Merge remote-tracking branch 'origin/master' into worktree/jobs-api-review-3052

# Conflicts:
#	apps/web/tsconfig.json
#	docs/event-producer-consumer.i18n.yaml
#	docs/event-producer-consumer.md
#	docs/event-producer-consumer.zh.md
#	docs/module-graph.i18n.yaml
#	docs/module-graph.md
#	docs/module-graph.zh.md
#	packages/api/remotes/package.json
#	packages/api/remotes/src/client/index.ts
#	packages/api/session-controller/tests/client-apply.client.spec.ts
#	packages/api/session-controller/tests/control-jobs.host.spec.ts
#	packages/api/session-controller/tests/fake-api.client.ts
#	packages/api/session-controller/tests/manager.client.spec.ts
#	packages/client/ui-jobs/package.json
#	packages/extensions/cordis-client-runner/src/client/slot-catalog.ts
#	packages/jobs/jobs-local/README.i18n.yaml
#	packages/jobs/jobs-local/README.md
#	packages/jobs/jobs-local/README.zh.md
#	packages/jobs/jobs-local/src/index.ts
#	packages/jobs/jobs-local/tests/jobs.spec.ts
#	packages/jobs/tool-jobs/tests/tool-jobs.spec.ts
#	packages/shell/shell/README.i18n.yaml
#	packages/shell/shell/README.md
#	packages/shell/shell/README.zh.md
#	packages/shell/tool-bash/README.i18n.yaml
#	packages/shell/tool-bash/README.md
#	packages/shell/tool-bash/README.zh.md
#	packages/shell/tool-bash/src/background.ts
#	packages/shell/tool-bash/src/index.ts
#	packages/shell/tool-pwsh/src/background.ts
#	packages/shell/tool-pwsh/src/index.ts
This commit is contained in:
Yichen Jiang committed 2026-09-16 15:53:02 +08:00
commit 1596097147
3429 files changed
+775482 -44018

No files matched your search

@@ -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
@@ -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.
@@ -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。
+3
View File
@@ -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",
@@ -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/2026-06-11-dev-invariants-over-deep-readonly.md
2026-06-11-dev-invariants-over-deep-readonly.md: 9adb741db1e26e3a9f750fed60524351159cfeaa
2026-06-11-dev-invariants-over-deep-readonly.zh.md: d352eca985b6594ec5d10c9b73c2ecc18ea458b5
2026-06-11-dev-invariants-over-deep-readonly.md: 0784960397f4d198e136e284ef9232b4ed3e2c2c
2026-06-11-dev-invariants-over-deep-readonly.zh.md: 7183813130560bf82426c8719570b668ba86b7e5
@@ -22,7 +22,7 @@ Responsibility is split between an always-on storage boundary and optional devel
`Session` accepts an event only after one recursive pass has materialized a lossless JSON snapshot. That pass rejects unsupported values and produces the exact detached record that enters the log, so validation and storage cannot observe different values from a stateful getter or retain caller-owned nested references.
The accepted event and all of its descendants are deep-frozen before publication. `append()` returns that owned frozen event, and `session/event` observers and `eventAt(seq)` receive the same record. `snapshotEvents(fromSeq?, toSeqExclusive?)` returns a frozen array snapshot; a previously returned array does not grow after a later append. `seq` and `eventAt()` avoid array materialization when a caller needs only the current length or one event. Seed records pass through the same validation, snapshot, and freeze boundary before construction succeeds.
The accepted event and all of its descendants are deep-frozen before publication. `append()` returns that owned frozen event, and `session/event` observers and `eventAt(seq)` receive the same record. `snapshotEvents(fromSeq?, toSeqExclusive?)` returns a frozen array snapshot; a previously returned array does not grow after a later append. `seq` reads the current length without materializing an array. Synchronous historical readers are deprecated under the [event-read policy](2026-09-09-deprecate-synchronous-session-event-reads.md). Seed records pass through the same validation, snapshot, and freeze boundary before construction succeeds.
This guarantee belongs in `Session`, not in an optional listener, because every composition relies on trustworthy history. A production deployment, a focused test, or a custom embedding receives the same storage semantics whether or not development support plugins are registered.
@@ -53,7 +53,7 @@ Detaching `deriveMessages()` would protect the most common request path but leav
## Consequences
- Every accepted live or seeded session event is detached from caller-owned inputs and deeply immutable before any observer can receive it.
- `snapshotEvents()` exposes stable immutable snapshots instead of the private growing array; `seq` and `eventAt()` serve scalar reads without copying that array.
- Existing `snapshotEvents()` and `eventAt()` callers retain immutable read results while their migration is deferred; `seq` reads the log length without copying the array.
- Request-side mutation cannot reach stored history through derived messages.
- Development builds can enable relational assertions without changing storage behavior, and disposing or filtering a companion does not weaken log immutability.
- `dsh-invariants` configures global enablement plus package allow/block regex lists; each check remains owned and tested by its product package.
@@ -22,7 +22,7 @@ TypeScript readonly 类型不是充分的运行时边界。它们在程序运行
`Session` 仅在一次递归遍历完成无损 JSON 快照的物化之后才接受事件。该遍历拒绝不支持的值,并产出进入日志的已分离的确切记录,因此验证与存储不会从有状态的 getter 观察到不同的值,也不会保留调用方拥有的嵌套引用。
被接受的事件及其所有后代在发布前被深度冻结。`append()` 返回由 Session 拥有的冻结事件,`session/event` 观察者和 `eventAt(seq)` 接收同一记录。`snapshotEvents(fromSeq?, toSeqExclusive?)` 返回冻结的数组快照;先前返回的数组不会因后续 append 而增长。调用方只需要当前长度或单个事件时,`seq` 和 `eventAt()` 不会物化数组。种子记录在构造成功前经过相同的验证、快照与冻结边界。
被接受的事件及其所有后代在发布前被深度冻结。`append()` 返回由 Session 拥有的冻结事件,`session/event` 观察者和 `eventAt(seq)` 接收同一记录。`snapshotEvents(fromSeq?, toSeqExclusive?)` 返回冻结的数组快照;先前返回的数组不会因后续 append 而增长。`seq` 无需物化数组即可读取当前长度。同步历史读取方法按[事件读取策略](2026-09-09-deprecate-synchronous-session-event-reads.zh.md)弃用。种子记录在构造成功前经过相同的验证、快照与冻结边界。
此保证属于 `Session` 而非可选监听器,因为每种组合都依赖可信的历史。无论是否注册了开发支持插件,生产部署、聚焦测试或自定义嵌入都获得相同的存储语义。
@@ -53,7 +53,7 @@ TypeScript readonly 类型不是充分的运行时边界。它们在程序运行
## 后果
- 每个被接受的实时或种子会话事件在任何观察者接收之前,都已从调用方拥有的输入中分离并深度不可变。
- `snapshotEvents()` 暴露稳定的不可变快照,而非持续增长的私有数组;`seq` 和 `eventAt()` 为标量读取提供无需复制数组的路径。
- 现有 `snapshotEvents()` 和 `eventAt()` 调用方在暂缓迁移期间仍获得不可变的读取结果;`seq` 无需复制数组即可读取日志长度。
- 请求侧的修改无法通过派生消息触及已存储的历史。
- 开发构建可以启用关系断言而不改变存储行为;dispose 或过滤一个配套插件不会削弱日志不可变性。
- `dsh-invariants` 配置全局启用状态以及包名允许/阻止 regex 列表;每项检查仍由其产品包拥有并测试。
@@ -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/2026-06-13-capability-seams.md
2026-06-13-capability-seams.md: 3c552c474b499b9f1c9f60242f4773750faadafb
2026-06-13-capability-seams.zh.md: 4e6100a2ce557d91fa18b5630c267f37f1bc0f00
2026-06-13-capability-seams.md: 798739b7ef203b10ea657f9e1c178ea7eab094ad
2026-06-13-capability-seams.zh.md: c8f568aa569fe74146f1e7fcc46e00774a39fb84
@@ -8,7 +8,7 @@ English | [中文](2026-06-13-capability-seams.zh.md)
The harness has swappable capabilities, including shell execution and model providers. A capability has three concerns that change at different rates and for different reasons: the *contract* (what the capability is), the *implementation* (how it runs), and the *consumer API* (what the model and other plugins program against). Bundling them in one package couples those rates of change — swapping a local executor for a sandboxed one would churn the tool schemas the model sees, even though the model-facing contract never changed.
This is distinct from "who provides vs. needs a capability at runtime", which Cordis already answers with services + `inject` (a provider registers `ctx.shell`; a consumer declares `inject: ['bash']` and its fiber pends until the service exists). That mechanism is necessary but doesn't dictate package boundaries; this Agent Note does.
This is distinct from "who provides vs. needs a capability at runtime", which Cordis already answers with services + `inject` (a provider registers `ctx.shell`; a consumer declares `inject: ['shell']` and its fiber pends until the service exists). That mechanism is necessary but doesn't dictate package boundaries; this Agent Note does.
## Decision
@@ -8,7 +8,7 @@ Status: implemented
harness 具有可替换的能力,包括 shell 执行和模型提供方。一项能力涉及三个关注点,它们以不同速率、因不同原因变化:*约定*(这项能力是什么)、*实现*(它如何运行)、*消费方 API*(模型和其他插件面向什么编程)。将三者捆绑在一个包中会耦合这些变化速率——把本地执行器换成沙箱化执行器时,模型看到的工具 schema 也会被搅动,尽管面向模型的约定从未改变。
这与「谁在运行时提供、谁需要一项能力」是不同的问题,后者 Cordis 已通过服务 + `inject` 解决(提供方注册 `ctx.shell`;消费方声明 `inject: ['bash']`,其 fiber 挂起直到服务存在)。该机制是必要的,但不决定包的边界;本 Agent Note 决定的是包的边界。
这与「谁在运行时提供、谁需要一项能力」是不同的问题,后者 Cordis 已通过服务 + `inject` 解决(提供方注册 `ctx.shell`;消费方声明 `inject: ['shell']`,其 fiber 挂起直到服务存在)。该机制是必要的,但不决定包的边界;本 Agent Note 决定的是包的边界。
## 决策
@@ -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/2026-06-13-twin-llm-adapters.md
2026-06-13-twin-llm-adapters.md: a4c87325a0b0d1ebe6cf8f95672e5de74ef37d57
2026-06-13-twin-llm-adapters.zh.md: 36996750a16cc95393d727cffee3bcc53573eaf2
2026-06-13-twin-llm-adapters.md: fe8b0b55e0e027e29eb0920e64a1760d0dc35aee
2026-06-13-twin-llm-adapters.zh.md: 4248b5afeb9f3fd4503914e53203c674e3a80dbf
@@ -25,3 +25,5 @@ The rule they enforce: **anything the StreamChunk vocabulary cannot express for
## Consequences
The twin doubles adapter and key-gated e2e maintenance—both cover V4 Flash and Pro across representative reasoning modes—in exchange for continuous seam-neutrality validation and a second implementation example. Both use `apiKey`, `baseURL`, and `models`; the direct-fetch adapter exposes `thinking`/`reasoningEffort`, while pi-ai exposes one `reasoning` level. A future conformance suite could justify retiring one adapter through a superseding Agent Note.
The [Messages adapter](../feature/2026-09-07-deepseek-messages-adapter.md) adds an Anthropic-protocol implementation inside `llm-deepseek`; it preserves the same stream conventions.
@@ -25,3 +25,5 @@ Status: implemented
## 后果
孪生体使适配器和需要密钥的 e2e 维护量翻倍——两者都覆盖 V4 Flash 和 Pro 在各代表性推理(reasoning)模式下的行为——换来的是持续的 seam 中立性验证和第二份实现示例。两个适配器均使用 `apiKey`、`baseURL` 和 `models`;直接 fetch 适配器暴露 `thinking`/`reasoningEffort`,pi-ai 适配器暴露一个 `reasoning` 级别。未来如果有一致性测试套件,可以通过后续 Agent Note 论证退役其中一个适配器。
[Messages 适配器](../feature/2026-09-07-deepseek-messages-adapter.zh.md) 在 `llm-deepseek` 内增加 Anthropic 协议实现,并遵守相同的流约定。
@@ -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/2026-06-18-session-surface.md
2026-06-18-session-surface.md: 93ea55883dedd943fe1ffac67a9842c962ca6dac
2026-06-18-session-surface.zh.md: 54ecb1162bc46007dfcbb7d8cb39075d52171567
2026-06-18-session-surface.md: 26780d54a1ea12283900572c6fd0e8f740de5524
2026-06-18-session-surface.zh.md: 328a9ee9f1bed9c52996cfd307549a7745e16d4b
@@ -37,7 +37,7 @@ Delta processing is O(1) when no new events and O(new events) when new events ar
### Persistence
The fields are serialized as top-level JSON properties. JSONL preserves placement and provenance without a separate column mapping. The [V3 canonical-envelope decision](2026-09-06-v3-canonical-session-envelopes.md) owns exact replacement keys and strict-acceptance rationale; the [V2-to-V3 specification](../../../../packages/session/session-format-v2-to-v3/README.md#canonical-envelopes) owns historical conversion. This note retains ordered-projection ownership and replacement rationale.
