappeared but its page-stored key stayed in `.env` and silently became active when the provider was added again. Generic row actions and confirmation copy did not identify which provider would be changed.
+
+## Decision
+
+Provider save remains a two-stage settings-then-credentials operation over the existing wire domains, but the card treats the successful settings response as a commit checkpoint. It replaces its comparison subtree and expected revision with the returned redacted descriptor before attempting `credentials.set`; if that second stage fails, the draft key and card stay visible, and retry produces no settings ops and repeats only the credential write. Genuine concurrent changes before the first settings commit still fail with `settings-conflict`. Typed keys are trimmed at the UI and direct DeepSeek resolver boundaries, and pi-ai records a derived reference only when the normalized key is non-empty; saving a blank key materializes an empty, reference-free profile for provider-native discovery.
+
+Deletion removes a credential only when the joined row identifies the exact `<ROUTE>_API_KEY` reference derived by this page and reports it configured and writable. It unsets that credential before the user-layer profile so a settings-stage failure leaves the row and its frozen target visible for retry; both unsets are idempotent. Custom references, environment credentials, missing credentials, and targets the join cannot identify are retained. The row's accessible Edit/Delete names and the destructive dialog title, description, and final action all use the same stable `Display Name (route-id)` identity, collapsing to the route id when both strings match. The dialog states whether the stored key will be removed and owns operation failures instead of replacing the whole page with a load-error banner. Rows expose API-key state only from the value-free join: a confirmed literal or referenced credential is a green solid dot, a confirmed missing named reference is a red solid dot, and reference-free provider-native authentication or unavailable credential enrichment has no dot. Each dot has accessible copy and a tooltip, while successful Apply uses the same provider identity in a local status message and never echoes secret material.
+
+## Alternatives considered
+
+**Add a cross-domain transaction RPC.** Settings and credentials have separate owning services and durable stores; introducing a new host transaction would broaden the public wire and still require compensation for provider-specific persistence failures. The UI checkpoint makes the current ordered stages recoverable without adding a fourth configuration contract.
+
+**Delete every credential reference named by a removed profile.** A custom reference can be shared, externally managed, or intentionally survive profile churn. Exact equality with this page's derived target plus configured+writable state is the narrow evidence available to the page; anything weaker risks deleting a credential it does not own.
+
+**Remove settings first and compensate by recreating the profile.** The browser holds only a redacted subtree and cannot faithfully reconstruct stored literal secrets or concurrent edits. Credential-first deletion leaves the authoritative profile visible on partial failure and makes retry safe without synthesizing configuration.
+
+## Consequences
+
+The Models page can recover from either second-stage failure without reload, secret disclosure, or a false concurrency conflict, and blank-key pi-ai profiles preserve Bedrock, Vertex, and other provider-native authentication. Confirmed status is visible without turning route liveness, native authentication, or a failed credential lookup into a false error, and a successful replacement remains observable even when the row stays green. Deleting a page-managed provider no longer leaves a reusable local key, while ambiguous credentials deliberately remain for manual management. Save and delete are still not atomic across durable stores: a process crash can stop between stages, but their order and idempotence leave an observable, retryable state. Component tests pin partial-success retries, empty-key native auth, normalized literals, status visibility, target identity, cleanup ownership, and credential/settings rejection ordering; the keyless browser scenario pins bilingual accessible copy and verifies that confirmed deletion removes both `settings.yaml` profile and `.env` credential. This decision refines the Models apply semantics recorded in the [web configuration plane note](../architecture/2026-07-30-web-config-plane.md).
@@ -0,0 +1,27 @@
+# Agent Note: 可恢复的提供方凭据生命周期
+
+Status: implemented
+
+[English](2026-08-06-provider-credential-lifecycle.md) | 中文
+
+## 问题
+
+Models 编辑器横跨互相独立的 settings 与凭据 RPC 领域。之前它先提交提供方 settings，再存储 API 密钥，却一直保留卡片打开时的 revision 和原始子树。如果凭据写入失败，重试会用陈旧 revision 重放已提交的 settings 变更，并产生冲突，导致用户无法从同一张卡片完成第二个阶段。空的 pi-ai 密钥还会写入派生的 `apiKeyEnv`，却不写入凭据，从而阻止 pi-ai 使用提供方原生凭据发现。删除时则存在相反的残留问题：profile 消失了，页面存储的密钥却保留在 `.env` 中，并在重新添加提供方时静默地恢复作用。笼统的行操作与确认文案也没有标明要更改哪个提供方。
+
+## 决策
+
+提供方保存仍在现有 wire 领域上按先 settings、后凭据的两阶段顺序执行，但卡片会把成功的 settings 响应视为提交检查点。它会在尝试 `credentials.set` 之前，用返回的脱敏 descriptor 替换比较基准子树与预期 revision；如果第二阶段失败，草稿密钥与卡片会继续显示，重试不会产生 settings op，只会再次写入凭据。首次 settings 提交之前发生的真实并发变更仍会以 `settings-conflict` 失败。UI 与 DeepSeek 直连 resolver 边界均会去除所输密钥的首尾空白，且只有标准化密钥非空时，pi-ai 才会记录派生引用；留空密钥会具化一个空的、不带引用的 profile，以便使用提供方原生凭据发现。
+
+只有当联接所得的行识别出该页面派生的精确 `<ROUTE>_API_KEY` 引用，并将其报告为已配置且可写时，删除操作才会清除该凭据。它会先取消设置该凭据，再取消设置用户层 profile；如果 settings 阶段失败，该行及其已冻结的目标仍可见，便于重试。两项 unset 都具备幂等性。自定义引用、环境凭据、缺失的凭据，以及联接无法识别目标的凭据均会保留。行的无障碍 Edit/Delete 名称以及破坏性对话框的标题、说明和最终操作都使用同一个稳定的 `Display Name (route-id)` 标识；当两个字符串相同时，标识会简化为路由 id。对话框会说明是否一并删除已存密钥，并在自身内显示操作失败，而不是用加载错误横幅替换整个页面。行只根据不含值的联接结果展示 API 密钥状态：确认已配置的字面密钥或引用凭据显示为绿色实心点，确认缺失的具名引用显示为红色实心点，无引用的提供方原生认证或无法取得凭据补充信息时则不显示状态点。每个状态点都有无障碍文案和工具提示；「应用」成功后的本地状态消息会使用同一个提供方标识，且绝不回显任何机密内容。
+
+## 曾考虑的替代方案
+
+**添加跨领域事务 RPC。**settings 与凭据分属不同的主管服务与持久存储；引入新的 Host 事务会扩大公开 wire 面，而且仍需要补偿提供方特定的持久化失败。UI 检查点让当前的有序阶段变得可恢复，无需添加第四项配置契约。
+
+**删除被移除 profile 所指定的每一个凭据引用。**自定义引用可能被共享、由外部管理，或有意在 profile 反复增删时存留。与该页面派生目标精确相等，再加上已配置且可写的状态，是页面所能获得的最小范围证据；比这更弱的判定都有可能删除不属于它的凭据。
+
+**先删除 settings，再重建 profile 以作补偿。**浏览器只持有脱敏后的子树，无法忠实重建已存的字面机密或并发编辑。先删除凭据可以让权威 profile 在部分失败时仍然可见，并且无需合成配置就能安全重试。
+
+## 后果
+
+Models 页可以从任一第二阶段失败中恢复，无需重新加载，也不会泄露机密或产生虚假的并发冲突；空密钥的 pi-ai profile 会保留 Bedrock、Vertex 与其他提供方原生认证。已确认的状态清晰可见，同时不会把路由存活状态、原生认证或凭据查询失败误报为错误；即使该行继续显示绿色，密钥替换成功也仍然可观察。删除由页面管理的提供方不再遗留可重用的本地密钥，而存在歧义的凭据会有意保留，交由手动管理。保存与删除在跨持久存储时仍非原子操作：进程可能在两个阶段之间崩溃，但它们的顺序与幂等性会留下可观察、可重试的状态。组件测试固定了部分成功后的重试、空密钥原生认证、标准化字面值、状态可见性、目标标识、清理所有权，以及凭据／settings 拒绝顺序；无密钥的浏览器场景固定了双语无障碍文案，并验证确认删除会同时清除 `settings.yaml` profile 与 `.env` 凭据。此决策细化了 [web 配置平面 note](../architecture/2026-07-30-web-config-plane.md) 中记录的 Models 应用语义。
@@ -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/2026-08-06-reader-scroll-attribution-observed-top-ledger.md
+2026-08-06-reader-scroll-attribution-observed-top-ledger.md: 2dcd0178e4216d4d3e2edcd3bfcc05b63606c333
+2026-08-06-reader-scroll-attribution-observed-top-ledger.zh.md: ffb03fdb68c9cb040e311077dd698b310a033fdd
@@ -0,0 +1,37 @@
+# Agent Note: Reader scroll attribution through the observed-top ledger
+
+Status: implemented
+
+English | [中文](2026-08-06-reader-scroll-attribution-observed-top-ledger.zh.md)
+
+## Problem
+
+ChatView's bottom-follow recognized only wheel/trackpad gestures as reader input: while pinned to the floor, a scroll event without matching wheel movement was treated as programmatic and snapped back. Touch panning, native-scrollbar dragging, and keyboard paging therefore could not leave the bottom of a streaming transcript — on a phone the tail was effectively locked. That wheel-only provenance was a deliberate deferral in the [sticky-composer note](2026-07-29-sticky-composer-conversation-scroll.md), which rejected a general input state machine "for this narrow fix" and left every other scroll source outside the model.
+
+## Decision
+
