Merge branch 'master' into fix/sidebar-document-preview-polish

.agents/notes/archived/architecture/2026-09-02-durable-image-offload.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 .agents/notes/archived/architecture/2026-09-02-durable-image-offload.md
+2026-09-02-durable-image-offload.md: db03e9f52499c1790b3a0c03853f3aea23315bec
+2026-09-02-durable-image-offload.zh.md: 2b6d0a4aa65f9a39b88888506f7e85b4cfff48cf

.agents/notes/archived/architecture/2026-09-02-durable-image-offload.md

@@ -0,0 +1,56 @@
+# Agent Note: Durable image offload by surface replacement
+
+Status: implemented
+Archived: 2026-09-10
+
+English | [中文](2026-09-02-durable-image-offload.zh.md)
+
+## Problem
+
+Request-size image offload was recomputed from scratch on every request. Each route collected every image occurrence on the derived surface, oldest first, and once the accumulated bytes exceeded its budget it rounded the excess up to a whole removal quantum and replaced that many oldest occurrences with placeholder text, per the [unified image request pipeline](../feature/2026-08-20-unified-image-request-pipeline.md). Nothing remembered where the previous request stopped; the prefix was stable only because the arithmetic over an append-only history repeated itself.
+
+That stability failed wherever the arithmetic inputs moved. The [Files inline fallback](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md) rebuilt a request under a 20 MiB inline budget with a 10 MiB quantum, offloading far more images for that one request, and the next request in file mode brought them back. The pi-ai route used a quantum of one byte, so its prefix moved on almost every request. Compaction lowered the total and returned previously offloaded images. A route switch moved every step boundary. Each move changed the model-visible prefix and invalidated the provider cache prefix.
+
+The same recomputation broke the repository invariant that model-visible input is reconstructable from the session log. Which representation was dispatched, the exact derived request-version byte lengths, and the route budgets and quanta were runtime or configuration facts that never entered the log; `request/header` records call config, system prompt, and tools only. Provider usage anchors token totals and cannot recover the image set, and the [route-priced estimate](../../archived/feature/2026-08-24-route-priced-image-request-pressure.md) documented that it did not reproduce the fallback budget. No consumer could pair a logged assistant response with the image set its request carried.
+
+## Decision
+
+The offload of an image occurrence is a durable surface fact recorded the way compaction records its reductions: a `surfaceOp: replace` node.
+
+**Marked copies on the surface.** `ImageBlock` gains `offloaded?: true`. An offloaded occurrence lives in a replacement event of the same type as the node that carried it (`user/message` or `tool/result`) whose content is a copy of the original with those blocks marked; `sourceEventSeqs` points at the replaced node. The session core, its event map, its derivation, and its validation are unchanged. `deriveMessages()` sends the marked block; serialization renders every marked block as `offloadedImageText` with the currently resolved access path and prepares only retained occurrences.
+
+**Only advances.** A replacement never reverts. When a budget grows, a route changes, or compaction lowers the total, the marked copies stay, so the model-visible prefix and the provider cache prefix move only forward.
+
+**Adapters project, never decide.** An image-capable route enforces an `LlmImageRequestBudget` (`representation`, `maxBytes`, `maxImages`, both quanta) over the retained occurrences' exact request-version bytes. When they still exceed the budget, in file mode, under the inline fallback's tighter budget, or under the pi-ai bound, the adapter fails the attempt with `IMAGE_OFFLOAD_REQUIRED` and `LlmFailure.offloadImages` naming how many more oldest occurrences must be offloaded, computed with the shared `requiredImageOffload()`. Nothing plans an offload before dispatch.
+
+**Recovery owned by `dsh-compaction-image-offload`.** Image offload is compaction in another capacity dimension: the provider rejects a request, durable history is reduced, the step retries. The executor is a sibling of `compaction-tool-result-pruner` in the compaction group and listens on the `agent/request-error` waterfall. On `IMAGE_OFFLOAD_REQUIRED` it walks the surface in model request order, marks the first `offloadImages` retained occurrences, and for each node that carried one appends the seam's `compaction/prune` shadow price followed by the marked copy, then returns the `retry` action without spending the provider retry budget or logging `llm/retry`. Assistant nodes carry model output, not input images, and are skipped. When nothing remains to offload it delegates and the failure reaches ordinary recovery. The loop re-runs the step over the replaced surface and logs the fresh `request/header` every surface replacement gets; the agent loop is unchanged.
+
+**Token accounting.** `priceImages` receives the surface's `ImageBlock`s and prices a marked one as its placeholder text; the DeepSeek and replay pricing no longer reproduce any offload arithmetic. The meter needs no new state: the `compaction/prune` event and the replacement reprice the node like a tool-result prune.
+
+**Other consumers.** Compaction reconstructs each selected event through `deriveEventMessage()` and sees the marks. Resume, fork, and replay reproduce the surface from the log. Text-only routes keep their separate whole-history substitution.
+
+## Alternatives considered
+
+**Keep recomputing the offload point per request.** Stable only while the arithmetic inputs held still; the inline fallback, the pi-ai quantum, compaction, and route switches all moved the prefix, and no consumer could reconstruct a historical request's image set.
+
+**Record each request's projection outcome as a log-only event.** Restores reconstructability but not stability: the recorded outcome is not a decision input, so every oscillation still happens and the log merely documents it, with two sources of truth for the offloaded set that can disagree.
+
+**Log the full projected request body.** Everything except the offload decision is already derivable; repeating the history per request grows the log quadratically to record one position.
+
+**A log-only `image/offload` watermark event applied by session derivation.** An earlier revision recorded the offload point as a position (event seq plus block path) and had `Session.deriveMessages()` mark every occurrence at or before it. That gave the session a second mechanism for changing model-visible history next to `surfaceOp: replace`, with its own validation, fold, cache invalidation, and required-on-read event, and placed the logic in the core instead of on the plugin that decides. Issue #3041 asks that a permanent eviction use a surface-changing event rather than a request-projection event; because the offload never reverts, it is one.
+
+**Let each adapter append the replacement.** The adapter owns the budgets but not the session surface; a surface change appended below the loop lets two adapters define the surface differently. Adapters report the count they need instead.
+
+**Plan the offload before dispatch, in the loop or in a plugin.** The loop knows the exact prepared route before deriving each request, so planning there never spends a failed attempt; but it puts a route-specific policy into the one component every profile shares, changes the documented step order, and bypasses the `agent/request-error` waterfall that context-overflow compaction and retry already use for the same repair-and-retry pattern. A pre-step plugin keeps the loop unchanged but cannot see the step's own messages or the first request's route, so the failure path stays necessary, and the plan needs every route to declare its budget on its model info. Handling only the failure costs one attempt per quantum crossing (64 MiB in DeepSeek file mode, 20 MiB on pi-ai) and matches the planned direction: routes will stop checking sizes locally, send every image, and the provider will report what it cannot cache, which is exactly a failure naming an offload point.
+
+**Keep a transient extra offload for the inline fallback and exact-byte overflow.** Would have sent an unlogged projection in exactly the cases the invariant exists for; the failure-and-advance path costs one serialization attempt and keeps every dispatched request derivable from the log.
+
+## Consequences
+
+Offloaded images never return automatically when budgets grow, a larger route is selected, or compaction lowers the total; recovery is the read-only path in the placeholder, which the model uses deliberately. A Files outage or a temporary switch to a small-budget route offloads permanently; both are accepted for the same reason. Each offload copies the carrying node's message into the log; images are references, so the copy is the node's text plus block metadata.
+
+Every dispatched request's image set is determined by the log alone, across file mode, inline fallback, resume, fork, retry, and compaction, and the provider cache prefix no longer oscillates. The execution-world access path embedded in placeholder and handle text is still resolved at serialization time; that gap exists for retained images too and belongs to a separate decision about recording the execution-world mapping. The route-local byte checks are provisional: once the provider reports uncacheable images itself, `requiredImageOffload()` and the route budgets go away while the marked copies and the recovery branch stay as they are.
+
+## Testing
+
+`packages/llm/llm/tests/content.spec.ts` pins arbitrary-depth image traversal, the offload count including the 129-to-64 MiB quantum example, and placeholder projection. `packages/compaction/compaction-image-offload/tests/image-offload.spec.ts` pins the `IMAGE_OFFLOAD_REQUIRED` replace-and-retry path with its `compaction/prune` shadow prices and without a retry event, the untouched original node, request-order counts after an earlier surface replacement, and the exhausted case that delegates downstream. Adapter specs pin placeholder projection, prepared-only-retained reads, and the exact-byte failure with its count; `route-pricing.spec.ts` pins placeholder pricing for a marked replacement. The `inline-image-prompt` TypeScript SDK snapshot replays an authored `IMAGE_OFFLOAD_REQUIRED` attempt through the shipped profile and pins the replacement and the retried request.

.agents/notes/archived/architecture/2026-09-02-durable-image-offload.zh.md

@@ -0,0 +1,56 @@
+# Agent Note: 以表层替换实现持久的图片 offload
+
+Status: implemented
+Archived: 2026-09-10
+
+[English](2026-09-02-durable-image-offload.md) | 中文
+
+## 问题
+
+请求级图片 offload 过去在每次请求时都从头重算。每条路由按最老优先的顺序收集派生表层上的全部图片出现位置，累计字节一旦超过预算，就把超出部分向上取整到整个删除量子，把这么多最老的出现位置替换为占位文本，见[统一图片请求管线](../feature/2026-08-20-unified-image-request-pipeline.zh.md)。没有任何东西记住上一次请求停在哪里，前缀稳定只是因为对 append-only 历史做同样的算术会得到同样的结果。
+
+算术的输入一动，这种稳定就失效。[Files 内联回退](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md)会用 20 MiB 内联预算和 10 MiB 量子重建请求，那一次省略多得多的图片，下一次 file 模式的请求又把它们带回来。pi-ai 路由的量子是一个字节，前缀几乎每次请求都会移动。compaction 降低总量，让先前省略的图片回归。切换路由会移动每个台阶边界。每一次移动都改变模型可见前缀，并让 provider 的缓存前缀失效。
+
+同样的重算也破坏了仓库不变量：模型可见输入必须能从 session log 重建。实际发出的表示方式、派生请求版本的精确字节长度、路由预算和量子都是运行时或配置事实，从不进入日志，`request/header` 只记录调用配置、系统提示词和工具。provider usage 只锚定 token 总量，恢复不了图片集合，[按路由定价的估计](../../archived/feature/2026-08-24-route-priced-image-request-pressure.md)也写明它不复现回退预算。没有任何消费方能把一条已记录的助手响应和它的请求携带的图片集合配对。
+
+## 决定
+
+一个图片出现位置的省略是一条持久的表层事实，记录方式和 compaction 记录它的缩减一样：一个 `surfaceOp: replace` 节点。
+
+**表层上带标记的副本。** `ImageBlock` 新增 `offloaded?: true`。被省略的出现位置存在于一条与原承载节点同类型的替换事件里（`user/message` 或 `tool/result`），内容是原消息的副本，只把那些块打上标记；`sourceEventSeqs` 指向被替换的节点。session 核心、它的事件表、派生和校验都不变。`deriveMessages()` 原样发送带标记的块；序列化把每个带标记的块渲染为带当前已解析访问路径的 `offloadedImageText`，只准备保留的出现位置。
+
+**只前进。** 替换永不回退。预算变大、路由切换或 compaction 降低总量时，带标记的副本留在原地，所以模型可见前缀和 provider 缓存前缀只向前移动。
+
+**adapter 只投影，不决定。** 支持图片的路由按保留的出现位置的精确请求版本字节执行一个 `LlmImageRequestBudget`（`representation`、`maxBytes`、`maxImages` 与两个量子）。当它们仍超过预算，无论是 file 模式、内联回退更紧的预算还是 pi-ai 上限，adapter 都以 `IMAGE_OFFLOAD_REQUIRED` 让本次尝试失败，并在 `LlmFailure.offloadImages` 中用共享的 `requiredImageOffload()` 算出还需省略多少最老的出现位置。发送前没有任何规划。
+
+**恢复归 `dsh-compaction-image-offload`。** 图片省略是 compaction 在另一个容量维度上的实例：provider 拒绝请求，持久历史被缩减，step 重试。执行器是 compaction 组里 `compaction-tool-result-pruner` 的兄弟包，监听 `agent/request-error` waterfall。收到 `IMAGE_OFFLOAD_REQUIRED` 时，它按模型请求顺序遍历表层，给前 `offloadImages` 个保留的出现位置打标记，为每个承载了其中任一位置的节点先追加 seam 的 `compaction/prune` 影子价格，再追加带标记的副本，然后返回 `retry` 动作，不占提供方重试预算，也不记录 `llm/retry`。assistant 节点承载的是模型输出而不是输入图片，直接跳过。没有可省略的出现位置时向下游委托，失败进入普通恢复路径。循环在替换后的表层上重跑该 step，并像每次表层替换后一样记录新的 `request/header`；agent loop 不变。
+
+**token 记账。** `priceImages` 接收表层的 `ImageBlock`，把带标记的按占位文本定价；DeepSeek 和 replay 的定价不再复现任何 offload 算术。meter 不需要新状态：`compaction/prune` 事件加替换节点，和工具结果剪枝一样重新为该节点定价。
+
+**其他消费方。** compaction 通过 `deriveEventMessage()` 重建每个选中事件，看得到标记。resume、fork 和重放从日志复现表层。纯文本路由保留各自的全历史替换。
+
+## 考虑过的替代方案
+
+**继续每次请求重算 offload 位置。** 只在算术输入不动时稳定，内联回退、pi-ai 量子、compaction 和路由切换都会移动前缀，且没有消费方能重建历史请求的图片集合。
+
+**用 log-only 事件记录每次请求的投影结果。** 恢复了可重建性但没有稳定性：记录的结果不是决策输入，每一种抖动照旧发生，日志只是把它记下来，且省略集合有两个可能不一致的事实来源。
+
+**记录完整的投影后请求体。** 除 offload 决定外一切都已可派生，为记录一个位置而每次请求重复整段历史会让日志平方级增长。
+
+**由 session 派生应用的 log-only `image/offload` 水位事件。** 更早的一版把 offload 点记成一个位置（事件序号加块路径），由 `Session.deriveMessages()` 给位于它及之前的每个出现位置打标记。这让 session 在 `surfaceOp: replace` 之外多出第二种改变模型可见历史的机制，带着自己的校验、折叠、缓存失效和读取时必须识别的事件，并且把逻辑放进了核心而不是做决定的插件。issue #3041 要求永久淘汰使用改变表层的事件而不是请求投影事件；省略永不回退，它就是永久淘汰。
+
+**让各个 adapter 自己追加替换。** adapter 拥有预算，但不拥有会话表层；在循环之下追加表层变更会让两个 adapter 对表层做出不同定义。adapter 改为上报它需要的数量。
+
+**在发送前规划省略，无论放在循环里还是插件里。** 循环在派生每个请求之前就知道精确的已准备路由，在那里规划永远不会多花一次失败的尝试；但这会把一条路由专属的策略放进所有 profile 共用的那个组件，改变已记录的 step 顺序，还绕过了上下文溢出 compaction 和重试已经在用的同一套 `agent/request-error` waterfall。pre-step 插件不改循环，但看不到 step 自己的消息和第一个请求的路由，失败路径仍然必需，而且规划要求每条路由在模型信息上声明预算。只处理失败的代价是每越过一个量子多一次尝试（DeepSeek file 模式 64 MiB，pi-ai 20 MiB），并且和既定方向一致：路由将不再本地检查大小，全部发送，由 provider 报告无法缓存的部分，那正是一个指明省略点的失败。
+
+**为内联回退和精确字节溢出保留临时的额外省略。** 恰好会在不变量所针对的场景发送未记录的投影；失败再推进的路径只多花一次序列化尝试，且让每个已发出请求都可由日志派生。
+
+## 后果
+
+预算变大、选中更大的路由或 compaction 降低总量时，被省略的图片不会自动回归；恢复手段是占位文本中的只读路径，模型需要时主动使用。一次 Files 故障或临时切到小预算路由会永久省略，两者出于同一理由被接受。每次省略会把承载节点的消息复制一份进日志；图片只是引用，复制的是该节点的文本和块元数据。
+
+每个已发出请求的图片集合都仅由日志决定，覆盖 file 模式、内联回退、resume、fork、重试和 compaction，provider 缓存前缀不再来回变化。占位文本和句柄文本里嵌入的执行世界访问路径仍在序列化时解析；这个缺口对保留的图片同样存在，属于另一个关于记录执行世界映射的决定。路由本地的字节检查是过渡实现：provider 自己报告无法缓存的图片之后，`requiredImageOffload()` 和路由预算会被删掉，带标记的副本和恢复分支保持不变。
+
+## 测试
+
+`packages/llm/llm/tests/content.spec.ts` 钉住任意深度的图片遍历、包括 129 到 64 MiB 量子示例在内的省略计数，以及占位投影。`packages/compaction/compaction-image-offload/tests/image-offload.spec.ts` 钉住 `IMAGE_OFFLOAD_REQUIRED` 替换并重试、带 `compaction/prune` 影子价格且不追加重试事件的路径、原节点保持不变、更早一次表层替换之后的请求顺序计数，以及向下游委托的耗尽情况。adapter 测试钉住占位投影、只读取保留图片以及带数量的精确字节失败；`route-pricing.spec.ts` 钉住带标记替换的占位定价。`inline-image-prompt` TypeScript SDK 快照通过发布的 profile 重放一次手工编写的 `IMAGE_OFFLOAD_REQUIRED` 尝试，钉住替换和重试后的请求。

.agents/notes/archived/manifest.json

@@ -358,6 +358,9 @@
     "architecture/2026-09-01-streamed-tool-call-identity.i18n.yaml": "sha256:5193bf7429a0300f743ea7e6087c3175a7d28c2c468ca1cfd1f77702b68b6399",
     "architecture/2026-09-01-streamed-tool-call-identity.md": "sha256:303433b95c22571bf6a801da604b42a4c7231f5f6d07daeb518354f0d7019141",
     "architecture/2026-09-01-streamed-tool-call-identity.zh.md": "sha256:53728d3ee06b29d6023e28024fb3ab6fa4a7473ae5e55368cd19690719d120f3",
+    "architecture/2026-09-02-durable-image-offload.i18n.yaml": "sha256:d06155597e7df4ca4c15a2902ab664d69cc2fde1e7963e61543d20a49930d2b9",
+    "architecture/2026-09-02-durable-image-offload.md": "sha256:02b7becba3dd73a4e627b85705e8bd5dd55978cb041f4d3a0fd1420a76e4de92",
+    "architecture/2026-09-02-durable-image-offload.zh.md": "sha256:a0464b8272086d663ee698efd0b2edd6224d40770d7a8fd88a6d8bfe030b2a1d",
     "architecture/2026-09-02-protocol-specific-model-listing-discovery.i18n.yaml": "sha256:ba8a62a9fa3263de4162709cb1aaf81f6ad4feee5011d7a5d8b24eb92ad81d30",
     "architecture/2026-09-02-protocol-specific-model-listing-discovery.md": "sha256:9db9c47559c17f0e2931b9e646f1d2d17b590df4d5699d4e779ed6d7f65fc371",
     "architecture/2026-09-02-protocol-specific-model-listing-discovery.zh.md": "sha256:9a266590d49093a097bcdb5d09865757cdd07625cfaaddce1303046d580551f1",

.agents/notes/implemented/architecture/2026-07-10-single-file-executable-sdk-runtime-distribution.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 .agents/notes/implemented/architecture/2026-07-10-single-file-executable-sdk-runtime-distribution.md
-2026-07-10-single-file-executable-sdk-runtime-distribution.md: aba6f09a3d470abdf51683a1e7efd15803ff01d2
-2026-07-10-single-file-executable-sdk-runtime-distribution.zh.md: ffa9071d738fb658f3423d2165e4fc4e004d6a23
+2026-07-10-single-file-executable-sdk-runtime-distribution.md: 665364e6d39a78f7ac497198f18ed9aa160a4cbd
+2026-07-10-single-file-executable-sdk-runtime-distribution.zh.md: 8b39df71c615dc59f3ef1bf622a5736a70b085b3

.agents/notes/implemented/architecture/2026-07-12-agent-scope-runtime-design.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 .agents/notes/implemented/architecture/2026-07-12-agent-scope-runtime-design.md
-2026-07-12-agent-scope-runtime-design.md: ca300d4eeeab878a4e41b8e68a669be418617181
-2026-07-12-agent-scope-runtime-design.zh.md: 870690d6ace9fefd859557a2e73e88b9b1da6206
+2026-07-12-agent-scope-runtime-design.md: b9b2c837f560c2b8b98da37e3656caf298b705e2
+2026-07-12-agent-scope-runtime-design.zh.md: 4b94d1b8449bd619c47f6bee0712ec8ed0b39593

.agents/notes/implemented/architecture/2026-07-12-agent-scope-runtime-design.md

@@ -60,6 +60,8 @@ Product helpers therefore construct the carrier and pass the domain subject sepa
 
 A Cordis waterfall is middleware-style dispatch. Each listener receives `next()`: calling it delegates to the remaining listeners and base operation, while returning without it short-circuits or replaces the downstream result. Waterfalls power prompt assembly and tool policy; ordinary emit events notify synchronously, and parallel events await all listeners without a veto result.
 
+<a id="scope-routing-one-opaque-key-selects-one-layer"></a>
+
 ## Scope routing: one opaque key selects one layer
 
 The scope package implements the smallest object needed for Cordis routing. Its carrier holds only a composed service filter and scope predicate, while the package records the opaque key privately and exposes the scope fiber's quiescent disposer separately.
@@ -117,30 +119,28 @@ Publication admits and announces resources in the order required by observers:
 1. Enter the session.
 2. Enter the agent.
 3. Announce `session/created`.
-4. Announce `agent/created`.
-5. Enable public driving.
-6. Emit `agent/session-start`.
-7. Start the driver.
+4. Await serial `agent/created` listeners.
+5. Release queued input to the driver.
 
-The agent never drives before both registries and creation notifications agree. A synchronous listener may veto or dispose an owner; the transaction records publication in progress and waits for that callback stack to unwind before teardown continues. Every creation announcement that begins has a matching disposal announcement during rollback.
+The agent never drives before both registries and creation listeners finish. A listener may reject or dispose an owner; the transaction retains the scope and session until dispatch settles before teardown continues. Every creation announcement that begins has a matching disposal announcement during rollback. The [awaited creation decision](2026-09-09-awaited-agent-creation.md) owns asynchronous initializer timing.
 
-The sequence diagram isolates the non-obvious race: a synchronous creation listener can request disposal while the publication call stack still owns both registry entries. Teardown must deactivate immediately but wait for that stack to unwind before stopping and detaching anything.
+A creation listener can request disposal while publication still owns both registry entries. Teardown deactivates immediately and waits for the awaited dispatch before stopping and detaching anything.
 
 ```mermaid
 sequenceDiagram
   participant Tx as AgentCreationTransaction
   participant Registries
-  participant Listener as Synchronous listener
+  participant Listener as Creation listener
   participant Driver
 
   Tx->>Tx: mark publication in progress
   Tx->>Registries: announce agent/created
-  Registries->>Listener: invoke inside the same call stack
+  Registries->>Listener: await listener
   Listener->>Tx: dispose reentrantly
   Tx->>Tx: deactivate, teardown waits for publication
   Tx-->>Listener: disposal request accepted
-  Listener-->>Registries: return
-  Registries-->>Tx: announcement unwound
+  Listener-->>Registries: settle
+  Registries-->>Tx: dispatch settled
   Tx->>Tx: resolve publication settlement
   Tx->>Driver: stop and drain
   Tx->>Registries: detach agent, then session
@@ -151,7 +151,7 @@ sequenceDiagram
 
 Every teardown request joins one memoized path. The order is:
 
-1. Deactivate creation or driving and let synchronous publication finish.
+1. Deactivate creation or driving and await creation dispatch.
 2. Stop and drain the driver, discarding any injection that remains pending.
 3. Detach the agent.
 4. Detach the session.
@@ -160,6 +160,8 @@ Every teardown request joins one memoized path. The order is:
 
 This order lets final agent and session events use the matching scoped listeners and keeps persistence observers attached through the final flush. Scope disposal comes last because registration revocation is the externally visible lifetime boundary.
 
+<a id="session-append-materialize-validate-commit-notify"></a>
+
 ## Session append: materialize, validate, commit, notify
 
 Session events cross a durable boundary, so append owns their data. The rest of the algorithm uses one attached entry and one commit point.
@@ -232,6 +234,8 @@ This is a trusted same-process extension point, not an authority boundary. A lis
 
 Scope solves the real isolation problem directly. Structured-output contributions register in the child's exact scope, while PTC mode derives its transport and SDK from the same resolved tool view. A second named-protection system would need another ownership and collision rule across arbitrary schema providers—including providers that intentionally contribute duplicate names—without creating a new trust boundary.
 
+<a id="structured-output-commits-only-authoritative-outcomes"></a>
+
 ### Structured output commits only authoritative outcomes
 
 Structured output combines child-scoped composition with a two-phase execution commit. The child registers its `structured_output` tool and instruction before publication; a trusted assembly listener may transform those ordinary contributions and is responsible for preserving the protocol if the child is expected to complete. The tool body validates a candidate and stages it by the current `ToolExecution`, but successful capture is decided only by immutable `tools/result` observations.
@@ -244,6 +248,8 @@ Once a value is pending or committed, a scoped monotonic guard denies later tool
 
 Pure PTC mode's registry contribution omits `structured_output` from native wire schemas and exposes it through the generated SDK. The assembly waterfall may deliberately change that presentation; execution still validates against the child-scoped definition, and the listener owns the consistency of any alternate model-visible route it creates.
 
+<a id="three-execution-boundaries-are-deliberately-one-way"></a>
+
 ### Three execution boundaries are deliberately one-way
 
 Prompt assembly is intentionally cooperative, but three execution facts need one-way settlement after their extensible stages:
@@ -294,19 +300,21 @@ Start resolves only after `initialize` and `newSession` succeed. Abort, spawn fa
 
 Worker and child-process bridges need more state than same-process registries because messages, process death, and cleanup can settle independently. Their state is organized around those real facts rather than duplicate cancellation protocols.
 
+<a id="workflow-children-are-pending-starts-or-published-records"></a>
+
 ### Workflow children are pending starts or published records
 
 The workflow host keeps pending provider-start promises and published child records. A child moves from pending to published only when async `SubagentRuntime.start()` fulfills; rejected starts clean their partial provider work and produce no child lifecycle pair.
 
-One host-owned AbortController supplies the required signal to pending and live children. Closing workflow admission aborts that signal, so there is no duplicate `ChildCancel` worker RPC or explicit host-side `run.cancel()` fanout. Quiescence waits for both pending starts and published child disposal.
+One host-owned AbortController supplies the required signal to pending and live children. Closing workflow admission aborts that signal; quiescence waits for both pending starts and published child disposal. [Workflow sandbox reuse](2026-09-13-workflow-ptc-sandbox-reuse.md) owns PTC process cancellation and the absence of a separate workflow cleanup timer.
 
-The worker boundary still serializes requests and outcomes. The host retains first-terminal-outcome arbitration, exact child accounting, worker-death handling, grace termination, late/duplicate message rejection, and bounded cleanup because result receipt, worker exit, and child quiescence are genuinely independent facts.
+PTC serializes requests and outcomes and owns process termination. The workflow adapter retains terminal-outcome arbitration and child ownership because program settlement, process exit and child quiescence remain independent facts.
 
 ### Terminal result and physical cleanup remain separate
 
-The workflow result records the first accepted terminal outcome according to the public precedence rules. Cleanup can continue after that result is chosen: live children still need disposal, a worker still needs termination, and a slow external backend may outlive the configured grace bound.
+The workflow result records the first accepted terminal outcome according to the public precedence rules. Choosing that outcome does not release resources: the PTC process and live children still need cleanup, and child disposal must fulfill its provider contract.
 
-Public disposal claims its memoized promise before invoking callbacks. Worker death closes admission before processing any queued late child request, synthesizes missing lifecycle ends, and starts child/process cleanup without rewriting an outcome already claimed.
+Public disposal joins one cleanup operation. Run settlement closes child admission, synthesizes missing lifecycle ends and cleans up children without rewriting an outcome already claimed.
 
 ### ACP prompt settlement does not depend on update delivery
 
@@ -334,7 +342,7 @@ The plugin does not police trusted setup by scanning registries or reject prompt
 
 The event catalog, service catalog, producer/consumer matrix, configuration catalog, module graph, tool catalog, type-equivalence blocks, and scoped-event resolver map are generated or freshness-gated from source. The [TypeScript semantic-gates Agent Note](../../archived/process/2026-07-14-typescript-program-backed-semantic-gates.md) owns Program construction, semantic event discovery, and resolver-generation rules.
 
-Behavioral tests pin scoped routing and disposal, final-entry collision cleanup, publication rollback, ordered quiescence, durable pre/post-commit behavior, live tool filtering across presentation and execution, cooperative prompt assembly, structured-output commit in native and PTC mode, async subagent startup and signal cancellation, worker terminal arbitration, ACP settlement, and process teardown.
+Behavioral tests pin scoped routing and disposal, final-entry collision cleanup, publication rollback, ordered quiescence, durable pre/post-commit behavior, live tool filtering across presentation and execution, cooperative prompt assembly, structured-output commit in native and PTC mode, async subagent startup and signal cancellation, workflow terminal arbitration, ACP settlement, and process teardown.
 
 ## Alternatives considered
 

.agents/notes/implemented/architecture/2026-07-12-agent-scope-runtime-design.zh.md

@@ -119,30 +119,28 @@ Setup 接收完整的子上下文和确切的未发布 Agent，可以等待插
 1. 将会话写入注册表。
 2. 将 agent 写入注册表。
 3. 宣告 `session/created`。
-4. 宣告 `agent/created`。
-5. 启用公开驱动。
-6. 发射 `agent/session-start`。
-7. 启动 driver。
+4. 等待串行 `agent/created` 监听器。
+5. 向驱动器释放已排队输入。
 
-Agent 在两个注册表和创建通知都达成一致之前绝不驱动。同步监听器可以否决或 dispose 一个所有者；事务记录发布进行中，并等待该回调栈展开后再继续拆除。每个已开始的创建宣告在回滚期间都有匹配的销毁宣告。
+Agent 在两个注册表与创建监听器都完成前绝不驱动。监听器可以拒绝或 dispose 一个所有者；事务保留作用域与会话，等待分发结算后再继续拆除。每个已开始的创建宣告在回滚期间都有匹配的销毁宣告。[可等待创建决策](2026-09-09-awaited-agent-creation.zh.md)拥有异步初始化器时序。
 
-以下序列图隔离了非显而易见的竞态：同步创建监听器可以在发布调用栈仍拥有两个注册表条目时请求 dispose。拆除必须立即停用，但要等待该栈展开后才停止和分离任何东西。
+创建监听器可以在发布仍拥有两个注册表条目时请求 dispose。Teardown 会立即停用，并等待所调用的异步分发完成后才停止和分离资源。
 
 ```mermaid
 sequenceDiagram
   participant Tx as AgentCreationTransaction
   participant Registries
-  participant Listener as Synchronous listener
+  participant Listener as Creation listener
   participant Driver
 
   Tx->>Tx: mark publication in progress
   Tx->>Registries: announce agent/created
-  Registries->>Listener: invoke inside the same call stack
+  Registries->>Listener: await listener
   Listener->>Tx: dispose reentrantly
   Tx->>Tx: deactivate, teardown waits for publication
   Tx-->>Listener: disposal request accepted
-  Listener-->>Registries: return
-  Registries-->>Tx: announcement unwound
+  Listener-->>Registries: settle
+  Registries-->>Tx: dispatch settled
   Tx->>Tx: resolve publication settlement
   Tx->>Driver: stop and drain
   Tx->>Registries: detach agent, then session
@@ -153,7 +151,7 @@ sequenceDiagram
 
 每个拆除请求加入一条记忆化路径。顺序为：
 
-1. 停用创建或驱动，让同步发布完成。
+1. 停用创建或驱动，并等待创建分发。
 2. 停止并排空 driver，丢弃仍处于待处理状态的注入。
 3. 分离 agent。
 4. 分离会话。
@@ -308,15 +306,15 @@ Worker 和子进程桥接比同进程注册表需要更多状态，因为消息
 
 工作流宿主保持待定的提供方 start promise 和已发布的子级记录。子级仅在异步 `SubagentRuntime.start()` 兑现时才从待定变为已发布；被拒绝的 start 清理其部分提供方工作且不产生子级生命周期对。
 
-一个宿主拥有的 AbortController 向待定和活跃子级提供必需的 signal。关闭工作流准入中止该 signal，因此没有重复的 `ChildCancel` worker RPC 或显式的宿主侧 `run.cancel()` 扇出。完全停稳需要等待待定 start 和已发布子级 dispose 两者。
+一个宿主拥有的 AbortController 向待定和活跃子级提供必需的 signal。关闭工作流准入中止该 signal；完全停稳需要等待待定 start 和已发布子级 dispose 两者。[工作流沙箱复用](2026-09-13-workflow-ptc-sandbox-reuse.zh.md)负责 PTC 进程取消和不另设工作流清理定时器的规则。
 
-Worker 边界仍然序列化请求和结果。宿主保留首个终端结果仲裁、精确的子级计数、worker 死亡处理、优雅终止、迟到/重复消息拒绝和有界清理，因为结果接收、worker 退出和子级完全停稳是真正独立的事实。
+PTC 序列化请求和结果并负责进程终止。工作流适配器保留终态结果仲裁和子级归属，因为程序结算、进程退出和子级完全停稳仍是独立事实。
 
 ### 终端结果与物理清理保持分离
 
-工作流结果按公开优先级规则记录首个被接受的终端结果。该结果选定后清理可以继续：活跃子级仍需 dispose，worker 仍需终止，慢速外部后端可能超出配置的优雅期限。
+工作流结果按公开优先级规则记录首个被接受的终端结果。选定结果不会释放资源：PTC 进程和活跃子级仍需清理，子级资源释放必须履行其提供方约定。
 
-公开 dispose 在调用回调之前取得其记忆化 promise 的所有权。Worker 死亡在处理任何排队的迟到子级请求之前关闭准入，合成缺失的生命周期结束，并启动子级/进程清理而不重写已声明的结果。
+公开 dispose 汇入同一个清理操作。运行结算关闭子级准入，合成缺失的生命周期结束并清理子级，不重写已经认领的结果。
 
 ### ACP 提示词结算不依赖更新投递
 
@@ -344,7 +342,7 @@ TypeScript 无法管控 JavaScript 强制转换、直接 Cordis dispatch、进
 
 事件目录、服务目录、生产者/消费方矩阵、配置目录、模块图、工具目录、type-equiv 块和作用域事件解析器映射都是从源码生成或受新鲜度门禁约束的。[TypeScript 语义门禁 Agent Note](../../archived/process/2026-07-14-typescript-program-backed-semantic-gates.md) 拥有 Program 构造、语义事件发现和解析器生成规则。
 
-行为测试固定了作用域路由和 dispose、最终写入注册表时的碰撞清理、发布回滚、有序完全停稳、持久化前/后提交行为、跨展示和执行的活跃工具过滤、协作式提示词组装、原生和 PTC mode 中的结构化输出提交、异步 subagent 启动和信号取消、worker 终端仲裁、ACP 结算和进程拆除。
+行为测试固定了作用域路由和 dispose、最终写入注册表时的碰撞清理、发布回滚、有序完全停稳、持久化前/后提交行为、跨展示和执行的活跃工具过滤、协作式提示词组装、原生和 PTC mode 中的结构化输出提交、异步 subagent 启动和信号取消、工作流终态仲裁、ACP 结算和进程拆除。
 
 ## 曾考虑的替代方案
 

.agents/notes/implemented/architecture/2026-07-23-client-plugin-loading-model.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 .agents/notes/implemented/architecture/2026-07-23-client-plugin-loading-model.md
-2026-07-23-client-plugin-loading-model.md: c5576b148c5991dd498d3aed605e3c2e3395774b
-2026-07-23-client-plugin-loading-model.zh.md: c7d6982c2680995bd4698ddbff052a7708f69997
+2026-07-23-client-plugin-loading-model.md: d0f9b20f0adabda6cc7132e5411bcadd60c9e108
+2026-07-23-client-plugin-loading-model.zh.md: 27d5026ed03509d6408a16389246cb1321a37959

.agents/notes/implemented/architecture/2026-07-23-client-plugin-loading-model.md

@@ -58,12 +58,12 @@ What happens between `dsh web` starting and the UI appearing? Three stages: the
 
 Why is the roster yml rows and not a scan? Because which plugins compose into a deployment is a composition decision, not a package property — a package declaring `dsh.client` in the repo does not mean this deployment mounts it, so discovery-by-scan cannot make that call; the node half scans only what the tree actually mounted.
 