The fields are serialized as top-level JSON properties. JSONL preserves placement and source-event references without a separate column mapping. The [V3 canonical-envelope decision](2026-09-06-v3-canonical-session-envelopes.md) owns exact replacement keys and strict-acceptance rationale; the [V2-to-V3 specification](../../../../packages/session/session-format-v2-to-v3/README.md#canonical-envelopes) owns historical conversion. This note retains ordered-projection ownership and replacement rationale.
### Crash recovery
@@ -37,7 +37,7 @@ surface 元数据仅属于四种 surface 事件类型(`system/message`、`user
### 持久化
这些字段作为顶层 JSON 属性序列化。JSONL 无需单独列映射即可保留位置与来源。[V3 规范信封决策](2026-09-06-v3-canonical-session-envelopes.zh.md)负责精确替换键与严格准入依据;[V2 到 V3 规范](../../../../packages/session/session-format-v2-to-v3/README.zh.md#canonical-envelopes)负责历史转换。本文继续负责有序投影的所有权与替换依据。
这些字段作为顶层 JSON 属性序列化。JSONL 无需单独列映射即可保留位置与源事件引用。[V3 规范信封决策](2026-09-06-v3-canonical-session-envelopes.zh.md)负责精确替换键与严格准入依据;[V2 到 V3 规范](../../../../packages/session/session-format-v2-to-v3/README.zh.md#canonical-envelopes)负责历史转换。本文继续负责有序投影的所有权与替换依据。
### 崩溃恢复
@@ -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/2026-07-05-prompt-variables-and-tool-guidance-ownership.md
2026-07-05-prompt-variables-and-tool-guidance-ownership.md: 35bb7c6fabc85ae6f93bdbb67e13910eea627ca3
2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md: 53d69cf45c02f6534334561b626d2c2ae6087c05
2026-07-05-prompt-variables-and-tool-guidance-ownership.md: 8f5542cb09930e62fd1e26960166fbdd6d3f6745
2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md: 348ae36cf6549d412adab1d3cdb5cc03f0badd01
@@ -26,7 +26,7 @@ The assembled system prompt had four defects, all of one family: facts the harne
### Prompt variables
Plugins register `{{name}}` values through `ctx.systemPrompt.variable(name, provider)`. Assembly resolves them into the waterfall-visible variable map. Rendering rejects unknown own-property references, registered providers that return `undefined`, malformed complete references, and unbalanced references that still contain a closing `}}`; a lone unmatched `{{` remains prose, and substituted values are not rescanned. Registration rejects invalid or duplicate variable names, and section names are unique.
Plugins register `{{name}}` values through `ctx.systemPrompt.variable(name, provider)`. Assembly resolves them into the waterfall-visible variable map. Rendering rejects unknown own-property references, registered providers that return `undefined`, malformed complete references, and unbalanced references that still contain a closing `}}`; a lone unmatched `{{` remains prose, and substituted values are not rescanned. Registration rejects invalid or duplicate variable names, and section names are unique. Sections may set `interpolate: false` to preserve generated documents literally; `tools:sdk` does so because tool descriptions and schemas may document their own `{{…}}` syntax.
`dsh-agent-loop` registers the two built-ins, both pure projections of the context agent: `model` (= `options.model`) and `cwd` (= `session.header.cwd`). The example personas write `powered by the {{model}} model` — the model name is stated once, in the `model:` config key. `{{cwd}}` is demonstrated in the ACP example only: every ACP session carries the client's cwd, while config-pre-created stdio agents have none (a persona claiming `{{cwd}}` there fails the turn — by design). The variables stay on the loop plugin (unlike the sections below): they are runtime facts of the agents THIS loop drives, and a replacement loop supplies its own.
@@ -47,7 +47,7 @@ Per-tool semantics and selection guidance live in tool descriptions. Prompt sect
- **The loop composes an identity line itself** — hardcodes model-facing prose in the one package that must stay thin ("plugins, not loop changes"), and outside the section pipeline it would be a second composition path. (The identity DOES ship as a code literal — but as an ordinary section registered by `dsh-system-prompt`, whose `system-prompt/assemble` waterfall remains the escape valve for a deployment that must drop it.)
- **Inject the model name via the `agent/request` waterfall** — prompt text would be composed in two places and the earlier rendered persona could disagree with the final routed header. The request plugin that owns late routing must also own any earlier prompt claim about that model.
- **Hand-write the model name in each persona** — duplicates the `model:` key one line above and silently lies after a config edit; the exact disease this decision cures.
- **Lenient interpolation (leave unknown refs verbatim, or substitute empty)** — a typo ships `{{modle}}` (or a hole) to the model and nobody notices until transcript review.
- **Lenient interpolation (leave unknown refs verbatim, or substitute empty)** — a typo ships `{{modle}}` (or a hole) to the model and nobody notices until transcript review. Leaving only unknown names unchanged would still substitute registered names inside tool documentation.
- **Per-instance subagent wording in config** — returns model-facing prose to every deployment × instance, reviving the hand-written-guidance-in-leaf-YAML drift. **Keying wording off the provider NAME** — `providerName` is itself config, so a renamed provider silently gets the wrong words.
- **Resolving the provider at `apply` time (a load-order requirement)** and **section-only subagent wording (lazily resolved at assemble)** — the alternatives to the provider-lifecycle events; both rejected in [the provider-lifecycle-events Agent Note](../../archived/architecture/2026-07-05-subagent-provider-lifecycle-events.md).
@@ -60,7 +60,7 @@ Per-tool semantics and selection guidance live in tool descriptions. Prompt sect
- The tui-agent prompt renders identity, persona with the interpolated model, then fs/shell/web guidance through one assembly path.
- Fork and fresh subagent descriptions reflect whether the provider inherits completed conversation turns; the tool appears, disappears, and is reworded with provider lifecycle changes.
- Unknown, valueless, malformed, or unbalanced variable references name the section and throw; duplicate section, variable, and tool registrations also throw.
- In interpolated sections, unknown, valueless, malformed, or unbalanced variable references name the section and throw; duplicate section, variable, and tool registrations also throw.
- Snapshot replay is prompt-independent: it keys recorded chunk streams by turn and step without comparing the outgoing request.
## Consequences
@@ -69,4 +69,4 @@ Per-tool semantics and selection guidance live in tool descriptions. Prompt sect
- `{{model}}` reflects `AgentOptions.model` at assembly time. A plugin that switches models in the `agent/request` waterfall makes the prompt's claim stale for that step, and one that SUPPLIES the model there (options.model unset — the loop's documented fallback) leaves the variable valueless at render, failing a `{{model}}` persona before the waterfall runs. Both have the same remedy, and it is the ownership rule itself: the plugin that owns the late-bound model fact states it early on the `system-prompt/assemble` waterfall (`assembly.variables['model'] = …`) — one owner, both statements; a loop test pins the supply path end-to-end. Accepted.
- While a bound provider is absent (not yet activated, unloaded, mid-HMR-reload), the subagent tool does not exist and a model request in that window simply lacks it. That is the honest state — the alternative was a registered tool whose description or execution could not be trusted.
- Strictness means a persona can fail a turn at render (e.g. `{{cwd}}` on a cwd-less session). The failure is contained — the turn ends `error`, the loop survives — and it is an authoring error we WANT loud.
- No escape syntax for a literal `{{name}}` in prompt prose yet; add one if a real prompt ever needs it.
- Inline escapes remain unsupported in interpolated text; literal sections need no escaping. PTC unit tests cover both modes and runtime languages, and the recorded `ptc-turn` scenario preserves tool-template examples in the model-visible prompt.
@@ -26,7 +26,7 @@ Status: implemented
### 提示词变量
插件通过 `ctx.systemPrompt.variable(name, provider)` 注册 `{{name}}` 值。组装过程将它们解析到 waterfall 可见的变量映射中。渲染阶段拒绝以下情况:引用未知的自有属性、已注册的提供方返回 `undefined`、格式错误的完整引用、以及仍包含闭合 `}}` 的不平衡引用;孤立的未匹配 `{{` 保留为行文,替换后的值不会被重新扫描。注册阶段拒绝无效或重复的变量名,section 名称也必须唯一。
插件通过 `ctx.systemPrompt.variable(name, provider)` 注册 `{{name}}` 值。组装过程将它们解析到 waterfall 可见的变量映射中。渲染阶段拒绝以下情况:引用未知的自有属性、已注册的提供方返回 `undefined`、格式错误的完整引用、以及仍包含闭合 `}}` 的不平衡引用;孤立的未匹配 `{{` 保留为行文,替换后的值不会被重新扫描。注册阶段拒绝无效或重复的变量名,section 名称也必须唯一。段可设置 `interpolate: false` 来原样保留生成的文档;`tools:sdk` 使用此设置,因为工具描述和 schema 可能会介绍自身的 `{{…}}` 语法。
`dsh-agent-loop` 注册两个内置变量,均为上下文 agent 的纯投影:`model`(= `options.model`)和 `cwd`(= `session.header.cwd`)。示例 persona 写 `powered by the {{model}} model`——模型名称只在 `model:` 配置键中声明一次。`{{cwd}}` 仅在 ACP 示例中演示:每个 ACP 会话携带客户端的 cwd,而配置预创建的 stdio agent 没有 cwd(在那里声称 `{{cwd}}` 的 persona 会导致该轮次失败——这是有意为之)。变量留在 loop 插件上(不同于下面的 section):它们是本循环驱动的 agent 的运行时事实,替换循环自行提供自己的变量。
@@ -47,7 +47,7 @@ Status: implemented
- **循环自行组合一行 identity 文本**:在必须保持精简的那个包(「用插件,不改循环」)中硬编码面向模型的行文,且在 section 流水线之外构成第二条组合路径。(identity 确实以代码字面量交付——但作为 `dsh-system-prompt` 注册的普通 section,其 `system-prompt/assemble` waterfall 仍是部署需要移除它时的逃生阀。)
- **通过 `agent/request` waterfall 注入模型名称**:提示词文本会在两处组合,更早渲染的 persona 也可能与最终已路由 header 不一致。拥有延迟路由的请求插件还必须拥有该模型在提示词中更早出现的声明。
- **在每个 persona 中手写模型名称**:与上方一行的 `model:` 键重复,配置修改后静默失实;正是本决策要治愈的病症。
- **宽松插值(未知引用保留原样或替换为空)**:一个拼写错误 `{{modle}}`(或一个空洞)会被发送给模型,直到 transcript(文本记录)审查时才会被发现。
- **宽松插值(未知引用保留原样或替换为空)**:一个拼写错误 `{{modle}}`(或一个空洞)会被发送给模型,直到 transcript(文本记录)审查时才会被发现。仅保留未知名称仍会替换工具文档中的已注册名称。
- **在配置中为每个 subagent 实例编写措辞**:面向模型的行文回到每个部署 × 实例中,重蹈在 leaf YAML 中手写指导的漂移。**根据提供方名称选择措辞**:`providerName` 本身是配置,重命名提供方后会静默获得错误的措辞。
- **在 `apply` 时解析提供方(加载顺序要求)**与**仅用 section 承载 subagent 措辞(在 assemble 时惰性解析)**:提供方生命周期事件的替代方案;两者均在[提供方生命周期事件 Agent Note](../../archived/architecture/2026-07-05-subagent-provider-lifecycle-events.md)中被否决。
@@ -60,7 +60,7 @@ Status: implemented
- tui-agent 的提示词通过一条组装路径依次渲染 identity、带插值模型名的 persona,然后是 fs/shell/web 指导。
- fork 和 fresh subagent 的描述反映提供方是否继承已完成的对话轮次;工具随提供方生命周期变化而出现、消失和重新措辞。
- 未知、无值、格式错误或不平衡的变量引用会指明 section 名称并抛出异常;重复的 section、变量和工具注册同样抛出异常。
- 在启用插值的段中,未知、无值、格式错误或不平衡的变量引用会指明 section 名称并抛出异常;重复的 section、变量和工具注册同样抛出异常。
- 快照回放与提示词无关:它按轮次和步骤索引已记录的分片流,不比较发出的请求。
## 后果
@@ -69,4 +69,4 @@ Status: implemented
- `{{model}}` 在组装时反映 `AgentOptions.model`。如果一个插件在 `agent/request` waterfall 中切换模型,提示词对该步骤的声明就会过时;如果一个插件在那里提供模型(options.model 未设置——循环文档中记载的回退路径),变量在渲染时无值,包含 `{{model}}` 的 persona 会在 waterfall 运行前失败。两者的补救方式相同,就是归属规则本身:拥有延迟绑定模型事实的插件在 `system-prompt/assemble` waterfall 上提前声明它(`assembly.variables['model'] = …`)——一个归属方,两处声明;一个循环测试端到端固定了 supply 路径。已接受。
- 当一个已绑定的提供方不存在时(尚未激活、已卸载、HMR(热模块替换)重载中),subagent 工具不存在,该窗口内的模型请求中不会包含它。这是诚实的状态——替代方案是注册一个 description 或执行都不可信的工具。
- 严格性意味着 persona 可能在渲染时导致轮次失败(例如在无 cwd 的会话上使用 `{{cwd}}`)。失败是受控的——该轮次以 `error` 结束,循环存活——且这是一个我们希望明确暴露的撰写错误。
- 目前没有在提示词行文中转义字面 `{{name}}` 的语法;如果真实提示词确实需要,再行添加。
- 插值文本仍不支持行内转义;字面文本段无需转义。PTC 单元测试覆盖两种模式和运行时语言,录制的 `ptc-turn` 场景在模型可见的提示词中保留工具模板示例。
@@ -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/2026-07-05-reconstructable-requests.md
2026-07-05-reconstructable-requests.md: 2f88675a75a72e7fbf105dfbf4f337a4dd80948a
2026-07-05-reconstructable-requests.zh.md: 90008502a6651e38c142b7fb88052c05d46dea76
2026-07-05-reconstructable-requests.md: f47d0938b91edf6c9b544223ac44925f02dd8c1f
2026-07-05-reconstructable-requests.zh.md: d2499a6264a3e6bebf31944d8eb056265e082c68
@@ -22,9 +22,9 @@ Prefix-cache stability is corollary #1, not the headline: an append-only log pro
**Messages.** `Session.deriveMessages()` is cached: each surface entry is projected exactly once, when first seen, through the public per-event function `deriveEventMessage(event)`; a surface rewrite (a compaction `replace` — `SurfaceManager.replaceGeneration`) rebuilds. Callers get a fresh array per call over shared, deep-frozen messages: mutating logged history through a projection is unrepresentable (it throws), replacing the old clone-per-call isolation. External reconstructors fold the same public function over a log prefix, so no two paths can disagree.