+Reader input is no longer identified by device. ChatView keeps an observed-top ledger (`observedTopRef`): the last `scrollTop` either delivered on the main thread or written by the component, recorded synchronously at every programmatic write site — bottom follow, open restore, prepend anchoring, resize follow, and scroll delivery itself. When a scroll event arrives, a position that deviates from `min(ledger, floor)` by more than half a pixel is reader input; a position on the ledger (a delayed programmatic delivery) or exactly on the shrunken floor (a browser clamp after content shrank) preserves the current ownership state. Ownership then changes only through reader input under the existing threshold rule: within `FOLLOW_THRESHOLD` of the floor re-pins, beyond it releases follow and shows Back to bottom. The wheel listener and its epoch bookkeeping are deleted; the component listens to `scroll` alone, so wheel, touch, scrollbar, keyboard, and any future input source are covered by one rule.
+
+## Contract change: coalesced shrink-plus-regrow clamps
+
+A shrink clamp whose layout regrows within the same rendering update before the clamp's scroll event is delivered is geometrically indistinguishable from reader input, so it now reads as the reader and releases follow (Back to bottom recovers). Realistic React-commit-driven shrink and regrow is still absorbed: the layout-effect follow re-pins and re-records the ledger per commit, and a shrink-only clamp lands exactly on `min(ledger, floor)`. Only a non-React reflow that shrinks and regrows inside one update mis-attributes. The previous wheel model kept following in that raced case; the unit contract was rewritten to the absorbed-shrink-only guarantee in the same change.
+
+## Testing
+
+Unit specs in `packages/client/ui-conversation/tests/chat-view.spec.tsx` pin the ledger contract directly: a `readerScroll` helper delivers a position the component never wrote, programmatic deliveries land on the ledger, and the stream-finalization shrink clamp keeps following. Two scenarios in `apps/web/tests/chat-scroll-contract.e2e.ts` extend the [browser e2e lane](../testing/2026-07-24-web-gui-browser-e2e-lane.md): keyboard paging over a settled transcript and a touch-style momentum fling against paced streaming, both red under the wheel-only implementation and green under the ledger.
+
+The lane's Chromium cannot synthesize any non-wheel device scrolling, which bounds what the e2e can drive for real: `Input.synthesizeScrollGesture` with a touch source and hand-rolled `Input.dispatchTouchEvent` sequences deliver DOM events but never move a scroller (headless and headed-under-Xvfb alike); the `default` gesture source synthesizes wheel events; and compositor scrollbars ignore synthetic mouse input entirely, with a gutter visible only when `--hide-scrollbars` is removed. Keyboard is the one working non-wheel primitive, so it carries the real-input-pipeline proof, and the fling scenario replays touch's signature — per-frame decaying displacements the component never authored — through the scrollport directly.
+
+## Alternatives considered
+
+**Keep the wheel-only model.** Rejected: it is the defect. Touch, scrollbar, and keyboard readers cannot take ownership away from a streaming tail, and each newly supported device would need its own carve-out.
+
+**Enumerate input devices.** Adding `touchstart`/`pointerdown`/`keydown` listeners beside the wheel epoch was the obvious extension. Rejected: native-scrollbar dragging exposes no input event to latch before its scrolls arrive, device lists rot as browsers add sources, and every listener would need its own compositor-delivery grace window — the input state machine the sticky-composer note already declined to build.
+
+**Absorb the coalesced shrink-plus-regrow clamp with heuristics.** Floor-mismatch grace windows or deferred rAF re-checks could keep the raced clamp from reading as the reader. Rejected: streaming rewrites the floor at chunk pace (24 ms) against ~16 ms frames, so any grace window either swallows genuine touch input during streaming — reopening the bug this change fixes — or is too short to cover the race it targets. The mis-attribution is accepted and recoverable instead.
+
+**Drive real touch and scrollbar devices in e2e.** Rejected by the environment, not by preference: every synthesis path (CDP touch gestures, touch event sequences, synthetic mouse on classic scrollbars, headed under Xvfb) was probed and cannot scroll; the details live in Testing above.
+
+## Consequences
+
+Every reader input owns bottom-follow uniformly, with less code: the wheel listener, its epoch counter, and the pre-input baseline bookkeeping are gone, and attribution rides state the component already maintained. The sticky-composer note's layout, wheel chaining, and prepend-anchoring decisions are untouched and remain authoritative; its narrow provenance rule is superseded by this note. The cost is the contract change above — a coalesced non-React shrink-plus-regrow clamp now pauses follow until the reader returns to the floor or presses Back to bottom — traded for touch, scrollbar, and keyboard correctness during streaming. The e2e lane gains non-wheel coverage only within what its browser can synthesize; if gesture synthesis starts working in a future Chromium, the fling emulation can be replaced by real touch strokes without changing the asserted contract.
@@ -0,0 +1,37 @@
+# Agent Note: 经由 observed-top ledger 的读者滚动归因
+
+Status: implemented
+
+[English](2026-08-06-reader-scroll-attribution-observed-top-ledger.md) | 中文
+
+## 问题
+
+ChatView 的贴底跟随此前只把滚轮／触控板手势识别为读者输入：钉在底部（floor）期间，一个没有对应滚轮位移的滚动事件会被视为程序化滚动并被拉回底部。因此触控平移、拖动原生滚动条与键盘翻页都无法离开流式 transcript（文本记录）的底部，在手机上尾部实际上被锁死。这种仅认滚轮的输入来源判定是 [sticky-composer 笔记](2026-07-29-sticky-composer-conversation-scroll.md)中有意的暂缓：该笔记拒绝为「此次窄范围修复」建立通用输入状态机，把其余所有滚动来源都留在模型之外。
+
+## 决策
+
+读者输入不再依据设备来识别。ChatView 维护一份 observed-top ledger（`observedTopRef`）：即最近一次由主线程交付、或由组件自身写入的 `scrollTop`，并在每一个程序化写入点（贴底跟随、打开时恢复、前置锚定、尺寸变化跟随以及滚动交付本身）同步记录。滚动事件到达时，偏离 `min(ledger, floor)` 超过半像素的位置即为读者输入；落在 ledger 上的位置（迟到的程序化交付），或恰好落在收缩后底部上的位置（内容收缩后的浏览器钳制），则维持当前的所有权状态。此后所有权只经由读者输入、按既有阈值规则变化：位置距底部在 `FOLLOW_THRESHOLD` 以内则重新贴底，超出则释放跟随并显示「回到底部」。滚轮监听器及其 epoch 簿记已删除；组件只监听 `scroll`，因此滚轮、触控、滚动条、键盘以及未来任何输入来源都由同一条规则覆盖。
+
+## 契约变更：收缩与重新增长被合并的钳制
+
+如果一次收缩钳制的布局在同一次渲染更新内、赶在该钳制的滚动事件交付之前重新增长，那么这个事件在几何上与读者输入无法区分，因此它现在会被判读为读者并释放跟随（可经「回到底部」恢复）。现实中由 React 提交（commit）驱动的收缩与重新增长仍会被吸收：layout effect 中的跟随会在每次提交后重新贴底并重新记录 ledger，而仅收缩的钳制会恰好落在 `min(ledger, floor)` 上。只有在同一次更新内先收缩再重新增长的非 React 重排会被误归因。旧的滚轮模型在这种竞态情形下会保持跟随；单元测试契约已在同一变更中改写为只吸收纯收缩的保证。
+
+## 测试
+
+`packages/client/ui-conversation/tests/chat-view.spec.tsx` 中的单元测试直接钉住 ledger 契约：`readerScroll` 辅助函数交付一个组件从未写入过的位置，程序化交付落在 ledger 上，流收尾阶段的收缩钳制保持跟随。`apps/web/tests/chat-scroll-contract.e2e.ts` 中的两个场景扩展了[浏览器 e2e 车道](../testing/2026-07-24-web-gui-browser-e2e-lane.md)：在已停稳的 transcript 上做键盘翻页，以及对着按节奏推进的流式输出做一次触控式惯性快滑（momentum fling）；两者在仅认滚轮的实现下均为红、在 ledger 下均为绿。
+
+该车道的 Chromium 无法合成任何非滚轮的设备滚动，这限定了 e2e 能真实驱动的范围：触控来源的 `Input.synthesizeScrollGesture` 与手工构造的 `Input.dispatchTouchEvent` 序列都能交付 DOM 事件，却从不移动滚动容器（无头模式与 Xvfb 下的有头模式皆然）；`default` 手势来源合成的是滚轮事件；合成器滚动条则完全无视合成的鼠标输入，且只有移除 `--hide-scrollbars` 后才能看到滚动条槽。键盘是唯一可用的非滚轮原语，因此由它承担真实输入流水线的证明；快滑场景则把触控的特征（组件从未写入过的逐帧衰减位移）直接回放进滚动容器。
+
+## 曾考虑的替代方案
+
+**保留仅认滚轮的模型。** 否决：它本身就是缺陷所在。触控、滚动条与键盘读者无法从流式尾部夺走所有权，而每种新支持的设备都需要单独开一个特例。
+
+**逐一枚举输入设备。** 在滚轮 epoch 旁边加挂 `touchstart`/`pointerdown`/`keydown` 监听器是最显而易见的扩展。否决：拖动原生滚动条在其滚动事件到达之前不暴露任何可供锁存的输入事件；设备清单会随浏览器新增输入来源而逐渐腐化；而且每个监听器都需要自己的合成器交付宽限窗口——这正是 sticky-composer 笔记当初就拒绝构建的那个输入状态机。
+
+**用启发式吸收收缩与重新增长被合并的钳制。** 针对底部失配的宽限窗口，或推迟到 rAF 的复查，本可让这种竞态下的钳制不被判读为读者。否决：流式输出以分片节奏（24 ms）改写底部，而帧间隔约 16 ms，因此任何宽限窗口要么会在流式输出期间吞掉真实的触控输入（重新打开本次变更所修复的缺陷），要么短到盖不住它想针对的竞态。转而接受这一误归因，它是可恢复的。
+
+**在 e2e 中驱动真实的触控与滚动条设备。** 否决来自环境，而非偏好取舍：每条合成路径（CDP 触控手势、触控事件序列、经典滚动条上的合成鼠标、Xvfb 下的有头模式）都逐一试过，均无法滚动；细节见上文「测试」一节。
+
+## 后果
+
+每种读者输入现在都以同一方式拥有贴底跟随，而代码更少：滚轮监听器、它的 epoch 计数器以及输入前基线簿记均已移除，归因搭载在组件本就维护的状态之上。sticky-composer 笔记中的布局、滚轮链式处理与前置锚定决策原样保留，仍为权威；其窄范围的输入来源规则由本笔记取代。代价就是上文的契约变更：一次收缩与重新增长被合并的非 React 钳制现在会暂停跟随，直到读者回到底部或按下「回到底部」；以此换来流式输出期间触控、滚动条与键盘的正确性。e2e 车道获得的非滚轮覆盖仅限其浏览器能够合成的范围；若手势合成在未来某个 Chromium 版本中开始可用，可以在不改变所断言契约的前提下，把快滑模拟替换为真实的触控划动。
@@ -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/2026-06-15-code-mode.md
-2026-06-15-code-mode.md: b6a24ecd9700e32912b8112b59cbd8b6ab131eb5
-2026-06-15-code-mode.zh.md: a00a43ece1e581190de6096be8138df25a23f07f
+2026-06-15-code-mode.md: 99bbed3edab32512f88ece9694d6519a1f89c2dd
+2026-06-15-code-mode.zh.md: ca1bbe9ed3e412186763d1ed4fca9ed06669d4c3
@@ -6,7 +6,7 @@ English | [中文](2026-06-15-code-mode.zh.md)