-**Phase one — the module face.** The injected HTML installs `window.__ModuleLoader__` in queue mode, starts preloading every application combo URL, executes every bootstrap combo URL as a blocking classic script, assigns `window.__DSH_BOOT__`, and then starts the Vite main module. The kernel calls the facade's `create()` with the raw graph and shell seeds. The facade removes and materializes the modules registration with a bootstrap `require` that rejects every external, then calls its `createClientModuleSystem` export. The modules bundle parses the graph, constructs the system, memoizes its own exports, retains the instance in its module closure, and switches the same facade to live registration. The kernel then prefetches every `immediately` row in parallel. Rows in the same application combo share its execution; separate combos load independently when an immediate row, a requested dependency, or ordinary entry import reaches them. A prefetch failure is swallowed here because phase two's import retries and owns the loud failure. `immediately` remains a registration barrier, not a package identity.
+**Phase one — the module face.** The injected HTML installs `window.__ModuleLoader__` in queue mode, starts preloading every application combo URL, executes every bootstrap combo URL as a blocking classic script, assigns `window.__DSH_BOOT__`, and then starts the Vite main module. The kernel calls the facade's `create()` with the raw graph and shell seeds. The facade removes and materializes the modules registration with a bootstrap `require` that rejects every external, then calls its `createClientModuleSystem` export. The modules bundle parses the graph, constructs and returns the system, memoizes its own exports, and switches the same facade to live registration. The kernel then prefetches every `immediately` row in parallel. Rows in the same application combo share its execution; separate combos load independently when an immediate row, a requested dependency, or ordinary entry import reaches them. A prefetch failure is swallowed here because phase two's import retries and owns the loud failure. `immediately` remains a registration barrier, not a package identity.
 
 **Phase two — the plugin face.**
 
 1. The kernel mounts the vendored Loader and injects the module system as `internal` before any entry exists. Ordering matters: `tree.import`'s bare-import fallback must never run in a browser.
-2. It creates every graph row uniformly. Importing the modules row returns the memoized bootstrap exports, whose `apply()` provides the closed-over system as `ctx.modules`; rows that require that service remain PENDING until then, so the modules row needs no special creation position. Render assembly is an ordinary host-graph row provided by `dsh-client-ui-renderer`; the kernel appends no assembly pseudo-entry.
+2. It creates every graph row uniformly. Importing the modules row returns the memoized bootstrap exports, whose `apply()` reads that tree's `Loader.internal` and provides the same instance as `ctx.modules`; rows that require that service remain PENDING until then, so the modules row needs no special creation position. Render assembly is an ordinary host-graph row provided by `dsh-client-ui-renderer`; the kernel appends no assembly pseudo-entry.
 3. Graph order governs synchronous factory availability; Cordis activation remains independent and proceeds through service waiting.
 4. `settled` = every entry created + `loader.await()` quiescent + an all-ACTIVE sweep. The sweep lists each import-failed, FAILED, or PENDING fiber with its missing services. It exists because cordis inject waits have no timeout — the sweep is the fail-loud floor.
 5. The framework-free loading page projects real fiber states via `internal/status`. After the sweep, the kernel calls `ctx.uiRenderer.mount(container)` and replaces the page with the real UI in one pass.

.agents/notes/implemented/architecture/2026-07-23-client-plugin-loading-model.zh.md

@@ -58,12 +58,12 @@ Host 会快照每个已构建插件产物，并把每个调度阶段的有序 ro
 
 为什么名册是 yml 行而不是扫描？因为哪些插件组合进一次部署是组合决策，不是包属性——一个在仓库中声明了 dsh.client 的包，不代表这次部署要挂载它，扫描发现无从替人做这个决定；node 半只扫描配置树实际挂载了的东西。
 
-**第一阶段——模块面。**注入的 HTML 以 queue 模式安装 `window.__ModuleLoader__`，开始预加载所有 application combo URL，以阻塞式 classic script 依次执行所有 bootstrap combo URL，赋值 `window.__DSH_BOOT__`，然后启动 Vite 主模块。内核把原始图和外壳 seed 传给 facade 的 `create()`。Facade 移除 modules registration，用拒绝全部 external 的 bootstrap `require` 将其物化，再调用其 `createClientModuleSystem` 导出。Modules bundle 解析图、构造系统、记忆化自身 exports、在模块闭包中保留该实例，并把同一 facade 切换到 live registration。随后内核并行预取每个 `immediately` row；同一 application combo 中的 row 共享一次执行，不同 combo 会在 immediate row、被请求依赖或普通 entry import 首次触及时独立加载。预取失败在这里被吞下，因为第二阶段 import 会重试并拥有那次大声失败。`immediately` 仍是 registration barrier，不是包身份。
+**第一阶段——模块面。**注入的 HTML 以 queue 模式安装 `window.__ModuleLoader__`，开始预加载所有 application combo URL，以阻塞式 classic script 依次执行所有 bootstrap combo URL，赋值 `window.__DSH_BOOT__`，然后启动 Vite 主模块。内核把原始图和外壳 seed 传给 facade 的 `create()`。Facade 移除 modules registration，用拒绝全部 external 的 bootstrap `require` 将其物化，再调用其 `createClientModuleSystem` 导出。Modules bundle 解析图、构造并返回系统、记忆化自身 exports，并把同一 facade 切换到 live registration。随后内核并行预取每个 `immediately` row；同一 application combo 中的 row 共享一次执行，不同 combo 会在 immediate row、被请求依赖或普通 entry import 首次触及时独立加载。预取失败在这里被吞下，因为第二阶段 import 会重试并拥有那次大声失败。`immediately` 仍是 registration barrier，不是包身份。
 
 **第二阶段——插件面。**
 
 1. 内核挂载 vendored Loader，在任何 entry 存在之前就把模块系统注入为 `internal`。顺序有讲究：`tree.import` 的裸 import 兜底分支在浏览器里绝不能跑到。
-2. 它统一创建每个 graph row。Import modules row 会返回记忆化的 bootstrap exports，其 `apply()` 把闭包中的系统提供为 `ctx.modules`；需要该 service 的 row 会保持 PENDING 直至此时，因此 modules row 无需特殊创建位置。渲染组装是由 `dsh-client-ui-renderer` 提供的普通 host graph row；内核不追加组装伪 entry。
+2. 它统一创建每个 graph row。Import modules row 会返回记忆化的 bootstrap exports，其 `apply()` 读取该树的 `Loader.internal`，把同一个实例提供为 `ctx.modules`；需要该 service 的 row 会保持 PENDING 直至此时，因此 modules row 无需特殊创建位置。渲染组装是由 `dsh-client-ui-renderer` 提供的普通 host graph row；内核不追加组装伪 entry。
 3. Graph 顺序治理同步 factory 可用性；Cordis 激活与之独立，仍经服务等待推进。
 4. `settled` = 每个 entry 已创建 + `loader.await()` 完全停稳 + 一次全 ACTIVE 扫描。扫描列出每个 import 失败、FAILED 或 PENDING 的 fiber 及其缺失的服务。它存在的理由：cordis 的 inject 等待没有超时——这次扫描就是大声失败的兜底线。
 5. 不依赖框架的 loading 页经 `internal/status` 投影真实 fiber 状态。检查完成后，内核调用 `ctx.uiRenderer.mount(container)`，一次切换到真实 UI。

.agents/notes/implemented/architecture/2026-07-29-projected-token-usage-and-request-context.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 .agents/notes/implemented/architecture/2026-07-29-projected-token-usage-and-request-context.md
-2026-07-29-projected-token-usage-and-request-context.md: f96257243bef91ff6a73418231de5e777d8edb2e
-2026-07-29-projected-token-usage-and-request-context.zh.md: 7365d5d816f9f9b324f3e3d3b4db0d3346851bdf
+2026-07-29-projected-token-usage-and-request-context.md: 10056ac8ef85562035d4fc4591e90aec603a7dd8
+2026-07-29-projected-token-usage-and-request-context.zh.md: 0e4cd68791f0ee65dd700b64493be3c33b78a920

.agents/notes/implemented/architecture/2026-07-29-projected-token-usage-and-request-context.md

@@ -58,4 +58,4 @@ Token totals stay stable across pagination, compaction, replay, restart, and rec
 
 Occupancy is approximate in the ways documented above. It is available immediately after restore or reconnect, since both fields are durable, at the cost of describing the last recorded request rather than an exact current boundary.
 
-Each session log gains one small `request/context` record per route or advertised-capacity change. Token-meter is the canonical owner of durable usage semantics, including retry-attempt separation in the cumulative projection and the reusable exact attempt/Turn fold; Web Chat only selects a complete loaded Turn and renders the fold result. The TUI retains its live per-step map because it does not mount the generic projection seam, and the standalone browser fixture mirrors the unit. Connection and API Gateway carry no token-specific code, own no per-session metrics cache, and perform no measurement. The browser keeps two generic projection values and no connection-local telemetry; streaming text deltas do not force the stats line to recompute or churn layout-observer subscriptions.
+Each session log gains one small `request/context` record per route or advertised-capacity change. Token-meter is the canonical owner of durable usage semantics, including retry-attempt separation in the cumulative projection and the reusable exact attempt/Turn fold; Web Chat only selects a complete loaded Turn and renders the fold result. The TUI retains its live per-step map because it does not mount the generic projection seam, and the assembled RemoteMock scenario supplies the same projection values. Connection and API Gateway carry no token-specific code, own no per-session metrics cache, and perform no measurement. The browser keeps two generic projection values and no connection-local telemetry; streaming text deltas do not force the stats line to recompute or churn layout-observer subscriptions.

.agents/notes/implemented/architecture/2026-07-29-projected-token-usage-and-request-context.zh.md

@@ -58,4 +58,4 @@ token 总量在分页、压缩、回放、重启和重连期间保持稳定，
 
 占用率在上文记录的意义上是近似值。由于两个字段都是持久的，它在恢复或重连后立即可用；代价是它描述的是最后一条已记录的请求，而不是精确的当前边界。
 
-每个会话日志会为每次路由或已公布容量变化增加一条小型 `request/context` 记录。token-meter 是持久用量语义的正典所有方，包括累计投影中的重试 attempt 分离，以及可复用的精确 attempt／Turn fold；Web Chat 只选择已完整加载的 Turn 并渲染 fold 结果。TUI 未挂载通用投影 seam，因此保留自己的实时逐步骤 map，而独立浏览器 fixture（测试前置数据）会镜像该单元。Connection 与 API Gateway 不携带任何 token 专用代码，不拥有逐会话指标缓存，也不执行测量。浏览器只保留两个通用投影值，不保留连接本地的遥测数据；流式文本增量不会迫使统计行重新计算或反复替换布局 observer 订阅。
+每个会话日志会为每次路由或已公布容量变化增加一条小型 `request/context` 记录。token-meter 是持久用量语义的正典所有方，包括累计投影中的重试 attempt 分离，以及可复用的精确 attempt／Turn fold；Web Chat 只选择已完整加载的 Turn 并渲染 fold 结果。TUI 未挂载通用投影 seam，因此保留自己的实时逐步骤 map，而组装 RemoteMock 场景提供相同投影值。Connection 与 API Gateway 不携带任何 token 专用代码，不拥有逐会话指标缓存，也不执行测量。浏览器只保留两个通用投影值，不保留连接本地的遥测数据；流式文本增量不会迫使统计行重新计算或反复替换布局 observer 订阅。

.agents/notes/implemented/architecture/2026-07-30-session-end-seed-log-boundary.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 .agents/notes/implemented/architecture/2026-07-30-session-end-seed-log-boundary.md
-2026-07-30-session-end-seed-log-boundary.md: 76f8904f75f6c4b5a1a6acab8d69ef2290be718e
-2026-07-30-session-end-seed-log-boundary.zh.md: 4310654c841305a7f0de2f46434d0f839085c482
+2026-07-30-session-end-seed-log-boundary.md: 9cf7c7f20701688208e8b1c343904f2a9f2cab9e
+2026-07-30-session-end-seed-log-boundary.zh.md: fdf4ec9de610f17e792de98334dd5d37f5e4c355

.agents/notes/implemented/architecture/2026-07-30-session-end-seed-log-boundary.md

@@ -8,7 +8,7 @@ English | [中文](2026-07-30-session-end-seed-log-boundary.zh.md)
 
 A plugin that owns a standalone open/close bracket in the session log cannot tell a dead marker from a live one. `compaction/start` … `compaction/end` is the shipped case: on picking up a log whose last compaction event is an unmatched `compaction/start`, "the previous writer died mid-compaction" and "a compaction is running right now" are byte-identical stored history. The owner must either refuse to compact a log that is actually free (wedging the session) or proceed over one that is genuinely busy.
 
-Nothing in the log marked where inherited history ended. `session/created`, `session/disposed`, and `session/flush` are cordis runtime signals, not log events; `agent/session-start` is emit-only. `Session.firstLiveSeq` already held the answer exactly — the seq of this lifecycle's first own write — but only in memory, so a consumer reading stored bytes could not see it.
+Nothing in the log marked where inherited history ended. `session/created`, `session/disposed`, and `session/flush` are cordis runtime signals, not log events; `agent/created` is also a non-durable runtime event. `Session.firstLiveSeq` already held the answer exactly — the seq of this lifecycle's first own write — but only in memory, so a consumer reading stored bytes could not see it.
 
 Crash repair does not close the gap and must not: `interruptedTurnClosers` synthesizes turn, step, and tool boundaries because core owns that vocabulary, and `compaction/*` belongs to the compaction seam. A core repair pass that closed plugin brackets would put every plugin's bracket semantics in core.
 

.agents/notes/implemented/architecture/2026-07-30-session-end-seed-log-boundary.zh.md

@@ -8,7 +8,7 @@ Status: implemented
 
 在会话日志中拥有独立开／闭括号的插件无法区分一个已死的标记和一个存活的标记。`compaction/start` … `compaction/end` 就是已发布的实例：当接手一份日志、而它最后的压缩（compaction）事件是一个未配对的 `compaction/start` 时，「上一个写入方在压缩中途死掉了」与「此刻正有一次压缩在运行」在存储历史中是逐字节相同的。该括号所有方只能二选一：拒绝压缩一份其实空闲的日志（把会话卡死），或者在一份确实繁忙的日志上继续压缩。
 
-日志中没有任何东西标出继承历史在哪里结束。`session/created`、`session/disposed` 与 `session/flush` 是 Cordis 运行时信号，不是日志事件；`agent/session-start` 只发射不落盘。`Session.firstLiveSeq` 本来就精确地持有这个答案——本生命周期第一次自有写入的 seq——但只存在于内存中，因此读取存储字节的消费方看不到它。
+日志中没有任何东西标出继承历史在哪里结束。`session/created`、`session/disposed` 与 `session/flush` 是 Cordis 运行时信号，不是日志事件；`agent/created` 同样是不落盘的运行时事件。`Session.firstLiveSeq` 本来就精确地持有这个答案——本生命周期第一次自有写入的 seq——但只存在于内存中，因此读取存储字节的消费方看不到它。
 
 崩溃修复既没有填上这个缺口，也不应该去填：`interruptedTurnClosers` 合成轮次、步骤与工具边界，是因为核心拥有那套词汇表，而 `compaction/*` 属于压缩 seam。一个会关闭插件括号的核心修复流程，等于把每个插件的括号语义都搬进核心。
 

.agents/notes/implemented/architecture/2026-07-31-goal-owned-durable-events.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 .agents/notes/implemented/architecture/2026-07-31-goal-owned-durable-events.md
-2026-07-31-goal-owned-durable-events.md: 8f7d3b74a83e063fbfc5a530f9e51a2a3eea8197
-2026-07-31-goal-owned-durable-events.zh.md: 8ee0099209c9c98a88549073b33c98ddb4b12bbd
+2026-07-31-goal-owned-durable-events.md: 0d3e632ba275f38e48fdf1f9ee6747f4bc460aab
+2026-07-31-goal-owned-durable-events.zh.md: fd29a6f36eed42f5ba26a46a32acb165d1e42873

.agents/notes/implemented/architecture/2026-07-31-goal-owned-durable-events.md

@@ -30,4 +30,4 @@ The domain does not automatically project each mutation into model input. Goal t
 
 Goal state is independent of inbox placement and admission. Replay has one mutation path, projections advance directly on `goal/change`, and continuation messages carry only round attribution. The model does not receive a mutation-only `<goal_state>` message; model-visible state appears through goal tools and scheduled continuation prompts. Direct session writers remain trusted and can append malformed changes, which the strict fold and invariant companion reject.
 
-Focused goal, goal-round-driver, command, TUI, and client-fixture tests pin durable replay, positive-round accounting, inbox independence, projection updates, and restored-session behavior. The keyless process test inspects the persisted `goal/change` event and verifies that creation alone starts no continuation round.
+Focused goal, goal-round-driver, command, TUI, and Client RemoteMock tests pin durable replay, positive-round accounting, inbox independence, projection updates, and restored-session behavior. The keyless process test inspects the persisted `goal/change` event and verifies that creation alone starts no continuation round.

.agents/notes/implemented/architecture/2026-07-31-goal-owned-durable-events.zh.md

@@ -30,4 +30,4 @@ Goal 领域需要持久状态，但不需要拥有待处理的模型输入。继
 
 Goal 状态不依赖 inbox 放置与准入。回放只有一条变更路径，投影直接由 `goal/change` 推进，继续执行消息只携带 Round 归属。模型不会收到仅用于变更的 `<goal_state>` 消息；模型可见状态来自 goal 工具与已调度的继续执行提示词。直接写入会话的写入方仍受信任，并且可以追加畸形变更；严格折叠与 invariant 配套模块会拒绝这些变更。
 
-聚焦的 goal、goal-round-driver、command、TUI 与 client fixture（测试前置数据）测试固定持久回放、正数 Round 计数、inbox 独立性、投影更新和恢复会话行为。无密钥进程测试检查持久的 `goal/change` 事件，并验证仅创建 goal 不会启动继续执行 Round。
+聚焦的 goal、goal-round-driver、command、TUI 与 Client RemoteMock 测试固定持久回放、正数 Round 计数、inbox 独立性、投影更新和恢复会话行为。无密钥进程测试检查持久的 `goal/change` 事件，并验证仅创建 goal 不会启动继续执行 Round。

.agents/notes/implemented/architecture/2026-08-15-client-shells-and-dynamic-packages.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 .agents/notes/implemented/architecture/2026-08-15-client-shells-and-dynamic-packages.md
-2026-08-15-client-shells-and-dynamic-packages.md: 1d67c778b6a06849324dd6095a98d57dc41b94f9
-2026-08-15-client-shells-and-dynamic-packages.zh.md: db1e4e7e7b319c283ae39a94535d88d4dc71d60a
+2026-08-15-client-shells-and-dynamic-packages.md: 89f89784512b86b70ee1b9460850e6f2f951a082
+2026-08-15-client-shells-and-dynamic-packages.zh.md: 5512d9262e979a94a65c25a6c1271b030e23ec70

.agents/notes/implemented/architecture/2026-08-15-client-shells-and-dynamic-packages.md

@@ -51,7 +51,7 @@ The modules Node half injects the startup protocol into the served HTML in this
 4. Assign `window.__DSH_BOOT__`, including all scheduling descriptors and every row's one-resource HMR combo URL.
 5. Execute the Vite main module.
 
-The bootstrap combo currently registers only the modules factory. The startup kernel passes the raw graph and shell seeds to `__ModuleLoader__.create()`. The facade removes the modules registration, materializes it with a `require` function that rejects every external, and invokes its `createClientModuleSystem` export. The modules bundle parses the graph, constructs `ClientModuleSystem`, caches its own exports as the modules row, retains the system in a module closure, and switches the same facade to live mode. The modules client face consequently has a zero-external bootstrap requirement.
+The bootstrap combo currently registers only the modules factory. The startup kernel passes the raw graph and shell seeds to `__ModuleLoader__.create()`. The facade removes the modules registration, materializes it with a `require` function that rejects every external, and invokes its `createClientModuleSystem` export. The modules bundle parses the graph, constructs and returns `ClientModuleSystem`, caches its own exports as the modules row, and switches the same facade to live mode. The kernel installs that instance as its Loader's `internal`, and the modules plugin reads it there when it provides `ctx.modules`. The modules client face consequently has a zero-external bootstrap requirement and no module-global system identity.
 
 After the `immediately` tier has registered its factories, the kernel creates all Loader entries, awaits Cordis quiescence, and requires every fiber to be ACTIVE. It then calls `ctx.uiRenderer.mount(container)`. The dynamic `ui-renderer` package owns React, slot rendering, hydration of the existing boot DOM, and the React root lifecycle; the startup kernel and failure page remain React-free.
 

.agents/notes/implemented/architecture/2026-08-15-client-shells-and-dynamic-packages.zh.md

@@ -51,7 +51,7 @@ Modules Node 半按以下顺序向实际返回的 HTML 注入启动协议：
 4. 赋值 `window.__DSH_BOOT__`，其中包含全部调度描述及每个 row 的单资源 HMR combo URL。
 5. 执行 Vite 主模块。
 
-Bootstrap combo 当前只登记 modules factory。启动内核把原始图与外壳 seed 传给 `__ModuleLoader__.create()`。Facade 移除 modules registration，用拒绝全部 external 的 `require` 函数将其物化，再调用其 `createClientModuleSystem` 导出。Modules bundle 解析图、构造 `ClientModuleSystem`、把自身 exports 缓存为 modules row、在模块闭包中保留该系统，并把同一 facade 切换到 live 模式。因此 modules client face 必须满足零 external 的自举要求。
+Bootstrap combo 当前只登记 modules factory。启动内核把原始图与外壳 seed 传给 `__ModuleLoader__.create()`。Facade 移除 modules registration，用拒绝全部 external 的 `require` 函数将其物化，再调用其 `createClientModuleSystem` 导出。Modules bundle 解析图、构造并返回 `ClientModuleSystem`、把自身 exports 缓存为 modules row，并把同一 facade 切换到 live 模式。内核把该实例装成自身 Loader 的 `internal`，modules 插件从这里读取并提供 `ctx.modules`。因此 modules client face 保持零 external 的自举要求，也没有模块级系统身份。
 
 `immediately` 层级完成 factory 注册后，内核创建全部 Loader entry，等待 Cordis 静止，并要求每个 fiber 都进入 ACTIVE。随后调用 `ctx.uiRenderer.mount(container)`。动态 `ui-renderer` 包拥有 React、slot 渲染、已有启动 DOM 的 hydrate 和 React root 生命周期；启动内核与失败页保持 React-free。
 

.agents/notes/implemented/architecture/2026-08-18-session-history-and-event-transport.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 .agents/notes/implemented/architecture/2026-08-18-session-history-and-event-transport.md
-2026-08-18-session-history-and-event-transport.md: 9a593ca9bfe4a80657d3eeeaa89cf0dcc73dd829
-2026-08-18-session-history-and-event-transport.zh.md: a7f8b8f27153a1814ca65bbf9042d78717b50849
+2026-08-18-session-history-and-event-transport.md: 8781ea265798ff200a6e0a58542b7d03693d8bd3
+2026-08-18-session-history-and-event-transport.zh.md: 1603f5aed8adb7a42fecab92511176d2147be5c3

.agents/notes/implemented/architecture/2026-08-18-session-history-and-event-transport.md

@@ -220,7 +220,7 @@ Session-list `updatedAt` is `max(header.createdAt, sessionListMetadata.lastPromp
 
 `packages/api/workspace-controller` provides Host `ctx.workspaceController` and the generated `ctx.remote.workspace` namespace.
 
-It owns create, rename, delete, insertBefore, insertSessionBefore, archiveSession, and `follow`. Workspace registry remains the durable source of truth; the Controller owns Remote commands, projection, and error mapping.
+It owns create, rename, delete, insertBefore, insertSessionBefore, archiveSession, unarchiveSession, and `follow`. Workspace registry remains the durable source of truth; the Controller owns Remote commands, projection, and error mapping.
 
 `WorkspaceFeed` synchronously observes storage `domain/changed`, and each follow generation emits a complete baseline before `upsert`, `remove`, `order`, and `archived` deltas.
 

.agents/notes/implemented/architecture/2026-08-18-session-history-and-event-transport.zh.md

@@ -220,7 +220,7 @@ Session 列表的 `updatedAt` 取 `max(header.createdAt, sessionListMetadata.las
 
 `packages/api/workspace-controller` 提供 Host `ctx.workspaceController` 与生成的 `ctx.remote.workspace` namespace。
 
-它拥有 create、rename、delete、insertBefore、insertSessionBefore、archiveSession 与 `follow`。Workspace registry 仍是持久事实来源，Controller 负责 Remote 命令、投影和错误映射。
+它拥有 create、rename、delete、insertBefore、insertSessionBefore、archiveSession、unarchiveSession 与 `follow`。Workspace registry 仍是持久事实来源，Controller 负责 Remote 命令、投影和错误映射。
 
 `WorkspaceFeed` 同步观察 storage `domain/changed`，并为每个 follow generation 先发送完整 baseline，再发送 `upsert`、`remove`、`order` 与 `archived` 增量。
 

.agents/notes/implemented/architecture/2026-08-23-client-derived-tool-presentation.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 .agents/notes/implemented/architecture/2026-08-23-client-derived-tool-presentation.md
-2026-08-23-client-derived-tool-presentation.md: 895618099c6669058d23fc0a58335a90d5fbe4ca
-2026-08-23-client-derived-tool-presentation.zh.md: 38572f17c41ea9b68dc8c8c8cdc26a74d9dbb757
+2026-08-23-client-derived-tool-presentation.md: 703960cb24f90eb678fda381f929efb3f94bbd28
+2026-08-23-client-derived-tool-presentation.zh.md: 7f161379553f3aca33c72c2edf93e114d27eb7f9

.agents/notes/implemented/architecture/2026-08-23-client-derived-tool-presentation.md

@@ -387,7 +387,7 @@ This change does not add a general tool-side-effect registry. The ability for a
 
 ## Fixtures and Test Data
 
-The Client fixture deletes its handwritten `presentCall()`, `presentResult()`, `viewFor()`, and fixture tool-view types. It continues producing the same raw calls, result content, and result metadata as a real log.
+The assembled RemoteMock scenario contains no handwritten `presentCall()`, `presentResult()`, `viewFor()`, or tool-view types. It supplies the same raw calls, result content, and result metadata as a real log.
 
 | Fixture | Raw facts that must remain |
 |---|---|
@@ -398,7 +398,7 @@ The Client fixture deletes its handwritten `presentCall()`, `presentResult()`, `
 | web | result metadata sources/answer or url/statusCode/truncated |
 | generic/custom | name, argsRaw, content, and error |
 
-The fixture does not import Host tool packages to compute page presentation and retains no presenter mirror. The same raw fixture continues to drive jsdom, built Web snapshots, and the `?fixture` browser path.
+The scenario does not import Host tool packages to compute page presentation and retains no presenter mirror. The same raw scenario drives built Web snapshots under jsdom; real-Host browser cases independently cover the network path.
 
 ## Presentation-Equivalence Matrix
 
@@ -566,7 +566,7 @@ Changes to this decision use `dsh-pre-push-checks` to select commands for the fi
 - ui-chat and ui-trajectory Tool Definition tests;
 - ui-tool terminal, diff, read, search, web, row, tree, and details tests;
 - ui-deliverables produced-file tests;
-- connection fixture and Client runtime tests;
+- assembled RemoteMock and Client runtime tests;
 - affected Host and Client TypeScript faces;
 - lint and duplication;
 - per-file 100% coverage for affected source files;

.agents/notes/implemented/architecture/2026-08-23-client-derived-tool-presentation.zh.md

@@ -387,7 +387,7 @@ Deliverables Definition 按 callId 观察 root `tool/call` 与成功 `tool/resul
 
 ## Fixture 与测试数据
 
-Client fixture 删除手写 `presentCall()`、`presentResult()`、`viewFor()` 与 fixture tool-view 类型。它继续产生与真实日志相同的 raw call、result content 和 result meta。
+组装 RemoteMock 场景不包含手写 `presentCall()`、`presentResult()`、`viewFor()` 或 tool-view 类型。它提供与真实日志相同的 raw call、result content 和 result meta。
 
 | Fixture | 必须保留的原始事实 |
 |---|---|
@@ -398,7 +398,7 @@ Client fixture 删除手写 `presentCall()`、`presentResult()`、`viewFor()`
 | web | result meta 的 sources/answer 或 url/statusCode/truncated |
 | generic/custom | name、argsRaw、content、error |
 
-fixture 不导入 Host 工具包来计算页面展示，也不保留 presenter 镜像。同一 raw fixture 继续驱动 jsdom、built Web snapshot 与 `?fixture` 浏览器路径。
+该场景不导入 Host 工具包来计算页面展示，也不保留 presenter 镜像。同一 raw 场景在 jsdom 下驱动 built Web snapshot；真实 Host 浏览器用例独立覆盖网络路径。
 
 ## 展示等价矩阵
 
@@ -566,7 +566,7 @@ Host registry 允许不同 scope 为同一 tool name 提供不同定义；Sessio
 - ui-chat 与 ui-trajectory Tool Definition 测试；
 - ui-tool terminal、diff、read、search、web、row、tree 与 details 测试；
 - ui-deliverables produced-files 测试；
-- connection fixture 与 Client runtime 测试；
+- 组装 RemoteMock 与 Client runtime 测试；
 - 受影响 Host/Client TypeScript face；
 - lint 与 duplication；
 - 受影响源文件 per-file 100% coverage；

.agents/notes/implemented/architecture/2026-08-27-outbound-proxy-policy.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 .agents/notes/implemented/architecture/2026-08-27-outbound-proxy-policy.md
-2026-08-27-outbound-proxy-policy.md: ef6ad24cead3b0c2c2f4ac5fff9ec1dc9b602a5f
-2026-08-27-outbound-proxy-policy.zh.md: d824fc705723b298b1ffc05b6d7996219d01459e
+2026-08-27-outbound-proxy-policy.md: e588ff751fd18762fbe5e24aa17f9bbbc4ff8deb
+2026-08-27-outbound-proxy-policy.zh.md: 57faea256120f5f76ecc4dab1bd8709003672fd0

.agents/notes/implemented/architecture/2026-08-27-outbound-proxy-policy.md

@@ -40,7 +40,7 @@ This keeps `proxyForUrl()` and the dispatcher answering from one set of values.
 
 The URL-level policy is untouched: `http(s)` only, no embedded credentials, the length cap, and the cross-origin redirect refusal all still apply on every hop.
 
-**A spawned child gets the policy through its environment; a model-executing worker gets nothing.** `proxyEnvironmentForChild()` merges into `scrubbedParentEnv()`, the one function every spawner already shares. The workflow worker does NOT receive it: it executes the model-authored script body, and a proxy URL may carry `user:password`. That is the same containment the PTC runtime keeps and `docs/defensive-patterns.md` requires, so a workflow's own requests go direct.
+**Ordinary child processes receive proxy policy; PTC program environments omit it.** `proxyEnvironmentForChild()` merges into `scrubbedParentEnv()`. PTC also executes workflow scripts and excludes these settings from program environments because a proxy URL may carry `user:password`; direct program requests go direct.
 
 The child keeps the user's own values, and that is what once broke it. Node parses `HTTP_PROXY` and `HTTPS_PROXY` under `NODE_USE_ENV_PROXY` before running the program and exits on any scheme other than `http:` or `https:`; a `socks4://` kept for `curl` therefore ended every Node child — MCP servers, subagent CLIs, `npm` — before its first line, while this process had reported only that the scheme stayed direct. Measured on Node 24.17: `socks4://`, `ftp://`, and a malformed value all exit 1; `socks5://` happens to be accepted there. The flag is now withheld whenever a value the child receives is one this package refused, so such a child connects directly and `curl` still reads the value it was kept for. Handing the child the resolved value instead would have kept Node proxied at the price of silently rewriting what the user set for another tool.
 
@@ -70,7 +70,7 @@ Weighed against that, telemetry is the one outbound channel whose loss costs the
 
 **Read the operating system's proxy settings.** Rejected for this change. Only Codex and Reasonix among six surveyed products do it, and Codex keeps it behind a default-off flag. Measured on the author's machine, it would have found nothing: the proxy application had written the setting to the Wi-Fi service while the primary interface was a USB ethernet adapter with no proxy, so `scutil --proxy` reported none while the exported variables worked. It also needs its own bypass matcher, because an operating system list carries CIDR entries that neither undici nor Node matches.
 
-**Give model-authored code the proxy too.** Rejected because a proxy URL may carry credentials. Node ptc-runtime processes and workflow workers keep those settings outside the program environment; direct network use remains subject to the program's execution policy.
+**Give model-authored code the proxy too.** Rejected because a proxy URL may carry credentials. PTC processes, including workflow execution, keep those settings outside the program environment; direct network use remains subject to the program's execution policy.
 