`EpochHeader` records the request's non-history state: call config and tool schemas. Writers omit `tools: []` and `adapterDefaults: {}`; current acceptance rejects those fields and any `header.system`, rather than repairing them. Whitespace-only system-message content, `config.stop: []`, and nested extensions remain intact. The [V3 canonical-envelope decision](2026-09-06-v3-canonical-session-envelopes.md) owns historical conversion. The rendered system prompt is derived history — the `system/message` event at surface node 0, per the [surface-node Agent Note](2026-09-02-system-prompt-as-surface-node.md) — so a prompt change is a surface replacement rather than a header change. Adapter-supplied effort and token defaults retain their `adapterDefaults` provenance; a Web model selection restored from the log omits an adapter-owned effort so the next resolution cannot reclassify the same effective config as an explicit selection and a false change. `request/header` always writes a full snapshot: the first loop instance uses reason `initial`, later instances use `resume`, an in-instance change uses `change`, and an unchanged envelope beginning an explicitly declared message series or following a surface replacement uses `series`. A `change` snapshot carries `startsSeries: true` when the changed request also starts a series, preserving the two independent facts without a duplicate header. Ordinary append-only later Turns, further same-series Steps, and retries inherit the latest snapshot. `foldRequestHeader` selects the latest snapshot. Legacy `request/header-delta` events and the removed `fallback` reason are rejected when appended or loaded.
`EpochHeader` records the request's non-history state: call config and tool schemas. Writers omit `tools: []` and `adapterDefaults: {}`; current acceptance rejects those fields and any `header.system`, rather than repairing them. Whitespace-only system-message content, `config.stop: []`, and nested extensions remain intact. The [V3 canonical-envelope decision](2026-09-06-v3-canonical-session-envelopes.md) owns historical conversion. The rendered system prompt is derived history — the `system/message` event at surface node 0, per the [surface-node Agent Note](2026-09-02-system-prompt-as-surface-node.md) — so a prompt change is a surface replacement rather than a header change. Adapter-supplied effort and token defaults retain their `adapterDefaults` ownership; a Web model selection restored from the log omits an adapter-owned effort so the next resolution cannot reclassify the same effective config as an explicit selection and a false change. `request/header` always writes a full snapshot: the first loop instance uses reason `initial`, later instances use `resume`, an in-instance change uses `change`, and an unchanged envelope beginning an explicitly declared message series or following a surface replacement uses `series`. A `change` snapshot carries `startsSeries: true` when the changed request also starts a series, preserving the two independent facts without a duplicate header. Ordinary append-only later Turns, further same-series Steps, and retries inherit the latest snapshot. `foldRequestHeader` selects the latest snapshot. Legacy `request/header-delta` events and the removed `fallback` reason are rejected when appended or loaded.
Each proposed step first claims its inbox batch, assembles the system prompt and tools, projects the rendered prompt against the surviving `system/message` node, and runs `agent/pre-step`. Rejection opens no step; enter opens `step/start`, commits a changed prompt as the `system/message` append or node-0 replacement, records the final message batch as `user/message` events, and may use `startsRequestSeries: true` to declare a distinct series. `agent/request` may replace only the frozen call-config seed. The loop records the owed initial, resume, change, or series full snapshot, builds `GenerateOptions` from derived messages (system message first) and that header with no `system` field, and freezes it while leaving `AbortSignal` live. The [request-freeze provenance decision](../simplification/2026-09-06-agent-request-freeze-provenance.md) owns reuse of completed message freezes and per-request local header freezing. The first call config starts from explicit `AgentOptions`, preserving fork overrides and resume reconfiguration; later calls start from the folded header.
Each proposed step first claims its inbox batch, assembles the system prompt and tools, projects the rendered prompt against the surviving `system/message` node, and runs `agent/pre-step`. Rejection opens no step; enter opens `step/start`, commits a changed prompt as the `system/message` append or node-0 replacement, records the final message batch as `user/message` events, and may use `startsRequestSeries: true` to declare a distinct series. `agent/request` may replace only the frozen call-config seed. The loop records the owed initial, resume, change, or series full snapshot, builds `GenerateOptions` from derived messages (system message first) and that header with no `system` field, and freezes it while leaving `AbortSignal` live. The [request-freeze evidence decision](../simplification/2026-09-06-agent-request-freeze-evidence.md) owns reuse of completed message freezes and per-request local header freezing. The first call config starts from explicit `AgentOptions`, preserving fork overrides and resume reconfiguration; later calls start from the folded header.
**The open step is the reconstruction boundary.** Its entered `user/message` batch and any newly written `request/header` precede request dispatch. Injection after the atomic claim joins a later request, while a listener that must affect this request returns messages through `agent/pre-step`. Header reconstruction selects the step's `request/header`, or carries the prior snapshot when no new header is written.
@@ -24,7 +24,7 @@ Status: implemented
`EpochHeader` 记录请求的非历史状态:调用配置和工具 schema。写入方省略 `tools: []` 与 `adapterDefaults: {}`;当前接纳拒绝这些字段以及任何 `header.system`,而不修复它们。仅含空白的系统消息内容、`config.stop: []` 与嵌套扩展保持原样。[V3 规范信封决策](2026-09-06-v3-canonical-session-envelopes.zh.md)负责历史转换。渲染后的系统提示词是派生历史——surface 第 0 号节点上的 `system/message` 事件,见[surface 节点 Agent Note](2026-09-02-system-prompt-as-surface-node.zh.md)——因此提示词变更是 surface 替换而不是 header 变更。适配器提供的推理强度与 token 默认值会保留其 `adapterDefaults` 来源信息;Web 从日志恢复模型选择时会省略适配器持有的推理强度,因此下一次解析不会把相同的有效配置重新归类为显式选择并产生虚假变更。`request/header` 始终写入完整快照:首个循环实例使用 reason `initial`,后续实例使用 `resume`,实例内变更使用 `change`,内容未变的封装显式开启消息序列或跟随表层替换时使用 `series`。如果发生变化的请求同时开启序列,`change` 快照会携带 `startsSeries: true`,无需重复 header 即可保留这两个独立事实。普通的仅追加后续 Turn、同一序列内后续的 Step 与重试沿用最新快照。`foldRequestHeader` 选择最新快照。旧的 `request/header-delta` 事件和已移除的 `fallback` reason 在追加或加载时都会被拒绝。
每个拟议步骤先领取其 inbox 批次,组装系统提示词与工具,把渲染后的提示词与存活的 `system/message` 节点比对投影,再运行 `agent/pre-step`。reject 不打开步骤;enter 打开 `step/start`,把变化的提示词作为 `system/message` 追加或第 0 号节点替换提交,把最终消息批次记录为 `user/message` 事件,并可使用 `startsRequestSeries: true` 声明独立序列。`agent/request` 只能替换冻结的调用配置种子。循环记录所需的 initial、resume、change 或 series 完整快照,从派生消息(系统消息在先)与该不含 `system` 字段的 header 构建 `GenerateOptions`,冻结请求但保持 `AbortSignal` 活跃。[请求冻结来源证明决策](../simplification/2026-09-06-agent-request-freeze-provenance.zh.md)拥有消息完整冻结的复用规则和每次请求的本地 header 冻结规则。首次调用配置从显式的 `AgentOptions` 出发,保留 fork 覆盖和恢复重配置;后续调用从折叠后的 header 出发。
每个拟议步骤先领取其 inbox 批次,组装系统提示词与工具,把渲染后的提示词与存活的 `system/message` 节点比对投影,再运行 `agent/pre-step`。reject 不打开步骤;enter 打开 `step/start`,把变化的提示词作为 `system/message` 追加或第 0 号节点替换提交,把最终消息批次记录为 `user/message` 事件,并可使用 `startsRequestSeries: true` 声明独立序列。`agent/request` 只能替换冻结的调用配置种子。循环记录所需的 initial、resume、change 或 series 完整快照,从派生消息(系统消息在先)与该不含 `system` 字段的 header 构建 `GenerateOptions`,冻结请求但保持 `AbortSignal` 活跃。[请求冻结证据决策](../simplification/2026-09-06-agent-request-freeze-evidence.zh.md)拥有消息完整冻结的复用规则和每次请求的本地 header 冻结规则。首次调用配置从显式的 `AgentOptions` 出发,保留 fork 覆盖和恢复重配置;后续调用从折叠后的 header 出发。
**已打开步骤是重建边界。** 进入步骤的 `user/message` 批次与任何新写入的 `request/header` 都位于请求分派之前。原子领取后发生的注入加入后续请求;必须影响本次请求的监听器则通过 `agent/pre-step` 返回消息。header 重建选择该步骤的 `request/header`,或在无新 header 写入时沿用前一个快照。
@@ -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/2026-07-08-tool-output-spill-files.md
2026-07-08-tool-output-spill-files.md: a80b6cdcb0e731a687d511d38ea288a68a298173
2026-07-08-tool-output-spill-files.md: fa6697aad63c97b0ef580fe1ca46e50541e3d682
2026-07-08-tool-output-spill-files.zh.md: a995eafc7878b342a2164c5116da1f43ada791a5
@@ -22,7 +22,7 @@ A thin spill storage seam plus a default spill policy plugin, in a new `packages
| `@deepseek-ai/dsh-spill-local` | Local backend: private, session-scoped file storage on the host filesystem. |
| `@deepseek-ai/dsh-spill-policy` | Tool-result policy plugin: wraps final text results after dispatch and replaces oversized results with a retained preview plus a spill locator. |
The tool-result Consumer is `dsh-spill-policy`, which consumes final tool results through the `tools/post-execute` waterfall. The model follows the backend-supplied retrieval hint for the returned locator. [Session-reference spill reuse](../bug-fix/2026-09-05-session-reference-spill-reuse.md) adds a direct storage consumer with separate preview, provenance, and failure semantics; it does not change the tool-result policy.
