Merge remote-tracking branch 'origin/master' into xtr/agent-loop-message-machine

# Conflicts:
#	docs/cordis-catalog/services.md
#	examples/acp-agent/tests/snapshots/code-mode-workspace-context/session.jsonl
#	packages/core/tools/README.i18n.yaml
#	packages/core/tools/src/code-mode.ts

.agents/notes/implemented/feature/2026-06-15-code-mode.i18n.yaml

@@ -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
-2026-06-15-code-mode.md: be5e29fdafd72c54d5801cc29597a27f401e4872
-2026-06-15-code-mode.zh.md: 55b47cc93bd953a610b5e2d9167e5d3cba1ed90e
+2026-06-15-code-mode.md: 1170fa4f9fa778fa9176097317477ea336588d32
+2026-06-15-code-mode.zh.md: 4362efa332d0ed24a6383a4cf75b83c0dba22e7e

.agents/notes/implemented/feature/2026-06-15-code-mode.md

@@ -42,7 +42,7 @@ This note owns Code Mode's presentation, composition, isolation, and settlement
 
 Under `'code'` and `'both'` the registry owns `run_code` as a reserved presentation transport with two required parameters, `{ code: string; description: string }` (the description labels the call in UIs, the bash precedent). It is represented by a normal `ToolDefinition` for dispatch but stays outside the filterable capability layers, so restrictions cannot accidentally remove Code Mode's only entry point. Calls traverse the complete tool pipeline — `tools/pre-execute` → monotonic guards → `tools/execute` around dispatch → `tools/post-execute` → optional definition-owned `finalizeContent` → immutable `tools/result` notification — exactly like native calls; a permission plugin can inspect the program text before it runs, and final-result observers see the normalized outer outcome. Its `execute(args, exec)`:
 
-1. **Build bindings.** One run-scoped signal follows outer cancellation and is aborted whenever the run settles. Each visible tool binding snapshots lossless-JSON arguments, waits on the serialization queue, executes with a deterministic call id and the outer token as `parent`, defers returned contexts through the outer execution, and logs `tool/code-dispatch` with the full rendered result content. Success returns the tool's final canonical JSON value; failure becomes the program-visible `ToolCallError`. Every sub-call retains its own immutable execution identity and traverses the full tool pipeline.
+1. **Build bindings.** One run-scoped signal follows outer cancellation and is aborted whenever the run settles. Each visible tool binding snapshots lossless-JSON arguments, enters the native-contract dispatch pool (the [live-parallel note](2026-07-26-code-mode-live-parallel-dispatch.md) owns the scheduling design), executes with a deterministic call id and the outer token as `parent`, defers returned contexts through the outer execution, and logs the `tool/code-dispatch-start`/`tool/code-dispatch` pair, the settle side carrying the full rendered result content. Success returns the tool's final canonical JSON value; failure becomes the program-visible `ToolCallError`. Every sub-call retains its own immutable execution identity and traverses the full tool pipeline.
 2. **Runs the program**: `ctx.codeRuntime.run({ program: args.code, bindings: [{ global: 'tools', functions }], signal: runController.signal })`. The runtime receives the run-scoped signal, not only the caller's outer signal, so any way the outer run settles also aborts work inside the runtime.
 3. **Settle after quiescence.** When the runtime settles, the bridge aborts outstanding work and drains the dispatch queue before returning. Success returns captured logs and the completion value as canonical output; the registry renders that value into durable `tool/result.content`, which the result card reads directly. A runtime failure becomes `CodeRunFailedError`; backend rejection uses the registry's normal error boundary. Both produce structured error results, and no sub-call can append after `run_code` settles.
 
@@ -54,7 +54,7 @@ Under `'code'` and `'both'` the registry owns `run_code` as a reserved presentat
 
 ### Observability: `tool/code-dispatch`
 
-Each sub-dispatch appends a log-only `tool/code-dispatch` event containing parent and child call ids, tool identity, normalized arguments, and the complete rendered `content`/`isError` outcome. It remains outside model history but available to persistence and UIs. Appends occur inside the open `run_code` turn. Direct executions without an agent still run but cannot log the event.
+Each sub-dispatch appends a log-only `tool/code-dispatch-start` event at pool entry and a `tool/code-dispatch` settle event containing parent and child call ids, tool identity, normalized arguments, and the complete rendered `content`/`isError` outcome. It remains outside model history but available to persistence and UIs. Appends occur inside the open `run_code` turn. Direct executions without an agent still run but cannot log the event.
 