 ## Consequences
 

.agents/notes/implemented/architecture/2026-08-27-outbound-proxy-policy.zh.md

@@ -40,7 +40,7 @@ Node 内置的 `fetch` 会忽略 `HTTP_PROXY` 与 `HTTPS_PROXY`。开发者运
 
 URL 层策略未受影响：仅 `http(s)`、禁止内嵌凭据、长度上限与跨域重定向拒绝在每一跳上依然生效。
 
-**派生的子进程通过环境获得策略；执行模型代码的 worker 什么也不获得。** `proxyEnvironmentForChild()` 并入 `scrubbedParentEnv()`——每个 spawner 本就共享的那一个函数。workflow worker **不**接收它：它执行的是模型编写的脚本体，而代理 URL 可能携带 `user:password`。这与 PTC runtime 保持的隔离相同，也是 `docs/defensive-patterns.md` 的要求，因此 workflow 自身的请求直连。
+**普通子进程接收代理策略；PTC 程序环境省略它。** `proxyEnvironmentForChild()` 并入 `scrubbedParentEnv()`。PTC 也执行工作流脚本，并从程序环境中排除这些设置，因为代理 URL 可能携带 `user:password`；程序的直接请求采用直连。
 
 子进程拿到的是用户自己的值，而这恰恰曾把它弄坏。Node 在 `NODE_USE_ENV_PROXY` 下会在运行程序之前先解析 `HTTP_PROXY` 与 `HTTPS_PROXY`，遇到 `http:`/`https:` 之外的协议直接退出；于是一个为 `curl` 保留的 `socks4://` 会让每个 Node 子进程——MCP server、subagent CLI、`npm`——在第一行之前就终结，而本进程此前只报告过该协议保持直连。在 Node 24.17 上实测：`socks4://`、`ftp://` 与畸形值均以 1 退出；`socks5://` 恰好在该版本被接受。现在只要子进程收到的某个值是本包拒绝过的，就扣下该标志，这样的子进程直连，`curl` 仍读到为它保留的值。若改为把解析后的值交给子进程，Node 固然能继续走代理，代价却是悄悄改写用户为另一工具设置的值。
 
@@ -70,7 +70,7 @@ URL 层策略未受影响：仅 `http(s)`、禁止内嵌凭据、长度上限与
 
 **读取操作系统的代理设置。** 本次变更中被否决。所调研的六个产品中只有 Codex 与 Reasonix 这样做，且 Codex 把它放在默认关闭的开关之后。在作者机器上实测，它什么也读不到：代理软件把设置写在了 Wi-Fi 服务上，而主接口是一块没有代理的 USB 以太网卡，因此 `scutil --proxy` 报告无代理，而导出的环境变量却工作正常。它还需要自带的绕过匹配器，因为操作系统的列表含有 undici 与 Node 都不匹配的 CIDR 条目。
 
-**也把代理配置交给模型编写的代码。** 不采纳，因为代理 URL 可能携带凭据。Node ptc-runtime 进程与 workflow worker 不在程序环境中提供这些设置；直接网络访问仍受程序执行策略约束。
+**也把代理配置交给模型编写的代码。** 不采纳，因为代理 URL 可能携带凭据。PTC 进程（包括工作流执行）不在程序环境中提供这些设置；直接网络访问仍受程序执行策略约束。
 
 ## Consequences
 

.agents/notes/implemented/architecture/2026-09-06-embedded-stream-record-readers.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 .agents/notes/implemented/architecture/2026-09-06-embedded-stream-record-readers.md
-2026-09-06-embedded-stream-record-readers.md: 972fee634833cef5fd7b0a54f69780b9370f0cc3
-2026-09-06-embedded-stream-record-readers.zh.md: faba6e179887a9943926f2c73aa8e42a10cee300
+2026-09-06-embedded-stream-record-readers.md: 76e109de577d070093bb7123aa50d9a4ae7fcbe5
+2026-09-06-embedded-stream-record-readers.zh.md: 33f4b31117197c47e7f2ff1637bc13fa2cbf1a8c

.agents/notes/implemented/architecture/2026-09-06-embedded-stream-record-readers.md

@@ -20,7 +20,7 @@ After v2 embedded streams settlement widened with the message content and Chat a
 - Run readers: `runFirstTokenTime` and `runFirstVisibleTime` reconstruct the first qualifying member's time from `time0` and the `dt` gaps and stop scanning there; a name-bearing Tool-call run yields `time0` without reading a fragment.
 - Stream readers: `assistantStreamFirstTokenTime`, `assistantStreamHasVisibleContent`, `assistantStreamHasVisibleText`, `lastAssistantStreamChunk(stream, type)` (backward scan), `assistantStreamChunks(stream, type)`, `joinAssistantStreamText`, and `assembleAssistantStream`, which feeds a `BlockAssembler` one joined delta per run (assembly only concatenates, so blocks, usage, finish, and replay state equal the per-member result). `RawStreamChunkType` excludes the delta types, so a raw-chunk lookup can never silently skip packed members.
 
-Session Stats reads `assistantStreamFirstTokenTime`; the token meter reads `lastAssistantStreamChunk(stream, 'usage')` and assembles provider output through `assembleAssistantStream`; the subagent output fold appends `joinAssistantStreamText`; the Session Controller scans `assistantStreamChunks(stream, 'block-end')` for images.
+Session Stats, Chat, and Trajectory read `assistantStreamFirstTokenTime` from both `assistant/attempt` and `assistant/message`, retaining the Step's first token across retries. Chat and Trajectory settle content from the assembled message while reading timing independently, so reopening history retains TTFT and decoding metrics without expanding streams. The token meter reads `lastAssistantStreamChunk(stream, 'usage')` and assembles provider output through `assembleAssistantStream`; the subagent output fold appends `joinAssistantStreamText`; the Session Controller scans `assistantStreamChunks(stream, 'block-end')` for images.
 
 `expandAssistantStream` keeps its strict validation and its remaining callers, which need every member or validate the stream at a durable boundary: Session restore validation, the v1-to-v2 migration validator and publication Worker replay, the reconnect baseline, and test support.
 

.agents/notes/implemented/architecture/2026-09-06-embedded-stream-record-readers.zh.md

@@ -20,7 +20,7 @@ Session 格式 v2 将每次模型尝试的紧凑流（`AssistantStreamRecord[]`
 - Run 读取器：`runFirstTokenTime` 与 `runFirstVisibleTime` 从 `time0` 与 `dt` 间隔重建首个合格成员的时间并停止扫描；带名称的 Tool-call run 直接产出 `time0`，不读片段。
 - 流读取器：`assistantStreamFirstTokenTime`、`assistantStreamHasVisibleContent`、`assistantStreamHasVisibleText`、`lastAssistantStreamChunk(stream, type)`（逆向扫描）、`assistantStreamChunks(stream, type)`、`joinAssistantStreamText` 与 `assembleAssistantStream`（每个 run 向 `BlockAssembler` 喂入一个拼接后的 delta；组装只做拼接，因此 blocks、usage、finish 与 replay state 与逐成员结果一致）。`RawStreamChunkType` 排除 delta 类型，因此原始 chunk 查找不可能静默跳过打包成员。
 
-Session Stats 读取 `assistantStreamFirstTokenTime`；token 计量读取 `lastAssistantStreamChunk(stream, 'usage')` 并通过 `assembleAssistantStream` 组装提供商输出；子代理输出折叠追加 `joinAssistantStreamText`；Session Controller 用 `assistantStreamChunks(stream, 'block-end')` 扫描镜像。
+Session Stats、Chat 与 Trajectory 从 `assistant/attempt` 和 `assistant/message` 读取 `assistantStreamFirstTokenTime`，跨重试保留步骤的首个 token。Chat 与 Trajectory 从组装后的消息结算内容，并独立读取计时，因此重新打开历史时无需展开流便能保留 TTFT 与解码指标。token 计量读取 `lastAssistantStreamChunk(stream, 'usage')` 并通过 `assembleAssistantStream` 组装提供商输出；子代理输出折叠追加 `joinAssistantStreamText`；Session Controller 用 `assistantStreamChunks(stream, 'block-end')` 扫描镜像。
 
 `expandAssistantStream` 保留其严格校验与其余调用方（需要每个成员或在持久边界校验流）：Session 恢复校验、v1-to-v2 迁移校验器与发布 Worker 重放、重连基线、测试支撑。
 

.agents/notes/implemented/architecture/2026-09-09-awaited-agent-creation.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 .agents/notes/implemented/architecture/2026-09-09-awaited-agent-creation.md
+2026-09-09-awaited-agent-creation.md: 27353225a8d99d038d5c61b1eea0fa67d9120a55
+2026-09-09-awaited-agent-creation.zh.md: f664e89da0794acd1d2a6c0c8a4820a7056dcac6

.agents/notes/implemented/architecture/2026-09-09-awaited-agent-creation.md

@@ -0,0 +1,31 @@
+# Agent Note: Awaited per-agent initialization
+
+Status: implemented
+
+English | [中文](2026-09-09-awaited-agent-creation.zh.md)
+
+## Problem
+
+Shared presets install tools and prompt sections separately for each Agent. That installation can await plugin activation, and external SessionStart hooks can produce context asynchronously. A creator must know that these contributions have finished before the Agent's first model request.
+
+## Decision
+
+`agent/created` is the serial initialization event after factory setup and registry entry. Each listener finishes before the next starts; a throw or rejection fails creation and skips later listeners. The payload retains `SessionStartSource` and accepts the factory's cancellation signal. `register()` and `announce()` are awaited by their callers. Lifecycle source selection belongs to factory publication through `announce()`; `register()` announces fresh startup.
+
+AgentLoop holds its existing maintenance activity through setup and creation dispatch. Input may enter the inbox during initialization, but the driver starts only after successful completion. Failure cancels that activity without waking queued input; ordered teardown owns inbox cleanup. Keeping these operations separate preserves the initialization error when another teardown has already removed the inbox projection.
+
+Creation dispatch retains the scope and Session while listeners await. Disposal cancels initialization and joins the dispatch before releasing those resources. A listener must not await its own Agent's idle state or its owner's disposal, because each waits for that listener to finish. Background work on another Agent follows that Agent's own initialization lifecycle.
+
+This decision owns asynchronous creation timing. The [scope runtime decision](2026-07-12-agent-scope-runtime-design.md) retains registry identity and teardown ownership, while the [interception decision](../feature/2026-06-30-interception-extension-points.md) retains policy and tool-event semantics.
+
+## Alternatives considered
+
+**A separate setup event.** Existing creation listeners already install per-agent contributions. A second initialization event splits that responsibility without a distinct consumer need.
+
+**Detached initialization.** Returning before plugin activation or hook context settles lets the first request omit required tools or context and disconnects initialization failure from the creator.
+
+## Consequences
+
+Creation latency includes asynchronous plugin and SessionStart work. Initializer failures become caller-visible creation failures, and cancellation relies on listeners settling cooperatively. Notifications already delivered cannot be undone; rollback pairs them with disposal notifications.
+
+Scoped installer tests verify rollback, lifecycle tests verify ordered completion and cancellation, and the SDK serial-created scenario verifies that asynchronous prompt context reaches the first request. Hook tests cover awaited context injection and subprocess disposal.

.agents/notes/implemented/architecture/2026-09-09-awaited-agent-creation.zh.md

@@ -0,0 +1,31 @@
+# Agent Note: 可等待的单个 agent 初始化
+
+Status: implemented
+
+[English](2026-09-09-awaited-agent-creation.md) | 中文
+
+## 问题
+
+共享 preset 为每个 Agent（智能体）分别安装工具与提示词段。安装可能要等待插件激活，外部 SessionStart 钩子也可能异步产生上下文。创建方必须知道这些贡献已完成，才能允许 Agent 发起首次模型请求。
+
+## 决策
+
+`agent/created` 是工厂 setup 完成且进入注册表后的串行初始化事件。每个监听器完成后才启动下一个；抛出或拒绝会使创建失败，并跳过后续监听器。载荷保留 `SessionStartSource`，并接受工厂的取消信号。调用方等待 `register()` 与 `announce()` 完成。生命周期来源由工厂发布时通过 `announce()` 指定；`register()` 宣告全新启动。
+
+AgentLoop 在 setup 与创建分发期间保留现有的维护活动。初始化期间输入可以进入收件箱，但只有成功完成后驱动器才会启动。失败会取消该活动而不唤醒已排队输入；有序 teardown 拥有收件箱清理。区分这两项操作，能够在另一条 teardown 已移除收件箱投影时保留初始化错误。
+
+创建分发在监听器等待期间保留作用域与 Session。Dispose（资源释放）会取消初始化，并在释放这些资源前等待分发结束。监听器不得等待自身 Agent 的空闲状态或自身所有者的 dispose，因为两者都要等待该监听器完成。另一个 Agent 上的后台工作遵循该 Agent 自己的初始化生命周期。
+
+本决策拥有异步创建时序。[作用域运行时决策](2026-07-12-agent-scope-runtime-design.zh.md)继续拥有注册表身份与 teardown 归属，[拦截决策](../feature/2026-06-30-interception-extension-points.zh.md)继续拥有策略与工具事件语义。
+
+## 考虑过的替代方案
+
+**单独的 setup 事件。** 现有创建监听器已经安装针对单个 agent 的贡献。第二个初始化事件会拆分这一职责，却没有独立的消费方需求。
+
+**分离运行的初始化。** 在插件激活或钩子上下文结算前返回，会让首次请求缺少所需工具或上下文，也会使初始化失败脱离创建方。
+
+## 后果
+
+创建延迟包括异步插件与 SessionStart 工作。初始化器失败成为调用方可见的创建失败，取消依赖监听器协作结算。已经送达的通知无法撤回；回滚会为其配对销毁通知。
+
+作用域安装器测试验证回滚，生命周期测试验证有序完成与取消，SDK serial-created 场景验证异步提示词上下文进入首次请求。钩子测试覆盖需等待的上下文注入与子进程 dispose。

.agents/notes/implemented/architecture/2026-09-10-image-offload-events.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 .agents/notes/implemented/architecture/2026-09-10-image-offload-events.md
+2026-09-10-image-offload-events.md: f540dc98cee8daf3e360b14234a2ac21d0f53aab
+2026-09-10-image-offload-events.zh.md: 868fca59ba5fc1941de752127f61b091aedb234f

.agents/notes/implemented/architecture/2026-09-10-image-offload-events.md

@@ -0,0 +1,49 @@
+# Agent Note: Durable image offload events
+
+Status: implemented
+
+English | [中文](2026-09-10-image-offload-events.zh.md)
+
+## Problem
+
+A request can exceed an image-capable route's byte or image-count budget. Recomputing the omitted prefix for each request lets a route switch, inline fallback, or compaction restore previously omitted images. The model-visible image set then depends on unlogged request preparation. The [unified image pipeline](../feature/2026-08-20-unified-image-request-pipeline.md) owns attachment normalization and request versions; durable omission needs its own recorded decision.
+
+## Decision
+
+`dsh-compaction-image-offload` owns recovery on the `agent/request-error` waterfall. An `IMAGE_OFFLOAD_REQUIRED` failure supplies `offloadImages`; the plugin selects that many oldest retained input-image occurrences in current surface order, appends one `image/offload` event, and returns `retry`. Assistant output images are excluded. If no retained input image remains, the plugin delegates. Other failures do not enter this recovery. Offload does not consume the provider retry budget or emit `llm/retry`.
+
+Summary requests bypass the agent error waterfall. `dsh-compaction-basic` preserves the full `LlmFailure`, checks cancellation and selection stability, and dispatches synchronous `compaction/summary-error` with the selected event seqs. The same image-offload plugin records omissions only within that selection. Synchronous recovery keeps the stability check and decision adjacent. The backend then re-derives input and pricing, including the shrink baseline. Each retry requires additional omitted occurrences; exhaustion delegates the failure. A later summary failure or cancellation preserves any recorded omissions, so command errors do not claim the conversation is unchanged.
+
+The event payload is `{ targets: [{ seq, imageIndexes }] }`. Each target identifies a current `user/message` or `tool/result` node. Its strictly increasing, zero-based indexes enumerate every image in depth-first content order, including nested tool results and previously omitted images. Equal attachment ids remain distinct occurrences. Exact selections remain unambiguous after positional replacements reorder the surface; an attachment id or a raw-log prefix cannot identify the selected set.
+
+The event has no `surfaceOp`. It neither creates a message nor replaces a message node. The plugin's pure projection validates all targets before Session commits the event and rejects missing or shadowed nodes, output images, duplicate targets, invalid indexes, and already omitted occurrences. Reconstruction derives immutable message copies with `ImageBlock.offloaded: true` at the recorded positions. Original events, message ids, sources, and unaffected blocks remain unchanged. Resume, fork, and replay apply the same selections from the log.
+
+The compaction plugin owns selection, event declaration, image validation and projection, and retry policy. [Plugin-owned message projections](2026-09-11-plugin-owned-message-projections.md) owns the generic Session integration and explicit detached assembly, superseding core-owned interpretation. The instance `deriveEventMessage()` and `deriveMessages()` return projected messages; pure reconstructors pass `foldSurface(events, projections).projectedMessages` to the exported `deriveEventMessage()`. Rewriters preserving images must consume projected messages, including tool-result pruning and summarizer input.
+
+`contentGeneration` advances for message replacements and image-offload events so cached history and request series are refreshed. `replaceGeneration` advances only for replacements; image omission cannot masquerade as successful text compaction. An unchanged request envelope gets a `request/header` with reason `series`; a simultaneous envelope change carries `startsSeries: true`.
+
+Adapters prepare only retained images and render marked occurrences as `offloadedImageText` with the current execution-world access path. Route-local budget checks report required counts, including the tighter inline fallback budget, but do not edit history. These checks are provisional: a provider failure that identifies uncacheable images can replace local counting without moving the durable decision into the adapter.
+
+Token accounting applies selections to the affected nodes' image occurrences without replacing their identities or changing prior usage-anchor snapshots. Route pricing substitutes placeholder text for image cost. The fixed structural heuristic excludes the `offloaded` marker, so image omission does not change that heuristic or require a `compaction/prune` shadow price. The context-pressure and context-breakdown checkpoint versions advance to reject cached estimates that counted the marker.
+
+`image/offload` is required on read. Its addition changes the recognized event vocabulary, not the Session envelope or operation grammar, so it does not advance `SESSION_FORMAT_VERSION`. A reader without this event type refuses the log instead of silently restoring omitted images.
+
+## Alternatives considered
+
+**Reuse `compaction/prune` and `surfaceOp: replace`.** The [archived replacement design](../../archived/architecture/2026-09-02-durable-image-offload.md) records full carrying-message copies and couples omission to node replacement and shadow-price accounting. A dedicated event expresses which images are omitted while retaining message identity and keeping retry policy in the same plugin.
+
+**Record a request-local projection outcome.** A record that does not affect subsequent derivation cannot prevent omitted images from returning. Logging whole request bodies repeats message history for a decision that needs only occurrence indexes.
+
+**Use an attachment id or a single watermark.** Attachments can occur repeatedly, and surface order can differ from event-sequence order. Explicit per-node indexes distinguish these cases without redefining a watermark after every replacement.
+
+**Move selection into the loop or adapters.** The request-error extension already owns repair and retry. The adapter knows the route budget; the plugin knows the session input order. Neither requires route-specific policy in the shared loop.
+
+## Consequences
+
+Omission remains effective when a route budget grows, an inline fallback ends, or compaction reduces context. It does not delete attachment bytes. Reading the placeholder's access path creates a new image occurrence that can be retained independently. The execution-world path is still resolved during serialization, as it is for retained image handles; this event records the selected image set, not filesystem mappings.
+
+Session invokes the registered pure projection and maintains one content-generation signal. Consumers reconstructing model input must use projected messages. Human views can still show original images from immutable events. The plugin requires neither token-meter nor compaction services merely to record omission.
+
+## Testing
+
+Plugin projection tests cover nested and repeated occurrences, atomic rejection, cache invalidation, pure folding, restore, fork, and untouched source events. Recovery tests cover consecutive failures, current surface order, exhausted recovery, fresh request headers, and disposal. Token-meter tests preserve node identity and prior usage anchors while repricing only selected occurrences. Tool-result pruning and compaction tests ensure later rewrites consume the projected history. TypeScript and Python SDK expected outputs exercise the required event through shipped profiles.

.agents/notes/implemented/architecture/2026-09-10-image-offload-events.zh.md

@@ -0,0 +1,49 @@
+# Agent Note：持久图片省略事件
+
+Status: implemented
+
+[English](2026-09-10-image-offload-events.md) | 中文
+
+## 问题
+
+请求可能超过图片路由的字节或图片数量预算。每次请求重新计算省略前缀，会让路由切换、内联回退或压缩恢复先前省略的图片。模型可见的图片集合因此取决于未记录的请求准备过程。[统一图片流水线](../feature/2026-08-20-unified-image-request-pipeline.zh.md)负责附件归一化与请求版本，持久省略需要单独记录决策。
+
+## 决策
+
+`dsh-compaction-image-offload` 在 `agent/request-error` waterfall 上负责恢复。`IMAGE_OFFLOAD_REQUIRED` 失败提供 `offloadImages`，插件按当前 surface 顺序选择该数量最旧且仍保留的输入图片出现位置，追加一个 `image/offload` 事件，再返回 `retry`。Assistant 输出图片不参与选择。没有保留的输入图片时，插件委托后续处理。其他失败不进入这项恢复。图片省略不消耗提供方重试预算，也不产生 `llm/retry`。
+
+摘要请求不经过 agent 错误 waterfall。`dsh-compaction-basic` 保留完整的 `LlmFailure`，检查取消和选区稳定性，再携带所选事件序号触发同步的 `compaction/summary-error`。同一个图片省略插件只在选区内记录省略。同步恢复使稳定性检查与决策相邻执行。后端随后重新派生输入并计价，包括判断摘要是否缩短输入的基准。每次重试都必须继续省略图片，无图可省略时委派失败。后续摘要失败或取消仍保留已经记录的省略，因此命令错误不声称会话未变。
+
+事件载荷为 `{ targets: [{ seq, imageIndexes }] }`。每个目标指向当前的 `user/message` 或 `tool/result` 节点。索引从零开始且严格递增，按内容深度优先顺序枚举所有图片，包括嵌套工具结果和先前省略的图片。相同附件 ID 的多次出现分别计数。位置替换改变 surface 顺序后，明确选择仍没有歧义，附件 ID 或原始日志前缀无法标识这个集合。
+
+事件不带 `surfaceOp`，不创建消息，也不替换消息节点。插件的纯投影在 Session 提交事件前校验全部目标，拒绝缺失或已被遮蔽的节点、输出图片、重复目标、无效索引和已被省略的位置。重建过程派生不可变消息副本，将记录的位置标为 `ImageBlock.offloaded: true`。原始事件、消息 ID、来源和未受影响的内容块保持不变。恢复、分叉和回放从日志应用相同的选择。
+
+compaction 插件拥有选图、事件声明、图片校验与投影，以及重试策略。[插件拥有消息投影](2026-09-11-plugin-owned-message-projections.zh.md)负责通用 Session 集成和显式独立装配，取代由核心负责解释的职责划分。实例方法 `deriveEventMessage()` 与 `deriveMessages()` 返回派生消息，纯重建函数将 `foldSurface(events, projections).projectedMessages` 传给导出的 `deriveEventMessage()`。需要保留图片的改写必须读取派生消息，包括工具结果裁剪和摘要输入。
+
+消息替换和图片省略事件都会推进 `contentGeneration`，使历史缓存和请求系列刷新。`replaceGeneration` 只随替换推进，图片省略不能被判断为文本压缩成功。请求配置未变时写入原因是 `series` 的 `request/header`，配置同时变化时携带 `startsSeries: true`。
+
+适配器只准备保留的图片，将带标记的位置渲染为包含当前执行环境访问路径的 `offloadedImageText`。路由本地预算检查报告所需数量，包括内联回退的更低预算，但不编辑历史。这些检查是暂时方案，未来提供方报告无法缓存的图片时，可以替代本地计数，无需把持久决策移入适配器。
+
+Token 计量将选择应用到受影响节点的图片出现位置，不替换节点身份，也不修改已有用量锚点快照。路由计价以占位文本替代图片费用。固定结构启发式不计 `offloaded` 标记，因此图片省略不改变该启发式，也不需要 `compaction/prune` 的遮蔽价格。上下文压力和分类明细检查点版本递增，拒绝曾将该标记计入的缓存估算。
+
+`image/offload` 是读取时必须识别的事件。它增加已知事件类型，不改变 Session 信封或操作语法，因此不推进 `SESSION_FORMAT_VERSION`。不认识该类型的读取器会拒绝日志，避免静默恢复已省略的图片。
+
+## 考虑过的替代方案
+
+**复用 `compaction/prune` 和 `surfaceOp: replace`。** [已归档的替换设计](../../archived/architecture/2026-09-02-durable-image-offload.md)记录携带图片的完整消息副本，把图片省略与节点替换和遮蔽价格计量绑定。专用事件表达被省略的图片，保留消息身份，重试策略仍留在同一个插件中。
+
+**记录单次请求的投影结果。** 不影响后续派生的记录无法阻止省略的图片返回。记录完整请求体会重复消息历史，而决策只需要图片出现位置索引。
+
+**使用附件 ID 或单个分界点。** 附件可能重复出现，surface 顺序也可能与事件序号顺序不同。逐节点记录明确索引能够区分这些情况，无需在每次替换后重新定义分界点。
+
+**把选择移入循环或适配器。** 请求错误扩展点已经负责修复与重试。适配器知道路由预算，插件知道会话输入顺序，两者都不需要在共享循环中加入路由策略。
+
+## 影响
+
+路由预算增加、内联回退结束或压缩减少上下文时，省略仍然有效。省略不删除附件字节。读取占位文本中的访问路径会创建新的图片出现位置，可以独立保留。执行环境路径仍在序列化时解析，与保留图片的标识文本一致。这个事件记录选择的图片集合，不记录文件系统映射。
+
+Session 调用已注册的纯投影，并维护一个内容版本信号。重建模型输入的消费方必须使用派生消息。面向人的视图仍可从不可变事件显示原始图片。插件记录省略时不需要 token-meter 或 compaction 服务。
+
+## 测试
+
+插件的投影测试覆盖嵌套和重复图片、原子拒绝、缓存失效、纯折叠、恢复、分叉及原始事件不变。恢复测试覆盖连续失败、当前 surface 顺序、无图可省略时的恢复、新请求头和释放。Token-meter 测试在仅重计所选位置价格时保留节点身份和已有用量锚点。工具结果裁剪和压缩测试保证后续改写读取派生历史。TypeScript 和 Python SDK 预期输出通过发布的 profile 覆盖必需事件。

.agents/notes/implemented/architecture/2026-09-11-plugin-owned-message-projections.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 .agents/notes/implemented/architecture/2026-09-11-plugin-owned-message-projections.md
+2026-09-11-plugin-owned-message-projections.md: 9e1eca92567fc035d48c1c42ad86d371f2542c5b
+2026-09-11-plugin-owned-message-projections.zh.md: b26b5d4549294b37665ac933944b7f51ef7a5717

.agents/notes/implemented/architecture/2026-09-11-plugin-owned-message-projections.md

@@ -0,0 +1,35 @@
+# Agent Note: Plugin-owned message projections
+
+Status: implemented
+
+English | [中文](2026-09-11-plugin-owned-message-projections.zh.md)
+
+## Problem
+
+The dedicated [image-offload event](2026-09-10-image-offload-events.md) changes derived message content without replacing nodes. Implementing its image traversal and target validation inside Session makes each feature-specific message transformation a core change. The ordinary session-projection registry derives domain state but does not participate in canonical model-history derivation.
+
+## Decision
+
+The event-owning plugin supplies a pure `SessionMessageProjection`: one event type and a synchronous interpreter of the preceding history. It validates the complete durable payload and returns immutable updates for existing messages, preserving their identities. Session owns atomic acceptance, shared live and detached folding, and content-generation cache invalidation. It contains no image selection or image projection implementation.
+
+The `@messageProjection` tag on the owning `SessionEventMap` member generates the required-interpreter inventory. Missing definitions reject append, seeded creation, restore, and pure folding. This inventory does not make unknown third-party event names readable or change `ignorable` compatibility. A projection event cannot also declare a surface operation.
+
+The compaction-image-offload plugin registers its definition through a fiber-owned `ctx.sessions.registerMessageProjection()` effect. Each event has one registered owner. Removing a definition invalidates pending committed decisions and cached reads that used it; a replacement definition requires restoring the session. Rejected, uncommitted candidates retain no interpreter dependency.
+
+Detached readers pass definitions explicitly. The current Session format catalog assembles the same browser-safe plugin exports for persistence validation and offline queries. This static assembly does not mount recovery listeners. Live sessions use the registered composition, while compaction invariants borrow that composition's definitions. No process-global registry or import-time registration is involved.
+
+The image event payload and required-on-read semantics remain owned by the image-offload note. Its indexes, selection policy, retry behavior, and SDK recordings are unchanged; this decision partially supersedes that note's core-owned interpretation.
+
+## Alternatives considered
+
+**Keep image interpretation in Session.** That gives detached callers an implicit built-in, but turns a plugin-owned transformation into core event-specific code. Explicit assembly preserves deterministic replay without assigning the algorithm to core.
+
+**Use the ordinary session-projection registry.** Its folds expose domain state after commit. They cannot veto an invalid append or supply canonical `deriveMessages()` content without another integration point.
+
+**Transform only the outgoing request.** Compaction, fork, offline replay, and request invariants also derive messages. A send-time transformation does not cover these consumers.
+
+## Consequences
+
+Adding a content-changing event requires its declaration, pure interpreter, live registration, and detached assembly entry. Ordinary log-only events need none of this. Consumers of the pure fold use the generic `projectedMessages` result. The event envelope and stored image decisions remain unchanged, so no structural Session format version is added.
+
+Tests cover missing interpreters, generic atomic projection, loaded windows, restore and fork, duplicate registration, fiber disposal, cached and pending reads, and rejected candidates. Image-specific tests remain with their plugin. The catalog checks that every generated required type has a detached interpreter and rejects invalid image references. Existing TypeScript and Python image SDK recordings exercise the same durable event through shipped profiles.

.agents/notes/implemented/architecture/2026-09-11-plugin-owned-message-projections.zh.md

@@ -0,0 +1,35 @@
+# Agent Note: 插件拥有消息投影
+
+Status: implemented
+
+[English](2026-09-11-plugin-owned-message-projections.md) | 中文
+
+## 问题
+
+独立的[图片省略事件](2026-09-10-image-offload-events.zh.md)修改派生消息内容，不替换节点。如果图片遍历和目标校验在 Session 内实现，每种功能专用的消息转换都需要修改核心代码。普通的 session-projection 注册表派生领域状态，不参与模型历史的标准派生过程。
+
+## 决策
+
+事件所属的插件提供纯 `SessionMessageProjection`，包含一个事件类型和同步解释此前历史的处理器。它校验完整的持久载荷，返回现有消息的不可变更新，并保留消息身份。Session 负责原子接受、实时与独立回放共用的折叠，以及内容版本驱动的缓存失效。它不包含选图或图片投影实现。
+
+所属 `SessionEventMap` 成员上的 `@messageProjection` 标记生成必需处理器清单。缺少处理器时，追加、带历史的创建、恢复和纯折叠都会被拒绝。这份清单不会让未知的第三方事件名变得可读，也不改变 `ignorable` 的兼容规则。投影事件不能同时声明 surface 操作。
+
+compaction-image-offload 插件通过 fiber 拥有的 `ctx.sessions.registerMessageProjection()` effect 注册处理器。每种事件只有一个注册者。移除处理器后，尚未应用的已提交决策和使用过它的缓存读取都会失效。替换处理器需要重新恢复会话。被拒绝且未提交的候选事件不会保留对处理器的依赖。
+
+独立读取器显式传入处理器。当前 Session 格式目录装配相同的浏览器安全插件导出，供持久数据校验和离线查询使用。这项静态装配不挂载恢复监听器。实时会话使用已注册的组合，压缩 invariant 借用该组合的处理器。不使用进程全局注册表，也不在导入模块时注册。
+
+图片事件载荷和读取时必须识别的语义仍由图片省略记录负责。索引、选图策略、重试行为和 SDK 录制保持不变。本决策部分取代该记录中由核心负责解释事件的职责划分。
+
+## 考虑过的替代方案
+
+**保留 Session 内的图片解释逻辑。** 独立调用者可以隐式获得内置处理器，但插件拥有的转换会变成核心中特定事件的代码。显式装配能够保留确定性回放，无需让核心拥有算法。
+
+**使用普通的 session-projection 注册表。** 它的折叠在提交后公开领域状态。没有额外集成点，它无法否决无效追加，也无法提供标准 `deriveMessages()` 内容。
+
+**只转换发出的请求。** 压缩、分叉、离线回放和请求 invariant 也会派生消息。发送时转换无法覆盖这些消费方。
+
+## 影响
+
+增加修改内容的事件，需要事件声明、纯处理器、实时注册和独立装配项。普通的仅记录日志事件不需要这些机制。纯折叠的消费方使用通用的 `projectedMessages` 结果。事件信封和已经存储的图片决策保持不变，因此不增加结构性的 Session 格式版本。
+
+测试覆盖缺少处理器、通用原子投影、已加载窗口、恢复与分叉、重复注册、fiber 释放、缓存与待应用读取，以及被拒绝的候选事件。图片专用测试保留在插件包。目录检查每种生成的必需类型都有独立处理器，并拒绝无效图片引用。现有 TypeScript 和 Python 图片 SDK 录制通过发布的 profile 覆盖相同的持久事件。

.agents/notes/implemented/architecture/2026-09-11-sandboxed-node-ptc-runtime.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 .agents/notes/implemented/architecture/2026-09-11-sandboxed-node-ptc-runtime.md
-2026-09-11-sandboxed-node-ptc-runtime.md: 02d76bd607ab8352e208d78f3cf0442fc0794917
-2026-09-11-sandboxed-node-ptc-runtime.zh.md: 40390243b6a15404a2bd9fd2f9d5f9700ad883e6
+2026-09-11-sandboxed-node-ptc-runtime.md: 242b3cfedb49b7ab60c47c6ee03005ddb821bb33
+2026-09-11-sandboxed-node-ptc-runtime.zh.md: d1ab47550e49129ee3e1fefbdfa54cda397374a3

.agents/notes/implemented/architecture/2026-09-11-sandboxed-node-ptc-runtime.md

@@ -30,7 +30,7 @@ Program completion, timeout, cancellation and protocol failure all close executi
 
 ### Resource limits
 
-The default elapsed deadline is 120 seconds, capped at 600 seconds by default. It includes runtime setup and nested tool or approval waits. V8 old-generation memory, serialized outer output and control traffic have separate configured bounds. The heap limit excludes native allocations and descendant memory, and elapsed time is not a process-tree CPU budget.
+The default elapsed deadline is 120 seconds, capped at 600 seconds by default. Trusted service consumers may request `timeoutMs: null` to disable this timer; [workflow sandbox reuse](2026-09-13-workflow-ptc-sandbox-reuse.md) owns that caller-controlled lifetime. Omitted and numeric requests, including model-facing `run_code`, retain the numeric defaults and caps. An enabled deadline includes runtime setup and nested tool or approval waits. V8 old-generation memory, serialized outer output and control traffic have separate configured bounds. The heap limit excludes native allocations and descendant memory, and elapsed time is not a process-tree CPU budget.
 