The tool-result Consumer is `dsh-spill-policy`, which consumes final tool results through the `tools/post-execute` waterfall. The model follows the backend-supplied retrieval hint for the returned locator. [Session-reference spill reuse](../bug-fix/2026-09-05-session-reference-spill-reuse.md) adds a direct storage consumer with separate preview, source-description, and failure semantics; it does not change the tool-result policy.
### Spill seam
@@ -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: 7792e24a5869be6b7bae7787a6a481f3075ba740
2026-07-10-single-file-executable-sdk-runtime-distribution.zh.md: 857bfaec80da962fac0403f2239e0a5e71954194
2026-07-10-single-file-executable-sdk-runtime-distribution.md: 665364e6d39a78f7ac497198f18ed9aa160a4cbd
2026-07-10-single-file-executable-sdk-runtime-distribution.zh.md: 8b39df71c615dc59f3ef1bf622a5736a70b085b3
@@ -40,13 +40,13 @@ Inside the exe's VFS sits a **real package tree in build-artifact form** (each p
The deploy root is [`python/sdk-runtime/package.json`](../../../../python/sdk-runtime/package.json) (`dsh-python-runtime-closure`, a pnpm workspace member and a zero-code pure dependency manifest) — the unified source of truth for "which plugins the exe ships" and "what the Python runtime distributes". Adding a plugin to the exe = adding one dependency line to the manifest and repackaging. [`scripts/verify-runtime-closure.ts`](../../../../scripts/verify-runtime-closure.ts) reads every shipped `packages/preset/agent-presets/presets/*/agent.cordis.yml`, evaluates `disabled` conditions that compare `process.platform` for every target in `python/sdk-runtime/platforms.json`, and requires each active workspace plugin at the runtime root through an explicit `workspace:` dependency. It also traverses every workspace package covered by that manifest and requires every non-optional workspace peer, reporting the complete preset or referencing-package → missing-dependency chain; unknown platform conditions remain active so a plugin cannot be omitted by an unsupported expression. `pnpm run hygiene`, CI static, and the single-exe build run it before packaging. Deploy also packs by each package's `files`, so the shared chunks tsdown splits out must be covered by `files`.
The deploy root includes `@deepseek-ai/dsh-mcp-client` as an explicitly supported custom-configuration plugin even though no shipped preset mounts it. An external config can therefore connect to user-supplied stdio and Streamable HTTP MCP servers and register their tools; the distribution does not carry those servers or extend the bridge to MCP Resources and Prompts. The executable and installed-wheel smokes start a temporary stdio server, discover its tool, and complete one model-requested call.
The deploy root includes `@deepseek-ai/dsh-mcp-client` as an explicitly supported custom-configuration plugin even though no shipped preset mounts it. An external config can therefore connect to user-supplied stdio and Streamable HTTP MCP servers and register their tools; the distribution does not carry those servers. The optional resource service provides MCP Resources; MCP Prompts remain unsupported. The executable and installed-wheel smokes start a temporary stdio server, discover its tool, and complete one model-requested call.
### Build pipeline and artifacts
[`scripts/build-exe-for-python-sdk.ts`](../../../../scripts/build-exe-for-python-sdk.ts): runtime closure verification → `pnpm run build` → (after clearing) `pnpm --filter dsh-python-runtime-closure deploy --legacy --prod --config.node-linker=hoisted --config.auto-install-peers=false --config.link-workspace-packages=true` **directly into** `python/sdk-runtime/src/deepseek_harness_runtime/runtime/node/`, including [`python/sdk-runtime/runtime-bootstrap.mjs`](../../../../python/sdk-runtime/runtime-bootstrap.mjs) as the carrier-root `runtime-bootstrap.mjs` → restore direct workspace packages omitted by legacy deploy and reject any remaining manifest gap → replace staged dependency symlinks with their target bytes, remove package-manager `.bin` links, and fail if any symlink remains → verify the deployed bootstrap and inject pkg configuration with that bin plus assets covering dynamic profile, bundle, frontend, preset, native-library, and configuration reads → stage the target `node-pty` addon → invoke `pkg --sea` once per target → write `deepseek-harness-sdk-runtime-<platform>-<arch>` under `dist-exe/` and copy it into the runtime directory. The Python runtime owns that bootstrap. It calls the public CLI export for ordinary launches and dispatches a provider-private selection to the same `@deepseek-ai/dsh-subprocess-local/runner` core without changing CLI grammar or adding another executable; the [native-containment decision](2026-08-28-subprocess-native-containment.md) owns that private path. Linux CI rebuilds `pty.node` inside the matching manylinux 2.28 container because legacy deploy omits that install side effect. Every target copies its native `@vscode/ripgrep` binary beside the executable as the required `-rg` sidecar; pkg runtimes select that sidecar through `process.pkg`, while ordinary Node execution uses `@vscode/ripgrep` directly. macOS uses its target prebuild and also emits the required `-spawn-helper`. All four deploy flags are grounded in measurement: `--legacy` is the mandatory path with inject-workspace-packages off; hoisted gives pkg a stable single-instance layout that the explicit materialization pass makes symlink-free; disabling automatic peer installation prevents undeclared peers from expanding the closure; link-workspace-packages selects direct workspace dependencies. [`pnpm-workspace.yaml`](../../../../pnpm-workspace.yaml) overrides the transitive `@deepseek-ai/cosmokit` and `@deepseek-ai/schemastery` semver requests to the pinned vendor sources so legacy deploy never resolves those unpublished names from a registry.
[`scripts/build-exe-for-python-sdk.ts`](../../../../scripts/build-exe-for-python-sdk.ts): runtime closure verification → `pnpm run build` → (after clearing) `pnpm --filter dsh-python-runtime-closure deploy --legacy --prod --config.allow-unused-patches=true --config.node-linker=hoisted --config.auto-install-peers=false --config.link-workspace-packages=true` **directly into** `python/sdk-runtime/src/deepseek_harness_runtime/runtime/node/`, including [`python/sdk-runtime/runtime-bootstrap.mjs`](../../../../python/sdk-runtime/runtime-bootstrap.mjs) as the carrier-root `runtime-bootstrap.mjs` → restore direct workspace packages omitted by legacy deploy and reject any remaining manifest gap → replace staged dependency symlinks with their target bytes, remove package-manager `.bin` links, and fail if any symlink remains → verify the deployed bootstrap and inject pkg configuration with that bin plus assets covering dynamic profile, bundle, frontend, preset, native-library, and configuration reads → stage the target `node-pty` addon → invoke `pkg --sea` once per target → write `deepseek-harness-sdk-runtime-<platform>-<arch>` under `dist-exe/` and copy it into the runtime directory. The Python runtime owns that bootstrap. It calls the public CLI export for ordinary launches and dispatches a provider-private selection to the same `@deepseek-ai/dsh-subprocess-local/runner` core without changing CLI grammar or adding another executable; the [native-containment decision](2026-08-28-subprocess-native-containment.md) owns that private path. Linux CI rebuilds `pty.node` inside the matching manylinux 2.28 container because legacy deploy omits that install side effect. Every target copies its native `@vscode/ripgrep` binary beside the executable as the required `-rg` sidecar; pkg runtimes select that sidecar through `process.pkg`, while ordinary Node execution uses `@vscode/ripgrep` directly. macOS uses its target prebuild and also emits the required `-spawn-helper`. The deploy flags preserve the measured runtime layout: `--legacy` is the mandatory path with inject-workspace-packages off; hoisted gives pkg a stable single-instance layout that the explicit materialization pass makes symlink-free; disabling automatic peer installation prevents undeclared peers from expanding the closure; link-workspace-packages selects direct workspace dependencies. [`pnpm-workspace.yaml`](../../../../pnpm-workspace.yaml) overrides the transitive `@deepseek-ai/cosmokit` and `@deepseek-ai/schemastery` semver requests to the pinned vendor sources so legacy deploy never resolves those unpublished names from a registry. Production deployment permits unused workspace patches because development-only tools can be absent from its closure; included-package patch failures still fail deployment. Workspace installs retain strict unused-patch validation.
CI: [`.github/workflows/build-exe-for-python-sdk.yml`](../../../../.github/workflows/build-exe-for-python-sdk.yml) runs [installed-wheel validation](../testing/2026-08-23-installed-python-wheel-black-box-ci.md) on Linux/Windows x64 for pull requests and Linux ARM64 plus both macOS architectures for master pushes. The [public publication workflow](../../archived/process/2026-08-11-python-publication-workflow.md) calls it for all five targets; `workflow_dispatch` can still select a subset. Native builds run on linux-x64 / linux-arm64 (`ubuntu-24.04-arm`) / macos-arm64 / macos-x64 (`macos-15-intel`) / win-x64 (`windows-2025`), with `~/.pkg-cache` cached where applicable, and pkg handles macOS ad-hoc signing. Each leg installs the release-shaped SDK and runtime wheels into a clean venv outside the checkout, proves their package and executable provenance, then drives the complete keyless scenario set through the public SDK and direct NDJSON JSON-RPC. Trusted pull requests and master pushes additionally run a real DeepSeek two-turn tool smoke on their selected targets; fork and Dependabot heads receive no key. Linux inspects the executable and native addon's GLIBC requirements and runs an additional manylinux 2.28 smoke, while macOS checks the runtime, ripgrep, and PTY helper architectures and verifies that all three deployment targets fit the wheel tag. A full five-target run retains six artifacts, each containing one release file: the platform-independent SDK wheel and five native runtime wheels; a subset dispatch retains the SDK wheel and selected runtime wheels. Bare executables and source bundles remain intermediate test inputs. [`.gitlab-ci.yml`](../../../../.gitlab-ci.yml) accepts `python-v<repository-version>` tag pipelines whose version matches the root `package.json`, builds one SDK wheel and five native runtime wheels, then a single serialized job checks and publishes all six to the project PyPI registry. The [`python/sdk-runtime` README](../../../../python/sdk-runtime/README.md) owns the Windows target and the explicit exclusion of Windows arm64.
CI: [`.github/workflows/build-exe-for-python-sdk.yml`](../../../../.github/workflows/build-exe-for-python-sdk.yml) runs [installed-wheel validation](../testing/2026-08-23-installed-python-wheel-black-box-ci.md) on Linux/Windows x64 for pull requests and Linux ARM64 plus both macOS architectures for master pushes. The [public publication workflow](../../archived/process/2026-08-11-python-publication-workflow.md) calls it for all five targets; `workflow_dispatch` can still select a subset. Native builds run on linux-x64 / linux-arm64 (`ubuntu-24.04-arm`) / macos-arm64 / macos-x64 (`macos-15-intel`) / win-x64 (`windows-2025`), with `~/.pkg-cache` cached where applicable, and pkg handles macOS ad-hoc signing. Each leg installs the release-shaped SDK and runtime wheels into a clean venv outside the checkout, verifies their package and executable origin, then drives the complete keyless scenario set through the public SDK and direct NDJSON JSON-RPC. Trusted pull requests and master pushes additionally run a real DeepSeek two-turn tool smoke on their selected targets; fork and Dependabot heads receive no key. Linux inspects the executable and native addon's GLIBC requirements and runs an additional manylinux 2.28 smoke, while macOS checks the runtime, ripgrep, and PTY helper architectures and verifies that all three deployment targets fit the wheel tag. A full five-target run retains six artifacts, each containing one release file: the platform-independent SDK wheel and five native runtime wheels; a subset dispatch retains the SDK wheel and selected runtime wheels. Bare executables and source bundles remain intermediate test inputs. [`.gitlab-ci.yml`](../../../../.gitlab-ci.yml) accepts `python-v<repository-version>` tag pipelines whose version matches the root `package.json`, builds one SDK wheel and five native runtime wheels, then a single serialized job checks and publishes all six to the project PyPI registry. The [`python/sdk-runtime` README](../../../../python/sdk-runtime/README.md) owns the Windows target and the explicit exclusion of Windows arm64.
### Python SDK distribution: two carriers, exe for production, node for development
@@ -60,13 +60,13 @@ The Python client launches the packaged `dsh` command with the selected profile
`dsh-python-runtime-closure` is the private deploy manifest and `deepseek-harness-sdk-runtime-<platform>-<arch>` is the executable family. The wire `serverInfo.name` is `deepseek-harness-sdk-runtime`; the Python distribution names are `deepseek-harness-sdk` / `deepseek-harness-runtime-bin`, while the import modules are `deepseek_harness` / `deepseek_harness_runtime`.