 ### The code-runtime seam
 

.agents/notes/implemented/feature/2026-06-15-code-mode.zh.md

@@ -42,7 +42,7 @@ Cloudflare 的 [Code Mode](https://blog.cloudflare.com/code-mode/) 提出了一
 
 在 `'code'` 和 `'both'` 下，注册表拥有 `run_code` 作为保留的呈现传输通道，带两个必需参数 `{ code: string; description: string }`（description 为 UI 标注该调用，沿用 bash 的先例）。它由一个正常的 `ToolDefinition` 表示以供分发，但位于可过滤的能力层之外，因此限制规则不会意外移除 Code Mode 的唯一入口。调用遍历完整的工具流水线——`tools/pre-execute` → 单调守卫 → `tools/execute` 包裹分发 → `tools/post-execute` → 由定义拥有的可选 `finalizeContent` → 不可变的 `tools/result` 通知——与原生调用完全一致；权限插件可以在程序运行前检查程序文本，最终结果观察者看到的是规范化的外层结果。其 `execute(args, exec)`：
 
-1. **构建绑定。** 一个 run 级别的 signal 跟随外层取消，并在 run 结算时被 abort。每个可见工具绑定都会对无损 JSON 参数创建快照，等待序列化队列，以确定性的 call id 和外层 token 作为 `parent` 执行，通过外层 execution 延后返回的上下文，并连同完整渲染后的结果内容记录 `tool/code-dispatch`。成功时返回工具最终的规范 JSON 值；失败则变为程序可见的 `ToolCallError`。每个子调用保留自己不可变的执行标识，并遍历完整的工具流水线。
+1. **构建绑定。** 一个 run 级别的 signal 跟随外层取消，并在 run 结算时被 abort。每个可见工具绑定都会对无损 JSON 参数创建快照，进入原生契约的分发池（调度设计由[实时并行 Agent Note](2026-07-26-code-mode-live-parallel-dispatch.md) 负责），以确定性的 call id 和外层 token 作为 `parent` 执行，通过外层 execution 延后返回的上下文，并记录 `tool/code-dispatch-start`/`tool/code-dispatch` 事件对，其中结算侧携带完整渲染后的结果内容。成功时返回工具最终的规范 JSON 值；失败则变为程序可见的 `ToolCallError`。每个子调用保留自己不可变的执行标识，并遍历完整的工具流水线。
 2. **运行程序**：`ctx.codeRuntime.run({ program: args.code, bindings: [{ global: 'tools', functions }], signal: runController.signal })`。运行时接收的是 run 级别的 signal 而非仅调用方的外层 signal，因此外层 run 以任何方式结算都会同时 abort 运行时内部的工作。
 3. **完全停稳后结算。** 运行时结算后，桥 abort 未完成的工作并排空分发队列后再返回。成功时返回捕获的日志和完成值，将其作为规范输出；注册表再把该值渲染为持久化的 `tool/result.content`，供结果卡片直接读取。运行时失败变为 `CodeRunFailedError`；后端拒绝使用注册表的正常错误边界。两者都产生结构化的错误结果，且 `run_code` 结算后不允许子调用追加。
 
@@ -54,7 +54,7 @@ Cloudflare 的 [Code Mode](https://blog.cloudflare.com/code-mode/) 提出了一
 
 ### 可观测性：`tool/code-dispatch`
 
-每次子分发追加一个仅日志的 `tool/code-dispatch` 事件，包含父子 call id、工具标识、规范化参数以及完整渲染后的 `content`/`isError` 结果。它不进入模型历史，但可供持久化和 UI 使用。追加发生在开放的 `run_code` 轮次内。没有 agent 的直接执行仍然运行，但无法记录该事件。
+每次子分发在进入分发池时追加一个仅日志的 `tool/code-dispatch-start` 事件，并以一个 `tool/code-dispatch` 结算事件收尾，后者包含父子 call id、工具标识、规范化参数以及完整渲染后的 `content`/`isError` 结果。它不进入模型历史，但可供持久化和 UI 使用。追加发生在开放的 `run_code` 轮次内。没有 agent 的直接执行仍然运行，但无法记录该事件。
 
 ### code-runtime seam
 

.agents/notes/implemented/feature/2026-07-26-code-mode-chat-subcall-rows.i18n.yaml

@@ -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
+2026-07-26-code-mode-chat-subcall-rows.md: 7d666f0a9e4b8bdb9bd6f5d0d0984fee0c4b21e2
+2026-07-26-code-mode-chat-subcall-rows.zh.md: fb9b0c62bb702cfdb7ba3c8ccce73d8e43f29c1b