 ## Alternatives considered
 

.agents/notes/implemented/architecture/2026-09-11-sandboxed-node-ptc-runtime.zh.md

@@ -30,7 +30,7 @@ Node worker 隔离 JavaScript 状态，但不应用调用 Session 的 OS 沙箱
 
 ### 资源限制
 
-默认经过时间截止为 120 秒，默认上限为 600 秒。它包括运行时准备以及嵌套工具或审批等待。V8 老生代内存、序列化外层输出与控制通信具有独立配置的上限。堆限制不包含原生分配和后代内存，经过时间也不是进程树 CPU 预算。
+默认经过时间截止为 120 秒，默认上限为 600 秒。可信服务消费方可以请求 `timeoutMs: null` 来禁用该定时器；[工作流沙箱复用](2026-09-13-workflow-ptc-sandbox-reuse.zh.md)负责这种由调用方控制的生命周期。省略 timeout 或使用数值的请求（包括面向模型的 `run_code`）保留数值默认值与上限。启用的截止包括运行时准备以及嵌套工具或审批等待。V8 老生代内存、序列化外层输出与控制通信具有独立配置的上限。堆限制不包含原生分配和后代内存，经过时间也不是进程树 CPU 预算。
 
 ## 考虑过的替代方案
 

.agents/notes/implemented/architecture/2026-09-12-browser-use-provider-registration.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 .agents/notes/implemented/architecture/2026-09-12-browser-use-provider-registration.md
+2026-09-12-browser-use-provider-registration.md: a0d3cfdbd798dadee6a9514dffdc619d95811c4d
+2026-09-12-browser-use-provider-registration.zh.md: b0016b6655292ccb5259adcb5a07decedda2e21e

.agents/notes/implemented/architecture/2026-09-12-browser-use-provider-registration.md

@@ -0,0 +1,53 @@
+# Agent Note: Browser-use provider registration and Session ownership
+
+Status: implemented
+
+English | [中文](2026-09-12-browser-use-provider-registration.zh.md)
+
+## Problem
+
+Browser-control backends expose different operations and observation formats. A common browser action API would constrain those experiments before a portable consumer exists. Browser sessions can be isolated, while attaching an existing logged-in browser must preserve its state and prevent concurrent ownership inside a provider.
+
+## Decision
+
+[`dsh-browser-use`](../../../../packages/browser-use/browser-use/README.md) owns `ctx.browserUse`, which registers one provider-owned name and returns its effect disposer. A second registration fails regardless of its name. The service contains no browser object, shared operation type, dispatch method, resource lifecycle, or runtime selector. The [computer-use registration decision](2026-09-12-computer-use-provider-registration.md) remains the independent owner of desktop-provider registration and shared-desktop coordination.
+
+[Playwright MCP](../../../../packages/experimental/browser-use-playwright-mcp/README.md), [Chrome DevTools MCP](../../../../packages/experimental/browser-use-chrome-devtools-mcp/README.md), and [native Stagehand](../../../../packages/experimental/browser-use-stagehand-native/README.md) own their browser tools and integrate through the normal DSH tool pipeline. They are public experimental opt-ins. DSH owns task planning and the task loop; Stagehand contributes individual AI-assisted operations. Profile or preset configuration selects launch or attachment for each provider activation.
+
+Browser resources belong to the exact live Agent and Session, not merely a reusable Session id. Calls retain state across turns. Runtime disposal closes launched resources, and reload or fork does not inherit a launched profile. Attachment preserves existing browser state and reserves the external browser exclusively for one Session within that provider instance. Cleanup disconnects without closing the external browser.
+
+The [experimental runtime helper](../../../../packages/experimental/browser-use-runtime/README.md) owns shared resource lifetime and attachment reservation without introducing those methods into the browser-use service. Canceling a caller's acquisition wait leaves initialization and its reservation owned by the Session. An active operation's Agent-disposal cancellation starts resource shutdown before the Agent waits for idle, because browser calls may settle only after their connections close. Provider teardown stops tool admission and retains its registration until resource cleanup and active calls settle. The service remains independent of all experimental packages.
+
+Stagehand's launcher inherits its process environment, and SDK initialization can time out before its cleanup settles. The provider host uses `@puppeteer/browsers` to own launched Chromium and its temporary profile before CDP or SDK readiness, with a scrubbed child environment. For both launch and attachment, an isolated Worker runs the SDK and only connects over CDP. Native inference has no abort signal. SDK close waits for active work; successful cleanup permits later reconnection while retaining browser state. Failed SDK drain blocks reuse while Chromium remains alive. Final cleanup can release a launched-browser reservation after owned Chromium and its Worker terminate. Failed SDK drain during attachment, failed Worker termination, or failed owned-process cleanup retains the reservation. The host kills only its own Chromium process and waits for child closure before removing the profile; an externally owned browser remains running.
+
+Stagehand uses an explicitly configured native model from its pinned SDK catalog. The configured API key and optional headers are forwarded into the browser extension, where Stagehand performs inference. Browser tool inputs and returned data, including SDK result metadata, use the existing Session log. DSH model routing, credential reuse, underlying inference request/response capture, and integration into Session usage accounting remain deferred; this integration adds no persistence events or Session schema changes.
+
+MCP client activation waits for connection and tool discovery, but provider activation can finish before any Session exists. That activation promise cannot represent the clients owned by future Sessions. Each MCP browser provider awaits one client startup attempt within the existing serial `agent/created` event. The [awaited Agent creation decision](2026-09-09-awaited-agent-creation.md) owns queued-input ordering and creation rollback. Successful creation or resume exposes the completed catalog to prompt assembly and direct callers.
+
+Startup failure or cancellation rejects creation or resume and triggers rollback of the Agent and its client resources. A busy attachment skips startup permanently for the activation while its other work continues; a newly created or resumed Agent can acquire the attachment after release. Late installation and reload apply only to future activations, following the [Schedule mounting policy](../../../../packages/schedule/schedule/README.md#use-this-package). Browser tools and resource requests for a successful client share the Session queue; other Sessions cannot execute those requests or receive that server's instructions.
+
+## Alternatives considered
+
+**Unified browser action API.** Playwright, Chrome DevTools, and Stagehand have different native semantics. No current consumer requires interchangeable action methods, so provider-owned tools retain those semantics.
+
+**A provider-owned startup phase.** Existing serial `agent/created` awaits Session-owned setup and makes failures visible to the creator. A separate maintenance task would duplicate that lifecycle ownership.
+
+**Discovery during prompt assembly.** Prompt and tool collection need ready registrations. Starting discovery there either exposes an incomplete catalog or requires another collection pass; awaited creation completes discovery before a turn begins.
+
+**One shared browser across Sessions.** Browser tabs, navigation, and login state can be isolated per Session. Sharing them would introduce cross-Session interference that the desktop integrations cannot generally avoid.
+
+**Fresh context when attaching.** A new browser context does not inherit the existing logged-in state. Exclusive use of the attached browser preserves the workflow that attachment enables.
+
+**Browser profiles restored with Sessions.** Durable browser state introduces profile storage and migration ownership beyond the Session log. Launched state lasts only for the live runtime; externally owned browsers retain their own persistence policy.
+
+**Only isolated launch.** Users need both clean browser sessions and access to existing authenticated state. Configuration selects the ownership policy explicitly.
+
+**Delegated browser agents.** The experiments compare browser-control backends. Delegating the whole task to another planner would alter DSH's control of the task loop.
+
+**DSH model bridge.** Native Stagehand configuration keeps its browser integration independent of DSH model-request adaptation and persistence. Session model selection, credential reuse, underlying inference capture, and Session usage accounting are deferred as one coordinated integration.
+
+## Consequences
+
+Providers evolve their tools independently while the shared service remains a name-only registry. The three providers and their runtime helper publish as experimental packages without enabling them in shipped defaults. The browser package group has no dependency on experimental runtime code.
+
+Attachment reservations apply within one provider instance; they do not coordinate separate DSH processes or external browser clients. Browser state is absent from Session replay, and cancellation does not undo delivered browser actions. Provider READMEs own engine support, model requirements, and upstream restrictions.

.agents/notes/implemented/architecture/2026-09-12-browser-use-provider-registration.zh.md

@@ -0,0 +1,53 @@
+# Agent Note: 浏览器操作提供方注册与 Session 所有权
+
+Status: implemented
+
+[English](2026-09-12-browser-use-provider-registration.md) | 中文
+
+## 问题
+
+浏览器控制后端暴露不同的操作与观测格式。在可移植消费方存在之前，通用浏览器动作 API 会约束这些实验。浏览器会话可以隔离，而附加现有已登录浏览器必须保留其状态，并防止一个提供方内出现并发所有权。
+
+## 决策
+
+[`dsh-browser-use`](../../../../packages/browser-use/browser-use/README.zh.md) 拥有 `ctx.browserUse`，注册一个提供方拥有的名称并返回其 effect 清理器。第二次注册无论名称为何都会失败。服务不包含浏览器对象、共享操作类型、分派方法、资源生命周期或运行时选择器。[计算机操作注册决策](2026-09-12-computer-use-provider-registration.zh.md)仍独立拥有桌面提供方注册与共享桌面协调规则。
+
+[Playwright MCP](../../../../packages/experimental/browser-use-playwright-mcp/README.zh.md)、[Chrome DevTools MCP](../../../../packages/experimental/browser-use-chrome-devtools-mcp/README.zh.md) 与[原生 Stagehand](../../../../packages/experimental/browser-use-stagehand-native/README.zh.md) 拥有自己的浏览器工具，并通过常规 DSH 工具管线集成。它们是公共实验性可选功能。DSH 拥有任务规划与任务循环；Stagehand 提供单项 AI（人工智能）辅助操作。Profile 或 preset 配置为每次提供方激活选择启动或附加模式。
+
+浏览器资源属于确切的实时 Agent 与 Session，而非仅凭可复用的 Session id。调用跨轮次保留状态。运行时释放会关闭启动的资源，重新加载或 fork 不会继承启动的 profile。附加保留现有浏览器状态，并在该提供方实例内将外部浏览器独占保留给一个 Session。清理断开连接而不关闭外部浏览器。
+
+[实验性运行时辅助库](../../../../packages/experimental/browser-use-runtime/README.zh.md)拥有共享资源生命周期与附加保留机制，而不向浏览器操作服务引入这些方法。取消调用方对资源获取的等待后，初始化及其保留仍归 Session 所有。活动操作收到 Agent 释放的取消信号时，在 Agent 等待空闲之前启动资源关闭，因为浏览器调用可能只有在连接关闭后才能结束。提供方清理停止接收工具调用，并保留注册，直到资源清理与活动调用完成。服务保持独立于所有实验包。
+
+Stagehand 的启动器继承其进程环境，SDK 初始化可能在清理完成前超时。提供方 Host 使用 `@puppeteer/browsers` 在 CDP 或 SDK 就绪前取得所启动 Chromium 及其临时 profile 的所有权，并清理子进程环境。启动和附加模式都由隔离的 Worker 运行 SDK，且只通过 CDP 连接。原生推理不接受 abort signal。SDK 关闭会等待活动工作；清理成功后可重新连接并保留浏览器状态。SDK 工作未能结束时，只要 Chromium 仍在运行，就阻止复用。最终清理可在自有 Chromium 和 Worker 终止后释放启动浏览器的占用。附加模式下 SDK 工作未能结束、Worker 终止失败或自有进程清理失败时，保留占用。Host 仅终止自己拥有的 Chromium 进程，等待子进程关闭后才删除 profile；外部拥有的浏览器保持运行。
+
+Stagehand 使用显式配置的固定版本 SDK 目录内原生模型。配置的 API 密钥与可选请求头会转发到浏览器扩展，由 Stagehand 在扩展内执行推理。浏览器工具输入和返回数据使用现有 Session 日志，包括 SDK 结果元数据。DSH 模型路由、凭据复用、底层推理请求/响应捕获，以及与 Session 用量计量的集成仍属暂缓工作；本集成不增加持久化事件或 Session schema 变更。
+
+MCP 客户端激活会等待连接和工具发现，但提供方可能在任何 Session 存在之前就完成激活。该激活 promise 无法代表未来各 Session 拥有的客户端。每个 MCP 浏览器提供方在现有的串行 `agent/created` 事件中等待一次客户端启动尝试。[等待 Agent 创建的决策](2026-09-09-awaited-agent-creation.zh.md)负责排队输入顺序与创建回滚。创建或恢复成功后，提示词组装与直接调用方即可读取完整目录。
+
+启动失败或取消会拒绝创建或恢复，并触发 Agent 及其客户端资源的回滚。附加连接被占用时，本次激活永久跳过启动，但其他工作继续运行；连接释放后，新创建或恢复的 Agent 可以获取连接。较晚安装和重新加载只作用于后续激活，与 [Schedule 的挂载策略](../../../../packages/schedule/schedule/README.zh.md#use-this-package)一致。成功客户端的浏览器工具与资源请求共享 Session 队列；其他 Session 不能执行这些请求，也不能收到该服务器的指导。
+
+## 考虑过的替代方案
+
+**统一浏览器动作 API。** Playwright、Chrome DevTools 与 Stagehand 的原生语义不同。当前没有消费方要求可互换的动作方法，因此由提供方拥有工具以保留这些语义。
+
+**提供方自有的启动阶段。** 现有的串行 `agent/created` 等待 Session 自有设置，并向创建方报告失败。独立维护任务会重复这份生命周期所有权。
+
+**在提示词组装期间发现。** 提示词和工具收集需要已就绪的注册。此时启动发现要么暴露不完整目录，要么要求再次收集；等待创建会在轮次开始前完成发现。
+
+**多个 Session 共享一个浏览器。** 浏览器标签页、导航与登录状态可以按 Session 隔离。共享它们会引入桌面集成通常无法避免的跨 Session 干扰。
+
+**附加时创建全新 context。** 新浏览器 context 不继承现有登录状态。独占使用附加的浏览器可保留附加所支持的工作流。
+
+**随 Session 恢复浏览器 profile。** 持久浏览器状态引入 Session 日志之外的 profile 存储与迁移所有权。启动的状态仅在实时运行时存续；外部拥有的浏览器保留自己的持久化策略。
+
+**仅支持隔离启动。** 用户既需要干净的浏览器会话，也需要访问现有身份验证状态。配置显式选择所有权策略。
+
+**委托浏览器 agent。** 这些实验比较浏览器控制后端。将整个任务委托给另一规划器会改变 DSH 对任务循环的控制。
+
+**DSH 模型桥接。** 原生 Stagehand 配置使浏览器集成独立于 DSH 模型请求适配和持久化。Session 模型选择、凭据复用、底层推理捕获与 Session 用量计量作为一个协调集成暂缓。
+
+## 影响
+
+提供方独立演进自己的工具，而共享服务保持仅注册名称。三个提供方及其运行时辅助库作为实验性包发布，但不在内置默认配置中启用。浏览器包组不依赖实验性运行时代码。
+
+附加保留机制作用于单个提供方实例；它不协调独立 DSH 进程或外部浏览器客户端。Session 重放不包含浏览器状态，取消不会撤销已交付的浏览器操作。提供方 README 拥有引擎支持、模型要求与上游限制。

.agents/notes/implemented/architecture/2026-09-13-workflow-ptc-sandbox-reuse.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 .agents/notes/implemented/architecture/2026-09-13-workflow-ptc-sandbox-reuse.md
+2026-09-13-workflow-ptc-sandbox-reuse.md: 9454ffef28e1ab5ba7791a30f4bcd67093a34ba3
+2026-09-13-workflow-ptc-sandbox-reuse.zh.md: c51706065703866520bfcad11beffc02eb1f5551

.agents/notes/implemented/architecture/2026-09-13-workflow-ptc-sandbox-reuse.md

@@ -0,0 +1,41 @@
+# Agent Note: Reuse the PTC Node sandbox for workflows
+
+Status: implemented
+
+English | [中文](2026-09-13-workflow-ptc-sandbox-reuse.zh.md)
+
+## Problem
+
+Dynamic workflows evaluate model-written JavaScript and start subagents. A worker thread keeps script execution off the host event loop, but code escaping its VM can use Node with the host process's file authority. PTC already owns a Node process implementation with OS file confinement, isolated program state, bounded output and control traffic, and managed cleanup. Maintaining a second launcher would duplicate those responsibilities.
+
+## Decision
+
+`dsh-workflow-ptc` implements `WorkflowEngine` through the shared Node `PtcRuntime`. Each run keeps the existing VM and workflow helpers inside one PTC process. Host bindings connect the guest to the configured subagent provider and workflow observers; the host supplies the calling Agent and resolves its Session's standing file policy and cwd.
+
+The VM defines the helper API and cooperative concurrency, total-agent and item caps. It is not a security boundary, and those counters are not host-enforced security quotas. File enforcement, V8 heap limits, output and control limits, and managed process cleanup remain owned by PTC and its sandbox/subprocess providers. Network access and provider-specific containment limits remain the same as PTC.
+
+Workflow execution passes `timeoutMs: null`, which explicitly disables the elapsed timer in the Node runtime. Omitted and numeric PTC requests keep their configured defaults and caps; `run_code` continues to accept only positive numeric overrides. The initial VM slice retains its own synchronous timeout. A caller's abort signal, including an enclosing tool deadline, still cancels the workflow.
+
+Cancellation immediately aborts the PTC process and the signal shared by pending and active child agents. The adapter awaits pending starts and child disposal, including a child that publishes after cancellation. PTC stops the program; its caller remains responsible for host bindings already in flight. There is no additional workflow cleanup timer or guest cancellation acknowledgement.
+
+Progress uses one binding call at a time. The first batch starts synchronously; later events queue in order and drain before child disposal and the final result. This prevents ordinary log bursts from exhausting PTC's pending-call limit. Child-result waits stop on cancellation while child disposal remains awaited.
+
+The Node bootstrap keeps its control pipe open after sending the terminal frame until the host closes it. Unawaited binding replies may still be in flight, so eager child-side close would let an `EPIPE` race an already completed program.
+
+The [dynamic-workflows decision](../feature/2026-07-05-dynamic-workflows.md) retains the script, structured-output, event and tool semantics; this note supersedes only its execution substrate and trust realization. The [sandboxed Node PTC decision](2026-09-11-sandboxed-node-ptc-runtime.md) retains execution and control guarantees; the explicit null deadline extends its service options. The [agent-scope runtime design](2026-07-12-agent-scope-runtime-design.md#workflow-children-are-pending-starts-or-published-records) retains pending-start and child-cleanup ownership.
+
+## Alternatives considered
+
+**Retain worker-thread execution.** Worker termination cannot apply the Session's OS file policy or provide the managed process cleanup already required by direct Node code.
+
+**Create a separate workflow subprocess runtime.** Another launcher, control transport and sandbox adapter would maintain the same execution responsibilities twice. PTC already accepts programs and named asynchronous host bindings without knowing about tools or Sessions.
+
+**Replace the VM with direct Node workflow APIs.** The VM and helpers preserve the existing script semantics, synchronous-slice timeout and JSON materialization. Removing them is unnecessary for OS confinement.
+
+**Apply PTC's numeric deadline to workflows.** A workflow may await a long series of subagents. Explicit `null` preserves caller-controlled lifetime without changing ordinary PTC defaults or treating an arbitrarily large number as no deadline.
+
+## Consequences
+
+Workflow and opt-in Ralph execution share PTC's security and process lifecycle implementation. Ralph remains disabled in shipped defaults. Scripts still use the same hooks and result envelope; no new authoritative progress ledger or host child-count quota is introduced.
+
+Cancellation does not wait for cooperative script progress. The adapter waits for child cleanup, so a subagent provider that does not fulfill its lifecycle contract can delay disposal. Process cleanup retains the selected subprocess provider's managed-range limitations; this change does not claim process-tree CPU/RSS accounting or universal descendant termination.

.agents/notes/implemented/architecture/2026-09-13-workflow-ptc-sandbox-reuse.zh.md

@@ -0,0 +1,41 @@
+# Agent Note: 工作流复用 PTC Node 沙箱
+
+Status: implemented
+
+[English](2026-09-13-workflow-ptc-sandbox-reuse.md) | 中文
+
+## 问题
+
+动态工作流执行模型编写的 JavaScript 并启动 subagent。worker 线程把脚本执行移出宿主事件循环，但逃逸 VM 的代码可以使用 Node 并拥有宿主进程的文件权限。PTC 已有 Node 进程实现，负责 OS 文件约束、隔离的程序状态、有界输出与控制通信，以及受管清理。维护第二套启动器会重复这些职责。
+
+## 决策
+
+`dsh-workflow-ptc` 通过共享的 Node `PtcRuntime` 实现 `WorkflowEngine`。每次运行在一个 PTC 进程中保留既有 VM 与工作流辅助函数。Host 绑定将 guest 连接到配置的 subagent 提供方及工作流观察器；Host 提供发起调用的 Agent，并解析其 Session 的常设文件策略与 cwd。
+
+VM 定义辅助 API，以及协作式并发、agent 总数和条目上限。它不是安全边界，这些计数器也不是 Host 强制的安全配额。文件强制、V8 堆限制、输出与控制限制、受管进程清理仍由 PTC 及其沙箱／子进程提供方负责。网络访问与提供方特有的约束限制保持与 PTC 相同。
+
+工作流执行传入 `timeoutMs: null`，显式禁用 Node 运行时的经过时间定时器。省略 timeout 或使用数值的 PTC 请求保留配置的默认值与上限；`run_code` 仍只接受正数覆盖值。最初的 VM 片段保留独立的同步超时。调用方的中止信号（包括外层工具截止）仍会取消工作流。
+
+取消立即中止 PTC 进程，以及待启动和活跃子 agent 共享的信号。适配器等待待完成启动与子 agent 资源释放，包括取消后才发布的子 agent。PTC 停止程序；已经进行中的 Host 绑定仍由其调用方负责。不增加工作流清理定时器，也不等待 guest 取消确认。
+
+进度同时只使用一个绑定调用。首批同步发起，后续事件按序排队，并在子 agent 资源释放及最终结果之前完成传递。这避免普通日志突发耗尽 PTC 的待完成调用上限。取消会停止对子 agent 结果的等待，但仍等待子 agent 资源释放。
+
+Node 引导程序发送终态帧后保持控制管道打开，直到 Host 将其关闭。未被等待的绑定回复可能仍在传输，因此子进程提前关闭会让 `EPIPE` 与已经完成的程序结果发生竞争。
+
+[动态工作流决策](../feature/2026-07-05-dynamic-workflows.zh.md)保留脚本、结构化输出、事件与工具语义；本文只取代其执行基底与信任实现。[沙箱化 Node PTC 决策](2026-09-11-sandboxed-node-ptc-runtime.zh.md)保留执行与控制保证；显式 null 截止扩展其服务选项。[agent 作用域运行时设计](2026-07-12-agent-scope-runtime-design.zh.md#workflow-children-are-pending-starts-or-published-records)保留待启动与子 agent 清理的归属规则。
+
+## 曾考虑的替代方案
+
+**保留 worker-thread 执行。** 终止 worker 无法应用 Session 的 OS 文件策略，也无法提供直接 Node 代码所需的受管进程清理。
+
+**创建独立的工作流子进程运行时。** 另一套启动器、控制传输与沙箱适配器会重复维护相同执行职责。PTC 已经接受程序和具名异步 Host 绑定，无需了解工具或 Session。
+
+**用直接 Node 工作流 API 替换 VM。** VM 与辅助函数保留既有脚本语义、同步片段超时与 JSON 物化。实现 OS 约束不需要移除它们。
+
+**对工作流应用 PTC 的数值截止。** 工作流可能等待一长串 subagent。显式 `null` 保留调用方控制的生命周期，不改变普通 PTC 默认值，也不把任意大的数值当作没有截止。
+
+## 后果
+
+工作流与显式启用的 Ralph 执行共享 PTC 的安全与进程生命周期实现。Ralph 在已发布默认组合中保持禁用。脚本仍使用相同钩子与结果信封；不增加新的权威进度台账或 Host 子 agent 数量配额。
+
+取消不等待脚本协作推进。适配器等待子 agent 清理，因此不履行生命周期约定的 subagent 提供方可能延迟资源释放。进程清理保留所选子进程提供方对受管范围的限制；本变更不宣称进程树 CPU/RSS 计量或普遍的后代终止保证。

.agents/notes/implemented/bug-fix/2026-09-13-partitioned-coverage-location-canonicalization.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 .agents/notes/implemented/bug-fix/2026-09-13-partitioned-coverage-location-canonicalization.md
+2026-09-13-partitioned-coverage-location-canonicalization.md: 54e5aaefe9551d1357f1c60201212873a1c7aef7
+2026-09-13-partitioned-coverage-location-canonicalization.zh.md: 1aa4b8de844a1f0e544d2bbb223641dbbfeff3da

.agents/notes/implemented/bug-fix/2026-09-13-partitioned-coverage-location-canonicalization.md

@@ -0,0 +1,43 @@
+# Agent Note: Canonicalize partition coverage locations before the blob merge
+
+Status: implemented
+
+English | [中文](2026-09-13-partitioned-coverage-location-canonicalization.zh.md)
+
+## Problem
+
+The coverage gate reported [packages/util/home-paths/src/index.ts](../../../../packages/util/home-paths/src/index.ts) at 96.96% statements with one uncovered statement at `48:22`, while branches, functions, and lines stayed at 100% and the same tests reported the file at 100% in an unpartitioned run. Line 48 holds one statement, so the merged report counted a second, unhit statement the source does not contain. The branch that exposed the failure only added a jsdom suite that loads a node-tested module; it did not touch that module or its package.
+
+A source file reaches the merged report through one statement map per Vite environment, and the serialized partition blob strips the location data istanbul-lib-coverage reconciles those maps with. A file executed under two environments can therefore fail the per-file 100% gate although every statement ran, and the failure follows the test environments and the partition count rather than the file.
+
+## Decision
+
+[scripts/coverage-partitions.ts](../../../../scripts/coverage-partitions.ts) passes every partition `--reporter=./scripts/coverage-canonical-locations.ts`, whose `onCoverage` hook rewrites each non-finite end column of the finished run's coverage map to `Number.MAX_SAFE_INTEGER`. That column keeps the meaning of a location that ends at its line's end, serializes as a number, and keys identically in every blob, so the merge command reconciles environment-specific spellings exactly as an in-process merge does. The per-file 100% gate keeps its full strength for every file `coverage.include` matches: the canonicalization adds hits through istanbul's containment rule and never removes a statement from the report. A payload that carries no istanbul `data` record fails the partition instead of leaving every location uncanonicalized.
+
+[scripts/coverage-uncovered-locations.cjs](../../../../scripts/coverage-uncovered-locations.cjs) reads the same column as a line end, so an uncovered record prints the same `path:line:col` with or without the canonicalization. The [in-job partitioned coverage](../process/2026-08-18-in-job-partitioned-coverage.md) coordinator owns the partition and merge commands this reporter joins, and it keeps its single merged threshold check.
+
+## Divergent statement maps across Vite environments
+
+A node suite maps a source file through the `ssr` environment and a `@vitest-environment jsdom` suite maps it through the `client` environment. The AST-based V8 remapper positions a statement at the node it finds in the transformed code, so one declaration enters the merged map twice: at its declared identifier from the ssr transform, at the nested call expression of the client transform. istanbul-lib-coverage attributes the hits of the narrowest containing range to an entry no other record names, which covers whichever spelling the client record introduces.
+
+That reconciliation runs on locations `getLoc()` accepts, which requires numeric line and column values. ast-v8-to-istanbul ends a whole-line statement at column `Infinity`; a partition blob serializes `Infinity` as `null`, so the merge command holds a location it cannot compare and keeps the client-only spelling as an extra, unhit statement. The same records merged inside one process, where `Infinity` survives, report no such statement.
+
+## Testing
+
+`scripts/coverage-partitions.spec.ts` merges two records that spell one statement both ways through the JSON hop a blob performs, and asserts that the canonicalized merge reports no uncovered statement where the raw merge reports the phantom. A second case pins the canonicalization across statement, function, and branch locations, a third pins the canonicalizing reporter on every partition command, and a fourth pins the loud rejection of a payload that carries no coverage data.
+
+## Alternatives considered
+
+**Drop the untested-file maps from partitions.** Rejected because those maps are what fails a file no test runs; removing them silences a real coverage gap in exchange for the phantom.
+
+**Reconcile statements by line in the merge command.** Rejected because several statements can share a line, so line-level merging hides genuinely uncovered code.
+
+**Keep jsdom suites from loading node-only modules.** Rejected because attribution must not depend on which environment a suite happens to load a module in, and any later cross-environment load would return the defect.
+
+**Repair the merged map in the report phase.** Rejected because the records are already fused by then, so re-adding hits cannot separate a phantom spelling from a genuinely unhit nested statement.
+
+**Exempt the file or relax the per-file gate.** Rejected because the file is covered and the gate is correct; the attribution was not.
+
+## Consequences
+
+Partitioned runs attribute for a cross-environment file what one process attributes, so the gate holds 100% per file without exempting anything. Canonicalization runs inside partitions only, which leaves unpartitioned runs and their reports as they are. Blobs carry a finite sentinel column for a line-end position, and both the partition reporter and the uncovered-locations reporter name that sentinel as the line-end convention. The canonicalization also depends on Vitest's reporter ordering: the blob reporter stores the map in its own `onCoverage` and serializes it in `onTestRunEnd`, so a Vitest upgrade that reordered those hooks would first show up as the phantom statement returning.

.agents/notes/implemented/bug-fix/2026-09-13-partitioned-coverage-location-canonicalization.zh.md

@@ -0,0 +1,43 @@
+# Agent Note: 在 blob 合并前规范化分区覆盖率位置
+
+Status: implemented
+
+[English](2026-09-13-partitioned-coverage-location-canonicalization.md) | 中文
+
+## 问题
+
+覆盖率门禁把 [packages/util/home-paths/src/index.ts](../../../../packages/util/home-paths/src/index.ts) 报为语句 96.96%，并在 `48:22` 标出 1 条未覆盖语句，而分支、函数与行都保持 100%，同一批测试在非分区运行中把该文件报为 100%。第 48 行只有 1 条语句，因此合并报告计入了源码中并不存在的第 2 条未命中语句。暴露该失败的分支只新增了一个加载受节点侧测试覆盖模块的 jsdom 套件，并未改动该模块或其所属包。
+
+一个源文件在每个 Vite 环境中各有一份语句映射并据此进入合并报告，而序列化后的分区 blob 抹掉了 istanbul-lib-coverage 用来调和这些映射的位置数据。因此在两个环境中执行过的文件可能在每条语句都跑到的情况下仍未通过逐文件 100% 门禁，且该失败跟随测试环境与分区数量，而不跟随文件本身。
+
+## 决策
+
+[scripts/coverage-partitions.ts](../../../../scripts/coverage-partitions.ts) 为每个分区传入 `--reporter=./scripts/coverage-canonical-locations.ts`；该报告器的 `onCoverage` 钩子把当次运行覆盖率映射中所有非有限的结束列改写为 `Number.MAX_SAFE_INTEGER`。该列保留“位置结束于所在行行尾”的含义，序列化后仍是数字，并且在每个 blob 中生成相同的键，因此合并命令会像进程内合并那样调和各环境特有的写法。逐文件 100% 门禁对 `coverage.include` 命中的每个文件保持完整强度：规范化只会通过 istanbul 的包含规则增加命中，绝不会把语句移出报告。载荷若不含 istanbul 的 `data` 记录，分区会直接失败，而不是让所有位置保持未规范化。
+
+[scripts/coverage-uncovered-locations.cjs](../../../../scripts/coverage-uncovered-locations.cjs) 把同一列读作行尾，因此无论是否经过规范化，未覆盖记录打印的 `path:line:col` 都相同。[单 job 分区覆盖率](../process/2026-08-18-in-job-partitioned-coverage.zh.md)协调器拥有该报告器所加入的分区与合并命令，并保留其唯一一次合并阈值判定。
+
+## 跨 Vite 环境的语句映射分歧
+
+节点侧套件把源文件映射到 `ssr` 环境，`@vitest-environment jsdom` 套件把它映射到 `client` 环境。基于 AST 的 V8 重映射器按它在转换后代码中找到的节点定位语句，因此同一条声明会以两种写法进入合并映射：ssr 转换给出其声明标识符的位置，client 转换给出其嵌套调用表达式的位置。istanbul-lib-coverage 会把最窄包含范围的命中计入其他记录都未命名的条目，从而覆盖 client 记录引入的那种写法。
+
+该调和只作用于 `getLoc()` 接受的位置，这要求行列值都是数字。ast-v8-to-istanbul 把整行语句的结束列标为 `Infinity`；分区 blob 会把 `Infinity` 序列化为 `null`，于是合并命令拿到无法比较的位置，并把仅存在于 client 的写法保留为额外的未命中语句。同样的记录在单个进程内合并时 `Infinity` 得以保留，因此不会报出该语句。
+
+## 验证
+
+`scripts/coverage-partitions.spec.ts` 让两条以两种写法表示同一语句的记录经过 blob 实际执行的 JSON 跳转后再合并，断言规范化后的合并没有未覆盖语句，而未经规范化的合并会报出该幻影语句。第二个用例固定语句、函数与分支位置的规范化，第三个用例固定每个分区命令都带上该规范化报告器，第四个用例固定对不含覆盖率数据的载荷的显式拒绝。
+
+## 曾考虑的替代方案
+
+**在分区中丢弃未测试文件映射。** 不予采用，因为这些映射正是让没有任何测试执行的文件失败的依据；去掉它们等于用真实的覆盖率缺口换取该幻影语句。
+
+**在合并命令中按行调和语句。** 不予采用，因为同一行可以承载多条语句，按行合并会掩盖真正未覆盖的代码。
+
+**禁止 jsdom 套件加载仅节点侧模块。** 不予采用，因为归因不能取决于套件恰好把模块加载到哪个环境，而且此后任何跨环境加载都会让该缺陷重现。
+
+**在报告阶段修复合并后的映射。** 不予采用，因为此时各记录已经融合，补回命中无法区分幻影写法与真正未命中的嵌套语句。
+
+**豁免该文件或放宽逐文件门禁。** 不予采用，因为该文件确有覆盖，门禁本身正确，错的是归因。
+
+## 后果
+
+分区运行对跨环境文件给出与单进程相同的归因，因此门禁在不豁免任何文件的前提下保持逐文件 100%。规范化只在分区内部运行，未分区运行及其报告保持原样。blob 在行尾位置携带一个有限哨兵列，分区报告器与未覆盖位置报告器都把该哨兵命名为行尾约定。该规范化还依赖 Vitest 的报告器次序：blob 报告器在自己的 `onCoverage` 中保存该映射，在 `onTestRunEnd` 中序列化它，因此升级 Vitest 后若次序改变，最先表现为幻影语句复现。

.agents/notes/implemented/feature/2026-06-30-interception-extension-points.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 .agents/notes/implemented/feature/2026-06-30-interception-extension-points.md
-2026-06-30-interception-extension-points.md: d628cdce0bc6c13a90a25eeebcfd977602dd76e3
-2026-06-30-interception-extension-points.zh.md: 689d7f6ee6d5a552c2eaf950ce5930be559eb5bb
+2026-06-30-interception-extension-points.md: 5adffd1679c97cc69cf18991913113b84d88ba3c
+2026-06-30-interception-extension-points.zh.md: c3aa374a22d58c5185a3bd8e79b4040a04bd43e3

.agents/notes/implemented/feature/2026-06-30-interception-extension-points.md

@@ -15,7 +15,7 @@ The surface needs distinct contracts for per-prompt policy (CC's `UserPromptSubm
 The canonical surface separates transformable policy, around-dispatch control, and observe-only notification. Policy waterfalls return small extension-point-specific **typed Decision unions**; wrappers return normalized results; notifications receive immutable snapshots and cannot affect the outcome. The set covers the hook points in scope (`session-start`, `prompt-submit`, `pre-tool`, `post-tool`, `stop`-via-continuation) while leaving non-hook execution policy independently composable.
 