## Disposition of worker-style plugins
## Packaged workflow and code execution
`dsh-workflow-worker-thread` and `dsh-code-runtime-worker-thread` are supported inside the exe. Their built hosts convert the sibling `lib/worker.cjs` URL with `fileURLToPath()` and pass the resulting filesystem string to `Worker`, which is the form pkg's Worker hook resolves inside the VFS. The worker entries are CommonJS because that hook compiles VFS worker files as CommonJS. The workflow engine keeps its data-URL bootstrap for unbuilt source execution; only its built sibling entry uses the filesystem string. The custom-config executable smoke loads both backends, invokes a real `run_code` call and a zero-agent `workflow` call, and requires each worker to return `42` from inside pkg's VFS.
`dsh-workflow-ptc` sends its self-contained guest program through `dsh-ptc-runtime-node`. The shared provider starts a managed child through the executable's private Node bootstrap dispatch, with a dedicated control channel; no separate workflow worker asset is required. The [sandboxed Node decision](2026-09-11-sandboxed-node-ptc-runtime.md) owns process and policy lifetime, and [workflow sandbox reuse](2026-09-13-workflow-ptc-sandbox-reuse.md) owns the workflow adapter.
## Testing
The verification surface has three tiers. Mechanism tier: the measured conclusions for the `--sea` chain are embedded in the Decision sections (ESM dynamic import inside the VFS, single cordis instance, fail-loud config chain, `node:sqlite`, macOS ad-hoc signing runs). SDK tier: the complete keyless pytest suite covers the client protocol against a fake runtime peer, subprocess cleanup, absolute cwd propagation, dual-carrier launch, and carrier resolution; root CI runs it on Python 3.10. End-to-end tier: every platform build installs both wheels into a clean venv outside the checkout, proves matching versions and installed module/executable locations, then completes turns against a mock endpoint through the default SDK path, a custom config, the checked-in standalone minimal composition, and the direct binary protocol, with final text and JSONL checked. The minimal run asserts its exact system prompt and single shell-tool catalog and retains shell state across calls. The custom config additionally drives `run_code` and a zero-agent `workflow` through their real worker files inside the packaged VFS. The filesystem-search scenario requires the model to call both `glob` and `grep` through the target-native `-rg` sidecar. The spawn-node scenario drives the platform shell tool through a command starting with `node` and requires the machine's own Node version in the tool result with no `PKG_EXECPATH` in the child environment, pinning the packaged runtime against a pkg upgrade that re-records the child-process patch. The MCP scenario starts a temporary external stdio server, deliberately delays its initial `tools/list` response, then immediately starts the first SDK prompt; the prompt must see and call the discovered tool, proving that `initialize` is a real Loader-settlement readiness boundary rather than a timing sleep. The same installed run compares a committed executable-specific snapshot through the Python SDK: a keyless scripted model mounts a Cordis plugin that registers a tool, invokes that tool from `run_code`, runs a direct spawn subagent and a workflow that starts a second spawn child, then unmounts the plugin. The fixture explicitly disables its unused bundled Bash and local skill discovery so its tool set does not depend on repository-external state, and the comparison normalizes opaque message, agent, workflow-run, and session IDs across the SDK result and notification stream plus the parent and two child JSONL logs. Trusted pull requests add a real-provider two-turn file write/read whose external bytes, tool calls, completed reasons, and persisted log must agree. This harness stays separate from ACP's `pnpm run test:snapshot` because the protocols and build artifacts differ.
The verification surface has three tiers. Mechanism tier: the measured conclusions for the `--sea` chain are embedded in the Decision sections (ESM dynamic import inside the VFS, single cordis instance, fail-loud config chain, `node:sqlite`, macOS ad-hoc signing runs). SDK tier: the complete keyless pytest suite covers the client protocol against a fake runtime peer, subprocess cleanup, absolute cwd propagation, dual-carrier launch, and carrier resolution; root CI runs it on Python 3.10. End-to-end tier: every platform build installs both wheels into a clean venv outside the checkout, proves matching versions and installed module/executable locations, then completes turns against a mock endpoint through the default SDK path, a custom config, the checked-in standalone minimal composition, and the direct binary protocol, with final text and JSONL checked. The minimal run asserts its exact system prompt and single shell-tool catalog and retains shell state across calls. The custom config additionally drives `run_code` and a zero-agent `workflow` through the shared packaged Node process bootstrap. The filesystem-search scenario requires the model to call both `glob` and `grep` through the target-native `-rg` sidecar. The spawn-node scenario drives the platform shell tool through a command starting with `node` and requires the machine's own Node version in the tool result with no `PKG_EXECPATH` in the child environment, pinning the packaged runtime against a pkg upgrade that re-records the child-process patch. The MCP scenario starts a temporary external stdio server, deliberately delays its initial `tools/list` response, then immediately starts the first SDK prompt; the prompt must see and call the discovered tool, proving that `initialize` is a real Loader-settlement readiness boundary rather than a timing sleep. The same installed run compares a committed executable-specific snapshot through the Python SDK: a keyless scripted model mounts a Cordis plugin that registers a tool, invokes that tool from `run_code`, runs a direct spawn subagent and a workflow that starts a second spawn child, then unmounts the plugin. The fixture explicitly disables its unused bundled Bash and local skill discovery so its tool set does not depend on repository-external state, and the comparison normalizes opaque message, agent, workflow-run, and session IDs across the SDK result and notification stream plus the parent and two child JSONL logs. Trusted pull requests add a real-provider two-turn file write/read whose external bytes, tool calls, completed reasons, and persisted log must agree. This harness stays separate from ACP's `pnpm run test:snapshot` because the protocols and build artifacts differ.
Manual-driving caveat: the bin treats stdin EOF as "the client is gone" and disposes immediately, so a short-lived pipe aborts an in-flight turn — pipe-driven runs must keep stdin open until the turn ends.
@@ -40,11 +40,11 @@ exe 的 VFS 内是**构建产物形态的真实包树**(各包的 `lib/` + 真
部署根目录是 [`python/sdk-runtime/package.json`](../../../../python/sdk-runtime/package.json)(`dsh-python-runtime-closure`,pnpm 工作区成员、零代码纯依赖 manifest),也是「exe 安装哪些插件」与「Python 运行时分发什么」的统一真源。向 exe 添加插件,就是在 manifest 中增加一行依赖后重新打包。[`scripts/verify-runtime-closure.ts`](../../../../scripts/verify-runtime-closure.ts) 读取每个已发布的 `packages/preset/agent-presets/presets/*/agent.cordis.yml`,针对 `python/sdk-runtime/platforms.json` 中的每个目标解析比较 `process.platform` 的 `disabled` 条件,并要求该目标启用的每个工作区插件都通过显式的 `workspace:` 依赖列在运行时根目录。它还遍历该 manifest 覆盖的全部工作区包,要求每个非可选的工作区对等依赖(peer dependency)都显式列出,并报告“preset 或引用包 → 缺失依赖”的完整链路;无法识别的平台条件会保持启用,避免因不支持的表达式遗漏插件。`pnpm run hygiene`、CI 静态检查与 single-exe 构建都会在打包前运行该门禁。部署还会依据各包的 `files` 字段打包,因此 tsdown 拆出的共享分片必须被 `files` 覆盖。
部署根目录显式包含 `@deepseek-ai/dsh-mcp-client`,将其作为自定义配置可用的插件,即使随附 preset 均未挂载该插件。外部配置因此可以连接由用户提供的 stdio 与 Streamable HTTP MCP server 并注册其工具;分发物不包含这些 server,也不将桥接范围扩展到 MCP Resources 和 Prompts。可执行程序与已安装 wheel 包的冒烟测试会启动临时 stdio server,发现其工具,并完成一次由模型请求的调用。
部署根目录显式包含 `@deepseek-ai/dsh-mcp-client`,将其作为自定义配置可用的插件,即使随附 preset 均未挂载该插件。外部配置因此可以连接由用户提供的 stdio 与 Streamable HTTP MCP server 并注册其工具;分发物不包含这些 server。可选资源服务提供 MCP Resources;MCP Prompts 仍不受支持。可执行程序与已安装 wheel 包的冒烟测试会启动临时 stdio server,发现其工具,并完成一次由模型请求的调用。