 ## Problem

-In the registry's native presentation, the agent loop advertises every visible capability as a JSON-schema function definition. `ToolRegistry` contributes its schemas to the system-prompt assembly, the assembly's `tools` land on the wire (and in the logged request header), the model invokes one `tool-call` block per step, and the loop dispatches each call through `ctx.tools.execute()` **sequentially** (parallel tool execution is an explicit open TODO in `dsh-tools` and [docs/architecture.md](../../../../docs/architecture.md)), with **every** intermediate `tool-result` re-entering the model's context on the next request.
+In the registry's native presentation, the agent loop advertises every visible capability as a JSON-schema function definition. `ToolRegistry` contributes its schemas to the system-prompt assembly, the assembly's `tools` land on the wire (and in the logged request header), the model invokes one `tool-call` block per step, and at the time of this note the loop dispatched each call through `ctx.tools.execute()` **sequentially** (parallel tool execution was an open TODO then; bounded parallel dispatch has since shipped — the [parallel tool-call note](2026-07-10-parallel-tool-call-execution.md), the rolling pool in [docs/architecture.md](../../../../docs/architecture.md)) — with **every** intermediate `tool-result` re-entering the model's context on the next request.

 For multi-step tool work this is token-heavy and serial. The model cannot compose tools — loop over a result set, branch on an intermediate value, fan out, post-process — without a full model round-trip per call, and each round-trip drags the entire intermediate result back into context whether the model needs it or not.

@@ -32,7 +32,7 @@ This note owns Code Mode's presentation, composition, isolation, and settlement

 **Interaction with `toolOrder`, stated up front:** a configured `systemPrompt.toolOrder` naming native capabilities rejects every assembly under `mode: 'code'`, because those names are outside that mode's wire-validation universe. This is correct behavior, not a bug: a deployment using Code Mode updates its order config or drops it.

-**SDK prompt section.** In `'code'` and `'both'`, the lazy `tools:sdk` section in the tool-guidance order band renders TypeScript declarations plus fixed usage instructions for the scope's visible capabilities. It shares lookup and execution visibility, excludes `run_code`, and sorts tools lexicographically for byte-stable output.
+**SDK prompt section.** In `'code'` and `'both'`, the lazy `tools:sdk` section in the tool-guidance order band renders the loaded runtime's language declarations plus fixed usage instructions for the scope's visible capabilities (TypeScript by default; the [language-dispatch note](2026-07-31-code-mode-language-dispatch.md) added Python and the `ctx.codeRuntime.language` renderer table). It shares lookup and execution visibility, excludes `run_code`, and sorts tools lexicographically for byte-stable output.

 **Assembly ownership.** `run_code` and `tools:sdk` enter the trusted `system-prompt/assemble` waterfall as normal assembly inputs. A scoped `tools:sdk` section may shadow the global default before dispatch, and a listener may remove or replace either contribution. The waterfall's returned assembly is final, so whoever changes these inputs owns preserving a viable Code Mode protocol when the deployment expects Code Mode to remain usable; no restoration pass overrides deliberate composition.

@@ -48,7 +48,7 @@ Under `'code'` and `'both'` the registry owns `run_code` as a reserved presentat

 **Sub-call contexts are deferred through the parent.** Injecting inside `run_code` would break parent call/result adjacency, so `ToolRunContext.deferContext()` collects every sub-result `additionalContexts` entry in dispatch order. The registry carries that array even when the program later throws, and the loop appends each entry only after the outer result and every sibling result in the step. An outer post-execute block discards tool-deferred entries and exposes only contexts explicitly attached by the blocking decision.

-**Concurrency is serialized.** Each run owns a dispatch queue, so even `Promise.all` executes tool calls in submission order. Settlement abandons queued calls that have not started. Parallelism requires per-tool concurrency-safety metadata.
+**Concurrency is bounded, not serialized.** Each run owns a dispatch queue that starts calls strictly in submission order and classifies each one through `registry.executionMode`, the same fail-closed `isConcurrencySafe` contract the native loop uses. Consecutive parallel-classified calls overlap up to `maxParallelSubCalls` (default 10; `1` restores serial dispatch); an exclusive call drains the pool and runs alone. Settlement abandons queued calls that have not started. This note shipped the serialized placeholder; the [live-parallel Agent Note](2026-07-26-code-mode-live-parallel-dispatch.md) owns the scheduler that replaced it.

 **Presentation.** `run_code`'s render intent is decided here per the [render-intent Agent Note](../architecture/2026-07-02-tool-render-intent-union.md): `presentCall` creates a `generic` card with `kind: 'execute'`, the program text as its title, and the same program text as `rawInput`; `run_code` intentionally declares no `presentResult`, so the TUI and host/client runtime (Web) complete that card through their generic raw-content fallback using the final durable `tool/result.content`, including captured logs plus the returned value, failure, or post-policy spill preview. This is not a `terminal` card: that card's semantics are "a shell command in a working directory", which a program is not. See the [result-card completeness note](../../archived/bug-fix/2026-07-20-code-mode-result-card-completeness.md).

@@ -64,7 +64,7 @@ Each sub-dispatch appends a log-only `tool/code-dispatch-start` event at pool en
 - `CodeBindingNamespace = { global: string; functions: Record<string, (args: unknown) => Promise<CodeJsonValue>>; errorClass?: { name: string; memberNameProperty: string } }` — the runtime exposes each namespace as a global object of async functions inside the program; the optional descriptor asks the runtime to inject a real program-visible rejection class without teaching the seam consumer-specific names. `CodeJsonValue` is this dependency-light seam's structural lossless-JSON type, so binding arguments and resolutions cross the implementation's serialization boundary whole.
 - `CodeRunResult = { value?: CodeJsonValue; logs: string[]; error?: CodeRunFailure }` — program execution outcomes resolve as the `error` field. `run()` may reject only for caller/seam misuse (for example a duplicate binding namespace); consumers still contain a non-conforming backend rejection at their own error boundary.
 - `CodeRunFailure = { kind: 'exception' | 'timeout' | 'abort' | 'worker-exit' | 'invalid-output' | 'output-limit'; message: string }` — orthogonal outcomes reported independently per [defensive patterns](../../../../docs/defensive-patterns.md); a timed-out run is not an exception, an abort is not a timeout, a lossy completion is not an overflow, and a substrate exit is none of them.
- Two readonly backend descriptors, informational not gating: `language` (what the program must be written in — `'typescript'` for the shipped backend; a Python backend would say so, and pair with its own SDK generator on the presentation side) and `isolation` (`'worker-thread'` for the shipped backend; `'process'`, `'container'`, … for future ones). `dsh-tools` requires `language === 'typescript'` in the MVP — its codegen emits TS — and fails the assembly loudly otherwise, the same misconfiguration idiom as `toolOrder` violations (as when `mode` is non-native with no `ctx.codeRuntime` loaded at all).
+- Two readonly backend descriptors, informational not gating: `language` (what the program must be written in — `'typescript'` for the first backend; a Python backend says `'python'` and pairs with its own SDK generator on the presentation side) and `isolation` (`'worker-thread'` for the shipped backend; `'process'`, `'container'`, … for future ones). `dsh-tools` accepts any `language` with a registered SDK renderer and `run_code` flavor (TypeScript and Python ship; see the [language-dispatch note](2026-07-31-code-mode-language-dispatch.md)) and fails the assembly loudly otherwise, the same misconfiguration idiom as `toolOrder` violations (as when `mode` is non-native with no `ctx.codeRuntime` loaded at all).

 Requests contain every runtime input; implementations own validated timeout and cap defaults. The registry looks up the optional runtime only when Code Mode is assembled, so native mode does not depend on one. Missing or language-incompatible runtimes fail loudly. Alternate substrates or languages can replace the implementation behind the same seam, paired with the appropriate SDK generator.

@@ -85,11 +85,11 @@ The worker runtime provides containment, not a security boundary: model code can

 ### What the model sees

-The SDK instructs the model to write an async erasable-TypeScript body, call tools through `await tools.name(args)`, catch rejected tool calls when needed, and return or log only the output that should re-enter context. Calls remain sequential even under `Promise.all`. The declaration prefix can be as large as native schemas, especially in `'both'`, but remains stable for provider caching.
+The SDK instructs the model to write an async body in the loaded runtime's language (an erasable-TypeScript body by default; a Python `async` body under a Python runtime — see the [language-dispatch note](2026-07-31-code-mode-language-dispatch.md)), call tools through `await tools.name(args)`, catch rejected tool calls when needed, and return or log only the output that should re-enter context. Both flavors state the same contract in their own primitive: independent read-only calls MAY overlap under `Promise.all` (TypeScript) or `asyncio.gather` (Python), mutating calls run alone in submission order, and dependent work sequences with `await`. The declaration prefix can be as large as native schemas, especially in `'both'`, but remains stable for provider caching.

 ## Consequences

-Deployments switching to `'code'` must update any native-only `toolOrder`. Assembly listeners own the integrity of any rewritten protocol surface. Sub-dispatch remains serialized, while per-call contexts retain their source, envelope, and metadata through the outer result.
+Deployments switching to `'code'` must update any native-only `toolOrder`. Assembly listeners own the integrity of any rewritten protocol surface. Sub-dispatch starts in submission order under a bounded overlap pool, while per-call contexts retain their source, envelope, and metadata through the outer result.

 ## Testing

@@ -106,7 +106,7 @@ Deployments switching to `'code'` must update any native-only `toolOrder`. Assem

 **Result elision / summarization over native tool-calling.** Addresses only the context-bloat half of the problem: trimming old `tool-result`s is cheap to add as a logged surface replacement under reconstructable requests, but still pays one model round-trip per call and cannot express loops, branches, or joins. Complementary, not competing; it can layer under Code Mode for residual native calls.

-**Parallel native dispatch in the loop.** The other answer to round-trip cost; still valid future work (the open TODO), still blocked on concurrency-safety metadata, and still no composition — it parallelizes calls the model already decided on in one step. Code Mode's serialized-queue decision keeps the two compatible: when the metadata lands, both native parallel dispatch and per-tool binding parallelism unlock together.
+**Parallel native dispatch in the loop.** The other answer to round-trip cost at decision time; it was blocked on concurrency-safety metadata and offers no composition either way — it parallelizes calls the model already decided on in one step. Code Mode's queue decision kept the two compatible, and that is how it played out: the metadata landed as `isConcurrencySafe` (the [parallel tool-call note](2026-07-10-parallel-tool-call-execution.md)), and native rolling-pool dispatch and per-tool binding parallelism unlocked on the same classifier.

 **Always-exclusive (Cloudflare-faithful, no mode).** Rejected for this SDK's primary consumer: a coding agent's bread-and-butter single calls (`bash`, `read`, `edit`) are already ideal as native calls, and forcing every edit through a program taxes the common case. The mode config keeps the faithful form (`'code'`) one line away without imposing it.

@@ -128,6 +128,6 @@ Deployments switching to `'code'` must update any native-only `toolOrder`. Assem

 **Large lossless JSON values can exhaust memory.** Tool bindings snapshot lossless JSON before dispatch and return canonical JSON resolutions whole. The runtime validates both sides of the worker port and applies no per-binding byte cap; structured-clone cost and process or worker memory are the practical bounds. The combined outer-output ledger for logs, the completion value, and a failure diagnostic is the only byte-capped boundary.

-**Serialized-only sub-dispatch.** `Promise.all` gains no wall-clock parallelism yet, only fewer round-trips; models may over-expect. The instructions state it; lifting it is tied to the same concurrency-safety metadata the native parallel-dispatch TODO needs.
+**Sub-dispatch overlap is bounded by tool safety claims, not by the caller.** A program's `Promise.all` or `asyncio.gather` buys wall-clock parallelism only across calls the tool itself classifies concurrency-safe; a run of exclusive calls still costs its round-trips in sequence, and models may over-expect. Both flavors' SDK instructions state the real contract. This note shipped the serialized placeholder that made the risk absolute; the [live-parallel Agent Note](2026-07-26-code-mode-live-parallel-dispatch.md) owns the scheduler and its overlap cap.