 **Agent events** (`dsh-agent`):
-- `agent/session-start({ agent, source })` — emit, once before turn 1, carrying a `SessionStartSource` (`startup` for a fresh/forked create, `resume` for a reloaded persisted session; `clear`/`compact` reserved). A pure notification — it CANNOT block startup (a deliberate gap: a bridge logs/injects, it does not gate startup). A listener seeds context via `agent.inject()`.
+- `agent/created({ agent, source, signal? })` — serial per-agent initialization before the first turn, carrying `SessionStartSource`. Listeners may install tools and seed context through `agent.inject()`; creation awaits them and rejects on failure. The [awaited creation decision](../architecture/2026-09-09-awaited-agent-creation.md) owns this timing and supersedes the observe-only startup policy.
 - `agent/pre-step({ agent, messages, turn, step, signal }, next) → PreStepDecision` — waterfall, fired before every proposed step after the loop has atomically removed its exclusive inbox batch. The payload carries the request's `turn`, `step`, and cancellation `signal` (the retired `PreStepContext` fields live in the payload; see the [payload-object events decision](../../archived/architecture/2026-08-06-agent-event-payload-objects.md)); `messages` is empty for a tool continuation with no intervening input. `enter` returns the complete message batch, including any current-request context a listener contributes; `reject` opens no step and leaves the claimed messages removed.
 
 **`agent/turn-stopping`** is an awaited notification at the natural stop boundary. A listener that needs another step calls `agent.steer()` with explicitly sourced model-facing content; the loop then re-reads the outbox and either continues or closes the turn.

.agents/notes/implemented/feature/2026-06-30-interception-extension-points.zh.md

@@ -15,7 +15,7 @@ harness 需要一套钩子子系统：用户像 Claude Code（CC）和 Codex 那
 规范接口将可变换策略、环绕调度控制与仅观测通知分离。策略 waterfall（瀑布式事件）返回小型的、扩展点专属的**类型化 Decision 联合类型**；包装层返回规范化结果；通知接收不可变快照，无法影响结果。覆盖的钩子点包括 `session-start`、`prompt-submit`、`pre-tool`、`post-tool`、通过 continuation 实现的 `stop`，同时将非钩子的执行策略留作独立可组合。
 
 **Agent 事件**（`dsh-agent`）：
-- `agent/session-start({ agent, source })` ——emit，在第 1 轮次之前触发一次，携带 `SessionStartSource`（`startup` 表示全新/fork 创建，`resume` 表示重新加载的持久化会话；`clear`/`compact` 保留）。纯通知，不能阻塞启动（这是有意的空白：桥接可以记录/注入，但不管控启动）。监听器通过 `agent.inject()` 注入上下文。
+- `agent/created({ agent, source, signal? })` ——首个轮次前针对单个 agent 的串行初始化，携带 `SessionStartSource`。监听器可以安装工具，并通过 `agent.inject()` 注入上下文；创建等待监听器完成，失败时拒绝。[可等待创建决策](../architecture/2026-09-09-awaited-agent-creation.zh.md)拥有这一时序，并取代只观察启动的策略。
 - `agent/pre-step({ agent, messages, turn, step, signal }, next) → PreStepDecision` ——waterfall，在每个拟议步骤之前、循环原子移除其独占 inbox 批次后触发。payload 携带该请求的 `turn`、`step` 与取消 `signal`（已退役的 `PreStepContext` 字段位于 payload 中；参见 [payload-object 事件决策](../../archived/architecture/2026-08-06-agent-event-payload-objects.md)）；没有中途输入的工具续步会收到空批次。`enter` 返回完整消息批次，其中包括监听器为当前请求贡献的上下文；`reject` 不打开步骤，并让已领取消息保持已删除。
 
 **`agent/turn-stopping`** 是自然停止边界上的一次 awaited 通知。需要再执行一步的监听器调用 `agent.steer()`，传入来源显式的 steering（中途引导）内容供模型使用；循环随后重新读取 outbox，继续执行或关闭轮次。

.agents/notes/implemented/feature/2026-07-05-dynamic-workflows.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 .agents/notes/implemented/feature/2026-07-05-dynamic-workflows.md
-2026-07-05-dynamic-workflows.md: 6d7dc43121df251d71a3305c083055b43d63f0f5
-2026-07-05-dynamic-workflows.zh.md: aaf527637f6fb5f58e088ab6021f5dcb9d4e0b5a
+2026-07-05-dynamic-workflows.md: c7245704b0734c1508fe7f001e6429f594ebdd1d
+2026-07-05-dynamic-workflows.zh.md: a4502d7d6b3acf1aebc4a5b36caccf50859bab01

.agents/notes/implemented/feature/2026-07-05-dynamic-workflows.md

@@ -22,19 +22,13 @@ One deliberate strictness DIVERGENCE from CC: hook misuse — unknown or deferre
 
 `ctx.workflowEngine` is an abstract `WorkflowEngine` in the bash shape — one engine per context, no named-provider registry (engines are deployment swaps, not co-residents). `start(request)` throws synchronously for a script that cannot begin; a returned `WorkflowRun`'s `result` NEVER rejects (failures resolve as `stopReason: 'error' | 'cancelled'`). The `workflow/*` events are observe-only emits carrying DATA SNAPSHOTS (id + meta; `workflow/end` omits the result value), per-listener contained, mirroring `subagent/start`/`subagent/end` — control stays with the run's holder. Vocabulary details: [subsystems/workflow.md](../../../../docs/subsystems/workflow.md).
 
-### The engine (dsh-workflow-worker-thread): one worker thread per run
+### The engine (dsh-workflow-ptc): shared Node process execution
 
-**Trust premise**: workflow scripts have the same trust as the model's bash access. The engine contains buggy scripts and guarantees settled results, JSON-safe values, and cancellation quiescence; it does not defend against hostile code. A vm context and worker thread are not security boundaries: a script can escape to Node APIs with process-wide authority. Sandboxing requires a separate-process or isolated-vm engine behind this seam.
+The [workflow sandbox reuse decision](../architecture/2026-09-13-workflow-ptc-sandbox-reuse.md) supersedes the worker-thread execution and trust realization. The engine retains the VM and helpers inside a sandboxed PTC Node process. The VM defines the script API; OS file policy and managed process cleanup belong to the shared execution provider.
 
-**Why `node:worker_threads`**: each run gets one unpooled worker. A vm context limits the documented script API, while message-port RPC bridges `agent()` to host-side child loops. The worker prevents synchronous script work from blocking the host, provides a serialization boundary, and permits forced termination after cancellation. `isolated-vm` was rejected because of its maintenance state and deployment requirements.
+The host validates metadata and parses the body before publication. Host bindings connect the guest to subagents and workflow observers. Pending starts and published child records share a cancellation signal; the [agent-scope runtime-design Agent Note](../architecture/2026-07-12-agent-scope-runtime-design.md#workflow-children-are-pending-starts-or-published-records) owns their lifecycle rules.
 
-The host validates metadata and parses the body before publication. Private enum-keyed payload maps define the wire protocol; pending starts, published child records, one cancellation signal, worker-death reaping, result precedence, and disposal quiescence preserve the subagent run contract across it. The [agent-scope runtime-design Agent Note](../architecture/2026-07-12-agent-scope-runtime-design.md#workflow-children-are-pending-starts-or-published-records) owns those race algorithms.
-
-The engine exposes an in-process `MessageChannel` test path because main-process V8 coverage cannot see worker execution.
-
-**Meta is data**: the schema-validated `meta` field reaches the seam as JSON and is only shape-validated. The host never evaluates a metadata literal, which would let script-controlled accessors run outside the worker's isolation.
-
-**Value boundary**: `materializeFromRealm` copies outbound values and rejects functions, symbols, nested `undefined`, exotic prototypes, cycles, sparse arrays, and non-finite numbers. Data-property copies make `"__proto__"` safe; getters are read normally and a throwing getter fails loudly. `args` crosses through `workerData` and is cloned again before exposure. Realm functions are invoked rather than copied, and thrown values use a total renderer so `result` cannot reject. Hook errors are host-realm `WorkflowError`s, so scripts branch on `name` or `code` rather than `instanceof Error`, as documented in the engine README. Concurrency, total-agent, item, timeout, and grace limits are validated config.
+**Meta is data**: the host never evaluates a metadata literal. **Values are lossless JSON**: guest-side realm materialization rejects unsupported values before PTC transport. Getters run inside the confined process, and hook errors retain their stable `name` and `code` fields across realms. Cooperative helper caps and the initial synchronous-slice timeout remain; no overall workflow elapsed timer is added.
 
 ### The Consumer (`dsh-tool-workflow`)
 
@@ -52,7 +46,7 @@ An output schema makes a schema-valid committed capture mandatory for successful
 
 ## Testing
 
-Worker-side logic runs through an in-process `MessageChannel` so V8 coverage measures it. Unit tests cover script helpers, fatal and nullable failures, JSON boundaries, caps, cancellation, child ownership, and structured output through real loops. A built-bin smoke runs the separately bundled `lib/worker.cjs` under plain Node, a with-key e2e drives real child agents, and model-facing workflow behavior is snapshot-covered through its owning example.
+Verification belongs to the workflow helper and host-lifecycle tests, the shared Node PTC confinement tests, and source/built workflow execution through the shipped profile. Recorded workflow and opt-in Ralph scenarios own the assembled model transcript; replay configurations using a passthrough sandbox do not establish OS enforcement.
 
 ## Deferred (documented non-goals)
 
@@ -60,14 +54,14 @@ Worker-side logic runs through an in-process `MessageChannel` so V8 coverage mea
 - **Journaling + resume** (`resumeFromRunId`, cached agent() prefixes) — implementing it reintroduces CC's determinism bans as a script-contract tightening (scripts may read the clock).
 - **Saved/bundled workflows** (a `.deepseek/workflows/` registry, slash-command API) and **script persistence to a run directory** (the tool-call event already records the script durably).
 - **Nested `workflow()`**, **token `budget`**, and the `effort`/`isolation`/`agentType` agent options (each rejects loud with a message naming it deferred).
-- **An overall run wall-clock timeout** — cancellation always frees the caller (result settles within the grace), so a cap on total run time is a policy knob for the background redesign, not a correctness need here.
-- **Engine hardening beyond worker threads**: an isolated-vm or separate-process engine behind the same seam (actual sandboxing; memory limits).
+- **An overall run wall-clock timeout** — workflow lifetime remains caller-controlled; explicit cancellation stops PTC execution and awaits child cleanup.
 - **ACP-backend structured output** and **`toolFilter`** (both still capability-gated `false`).
 
 ## Alternatives considered
 
-- **Hostile-value containment in the host** (trap-free proxy rejection, accessor-never-invoked descriptor walks, realm-side pre-rendering of thrown values, realm-built promises/arrays/error clones with structural fatal recognition): rejected because every defense targets an author the trust premise accepts, while the thread's serialization boundary already makes cross-realm values total by construction.
-- **In-process `node:vm` execution**: mechanically simplest — no RPC, no thread — but `start()` blocks the caller for the script's initial synchronous slice, a synchronous spin past the first await cannot be killed in-process (the vm `timeout` covers only that first slice), and `dispose()` could only abandon an unsettling script on the host loop. The worker-thread engine keeps the same vm-context script API while unblocking the host and making termination real.
+- **Host-side defenses for VM values** (proxy rejection, descriptor walks and cross-realm clones): these cannot enforce OS file authority. VM evaluation and materialization belong inside the confined process; the shared PTC provider owns validation at the process transport.
+- **In-process `node:vm` execution**: mechanically simplest — no RPC, no thread — but `start()` blocks the caller for the script's initial synchronous slice, a synchronous spin past the first await cannot be killed in-process (the vm `timeout` covers only that first slice), and `dispose()` could only abandon an unsettling script on the host loop. The PTC process keeps the vm-context script API while unblocking the host and providing managed termination.
+- **`isolated-vm`**: adding a separate JavaScript engine brings native dependency and deployment requirements; the shared PTC provider already supplies process confinement.
 - **Background execution as the default** (CC's shape): deferred; foreground-synchronous matches `dsh-tool-subagent`'s cut, and background semantics should be designed ONCE across shell/subagent/workflow rather than per-tool.
 - **Workflow-layer JSON parsing for `agent({schema})`**: duplicating a seam concern at one consumer while the seam's capability flag stayed dishonestly `false`.
 - **Meta embedded in the script as `export const meta = {...}`** (CC's exact format): keeps scripts self-contained and CC scripts drop-in, but obtaining meta requires evaluating model-written text on the host. Even an empty timed vm context cannot bound script-controlled getters when the host reads the resulting object. A JSON parameter removes the scanner, evaluation, and host-spin hole; the cost is that a CC script's meta header must move into the parameter (the body stays drop-in).
@@ -78,4 +72,4 @@ Worker-side logic runs through an in-process `MessageChannel` so V8 coverage mea
 
 ## Consequences
 
-Fan-out plans now live in rerunnable scripts, and `outputSchema` provides authoritative structured child results. Each run pays worker startup and message-port RPC costs, but host startup stays non-blocking, cancellation can terminate the worker, and serialization enforces the value boundary. Worker threads are not a security boundary. Invalid options fail rather than degrading to Claude Code's `null`; consumers retain control through the run handle while observers receive snapshots only. Top-level Web users also receive a durable, replayable workflow record without widening the execution seam or coupling the original tool card to workflow-specific UI.
+Fan-out plans now live in rerunnable scripts, and `outputSchema` provides authoritative structured child results. Each run pays PTC process startup and binding RPC costs. Host execution stays non-blocking, cancellation stops the managed process, and JSON serialization separates guest values from the host. Invalid options fail rather than degrading to Claude Code's `null`; consumers retain control through the run handle while observers receive snapshots only. Top-level Web users also receive a durable, replayable workflow record without widening the execution seam or coupling the original tool card to workflow-specific UI.

.agents/notes/implemented/feature/2026-07-05-dynamic-workflows.zh.md

@@ -22,19 +22,13 @@ harness 可以通过 `dsh-tool-subagent` 将一个任务委派给一个子 agent
 
 `ctx.workflowEngine` 是 bash 形态的抽象 `WorkflowEngine`——每个上下文一个引擎，无命名提供方注册表（引擎是部署级替换，不是共存者）。`start(request)` 对无法启动的脚本同步抛出；返回的 `WorkflowRun` 的 `result` 永不 reject（失败时结算为 `stopReason: 'error' | 'cancelled'`）。`workflow/*` 事件是仅观察的 emit，携带数据快照（id + meta；`workflow/end` 省略 result 值），按监听器隔离，与 `subagent/start`/`subagent/end` 对称——控制权留在 run 的持有者手中。词汇详情见 [subsystems/workflow.md](../../../../docs/subsystems/workflow.zh.md)。
 
-### 引擎（dsh-workflow-worker-thread）：每次运行一个 worker 线程
+### 引擎（dsh-workflow-ptc）：共享 Node 进程执行
 
-**信任前提**：工作流脚本与模型的 bash 访问具有相同的信任级别。引擎会约束有缺陷脚本的影响，并保证结果已 settled、值可安全表示为 JSON、取消后完全停稳；它不防御恶意代码。vm 上下文和 worker 线程不是安全边界：脚本可以逃逸到具有进程级权限的 Node API。沙箱化需要在此 seam 背后使用独立进程或 isolated-vm 引擎。
+[工作流沙箱复用决策](../architecture/2026-09-13-workflow-ptc-sandbox-reuse.zh.md)取代 worker-thread 执行与信任实现。引擎在沙箱化 PTC Node 进程中保留 VM 与辅助函数。VM 定义脚本 API；OS 文件策略和受管进程清理由共享执行提供方负责。
 
-**为何选择 `node:worker_threads`**：每次运行获得一个非池化的 worker。vm 上下文限定了文档中说明的脚本 API，而消息端口 RPC 将 `agent()` 桥接到宿主侧的子循环。worker 防止脚本的同步工作阻塞宿主，提供序列化边界，并允许取消后强制终止。`isolated-vm` 因其维护状态和部署要求被否决。
+宿主在发布前校验元数据并解析正文。Host 绑定将 guest 连接到 subagent 和工作流观察器。待启动与已发布子记录共享取消信号；[agent 作用域运行时设计 Agent Note](../architecture/2026-07-12-agent-scope-runtime-design.zh.md#workflow-children-are-pending-starts-or-published-records)负责其生命周期规则。
 
-宿主在发布前校验元数据并解析正文。私有枚举键 payload 映射定义协议格式；待启动记录、已发布子记录、单一取消信号、worker 死亡回收、结果优先级与 dispose（资源释放）时的完全停稳，在此协议上保持 subagent run 约定。这些竞态算法由 [agent 作用域运行时设计 Agent Note](../architecture/2026-07-12-agent-scope-runtime-design.zh.md#workflow-children-are-pending-starts-or-published-records) 定义。
-
-引擎暴露一条进程内 `MessageChannel` 测试路径，因为主进程 V8 覆盖率无法观测 worker 执行。
-
-**Meta 是数据**：经 schema 校验的 `meta` 字段以 JSON 形式到达 seam，仅做形状校验。宿主从不执行元数据字面量，否则脚本控制的访问器可以在 worker 隔离之外运行。
-
-**值边界**：`materializeFromRealm` 复制出站值，并拒绝函数、symbol、嵌套 `undefined`、异域原型、循环引用、稀疏数组和非有限数字。数据属性复制使 `"__proto__"` 安全；getter 正常读取，抛出异常的 getter 会明确报错。`args` 通过 `workerData` 传入，暴露前再次克隆。realm 函数被调用而非复制，抛出的值使用对所有输入均有定义的渲染器，因此 `result` 不会 reject。钩子错误是宿主 realm 的 `WorkflowError`，脚本应基于 `name` 或 `code` 分支而非 `instanceof Error`，如引擎 README 所述。并发、total-agent、item、超时和宽限限制均为经校验的配置。
+**Meta 是数据**：Host 从不执行元数据字面量。**值是无损 JSON**：guest 侧 realm 物化在 PTC 传输前拒绝不支持的值。getter 在受限进程内运行，钩子错误跨 realm 保留稳定的 `name` 与 `code` 字段。保留协作式辅助函数上限与最初同步片段超时；不增加整体工作流经过时间定时器。
 
 ### Consumer（`dsh-tool-workflow`）
 
@@ -52,7 +46,7 @@ harness 可以通过 `dsh-tool-subagent` 将一个任务委派给一个子 agent
 
 ## 测试
 
-worker 侧逻辑通过进程内 `MessageChannel` 运行，使 V8 覆盖率能够度量它。单元测试覆盖脚本辅助函数、fatal 与 nullable 失败、JSON 边界、上限、取消、子 agent 所有权和通过真实循环的结构化输出。构建后二进制文件的冒烟测试在纯 Node 下运行单独打包的 `lib/worker.cjs`，带密钥的 e2e 驱动真实子 agent，面向模型的工作流行为通过其所属示例进行快照覆盖。
+验证由工作流辅助函数和 Host 生命周期测试、共享 Node PTC 约束测试，以及通过已发布 profile 的源码／构建后工作流执行负责。已记录工作流与显式启用的 Ralph 场景负责组装后的模型转录；使用 passthrough 沙箱的回放配置不能证明 OS 强制能力。
 
 ## 延迟（明确的非目标）
 
@@ -60,14 +54,14 @@ worker 侧逻辑通过进程内 `MessageChannel` 运行，使 V8 覆盖率能够
 - **日志化 + 恢复**（`resumeFromRunId`、缓存的 agent() 前缀）：实现它会以脚本约定收紧的形式重新引入 CC 的确定性禁令（脚本可以读取时钟）。
 - **保存／打包的工作流**（`.deepseek/workflows/` 注册表、斜杠命令 API）和**脚本持久化到运行目录**（工具调用事件已经持久记录了脚本）。
 - **嵌套 `workflow()`**、**token `budget`**，以及 `effort`/`isolation`/`agentType` agent 选项（每个都会明确拒绝，并在消息中注明其已延迟实现）。
-- **整体运行的挂钟超时**：取消总能释放调用方（result 在宽限期内 settle），因此总运行时间上限是后台重设计的策略旋钮，不是此处的正确性需求。
-- **超越 worker 线程的引擎加固**：在同一 seam 背后使用 isolated-vm 或独立进程引擎（真正的沙箱化；内存限制）。
+- **整体运行的挂钟超时**：工作流生命周期仍由调用方控制；显式取消停止 PTC 执行并等待子 agent 清理。
 - **ACP（Agent Client Protocol）后端结构化输出**和 **`toolFilter`**（两者仍以能力标志 `false` 门控）。
 
 ## 曾考虑的替代方案
 
-- **宿主侧的恶意值防护**（无 trap 代理拒绝、从不调用访问器的描述符遍历、realm 侧预渲染抛出值、realm 构建的 promise/array/error 克隆加结构化 fatal 识别）：否决。每项防御针对的都是信任前提所接受的作者，而线程的序列化边界已经从构造上保证跨 realm 值的处理对所有输入都有确定结果。
-- **进程内 `node:vm` 执行**：机械上最简——无 RPC、无线程——但 `start()` 会在脚本的初始同步切片期间阻塞调用方，第一个 await 之后的同步自旋无法在进程内终止（vm `timeout` 仅覆盖第一个切片），且 `dispose()` 只能在宿主循环上放弃一个未 settle 的脚本。worker 线程引擎保持相同的 vm 上下文脚本 API，同时解除宿主阻塞并使终止成为现实。
+- **VM 值的 Host 侧防护**（代理拒绝、描述符遍历和跨 realm 克隆）：这些无法强制 OS 文件权限。VM 求值与物化属于受限进程内部；共享 PTC 提供方负责进程传输处的验证。
+- **进程内 `node:vm` 执行**：机械上最简——无 RPC、无线程——但 `start()` 会在脚本的初始同步切片期间阻塞调用方，第一个 await 之后的同步自旋无法在进程内终止（vm `timeout` 仅覆盖第一个切片），且 `dispose()` 只能在宿主循环上放弃一个未 settle 的脚本。PTC 进程保持 vm 上下文脚本 API，同时解除宿主阻塞并提供受管终止。
+- **`isolated-vm`**：引入另一套 JavaScript 引擎会增加原生依赖与部署要求；共享 PTC 提供方已提供进程约束。
 - **后台执行作为默认**（CC 的形态）：延迟。前台同步与 `dsh-tool-subagent` 的当前形态一致，后台语义应在 bash、subagent 和工作流之间统一设计一次，而非逐工具设计。
 - **工作流层为 `agent({schema})` 做 JSON 解析**：在一个消费方重复 seam 关注点，而 seam 的能力标志仍不诚实地为 `false`。
 - **Meta 嵌入脚本中作为 `export const meta = {...}`**（CC 的确切格式）：保持脚本自包含且 CC 脚本可直接使用，但获取 meta 需要在宿主上执行模型编写的文本。即使一个空的限时 vm 上下文也无法约束脚本控制的 getter（当宿主读取结果对象时）。JSON 参数消除了扫描器、执行和宿主自旋漏洞；代价是 CC 脚本的 meta 头必须移入参数（正文保持可直接使用）。
@@ -78,4 +72,4 @@ worker 侧逻辑通过进程内 `MessageChannel` 运行，使 V8 覆盖率能够
 
 ## 后果
 
-扇出计划现在存在于可重运行的脚本中，`outputSchema` 提供权威的结构化子 agent 结果。每次运行付出 worker 启动和消息端口 RPC 成本，但宿主启动保持非阻塞，取消可以终止 worker，序列化强制执行值边界。worker 线程不是安全边界。无效选项会失败而非退化为 Claude Code 的 `null`；消费方通过 run handle 保持控制权，观察者仅接收快照。顶层 Web 用户还会得到持久、可回放的工作流记录，同时不扩宽执行 seam，也不把原工具卡耦合到工作流专属 UI。
+扇出计划现在存在于可重运行的脚本中，`outputSchema` 提供权威的结构化子 agent 结果。每次运行付出 PTC 进程启动与绑定 RPC 成本。Host 执行保持非阻塞，取消停止受管进程，JSON 序列化将 guest 值与 Host 分开。无效选项会失败而非退化为 Claude Code 的 `null`；消费方通过 run handle 保持控制权，观察者仅接收快照。顶层 Web 用户还会得到持久、可回放的工作流记录，同时不扩宽执行 seam，也不把原工具卡耦合到工作流专属 UI。

.agents/notes/implemented/feature/2026-07-16-harness-level-loop.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 .agents/notes/implemented/feature/2026-07-16-harness-level-loop.md
-2026-07-16-harness-level-loop.md: 43b8f867ae9af92f22fdae0cecef37803b49be30
-2026-07-16-harness-level-loop.zh.md: 40af80d95127974bcbed4f7114dec048e0dbdc57
+2026-07-16-harness-level-loop.md: 54adc1aad63e8968c9e0d276a075cd42abf021fd
+2026-07-16-harness-level-loop.zh.md: 9e1a7e73fbf1fab1807b59b3b7816cdfcc6f45df

.agents/notes/implemented/feature/2026-07-16-harness-level-loop.md

@@ -76,7 +76,7 @@ Base-backed profiles mount the shared command registry and complete goal stack b
 
 Ralph is a first-class model tool in its own plugin, demonstrating that a sophisticated fixed execution policy can be composed without a new loop core. The plugin owns a fixed workflow script over `ctx.workflowEngine` and `ctx.subagents`; it does not create session-goal state or add a branch to `dsh-agent-loop`.
 
-Each round uses an explicit `WorkflowStartRequest.subagentProvider`, defaulting to `spawn`. The provider must exist, support structured output, and declare that it does not inherit parent context. Ralph also passes its resolved round cap as `WorkflowStartRequest.maxTotalAgents`; the worker engine validates both per-run policies before publishing work, so provider misconfiguration or an engine ceiling below the requested Ralph scale fails before a run exists. The child inherits cwd and lineage but receives only the immutable objective, round/cap, workspace-as-authority instruction, and previous normalized report.
+Each round uses an explicit `WorkflowStartRequest.subagentProvider`, defaulting to `spawn`. The provider must exist, support structured output, and declare that it does not inherit parent context. Ralph also passes its resolved round cap as `WorkflowStartRequest.maxTotalAgents`; the PTC workflow engine validates both per-run policies before publishing work, so provider misconfiguration or an engine ceiling below the requested Ralph scale fails before a run exists. The child inherits cwd and lineage but receives only the immutable objective, round/cap, workspace-as-authority instruction, and previous normalized report.
 
 A report contains status, summary, evidence, next steps, and blocker text. Status-specific invariants and serialized size are validated inside the fixed script and again at the consumer boundary. `maxRounds` is configurable, defaults to `256`, and is the ceiling for a call override. `maxHandoffChars` defaults to `16384`; oversized reports fail rather than being silently truncated. `maxResultChars` separately defaults to `16384` and bounds the complete successful parent-facing text, including its envelope and truncation marker.
 

.agents/notes/implemented/feature/2026-07-16-harness-level-loop.zh.md

@@ -76,7 +76,7 @@ Goal Round 驱动器为每个特定的实时 agent 至多拥有一个待定预
 
 Ralph 是位于自有插件中的一等模型工具，展示了复杂固定执行策略可以在没有新 loop 核心的情况下组合完成。该插件拥有构建在 `ctx.workflowEngine` 与 `ctx.subagents` 之上的固定工作流脚本；它不会创建会话目标状态，也不会为 `dsh-agent-loop` 增加分支。
 
-每个 Round 都使用显式 `WorkflowStartRequest.subagentProvider`，默认为 `spawn`。该提供方必须存在、支持结构化输出，并声明不继承父上下文。Ralph 还会把解析后的 Round 上限作为 `WorkflowStartRequest.maxTotalAgents` 传递；工作线程引擎会在发布工作前验证两项每次运行策略，因此提供方配置错误或低于所请求 Ralph 规模的引擎上限会在运行创建前失败。子 agent 继承 cwd 与谱系，但只接收不可变目标、当前 Round/上限、以工作区为权威的指令和上一份规范化报告。
+每个 Round 都使用显式 `WorkflowStartRequest.subagentProvider`，默认为 `spawn`。该提供方必须存在、支持结构化输出，并声明不继承父上下文。Ralph 还会把解析后的 Round 上限作为 `WorkflowStartRequest.maxTotalAgents` 传递；PTC 工作流引擎会在发布工作前验证两项每次运行策略，因此提供方配置错误或低于所请求 Ralph 规模的引擎上限会在运行创建前失败。子 agent 继承 cwd 与谱系，但只接收不可变目标、当前 Round/上限、以工作区为权威的指令和上一份规范化报告。
 
 报告包含状态、摘要、证据、下一步与阻塞文本。固定脚本内部和消费方边界都会验证状态专用不变量与序列化大小。`maxRounds` 可配置，默认为 `256`，并作为调用覆盖值的上限。`maxHandoffChars` 默认为 `16384`；过大报告会失败，而不会被静默截断。`maxResultChars` 单独默认为 `16384`，并限制面向父级的完整成功文本，包括外层文本与截断标记。
 

.agents/notes/implemented/feature/2026-07-19-persisted-same-session-goal-domain.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 .agents/notes/implemented/feature/2026-07-19-persisted-same-session-goal-domain.md
-2026-07-19-persisted-same-session-goal-domain.md: bc6c9ec14c16b20bc8434bac0f1eb6070ea8920d
-2026-07-19-persisted-same-session-goal-domain.zh.md: 30b9d22633f37dda971104f8d2d5be0d3c8c3163
+2026-07-19-persisted-same-session-goal-domain.md: 4ddc02c3bd3b7bbf04b3c6867a76673431309460
+2026-07-19-persisted-same-session-goal-domain.zh.md: 2bb937b0f637e54ab6d9a6e10ebe237849e08171

.agents/notes/implemented/feature/2026-07-19-persisted-same-session-goal-domain.md

@@ -28,7 +28,7 @@ Incremental replay advances its cursor after each valid event and remains positi
 
 At most one goal is current. Create requires no current non-complete goal and always generates a revision-one id not used earlier in the session; a completed goal may be replaced. Every other mutation carries the expected `GoalRef`, and stale ids or revisions reject. Resume accepts a paused or blocked phase, or a disarmed active goal, only when the round cap has remaining capacity. The domain validates blocker reason shape but deliberately leaves reason codes and the decision to block to policy consumers.
 
-A cache built from any seed starts disarmed, and every `agent/session-start` edge disarms it again. `GoalService.disarm(agent)` also lets a lifecycle owner remove process-local authority without a session event, revision change, or `goal/changed` notification. Resume, fork, and continuation-driver replacement therefore preserve the durable objective and history but never initiate work on their own. A later human prompt can be interpreted by the model, whose policy API may explicitly call resume and arm the goal.
+A cache built from any seed starts disarmed, and every `agent/created` edge disarms it again. `GoalService.disarm(agent)` also lets a lifecycle owner remove process-local authority without a session event, revision change, or `goal/changed` notification. Resume, fork, and continuation-driver replacement therefore preserve the durable objective and history but never initiate work on their own. A later human prompt can be interpreted by the model, whose policy API may explicitly call resume and arm the goal.
 
 ### Service boundary
 

.agents/notes/implemented/feature/2026-07-19-persisted-same-session-goal-domain.zh.md

@@ -28,7 +28,7 @@ Status: implemented
 
 最多只有一个当前目标。创建要求不存在未完成的当前目标，并始终生成该会话此前未使用过、修订号为一的 id；已完成目标可以被替换。其他每次变更都携带预期的 `GoalRef`，陈旧的 id 或修订号会被拒绝。仅当 Goal Round 上限仍有余量时，暂停或阻塞阶段以及已解除激活的活跃目标才能恢复。领域层校验阻塞原因的形状，但会把原因代码和是否阻塞的决策留给策略消费方。
 
-从任何种子构建的缓存都以未激活状态开始，每次 `agent/session-start` 边沿也会再次解除激活。`GoalService.disarm(agent)` 还允许生命周期所有者移除进程内权限，而不写入会话事件、不改变修订号，也不发出 `goal/changed` 通知。因此，会话恢复、fork 和继续执行驱动器替换都会保留持久化目标与历史，但绝不会自行启动工作。后续人类提示词可由模型解释，其策略 API 可以显式调用恢复操作并激活目标。
+从任何种子构建的缓存都以未激活状态开始，每次 `agent/created` 边沿也会再次解除激活。`GoalService.disarm(agent)` 还允许生命周期所有者移除进程内权限，而不写入会话事件、不改变修订号，也不发出 `goal/changed` 通知。因此，会话恢复、fork 和继续执行驱动器替换都会保留持久化目标与历史，但绝不会自行启动工作。后续人类提示词可由模型解释，其策略 API 可以显式调用恢复操作并激活目标。
 
 ### 服务边界
 

.agents/notes/implemented/feature/2026-08-06-manager-owned-subagent-settlement-delivery.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 .agents/notes/implemented/feature/2026-08-06-manager-owned-subagent-settlement-delivery.md
-2026-08-06-manager-owned-subagent-settlement-delivery.md: 9d14d4af775809a974349c753a7b4b593a145489
-2026-08-06-manager-owned-subagent-settlement-delivery.zh.md: 4bbee373aa2b50902af5319f9398a89da9cc3143
+2026-08-06-manager-owned-subagent-settlement-delivery.md: 5553d4c06a6776f0ef025e19ae7f4f67fe277a56
+2026-08-06-manager-owned-subagent-settlement-delivery.zh.md: 6d7ee756ad531e123508e8667adc4e5f99c2b9ea

.agents/notes/implemented/feature/2026-08-06-manager-owned-subagent-settlement-delivery.md

@@ -16,7 +16,11 @@ The signal already existed. `subagent/end` has carried `stopReason` and `lastAss
 
 The continuation manager delivers the account itself, from inside the disposal transaction that ends the Activation.
 
-When a resident Activation settles, `notifySettlement()` resolves the child's durable direct parent and sends it one user-role message: the epoch's outcome as a sentence the parent can act on, then the child's final assistant content, or a statement that it produced none. Delivery is unconditional for every child whose id a caller actually received. It does not consult whether the child reported, and it keeps no bookkeeping that could make the promise conditional — that unconditionality is what lets `tool-subagent` promise a runtime notice containing the outcome and any final assistant message. A materialization rolled back before its first accepted message stays silent, because the caller was told that child was not established.
+When a resident Activation settles, `notifySettlement()` resolves the child's durable direct parent and sends it one user-role message: the epoch's outcome as a sentence the parent can act on, then the text from the child's final assistant output, or a statement that it produced no closing text. Delivery is unconditional for every child whose id a caller actually received. It does not consult whether the child reported, and it keeps no bookkeeping that could make the promise conditional — that unconditionality is what lets `tool-subagent` promise a runtime notice containing `its outcome and any final assistant message`. A materialization rolled back before its first accepted message stays silent, because the caller was told that child was not established.
+
+### Closing text
+
+[`createSettlementMessage()`](../../../../packages/subagent/subagent/src/continuation-messages.ts) projects the selected assistant output to nonempty text blocks before creating the user-role notice. It preserves text bytes and block order, excludes every nontext block, and uses `It left no closing message.` when no nonempty text remains. Text-only notices work across parent providers without depending on their support for particular content types. Reasoning and tool calls, for example, cannot appear in a DeepSeek Messages user message; images are accepted by Messages but not by every parent model. The conversion belongs to notice construction; `AssistantOutputFold`, `SubagentResult.output`, and `subagent/end.lastAssistantMessage` retain complete child output for SDK and UI consumers.
 
 ### Runtime source
 
@@ -62,7 +66,7 @@ Both matter past the notice: `subagent/end` carries `stopReason` to the jsonrpc
 
 Three assembled ACP scenarios cover the notice: a child that sends no message, a child that sends a message first, and a child driven through several Agent-message turns. All three need an explicit fence. The notice arrives once the child's teardown finishes, which races whatever the parent is already doing, so each scenario holds the child behind the parent's spawn turn and then waits for the parent turn the notice opens (`waitForTurnStart` at that turn, then `waitForTurnEnd`) before the script continues. Waiting for a turn the run is not fenced to produce is not coverage: it is a timeout when the notice lands in the turn already running instead.
 