.agents/notes/implemented/feature/2026-07-26-code-mode-chat-subcall-rows.md

@@ -0,0 +1,32 @@
+# Agent Note: Code Mode chat rendering — sub-calls as native rows under the parent
+
+Status: implemented
+
+English | [中文](2026-07-26-code-mode-chat-subcall-rows.zh.md)
+
+> Scope: how the web chat view renders a `run_code` turn — the client-side half of the Code Mode UI stack, built on the [host foundation](2026-07-26-code-dispatch-ui-foundation.md) (full-content `tool/code-dispatch`, the required `description` parameter). The [toolview dissolution](../architecture/2026-07-23-toolview-dissolution.md) owns the slot model this rides on.
+
+## Problem
+
+With Code Mode enabled, the chat view showed one opaque `run_code` row: raw program text as the summary, sub-calls invisible everywhere. The settled product requirement is the opposite: each sub-call must render *identically* to a native tool call — same row components, same custom registrations, same details panel — while the transcript stays honest about the fact that the model made ONE call.
+
+## Decision
+
+**Sub-calls are `ToolResultNode`s indexed off the surface flow, rendered through the same keyed slot as native rows, nested always-visible under their parent.**
+
+- **Data layer**: `Session.applyEventSideEffects` folds each in-window `tool/code-dispatch` into `ConversationSnapshot.codeDispatches: ReadonlyMap<parentCallId, readonly CodeSubCall[]>`, where `CodeSubCall` IS `ToolResultNode` (the sub-call id as `callId`, the logged args JSON-stringified into `call.argsRaw`, the full logged `content`/`isError`). Live mux frames and history replay build the identical index (`rebuildDerivedFromWindow` clears and re-derives; copy-on-write per-parent arrays keep snapshot references memo-stable). Sub-calls never join `nodes` — the surface flow remains exactly the model-visible turn structure. The event is narrowed structurally at the wire-consumer boundary (dsh-tools' host types cannot enter the client program — the host/client `Context` merges collide), the same posture as every cross-wire payload.
+- **Render layer**: `ChatView`'s `CallRow` renders the parent, then — for parents present in the index — a `[data-subcalls]` nest of `SubCallRow`s, each dispatching through the SAME `'conversation.chat.toolview'` keyed hole with `entryKey = sub-tool name` and the same `GenericToolCard` fallback. Identity with native rows holds by construction: a keyed registration (e.g. the bash sample) takes over sub-rows exactly as it takes over top-level rows, with zero registration changes. Running parents (`runningCalls`) nest their so-far dispatches the same way, so sub-rows stream in live during the run (PR1 logs each dispatch as it completes).
+- **`run_code` presentation**: a new `code` row variant (classifier `run_code → code`, `Code` title, `IconCodeOutline16`) summarizes with the model-authored `description` and expands to the program itself (monospace on the markdown code-block fill) rather than the args JSON envelope.
+- **Details panel**: `materialFor` falls through nodes → runningCalls → the dispatch index, so a selected sub-callId resolves to full args and complete output through the identical rendering path as a native settled call.
+
+## Alternatives considered
+
+**Sub-calls flat in the surface flow (fold them into `nodes`).** Rejected: misrepresents the transcript — the model made one call; nesting under the parent preserves the code↔calls association and keeps the fold's model-visible-order invariant untouched.
+
+**Hidden until the parent row expands.** Rejected by product decision: the sub-calls ARE the story of a Code Mode turn; hiding them re-creates the opacity this feature removes. The parent's expand toggle reveals only the program.
+
+**A dedicated sub-call row component.** Rejected: the whole point is identity with native rows; a parallel component would drift. The nest wrapper (indent + left edge) is the only sub-call-specific chrome.
+
+## Consequences
+
+Custom toolview registrations apply to sub-calls for free — and deliberately: there is no per-registration opt-out short of the component reading its own context, which no current consumer needs. Selection highlighting reaches nested rows through the same `selectedCallId` channel (group membership tests both levels). Trajectory/waterfall still render `run_code` as a single row — their sub-call spans are deferred to the PR that adds dispatch timing (start/end events), without which a waterfall span would be a lie. Fixture turn 64 (`?fixture`) plus the `code-mode-round` browser e2e (recorded real round, keyless replay) pin the full surface; the jsdom suites pin the slot dispatch, error states, details resolution, and index reference stability.