 **Budget metering reads the event loop, not a flag.** Busy-time polling (`eventLoopUtilization()`) is coarser than an exact CPU meter — a budget expires up to one poll interval late — and its correctness claim ("a pending dispatch cannot pause it") is load-bearing against a hostile program. Both sides are unit-tested (hot loop with a pending decoy dispatch dies at `computeMs`; idle-on-slow-binding survives to `maxWallMs`), and the poll interval is an internal constant, not config — nothing a deployment could mis-tune into a bypass. `maxWallMs` is config, and it reaches `setTimeout`, which clamps a delay above `MAX_TIMER_DELAY_MS` (2^31-1 ms) to 1 ms; a positivity check alone therefore accepts a 25-day ceiling that expires on the first tick and times out every run. The worker runtime range-checks the field at load for that reason. `computeMs` needs no upper bound because it is compared against measured utilization instead of being handed to a timer.
@@ -6,7 +6,7 @@ Status: implemented

 ## 问题

-在注册表的原生呈现方式下，agent loop（智能体循环）将每个可见能力以 JSON Schema 函数定义的形式通告给模型。`ToolRegistry` 将其 schema 贡献给系统提示词组装，组装结果中的 `tools` 落到协议格式（wire format）上（也记录在请求头日志中），模型每步调用一个 `tool-call` 块，循环通过 `ctx.tools.execute()` **逐个**分发每次调用（并行工具执行是 `dsh-tools` 和 [docs/architecture.md](../../../../docs/architecture.md) 中明确标注的 open TODO），且**每一个**中间 `tool-result` 都会在下一次请求时重新进入模型上下文。
+在注册表的原生呈现方式下，agent loop（智能体循环）将每个可见能力以 JSON Schema 函数定义的形式通告给模型。`ToolRegistry` 将其 schema 贡献给系统提示词组装，组装结果中的 `tools` 落到协议格式（wire format）上（也记录在请求头日志中），模型每步调用一个 `tool-call` 块，而在本 note 写作时，循环通过 `ctx.tools.execute()` **逐个**分发每次调用（并行工具执行当时还是 open TODO；此后有界的并行分发已经交付——见[并行工具调用 note](2026-07-10-parallel-tool-call-execution.md)，以及 [docs/architecture.md](../../../../docs/architecture.md) 中的 rolling pool）——且**每一个**中间 `tool-result` 都会在下一次请求时重新进入模型上下文。

 对于多步工具操作，这种方式 token 开销大且串行。模型无法组合工具——遍历结果集、根据中间值分支、扇出、后处理——每次调用都需要一次完整的模型往返，而每次往返都会把完整的中间结果拖回上下文，不管模型是否需要。

@@ -32,7 +32,7 @@ Cloudflare 的 [Code Mode](https://blog.cloudflare.com/code-mode/) 提出了一

 **与 `toolOrder` 的交互，预先说明：** 如果配置的 `systemPrompt.toolOrder` 引用了原生能力名称，在 `mode: 'code'` 下会拒绝所有组装，因为那些名称不在该模式的协议校验范围内。这是正确行为而非 bug：使用 Code Mode 的部署需要更新其 order 配置或移除它。

-**SDK 提示词段。** 在 `'code'` 和 `'both'` 下，tool-guidance order band 中的惰性 `tools:sdk` 段为当前 scope 的可见能力渲染 TypeScript 声明加固定的使用说明。它共享查找和执行可见性，排除 `run_code`，并按字典序排列工具以获得字节稳定的输出。
+**SDK 提示词段。** 在 `'code'` 和 `'both'` 下，tool-guidance order band 中的惰性 `tools:sdk` 段为当前 scope 的可见能力渲染所加载运行时语言的声明加固定的使用说明（默认 TypeScript；[语言分发 note](2026-07-31-code-mode-language-dispatch.md) 加入了 Python 与按 `ctx.codeRuntime.language` 选择的渲染器表）。它共享查找和执行可见性，排除 `run_code`，并按字典序排列工具以获得字节稳定的输出。

 **组装所有权。** `run_code` 和 `tools:sdk` 作为正常的组装输入进入受信任的 `system-prompt/assemble` waterfall。一个 scoped 的 `tools:sdk` 段可以在分发前遮蔽全局默认值，监听器也可以移除或替换任一贡献。waterfall 返回的组装结果是最终的，因此修改这些输入的人有责任在部署期望 Code Mode 可用时保持协议面的完整性；没有恢复 pass 会覆盖有意的组合。

@@ -48,7 +48,7 @@ Cloudflare 的 [Code Mode](https://blog.cloudflare.com/code-mode/) 提出了一

 **子调用上下文通过父调用延后。** 在 `run_code` 内部注入会破坏父调用/结果的相邻性，因此 `ToolRunContext.deferContext()` 按分发顺序收集每个子结果的 `additionalContexts` 条目。即使程序后来抛出异常，注册表仍携带该数组；循环只在外层结果与步骤中所有兄弟结果之后追加每个条目。外层 post-execute 阻止会丢弃工具延后的条目，只暴露阻止 decision 显式附加的上下文。

-**并发被序列化。** 每次 run 拥有一个分发队列，因此即使 `Promise.all` 也按提交顺序执行工具调用。结算时放弃尚未开始的排队调用。并行化需要每个工具的并发安全元数据。
+**并发是有界的，而非被序列化。** 每次 run 拥有一个分发队列，严格按提交顺序启动调用，并通过 `registry.executionMode` 对每个调用分类——与原生循环所用的 fail-closed `isConcurrencySafe` 契约相同。连续的 parallel 类调用最多重叠 `maxParallelSubCalls` 个（默认 10；设为 `1` 恢复串行分发）；exclusive 类调用会排空池并单独运行。结算时放弃尚未开始的排队调用。本 note 交付的是被序列化的占位实现；取代它的调度器由[实时并行 Agent Note](2026-07-26-code-mode-live-parallel-dispatch.md) 负责。

 **呈现。** `run_code` 的 render intent 按[呈现意图 Agent Note](../architecture/2026-07-02-tool-render-intent-union.md)在此决定：`presentCall` 创建一个 `generic` 卡片，`kind: 'execute'`，以程序文本作为标题，并将同一程序文本作为 `rawInput`；`run_code` 有意不声明 `presentResult`，因此 TUI 和宿主/客户端运行时（Web）会通过通用原始内容回退机制，使用最终持久化的 `tool/result.content` 补全该卡片，其中包括捕获的日志，以及返回值、失败信息或 post-policy 输出落盘预览。这不是 `terminal` 卡片：该卡片的语义是「工作目录中的 shell 命令」，程序不是。参见[结果卡片完整性说明](../../archived/bug-fix/2026-07-20-code-mode-result-card-completeness.md)。

@@ -64,7 +64,7 @@ Cloudflare 的 [Code Mode](https://blog.cloudflare.com/code-mode/) 提出了一
 - `CodeBindingNamespace = { global: string; functions: Record<string, (args: unknown) => Promise<CodeJsonValue>>; errorClass?: { name: string; memberNameProperty: string } }`——运行时将每个命名空间作为程序内部的全局异步函数对象暴露；可选描述符要求运行时注入真正的、程序可见的 reject 类，而无需让 seam 获知消费方专用名称。`CodeJsonValue` 是这个低依赖 seam 的结构化无损 JSON 类型，因此绑定参数与返回值可以完整跨越实现的序列化边界。
 - `CodeRunResult = { value?: CodeJsonValue; logs: string[]; error?: CodeRunFailure }`——程序执行失败时，执行 promise 仍会 fulfill，并通过 `error` 字段返回失败结果。只有调用方/seam 误用（例如重复的绑定命名空间）时，`run()` 才会 reject；消费方仍在自己的错误边界处理不合规后端的拒绝。
 - `CodeRunFailure = { kind: 'exception' | 'timeout' | 'abort' | 'worker-exit' | 'invalid-output' | 'output-limit'; message: string }`——按[防御性模式](../../../../docs/defensive-patterns.md)独立报告的正交结果；超时的 run 不是异常，abort 不是超时，有损完成值不是溢出，基底退出也与上述情况相互独立。
- 两个只读的后端描述符，仅供信息参考而非门禁判定：`language`（程序必须使用的语言——交付的后端为 `'typescript'`；Python 后端会声明自己，并在呈现侧配对自己的 SDK 生成器）和 `isolation`（交付的后端为 `'worker-thread'`；未来可为 `'process'`、`'container'` 等）。`dsh-tools` 在 MVP 中要求 `language === 'typescript'`——其代码生成输出 TS——否则组装会大声失败，与 `toolOrder` 违规时的配置错误惯用法相同（如 `mode` 为非 native 但根本没有加载 `ctx.codeRuntime`）。
+- 两个只读的后端描述符，仅供信息参考而非门禁判定：`language`（程序必须使用的语言——首个后端为 `'typescript'`；Python 后端声明 `'python'`，并在呈现侧配对自己的 SDK 生成器）和 `isolation`（交付的后端为 `'worker-thread'`；未来可为 `'process'`、`'container'` 等）。`dsh-tools` 接受任何注册了 SDK 渲染器与 `run_code` flavor 的 `language`（TypeScript 与 Python 已交付；见[语言分发 note](2026-07-31-code-mode-language-dispatch.md)），否则组装会大声失败，与 `toolOrder` 违规时的配置错误惯用法相同（如 `mode` 为非 native 但根本没有加载 `ctx.codeRuntime`）。

 请求包含所有运行时输入；实现方拥有经校验的超时和上限默认值。注册表仅在组装 Code Mode 时查找可选的运行时，因此 native 模式不依赖它。缺失或语言不兼容的运行时会大声失败。替代基底或语言可以在同一 seam 背后替换实现，配对相应的 SDK 生成器。

@@ -85,11 +85,11 @@ worker 运行时只能约束程序的运行，而不构成安全边界：模型

 ### 模型看到的内容

-SDK 指示模型编写一个异步的可擦除 TypeScript 函数体，通过 `await tools.name(args)` 调用工具，在需要时捕获被拒绝的工具调用，并仅 return 或 log 应重新进入上下文的输出。即使在 `Promise.all` 下调用仍保持顺序。声明前缀可能与原生 schema 一样大，尤其在 `'both'` 下，但对提供方缓存保持稳定。
+SDK 指示模型编写一个所加载运行时语言的异步函数体（默认可擦除 TypeScript；Python 运行时下为 Python `async` 函数体——见[语言分发 note](2026-07-31-code-mode-language-dispatch.md)），通过 `await tools.name(args)` 调用工具，在需要时捕获被拒绝的工具调用，并仅 return 或 log 应重新进入上下文的输出。两种 flavor 用各自的原语陈述同一契约：相互独立的只读调用可以在 `Promise.all`（TypeScript）或 `asyncio.gather`（Python）下重叠，有副作用的调用按提交顺序单独运行，有依赖的工作用 `await` 排序。声明前缀可能与原生 schema 一样大，尤其在 `'both'` 下，但对提供方缓存保持稳定。