-`subagent-continuable` is the one that pins a failure. Its child's last turn dies on the forced durability checkpoint without entering a step, so that transcript is where the stop-reason rule above is visible end to end: the notice says the child *failed*, carries the earlier `SECOND_OK` as its last content rather than as a result, and the parent's own acknowledgement turn reaches the ACP client.
+`subagent-continuable` pins a completed SDK settlement with `SECOND_OK` and no child reasoning in the parent's notice; its child finishes in turn 1, so the configured turn-3 checkpoint failure is not exercised.
 
 A keyless headless Loader snapshot covers the user-visible path end to end. Its replay parent omits `run_in_background` to exercise the continuable background default, never calls `list_agents`, `send_message`, or Task tools, consumes the manager-authored `subagent-settled` notice, and produces its final answer. The child sends no Agent message, so the transcript depends only on the runtime notice. A test-only Loader fence holds the parent's post-spawn request until the real manager notice enters its inbox, removing platform scheduling from the transcript without synthesizing the notice.
 
@@ -84,6 +88,10 @@ The refusal and interruption wordings are pinned verbatim in unit tests rather t
 
 **Always use `followup`.** Simpler and uniform, but a fan-out of children settling together would cost one parent turn each. The step-boundary batch already exists; using it is free.
 
+**Filter the canonical child output.** Removing nontext blocks in `AssistantOutputFold` would discard assistant content needed by SDK and UI consumers. Only the parent notice requires a text projection.
+
+**Relax the Messages serializer.** Accepting or silently discarding invalid user-role blocks would conceal the producer's role conversion error. Notice construction supplies content that every parent provider can represent while protocol validation remains strict.
+
 ## Consequences
 
 - A continuable child's parent receives one message per settled Activation. Fan-out deployments therefore add parent turns; steering keeps a simultaneous batch to one step.

.agents/notes/implemented/feature/2026-08-06-manager-owned-subagent-settlement-delivery.zh.md

@@ -16,7 +16,11 @@ Status: implemented
 
 继续执行管理器自己投递这份记账，就在结束 Activation 的那笔 dispose 事务内部完成。
 
-当驻留 Activation 结算时，`notifySettlement()` 解析该 child 持久化的直接父级，并向它发送一条用户角色消息：先是父级可据以行动的一句结果说明，然后是 child 的最终 assistant 内容，或一句说明它没有产出内容。对每个调用方真正拿到过 id 的 child，投递都是无条件的。它不查询 child 是否上报过，也不保留任何可能让这项承诺变成有条件的记账——正是这种无条件性，才让 `tool-subagent` 能够承诺一条包含结局与可能存在的最终 assistant 消息的运行时通知。在第一条消息被接受之前就回滚的物化保持静默，因为调用方已被告知该 child 未建立。
+当驻留 Activation 结算时，`notifySettlement()` 解析该 child 持久化的直接父级，并向它发送一条用户角色消息：先是父级可据以行动的一句结果说明，然后是 child 最终 assistant 输出中的文本，或一句说明它没有产出收尾文本。对每个调用方真正拿到过 id 的 child，投递都是无条件的。它不查询 child 是否上报过，也不保留任何可能让这项承诺变成有条件的记账——正是这种无条件性，才让 `tool-subagent` 能够承诺一条包含 `its outcome and any final assistant message` 的运行时通知。在第一条消息被接受之前就回滚的物化保持静默，因为调用方已被告知该 child 未建立。
+
+### 收尾文本
+
+[`createSettlementMessage()`](../../../../packages/subagent/subagent/src/continuation-messages.ts) 在创建用户角色通知前，将选中的 assistant 输出投影为非空文本块。它保留文本字节与块顺序，排除所有非文本块，并在没有剩余非空文本时使用 `It left no closing message.`。纯文本通知适用于不同的父级提供方，不依赖它们对特定内容类型的支持。例如，推理与工具调用不能出现在 DeepSeek Messages 的用户消息中；Messages 接受图片，但并非每个父级模型都支持图片。这项转换归通知构建所有；`AssistantOutputFold`、`SubagentResult.output` 与 `subagent/end.lastAssistantMessage` 为 SDK 和 UI 消费方保留完整的子级输出。
 
 ### 来源信息
 
@@ -62,7 +66,7 @@ Status: implemented
 
 三个整体组装的 ACP 场景覆盖该通知：一个不发送消息的 child、一个先发送消息的 child，以及一个被多轮 Agent 消息驱动的 child。三者都需要显式栅栏。通知在 child 拆卸完成后才到达，会与父级当时正在做的事竞争，因此每个场景都会把 child 保持到父级启动轮次结束，再等待该通知开启的那个父级轮次（先 `waitForTurnStart` 到该轮次，再 `waitForTurnEnd`），然后脚本才继续。等待一个运行并未被栅栏保证会产生的轮次不算覆盖：一旦通知落进已经在跑的那个轮次，它就是一次超时。
 
-`subagent-continuable` 是其中固定失败结局的那个。它的 child 最后一个轮次在被强制的持久化检查点上死亡，且未进入任何 step，因此该 transcript 正是上面那条终止原因规则的端到端可见之处：通知说该 child **失败**，把此前的 `SECOND_OK` 作为它最后产出的内容而非结果携带，而父级自己的确认轮次会到达 ACP 客户端。
+`subagent-continuable` 固定了 SDK 中已完成的结算：父级通知包含 `SECOND_OK`，但不含子级推理；子级在第 1 轮结束，因此未覆盖已配置的第 3 轮检查点失败。
 
 另有一个无密钥的 headless Loader 快照端到端覆盖用户可见路径。其重放父级省略 `run_in_background` 以覆盖可继续后台默认路径，从不调用 `list_agents`、`send_message` 或 Task 工具，消费管理器写入的 `subagent-settled` 通知，并给出最终答案。child 不发送 Agent 消息，因此该 transcript 只依赖运行时通知。一个仅用于测试的 Loader 栅栏会把父级启动后的请求保持到真实管理器通知进入其 inbox 为止，从 transcript 中排除平台调度差异，但不会伪造该通知。
 
@@ -84,6 +88,10 @@ Status: implemented
 
 **始终使用 `followup`。** 更简单也更统一，但一批同时结算的 child 会各自消耗一个父级轮次。step 边界的批量语义本来就存在，用它是免费的。
 
+**过滤规范的子级输出。** 在 `AssistantOutputFold` 中删除非文本块，会丢弃 SDK 和 UI 消费方需要的 assistant 内容。只有父级通知需要文本投影。
+
+**放宽 Messages 序列化器。** 接受或静默丢弃用户角色中的无效块，会掩盖生产方的角色转换错误。通知构建提供每种父级提供方都能表示的内容，协议校验仍保持严格。
+
 ## 后果
 
 - 可继续 child 的父级会为每个已结算 Activation 收到一条消息。因此，做扇出的部署会增加父级轮次；steer 会把同时结算的一批压缩到一个 step。

.agents/notes/implemented/feature/2026-08-20-unified-image-request-pipeline.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 .agents/notes/implemented/feature/2026-08-20-unified-image-request-pipeline.md
-2026-08-20-unified-image-request-pipeline.md: 736c8fb7f6f58db137d2b3cb9f5a007db16dd67b
-2026-08-20-unified-image-request-pipeline.zh.md: 532662291ba2c7257906a2bf4f52792d5285ec1c
+2026-08-20-unified-image-request-pipeline.md: 7431a474b55a9dae82b1c9cd36b770b31fe39602
+2026-08-20-unified-image-request-pipeline.zh.md: 247a57b8017f513c552204d951ef6931f7f9d045

.agents/notes/implemented/feature/2026-08-20-unified-image-request-pipeline.md

@@ -26,7 +26,7 @@ Batch admission prepares and verifies every normalized attachment once before pu
 
 The `variantId` and cache path cover the normalized attachment id, transform version, route pixel and byte budgets, and fixed encoder parameters. A new cache entry is fully decoded before publication. Cache hits use a header probe to check format, 8-bit sRGB/sRGBA facts, dimensions, and alpha without decoding the complete raster again; a mismatch regenerates the entry. DeepSeek Files and pi-ai inline base64 therefore use the same deterministic bytes for the same policy. Inline accounting uses the derived byte length after base64 expansion, not the normalized attachment byte count. Equal in-process `variantId` calls share one transform and cache write. Each caller can cancel its own wait; the shared transform is aborted only after every waiter has cancelled. Callers preserve order by applying `Promise.all` to singular `readImageRequest` calls. The local implementation runs normalization and request transforms through one FIFO limiter; `imageCompressionConcurrency` is configurable from 1 through 8 and defaults to 2. Batch publication remains sequential after every normalized attachment has been prepared.
 
-Request-size offload is a deterministic oldest-first projection. Before reading attachments, each route uses `min(attachmentBytes, requestVersionMaxBytes)` as a conservative upper bound and removes the oldest over-budget prefix. Only retained attachments are read and transformed, so an omitted missing or corrupt object cannot block the request. A second projection uses exact derived lengths without bringing omitted images back. DeepSeek defaults to 128MiB and 600 referenced images. Its removed prefix advances past successive 64MiB byte boundaries and in 20-image count quanta, so 129 one-megabyte images remove the oldest 65, retain 64MiB, and keep that prefix stable until total history passes 192MiB. Pi-ai retains a configurable base64 request bound. Each omitted image becomes a per-image placeholder that retains its identity and access resolved for the current tool execution world, including nested tool-result images, while append-only session history keeps the original references.
+Request-size offload follows the [durable image-offload decision](../architecture/2026-09-10-image-offload-events.md). Routes check retained occurrences against their request-version byte and image-count budgets and report `IMAGE_OFFLOAD_REQUIRED` with a removal count. The compaction plugin records exact oldest-first occurrences in `image/offload`; Session derivation applies the marks. Adapters prepare only retained attachments and render each marked occurrence as placeholder text with its identity and current execution-world access path, including nested tool-result images. Original message events retain the attachment references.
 
 ### Stable handles
 
@@ -34,9 +34,9 @@ Every retained request image is preceded by its display name or complete attachm
 
 ### DeepSeek Files lifecycle
 
-The direct `deepseek-official` adapter normally uploads every retained request version through the OpenAI-compatible Files API and sends `file_id` content blocks. A [bounded inline fallback](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md) sends the same deterministic request versions when file resolution fails. The default catalog advertises `deepseek-v4-flash-vision-exp` as image-capable. Uploaded ids are indexed by endpoint and API-key scope plus `variantId`. Uploads request seven days by default and record the returned `expires_at`; a mapping with no more than one hour remaining is replaced without a preceding retrieve call. The index never stores the API key.
+The direct `deepseek-official` adapter normally uploads every retained request version through its selected protocol’s Files API and sends file-id references. [Messages](2026-09-07-deepseek-messages-adapter.md) and Chat Completions share the lifecycle while retaining their own endpoint and wire formats. A [bounded inline fallback](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md) sends the same deterministic request versions when file resolution fails. The default catalog advertises `deepseek-v4-flash-vision-exp` as image-capable. Uploaded ids are indexed by Files endpoint and API-key scope plus `variantId`. Uploads request seven days by default. Chat records the returned `expires_at`; Messages metadata omits expiry, so its local reuse deadline uses the original upload time plus the configured lifetime without promising remote deletion. A mapping with no more than one hour of reuse remaining is replaced without a preceding retrieve call. The index never stores the API key.
 
-An upload is indexed only after the response returns a complete file object, matching byte count, and `expires_at`. A missing or inconsistent response leaves no local mapping, so a later request uploads again. Concurrent upload resolution for one scoped `variantId` shares one provider operation; one waiter cannot cancel another, and the upload stops when every waiter has cancelled. A malformed upload index is an empty cache and is replaced on the next successful upload; filesystem I/O failures remain errors. If chat reports expired, deleted, missing, or invalid ids and names one or more ids used by the request, only those mappings are removed. A stale-file response without a specific id removes every mapping used by that chat attempt. The affected request bytes are uploaded again and chat is retried once. A second stale rejection clears the mappings identified by its response and returns the error without a third chat attempt. One upload quota error first lists the configured number of oldest harness-owned `dsh-` files, then deletes that collected set and retries once; deleting after pagination keeps provider cursors valid. Public file operations expose list, retrieve, delete, one-variant release, and namespace-wide release. Every Files request carries the shared Harness `User-Agent`. The client enforces the documented 128MiB upload limit, 32MiB chat-image limit, 10,000-file and 25GiB quotas, and one-hour to 30-day expiry range.
+An upload is indexed only after the response returns a valid native file object with a matching byte count and the client establishes the protocol’s reuse deadline. A missing or inconsistent response leaves no local mapping, so a later request uploads again. Concurrent upload resolution for one scoped `variantId` shares one provider operation; one waiter cannot cancel another, and the upload stops when every waiter has cancelled. A malformed upload index is an empty cache and is replaced on the next successful upload; filesystem I/O failures remain errors. If the model endpoint reports expired, deleted, missing, or invalid ids and names one or more ids used by the request, only those mappings are removed. A stale-file response without a specific id removes every mapping used by that model request. The affected request bytes are uploaded again and the model request is retried once. A second stale rejection clears the mappings identified by its response and returns the error without a third model request. One upload quota error first lists the configured number of oldest harness-owned `dsh-` files, then deletes that collected set and retries once; deleting after pagination keeps provider cursors valid. Public file operations expose list, retrieve, delete, one-variant release, and namespace-wide release. Every Files request carries the shared Harness `User-Agent`. The client enforces a 128MiB upload limit, a 32MiB request-image limit, 10,000-file and 25GiB quotas, and one-hour to 30-day expiry range.
 
 ### Diagnostics
 
@@ -52,7 +52,7 @@ Historical attachment objects that later disappear or fail integrity verificatio
 
 **Treat PNG as a screenshot and reject 16-bit PNG.** File format does not reveal pixel complexity, and 16-bit RGB/RGBA is a convertible sample depth rather than an unsupported image type. Pixel sampling and post-conversion probes give the required facts.
 
-**Keep DeepSeek data URLs as the primary transport.** Inline base64 repeats bytes on every request and caps usable image history by request-body size. Files API references reuse uploaded deterministic request bytes and provide explicit expiry and deletion; the bounded fallback uses data URLs only when file resolution fails.
+**Keep DeepSeek data URLs as the primary transport.** Inline base64 repeats bytes on every request and caps usable image history by request-body size. Files API references reuse uploaded deterministic request bytes with bounded local reuse and explicit deletion; the bounded fallback sends inline base64 only when file resolution fails.
 
 **Trust a locally indexed file id indefinitely.** Remote expiry, deletion, and lost upload responses make local and provider state diverge. Response-directed invalidation and one re-upload recover without an unbounded retry loop; an ambiguous stale-file response must invalidate every file used by that attempt because it provides no safe exact target.
 
@@ -62,7 +62,7 @@ Historical attachment objects that later disappear or fail integrity verificatio
 
 ## Verification
 
-Package tests generate 16-bit RGB and RGBA PNG fixtures, prove 8-bit conversion and clean 8-bit passthrough, retain alpha under byte pressure, stop lazy encoding after the first fitting candidate, keep the smallest ladder output above an unreachable byte target, cover square and wide 640,000-pixel projections, singleflight equal variants and uploads without shared-cancellation leaks, bound transform concurrency, preserve cache and upload identity, skip attachment reads for conservatively offloaded history, prepare batches once, reject inconsistent Files responses, refresh near-expiry ids without retrieve, recover once from single-id, multiple-id, and ambiguous stale responses, fall back to bounded all-inline requests after file resolution failure, paginate before quota deletion, normalize provider diagnostics, project text-only history, and share normal/compaction request bytes. Keyless assembled snapshots cover the real tool schemas and image request path. A credentialed test uses the built-in `deepseek-official` route and its configured endpoint, never a custom provider entry.
+Package tests generate 16-bit RGB and RGBA PNG fixtures, prove 8-bit conversion and clean 8-bit passthrough, retain alpha under byte pressure, stop lazy encoding after the first fitting candidate, keep the smallest ladder output above an unreachable byte target, cover square and wide 640,000-pixel projections, singleflight equal variants and uploads without shared-cancellation leaks, bound transform concurrency, preserve cache and upload identity, skip attachment reads for durably offloaded history, prepare batches once, reject inconsistent Files responses, refresh near-expiry ids without retrieve, recover once from single-id, multiple-id, and ambiguous stale responses, fall back to bounded all-inline requests after file resolution failure, paginate before quota deletion, normalize provider diagnostics, project text-only history, and share normal/compaction request bytes. Keyless assembled snapshots cover the real tool schemas and image request path. A credentialed test uses the built-in `deepseek-official` route and its configured endpoint, never a custom provider entry.
 
 ## Consequences
 

.agents/notes/implemented/feature/2026-08-20-unified-image-request-pipeline.zh.md

@@ -26,7 +26,7 @@ Status: implemented
 
 `variantId` 和缓存路径覆盖规范化附件 ID、变换策略版本、路由像素和字节预算及固定编码参数。新缓存条目在发布前会完整解码。缓存命中只探测文件头，校验格式、8-bit sRGB/sRGBA、尺寸和透明通道，不会再次完整解码光栅；不匹配时会重新生成。因此，同一策略下的 DeepSeek Files 和 pi-ai 内联 base64 使用相同的确定性字节。内联计量使用派生字节经过 base64 膨胀后的长度，不使用规范化附件字节数。同一进程内相同 `variantId` 的调用共享一次变换和缓存写入。每个调用方可以取消自己的等待；只有全部等待方都取消时，共享变换才会中止。调用方对单数 `readImageRequest` 使用 `Promise.all` 保持结果顺序。本地实现通过一个 FIFO 限流器运行规范化和请求变换，`imageCompressionConcurrency` 的可配置范围为 1 至 8，默认值为 2。全部规范化附件准备完成后，批次仍按顺序发布。
 
-请求大小 offload 是确定性的从旧到新投影。读取附件前，每条路由先以 `min(附件字节数, 请求版本字节上限)` 作为保守上界，移除超出预算的最旧前缀。系统只读取并转换保留的附件，因此已省略的缺失或损坏对象不会阻塞请求。第二次投影使用确切派生长度，但不会重新加入已省略图片。DeepSeek 默认上限为 128MiB 和 600 张引用图片。被移除前缀会越过连续的 64MiB 字节边界，并按 20 张图片数量步长递增，因此 129 张 1MiB 图片会移除最旧的 65 张并保留 64MiB；持久历史超过 192MiB 前，该前缀保持不变。Pi-ai 保留可配置的 base64 请求上限。每张省略图片都会变成逐图占位文本，保留自己的身份和本次工具执行环境解析出的访问方式，嵌套工具结果图片也使用相同规则；追加式会话历史继续保留原始引用。
+请求大小 offload 遵循[持久图片省略决策](../architecture/2026-09-10-image-offload-events.zh.md)。路由根据请求版本字节和图片数量预算检查保留的位置，通过 `IMAGE_OFFLOAD_REQUIRED` 报告所需移除数量。Compaction 插件在 `image/offload` 中记录明确的最旧图片出现位置，Session 派生时应用标记。适配器只准备保留的附件，将每个带标记的位置渲染为包含其身份和当前执行环境访问路径的占位文本，嵌套工具结果图片也适用。原始消息事件保留附件引用。
 
 ### 稳定句柄
 
@@ -34,9 +34,9 @@ Status: implemented
 
 ### DeepSeek Files 生命周期
 
-直接 `deepseek-official` 适配器通常通过 OpenAI 兼容 Files API 上传每张保留的请求版本，并发送 `file_id` 内容块。文件解析失败时，[有界内联回退](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md)会发送相同的确定性请求版本。默认 catalog 把 `deepseek-v4-flash-vision-exp` 公布为支持图片。上传 ID 按端点和 API key 作用域以及 `variantId` 写入索引。上传默认请求 7 天有效期，并记录返回的 `expires_at`；本地映射剩余时间不超过一小时时会直接替换，不会先查询远端文件。索引绝不存储 API key。
+直接 `deepseek-official` 适配器通常通过所选协议的 Files API 上传每张保留的请求版本，并发送 file-id 引用。[Messages](2026-09-07-deepseek-messages-adapter.zh.md) 与 Chat Completions 共享生命周期，各自保留端点和协议格式。文件解析失败时，[有界内联回退](../../archived/bug-fix/2026-08-21-deepseek-files-inline-fallback.md)会发送相同的确定性请求版本。默认 catalog 把 `deepseek-v4-flash-vision-exp` 公布为支持图片。上传 ID 按 Files 端点和 API key 作用域以及 `variantId` 写入索引。上传默认请求 7 天有效期。Chat 记录返回的 `expires_at`；Messages 元数据不含过期时间，因此本地复用期限使用原始上传时间加配置的生存期，但不保证远端文件删除。本地映射剩余复用时间不超过一小时时会直接替换，不会先查询远端文件。索引绝不存储 API key。
 
-只有上传响应返回完整文件对象、匹配的字节数和 `expires_at` 时，上传结果才会写入索引。缺失或不一致的响应不会留下本地映射，后续请求会重新上传。同一作用域和 `variantId` 的并发解析共享一次提供方上传；单个等待方无法取消其他等待方，全部等待方取消时才会停止上传。格式损坏的上传索引按空缓存处理，并在下一次成功上传时替换；文件系统 I/O 失败仍是错误。如果 chat 报告 ID 已过期、删除、缺失或无效，并指出本次请求使用的一个或多个 ID，适配器只删除这些映射。如果响应只说明文件状态失效而没有指出具体 ID，适配器会删除该次 chat 使用的全部映射。受影响的请求字节会重新上传，chat 只重试一次。第二次仍报告文件失效时，适配器会按响应清理映射并返回错误，不会发起第三次 chat。一次上传配额错误会先列出配置数量的最旧 `dsh-` 文件，再删除收集到的文件并重试一次；分页完成后才删除，避免游标失效。公开文件操作提供列表、查询、删除、单个变体释放和整个作用域释放。每个 Files 请求都携带 Harness 的共享 `User-Agent`。客户端执行文档规定的 Files 单次上传 128MiB、chat 单图 32MiB、10,000 个文件、25GiB，以及一小时到 30 天有效期限制。
+只有上传响应返回有效的原生文件对象、字节数匹配，且客户端确定了该协议的复用期限时，上传结果才会写入索引。缺失或不一致的响应不会留下本地映射，后续请求会重新上传。同一作用域和 `variantId` 的并发解析共享一次提供方上传；单个等待方无法取消其他等待方，全部等待方取消时才会停止上传。格式损坏的上传索引按空缓存处理，并在下一次成功上传时替换；文件系统 I/O 失败仍是错误。如果模型端点报告 ID 已过期、删除、缺失或无效，并指出本次请求使用的一个或多个 ID，适配器只删除这些映射。如果响应只说明文件状态失效而没有指出具体 ID，适配器会删除该次模型请求使用的全部映射。受影响的请求字节会重新上传，模型请求只重试一次。第二次仍报告文件失效时，适配器会按响应清理映射并返回错误，不会发起第三次模型请求。一次上传配额错误会先列出配置数量的最旧 `dsh-` 文件，再删除收集到的文件并重试一次；分页完成后才删除，避免游标失效。公开文件操作提供列表、查询、删除、单个变体释放和整个作用域释放。每个 Files 请求都携带 Harness 的共享 `User-Agent`。客户端执行 Files 单次上传 128MiB、请求单图 32MiB、10,000 个文件、25GiB，以及一小时到 30 天有效期限制。
 
 ### 诊断
 
@@ -52,7 +52,7 @@ Status: implemented
 
 **把 PNG 当作截图，并拒绝 16-bit PNG。** 文件格式不能说明像素复杂度，16-bit RGB/RGBA 是可转换位深，不是不支持的图片类型。像素采样和转换后探测能提供所需事实。
 
-**把 DeepSeek data URL 作为首选传输方式。** 内联 base64 会在每次请求中重复字节，并按请求正文大小限制可用图片历史。Files API 引用会复用上传后的确定性请求字节，并提供显式有效期和删除操作；有界回退只在文件解析失败时使用 data URL。
+**把 DeepSeek data URL 作为首选传输方式。** 内联 base64 会在每次请求中重复字节，并按请求正文大小限制可用图片历史。Files API 引用会复用上传后的确定性请求字节，并限制本地复用期限、提供显式删除操作；有界回退只在文件解析失败时发送内联 base64。
 
 **永久信任本地索引中的文件 ID。** 远端过期、删除和上传响应丢失会使本地与提供方状态不一致。按响应失效和一次重新上传可以恢复，同时避免无界重试；响应没有给出可安全使用的精确目标时，必须使该次请求使用的全部文件失效。
 
@@ -62,7 +62,7 @@ Status: implemented
 
 ## Verification
 
-包测试会生成 16-bit RGB 和 RGBA PNG fixture，验证 8-bit 转换与干净 8-bit 字节直通、字节压力下保留透明通道、首个候选合规后停止编码、字节目标不可达时保留最小阶梯产物、正方形和宽屏 640,000 像素投影、相同变体与上传 singleflight 且不会共享取消、变换并发上限、缓存与上传身份、跳过已保守 offload 的历史附件读取、批量只准备一次、Files 响应不一致、进入刷新余量时不查询远端并更新 ID、单个 ID、多个 ID 和模糊失效响应只恢复一次、文件解析失败后回退到有界全内联请求、删除配额文件前完成分页、规范化提供方诊断、纯文本投影，以及普通请求与压缩共享请求字节。无需密钥的组装快照覆盖真实工具 schema 和图片请求路径。使用凭据的测试只使用内置 `deepseek-official` 路由及其已配置端点，不使用自定义提供方条目。
+包测试会生成 16-bit RGB 和 RGBA PNG fixture，验证 8-bit 转换与干净 8-bit 字节直通、字节压力下保留透明通道、首个候选合规后停止编码、字节目标不可达时保留最小阶梯产物、正方形和宽屏 640,000 像素投影、相同变体与上传 singleflight 且不会共享取消、变换并发上限、缓存与上传身份、跳过已持久省略的历史附件读取、批量只准备一次、Files 响应不一致、进入刷新余量时不查询远端并更新 ID、单个 ID、多个 ID 和模糊失效响应只恢复一次、文件解析失败后回退到有界全内联请求、删除配额文件前完成分页、规范化提供方诊断、纯文本投影，以及普通请求与压缩共享请求字节。无需密钥的组装快照覆盖真实工具 schema 和图片请求路径。使用凭据的测试只使用内置 `deepseek-official` 路由及其已配置端点，不使用自定义提供方条目。
 
 ## Consequences
 

.agents/notes/implemented/feature/2026-09-07-deepseek-messages-adapter.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 .agents/notes/implemented/feature/2026-09-07-deepseek-messages-adapter.md
-2026-09-07-deepseek-messages-adapter.md: 981ca1ec47f7faa3bcabf5db54393a73474e846c
-2026-09-07-deepseek-messages-adapter.zh.md: 8a1059cfa561b6576518c1ce454f0857450682c8
+2026-09-07-deepseek-messages-adapter.md: 9a92f9e95931bebfa9fa6e7e64fbd7d27ec8306c
+2026-09-07-deepseek-messages-adapter.zh.md: 67f6c97c0a416cbddb7ec9d0de01e47e658036b7

.agents/notes/implemented/feature/2026-09-07-deepseek-messages-adapter.md

@@ -10,17 +10,19 @@ Deployments expose DeepSeek through Anthropic Messages gateways as well as chat-
 
 ## Decision
 
-The [DeepSeek adapter](../../../../packages/llm/llm-deepseek/README.md) serves multiple protocols under one `deepseek-official` route and `llm-deepseek` settings namespace. `common/` shares configuration, the model catalog, and capability resolution; `protocols/chat-completions/` and `protocols/messages/` own serialization, stream conversion, and transport. Cordis YAML selects the implementation through `protocol`, defaulting to `chat-completions`. The existing `PreparedAdapterCall` freezes protocol, endpoint, credential reference, and model capabilities; retries retain that generation while subsequent calls read new configuration.
+The [DeepSeek adapter](../../../../packages/llm/llm-deepseek/README.md) serves multiple protocols under one `deepseek-official` route and `llm-deepseek` settings namespace. `common/` shares configuration, the model catalog, capability resolution, and Files lifecycle; `protocols/chat-completions/` and `protocols/messages/` own serialization, stream conversion, and transport. Cordis YAML selects the implementation through `protocol`, defaulting to `messages`; shipped first-party compositions inherit that default. The existing `PreparedAdapterCall` freezes protocol, endpoint, credential reference, and model capabilities; retries retain that generation while subsequent calls read new configuration.
 
-The adapter follows the [DeepSeek compatibility documentation](https://api-docs.deepseek.com/zh-cn/guides/anthropic_api) and [Anthropic streaming protocol](https://platform.claude.com/docs/en/build-with-claude/streaming). The pi-ai Anthropic implementation informed the handling of adjacent user messages, cumulative usage, fragmented tool arguments, and optional thinking signatures. DeepSeek effort uses `output_config.effort`; an Anthropic thinking token budget does not control DeepSeek effort.
+The adapter follows the [DeepSeek compatibility documentation](https://api-docs.deepseek.com/zh-cn/guides/anthropic_api) and [Anthropic streaming protocol](https://platform.claude.com/docs/en/build-with-claude/streaming). The pi-ai Anthropic implementation informed the handling of adjacent user messages, cumulative usage, fragmented tool arguments, and optional thinking signatures. DeepSeek effort uses `output_config.effort`; an Anthropic thinking token budget does not control DeepSeek effort. Both protocols forward explicit `temperature` values; DeepSeek accepts that parameter with thinking enabled and ignores its value, so callers retain their existing thinking configuration.
 
 Assistant blocks remain the durable model-visible content. A versioned `ReplayEnvelope` stores only the protocol format, model identity, aligned block kinds, and signatures absent from those blocks. Same-model Messages continuation restores signatures verbatim, including empty signatures; foreign history carries no invented signature. Unusable metadata follows the existing [replay degradation rule](../architecture/2026-07-14-provider-routed-llm-adapters.md): the request omits signatures with a warning while preserving durable content; content validation such as tool argument parsing still fails explicitly. This keeps provider replay data opaque to the loop while preserving it through Session persistence and block pruning.
 
-Image requests use bounded inline base64 versions from the attachment service. Shared attachment offload and DeepSeek token measurement keep request and measurement policy consistent. Files uploads remain outside this adapter because their endpoints and cache ownership differ from chat-completions; adding them requires a Messages-specific lifetime and error policy.
+Both protocols prefer Files references for deterministic request images and share upload caching, refresh, quota recovery, and attachment offload. The Files client retains the selected protocol and configured endpoint: Messages uses `/v1/files` with its required beta header, while Chat Completions uses `/files`. Cached ids remain scoped by configured endpoint and credential. Messages metadata omits expiry, so local reuse is bounded from the original upload time without asserting remote deletion. A Files-resolution failure rebuilds the complete request under the independent inline-image budget; caller cancellation stops it. The shared image policy preserves the 128 MiB retained-image budget, 20 MiB inline base64 budget, and oldest-prefix offload in both requests and token measurement.
 
 System updates use the existing [route capability](2026-09-02-in-history-system-prompt-replacement.md) when explicitly declared for an endpoint/model. Messages retains the initial top-level system and emits later snapshots as native system turns after the corresponding user/tool-result turn, preserving previously sent prefixes. This placement differs from the loop's system-before-user admission; serialization changes neither the durable log nor conversation-turn order. Undeclared routes consolidate the latest snapshot at the top level, including direct compaction calls. Capability inference from protocol or model names is insufficient because support and update semantics depend on the deployed endpoint.
 
-Web always displays DeepSeek without a protocol selector. Both protocols share `baseURL` and `apiKeyEnv`, with no nested per-protocol configuration map. Without an endpoint override, resolution uses the selected protocol’s official default; Messages uses `https://api.deepseek.com/anthropic`. Switching retains existing endpoint overrides, whose compatibility belongs to the deployment. One model catalog includes `deepseek-flash` text/image and in-history system capabilities and retains the V4 entries.
+Web always displays DeepSeek without a protocol selector. Both protocols share `baseURL` and `apiKeyEnv`, with no nested per-protocol configuration map. Without an endpoint override, resolution uses the selected protocol’s official default; Messages uses `https://api.deepseek.com/anthropic`. Switching retains existing endpoint overrides, whose compatibility belongs to the deployment. One model catalog includes `deepseek-flash` text/image and in-history system capabilities and retains the V4 entries. Explicit `chat-completions` remains supported with its own official default; a custom `baseURL` or environment override is never rewritten to match a protocol.
+
+Both transports use the existing [request-extension registry](../architecture/2026-08-21-deepseek-llm-api-request-extensions.md) after native serialization and accept captured contributions after HTTP 2xx, before reading the stream. Session-log delivery and plugin inventory retain their owners and remain outside model input. The auxiliary [web-search provider](../../../../packages/web/web-search-deepseek/README.md) retains its separate endpoint, request, and settings.
 
 ## Alternatives considered
 
@@ -34,6 +36,6 @@ Web always displays DeepSeek without a protocol selector. Both protocols share `
 
 ## Consequences
 
-The package owns wire validation, stop-reason mapping, cancellation, and error classification, so protocol changes require adapter maintenance. Unsupported content and incomplete streams fail explicitly. The existing retry consumer owns retries; the existing assembler drops incomplete tool calls at the output limit. The shared base and Web default to Chat Completions; Messages requires explicit opt-in.
+The package owns wire validation, stop-reason mapping, cancellation, and error classification, so protocol changes require adapter maintenance. Unsupported content and incomplete streams fail explicitly. The existing retry consumer owns retries; the existing assembler drops incomplete tool calls at the output limit. Messages defaults cover the shared base, Web, and standalone first-party compositions. Protocol changes preserve the provider id and saved model selections, but an explicit endpoint override must support the selected protocol.
 
 Verification covers wire fixtures, real Loader composition, per-file unit coverage, [recorded Session replay](../../../../snapshots/session/deepseek-messages-replay/snapshot.yml) with [unknown replay versions](../../../../snapshots/session/deepseek-messages-degraded-replay/snapshot.yml), a Web Messages Session replay, and credential-gated text, thinking, tool continuation, image, and cancellation requests. Live gateway checks establish compatibility with the configured gateway; they do not establish compatibility with every Anthropic proxy.