.agents/notes/implemented/feature/2026-07-26-code-mode-chat-subcall-rows.zh.md

@@ -0,0 +1,32 @@
+# Agent Note：Code Mode 的 chat 渲染——子调用作为父行之下的原生行
+
+Status: implemented
+
+[English](2026-07-26-code-mode-chat-subcall-rows.md) | 中文
+
+> 范围：web chat 视图如何渲染一个 `run_code` 轮次，即 Code Mode UI 堆叠 PR（Pull Request）链的 client 侧一半，构建在[宿主侧基础](2026-07-26-code-dispatch-ui-foundation.md)之上（携带完整内容的 `tool/code-dispatch`、必填的 `description` 参数）。本篇所依托的 slot 模型归 [toolview 溶解](../architecture/2026-07-23-toolview-dissolution.md)所有。
+
+## 问题
+
+启用 Code Mode 后，chat 视图过去只显示一条不透明的 `run_code` 行：摘要就是原始程序文本，子调用则处处不可见。已敲定的产品要求恰恰相反：每个子调用都必须与原生工具调用渲染得*完全一致*——同样的行组件、同样的自定义注册、同样的 details 面板——同时 transcript（文本记录）仍须如实反映模型只发起了一次调用这一事实。
+
+## 决策
+
+**子调用是 surface 流之外单独索引的 `ToolResultNode`，经由与原生行相同的 keyed slot 渲染，以始终可见的方式嵌套在父行之下。**
+
+- **数据层**：`Session.applyEventSideEffects` 把窗口内的每条 `tool/code-dispatch` 折入 `ConversationSnapshot.codeDispatches: ReadonlyMap<parentCallId, readonly CodeSubCall[]>`，其中 `CodeSubCall` 本身就是 `ToolResultNode`（子调用 id 充当 `callId`，已记录的参数经 JSON 字符串化写入 `call.argsRaw`，完整记录的 `content`/`isError` 原样携带）。live mux 帧与历史回放构建出同一份索引（`rebuildDerivedFromWindow` 先清空再重新推导；逐父级的写时复制（copy-on-write）数组保持快照引用 memo 稳定）。子调用永不进入 `nodes`——surface 流始终精确等于模型可见的轮次结构。该事件在 wire 消费方边界作结构性收窄（dsh-tools 的 host 类型进不了 client 程序——host/client 两侧的 `Context` 声明合并会冲突），姿态与所有跨 wire 载荷一致。
+- **渲染层**：`ChatView` 的 `CallRow` 先渲染父行，随后对索引中出现的父级渲染一组 `[data-subcalls]` 嵌套的 `SubCallRow`，每一行都经由同一个 `'conversation.chat.toolview'` keyed 孔位、以 `entryKey = sub-tool name` 分发，并共用同一个 `GenericToolCard` fallback。与原生行的同一性由构造保证：一个 keyed 注册（例如 bash 样例）接管子行与接管顶层行的方式完全相同，注册本身零改动。运行中的父调用（`runningCalls`）也以同样的方式嵌套目前已产生的分发，因此子行在运行期间实时流入（PR1 在每次分发完成时即记录该分发）。
+- **`run_code` 的呈现**：新增一种 `code` 行变体（分类器映射 `run_code → code`、标题 `Code`、图标 `IconCodeOutline16`），以模型撰写的 `description` 作摘要，展开后显示程序本身（在 markdown 代码块的填充底色上以等宽字体呈现），而非参数的 JSON 信封。
+- **details 面板**：`materialFor` 按 nodes → runningCalls → 分发索引的顺序逐级回落，因此被选中的子调用 callId 会经由与已完结的原生调用完全相同的渲染路径，解析出完整参数与完整输出。
+
+## 曾考虑的替代方案
+
+**把子调用平铺进 surface 流（折入 `nodes`）。** 否决：这会歪曲 transcript——模型只发起了一次调用；嵌套在父行之下既保住代码↔调用的关联，也让 fold 的模型可见顺序不变式原封不动。
+
+**隐藏子调用，展开父行后才显示。** 由产品决策否决：子调用正是一个 Code Mode 轮次的核心内容；把它们藏起来，等于重新制造出本功能所要消除的那种不透明。父行的展开开关只用于显示程序本身。
+
+**专用的子调用行组件。** 否决：本功能的全部要义就在于与原生行保持同一性；一个平行组件必然漂移。嵌套包装层（缩进 + 左侧边线）是子调用唯一的专属 chrome。
+
+## 后果
+
+自定义 toolview 注册免费适用于子调用——而且是刻意为之：不存在按注册粒度的 opt-out，唯一的出路是组件自行读取自身上下文，而当前没有任何消费方需要这么做。选中高亮经由同一条 `selectedCallId` 通道到达嵌套行（分组归属判断会同时检验两个层级）。trajectory/waterfall 仍把 `run_code` 渲染为单独一行——它们的子调用 span 推迟到增加分发计时（start/end 事件）的那个 PR；缺少计时，waterfall 上的 span 就是在撒谎。fixture（测试前置数据）的轮次 64（`?fixture`），加上 `code-mode-round` 浏览器 e2e（录制的真实 round、无密钥回放），共同锁定整个表面；jsdom 套件则锁定 slot 分发、错误状态、details 解析与索引引用稳定性。

.agents/notes/implemented/feature/2026-07-26-code-mode-live-parallel-dispatch.i18n.yaml

@@ -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
+2026-07-26-code-mode-live-parallel-dispatch.md: b4afc21be902d8ed3e5bee2ad1a540413a864f25
+2026-07-26-code-mode-live-parallel-dispatch.zh.md: 409e4cbf9ea3d1b4d1bbe0cd86b494429ebb8a3d