 ## 后果

-切换到 `'code'` 的部署必须更新任何仅限 native 的 `toolOrder`。组装监听器有责任维护任何被重写的协议面的完整性。子分发保持序列化，而每次调用的上下文会通过外层结果保留其 source、信封与元数据。
+切换到 `'code'` 的部署必须更新任何仅限 native 的 `toolOrder`。组装监听器有责任维护任何被重写的协议面的完整性。子分发在有界的重叠池下按提交顺序启动，而每次调用的上下文会通过外层结果保留其 source、信封与元数据。

 ## 测试

@@ -106,7 +106,7 @@ SDK 指示模型编写一个异步的可擦除 TypeScript 函数体，通过 `aw

 **在原生工具调用上做结果省略/摘要。** 仅解决问题中上下文膨胀这一半：裁剪旧 `tool-result` 作为可重建请求下的日志化表面替换成本低，但仍需每次调用一次模型往返，且无法表达循环、分支或汇合。互补而非竞争；它可以在 Code Mode 下为残余的原生调用分层。

-**循环中的并行原生分发。** 往返成本的另一个答案；仍是有效的未来工作（open TODO），仍被并发安全元数据阻塞，且仍无组合能力——它并行化的是模型在一步中已经决定的调用。Code Mode 的序列化队列决策保持两者兼容：当元数据就绪时，原生并行分发和每工具绑定并行化一起解锁。
+**循环中的并行原生分发。** 决策当时对往返成本的另一个答案；它被并发安全元数据阻塞，且无论如何都不提供组合能力——它并行化的是模型在一步中已经决定的调用。Code Mode 的队列决策保持了两者兼容，后续也正是这样落地的：元数据以 `isConcurrencySafe` 的形式就绪（见[并行工具调用 note](2026-07-10-parallel-tool-call-execution.md)），原生 rolling-pool 分发与每工具绑定并行化基于同一个分类器一起解锁。

 **始终排他（忠于 Cloudflare，无模式）。** 否决，因为本 SDK 的主要消费方是编码 agent：其日常的单次调用（`bash`、`read`、`edit`）作为原生调用已经是最优的，强制每次编辑都通过程序会给常见场景增加负担。mode 配置让忠实形式（`'code'`）只需一行配置即可启用，而不强加于人。

@@ -128,6 +128,6 @@ SDK 指示模型编写一个异步的可擦除 TypeScript 函数体，通过 `aw

 **大型无损 JSON 值可能耗尽内存。** 工具绑定会在分发前对无损 JSON 创建快照，并完整返回规范 JSON 返回值。运行时会校验 worker 端口两侧，但不对单次绑定设置字节数上限；结构化克隆成本以及进程或 worker 内存构成实际边界。只有包含日志、完成值和失败诊断的组合外层输出账本受字节数上限约束。

-**仅序列化的子分发。** `Promise.all` 尚未获得挂钟并行性，仅减少往返次数；模型可能过度期望。说明中已声明；解除此限制与原生并行分发 TODO 所需的并发安全元数据绑定。
+**子分发的重叠由工具自身的安全声明限定，而非由调用方决定。** 程序里的 `Promise.all` 或 `asyncio.gather` 只在工具自己分类为并发安全的调用之间换来挂钟并行性；一串 exclusive 调用仍要按顺序付出各自的往返开销，模型可能过度期望。两种 flavor 的 SDK 说明都陈述了真实契约。本 note 交付的是使该风险绝对化的序列化占位实现；调度器及其重叠上限由[实时并行 Agent Note](2026-07-26-code-mode-live-parallel-dispatch.md) 负责。