.agents/notes/implemented/feature/2026-09-07-deepseek-messages-adapter.zh.md

@@ -10,17 +10,19 @@ Status: implemented
 
 ## 决策
 
-[DeepSeek 适配器](../../../../packages/llm/llm-deepseek/README.zh.md)通过一个 `deepseek-official` 路由和 `llm-deepseek` 设置命名空间支持多个协议。`common/` 共享配置、模型目录和能力解析；`protocols/chat-completions/` 与 `protocols/messages/` 分别负责协议序列化、流转换和传输。`protocol` 配置在 Cordis YAML 中选择实现，默认 `chat-completions`。已有 `PreparedAdapterCall` 冻结协议、端点、凭据引用与模型能力，重试保持同一代配置，后续调用读取新配置。
+[DeepSeek 适配器](../../../../packages/llm/llm-deepseek/README.zh.md)通过一个 `deepseek-official` 路由和 `llm-deepseek` 设置命名空间支持多个协议。`common/` 共享配置、模型目录、能力解析和 Files 生命周期；`protocols/chat-completions/` 与 `protocols/messages/` 分别负责协议序列化、流转换和传输。`protocol` 配置在 Cordis YAML 中选择实现，默认 `messages`；随产品交付的官方组合继承该默认值。已有 `PreparedAdapterCall` 冻结协议、端点、凭据引用与模型能力，重试保持同一代配置，后续调用读取新配置。
 
-适配器遵循 [DeepSeek 兼容文档](https://api-docs.deepseek.com/zh-cn/guides/anthropic_api) 和 [Anthropic 流协议](https://platform.claude.com/docs/en/build-with-claude/streaming)。pi-ai 的 Anthropic 实现为相邻用户消息、累计用量、工具参数分片和可选思考签名的处理提供参考。DeepSeek 通过 `output_config.effort` 设置思考强度；Anthropic 思考 token 预算不控制 DeepSeek 思考强度。
+适配器遵循 [DeepSeek 兼容文档](https://api-docs.deepseek.com/zh-cn/guides/anthropic_api) 和 [Anthropic 流协议](https://platform.claude.com/docs/en/build-with-claude/streaming)。pi-ai 的 Anthropic 实现为相邻用户消息、累计用量、工具参数分片和可选思考签名的处理提供参考。DeepSeek 通过 `output_config.effort` 设置思考强度；Anthropic 思考 token 预算不控制 DeepSeek 思考强度。两种协议都转发显式 `temperature` 值；DeepSeek 在启用思考时接受该参数但忽略其值，因此调用方可以保留已有思考配置。
 
 助手内容块保留持久化的模型可见内容。带版本的 `ReplayEnvelope` 仅保存协议格式、模型标识、对齐的块类型以及内容块未包含的签名。同模型续接原样恢复签名，包括空签名；外部历史不生成虚构签名。不可用的元数据遵循现有[回放降级规则](../architecture/2026-07-14-provider-routed-llm-adapters.zh.md)：请求省略签名并记录警告，保留持久化内容；工具参数等内容校验仍会正常报错。提供者回放数据对循环保持不透明，同时能够随 Session 持久化和内容块裁剪保留。
 
-图片请求使用附件服务生成的、有预算限制的内联 base64 版本。共享附件卸载机制和 DeepSeek token 计量使请求与计量策略保持一致。此适配器不负责 Files 上传，因为其端点和缓存所有权与 chat-completions 不同；增加上传支持需要定义 Messages 专属的生命周期和错误策略。
+两种协议均优先为确定性请求图片使用 Files 引用，并共享上传缓存、刷新、配额恢复和附件卸载。Files 客户端保留所选协议与已配置端点：Messages 使用 `/v1/files` 并携带必需的 beta 标头，Chat Completions 使用 `/files`。缓存 id 仍按配置的端点和凭据限定作用域。Messages 元数据不含过期时间，因此本地复用从原始上传时间起受限，但不宣称远端文件已删除。Files 解析失败会按独立的内联图片预算重建完整请求；调用方取消则停止请求。共享图片策略在请求与 token 计量中保留 128 MiB 的保留图片预算、20 MiB 的内联 base64 预算，以及最旧前缀卸载。
 
 系统提示词更新在端点与模型显式声明支持时，使用现有[路由能力](2026-09-02-in-history-system-prompt-replacement.zh.md)。Messages 保留初始顶层 system，在对应的用户或工具结果轮次之后，将后续快照发送为原生 system 轮次，保留此前发送的前缀。这个位置不同于循环先 system、后 user 的接纳顺序；序列化既不改写持久化日志，也不改变对话轮次的顺序。未声明能力的路由将最新快照归并到顶层，直接压缩调用也如此。仅凭协议或模型名称推断能力并不充分，因为支持情况和更新语义取决于实际部署的端点。
 
-Web 始终显示 DeepSeek，不提供协议选择器。两个协议共用 `baseURL` 与 `apiKeyEnv`，没有嵌套的协议配置表。未提供地址覆盖时使用当前协议的官方默认值；Messages 为 `https://api.deepseek.com/anthropic`。切换协议保留已有端点覆盖，部署者负责其兼容性。模型目录只维护一份，包含 `deepseek-flash` 的文本/图片和历史内 system 更新能力，也保留 V4 条目。
+Web 始终显示 DeepSeek，不提供协议选择器。两个协议共用 `baseURL` 与 `apiKeyEnv`，没有嵌套的协议配置表。未提供地址覆盖时使用当前协议的官方默认值；Messages 为 `https://api.deepseek.com/anthropic`。切换协议保留已有端点覆盖，部署者负责其兼容性。模型目录只维护一份，包含 `deepseek-flash` 的文本/图片和历史内 system 更新能力，也保留 V4 条目。显式 `chat-completions` 仍受支持，并使用自己的官方默认值；不会为匹配协议而改写自定义 `baseURL` 或环境覆盖。
+
+两种传输都在原生序列化后使用现有[请求扩展注册表](../architecture/2026-08-21-deepseek-llm-api-request-extensions.zh.md)，并在 HTTP 2xx 后、读取流之前接受已捕获贡献。会话日志投递和插件清单仍由原有包负责，并留在模型输入之外。辅助 [web 搜索提供方](../../../../packages/web/web-search-deepseek/README.zh.md)保留独立的端点、请求与设置。
 
 ## 考虑过的替代方案
 
@@ -34,6 +36,6 @@ Web 始终显示 DeepSeek，不提供协议选择器。两个协议共用 `baseU
 
 ## 结果
 
-该包负责协议校验、停止原因映射、取消和错误分类，因此协议变化需要维护适配器。不支持的内容和不完整的流会明确报错。现有重试消费者负责重试；现有装配器在输出达到上限时丢弃未完成的工具调用。共享 base 与 Web 默认使用 Chat Completions；Messages 需要显式启用。
+该包负责协议校验、停止原因映射、取消和错误分类，因此协议变化需要维护适配器。不支持的内容和不完整的流会明确报错。现有重试消费者负责重试；现有装配器在输出达到上限时丢弃未完成的工具调用。Messages 默认配置覆盖共享 base、Web 和独立的官方组合。协议变化保留 provider id 与已保存的模型选择，但显式端点覆盖必须支持所选协议。
 
 验证覆盖协议夹具、真实 Loader 组合、逐文件单元覆盖率、[已记录 Session 回放](../../../../snapshots/session/deepseek-messages-replay/snapshot.yml)与[未知回放版本](../../../../snapshots/session/deepseek-messages-degraded-replay/snapshot.yml)，Web Messages Session 回放，以及凭证控制的文本、思考、工具续接、图片和取消请求。真实网关检查证明与已配置网关的兼容性，不能证明与所有 Anthropic 代理兼容。

.agents/notes/implemented/feature/2026-09-09-ptc-trajectory-code-inspection.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 .agents/notes/implemented/feature/2026-09-09-ptc-trajectory-code-inspection.md
+2026-09-09-ptc-trajectory-code-inspection.md: 477d199f519b5d515e5d58430bd902d9d209e46b
+2026-09-09-ptc-trajectory-code-inspection.zh.md: f305569176214c63ac549b0ec5103289e3e57a88

.agents/notes/implemented/feature/2026-09-09-ptc-trajectory-code-inspection.md

@@ -0,0 +1,29 @@
+# Agent Note: Inspect recorded PTC source in Trajectory
+
+Status: implemented
+
+English | [中文](2026-09-09-ptc-trajectory-code-inspection.zh.md)
+
+## Problem
+
+PTC programs arrive as JSON string arguments. Escaping makes long programs difficult to read in a generic argument tree, while historical calls may use a different runtime language from the current deployment. Recorded result text can contain both printed output and a returned value without retaining their separation.
+
+## Decision
+
+[Trajectory](../../../../packages/client/ui-trajectory/README.md) identifies calls by their recorded tool name and derives a code inspector from validated `run_code` arguments. The validated program carries the original JSON text; copying preserves source and argument bytes and exposes the original JSON through a separate toggle. Syntax highlighting requires an unambiguous TypeScript or Python hint in that call's recorded parameter description; unknown or conflicting hints leave plain text. Unsupported arguments retain the generic inspector.
+
+The result view preserves recorded text and uses a tree only for complete JSON objects or arrays. Errors retain captured output. Code views sample the shared wrapping preference when opened and keep their own choice while mounted.
+
+The [PTC runtime decision](2026-06-15-ptc.md) still owns execution and settlement; the [client presentation decision](../architecture/2026-08-23-client-derived-tool-presentation.md) still owns deriving UI from recorded facts. This inspector adds no Session events or host presentation fields.
+
+## Alternatives considered
+
+**Keep source inside the argument tree.** JSON escaping obscures program structure and makes copying executable source cumbersome.
+
+**Use the active runtime language or infer it from source.** Either can mislabel a historical program. Recorded schema hints constrain highlighting without changing the recording.
+
+**Split printed output from return values.** Historical rendered text does not establish that distinction; a parser could invent a separation the producer never recorded.
+
+## Consequences
+
+Readers can inspect and copy recorded programs without changing replay data. Schemas with no recognizable language hint receive no syntax highlighting. Component tests cover recorded-name recognition, schema fallback, exact source and argument copying, output states, and independent wrapping. JSON-tree tests cover clipping geometry, missing `ResizeObserver`, clipboard settlement after hover changes or unmount, and value-read counts during hover. Thinking tests cover body arrival, manual disclosure, and switching records; the [PTC browser scenario](../../../../apps/web/tests/ptc-round.e2e.ts) pins the assembled inspector and verifies overflow and the original-JSON round trip.

.agents/notes/implemented/feature/2026-09-09-ptc-trajectory-code-inspection.zh.md

@@ -0,0 +1,29 @@
+# Agent Note: 在 Trajectory 中检查已记录的 PTC 源码
+
+Status: implemented
+
+[English](2026-09-09-ptc-trajectory-code-inspection.md) | 中文
+
+## 问题
+
+PTC 程序以 JSON 字符串参数传入。转义使长程序在通用参数树中难以阅读，而历史调用使用的运行时语言可能不同于当前部署。记录的结果文本可以同时包含打印输出与返回值，却未保留二者的分界。
+
+## 决策
+
+[Trajectory](../../../../packages/client/ui-trajectory/README.zh.md) 按已记录的工具名识别调用，并从已验证的 `run_code` 参数派生代码检查器。验证后的程序携带原始 JSON 文本；复制时保留源码与参数字节，并通过独立切换按钮展示原始 JSON。语法高亮要求该调用记录的参数说明包含明确且无冲突的 TypeScript 或 Python 提示；未知或冲突的提示使用纯文本。不支持的参数保留通用检查器。
+
+结果视图保留记录的文本，只对完整 JSON 对象或数组使用树形展示。错误保留已捕获的输出。代码视图在打开时读取共享换行偏好，并在挂载期间保留自身的选择。
+
+[PTC 运行时决策](2026-06-15-ptc.zh.md) 仍负责执行与结算；[客户端展示决策](../architecture/2026-08-23-client-derived-tool-presentation.zh.md) 仍负责从已记录事实派生 UI。此检查器不增加 Session 事件或宿主展示字段。
+
+## 考虑过的替代方案
+
+**把源码保留在参数树中。** JSON 转义遮蔽程序结构，也使复制可执行源码变得繁琐。
+
+**使用当前运行时语言，或从源码推断语言。** 两种方式都可能错误标注历史程序。已记录 Schema 的提示约束高亮，无需修改记录。
+
+**拆分打印输出与返回值。** 历史渲染文本无法确定二者的分界；解析器可能凭空添加生产方从未记录的分隔。
+
+## 后果
+
+读者可以检查和复制已记录的程序，无需修改回放数据。Schema 没有可识别的语言提示时不提供语法高亮。组件测试覆盖记录工具名识别、Schema 回退、源码与参数原样复制、输出状态及独立换行。JSON 树测试覆盖裁剪几何、缺少 `ResizeObserver`、悬停切换或卸载后剪贴板写入落定，以及悬停期间读取值的次数。思考测试覆盖正文到达、手动展开折叠及记录切换；[PTC 浏览器场景](../../../../apps/web/tests/ptc-round.e2e.ts) 固定组装后的检查器展示，并验证溢出和原始 JSON 的往返切换。

.agents/notes/implemented/feature/2026-09-09-web-sidebar-terminal.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 .agents/notes/implemented/feature/2026-09-09-web-sidebar-terminal.md
+2026-09-09-web-sidebar-terminal.md: ce1162a84d3f96336cbb217dbdfe62d12ac23280
+2026-09-09-web-sidebar-terminal.zh.md: 51e0f374b5fde389323868e56fbd86d3bc5720b3

.agents/notes/implemented/feature/2026-09-09-web-sidebar-terminal.md

@@ -0,0 +1,45 @@
+# Agent Note: Web sidebar terminals
+
+Status: implemented
+
+English | [中文](2026-09-09-web-sidebar-terminal.zh.md)
+
+## Problem
+
+Web users need an interactive shell beside a Session to inspect the workspace and run commands. The Agent's persistent terminal tools control prompts and wait for semantic results; a human terminal instead needs raw keyboard input, normal shell configuration and a full screen. Browser rendering and transport can disappear while a command is still running.
+
+## Decision
+
+`api-terminal-controller` owns user terminals by Session and exposes the `terminal` Remote namespace. `ui-sidebar-terminal` registers native right-sidebar tabs, xterm.js rendering and FitAddon sizing. A new terminal offers installed shells and waits for Start. Host discovery verifies the configured candidates, with the execution default first; creation accepts only a currently discovered path. The browser remembers the last selected shell path in origin-scoped localStorage and falls back to the current default if that path is unavailable. The terminal type declares independent instances, so ordinary page deduplication cannot collapse separate processes when opening or docking tabs. The existing sidebar controls open additional tabs; double-clicking a tab title renames its terminal. Terminal processes use the composed subprocess provider and Session sandbox policy. Shell resolution occurs during discovery and creation; reading limits and reconnecting an existing process do not depend on the default executable remaining available. Interactive shell configuration supplies Tab completion and optional inline suggestions.
+
+Close and replacement remove the tab synchronously and run process cleanup in the background. The Client first records the unfinished close request under a terminal-specific localStorage key; success removes it, and startup retries requests that remain. A cleanup failure produces a lightweight notification with a retry action without reopening the tab. Independent keys prevent another window from overwriting unrelated cleanup requests. Collapse, tab/Session switching, floating, fullscreen and browser disconnection preserve the process. Component cleanup and `TabDomain.signal` only detach browser work because the same lifetime can end during plugin reload. Failed process cleanup retains ownership, including failures after allocation but before create publication. Session owner disposal and Host plugin disposal also clean up terminals. A definitive missing-Session response retires its saved close request because the Session owns process cleanup; transport failures remain retryable. Client plugin disposal awaits every detached stream so a replacement plugin does not inherit unfinished Client cleanup.
+
+Sidebar layout, open-tab mappings, selection and process PIDs are not persisted. When the Session header mounts, the Client queries `terminal.list` and opens retained Host terminals as new tabs. Listing takes the Session ID directly because history can outlive its Agent and terminal owner; an offline Session has no retained terminals to restore. Their `params.terminalId` association exists only in the current page. New and recovered views use different `createWhenMissing` values: only a new view may allocate a process; a recovered target that disappears reports an error. Host state supplies recovery identities and titles, so the browser does not maintain a second active-terminal registry.
+
+The caller retains a terminal ID in memory before creating it. Repeating create with the same Session and open ID does not allocate a second process. Closing an uncertain create uses that ID even when no creation response arrived. The Host records closed IDs before awaiting allocation, preventing a delayed create from reviving a closed terminal. Each attachment begins with a consistent, bounded serialized xterm screen; ordered output follows through the Gateway's existing multiplexed Remote stream. Output and screen snapshots share one operation queue. Followers retain final output on normal closure and fail explicitly on buffer overflow. Browser render acknowledgement prevents React batching from dropping increments.
+
+The latest attachment controls input and dimensions; other attachments remain read-only. Every physical stream opening gets a fresh input attachment identity, including automatic transport recovery. Input RPCs are serialized, and results from a superseded attachment cannot downgrade a replacement connection. Terminal output creates no model input, Agent tool result or Session event. The existing [persistent Agent terminal decision](2026-07-16-persistent-pty-sessions.md) continues to govern model-owned sessions; this feature extends the [portable subprocess provider](../architecture/2026-07-28-portable-execution-world-consumers.md) with terminal environment facts and resize. Control-transfer and process-exit refusals only disable input, preserving the healthy output view. Client-owned error identifiers are translated by the terminal UI, including guidance to close retained exited terminals when the quota is full.
+
+## Alternatives considered
+
+**Persist complete shell profiles in the browser.** Only the selected path is a user preference. Arguments and availability belong to Host discovery; persisting them would allow stale profiles to bypass current execution configuration. The sidebar continues to own the terminal list and tab controls.
+
+**Keep the tab visible until process cleanup finishes.** A slow or failed termination would delay the user's close action. Saving the cleanup intent allows immediate removal while preserving failure reporting and retry.
+
+**Persist sidebar layout or an active tab-to-process registry.** Layout persistence is outside this feature. An additional active registry duplicates Host state and can restore stale associations. Only an unfinished close is an independent user request that must survive page reload.
+
+**Share the Agent terminal registry.** Its controlled prompts and semantic send/wait behavior would change human shell configuration and blur process ownership. User terminals share the subprocess capability instead.
+
+**Add a dedicated terminal WebSocket.** The existing Remote stream transport already owns authentication, cancellation and reconnection. A second carrier would duplicate those responsibilities.
+
+**Replay only a bounded raw byte tail.** A tail can begin inside an escape sequence or omit an alternate-screen transition. A serialized terminal screen provides a consistent recovery point with bounded history.
+
+**Kill when a React body or tab signal is disposed.** Unmount, Session switching and plugin reload can end those lifetimes without an explicit close. Cleanup must follow the user's close operation.
+
+**Build a Web completion engine.** Native shell completion already handles commands, paths and configured plugins through ordinary PTY input. An independent completion UI adds shell-specific parsing and synchronization; it is outside this feature.
+
+## Consequences
+
+A kept-open terminal retains a process and bounded screen memory. Reload restores Host-retained terminals rather than the previous sidebar layout; Host restart does not restore processes. An exited shell remains visible without automatic respawn. Background cleanup may outlive its tab, and unavailable browser storage limits retry recovery to the current page. Native PTY support and descendant cleanup guarantees remain provider-specific. One writable attachment avoids competing resize and input streams, while explicit takeover permits recovery from another page. Changing sandbox mode requires closing retained terminals first.
+
+The implementation retains the Agent-terminal and portable-execution notes because their ownership and provider decisions remain independently useful; neither is superseded by browser terminals.

.agents/notes/implemented/feature/2026-09-09-web-sidebar-terminal.zh.md

@@ -0,0 +1,45 @@
+# Agent Note: Web sidebar terminals
+
+Status: implemented
+
+[English](2026-09-09-web-sidebar-terminal.md) | 中文
+
+## 问题
+
+Web 用户需要在 Session 旁使用交互式 shell 检查工作区和运行命令。Agent 的持久终端工具控制提示符并等待语义结果；人工终端需要原始键盘输入、正常 shell 配置和完整屏幕。命令仍在运行时，浏览器渲染和网络连接可能消失。
+
+## 决定
+
+`api-terminal-controller` 按 Session 管理用户终端并提供 `terminal` Remote namespace。`ui-sidebar-terminal` 注册原生右侧栏标签页，使用 xterm.js 渲染和 FitAddon 测量尺寸。新终端提供已安装 shell 的选择并等待用户启动。Host 探测会验证配置的候选，并把执行环境默认项放在首位；创建只接受当前探测返回的路径。浏览器在当前站点 localStorage 中记住上次选择的 shell 路径，该路径不可用时回到当前默认项。终端类型声明独立实例，因此打开或停靠标签页时，普通页面的去重规则不会合并不同进程。已有侧栏控件负责打开更多标签页，双击标签页标题可重命名终端。终端进程使用组合的 subprocess provider 和 Session sandbox policy。shell 在探测和创建时解析；读取限制和重新连接已有进程不依赖默认可执行文件仍然可用。交互式 shell 配置提供 Tab 补全和可选的内联建议。
+
+关闭和替换会同步移除标签页，并在后台清理进程。Client 先以终端独立的 localStorage key 保存未完成的关闭请求；成功后删除，启动时重试剩余请求。清理失败时显示带重试操作的轻量通知，不重新打开标签页。独立 key 避免其他窗口覆盖无关的清理请求。折叠、切换标签页或 Session、浮动、全屏和浏览器断线均保留进程。组件清理和 `TabDomain.signal` 只停止浏览器工作，因为插件重新加载也会结束这些生命周期。进程清理失败时保留所有权，包括分配完成但 create 尚未发布时的失败。Session owner 和 Host 插件卸载也会清理终端。 明确的 Session 不存在响应会清除已保存的关闭请求，因为进程清理由 Session 负责；传输失败仍可重试。Client 插件卸载等待所有断开的流结束，避免替换插件继承未完成的 Client 清理。
+
+侧栏布局、打开标签页映射、选中项和进程 PID 不持久化。Session header 挂载时，Client 查询 `terminal.list`，把 Host 保留的终端打开为新标签页。列表直接使用 Session ID，因为历史记录可以比 Agent 和终端 owner 存活更久；离线 Session 没有需要恢复的保留终端。`params.terminalId` 关联只在当前页面中保留。新视图与恢复视图使用不同的 `createWhenMissing`：只有新视图可以分配进程，恢复目标消失时显示错误。恢复标识和标题来自 Host 状态，浏览器不维护第二份活跃终端注册表。
+
+调用者在创建之前把终端 ID 保留在内存中。相同 Session 和未关闭 ID 的重复 create 不再分配进程。创建结果不确定时，关闭仍使用该 ID，即使没有收到创建响应。Host 在等待分配完成前记录已关闭 ID，防止迟到的 create 复活已关闭终端。每次连接先接收一致、有界的 xterm 序列化屏幕，后续有序输出使用 Gateway 已有的复用 Remote stream。输出和屏幕快照共享操作队列。订阅者正常关闭时保留末尾输出，缓存超限时明确失败。浏览器在完成渲染后确认帧，避免 React 批处理丢失增量。
+
+最新连接控制输入和尺寸，其他连接保持只读。每次物理流建立都创建新的输入连接标识，包括传输自动恢复。输入 RPC 按序发送，旧连接的结果不能把新连接降级为失败。终端输出不产生模型输入、Agent 工具结果或 Session 事件。[Agent 持久终端决策](2026-07-16-persistent-pty-sessions.zh.md)仍约束模型拥有的终端；此功能为[可移植 subprocess provider](../architecture/2026-07-28-portable-execution-world-consumers.zh.md)增加执行环境事实和 resize。 控制权转移和进程退出导致的拒绝只禁用输入，保留健康的输出视图。Client 自产错误标识由终端 UI 翻译，包括名额用满时关闭已保留的退出终端的提示。
+
+## 考虑过的替代方案
+
+**在浏览器保存完整 shell profile。** 用户偏好只包含选中的路径。参数和可用性由 Host 探测决定；持久化这些信息会让过期 profile 绕过当前执行配置。终端列表和标签页控件仍由侧栏负责。
+
+**进程清理完成前保留标签页。** 缓慢或失败的终止会拖延用户关闭操作。保存清理意图后，可以立即移除标签页，同时保留错误反馈与重试。
+
+**持久化侧栏布局或活跃标签页到进程的注册表。** 布局持久化不属于此功能。额外的活跃注册表重复 Host 状态，可能恢复过期关联。只有未完成的关闭操作是需要跨页面刷新保留的独立用户请求。
+
+**共用 Agent 终端注册表。** 受控提示符和语义化 send/wait 会改变人工 shell 配置并混淆进程所有权。用户终端只共享 subprocess 能力。
+
+**增加专用终端 WebSocket。** 现有 Remote stream 已管理认证、取消和重连。第二条传输通道会重复这些职责。
+
+**只重放有界原始字节尾部。** 字节尾部可能从转义序列中间开始，或缺失备用屏幕切换。序列化终端屏幕能在保留有界历史的同时提供一致恢复点。
+
+**React 正文或 tab signal 卸载时 kill。** 组件卸载、Session 切换和插件重载都可能结束这些生命周期，而用户并未关闭终端。清理必须跟随显式关闭操作。
+
+**自建 Web 补全引擎。** shell 原生补全通过普通 PTY 输入处理命令、路径和已配置插件。独立补全界面需要针对 shell 解析和同步，不属于此功能。
+
+## 影响
+
+保留终端会保留进程和有界屏幕内存。刷新恢复的是 Host 保留的终端，不是此前的侧栏布局；Host 重启不恢复进程。shell 退出后保持可见，不自动重启。后台清理可能比标签页存活更久，浏览器存储不可用时只能在当前页面保留重试能力。原生 PTY 支持和后代进程清理保证仍由 provider 决定。单一可写连接避免竞争的输入和尺寸流，显式接管允许从另一页面恢复操作。改变 sandbox mode 前需要关闭保留的终端。
+
+Agent 终端和可移植执行环境两篇记录仍保留，其所有权与 provider 决策继续独立有效，不被浏览器终端取代。

.agents/notes/implemented/feature/2026-09-12-mcp-resources-and-instructions.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 .agents/notes/implemented/feature/2026-09-12-mcp-resources-and-instructions.md
-2026-09-12-mcp-resources-and-instructions.md: 298c334bf357048c43d1c17c4cabb7d31d77c5bb
-2026-09-12-mcp-resources-and-instructions.zh.md: c79b11d8be1a67cb5db2e716e1e9ec2923f76b63
+2026-09-12-mcp-resources-and-instructions.md: a5eabd6170507d5a4df295b8554e6da47b40b9be
+2026-09-12-mcp-resources-and-instructions.zh.md: 44890b23944f482a2e1a38483029f1e61caebbe1

.agents/notes/implemented/feature/2026-09-12-mcp-resources-and-instructions.md

@@ -12,7 +12,7 @@ MCP servers expose documents and URI templates separately from tools. A tools-on
 
 [`mcp-resources`](../../../../packages/mcp/mcp-resources/README.md) provides three shared tools for listing resources, listing templates, and reading a URI. Each requires an explicit configured server name. When system-prompt assembly is composed, a literal section derives the caller-visible names from the dispatch registry, so resource-only servers remain discoverable without server instructions. The existing system-message log records those names; provider disposal removes them from later assemblies. The execution path resolves that server in the calling agent's scope before dispatch; provider registrations use reversible Cordis effects.
 
-One opt-in service mount installs the shared tools. Each [`mcp-client`](../../../../packages/mcp/mcp-client/README.md) instance owns its connection and registers a resource provider when the service is mounted. Resource operations require the server's resource capability; servers need not advertise tools. The official SDK owns protocol operations; list cursors and resource URIs remain opaque, and an explicit cursor requests one page while an omitted cursor lets the SDK collect pages.
+The [profile availability decision](2026-09-13-mcp-resources-in-profiles.md) supersedes the separate opt-in resource mount and owns conditional tool visibility. This note retains the resource operations, result representation, and instruction decisions. Each [`mcp-client`](../../../../packages/mcp/mcp-client/README.md) instance owns its connection and registers a resource provider when the service is mounted. The SDK returns empty resource and template lists when the server lacks the resource capability; unsupported reads fail. Servers need not advertise tools. The official SDK owns protocol operations; list cursors and resource URIs remain opaque, and an explicit cursor requests one page while an omitted cursor lets the SDK collect pages.
 
 Resource results preserve the complete canonical JSON for programmatic callers. Native text includes the configured server name and returned URI metadata. String-valued `blob` fields become binary descriptions instead of inline base64. Existing tool-result logging records the model projection; this package does not create a parallel resource log or a binary attachment store.
 
@@ -36,4 +36,4 @@ The [resource tests](../../../../packages/mcp/mcp-resources/tests/resources.spec
 
 ## Consequences
 
-Resource-only servers become useful without adding per-server model tools. Caller-visible server names and server instructions add prompt tokens; resource documents add tokens only when read. Shared schemas stay stable as provider availability changes, but a call still fails when its selected server is unavailable. Binary resources remain programmatic values, and pagination follows the SDK.
+Resource-only servers become useful without adding per-server model tools. Caller-visible server names and server instructions add prompt tokens; resource documents add tokens only when read. Shared schemas stay stable during connection failures while a caller-visible client remains configured, but a call still fails when its selected server is unavailable. Binary resources remain programmatic values, and pagination follows the SDK.

.agents/notes/implemented/feature/2026-09-12-mcp-resources-and-instructions.zh.md

@@ -12,7 +12,7 @@ MCP 服务器将文档与 URI 模板作为独立于工具的能力暴露。只
 
 [`mcp-resources`](../../../../packages/mcp/mcp-resources/README.zh.md) 提供三个共享工具，用于列出资源、列出模板及读取 URI。每个工具都要求显式指定已配置的服务器名称。组合包含系统提示词装配时，字面段落从派发注册表获取调用方可见的名称，因此没有服务器指令的纯资源服务器仍可被发现。已有的系统消息日志记录这些名称；提供方释放后，后续组装会移除其名称。执行路径在派发前于调用 agent 的作用域中解析该服务器；提供方注册使用可撤销的 Cordis effect。
 
-一次显式启用的服务挂载安装共享工具。每个 [`mcp-client`](../../../../packages/mcp/mcp-client/README.zh.md) 实例拥有自己的连接，并在服务已挂载时注册资源提供方。资源操作要求服务器具备资源能力；服务器无需声明工具能力。官方 SDK 负责协议操作；列表游标与资源 URI 保持不透明；显式游标请求一页，省略游标则由 SDK 汇总各页。
+[Profile 可用性决策](2026-09-13-mcp-resources-in-profiles.zh.md)取代单独启用资源服务的挂载方式，并拥有工具条件可见性。本笔记保留资源操作、结果表示与指令决策。每个 [`mcp-client`](../../../../packages/mcp/mcp-client/README.zh.md) 实例拥有自己的连接，并在服务已挂载时注册资源提供方。服务器缺少资源能力时，SDK 返回空的资源列表与模板列表；不受支持的读取会失败。服务器无需声明工具能力。官方 SDK 负责协议操作；列表游标与资源 URI 保持不透明；显式游标请求一页，省略游标则由 SDK 汇总各页。
 
 资源结果为程序化调用方保留完整规范 JSON。Native 文本包含已配置的服务器名称及返回的 URI 元数据。字符串值的 `blob` 字段变为二进制说明文字，不内联 base64。已有的工具结果日志记录模型投影；本包不创建并行的资源日志或二进制附件存储。
 
@@ -36,4 +36,4 @@ MCP 服务器将文档与 URI 模板作为独立于工具的能力暴露。只
 
 ## 后果
 
-仅提供资源的服务器无需添加按服务器区分的模型工具即可使用。调用方可见的服务器名称和服务器指令增加提示词 token；资源文档仅在读取时增加 token。提供方可用性变化时，共享 schema 保持稳定，但所选服务器不可用时调用仍会失败。二进制资源仍是程序化值，分页遵循 SDK。
+仅提供资源的服务器无需添加按服务器区分的模型工具即可使用。调用方可见的服务器名称和服务器指令增加提示词 token；资源文档仅在读取时增加 token。只要调用方可见的客户端仍已配置，连接失败期间共享 schema 就保持稳定，但所选服务器不可用时调用仍会失败。二进制资源仍是程序化值，分页遵循 SDK。

.agents/notes/implemented/feature/2026-09-12-mcp-sdk-protocol-negotiation.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 .agents/notes/implemented/feature/2026-09-12-mcp-sdk-protocol-negotiation.md
-2026-09-12-mcp-sdk-protocol-negotiation.md: 8934639ce12105a4549f16c8b5110a1661d3735d
-2026-09-12-mcp-sdk-protocol-negotiation.zh.md: 3e11a1aa5a3f4d1ebd94b6b0725b31024a491dc8
+2026-09-12-mcp-sdk-protocol-negotiation.md: 74f027cead92688b0af71c1028d536b26f2a8059
+2026-09-12-mcp-sdk-protocol-negotiation.zh.md: bc8a9a41b8ccd8cbc844b4b9de8b39a646d1dcbf

.agents/notes/implemented/feature/2026-09-12-mcp-sdk-protocol-negotiation.md

@@ -28,6 +28,6 @@ The [tool bridge note](2026-07-07-mcp-client-plugin.md) retains the independent
 
 ## Consequences
 
-Stdio negotiation starts a disposable probe process and waits for its exit before starting the serving process. The SDK bounds discovery with its page limit, and malformed results fail before projection. Valid text, canonical JSON, image admission, cancellation, and registration ownership remain bridge contracts. Resources have an optional consumer; elicitation, MCP prompts, and task execution remain unsupported.
+Stdio negotiation starts a disposable probe process and waits for its exit before starting the serving process. The SDK bounds discovery with its page limit, and malformed results fail before projection. Valid text, canonical JSON, image admission, cancellation, and registration ownership remain bridge contracts. Shipped profiles include shared resource access; elicitation, MCP prompts, and task execution remain unsupported.
 
 Real-SDK lifecycle tests verify probe disposal, process ordering, HTTP probe retry budgets, and failed stdio spawns. The connection-supervisor tests retain attached-transport close barriers and bounded failure behavior.