.agents/notes/implemented/feature/2026-07-26-code-mode-live-parallel-dispatch.md

@@ -0,0 +1,32 @@
+# Agent Note: Code Mode live dispatch lifecycle and native-contract parallelism
+
+Status: implemented
+
+English | [中文](2026-07-26-code-mode-live-parallel-dispatch.zh.md)
+
+> Scope: the third PR of the Code Mode UI stack — the `tool/code-dispatch-start` event, per-sub-call running state in the web chat, and the bridge's scheduler reusing the native concurrency contract. Builds on the [host foundation](2026-07-26-code-dispatch-ui-foundation.md) and [chat sub-call rows](2026-07-26-code-mode-chat-subcall-rows.md); the native contract itself is owned by the [parallel tool-call note](2026-07-10-parallel-tool-call-execution.md).
+
+## Problem
+
+Two gaps remained after the first two PRs. Sub-call rows appeared only when each dispatch *settled* — while one ran, the UI showed nothing for it, so a slow sub-call read as a stalled parent. And the bridge serialized every binding call ("even `Promise.all` executes one at a time"), a placeholder from before tools carried concurrency metadata: `isConcurrencySafe` now exists, the loop scheduler already runs native siblings in bounded pools, and a Code Mode program awaiting three independent reads paid 3× the latency the native path would.
+
+## Decision
+
+**One lifecycle pair, one scheduling contract, shared with native.**
+
+- **Event pair**: `tool/code-dispatch-start` (parent/sub ids, name, normalized args) is appended when the scheduler actually starts a call — not at submission, so a queued call abandoned by run settlement logs nothing. The existing `tool/code-dispatch` settles the pair (same `subCallId`); every started call settles exactly once (aborts settle as `isError` outcomes through the pipeline). Timing = the two events' `time` fields. Both stay log-only; model context is untouched; format stays v0.
+- **Bridge scheduler**: submitted calls are classified at start time via `registry.executionMode` (the SAME fail-closed `isConcurrencySafe` contract the loop uses) and start strictly in submission order. One single-lane driver owns every ORDERED stage — the start append, `prepare` (pre-execute/guards), the head-of-line `finalize`/`finish` commit (post-execute + context deferral + settle append) — so ordered policy stages never overlap each other and only the around-dispatch/body stage runs concurrently, exactly the native loop's sequencing (`fillPool` awaits `startCall` then `commitReady`). Consecutive parallel-classified calls overlap up to `maxParallelSubCalls` (a `Config` field validated by the Loader schema AND re-validated at direct construction, default 10 — the loop scheduler's own default; `1` restores serial dispatch); an exclusive call drains the pool, runs alone, and holds its barrier until its COMMIT completes (post-execute included), like a native exclusive group. Run settlement aborts in-flight dispatches and abandons queued-unstarted ones (binding rejection, no events), then drains to quiescence — including a commit already mid-flight when the program returned — before the outer result closes the turn.
+- **Client**: `CodeSubCall` widens to `RunningToolCall | ToolResultNode` — a start event lands the running shape in the dispatch index (rows derive the running ring from the shape, exactly as for native in-flight calls), and its settle replaces the entry in place, preserving start order under parallel completion and carrying the start's `time` as `callTime` (duration source). A settle with no observed start (window cut mid-pair, or a pre-start-event log) appends directly, so old logs keep rendering.
+- **SDK prompt**: the model-facing "calls execute sequentially" sentence is replaced with the true contract (independent safe calls may overlap under `Promise.all`; dependent work sequences with `await`) — a model-visible change, re-recorded across every code-mode snapshot.
+
+## Alternatives considered
+
+**Unrestricted parallelism (let `Promise.all` overlap everything).** Rejected: writes could race; the native scheduler exists precisely because the tool, not the caller, owns the safety claim. One concurrency vocabulary across native and Code Mode was the settled requirement.
+
+**Emit the start event at submission instead of pool entry.** Rejected: a submission-time start would show queued-but-never-run calls as "running" and would force a third "abandoned" terminal event to reconcile the log. Start-at-entry keeps the invariant *started ⇔ settles exactly once* and needs no third event.
+
+**Reuse the loop scheduler's implementation directly.** Rejected: the loop schedules a fully-parsed batch with model-order result commitment; the bridge schedules an open-ended stream of submissions whose results return to the program (not the transcript), so only the *contract* (classification, pool, barriers) is shared, not the machinery.
+
+## Consequences
+
+Programs get native-grade latency for independent reads with no new model-side API — `Promise.all` simply works better, and prompt guidance changed accordingly. The web UI shows per-sub-call running rings live (fixture emits start/settle pairs; jsdom pins the running shape; the runtime spec pins in-place settlement, out-of-order completion, and callTime pairing). PR6 (trajectory/waterfall spans) can now draw truthful spans from the pair's timing. The spill PR (next) inherits the settle event as its single bounding point.