 **预算计量读取事件循环，而非 flag。** 忙碌时间轮询（`eventLoopUtilization()`）比精确 CPU 计量更粗糙——预算到期最多延迟一个轮询间隔——且其正确性声明（「pending 的分发不能暂停它」）是抵御恶意程序的关键。两种情况均有单元测试（带 pending 诱饵分发的热循环会在耗尽 `computeMs` 预算时终止；等待慢速绑定的空闲程序则会持续运行至 `maxWallMs`），轮询间隔是内部常量而非配置——部署无法将其误调为绕过手段。`maxWallMs` 是配置项，且会传入 `setTimeout`，后者会把超过 `MAX_TIMER_DELAY_MS`（2^31-1 ms）的延迟夹到 1 ms；因此仅有正数校验会放行一个 25 天的上限，它在第一个 tick 就到期，使每次运行都超时。worker 运行时正因如此在加载时对该字段做范围校验。`computeMs` 不需要上界，因为它对照的是实测占用率，而不是交给定时器。
@@ -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/2026-07-31-code-mode-language-dispatch.md
+2026-07-31-code-mode-language-dispatch.md: 96001252d6494d058a8df9974fb5a0d59e7d7112
+2026-07-31-code-mode-language-dispatch.zh.md: aa7eb2a6b4b9117f1d707b37afcdbe12b814bad2
@@ -0,0 +1,46 @@
+# Agent Note: Code Mode language dispatch and the Python SDK renderer
+
+Status: implemented
+
+English | [中文](2026-07-31-code-mode-language-dispatch.zh.md)
+
+## Problem
+
+Code Mode generated one SDK flavor: TypeScript. `ToolRegistry` hard-coded `renderToolsSdk` for the `tools:sdk` section and `requireCodeRuntime` rejected any `ctx.codeRuntime.language !== 'typescript'`. Adding a CPython backend means a program's source language is no longer fixed: the same visible tool registry must project a Python SDK when a Python runtime is loaded, and the model-facing `run_code` schema strings ("Execute a Python program …") must match the SDK section's language so the model never sees a TypeScript instruction over a Python runtime.
+
+This is the tool-facing half of the multi-language Code Mode split; the [code-runtime seam](../../../../packages/code-runtime/code-runtime/README.md) already carries `CodeRuntime.language`. This note owns only how `dsh-tools` dispatches on that field. The backend that implements `language: 'python'` is owned by its own note, delivered separately.
+
+## Decision
+
+Language selection is a lookup on `ctx.codeRuntime.language`, resolved lazily at prompt assembly, against two parallel tables in `dsh-tools`:
+
+- `SDK_RENDERERS` (index.ts) maps a language to its `tools:sdk` renderer — `typescript → renderToolsSdk`, `python → renderToolsSdkPy`. The `tools:sdk` section reads the loaded runtime's language and picks the renderer; `requireCodeRuntime` rejects a `mode: code`/`both` runtime whose language is absent from the table, naming the known languages.
+- `RUN_CODE_FLAVORS` (code-mode.ts) maps a language to its two model-facing `run_code` strings (tool `description` and the `code` parameter description), so a language's SDK section and its transport schema always agree.
+
+Both tables are read with `Object.hasOwn` before use so a language named `toString`/`constructor` cannot resolve an inherited `Object.prototype` member as a renderer. The two guards differ in reachability: `SDK_RENDERERS`' in-callback guard is unreachable because `requireCodeRuntime` validated the same `const` table earlier in the same callback (it carries a `/* v8 ignore */`), while `RUN_CODE_FLAVORS`' guard is the primary, publicly reachable rejection — any language absent from the flavor table hits it through `run_code`'s language-aware getters, which the public `schemas()` reaches without passing `requireCodeRuntime` first; the test reads one of those getters off the definition directly, under a language absent from both tables. A language present in `SDK_RENDERERS` but not `RUN_CODE_FLAVORS` is drift the shared `CodeSdkLanguage` `satisfies` pins reject at `typecheck`, so it is not an input either guard can see; what the guards still own is a mounted runtime reporting a language absent from both tables. Schema emission reads the runtime through `peekRuntime()` rather than `requireRuntime()`: `undefined` (no runtime mounted, reached by definition readers and `schemas()`, of which the doc-catalog harvest is the only shipped one and none of which feeds a model because assembly passes `requireCodeRuntime` first) degrades to the TypeScript flavor, whereas a mounted unknown language fails loud — this is NOT the silent fallback rejected below, which concerns emitting a wrong-language SDK for a real runtime. Adding a backend language is three parallel edits — a `CodeSdkLanguage` member and the two table entries — plus its renderer and the prose that names the well-known values instead of deriving them (the seam's `dsh-code-runtime` README pair, its `CodeRuntime.language` JSDoc, and the `docs/core-data-structures/code-runtime.md` pair; this package's own README pair and its `Config.mode` JSDoc — no gate checks any of it), with no `agent-loop` or registry-structure change.
+
+`code-mode.ts` depends only on the runtime seam (`@deepseek-ai/dsh-code-runtime`), never on a concrete backend; dispatch is by `runtime.language` at run time. The tool layer therefore lands independently of the protocol and backend PRs — it needs only the seam's `language` field, which is already on master.
+
+### The Python SDK renderer
+
+`py-types.ts` renders the same unified tool-schema vocabulary `jsonSchemaToTs` covers, targeting Python: `jsonSchemaToPy` emits a type expression per JSON-schema node, and `renderToolsSdkPy` assembles named `TypedDict`s for each visible tool's arguments and canonical output plus a `tools` object with usage instructions equivalent to the TypeScript flavor. Unsupported raw constructs degrade rather than throwing during assembly, matching the TypeScript renderer's contract. The output is deterministic — lexicographic tool order, byte-identical text for an unchanged tool set — so the prompt stays prefix-cache-friendly. Lexicographic means one ordered member stream: a tool whose name is not a legal attribute is listed as a `tools[name]` comment in its sorted position rather than partitioned to the end, matching how the TypeScript flavor quotes an exotic key in place. That stream forces one thing directly — comment lines are not statements, so a tool set that emits no method at all still needs an explicit `pass`. Three further rules are Python-specific rather than consequences of the ordering. The usage contract states that the declarations are static stubs and arguments are plain `dict`/`list` values: a `TypedDict` reads as a constructible class, so a model that writes `FooArgs(field=1)` gets a `NameError` — TypeScript's `interface` is visibly a type, and the TS flavor's "runs type-stripped" clause already covers it. A description becomes the method's docstring emitted as the FIRST statement of its body: above the `async def` the first one would document the `Tools` class and the rest would be dead expressions, leaving every method undocumented. And a `list[…]` chain degrades to `Any` past `MAX_LIST_NESTING`, because CPython's tokenizer rejects a line with more than 200 open brackets and the block must stay parseable Python — the same reason `docLines` escapes quotes and backslashes. `ts-types` needs neither: TypeScript attaches a leading `/** … */` to the member that follows it and bounds nesting nowhere in its grammar.
+
+The standard that cap serves is grammatical validity, and the boundary is deliberate: a long `A | B | …` union is valid Python at any length and is left uncapped, even though CPython's `compile()` exhausts its C recursion walking the left-nested `BinOp` spine (measured on 3.9: 1,000 branches compile, 5,000 raise `RecursionError`). Nothing compiles this block — it is prompt text — so that limit costs nothing, whereas capping union length would retire the deep-chain tests that pin the walk's linear time and the class-name propagation cap. A future renderer that does need compilable output should flatten unions rather than truncate them.
+
+`renderType` validates the whole schema once (`assertSupportedJsonSchema`) and then trusts it, wrapping the walk in one `try/catch` that degrades to `Any` — the same trusted-after-validation stance the sibling `ts-types` renderer takes at this typed same-process seam ([Trust TypeScript at typed same-process seams](../../../../AGENTS.md)). It deliberately carries NO defenses against a schema whose accessors mutate between reads (post-validation cycles, TOCTOU on `const`/`enum`, self-referential functions): the input is a first-party registration (a `defineTool` literal or a raw registration) or a wire-derived plain JSON schema — the former is trusted per AGENTS.md, the latter is a `JSON.parse` product that physically cannot carry accessors, and `renderType` re-validates the whole tree on every call regardless — so such inputs are unreachable, and adding per-shape guards here would break symmetry with `ts-types` (which has none) for values the static interface forbids. `jsonSchemaToPy(schema: unknown)` accepts `unknown` and returns `Any` on a malformed schema — the Python counterpart of the TS flavor's `unknown` — but its contract is "degrade an unsupported schema", not "survive an adversarial mutating one".
+
+## Alternatives considered
+
+- **A `language` config field on `ToolRegistry`.** Deployment would then have two places to name the language (the loaded runtime and the tools config) that can disagree; the loaded runtime is the single source of truth, so the registry reads it rather than duplicating it.
+- **Importing the Python backend into `code-mode.ts` to detect it.** That would couple the tool layer to a concrete backend and force the protocol/backend PRs to land first. Runtime dispatch on `language` keeps the layer backend-agnostic and independently shippable.
+- **A default renderer for an unknown language.** A silent fallback would emit a TypeScript SDK over, e.g., a Ruby runtime — the model would see instructions in the wrong language. Failing loud at assembly is the repository's misconfiguration stance.
+
+## Consequences
+
+Adding a backend language is three parallel edits — a `CodeSdkLanguage` member, an `SDK_RENDERERS` entry, and a `RUN_CODE_FLAVORS` entry — plus the renderer function the second points at, with no change to `agent-loop` or the registry structure. The two tables (`SDK_RENDERERS`, `RUN_CODE_FLAVORS`) must stay in step, and that invariant is checked statically rather than left to review: both are `satisfies`-checked against that one union, so a language added to one and not the other fails `typecheck`. This is the mechanical form the drift risk deserves — the runtime `Object.hasOwn` guards would catch it too, but only once a backend reporting that language ships: one PR after the drift, at the consumer's integration point rather than where it was introduced, and on this base never, since no second backend exists. The tables keep their `Record<string, …>` declared type because `CodeRuntime.language` is an unconstrained `string`; the union pins what the harness ships, the guards reject what a runtime reports. What stays outside that check is the prose that names the well-known values instead of deriving them: `dsh-code-runtime`'s README pair, its `CodeRuntime.language` JSDoc, and the `docs/core-data-structures/code-runtime.md` pair at the seam, plus this package's own README pair and its `Config.mode` JSDoc. Earlier notes name the values as the state at their own PR and are not on that list. Two separate reasons keep it ungated. Prose is not type-checked at all, wherever the union lives. And no type-level pin can stand in for it here: the interface package must not import its consumer's table, and `CodeRuntime.language` stays an unconstrained `string` by design, so moving the union into the seam would not apply it either. A unit test pinning the two key sets equal was rejected in favor of this: it would buy the same check at the cost of a test-only export of two private tables, and would run later than the compiler does. Which of the two runtime failures surfaces depends on the entry point, for a language absent from both tables: assembly reports the missing renderer, because `wireSchemas` calls `requireCodeRuntime` before projecting, while the public `schemas()` reaches `run_code`'s language-aware getters first and reports the missing flavor. The tool layer stays free of any concrete backend dependency, so it lands and is testable on master ahead of the Python protocol and backend.
+
+The cost is that the Python branch of both tables is unreachable on this base: `CodeRuntime.language` is set by the loaded backend, the only published backend is `dsh-code-runtime-worker` (`'typescript'`), and the registry reads the loaded runtime rather than a config field, so no assembled application can select `renderToolsSdkPy` or `PYTHON_FLAVOR`. The model-visible surface is therefore unchanged by this note's work until a backend reporting `'python'` is published, and this PR's coverage is unit-level — the renderer output plus the dispatch and rejection paths. The keyless snapshot for the Python model interface belongs to the PR that publishes that backend, because only there does a real `cordis.yml` over published plugins produce a Python assembly; a snapshot example that mounted a fixture runtime here would assert against a test double, which [docs/testing.md](../../../../docs/testing.md) rejects as a substitute for the assembled application transcript.