.agents/notes/implemented/feature/2026-09-12-mcp-sdk-protocol-negotiation.zh.md

@@ -28,6 +28,6 @@ MCP 服务器使用不同协议版本。围绕旧版 SDK 实现发现与执行
 
 ## 影响
 
-Stdio 协商启动可释放的探测进程，并等待其退出后才启动实际服务进程。SDK 通过页数上限约束发现，格式错误的结果会在投影前失败。有效文本、规范 JSON、图片接纳、取消与注册归属仍是桥接器的约定。资源有可选消费者；elicitation、MCP 提示模板及任务执行仍不受支持。
+Stdio 协商启动可释放的探测进程，并等待其退出后才启动实际服务进程。SDK 通过页数上限约束发现，格式错误的结果会在投影前失败。有效文本、规范 JSON、图片接纳、取消与注册归属仍是桥接器的约定。随附 profile 包含共享资源访问；elicitation、MCP 提示模板及任务执行仍不受支持。
 
 真实 SDK 生命周期测试验证探测释放、进程顺序、HTTP 探测重试预算及 stdio 启动失败。连接监督器测试保留已绑定传输的关闭屏障与有界失败行为。

.agents/notes/implemented/feature/2026-09-12-session-unarchive-settings-page.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 .agents/notes/implemented/feature/2026-09-12-session-unarchive-settings-page.md
+2026-09-12-session-unarchive-settings-page.md: 3b1e364e3abc6cea0a8b29ad78abb0ff7c274dbe
+2026-09-12-session-unarchive-settings-page.zh.md: 61e108dd50e61b9e6e9f72098973c067e3257407

.agents/notes/implemented/feature/2026-09-12-session-unarchive-settings-page.md

@@ -0,0 +1,45 @@
+# Agent Note: Session unarchive Settings page
+
+Status: implemented
+
+English | [中文](2026-09-12-session-unarchive-settings-page.zh.md)
+
+## Problem
+
+Archiving a Session removed it from every Workspace grouping surface, and nothing brought it back. The archive set is a durable display filter, so a hidden Session kept its log, its Workspace accounting slot, and its position, but the only recovered route was editing the domain state by hand. The archived feature record predicted the restore surface as "one UI surface plus one inverse RPC"; both now exist, and the Client had no place that listed archived Sessions at all.
+
+## Decision
+
+`WorkspaceRegistry.unarchiveSession(sessionId)` drops one id from the registry-global `archivedSessionIds` set in a single durable `setState`, running on the same operation chain as archive, create, and delete. The idempotent check-then-write pair sits inside one chain slot, so a concurrent archive cannot interleave and a lost race resolves as a no-op. An id that is not archived resolves without writing.
+
+Unarchiving runs no session-existence probe. Archive verifies that the Session is live or persisted because it adds a reference that must resolve; unarchive only removes an id, so it cannot introduce an unknown referent and an entry whose Session is gone still restores. `@Remote('unarchiveSession')` on `WorkspaceController` returns the complete `WorkspaceArchiveValue`, matching `archiveSession`, and `IWorkspaces.unarchiveSession` plus `UiWorkspace.unarchiveSession` carry the verb to the browser. Both verbs answer with the complete set, so `ClientWorkspaceModel` installs a reply only while it is still the latest archive-set request: a later request, or a set pushed by the follow stream, supersedes an in-flight answer and keeps the projection on the newer state.
+
+The new Web Settings page `@deepseek-ai/dsh-client-ui-settings-unarchive-sessions` owns the surface. It registers one localized `settings.section` contribution with id `archived-sessions` at nav order 25, joins the archive set from `useWorkspaces` with the loaded Session summaries from `useSessions`, lists the newest archive first with the owning Workspace title or the ungrouped label and a relative last-activity time, filters by title or Workspace name, and offers one Unarchive action per row. A rejected write is logged as a console diagnostic and leaves the row for another attempt. An archive entry whose Session summary is missing produces no row, so the page never renders an action that cannot restore anything.
+
+Restoring is a page whose subject is archived Sessions, and the archived Sessions themselves are hidden from every grouping surface, so no Session row can host the action. The archived-sessions page joins the Settings navigation beside General, Models, and Plugins, whose rail already carries the archive glyph this page declares in `SettingsRoot`.
+
+The `{ type: 'archived', archivedSessionIds }` follow increment already carries the complete set, so a restore reuses it: the Remote answer installs locally, and every other Client converges through the same increment. No frame type, persisted field, or `SESSION_FORMAT_VERSION` change accompanies the verb.
+
+## Alternatives considered
+
+**An Unarchive action in the Session row menu.** The row menu owns Archive, but a restored Session has no visible row until it is restored, so the action would have nowhere to live in the state that needs it; a disabled entry on every row would be decoration without a subject.
+
+**A second frame type carrying a removal delta.** The existing `archived` increment is a complete-set replacement, so a removal delta would add a redundant representation that every consumer would have to merge; the full-snapshot posture is what lets the archive echo be reused unchanged.
+
+**Rendering an archive entry whose Session is gone as a disabled row.** Such a row could explain the gap but not restore anything, and the registry and the Remote still accept the id, so the page lists what it can act on and leaves the orphan id to a future cleanup surface.
+
+**A Session-existence probe on unarchive, mirroring archive.** The check exists to keep added references resolvable, and a removal cannot break that invariant; probing would only turn restoring a deleted history into a failure with nothing to repair.
+
+## Consequences
+
+The archive set stays the only durable state a restore rewrites; the `workspace` domain version, the Session log, and the Workspace accounting slot are untouched, and a restored Session returns to its recorded position. Archive and unarchive now enforce different Session checks, a deliberate asymmetry recorded in the [Workspace registry limitations](../../../../packages/workspace/workspace/README.md#known-limitations-and-deferred-work).
+
+The restore surface is bounded by what the page can display: an archived id whose summary is not loaded has no row and no Unarchive action even though the registry method and the Remote accept it, so the page reports an unavailable set instead of an empty archive. Restoring through automation stays available, and a future cleanup surface can address the remaining orphan entries.
+
+## Testing
+
+`packages/workspace/workspace/tests/workspace.spec.ts` pins the registry method: removal keeps the surviving archive order, the accounting slot stays, a repeat and a never-archived id neither rewrite the medium nor emit a change, an entry whose Session is gone resolves without a persistence listing, and the surviving set reloads across a restart. `packages/api/workspace-controller/tests/workspace-controller.host.spec.ts` pins the verb's idempotent complete-set answer and the unchanged `archived` follow increment, and `packages/api/workspace-controller/tests/model.client.spec.ts` pins the Client model echo and its refusal to install a failed answer, plus the four stale-reply races: overlapping archive or unarchive requests settling out of order, and a follow increment or baseline arriving while a reply is in flight. `packages/client/ui-settings-unarchive-sessions/tests/components.client.spec.tsx` pins newest-first ordering, the ungrouped label, the hidden entry whose Session is gone, the reading and empty states, search, and the console diagnostic on rejection, while `tests/browser-plugin.client.spec.tsx` pins the section registration.
+
+## Related
+
+- [Session archive (registry-global set)](../../archived/feature/2026-07-31-session-archive-global-set.md) — the frozen record of the archive set, the follow increment, and the predicted restore surface.

.agents/notes/implemented/feature/2026-09-12-session-unarchive-settings-page.zh.md

@@ -0,0 +1,45 @@
+# Agent Note: Session unarchive Settings page
+
+Status: implemented
+
+[English](2026-09-12-session-unarchive-settings-page.md) | 中文
+
+## 问题
+
+归档会话会把它从每一个 Workspace 分组界面中移除，而没有任何东西能把它带回来。归档集合是持久的显示过滤器，因此被隐藏的会话保留其日志、Workspace 记账位置与所在位置，但唯一的恢复途径是手工编辑领域状态。归档功能记录把恢复界面预测为「一个 UI 界面加一个反向 RPC」；两者如今都已存在，而 Client 此前根本没有列出已归档会话的地方。
+
+## 决策
+
+`WorkspaceRegistry.unarchiveSession(sessionId)` 在一次持久化 `setState` 中把某个 id 从注册表全局的 `archivedSessionIds` 集合里移除，并与 archive、create、delete 跑在同一条操作链上。幂等的「先检查再写入」位于同一个链槽内，因此并发归档无法插入其间，输掉竞态的一方会解析为无操作。未被归档的 id 不写盘即解析完成。
+
+取消归档不做会话存在性探测。归档会校验会话处于实时或已持久化状态，因为它加入的引用必须可解析；取消归档只是移除一个 id，因此不可能引入未知引用，会话已不存在的条目也仍然能恢复。`WorkspaceController` 上的 `@Remote('unarchiveSession')` 返回完整的 `WorkspaceArchiveValue`，与 `archiveSession` 一致，而 `IWorkspaces.unarchiveSession` 与 `UiWorkspace.unarchiveSession` 把该动词带到浏览器。两个动词都返回完整集合，因此 `ClientWorkspaceModel` 只在该应答仍是最新归档集合请求时安装它：更晚的请求或 follow 流推送的集合都会取代在途应答，让投影停留在更新的状态上。
+
+新的 Web 设置页 `@deepseek-ai/dsh-client-ui-settings-unarchive-sessions` 拥有该界面。它注册一个 id 为 `archived-sessions`、导航顺序为 25 的本地化 `settings.section` 贡献，把来自 `useWorkspaces` 的归档集合与来自 `useSessions` 的已加载 Session 摘要合并，按归档时间由新到旧列出，并显示为其记账的 Workspace 标题或未分组标签，以及相对最近活动时间；它按标题或 Workspace 名称过滤，并为每行提供一个取消归档操作。写入被拒绝时会记录一条 console 诊断，并保留该行以便再次尝试。缺失 Session 摘要的归档条目不产生行，因此该页面绝不会渲染无法恢复任何东西的操作。
+
+恢复操作所在的页面以已归档会话为主题，而已归档会话本身从每一个分组界面中隐藏，因此没有任何 Session 行能承载该操作。「已归档会话」页在设置导航中与「通用」「模型」「插件」并列，而该导航轨道本就带有本页在 `SettingsRoot` 中声明的归档字形。
+
+`{ type: 'archived', archivedSessionIds }` follow 增量本就携带完整集合，因此恢复复用它：Remote 应答在本地安装，其他每个 Client 都通过同一个增量收敛。该动词不伴随任何帧类型、持久字段或 `SESSION_FORMAT_VERSION` 变更。
+
+## 考虑过的替代方案
+
+**在 Session 行菜单放一个取消归档操作。** 行菜单拥有 Archive，但被恢复的会话在恢复之前没有可见的行，因此该操作在真正需要它的状态里无处安放；给每一行加一个禁用条目只是没有主体的装饰。
+
+**新增一种携带移除增量的帧类型。** 现有的 `archived` 增量是完整集合替换，因此移除增量会增加一种每个消费方都必须合并的冗余表示；正是全快照姿态让归档回声得以原样复用。
+
+**把会话已不存在的归档条目渲染为禁用行。** 这样的行可以解释缺口，却无法恢复任何东西，而注册表与 Remote 仍然接受该 id，因此该页面只列出它能操作的条目，把遗留 id 留给未来的清理界面。
+
+**像归档一样，在取消归档时做会话存在性探测。** 该检查的存在是为了让新增的引用保持可解析，而移除不可能破坏这一不变式；探测只会把恢复已删除的历史变成一次无从修复的失败。
+
+## 后果
+
+归档集合仍是恢复唯一重写的持久状态；`workspace` 领域版本、会话日志与 Workspace 记账位置都不受影响，被恢复的会话回到其记录的位置。归档与取消归档现在执行不同的会话校验，这是刻意的不对称，并记录在 [Workspace registry 的限制](../../../../packages/workspace/workspace/README.zh.md#known-limitations-and-deferred-work)中。
+
+恢复界面受该页面能显示的内容限制：摘要未加载的已归档 id 没有行也没有取消归档操作，尽管注册表方法与 Remote 都接受它；因此页面报告没有可恢复条目，而不是把归档集合说成空的。通过自动化恢复仍然可用，未来的清理界面可以处理剩余的遗留条目。
+
+## 测试
+
+`packages/workspace/workspace/tests/workspace.spec.ts` 固定注册表方法：移除后幸存的归档顺序保持不变、记账位置仍在、重复取消归档与从未归档的 id 既不重写介质也不发出变更、会话已不存在的条目无需持久化列表即可解析、幸存集合在重启后重新加载。`packages/api/workspace-controller/tests/workspace-controller.host.spec.ts` 固定该动词幂等的完整集合应答与不变的 `archived` follow 增量，`packages/api/workspace-controller/tests/model.client.spec.ts` 固定 Client model 的回声及其拒绝安装失败应答的行为，以及四种过期应答竞态：归档或取消归档请求重叠且乱序返回，以及应答在途时到达的 follow 增量或 baseline。`packages/client/ui-settings-unarchive-sessions/tests/components.client.spec.tsx` 固定由新到旧的排序、未分组标签、会话已不存在条目的隐藏、读取与空状态、搜索，以及被拒绝时的 console 诊断，`tests/browser-plugin.client.spec.tsx` 则固定分区注册。
+
+## 相关
+
+- [会话归档（注册表全局集合）](../../archived/feature/2026-07-31-session-archive-global-set.md)——归档集合、follow 增量与被预测的恢复界面的冻结记录。

.agents/notes/implemented/feature/2026-09-13-mcp-resources-in-profiles.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 .agents/notes/implemented/feature/2026-09-13-mcp-resources-in-profiles.md
+2026-09-13-mcp-resources-in-profiles.md: 236cd2849383e32fee523c56a43aae2fcae1f07a
+2026-09-13-mcp-resources-in-profiles.zh.md: d2e0fbc45cdffe16ac482173d75c13329b98b868

.agents/notes/implemented/feature/2026-09-13-mcp-resources-in-profiles.md

@@ -0,0 +1,37 @@
+# Agent Note: MCP resource availability follows configured servers
+
+Status: implemented
+
+English | [中文](2026-09-13-mcp-resources-in-profiles.zh.md)
+
+## Problem
+
+Sessions without configured MCP servers need neither resource schemas nor MCP guidance. Requiring a separate resource-service entry also makes users configure shared resource access in addition to each connection. A shared tool set owned by the first server can disappear when that server unloads even though another server still needs it.
+
+## Decision
+
+Every shipped profile mounts `mcp-resources` once: base-backed profiles, including Desktop, inherit its row from `dsh-base`; standalone `sdk-minimal` owns its row. Users configure only their `mcp-client` entries. No MCP server is enabled by default.
+
+The resource service uses configured provider registrations in the caller's scope, including MCP clients mounted by another provider. An empty visible registry contributes no resource prompt, native tool schemas, PTC declarations, or PTC bindings. The first provider in a scope enables its shared tools; removal of the last removes those local registrations while preserving inherited providers and tools. The resource service owns the shared tool effects independently of any server plugin.
+
+Connection health does not determine this visibility. An active client remains configured through failed requests and reconnect attempts; its shared resource tools and server-name guidance stay available, and calls report connection failures. Server instructions retain their connection-owned publication rules.
+
+This decision partially supersedes the separate opt-in mount in the [resource and instruction decision](2026-09-12-mcp-resources-and-instructions.md). That note retains the operation, scope-selection, canonical-result, binary-rendering, and logged-instruction rationale.
+
+## Alternatives considered
+
+**Keep the separate resource mount.** It makes a shared capability a second user configuration task and permits different defaults across shipped profiles.
+
+**Keep resource tools visible without servers.** It adds unusable operations and prompt tokens to ordinary sessions, including the minimal SDK's single-shell default.
+
+**Filter providers by negotiated resource capability.** This omits resource guidance for servers that declare no resources, but visibility then requires a successful capability exchange. The configured-client policy uses one criterion for direct and provider-mounted clients, including before the first successful connection and during recovery.
+
+**Own shared tools under the first server plugin.** Disposing that server can remove tools still needed by another configured server. The service owns their lifetime instead.
+
+## Verification
+
+[Resource tests](../../../../packages/mcp/mcp-resources/tests/resources.spec.ts) cover empty native and PTC views, scoped inheritance, first/last-provider transitions, disposal, and failing configured providers. [Real SDK tests](../../../../packages/mcp/mcp-client/tests/protocol.spec.ts) pin empty discovery and unsupported read errors for a tools-only server. [Profile composition tests](../../../../apps/cli/tests/profile-mcp.spec.ts) resolve every shipped CLI template; [Desktop composition tests](../../../../apps/desktop/tests/profile-mcp.spec.ts) include its profile and Host overlay. The empty native and PTC recorded Sessions exercise the shipped headless composition without adding a resource entry.
+
+## Consequences
+
+An empty MCP configuration adds no MCP prompt or tool tokens. A configured server without resource capability still contributes its name and shared resource schemas. The SDK returns empty discovery lists; unsupported reads fail. Adding the first visible server or removing the last changes subsequent prompt and tool assembly. The minimal SDK advertises its single shell until the user adds MCP servers. Resource reads remain on demand, and no connection or durable Session format changes are required.

.agents/notes/implemented/feature/2026-09-13-mcp-resources-in-profiles.zh.md

@@ -0,0 +1,37 @@
+# Agent Note: MCP 资源可用性取决于已配置服务器
+
+Status: implemented
+
+[English](2026-09-13-mcp-resources-in-profiles.md) | 中文
+
+## 问题
+
+未配置 MCP 服务器的会话不需要资源 schema 或 MCP 指引。要求单独的资源服务条目，也让用户在配置每个连接之外还要配置共享资源访问。由首台服务器拥有的共享工具集可能随该服务器卸载而消失，即使其他服务器仍需要它。
+
+## 决策
+
+每个随附 profile 统一挂载 `mcp-resources` 一次：包括 Desktop 在内的基于 base 的 profile 从 `dsh-base` 继承该行；独立的 `sdk-minimal` 拥有自己的行。用户只需配置 `mcp-client` 条目。默认不启用任何 MCP 服务器。
+
+资源服务使用调用方作用域中的已配置提供方注册，包括由其他提供方挂载的 MCP 客户端。可见注册表为空时，不贡献资源提示词、native 工具 schema、PTC 声明或 PTC 绑定。作用域中的首个提供方启用共享工具；移除最后一个提供方时移除这些本地注册，同时保留继承的提供方与工具。资源服务独立于任何服务器插件拥有共享工具 effect。
+
+连接健康状态不决定这些内容的可见性。激活的客户端在请求失败或重连尝试期间仍属于已配置状态；共享资源工具与服务器名称指引保持可用，调用会报告连接失败。服务器指令保留由连接拥有的发布规则。
+
+本决策部分取代[资源与指令决策](2026-09-12-mcp-resources-and-instructions.zh.md)中单独启用资源服务的挂载方式。该笔记保留操作、作用域选择、规范结果、二进制渲染及已记录指令的设计依据。
+
+## 考虑过的替代方案
+
+**保留单独的资源挂载。** 这会让共享能力成为第二项用户配置任务，并允许随附 profile 使用不同默认值。
+
+**没有服务器时仍显示资源工具。** 这会向普通会话增加不可用操作与提示词 token，包括默认仅有一个 shell 的极简 SDK。
+
+**按协商出的资源能力过滤提供方。** 这能省略未声明资源的服务器对应的资源指引，但会让可见性取决于成功的能力交换。已配置客户端策略对直接配置和由提供方挂载的客户端使用同一标准，包括首次成功连接前与恢复期间。
+
+**由首台服务器插件拥有共享工具。** 释放该服务器可能移除其他已配置服务器仍需要的工具，因此由服务拥有这些工具的生命周期。
+
+## 验证
+
+[资源测试](../../../../packages/mcp/mcp-resources/tests/resources.spec.ts)覆盖空 native 与 PTC 视图、作用域继承、首个与最后一个提供方的变化、释放，以及已配置提供方调用失败。[真实 SDK 测试](../../../../packages/mcp/mcp-client/tests/protocol.spec.ts)固定了只提供工具的服务器返回空发现结果，以及不受支持的读取报错。[Profile 组合测试](../../../../apps/cli/tests/profile-mcp.spec.ts)解析每个随附 CLI 模板；[Desktop 组合测试](../../../../apps/desktop/tests/profile-mcp.spec.ts)包含其 profile 与 Host overlay。空 native 与 PTC 录制会话使用随附 headless 组合，不添加资源条目。
+
+## 后果
+
+空 MCP 配置不增加 MCP 提示词或工具 token。不具备资源能力的已配置服务器仍贡献名称与共享资源 schema。SDK 返回空的发现列表；不受支持的读取会失败。添加首个可见服务器或移除最后一个服务器，会改变后续提示词与工具组装。极简 SDK 在用户添加 MCP 服务器前只公布一个 shell。资源读取仍按需进行，无需更改连接或持久 Session 格式。

.agents/notes/implemented/process/2026-07-20-gui-testing-system.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 .agents/notes/implemented/process/2026-07-20-gui-testing-system.md
-2026-07-20-gui-testing-system.md: fa38b6e57fc0ff48f2604b000f6b9131897f2e4c
-2026-07-20-gui-testing-system.zh.md: 06ff9b81aad0d13abf51b2db093670d5e0ca57fb
+2026-07-20-gui-testing-system.md: 3fdb9a596dd18861ea8b765da129656942778f96
+2026-07-20-gui-testing-system.zh.md: e49d5a2f848e7f4847c7761ee0a487ce6fd3220e

.agents/notes/implemented/process/2026-07-20-gui-testing-system.md

@@ -20,12 +20,12 @@ Cut along the architecture's natural test hooks into three tiers, bottom-up:
 |---|---|---|---|
 | 1 Protocol isomorphism | Generated Typert Remote descriptors + `ApiGateway` + the Connection RPC carrier (arguments / results / errors / streams / cancellation) | **The full chain at the isomorphic point**: gateway host/client suites validate descriptor codecs and Remote dispatch in process; Connection host suites exercise the same `/api` carrier framing and trust checks without a browser | `packages/api/gateway/tests/`, `packages/client/connection/tests/` |
 | 2 Object-layer orchestration | `Session`/`SessionManager`/`ConnectionController` (state machines and timing: stitching / dedup / paging / optimistic draft clearing / pendingBuffers / reconnect / backoff) | **The "event sequence in → snapshot out" golden path**: programmable fakes + deferreds controlling timing + fake timers controlling backoff | `packages/client/{runtime,connection}/tests/` |
-| 3 Assembled presentation | Built artifacts × the real client loader and plugin composition | App-owned semantic snapshots boot all eight built client plugins under jsdom for deterministic cross-plugin state changes; bare Playwright smoke separately proves the real browser/carrier boundary, with real-host cases self-skipping without a key; the keyless browser e2e lane disables the shipped model-adapter row and replays recorded session fixtures through `dsh-llm-replay` in the real in-process web assembly against conversation aria goldens ([web e2e lane](../testing/2026-07-24-web-gui-browser-e2e-lane.md), [required CI gate](../testing/2026-07-30-web-browser-snapshot-ci-gate.md)) | `apps/web/tests/*.snapshot.ts`, `apps/web/tests/smoke-{fixture,real}.e2e.ts`, `apps/web/tests/{replay-round-trip,seeded-history}.e2e.ts` |
+| 3 Assembled presentation | Built artifacts × the real client loader and plugin composition | App-owned semantic snapshots boot the built client graph under jsdom with a test-owned `RemoteMock`; Playwright cases separately exercise the real browser and Host carriers, with recorded model sessions replayed through `dsh-llm-replay` ([whole-client tier](../testing/2026-09-06-client-assembly-test-line.md), [web e2e lane](../testing/2026-07-24-web-gui-browser-e2e-lane.md)) | `apps/web/tests/*.expected.e2e.ts`, `apps/web/tests/*.e2e.ts`, `apps/web/tests/*.snapshot.ts` |
 
 Inter-tier discipline: **each tier tests its own layer, upper tiers never re-test lower ones** — an app semantic snapshot pins only user-visible projection across the assembled plugin boundary, while Playwright smoke proves browser and carrier liveness; wire semantics belong to tier 1 and data semantics to tier 2. Pure-function layers (lineage/partial/notifier/transcript-adapter) are tested directly with zero fakes in the same package's tests/ alongside tier 2.
 
 - **Host and client source** are under the repo-wide per-file 100% coverage gate except the narrow browser-grade exclusions annotated in `vitest.config.ts`; component suites use per-file jsdom pragmas and Testing Library without changing Node suites.
-- **App-owned semantic snapshots** read built client bundles, execute them through the real loader, and drive only deterministic fixture hooks. They own stable visible state such as sidebar labels, breadcrumbs, and `document.title`, not CSS pixels or lower-layer state-machine details.
+- **App-owned semantic snapshots** read built client bundles, execute them through the real loader, and drive deterministic RemoteMock scenarios. They own stable visible state such as sidebar labels, breadcrumbs, and `document.title`, not CSS pixels or lower-layer state-machine details.
 
 ## Lane map
 
@@ -33,7 +33,7 @@ Inter-tier discipline: **each tier tests its own layer, upper tiers never re-tes
 |---|---|---|---|
 | Baseline | `pnpm run test:gui` | Tier 1+2 vitest (`packages/client packages/host`), seconds-fast, no browser, no server | Casually, after touching any GUI source |
 | Semantic snapshot | `DSH_EXAMPLE_MODE=lib pnpm run test:snapshot` | Keyless assembled-application semantics plus the repo's transport-specific expected outputs | After a human-visible GUI change; before delivery |
-| Browser end-to-end | `pnpm run test:web` | Rebuilds the front-end dist first, then runs the tier-3 browser set: the two-level smoke (fixture level + real-host level self-skip) plus the keyless replayed e2e scenarios (`DSH_SNAPSHOT=record`/`refresh` re-record fixtures / rewrite goldens) | After touching the build surface/boot/carriage; before delivery |
+| Browser end-to-end | `pnpm run test:web` | Rebuilds the front-end dist first, then runs built-client RemoteMock cases and real-Host browser scenarios, including keyless recorded-session replay (`DSH_SNAPSHOT=record`/`refresh` re-record fixtures / rewrite goldens) | After touching the build surface/boot/carriage; before delivery |
 | Browser expected-output gate | `DSH_SNAPSHOT=replay pnpm run test:web:built` | Reuses CI-built artifacts and compares every committed browser golden without writing | Every Linux pull request |
 | Gate | `pnpm run test:coverage` | The repo-wide gate (host and client GUI packages included, except annotated browser-grade exclusions) | The PR window |
 
@@ -42,7 +42,7 @@ Inter-tier discipline: **each tier tests its own layer, upper tiers never re-tes
 ## Anti-regression discipline
 
 - **Every bug fix pins an assertion**: a browser-visible bug is pinned into its owning browser spec (smoke or e2e scenario); a data-layer bug is pinned into the matching spec (precedent: the res-close misjudgment pinned in the webserver bridge suite — pure Node, reproduces in seconds, no longer needs the 12s browser sentinel as the only defense).
-- **All-green on fixture is not done, the real wire must pass too**: what the fixture short-circuits is exactly the wire carriage chain (node:http bridge close semantics, real network timing); both empirically confirmed bugs hid there. Changes touching connection/bridge/handler/SSE must run the browser lane (`pnpm run test:web`) — its keyless e2e scenarios drive the real HTTP/SSE carriage, and the with-key real-host smoke remains the live-model complement.
+- **All-green on RemoteMock is not done, the real wire must pass too**: a decoded in-process carrier deliberately omits the HTTP/WebSocket chain and its network timing. Changes touching connection, bridge, handler, or streaming transport must run the browser lane (`pnpm run test:web`), whose keyless e2e scenarios drive the real carrier; the with-key real-Host smoke remains the live-model complement.
 - The code-on-disk-is-the-answer reconciliation workflow: when a behavior change lands and turns existing cases red, reconcile on the spot (fix the test or fix the code, with the RFC/contract as arbiter); no red left hanging.
 
 ## Consequences
@@ -55,6 +55,6 @@ Each lane tests its own tier: touching any GUI source gets seconds-fast `test:gu
 |---|---|
 | Single e2e (everything through the browser) | Browser startup is seconds × N slower and timing is uncontrollable; wire/object-layer invariants can be fully asserted in milliseconds in node env |
 | Migrating the verify scripts to vitest | An ordered script shares one browser session; splitting the cases either formalizes it (sequential + shared page) or re-runs the preamble × N; streaming PASS/FAIL output is exactly the agent's locating interface |
-| Reusing FixtureApiClient in tests | The demo script runs on a real clock, tests need deferred hand-controlled timing — orthogonal purposes; forced reuse chains the tests to the demo's rhythm |
+| Keeping a production client fixture for tests | It couples shipped code to scenario data and a query-selected transport; RemoteMock and real-Host tests own the two required tiers directly |
 | A standalone vitest config for GUI packages (once designed as vitest.gui.config.ts) | Package-level tests/ are already scanned by the root include; `vitest run packages/client packages/host` path filtering is the tight loop — zero new config |
 | Deferring hooks/component-layer unit tests | jsdom remains the coverage mainline because it gives fast per-file component behavior; the required browser replay gate complements it at the assembled tier rather than replacing it ([CI gate decision](../testing/2026-07-30-web-browser-snapshot-ci-gate.md)) |

.agents/notes/implemented/process/2026-07-20-gui-testing-system.zh.md

@@ -20,12 +20,12 @@ GUI 栈需要考虑多种应用形态，同应用形态内的不同运行环境
 |---|---|---|---|
 | 1 协议同构层 | 生成的 Typert Remote 描述符 + `ApiGateway` + Connection RPC 承载（参数/结果/错误/流/取消） | **同构点全链**：gateway host/client 套件在进程内验证描述符 codec 与 Remote 分派；Connection host 套件不经浏览器即可运行相同的 `/api` 承载帧与信任检查 | `packages/api/gateway/tests/`、`packages/client/connection/tests/` |
 | 2 对象层编排 | `Session`/`SessionManager`/`ConnectionController`（状态机与时序：缝合/去重/翻页/乐观清稿/pendingBuffers/重连/退避） | **「事件序列进→快照出」黄金路径**：可编程假体 + deferred 控时序 + fake timers 控退避 | `packages/client/{runtime,connection}/tests/` |
-| 3 组装呈现层 | 构建产物 × 真实 client loader 与插件组合 | 归应用所有的语义快照会在 jsdom 下启动全部 8 个已构建的 client 插件，以确定性方式驱动跨插件状态变化；另有最简 Playwright 冒烟测试负责验证真实浏览器/承载层边界，真 host 用例在无密钥时自行跳过；无密钥浏览器 e2e 车道会禁用交付配置中的模型适配器行，并通过 `dsh-llm-replay` 在真实进程内 web 组装中回放录制的会话 fixture（测试前置数据），与会话区 aria 预期输出比对（[web e2e 车道](../testing/2026-07-24-web-gui-browser-e2e-lane.zh.md)、[必需 CI 门禁](../testing/2026-07-30-web-browser-snapshot-ci-gate.zh.md)） | `apps/web/tests/*.snapshot.ts`、`apps/web/tests/smoke-{fixture,real}.e2e.ts`、`apps/web/tests/{replay-round-trip,seeded-history}.e2e.ts` |
+| 3 组装呈现层 | 构建产物 × 真实 client loader 与插件组合 | 归应用所有的语义快照在 jsdom 下用测试持有的 `RemoteMock` 启动构建后 Client 图；Playwright 用例分别验证真实浏览器与 Host 载体，并通过 `dsh-llm-replay` 回放已录制的模型会话（[整机客户端测试档](../testing/2026-09-06-client-assembly-test-line.zh.md)、[web e2e 车道](../testing/2026-07-24-web-gui-browser-e2e-lane.zh.md)） | `apps/web/tests/*.expected.e2e.ts`、`apps/web/tests/*.e2e.ts`、`apps/web/tests/*.snapshot.ts` |
 
 层间纪律：**各层各测各的，上层不重测下层**：应用语义快照只固定组装后插件边界上的用户可见投影，Playwright 冒烟测试负责验证浏览器与承载层是否存活；wire 语义归 1 层，数据语义归 2 层。纯函数层（lineage/partial/notifier/transcript-adapter）随 2 层同包 tests/ 零假体直测。
 
 - **host 与 client 源码**均纳入全仓 per-file 100% 覆盖率门禁，仅排除 `vitest.config.ts` 中带注释的少量浏览器级例外；组件套件通过逐文件 jsdom pragma 和 Testing Library 运行，不会改变 Node 套件。
-- **归应用所有的语义快照**读取已构建的 client bundle，通过真实 loader 执行它们，并且只驱动确定性的 fixture 钩子。它们负责固定侧边栏标签、面包屑和 `document.title` 等稳定可见状态，而不固定 CSS 像素或下层状态机细节。
+- **归应用所有的语义快照**读取已构建的 client bundle，通过真实 loader 执行它们，并驱动确定性的 RemoteMock 场景。它们负责固定侧边栏标签、面包屑和 `document.title` 等稳定可见状态，而不固定 CSS 像素或下层状态机细节。
 
 ## 车道地图
 
@@ -33,7 +33,7 @@ GUI 栈需要考虑多种应用形态，同应用形态内的不同运行环境
 |---|---|---|---|
 | 基础 | `pnpm run test:gui` | 1+2 层 vitest（`packages/client packages/host`），秒级、无浏览器、无 server | 改 GUI 任意源码后随手跑 |
 | 语义快照 | `DSH_EXAMPLE_MODE=lib pnpm run test:snapshot` | 无需密钥的组装应用语义，以及仓库按传输形态划分的预期输出 | 用户可见的 GUI 变更后；交付前 |
-| 浏览器端到端 | `pnpm run test:web` | 先重建前端 dist，再跑 3 层浏览器全集：双级冒烟测试（fixture 级 + 真 host 级 self-skip）加上无密钥回放 e2e 场景（`DSH_SNAPSHOT=record`/`refresh` 重录 fixture / 重写期望输出） | 改构建面/boot/承载后；交付前 |
+| 浏览器端到端 | `pnpm run test:web` | 先重建前端 dist，再运行 built-client RemoteMock 用例与真实 Host 浏览器场景，其中包括无密钥的录制会话回放（`DSH_SNAPSHOT=record`/`refresh` 重录 fixture／重写预期输出） | 改构建面/boot/承载后；交付前 |
 | 浏览器预期输出门禁 | `DSH_SNAPSHOT=replay pnpm run test:web:built` | 复用 CI 构建的产物，并在不写入的情况下比较每份已提交的浏览器预期输出 | 每个 Linux 拉取请求 |
 | 门禁 | `pnpm run test:coverage` | 全仓门禁（host 与 client GUI 包均纳入，仅排除带注释的浏览器级例外） | PR（Pull Request）窗口 |
 
@@ -42,7 +42,7 @@ GUI 栈需要考虑多种应用形态，同应用形态内的不同运行环境
 ## 防回归纪律
 
 - **修一个 bug 钉一条断言**：浏览器可见的 bug 钉进所属浏览器 spec（冒烟测试或 e2e 场景）；数据层 bug 钉进对应 spec（先例：res-close 误判钉在 webserver 桥 suite——纯 Node 秒级复现，不再需要 12s 浏览器哨兵作唯一防线）。
-- **fixture 全绿不算完，真 wire 也要过**：fixture 短路的恰是 wire 承载链（node:http 桥 close 语义、真网络时序），两次实证 bug 都藏在那里。改动触及连接/桥/handler/SSE 的，浏览器车道（`pnpm run test:web`）必跑——其无密钥 e2e 场景驱动真实 HTTP/SSE 承载，带密钥的真 host 冒烟测试仍是真模型侧的补充。
+- **RemoteMock 全绿不算完，真 wire 也要过**：已解码的进程内载体会刻意绕过 HTTP/WebSocket 链及其网络时序。改动触及 connection、bridge、handler 或流式 transport 时必须运行浏览器车道（`pnpm run test:web`），由其中的无密钥 e2e 场景驱动真实载体；带密钥的真实 Host 冒烟仍是真模型侧的补充。
 - 落盘代码即答案的对表工作流：行为改动落盘打红既有用例时，当场对表校准（改测试还是改代码以 RFC/约定为裁），不留悬红。
 
 ## Consequences
@@ -55,6 +55,6 @@ GUI 栈需要考虑多种应用形态，同应用形态内的不同运行环境
 |---|---|
 | 单一 e2e（全走浏览器） | 浏览器起步秒级×N 倍慢+时序不可控；wire/对象层不变量在 node env 可毫秒级全断言 |
 | verify 脚本迁 vitest | 有序脚本共享浏览器会话，拆 case 要么形式化（sequential+共享 page）要么重走前置×N；PASS/FAIL 流式输出正是 agent（智能体）定位接口 |
-| 测试复用 FixtureApiClient | 演示脚本走真实时钟，测试需要 deferred 手控时序——用途正交，硬复用把测试绑死在演示节奏上 |
+| 为测试保留生产 Client fixture | 它把交付代码耦合到场景数据和 query 选择的 transport；RemoteMock 与真实 Host 测试分别直接持有两个所需层级 |
 | GUI 包独立 vitest config（曾设计 vitest.gui.config.ts） | 包级 tests/ 本就被根 include 扫到，`vitest run packages/client packages/host` 路径过滤即窄循环——零新 config |
 | 钩子/组件层暂缓单测 | jsdom 仍是覆盖率主线，因为它能快速验证逐文件组件行为；必需的浏览器回放门禁在组装层与之互补，而非取代它（[CI 门禁决策](../testing/2026-07-30-web-browser-snapshot-ci-gate.zh.md)） |