### 构建流水线与产物
[`scripts/build-exe-for-python-sdk.ts`](../../../../scripts/build-exe-for-python-sdk.ts):运行时闭包校验 → `pnpm run build` →(清空后)`pnpm --filter dsh-python-runtime-closure deploy --legacy --prod --config.node-linker=hoisted --config.auto-install-peers=false --config.link-workspace-packages=true` **直接写入** `python/sdk-runtime/src/deepseek_harness_runtime/runtime/node/`,其中包含作为载体根目录 `runtime-bootstrap.mjs` 的 [`python/sdk-runtime/runtime-bootstrap.mjs`](../../../../python/sdk-runtime/runtime-bootstrap.mjs) → 恢复 legacy deploy 遗漏的直接工作区包,并拒绝剩余的 manifest 缺口 → 将暂存依赖中的符号链接替换为目标文件内容,删除包管理器的 `.bin` 链接,并在仍有任何符号链接时失败 → 验证已部署的 bootstrap,并注入以该文件为 bin 的 pkg 配置及覆盖动态读取 profile、bundle、前端、preset、原生库与配置文件的 assets → 暂存目标平台的 `node-pty` addon → 每个构建目标调用一次 `pkg --sea` → 将 `deepseek-harness-sdk-runtime-<platform>-<arch>` 写入 `dist-exe/` 并拷回运行时目录。该 bootstrap 由 Python runtime 拥有;普通启动时它调用公开 CLI export,而提供方私有选择会分派到同一个 `@deepseek-ai/dsh-subprocess-local/runner` 核心,不改变 CLI 语法,也不增加另一个可执行文件;[原生 containment 决策](2026-08-28-subprocess-native-containment.zh.md)负责这条私有路径。Linux CI 会在匹配的 manylinux 2.28 容器中重新构建 `pty.node`,因为 legacy deploy 会遗漏这一安装副作用。每个目标都会把对应的原生 `@vscode/ripgrep` 二进制复制到可执行文件旁,作为必需的 `-rg` 伴随文件;pkg 运行时通过 `process.pkg` 选择该伴随文件,普通 Node 执行则直接使用 `@vscode/ripgrep`。macOS 使用对应目标的预构建产物,并额外生成所需的 `-spawn-helper`。四个部署标志都有实测依据:未启用 `inject-workspace-packages` 时必须使用 `--legacy`;`hoisted` 为 pkg 提供稳定的单实例布局,再由显式物化步骤消除符号链接;关闭对等依赖自动安装可防止未声明的对等依赖扩大闭包;`link-workspace-packages` 选择直接工作区依赖。[`pnpm-workspace.yaml`](../../../../pnpm-workspace.yaml) 将传递的 `@deepseek-ai/cosmokit` 与 `@deepseek-ai/schemastery` semver 请求覆盖到固定的 vendor 源码,使 legacy deploy 不会从注册表解析这些未发布名称。
[`scripts/build-exe-for-python-sdk.ts`](../../../../scripts/build-exe-for-python-sdk.ts):运行时闭包校验 → `pnpm run build` →(清空后)`pnpm --filter dsh-python-runtime-closure deploy --legacy --prod --config.allow-unused-patches=true --config.node-linker=hoisted --config.auto-install-peers=false --config.link-workspace-packages=true` **直接写入** `python/sdk-runtime/src/deepseek_harness_runtime/runtime/node/`,其中包含作为载体根目录 `runtime-bootstrap.mjs` 的 [`python/sdk-runtime/runtime-bootstrap.mjs`](../../../../python/sdk-runtime/runtime-bootstrap.mjs) → 恢复 legacy deploy 遗漏的直接工作区包,并拒绝剩余的 manifest 缺口 → 将暂存依赖中的符号链接替换为目标文件内容,删除包管理器的 `.bin` 链接,并在仍有任何符号链接时失败 → 验证已部署的 bootstrap,并注入以该文件为 bin 的 pkg 配置及覆盖动态读取 profile、bundle、前端、preset、原生库与配置文件的 assets → 暂存目标平台的 `node-pty` addon → 每个构建目标调用一次 `pkg --sea` → 将 `deepseek-harness-sdk-runtime-<platform>-<arch>` 写入 `dist-exe/` 并拷回运行时目录。该 bootstrap 由 Python runtime 拥有;普通启动时它调用公开 CLI export,而提供方私有选择会分派到同一个 `@deepseek-ai/dsh-subprocess-local/runner` 核心,不改变 CLI 语法,也不增加另一个可执行文件;[原生 containment 决策](2026-08-28-subprocess-native-containment.zh.md)负责这条私有路径。Linux CI 会在匹配的 manylinux 2.28 容器中重新构建 `pty.node`,因为 legacy deploy 会遗漏这一安装副作用。每个目标都会把对应的原生 `@vscode/ripgrep` 二进制复制到可执行文件旁,作为必需的 `-rg` 伴随文件;pkg 运行时通过 `process.pkg` 选择该伴随文件,普通 Node 执行则直接使用 `@vscode/ripgrep`。macOS 使用对应目标的预构建产物,并额外生成所需的 `-spawn-helper`。部署标志都有实测依据:未启用 `inject-workspace-packages` 时必须使用 `--legacy`;`hoisted` 为 pkg 提供稳定的单实例布局,再由显式物化步骤消除符号链接;关闭对等依赖自动安装可防止未声明的对等依赖扩大闭包;`link-workspace-packages` 选择直接工作区依赖。[`pnpm-workspace.yaml`](../../../../pnpm-workspace.yaml) 将传递的 `@deepseek-ai/cosmokit` 与 `@deepseek-ai/schemastery` semver 请求覆盖到固定的 vendor 源码,使 legacy deploy 不会从注册表解析这些未发布名称。 生产部署允许未使用的工作区补丁,因为开发工具可能不属于运行时闭包;闭包内包的补丁应用失败仍会中止部署。工作区安装继续严格校验未使用的补丁。
CI 使用 [`.github/workflows/build-exe-for-python-sdk.yml`](../../../../.github/workflows/build-exe-for-python-sdk.yml):[安装后 wheel 包验证](../testing/2026-08-23-installed-python-wheel-black-box-ci.zh.md)在拉取请求上运行 Linux/Windows x64,在 master 推送上运行 Linux ARM64 与两种 macOS 架构。[公开发布工作流](../../archived/process/2026-08-11-python-publication-workflow.md)调用它构建全部五个目标;`workflow_dispatch` 仍可选择部分目标。linux-x64、linux-arm64(`ubuntu-24.04-arm`)、macos-arm64、macos-x64(`macos-15-intel`)与 win-x64(`windows-2025`)分别进行原生构建,并在适用平台缓存 `~/.pkg-cache`;macOS 的 ad-hoc 签名由 pkg 处理。每个平台都把发布形态的 SDK wheel 包与运行时 wheel 包安装到 checkout 外的干净 venv,证明包与可执行文件来源,再通过公开 SDK 与直接 NDJSON JSON-RPC 运行完整 keyless 场景。可信拉取请求与 master 推送还会在各自选定的目标上运行真实 DeepSeek 双轮工具冒烟测试;fork 与 Dependabot head 不会获得密钥。Linux 会检查可执行文件和原生 addon 各自的 GLIBC 依赖,并额外运行 manylinux 2.28 冒烟测试;macOS 则检查 runtime、ripgrep 与 PTY helper 的架构,并验证三个载荷的部署目标都符合 wheel 包标签。完整构建五个目标时保留 6 个产物,每个产物只含一个发布文件:平台无关的 SDK wheel 包与 5 个原生运行时 wheel 包;手动选择部分目标时保留 SDK wheel 与所选运行时 wheel。裸 exe 与源码包只作为测试中间输入。[`.gitlab-ci.yml`](../../../../.gitlab-ci.yml) 只接受版本与根目录 `package.json` 匹配的 `python-v<repository-version>` 标签流水线,构建一个 SDK wheel 包和 5 个原生运行时 wheel 包,再由单个串行任务校验并将这 6 个文件发布到项目的 PyPI 注册表。[`python/sdk-runtime` README](../../../../python/sdk-runtime/README.zh.md)负责 Windows 目标及对 Windows arm64 的明确排除。
@@ -60,14 +60,13 @@ Python 客户端使用所选 profile(默认 `sdk`)、有序 patch 文件和
`dsh-python-runtime-closure` 是私有部署 manifest,`deepseek-harness-sdk-runtime-<platform>-<arch>` 是可执行文件族。协议字段 `serverInfo.name` 是 `deepseek-harness-sdk-runtime`;Python 分发包名是 `deepseek-harness-sdk` / `deepseek-harness-runtime-bin`,导入模块名是 `deepseek_harness` / `deepseek_harness_runtime`。
## 工作线程插件
## 打包后的工作流与代码执行
exe 内支持 `dsh-workflow-worker-thread` 与 `dsh-code-runtime-worker-thread`。两个后端构建后的宿主都通过 `fileURLToPath()` 转换相邻 `lib/worker.cjs` 的 URL,再将所得文件系统字符串传给 `Worker`;pkg 的 Worker 钩子可以用这种形式解析 VFS 内文件。该钩子会把 VFS 内的工作线程文件作为 CommonJS 编译,所以工作线程入口采用 CommonJS。工作流引擎在未构建的源码执行中仍保留 `data:` URL 引导程序,只有构建后的相邻入口使用文件系统字符串。自定义配置的可执行文件冒烟测试会加载两个后端,实际调用 `run_code` 与不启动 agent(智能体)的 `workflow`,并要求两个工作线程都从 pkg 的 VFS 内返回 `42`。
`dsh-workflow-ptc` 通过 `dsh-ptc-runtime-node` 发送自包含 guest 程序。共享提供方通过可执行文件的私有 Node bootstrap 分派启动受管子进程,并使用专用控制通道;不需要独立的工作流 worker 资源。[沙箱 Node 决策](2026-09-11-sandboxed-node-ptc-runtime.zh.md)负责进程与策略生命周期,[工作流沙箱复用](2026-09-13-workflow-ptc-sandbox-reuse.zh.md)负责工作流适配器。
## 测试
验证面分三层。机制层:`--sea` 链路的实测结论内嵌在「决策」各节(VFS 内 ESM 动态 `import()`、单一 Cordis 实例、明确报错的配置链路、`node:sqlite`、macOS ad-hoc 签名可运行)。SDK 层:完整的无密钥 pytest 套件以 mock 运行时对端覆盖客户端协议、子进程清理、绝对 `cwd` 传递、双载体启动与载体解析;根 CI 在 Python 3.10 上运行全部用例。端到端层:每个平台构建都会把两个 wheel 包安装进 checkout 外的干净 venv,证明版本相同以及已安装模块/可执行文件的位置,再通过默认 SDK 路径、自定义配置、仓库内置的独立 minimal 组合和直接二进制协议,对 mock 端点完成轮次,并校验最终文本与 JSONL。minimal 运行会断言其精确系统提示词与单个 shell 工具目录,并跨调用保留 shell 状态。自定义配置还会通过打包进 VFS 的真实工作线程文件执行 `run_code` 和不启动 agent 的 `workflow`。文件系统搜索场景要求模型通过目标平台的 `-rg` 伴随文件调用 `glob` 与 `grep`。spawn-node 场景驱动平台 shell 工具执行以 `node` 开头的命令,要求工具结果给出机器自身的 Node 版本且子进程环境中无 `PKG_EXECPATH`,把打包运行时钉死在「pkg 升级重录 child_process 补丁也不得回归」的行为上。MCP 场景会启动临时外部 stdio server,刻意延迟首次 `tools/list` 响应,随后立即启动第一个 SDK 提示词;该提示词必须看到并调用已发现的工具,从而证明 `initialize` 是真正以 Loader 插件树完全稳定为准的就绪边界,而不是依赖定时 sleep。同一项安装后运行还会经 Python SDK 比较一组检入的 exe 专用快照:无密钥脚本化模型挂载一个会注册工具的 Cordis 插件,从 `run_code` 调用该工具,运行一个直接 spawn 的 subagent 和一个会通过 spawn 启动第二个 subagent 的工作流,随后卸载该插件。该 fixture(测试前置数据)会显式禁用组合包中未使用的 Bash 和本地 skill(技能)发现,使其工具集不依赖仓库外部状态;比较时会规范化 SDK 结果与通知流,以及父会话和两个子会话 JSONL 日志中不透明的消息、agent、工作流运行与会话 ID。可信拉取请求会增加真实提供方双轮文件写入/读取,并要求外部字节、工具调用、已完成原因与持久化日志一致。该 harness 与 ACP 的 `pnpm run test:snapshot` 保持独立,因为二者的协议和构建产物不同。
验证面分三层。机制层:`--sea` 链路的实测结论内嵌在「决策」各节(VFS 内 ESM 动态 `import()`、单一 Cordis 实例、明确报错的配置链路、`node:sqlite`、macOS ad-hoc 签名可运行)。SDK 层:完整的无密钥 pytest 套件以 mock 运行时对端覆盖客户端协议、子进程清理、绝对 `cwd` 传递、双载体启动与载体解析;根 CI 在 Python 3.10 上运行全部用例。端到端层:每个平台构建都会把两个 wheel 包安装进 checkout 外的干净 venv,证明版本相同以及已安装模块/可执行文件的位置,再通过默认 SDK 路径、自定义配置、仓库内置的独立 minimal 组合和直接二进制协议,对 mock 端点完成轮次,并校验最终文本与 JSONL。minimal 运行会断言其精确系统提示词与单个 shell 工具目录,并跨调用保留 shell 状态。自定义配置还会通过共享的打包后 Node 进程 bootstrap 执行 `run_code` 和不启动 agent 的 `workflow`。文件系统搜索场景要求模型通过目标平台的 `-rg` 伴随文件调用 `glob` 与 `grep`。spawn-node 场景驱动平台 shell 工具执行以 `node` 开头的命令,要求工具结果给出机器自身的 Node 版本且子进程环境中无 `PKG_EXECPATH`,把打包运行时钉死在「pkg 升级重录 child_process 补丁也不得回归」的行为上。MCP 场景会启动临时外部 stdio server,刻意延迟首次 `tools/list` 响应,随后立即启动第一个 SDK 提示词;该提示词必须看到并调用已发现的工具,从而证明 `initialize` 是真正以 Loader 插件树完全稳定为准的就绪边界,而不是依赖定时 sleep。同一项安装后运行还会经 Python SDK 比较一组检入的 exe 专用快照:无密钥脚本化模型挂载一个会注册工具的 Cordis 插件,从 `run_code` 调用该工具,运行一个直接 spawn 的 subagent 和一个会通过 spawn 启动第二个 subagent 的工作流,随后卸载该插件。该 fixture(测试前置数据)会显式禁用组合包中未使用的 Bash 和本地 skill(技能)发现,使其工具集不依赖仓库外部状态;比较时会规范化 SDK 结果与通知流,以及父会话和两个子会话 JSONL 日志中不透明的消息、agent、工作流运行与会话 ID。可信拉取请求会增加真实提供方双轮文件写入/读取,并要求外部字节、工具调用、已完成原因与持久化日志一致。该 harness 与 ACP 的 `pnpm run test:snapshot` 保持独立,因为二者的协议和构建产物不同。
手工驱动注意:`bin` 将 stdin EOF 视为「客户端已离开」并立即 dispose,生命周期较短的管道会中止进行中的轮次——管道驱动必须保持 stdin 打开,直到轮次结束。
@@ -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
@@ -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
@@ -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
@@ -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: 21bad78792c6b5aad48b51f454f6c08c0400ad72
2026-07-23-client-plugin-loading-model.zh.md: 2758f28f3bd34131ece3bed74152fbfe0b36174e
2026-07-23-client-plugin-loading-model.md: d0f9b20f0adabda6cc7132e5411bcadd60c9e108
2026-07-23-client-plugin-loading-model.zh.md: 27d5026ed03509d6408a16389246cb1321a37959
@@ -52,18 +52,18 @@ What happens between `dsh web` starting and the UI appearing? Three stages: the
**Host side — compose the graph.**
1. The composing app (`apps/cli`) ships the roster as ordinary rows in its `cordis.yml` config tree — client plugin packages are entry rows like every host plugin, including the always-mounted `client-hmr` row. A roster row that fails to import is caught by `assertEntriesLoaded`; a row whose fiber rejects is reported with its original stack by `assertEntriesActivated` ([host boot decision](2026-07-24-web-config-tree-boot-and-transport-layering.md)).