.agents/notes/implemented/feature/2026-07-26-code-mode-live-parallel-dispatch.zh.md

@@ -0,0 +1,32 @@
+# Agent Note：Code Mode 的实时分发生命周期，以及复用原生契约的并行执行
+
+Status: implemented
+
+[English](2026-07-26-code-mode-live-parallel-dispatch.md) | 中文
+
+> 范围：Code Mode UI 堆叠 PR（Pull Request）链的第三个 PR，涵盖 `tool/code-dispatch-start` 事件、web chat 中每个子调用的运行状态，以及桥接层调度器对原生并发契约的复用。构建在[宿主侧基础](2026-07-26-code-dispatch-ui-foundation.md)与 [chat 子调用行](2026-07-26-code-mode-chat-subcall-rows.md)之上；原生契约本身归[并行工具调用 Agent Note](2026-07-10-parallel-tool-call-execution.md) 所有。
+
+## 问题
+
+前两个 PR 之后仍留有两个缺口。子调用行过去只在每次分发*结算*（settle）后才出现：某次分发运行期间，UI 对它毫无展示，于是一个慢的子调用看上去就像父调用卡住了。而桥接层过去把每一次绑定调用都串行化（「即使 `Promise.all` 也一次只执行一个」），这是工具尚未携带并发元数据时留下的占位实现：如今 `isConcurrencySafe` 已经存在，agent loop（智能体循环）调度器早已在有界并发池中运行原生兄弟调用，而一个等待三个独立读取的 Code Mode 程序，付出的延迟却是原生路径的 3 倍。
+
+## 决策
+
+**一对生命周期事件，一份调度契约，与原生共用。**
+
+- **事件对**：`tool/code-dispatch-start`（父/子 id、名称、规范化参数）在调度器真正启动某个调用时才追加，而非在提交时，因此因 run 结算而被放弃的排队调用不会留下任何日志。既有的 `tool/code-dispatch` 结算该事件对（`subCallId` 相同）；每个已启动的调用恰好结算一次（中止也会作为 `isError` 结果经由流水线结算）。计时即这两个事件的 `time` 字段。两个事件都保持仅日志；模型上下文不受影响；格式保持 v0。
+- **桥接层调度器**：已提交的调用在启动那一刻经 `registry.executionMode` 分类（与 loop 所用完全相同的 fail-closed `isConcurrencySafe` 契约），并严格按提交顺序启动。所有有序阶段——start 事件追加、`prepare`（pre-execute/守卫）、队首 `finalize`/`finish` 提交（post-execute + 上下文延迟提交 + settle 事件追加）——由单一驱动车道独占执行，因此有序策略阶段彼此绝不重叠，只有 around-dispatch/工具体阶段并发运行，与原生 loop 的时序完全一致（`fillPool` 先 await `startCall` 再 `commitReady`）。连续被分类为可并行的调用可以重叠执行，上限为 `maxParallelSubCalls`（`Config` 字段，Loader schema 校验之外直接构造时也重新校验，默认值 10，即 loop 调度器自身的默认值；设为 `1` 即恢复串行分发）；独占调用则先排空池、独自运行，且其屏障保持到自身提交（含 post-execute）完成为止，与原生独占分组一致。run 结算时会中止仍在运行的分发，并放弃已排队未启动的分发（绑定调用被拒绝，不产生事件），随后排空到完全停稳——包括程序返回时已在途的提交——之后外层结果才结束该轮次。
+- **client 侧**：`CodeSubCall` 拓宽为 `RunningToolCall | ToolResultNode`：start 事件把运行中形状写入分发索引（行组件从该形状推导出运行指示环，与原生运行中的调用处理完全一致），其结算事件则原位替换该条目，即使并行完成也保持启动顺序不变，并把 start 事件的 `time` 作为 `callTime`（时长来源）带入。未观察到对应 start 的结算事件（窗口切在事件对中间，或日志录制于 start 事件引入之前）会直接追加，因此旧日志仍能照常渲染。
+- **SDK 提示词**：面向模型的「调用按顺序执行」一句替换为真实契约（相互独立的安全调用可以在 `Promise.all` 下重叠执行；相互依赖的工作以 `await` 顺序衔接）；这是模型可见的变更，每一份 code-mode 快照都已重新录制。
+
+## 曾考虑的替代方案
+
+**不加限制的并行（让 `Promise.all` 重叠一切）。** 否决：写操作可能产生竞态；原生调度器之所以存在，正是因为安全性声明归工具所有，而不归调用方。原生与 Code Mode 使用同一套并发词汇，是已敲定的要求。
+
+**在提交时而非入池时发出 start 事件。** 否决：提交即发 start 会把排了队却从未运行的调用显示成「运行中」，还得强行引入第三种「已放弃」终态事件才能使日志自洽。入池才发 start 保住了*已启动 ⇔ 恰好结算一次*这一不变式，且不需要第三种事件。
+
+**直接复用 loop 调度器的实现。** 否决：loop 调度的是一个完整解析好的批次，并按模型顺序提交结果；桥接层调度的则是一条开放式的提交流，其结果返回给程序，而不是进入 transcript（文本记录）。因此两者共享的只是*契约*（分类、池、屏障），而不是实现机制。
+
+## 后果
+
+程序不需要任何新的模型侧 API，独立读取就获得了原生级的延迟：`Promise.all` 直接变得更好用，提示词指引也随之修改。web UI 实时显示每个子调用的运行指示环：fixture（测试前置数据）发出成对的 start/settle 事件；jsdom 锁定运行中形状；运行时 spec 锁定原位结算、乱序完成与 callTime 配对。PR6（trajectory/waterfall 的 span）现在可以依据这对事件的计时绘制如实的 span。spill PR（下一个）则继承结算事件，作为自己唯一施加边界的位置。