+
+Two runtime contracts the Python SDK text asserts are owed by that same backend PR. First, the instructions tell the model that exactly `tools` and `ToolCallError` are bound and that the declared `TypedDict` classes are not, so the backend must inject those two names — with `ToolCallError.toolName` populated per the seam's `errorClass` contract — and must NOT bind the declared class names into the program's globals; injecting them "helpfully" would make the SDK text false. Second, the language has to be bound to the request: `requireCodeRuntime` resolves `ctx.codeRuntime` separately at assembly and at `run_code` execution, so a reload that swapped the runtime between those two points would hand a program written against one flavor to the other. The split is finer than those two points — `run_code`'s `description` and `parameters` getters each call `resolveFlavor(peekRuntime())`, and `schemaOf` destructures both, so one projection reads the runtime twice; both reads are for `run_code`'s own schema, since the getters are installed on that one definition and every other definition carries plain data properties. A reload between those two reads yields a single schema whose two halves name different languages. Neither is reachable here — one published backend means both reads return the same flavor and no program ever runs against this renderer's output — and the cross-language rejection is not testable until a second language exists.
+
+Third, that PR owns the CPython floor, and with it the renderer's Unicode-table skew. Four expressions read the running engine's tables (Node 22.23.1: Unicode 17.0) while the interpreter uses its own (CPython 3.9.6: 13.0.0): `isBareIdentifier`'s `IDENTIFIER`, and `camelCase`'s split set, head test, and `toUpperCase()`. An interpreter older than the engine is the failing direction — the engine emits a character its tokenizer refuses, taking the whole block down — and it arrives by three independent paths. Through the predicate, a bare method or field name carrying a character added between the two versions — to `XID_Start` at its head, or to `XID_Continue` in any tail position, the middle of a name included. Through `camelCase`'s XID reads, a class name, which reaches emitted text whenever any object shape in the tool's schema declares a `TypedDict`, and which the predicate's verdict on the tool name does not gate: `zz-` plus U+1E4D0 never reaches the predicate's skew, since the `-` rejects it outright, yet it still declares `class Zz𞓐xArgs`. Through the case mapping, a class name derived from a tool the predicate accepted — a different table and a wider window than XID membership: U+019B is XID_Start and NFKC-stable, so `async def ƛ` compiles on 3.9.6, but Node uppercases it to U+A7DC (unassigned there; CPython's own `.upper()` is the identity) and `class ꟜArgs` fails with `invalid non-printable character U+A7DC`. The exposure window is the characters and mappings that changed between the two versions, so the PR that names a supported CPython range must decide explicitly between accepting it and pinning all four read points to tables for that floor — pinning the predicate alone leaves both class-name paths open. Nothing here can decide it: the floor does not exist yet, and a table pinned to a guess would be a deployment-varying constant with no configurability behind it. A second axis rides along with the floor and is not one of the four: the names and syntax the block would evaluate at definition time. `TypedDict` needs 3.8, the PEP 585 builtin generics `dict[str, Any]` and `list[…]` need 3.9, an `A | B` annotation 3.10, and `NotRequired` 3.11. These are not parse failures — the block parses on any version, which is the standard the `MAX_LIST_NESTING` cap serves — but definition-time evaluation failures, and nothing in the product evaluates this text. Recording them with the read points keeps "parseable on the supported range" from being read as "executable on it".
@@ -0,0 +1,46 @@
+# Agent Note: Code Mode 语言分发与 Python SDK 渲染器
+
+Status: implemented
+
+[English](2026-07-31-code-mode-language-dispatch.md) | 中文
+
+## 问题
+
+Code Mode 只生成一种 SDK 形态：TypeScript。`ToolRegistry` 为 `tools:sdk` 段硬编码了 `renderToolsSdk`，且 `requireCodeRuntime` 会拒绝任何 `ctx.codeRuntime.language !== 'typescript'`。引入 CPython 后端后，程序的源语言不再固定：同一个可见工具注册表在加载 Python 运行时时必须投射出 Python SDK，而面向模型的 `run_code` schema 字符串（"Execute a Python program …"）也必须与 SDK 段的语言一致，模型才不会在 Python 运行时下看到 TypeScript 指令。
+
+这是多语言 Code Mode 拆分中面向工具的那一半；[代码运行时 seam](../../../../packages/code-runtime/code-runtime/README.md) 已经携带 `CodeRuntime.language`。本 Note 只负责 `dsh-tools` 如何在该字段上分发。实现 `language: 'python'` 的后端由它自己的 Note 负责，单独交付。
+
+## 决策
+
+语言选择就是对 `ctx.codeRuntime.language` 的查表，在 prompt 装配时惰性解析，查 `dsh-tools` 里两张平行的表：
+
+- `SDK_RENDERERS`（index.ts）把语言映射到它的 `tools:sdk` 渲染器——`typescript → renderToolsSdk`、`python → renderToolsSdkPy`。`tools:sdk` 段读取所加载运行时的语言并选出渲染器；`requireCodeRuntime` 拒绝其语言不在表中的 `mode: code`/`both` 运行时，并列出已知语言。
+- `RUN_CODE_FLAVORS`（code-mode.ts）把语言映射到它那两条面向模型的 `run_code` 字符串（工具 `description` 与 `code` 参数描述），使一种语言的 SDK 段与它的传输 schema 始终一致。
+
+两张表在使用前都以 `Object.hasOwn` 读取，这样名为 `toString`/`constructor` 的语言不会把继承自 `Object.prototype` 的成员解析成渲染器。两个守卫的可达性不同：`SDK_RENDERERS` 的段内守卫不可达，因为 `requireCodeRuntime` 已在同一回调更早处校验过同一张 `const` 表（它带 `/* v8 ignore */`）；而 `RUN_CODE_FLAVORS` 的守卫是主要的、可公开到达的拒绝路径——任何缺席 flavor 表的语言都经 `run_code` 的语言感知 getter 到达它，而公共 `schemas()` 抵达那些 getter 时并未先过 `requireCodeRuntime`；测试直读 definition 上的其中一个 getter，用的是对两张表都缺席的语言。「在 `SDK_RENDERERS` 里却不在 `RUN_CODE_FLAVORS` 里」这种漂移已由共享的 `CodeSdkLanguage` `satisfies` 在 `typecheck` 处拒绝，两个守卫都看不到这种输入；它们如今负责的是所挂载运行时报告了一门两张表都缺席的语言。schema 发射通过 `peekRuntime()` 而非 `requireRuntime()` 读取运行时：`undefined`（无运行时，由直读 definition 的读者与 `schemas()` 到达，其中 doc-catalog 采集是唯一已交付的一个，而它们都不会喂给模型，因为组装路径先过 `requireCodeRuntime`）降级到 TypeScript flavor，而挂载了未知语言则 fail loud——这不是下方被否决的静默回退，那指的是为真实运行时发出错误语言的 SDK。新增一门后端语言是三处并列编辑——一个 `CodeSdkLanguage` 成员加两条表项——再加它的渲染器，以及点名已知值而非从中派生的散文（seam 侧的 `dsh-code-runtime` README 双语对、它的 `CodeRuntime.language` JSDoc 与 `docs/core-data-structures/code-runtime.md` 双语对；本包自己的 README 双语对与它的 `Config.mode` JSDoc，无任何 gate 检查其中任何一处），不动 `agent-loop`，也不动注册表结构。
+
+`code-mode.ts` 只依赖运行时 seam（`@deepseek-ai/dsh-code-runtime`），绝不依赖具体后端；分发在运行时按 `runtime.language` 进行。因此工具层独立于协议和后端 PR 落地——它只需要 seam 的 `language` 字段，而该字段已在 master 上。
+
+### Python SDK 渲染器
+
+`py-types.ts` 渲染 `jsonSchemaToTs` 所覆盖的同一套统一工具 schema 词汇，目标为 Python：`jsonSchemaToPy` 为每个 JSON-schema 节点发出一个类型表达式，`renderToolsSdkPy` 为每个可见工具的参数与规范输出装配具名 `TypedDict`，再加一个带用法说明的 `tools` 对象，与 TypeScript 形态等价。不支持的原始构造在装配时降级而非抛错，与 TypeScript 渲染器的契约一致。输出是确定性的——工具按字典序排列，工具集不变时文本逐字节相同——因此 prompt 保持 prefix-cache 友好。字典序意味着单一有序的成员流：名字不是合法属性的工具以 `tools[name]` 注释出现在它排序后的位置上，而不是被分拣到末尾，与 TypeScript 形态就地为异常键加引号的做法一致。这个成员流直接决定了一件事：注释行不是语句，所以一个不发出任何方法的工具集仍需显式 `pass`。另有三条规则并非源自排序，而是 Python 特有。其一，用法约定声明这些声明只是静态存根、参数为普通 `dict`/`list` 值：`TypedDict` 读起来像一个可构造的类，模型若写 `FooArgs(field=1)` 会得到 `NameError`——TypeScript 的 `interface` 一眼就是类型，且 TS 形态的「runs type-stripped」一句已经覆盖了它。其二，描述会成为方法的 docstring，且必须作为方法体的**第一条语句**发出：放在 `async def` 之上，第一条会变成 `Tools` 的类文档、其余都是无效果表达式，导致每个方法都没有文档。其三，`list[…]` 链超过 `MAX_LIST_NESTING` 后降级为 `Any`，因为 CPython 的 tokenizer 拒绝一行中超过 200 个同时未闭合的括号，而这个块必须是可解析的 Python——与 `docLines` 转义引号和反斜杠是同一个理由。`ts-types` 两者都不需要：TypeScript 会把前置的 `/** … */` 附着到其后的成员上，其语法也不对嵌套设限。
+
+该上限服务的标准是**语法合法性**，这条边界是有意划定的：长的 `A | B | …` union 在任何长度下都是合法 Python，故不设上限——尽管 CPython 的 `compile()` 在沿左嵌套 `BinOp` 脊柱下降时会耗尽 C 递归（在 3.9 上实测：1,000 个分支可编译，5,000 个抛 `RecursionError`）。没有任何东西会编译这个块——它是提示词文本——所以那条限制在这里没有代价；而给 union 长度封顶会作废那几个钉住 walk 线性时间与类名传播上限的深链测试。将来若有渲染器确实需要可编译的输出，应当把 union 拍平，而不是截断。
+
+`renderType` 先用 `assertSupportedJsonSchema` 整树校验一次、随后信任它，用单个 `try/catch` 把整个遍历兜住并降级为 `Any`——与姊妹渲染器 `ts-types` 在这个 typed 同进程 seam 上采取的「校验后信任」姿态一致（[Trust TypeScript at typed same-process seams](../../../../AGENTS.md)）。它有意不设任何针对「访问器在多次读取间变值」的防御（校验后成环、`const`/`enum` 的 TOCTOU、自引用函数）：输入是第一方注册（`defineTool` 字面量或 raw 注册）或从 wire 桥接而来的纯 JSON——前者按 AGENTS.md 受信任，后者是 `JSON.parse` 产物、物理上不可能携带访问器，且每次调用 `renderType` 都会整树重新校验——这类输入不可达，而在此加逐形态守卫会为静态接口所禁止的值破坏与 `ts-types`（没有这类守卫）的对称。`jsonSchemaToPy(schema: unknown)` 接受 `unknown` 并对畸形 schema 返回 `Any`——TypeScript 形态 `unknown` 的对应物——但它的契约是「降级不支持的 schema」，而非「扛住对抗性的可变 schema」。
+
+## Alternatives considered
+
+- **在 `ToolRegistry` 上加一个 `language` 配置字段。** 那样部署方就会有两处命名语言（所加载的运行时与 tools 配置）且可能相互矛盾；所加载的运行时是唯一真相来源，故注册表读取它而不复制它。
+- **把 Python 后端 import 进 `code-mode.ts` 来检测它。** 那会把工具层耦合到具体后端，并迫使协议/后端 PR 先落地。按 `language` 运行时分发使该层保持后端无关、可独立发布。
+- **为未知语言提供默认渲染器。** 静默回退会在比如 Ruby 运行时上发出 TypeScript SDK——模型会看到错误语言的指令。在装配处 fail loud 是本仓库对错误配置的立场。
+
+## Consequences
+
+新增一门后端语言是三处并列编辑——一个 `CodeSdkLanguage` 成员、一个 `SDK_RENDERERS` 表项、一个 `RUN_CODE_FLAVORS` 表项——再加第二处所指向的渲染器函数，不动 `agent-loop`，也不动注册表结构。两张表（`SDK_RENDERERS`、`RUN_CODE_FLAVORS`）必须同步，且这条不变式由静态检查把关，而非交给 review：两张表都以 `satisfies` 对上述同一个 union 校验，因此只加其一而漏掉另一会在 `typecheck` 处失败。这正是该漂移风险应有的机械形式——运行期的 `Object.hasOwn` 守卫同样能捕获，但要等到有后端报告该语言之后：晚于漂移引入一个 PR，且触发点在消费方的集成处而非漂移引入处；在当前 base 上则永远不会触发，因为不存在第二个后端。两张表的声明类型仍是 `Record<string, …>`，因为 `CodeRuntime.language` 是不受约束的 `string`：union 钉住本仓库交付了什么，守卫拒绝运行时报告了什么。落在这条检查之外的是点名已知值而非从中派生的散文：seam 侧的 `dsh-code-runtime` README 双语对、它的 `CodeRuntime.language` JSDoc 与 `docs/core-data-structures/code-runtime.md` 双语对，再加本包自己的 README 双语对与它的 `Config.mode` JSDoc。更早的 note 点名这些值时记的是其自身 PR 当时的状态，不在此列。让它无 gate 的是两条独立理由。其一，散文根本不受类型检查，union 放在哪里都一样。其二，类型级替代在这里也不可用：接口包不得 import 其消费方的表，而 `CodeRuntime.language` 按设计保持不受约束的 `string`，即便把 union 迁进 seam 也不会作用到它。用一个断言两张表键集相等的 unit test 的方案被否决：它买到的是同一条检查，代价却是把两张私有表做测试专用导出，且运行时机晚于编译器。对两张表都缺席的语言，两种运行期失败中报出哪一条随入口而异：组装路径报缺渲染器，因为 `wireSchemas` 在投影前先调 `requireCodeRuntime`；而公共 `schemas()` 先经过 `run_code` 的语言感知 getter，报的是缺 flavor 表项。工具层不依赖任何具体后端，因此它能先于 Python 协议和后端在 master 上落地并可测。
+
+代价是两张表的 Python 分支在当前 base 上不可达：`CodeRuntime.language` 由所加载的后端设定，已发布的后端只有 `dsh-code-runtime-worker`（`'typescript'`），而注册表读取的是所加载的运行时而非某个配置字段，因此没有任何一份组装好的应用能选中 `renderToolsSdkPy` 或 `PYTHON_FLAVOR`。也就是说，在报告 `'python'` 的后端发布之前，本 note 的工作不改变模型可见表面，本 PR 的覆盖因此是 unit 级——渲染器输出加分发与拒绝路径。Python 模型界面的 keyless snapshot 归属于发布该后端的那个 PR，因为只有在那里，一份基于已发布插件的真实 `cordis.yml` 才会产出 Python 组装；在此处挂载 fixture 运行时的快照示例断言的是测试替身，而 [docs/testing.md](../../../../docs/testing.md) 明确拒绝以此替代组装好的应用 transcript。
+
+Python SDK 文本断言的两条运行时契约同样归属那个 backend PR。其一，说明文字告诉模型运行时恰好绑定 `tools` 与 `ToolCallError` 两个名字、所声明的 `TypedDict` 类不绑定，因此后端必须注入这两个名字（并按 seam 的 `errorClass` 契约填充 `ToolCallError.toolName`），且**不得**把所声明的类名绑进程序全局——「好心」注入会使这段 SDK 文本变成假话。其二，语言必须绑定到请求上：`requireCodeRuntime` 在组装时与 `run_code` 执行时分别解析 `ctx.codeRuntime`，若在这两点之间发生重载并换掉运行时，就会把针对一种形态写成的程序交给另一种形态执行。分裂比这两点更细——`run_code` 的 `description` 与 `parameters` 两个 getter 各自调用 `resolveFlavor(peekRuntime())`，而 `schemaOf` 会解构这两个字段，因此一次投影读两次运行时；两次都属于 `run_code` 自己的 schema，因为这两个 getter 只装在那一个 definition 上，其余 definition 携带的都是普通数据属性。在这两次读取之间重载会产出单个 schema 的两半分属不同语言。两者在此处都不可达——只有一个已发布后端意味着两次读取返回同一形态，且没有任何程序会针对本渲染器的输出运行——而跨语言拒绝在第二门语言存在之前也无法测试。
+
+其三，那个 PR 拥有 CPython 版本下限，连带拥有本渲染器的 Unicode 表偏斜。有四处表达式读所运行引擎的表（Node 22.23.1：Unicode 17.0），而解释器用它自己的表（CPython 3.9.6：13.0.0）：`isBareIdentifier` 的 `IDENTIFIER`，以及 `camelCase` 的切分集、头部测试与 `toUpperCase()`。解释器旧于引擎是会失败的那个方向——引擎发出的字符被其 tokenizer 拒收，整个块随之不可解析——而它经三条独立路径抵达。经判据抵达的是裸发的方法名或字段名，其中带有一个在两个版本之间新增的字符——首位加进 `XID_Start`，或尾部任意位置（含名字中部）加进 `XID_Continue`。经 `camelCase` 的 XID 读取抵达的是类名：只要工具 schema 中有任一对象形态声明 `TypedDict`，该类名就进入发出的文本，且判据对工具名的裁决并不对它设闸——工具名 `zz-` 加 U+1E4D0 因 `-` 被判据直接拒绝、从不触及那里的偏斜，却照样声明 `class Zz𞓐xArgs`。经大写映射抵达的是由判据已接受的工具派生出的类名——这是另一张表，窗口也比 XID 归属更宽：U+019B 既是 XID_Start 又 NFKC 稳定，故 `async def ƛ` 在 3.9.6 上可编译，但 Node 将其大写为 U+A7DC（在那里未分配；CPython 自己的 `.upper()` 在此是恒等），于是 `class ꟜArgs` 以 `invalid non-printable character U+A7DC` 失败。暴露窗口是两个版本之间发生变化的那些字符与映射，所以宣布支持某个 CPython 范围的那个 PR 必须在「接受该暴露」与「按该下限的表钉住全部四个读取点」之间显式作出决定——只钉判据会同时留下两条类名路径。此处无法决定：下限尚不存在，而按猜测钉死一张表会成为一个随部署而变、却没有可配置性支撑的常量。还有第二条轴随该下限一同确定，且不属于那四个读取点：本块在定义期会被求值的那些名字与语法。`TypedDict` 需要 3.8，PEP 585 的内建泛型 `dict[str, Any]` 与 `list[…]` 需要 3.9，`A | B` 形式的注解需要 3.10，`NotRequired` 需要 3.11。这些不是解析失败——本块在任何版本上都能解析，这正是 `MAX_LIST_NESTING` 上限所服务的标准——而是定义期求值失败，且产品中没有任何东西会求值这段文本。把它们与那四个读取点记在一起，可避免把「在所支持范围上可解析」读成「在其上可执行」。
@@ -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/testing/2026-07-24-web-gui-browser-e2e-lane.md
-2026-07-24-web-gui-browser-e2e-lane.md: f8519a9622d2f7216226a695db95dbebdbf24ea1
-2026-07-24-web-gui-browser-e2e-lane.zh.md: 294f3e840e0242d9a0d9c53ac510d44d3b0d100f
+2026-07-24-web-gui-browser-e2e-lane.md: 095ad9055c8462ecec67609ed9af55df2a2d7adb
+2026-07-24-web-gui-browser-e2e-lane.zh.md: eb20ee378eb5154bafcb7a448202c66dacda9199
@@ -42,7 +42,7 @@ The typecheck plane split is structural: the host scaffold, its support module,