1. The composing app (`apps/cli`) ships the roster as ordinary rows in its `cordis.yml` config tree — client plugin packages are entry rows like every host plugin, including the always-mounted `client-hmr` row. `auditStartupEntries` reports failed imports, activation errors with their original stacks, and pending dependencies; optional entries warn and required entries reject startup ([startup policy](2026-09-09-consumer-owned-startup-strictness.md)).
2. The `dsh-client-modules` node half (the package is dual-face: its browser half is the module table) resolves each live Loader entry through the same `name` and owning-tree `baseUrl` inputs that imported its Host face, then reads the nearest owning package.json `dsh.client` declaration and composes `window.__DSH_BOOT__`: `{ rev, entries: [{ id, url, rev, inject?, immediately?, external? }], batches: [{ phase, url, rev, entries }] }`. The manifest package name is the browser module identity even when an overlay names a relative source or built entry file. Distinct active Loader sources resolving to one package name fail composition; after one source unloads, the surviving source supplies the row without a fiber restart. The row's three optional fields come from manifests, never hand-copied. Composition orders requested dynamic rows before their consumers, rejects synchronous request cycles, and assigns every row to exactly one initial batch. It refuses declared plugins without built `./client` bundles and groups their package/path rows under one required source-build instruction; malformed declaration fields also fail activation, and the Host audit reports either error from the FAILED fiber.
3. Scanning is incremental per package — there is no full-rescan code path. Each cordis `internal/plugin` emission marks the fiber's entry name dirty (entry-less fibers drop O(1)); a microtask flush reconciles each dirty name against live loader entries, with package metadata (including the negative "not a client package" verdict) cached per entry name and owning-tree base URL for the process lifetime and bundle re-hashing reachable only through `rebuilt(id)`. The activation pass seeds the same dirty set from current entries and flushes synchronously, so first scan and steady state share one implementation. Initial rows receive an opaque process nonce plus sequence without hashing their artifacts; startup combo revisions hash the combined script inputs plus indexed map, and the rows plus batch descriptors hash into `graph.rev`. The graph types are single-sourced in the modules package's `./client` export — the webserver knows nothing about the graph, while modules registers the combo route and contributes structured index-injection rows.
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.
@@ -52,18 +52,18 @@ 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
**host 侧——组合这张图。**
1. 负责组合的 app(`apps/cli`)把名册作为普通行放进它的 `cordis.yml` 配置树——client 插件包与每个 host 插件一样是 entry 行,包括无条件挂载的 `client-hmr` 行。名册行 import 失败由 `assertEntriesLoaded` 捕获;fiber reject 的行则由 `assertEntriesActivated` 报告原始 stack([host boot 决策](2026-07-24-web-config-tree-boot-and-transport-layering.zh.md))。
1. 负责组合的 app(`apps/cli`)把名册作为普通行放进它的 `cordis.yml` 配置树——client 插件包与每个 host 插件一样是 entry 行,包括无条件挂载的 `client-hmr` 行。`auditStartupEntries` 报告 import 失败、带原始 stack 的激活错误,以及待满足的依赖;optional entry 输出 warning,required entry 则使启动失败([启动策略](2026-09-09-consumer-owned-startup-strictness.zh.md))。
2. `dsh-client-modules` 的 node 半(该包是双面的:浏览器半就是模块表)使用 Host face import 时相同的 `name` 与所属 tree `baseUrl` 解析每个 live Loader entry,再读取最近归属 package.json 的 `dsh.client` 声明并组合出 `window.__DSH_BOOT__`:`{ rev, entries: [{ id, url, rev, inject?, immediately?, external? }], batches: [{ phase, url, rev, entries }] }`。即使 overlay 指向相对的 source 或 built entry 文件,manifest 包名仍是浏览器模块身份。若不同的 active Loader source 解析到同一包名,组合会失败;一个来源卸载后,仍存活的来源无需重启 fiber 即可提供该 row。Row 的三个可选字段都来自 manifest,永不人肉抄写。组合会把被请求的动态图 row 排到消费者之前、拒绝同步请求环,并把每个 row 恰好分配给一个初始批次。它会拒绝没有已构建 `./client` bundle 的已声明插件,并把它们的 package/path 行归到一条源码构建要求下;畸形声明字段同样会让激活失败,Host 检查会从 FAILED fiber 报告这两类错误。
3. 扫描是单包增量——不存在全量重扫代码路径。每次 cordis `internal/plugin` 发射把该 fiber 的 entry 名标脏(无 entry 的 fiber O(1) 丢弃);微任务 flush 把每个脏名对账 live loader entries,包元数据(含「非 client 包」的否定结论)按 entry 名与所属 tree base URL 缓存至进程结束,bundle 重哈希只经 `rebuilt(id)` 可达。激活趟从当前 entries 灌同一脏集合并同步 flush,初扫与稳态共享一条实现。初始 row 使用不透明的进程 nonce 加序号,不对其产物求哈希;启动 combo revision 对合并脚本输入及 indexed map 求哈希,row 与批次描述再共同哈希进 `graph.rev`。图类型单源在 modules 包的 `./client` 出口——webserver 对图一无所知;modules 会注册 combo 路由并贡献结构化 index 注入行。
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。
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-24-single-harness-home-resolver.md
2026-07-24-single-harness-home-resolver.md: 02578674b4547bae54b2edce99c6c4ef53748588
2026-07-24-single-harness-home-resolver.zh.md: 5764be3f85901196cd2a2db77d1fc3838668163f
2026-07-24-single-harness-home-resolver.md: 1191c52bfa222b807e9aa301b250d2efcc8b0953
2026-07-24-single-harness-home-resolver.zh.md: cefa124b7cb46bab92157af34576f59d3053da9f
@@ -23,6 +23,8 @@ explicit configured path > $DSH_HOME > ~/.dsh
An empty or whitespace-only `$DSH_HOME` is treated as unset; otherwise `resolve('')` would silently place the home at the current working directory. The harness keeps all user data under one root; there is no XDG config/data/cache split. `dshHomePath(...segments)` joins deployment-owned children onto that root, and `dsh-app-boot` exposes it to Loader `!!js` config expressions before mounting entries, so shipped compositions derive `sessions` and `storages` without copying the resolver. `dshHomeDisplay()` names a resolved root symbolically for user-facing paths — `~/.dsh` for the default home, `$DSH_HOME` for any configured home — so the user-global `AGENTS.md` label never leaks an absolute machine path. It replaces agent-instructions's bespoke default-vs-`$DSH_HOME` check.
`dshCachePath(...segments)` derives paths below the resolved home's `cache` directory. An initial `{ dshHome }` option preserves a provider's explicit home override. It resolves paths without creating directories; callers own directory creation. `attachment-local` uses this helper for regenerable request-image variants while retaining durable attachment objects in their versioned storage tree, so clearing the cache cannot remove Session attachments. Existing request-image cache entries are left in place and are not read or copied; a cache miss regenerates the variant from its durable attachment.
`@deepseek-ai/dsh-home` is deleted. Home-owning providers and boot packages import `resolveDshHome` from `dsh-home-paths`; composition bundles contain only the resolved configuration rows.
`dsh-telemetry` and its separate home policy are absent under the [SDK project toolchain removal](../../archived/simplification/2026-08-11-remove-sdk-project-toolchain.md), leaving this resolver as the sole home policy.
@@ -23,6 +23,8 @@ explicit configured path > $DSH_HOME > ~/.dsh
空或仅含空白的 `$DSH_HOME` 被当作未设置处理;否则,`resolve('')` 会悄悄把 home 落在当前工作目录。harness 把所有用户数据都放在同一个根目录下;不存在 XDG 的 config/data/cache 拆分。`dshHomePath(...segments)` 将部署负责的子路径拼接到该根目录下,`dsh-app-boot` 在挂载条目前向 Loader `!!js` 配置表达式暴露它,因此出厂组合无需复制解析器即可派生 `sessions` 和 `storages`。`dshHomeDisplay()` 为面向用户的路径以符号形式命名已解析的根目录——默认 home 显示为 `~/.dsh`,任何已配置的 home 显示为 `$DSH_HOME`——这样用户全局的 `AGENTS.md` 标签就绝不会泄露机器上的绝对路径。它取代了 agent-instructions 中自定义的「默认值 vs `$DSH_HOME`」判断。
`dshCachePath(...segments)` 在解析出的主目录下的 `cache` 目录中派生路径。首个 `{ dshHome }` 选项保留提供方显式配置的主目录覆盖值。它只解析路径,不创建目录;目录创建由调用方负责。`attachment-local` 将此函数用于可重新生成的请求图片版本,持久附件对象仍保留在其带版本的存储树中,因此清空缓存不会删除 Session 附件。已有请求图片缓存条目保留在原处,不再读取或复制;缓存未命中时从持久附件重新生成请求版本。
`@deepseek-ai/dsh-home` 被删除。拥有 home 配置的提供方与 boot 包从 `dsh-home-paths` 导入 `resolveDshHome`;组合包只包含解析后的配置行。
`dsh-telemetry` 及其独立 home 策略已随 [SDK 项目工具链移除](../../archived/simplification/2026-08-11-remove-sdk-project-toolchain.md)一并消失,因此该解析器是唯一的 home 策略。
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-24-web-config-tree-boot-and-transport-layering.md
2026-07-24-web-config-tree-boot-and-transport-layering.md: e7c0781e1504bce12a8b0d2197bd47b37d3e873a
2026-07-24-web-config-tree-boot-and-transport-layering.zh.md: 9a0affa397151a3994c5731bebda4b45a4139b6c
2026-07-24-web-config-tree-boot-and-transport-layering.md: c8a1db3cd7bd715d8912b22a0b41a9ef660c1316
2026-07-24-web-config-tree-boot-and-transport-layering.zh.md: 41fcf65b06829199c203faa806bd17a310b68bcf
@@ -12,7 +12,7 @@ English | [中文](2026-07-24-web-config-tree-boot-and-transport-layering.zh.md)
## Decision
**Composition is one flat assembled tree.** `apps/cli/config/base.cordis.yml` plus `apps/cli/config/web.cordis.yml` holds every row — the host runtime (32 rows), the `api-gateway` row, the `webserver` row, and the `dsh.client` rows (the browser roster; the modules row is simultaneously a host row). No spine bundle: every plugin is one row and every config field is yml-editable. That stance later became repository-wide, with the rows both surfaces share factored into `apps/cli/config/base.cordis.yml` and each surface reduced to an overlay ([shared-base overlays](../../archived/simplification/2026-07-29-shared-base-config-overlays.md)). The `dsh-client-hmr` row is an ordinary always-on bundle row (originally appended in code by `--dev`; the flag is retired). Row order carries no load semantics; activation is service-availability driven. The shared audit rejects imports with no fiber, awaits only failed fibers to recover original activation errors, and reports services that leave a fiber `PENDING`; before throwing, it marks those exact rejection reasons through one process checkpoint so `installFailLoud` coalesces Loader's duplicate notification while unrelated unhandled rejections remain fatal. The Node app-boot artifact embeds `@cordisjs/plugin-include` while leaving `@cordisjs/plugin-loader` external, so the include's `EntryTree` and the host bind to one Loader peer instead of splitting a config tree across two Loader implementations.
**Composition is one flat assembled tree.** `apps/cli/config/base.cordis.yml` plus `apps/cli/config/web.cordis.yml` holds every row — the host runtime (32 rows), the `api-gateway` row, the `webserver` row, and the `dsh.client` rows (the browser roster; the modules row is simultaneously a host row). No spine bundle: every plugin is one row and every config field is yml-editable. That stance later became repository-wide, with the rows both surfaces share factored into `apps/cli/config/base.cordis.yml` and each surface reduced to an overlay ([shared-base overlays](../../archived/simplification/2026-07-29-shared-base-config-overlays.md)). The `dsh-client-hmr` row is an ordinary always-on bundle row (originally appended in code by `--dev`; the flag is retired). Row order carries no load semantics; activation is service-availability driven. `auditStartupEntries` reports failed imports, reads failed fibers for original activation errors, and lists services that leave a fiber `PENDING`. The [startup policy](2026-09-09-consumer-owned-startup-strictness.md) makes optional failures warnings and required failures fatal. Reported rejection reasons stay marked through one process checkpoint so `installFailLoud` coalesces Loader's duplicate notification while unrelated unhandled rejections remain fatal. The Node app-boot artifact embeds `@cordisjs/plugin-include` while leaving `@cordisjs/plugin-loader` external, so the include's `EntryTree` and the host bind to one Loader peer instead of splitting a config tree across two Loader implementations.