.agents/notes/implemented/process/2026-06-17-ts-build-config.i18n.yaml

@@ -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
-2026-06-17-ts-build-config.md: 17036438b83a77f72b49f55abf29632af3f4ffef
-2026-06-17-ts-build-config.zh.md: 70e49c61deba418894a48be3016898d1d85c78a0
+2026-06-17-ts-build-config.md: 5bdfc5e170f12cd95a68f443ab8d02b16db554f3
+2026-06-17-ts-build-config.zh.md: 9f74fb6be9c8e00a070e27a609edf8421a2b5ca6

.agents/notes/implemented/process/2026-06-17-ts-build-config.md

@@ -43,7 +43,7 @@ In-package relative imports use explicit `.ts` specifiers.
 - Referenced package/vendor projects keep the same emit behavior as build, so typecheck refreshes their `lib/types` outputs instead of using a separate no-emit graph. Project-specific strictness changes live in the owning `packages/*/*/tsconfig.json` or `vendor/*/tsconfig.json`.
 - The no-emit aggregates disable `rewriteRelativeImportExtensions`; they emit nothing and include tests that import helpers across project-reference boundaries. Package/vendor emit projects keep the rewrite enabled.
 
-Composite projects keep their incremental build information inside their project-local `lib/` output. `pnpm run clean` derives live output directories from the root TypeScript project-reference graph, removes legacy root build information, and removes deleted `packages/*/*` directories that contain only known generated residue. It preserves `node_modules` for every package that still has a `package.json`, and refuses to remove a manifest-less directory containing unknown files. Build does not invoke clean automatically, so ordinary builds retain incremental state.
+Composite projects keep their incremental build information inside their project-local `lib/` output. `pnpm run clean` derives live output directories from the root TypeScript project-reference graph, removes legacy root build information, and removes deleted `packages/*/*` directories that contain only known generated residue. Before removing an existing target, it resolves the target's parent and refuses it if that resolved parent is outside the repository, so a symlinked project reference cannot redirect cleanup outside the checkout. It preserves `node_modules` for every package that still has a `package.json`, and refuses to remove a manifest-less directory containing unknown files. Build does not invoke clean automatically, so ordinary builds retain incremental state.
 
 The command orchestration shape is:
 

.agents/notes/implemented/process/2026-06-17-ts-build-config.zh.md

@@ -43,7 +43,7 @@ Status: implemented
 - 被引用的包/vendor 项目保持与构建相同的输出行为，因此类型检查会刷新它们的 `lib/types` 输出，而无需使用独立的 no-emit 图。项目特定的严格度变更放在各自的 `packages/*/*/tsconfig.json` 或 `vendor/*/tsconfig.json` 中。
 - 两个 no-emit 聚合禁用 `rewriteRelativeImportExtensions`；它们不输出任何文件，且包含跨 project-reference 边界导入 helper 的测试。包/vendor 的 emit 项目保持重写开启。
 
-复合项目将增量构建信息保存在各项目本地的 `lib/` 输出中。`pnpm run clean` 会根据根 TypeScript project-reference 图确定当前有效的输出目录，删除遗留的根目录构建信息，并删除已删除包留下且仅包含已知生成残留的 `packages/*/*` 目录。对于仍有 `package.json` 的每个包，该命令都会保留 `node_modules`；如果不含 `package.json` 的目录中存在未知文件，则拒绝删除。构建不会自动调用 clean，因此常规构建会保留增量状态。
+复合项目将增量构建信息保存在各项目本地的 `lib/` 输出中。`pnpm run clean` 会根据根 TypeScript project-reference 图确定当前有效的输出目录，删除遗留的根目录构建信息，并删除已删除包留下且仅包含已知生成残留的 `packages/*/*` 目录。在删除现有目标前，该命令会解析目标父目录的真实路径；如果解析后的父目录位于仓库之外，则拒绝删除，防止使用符号链接的 project reference 将清理操作重定向到工作副本之外。对于仍有 `package.json` 的每个包，该命令都会保留 `node_modules`；如果不含 `package.json` 的目录中存在未知文件，则拒绝删除。构建不会自动调用 clean，因此常规构建会保留增量状态。
 
 命令编排结构如下：
 

.agents/notes/implemented/process/2026-07-26-incremental-pr-base-retargeting.i18n.yaml

@@ -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
+2026-07-26-incremental-pr-base-retargeting.md: e2097ac4c32a926c8c0271df19dbc9796d0ed19d
+2026-07-26-incremental-pr-base-retargeting.zh.md: a6c94b66732b6c037fee1b0726b31ecb6f3b48c5