 ### Coverage contract

-The lane covers three behavior families. Live-turn scenarios pin ordinary tool execution, cancellation, non-retryable failure, transient retry, resident questions, and mid-turn steering; synchronization uses durable events, `whenIdle()`, or an explicit replay marker rather than delays. Cold-history scenarios seed through the real persistence API and cover history rendering, sidebar search, trajectory and waterfall views, and tool details without model calls. Browser-lifecycle scenarios cover first-send workspace materialization, reload recovery, layout reset, theme and locale preferences, and workspace create/rename/view operations. Each family asserts the browser surface and the authoritative host state; a stray model call or under-consumed fixture fails teardown. The required lane additionally carries an 88-turn synthetic Chat scroll contract mixing wrapped Markdown, fenced code, and paired bash calls/results. Real wheel, composer, tool, tab, session, and viewport interactions assert a named settled row's top relative to the transcript scrollport and distance from the true bottom across concurrent history prepend plus paced streaming, pinned/away streaming, tool-disclosure offscreen cycles, expanded-history view/session remount, width reflow, immediate pinned remount, composer resize, and textarea wheel chaining; it deliberately pins neither DOM cardinality nor absolute `scrollTop`, so the same contract can qualify a virtualized implementation. A separate interaction contract over the same fixture pins heterogeneous-row order, independent adjacent tool disclosure, exact user-message clipboard content, a turn-bounded message fork, source/child isolation, and a real follow-up turn in the child; wheel input only navigates to semantic targets and carries no geometry expectation. A short live-history contract starts from a blank workspace and drives consecutive composer turns, including real bash call/result rounds and a paced long final response, pinning one session identity, exact per-turn event ownership, browser echo uniqueness, and composer recovery without timing thresholds.
+The lane covers three behavior families. Live-turn scenarios pin ordinary tool execution, cancellation, non-retryable failure, transient retry, resident questions, and mid-turn steering; synchronization uses durable events, `whenIdle()`, or an explicit replay marker rather than delays. Cold-history scenarios seed through the real persistence API and cover history rendering, sidebar search, trajectory and waterfall views, and tool details without model calls. Browser-lifecycle scenarios cover first-send workspace materialization, reload recovery, layout reset, theme and locale preferences, and workspace create/rename/view operations. Each family asserts the browser surface and the authoritative host state; a stray model call or under-consumed fixture fails teardown. The required lane additionally carries an 88-turn synthetic Chat scroll contract mixing wrapped Markdown, fenced code, and paired bash calls/results. Real wheel, composer, tool, tab, session, and viewport interactions assert a named settled row's top relative to the transcript scrollport and distance from the true bottom across concurrent history prepend plus paced streaming, pinned/away streaming, tool-disclosure offscreen cycles, expanded-history view/session remount, width reflow, immediate pinned remount, composer resize, and textarea wheel chaining; real keyboard paging and a touch-style fling emulation additionally pin wheel-free bottom-follow ownership ([reader-scroll-attribution note](../bug-fix/2026-08-06-reader-scroll-attribution-observed-top-ledger.md)); it deliberately pins neither DOM cardinality nor absolute `scrollTop`, so the same contract can qualify a virtualized implementation. A separate interaction contract over the same fixture pins heterogeneous-row order, independent adjacent tool disclosure, exact user-message clipboard content, a turn-bounded message fork, source/child isolation, and a real follow-up turn in the child; wheel input only navigates to semantic targets and carries no geometry expectation. A short live-history contract starts from a blank workspace and drives consecutive composer turns, including real bash call/result rounds and a paced long final response, pinning one session identity, exact per-turn event ownership, browser echo uniqueness, and composer recovery without timing thresholds.

 ### CI stance

@@ -42,7 +42,7 @@ Web GUI 以一条真实组装链交付——chromium 页面 → client 插件 bu

 ### 覆盖契约

-该车道覆盖三类行为。实时轮次场景钉住普通工具执行、取消、不可重试失败、瞬态重试、常驻提问与轮次中途 steering；同步依赖持久事件、`whenIdle()` 或显式回放标记，而不使用延时。冷历史场景通过真实持久化 API 播种，在不调用模型的情况下覆盖历史渲染、侧栏搜索、Trajectory 与 Waterfall 视图及工具详情。浏览器生命周期场景覆盖首次发送时物化工作区、重新加载恢复、布局重置、主题与语言偏好，以及工作区的创建、重命名和视图操作。每类场景都断言浏览器表面和权威的 host 状态；离群的模型调用或未耗尽的 fixture 会使拆卸失败。必需车道还包含一份合成的 88 轮 Chat 滚动契约，其中混合了换行 Markdown、围栏代码以及成对的 bash 调用/结果。真实 wheel、输入框、工具、tab、会话与 viewport 交互会在并发历史前插加带节奏流式输出、贴底/离底流式输出、工具 disclosure 离屏循环、扩展历史后的视图/会话重新挂载、宽度重排、贴底后立即重新挂载、输入框尺寸变化以及 textarea wheel 链场景中，断言一个具名已结算行相对 transcript scrollport 的顶部位置和到真实底部的距离；它刻意不钉 DOM 基数或绝对 `scrollTop`，因此同一契约可以验收虚拟化实现。另一份基于同一 fixture 的交互契约钉住异构行顺序、相邻工具 disclosure 的独立状态、用户消息剪贴板内容的精确值、以轮次为边界的消息 fork、源会话/子会话隔离，以及子会话中的一次真实追问轮次；wheel 输入只用于导航到语义目标，不承载几何预期。一份简短的实时历史契约从空白工作区开始，连续驱动输入框轮次，其中包括真实的 bash 调用/结果轮次和一段带节奏的长篇最终响应；它钉住单一会话身份、每轮事件的精确归属、浏览器回显唯一性与输入框恢复，不设置时间阈值。
+该车道覆盖三类行为。实时轮次场景钉住普通工具执行、取消、不可重试失败、瞬态重试、常驻提问与轮次中途 steering；同步依赖持久事件、`whenIdle()` 或显式回放标记，而不使用延时。冷历史场景通过真实持久化 API 播种，在不调用模型的情况下覆盖历史渲染、侧栏搜索、Trajectory 与 Waterfall 视图及工具详情。浏览器生命周期场景覆盖首次发送时物化工作区、重新加载恢复、布局重置、主题与语言偏好，以及工作区的创建、重命名和视图操作。每类场景都断言浏览器表面和权威的 host 状态；离群的模型调用或未耗尽的 fixture 会使拆卸失败。必需车道还包含一份合成的 88 轮 Chat 滚动契约，其中混合了换行 Markdown、围栏代码以及成对的 bash 调用/结果。真实 wheel、输入框、工具、tab、会话与 viewport 交互会在并发历史前插加带节奏流式输出、贴底/离底流式输出、工具 disclosure 离屏循环、扩展历史后的视图/会话重新挂载、宽度重排、贴底后立即重新挂载、输入框尺寸变化以及 textarea wheel 链场景中，断言一个具名已结算行相对 transcript scrollport 的顶部位置和到真实底部的距离；真实键盘翻页与触摸式惯性滑动模拟额外钉住不依赖 wheel 的贴底跟随所有权（[读者滚动归因笔记](../bug-fix/2026-08-06-reader-scroll-attribution-observed-top-ledger.md)）；它刻意不钉 DOM 基数或绝对 `scrollTop`，因此同一契约可以验收虚拟化实现。另一份基于同一 fixture 的交互契约钉住异构行顺序、相邻工具 disclosure 的独立状态、用户消息剪贴板内容的精确值、以轮次为边界的消息 fork、源会话/子会话隔离，以及子会话中的一次真实追问轮次；wheel 输入只用于导航到语义目标，不承载几何预期。一份简短的实时历史契约从空白工作区开始，连续驱动输入框轮次，其中包括真实的 bash 调用/结果轮次和一段带节奏的长篇最终响应；它钉住单一会话身份、每轮事件的精确归属、浏览器回显唯一性与输入框恢复，不设置时间阈值。

 ### CI 立场