.agents/notes/implemented/process/2026-07-26-incremental-pr-base-retargeting.md

@@ -0,0 +1,27 @@
+# Agent Note: Retarget PR bases incrementally
+
+Status: implemented
+
+English | [中文](2026-07-26-incremental-pr-base-retargeting.zh.md)
+
+## Problem
+
+A PR base can advance while its current tip is being merged into the PR branch. Restarting from the newer tip discards completed conflict resolution and validation. Rewriting a merge that is already pushed also erases reviewable history.
+
+## Decision
+
+Each observed base tip gets its own merge checkpoint. If the base advances during the work, finish and validate the merge already in progress, commit it, and push it when the task authorizes a push. Only then fetch and merge the newer base in a separate merge commit. Never abandon, amend, rebase, or otherwise rewrite the earlier work.
+
+The root [AGENTS.md](../../../../AGENTS.md) states the standing order. The [stacked-PR landing skill](../../../skills/dsh-merging-stacked-prs/SKILL.md) applies it while retargeting dependent PRs, and the [stack review guide](../../../../docs/cookbook/responding-to-pr-review-on-a-stack.md) owns merging fixes down a stack.
+
+## Alternatives considered
+
+**Abort and restart from the newest base.** This discards resolved conflicts and completed validation, repeats work, and removes a useful recovery point.
+
+**Fold both base tips into one rewritten merge.** This hides the order in which conflicts were resolved and requires rewriting remote history if the first merge was pushed.
+
+## Consequences
+
+- A PR can carry several base-merge commits when its base advances repeatedly.
+- Completed work remains reviewable and recoverable instead of being discarded.
+- Merging a newer base changes the combined tree, so the relevant checks run again before the next push.

.agents/notes/implemented/process/2026-07-26-incremental-pr-base-retargeting.zh.md

@@ -0,0 +1,27 @@
+# Agent Note: 增量更新 PR 的 base 分支
+
+Status: implemented
+
+[English](2026-07-26-incremental-pr-base-retargeting.md) | 中文
+
+## 问题
+
+将 PR（Pull Request）的 base 分支当前顶端提交合入 PR 分支的过程中，base 分支可能继续前移。若改从新的顶端提交重新开始，就会丢弃已经完成的冲突解决和验证工作。重写已经推送的合并还会抹去可供评审的历史记录。
+
+## 决策
+
+每次观察到的 base 分支顶端提交都保留为独立的合并检查点。如果处理期间 base 分支继续前移，先完成并验证正在进行的合并，再将其提交；任务授权推送时，还要完成推送。完成这些步骤后，才能拉取较新的 base，并通过单独的合并提交将其合入。绝不放弃先前工作，也不通过 amend、rebase 或其他方式重写它。
+
+根 [AGENTS.md](../../../../AGENTS.md) 规定了这项常设指令。[堆叠 PR 落地 skill（技能）](../../../skills/dsh-merging-stacked-prs/SKILL.md)在调整依赖 PR 的 base 时执行这一规则，[堆叠评审指南](../../../../docs/cookbook/responding-to-pr-review-on-a-stack.md)则负责说明如何将修复沿堆叠向下合并。
+
+## 曾考虑的替代方案
+
+**中止当前工作，改从最新 base 重新开始。** 这会丢弃已经解决的冲突和完成的验证，重复劳动，并失去一个有用的恢复点。
+
+**重写为一次同时包含两个 base 分支顶端的合并。** 这会掩盖冲突解决的顺序；如果第一次合并已经推送，还必须重写远程历史。
+
+## 后果
+
+- PR 的 base 多次前移时，这个 PR 可以包含多个用于合并 base 的提交。
+- 已完成的工作不会被丢弃，而是保持可供评审和恢复。
+- 合入较新的 base 会改变合并后的文件树，因此相关检查会在下一次推送前重新运行。

.agents/skills/dsh-merging-stacked-prs/SKILL.md

@@ -20,6 +20,7 @@ Given `A ← B ← C` landing on `master`:
 2. **Retarget PR B, refresh it, then merge it — keeping its branch.**
    - `gh pr edit B --base master` (now that A is in master, B's base becomes master).
    - Merge the new master *into* branch B: check out B, `git fetch origin`, `git merge origin/master` — merge `origin/master`, not local `master`, because `gh pr merge` updated only GitHub and the local branch is stale — resolve any conflicts here, and push. This makes B current and surfaces conflicts in the working branch where they can be tested — not as a surprise at the GitHub merge.
+   - If `origin/master` moves during that work, finish and push the in-progress merge, then fetch and merge the newer tip in a separate commit. Never abandon or rewrite the earlier work ([rationale](../../notes/implemented/process/2026-07-26-incremental-pr-base-retargeting.md)).
    - `gh pr merge B --merge` — still no `--delete-branch` (PR C bases on branch B).
 
 3. **Retarget PR C, refresh it, then merge it — keeping its branch.** Same steps: `gh pr edit C --base master`, fetch and merge `origin/master` into branch C, resolve conflicts there and push, then `gh pr merge C --merge` without `--delete-branch`.