**Boot glue is a class pair.** `AppCLIEntry` (apps/cli) and `AppWebEntry` (the shell kernel) hold only what must exist independently of cordis: argv facts, the composed patch set, the parsed boot manifest, the module system instance, loading-page handles — everything else lives in plugins. `AppCLIEntry.run()` is three stages: layered env (ambient > cwd `.env` > `$DSH_HOME/.env`, closing the defect above) → patch composition → Loader include boot plus the activation audit. `AppWebEntry.run()` mirrors it browser-side: parse `window.__DSH_BOOT__` into a `BootManifest` (two views: npm-package rows for the module table, cordis-plugin rows for entry composition; malformed wire throws), build the module system, render the loading page, prefetch the `immediately` tier in parallel with Context/Loader setup, **await the prefetch before creating entries** (materialization is `tree.import`'s synchronous require, unprotected by fiber inject waiting; cross-package require edges such as i18n → runtime/client need every immediately-tier factory registered first — an empirically found 10–25% boot race otherwise), adopt the modules entry, create the graph rows, settle, sweep.
@@ -12,7 +12,7 @@ Status: implemented
## 决策
**组合结果是一棵平铺配置树。** `apps/cli/config/base.cordis.yml` 与 `apps/cli/config/web.cordis.yml` 共同持有全部行——host 运行时(32 行)、`api-gateway` 行、`webserver` 行、`dsh.client` 行(浏览器 roster;modules 行同时是 host 行)。不做主干 bundle:每插件一行、每个 config 字段 yml 可改。这一立场后来推广到全仓:两个 surface 共享的配置项被抽取进 `apps/cli/config/base.cordis.yml`,各 surface 则收敛为一份 overlay([共享 base overlay](../../archived/simplification/2026-07-29-shared-base-config-overlays.md))。`dsh-client-hmr` 行是普通的始终启用的 bundle 行(最初由 `--dev` 在代码中追加;该旗标已废除)。行序无装载语义;激活由服务可用性驱动。共享 audit 会拒绝没有 fiber 的 import、仅等待失败的 fiber 以恢复原始激活错误,并报告让 fiber 停在 `PENDING` 的服务;抛出错误前,审计会通过一个进程级检查点标记这些 rejection 的确切原因,从而让 `installFailLoud` 将 Loader 的重复通知合并为一次,而无关的未处理 rejection 仍然致命。Node app-boot 产物内嵌 `@cordisjs/plugin-include`,但将 `@cordisjs/plugin-loader` 保持为外部依赖,因此 include 的 `EntryTree` 与 host 会绑定到同一个 Loader peer,而不会让一棵配置树横跨两个 Loader 实现。
**组合结果是一棵平铺配置树。** `apps/cli/config/base.cordis.yml` 与 `apps/cli/config/web.cordis.yml` 共同持有全部行——host 运行时(32 行)、`api-gateway` 行、`webserver` 行、`dsh.client` 行(浏览器 roster;modules 行同时是 host 行)。不做主干 bundle:每插件一行、每个 config 字段 yml 可改。这一立场后来推广到全仓:两个 surface 共享的配置项被抽取进 `apps/cli/config/base.cordis.yml`,各 surface 则收敛为一份 overlay([共享 base overlay](../../archived/simplification/2026-07-29-shared-base-config-overlays.md))。`dsh-client-hmr` 行是普通的始终启用的 bundle 行(最初由 `--dev` 在代码中追加;该旗标已废除)。行序无装载语义;激活由服务可用性驱动。`auditStartupEntries` 报告 import 失败、读取失败 fiber 的原始激活错误,并列出让 fiber 停在 `PENDING` 的服务。[启动策略](2026-09-09-consumer-owned-startup-strictness.zh.md)将 optional failure 作为 warning,将 required failure 视为致命错误。已报告的 rejection 原因会保持标记至一个进程级检查点,使 `installFailLoud` 合并 Loader 的重复通知,而无关的未处理 rejection 仍然致命。Node app-boot 产物内嵌 `@cordisjs/plugin-include`,但将 `@cordisjs/plugin-loader` 保持为外部依赖,因此 include 的 `EntryTree` 与 host 会绑定到同一个 Loader peer,而不会让一棵配置树横跨两个 Loader 实现。
**boot 胶水由两个类组成。** `AppCLIEntry`(apps/cli)与 `AppWebEntry`(壳内核)只持有那些必须独立于 cordis、提前存在的东西:argv 事实、合成的 patch 集、解析出的 boot manifest(元数据清单)、模块系统实例、loading 页句柄——其余一律进插件。`AppCLIEntry.run()` 三段:分层 env(ambient > cwd `.env` > `$DSH_HOME/.env`,顺手关掉上述缺陷)→ patch 合成 → Loader include boot 加 activation audit。`AppWebEntry.run()` 在浏览器侧镜像它:把 `window.__DSH_BOOT__` 解析成 `BootManifest`(双视角:npm 包行给模块表、cordis 插件行给 entry 组合;畸形 wire 大声抛)、建模块系统、渲染 loading 页、immediately 层预取与 Context/Loader 准备并行、**create entry 之前等预取齐**(物化是 `tree.import` 的同步 require,不受 fiber inject 等待保护;i18n → runtime/client 这类跨包 require 边要求 immediately 层工厂全部注册完——否则有实测 10–25% 的 boot 竞态)、收编 modules entry、逐一创建图行、settle、sweep。
@@ -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/2026-07-28-portable-execution-world-consumers.md
2026-07-28-portable-execution-world-consumers.md: 787fe341e58cc212c99e0f35f07eea8e83daf000
2026-07-28-portable-execution-world-consumers.zh.md: a558a5af64437b8743e741ace4ccf27079501721
2026-07-28-portable-execution-world-consumers.md: 5a687a552f322f3929b3f3937caf36c3373c05e7
2026-07-28-portable-execution-world-consumers.zh.md: 6bc0e20bd29d0b0a01068c004655cc90ac3b6699
@@ -16,6 +16,8 @@ Ordinary pipes do not cover one requirement. A persistent terminal needs PTY all
`ctx.fs` and `ctx.subprocess` together define one execution world. Providers mounted together must describe the same path namespace, executables, processes, and terminal sessions; higher capabilities consume those two interfaces rather than name the provider.
Foreground Bash and PowerShell calls use one executor deadline for asynchronous confinement preparation and process execution. Preparation expiry returns an outcome without process-exit or signal facts, and late argv cannot start a process; background preparation follows caller cancellation only. The protected execution result tells subclasses whether the subprocess provider was called, so enforcement facts are attached only after publication.
The filesystem interface owns the path facts that another capability needs without exposing its opaque target identity: a canonical process path, canonical `file:` URI, and containment. Existing whole and streaming text operations remain filesystem-owned; protocol consumers enforce their own retention limits while consuming the stream.
The subprocess interface owns executable lookup and process primitives: ordinary raw or collected process spawning and `spawnTerminal()`. An ordinary handle keeps target identity private: `.done` reports the direct target, while `terminate()` and `waitForExit()` control and observe the same provider-managed range. The [native-containment decision](2026-08-28-subprocess-native-containment.md) owns local Linux scopes, Windows Jobs, and their disclosed fallbacks. The terminal operation is one deep primitive whose handle owns text I/O, foreground groups, signalling, and one awaited TERM-to-KILL operation that settles in-flight handle calls and reaches quiescence for every member of its provider-owned range; an observational fallback limits that range to identities it can still observe. Its signal cancels allocation only; the published handle owns its lifetime. Prompt detection, idle inference, scrollback, sandbox policy, and owner lifecycle remain in the PTY consumer.
@@ -26,19 +28,15 @@ Generic consumers use that execution world:
- `dsh-lsp-stdio` reads and contains source through `ctx.fs`, resolves and launches language servers through `ctx.subprocess`, and carries provider-owned file URIs through initialization and result rendering. One provider-lifetime signal aborts filesystem and protocol work during disposal, including workspace lookup before queue ownership; its JSON-RPC, pooling, synchronization, and normalization stay unchanged.
- `dsh-terminal-bash` maps persistent-shell semantics onto `ctx.subprocess.spawnTerminal()`. The local `node-pty` and process-inspection implementation moves into `dsh-subprocess-local`; another subprocess provider supplies the same primitive. `danger-full-access` needs no `ctx.sandbox`; a confined mode requires a same-world sandbox provider and fails before spawn when none is mounted. Prompt and silence evidence collected during asynchronous pre-write inspection is discarded when the provider write begins. Cancellation retains the send reservation while an in-flight write settles and then signals the foreground group, so late bytes or the signal cannot target a successor; an in-flight readiness poll cannot release that reservation, and a rejected write sends no signal. The absolute deadline remains armed throughout cancellation. A signal failure becomes terminal transport failure. Completion of a stale inspection resumes polling for the current send. Startup cancellation begins terminal rollback without waiting for a stalled readiness or signalling call. Close rejects new public signals and delegates provider-managed session quiescence to the handle's awaited termination operation.
## E2B POC boundary
## Remote provider ownership
The opt-in E2B realization has exactly three provider-specific packages under `packages/e2b/`: `dsh-e2b` creates one sandbox and deletes it on timeout or disposal, `dsh-fs-e2b` implements `ctx.fs`, and `dsh-subprocess-e2b` implements `ctx.subprocess` over E2B Commands, PTYs, and remote Linux process groups. The two adapters obtain the sole SDK handle from the owner and never create private sandboxes.
The [E2B provider removal](../simplification/2026-09-11-remove-e2b-providers.md) supersedes the E2B realization of this decision. The filesystem/subprocess agreement and asynchronous terminal contracts remain in force for remote implementations. The [POSIX SSH providers](2026-09-11-posix-ssh-runtime.md) realize them through one installed helper and independent stream channels.
E2B owns the mutable filesystem, managed command and Bash processes, terminal allocation and terminal-session groups, language-server processes and source reads, and adapter-private files under `.dsh-e2b`. The host owns Cordis and plugin objects, the agent loop, agent/session/goal state, session logs and persistence, LLM calls, prompts and tools, authority, skills, subagent orchestration, PTY buffers and readiness, LSP protocol state, and E2B SDK/network buffers. The overlay neither uploads nor synchronizes the host workspace.
The adapters retain only substrate mechanics. Filesystem canonicalization crosses the SDK's decoded command transport as strict base64-encoded NUL framing; streamed reads leave byte ceilings with consumers. Subprocess command output and environment snapshots use ASCII/base64 where SDK chunk decoding would otherwise lose bytes, while private control shells isolate profiles and later launches blank discovered credential-shaped names. Process and terminal cleanup uses remote groups and proves quiescence before settlement.
Sandbox state is deliberately ephemeral: timeout and disposal delete the remote files and unmanaged state. The POC adds no reconnect or pause/leave retention, session-persistence backend, template builder, volume, snapshot, network-policy layer, sandbox catalog, workspace synchronization, durable remote handles, or whole-harness execution.
A remote provider owns mutable files, command and terminal processes, language-server processes, and provider-private runtime files. The host owns Cordis and plugin objects, the agent loop, agent/session/goal state, session logs and persistence, model transport, authority, skills, subagent orchestration, terminal readiness and LSP protocol state. Moving execution does not imply workspace synchronization or durable remote handles.
## Verification
Focused package suites pin sandbox lifecycle, canonical path framing, filesystem metadata and atomic versions, subprocess publication/rollback, terminal text I/O and session cleanup, output limits, cancellation, disposal, and invariant registration. A credential-gated Loader composition exercises the same three-package provider through source imports and built exports, including FS/Bash visibility, hostile login profiles, byte-split UTF-8 output, process and terminal cleanup, LSP queries, host-workspace isolation, and final sandbox deletion.
The local filesystem, subprocess, terminal and LSP suites cover path identity, executable lookup, managed cleanup, output limits, cancellation and disposal. Terminal consumer tests retain delayed provider writes, foreground inspection and signalling to verify asynchronous ownership without requiring a remote service.
## Alternatives considered
@@ -60,14 +58,14 @@ Focused package suites pin sandbox lifecycle, canonical path framing, filesystem
**Implement remote filesystem operations only through shell commands.** Rejected because that discards structured filesystem identity, errors, streaming, version guards, and atomic mutation semantics already consumed by the file tools.
**Add a generic distributed-runtime abstraction or reconnect live handles.** Rejected because the existing capability seams carry the demonstrated contracts, while remote identity alone cannot reconstruct callbacks, pending promises, authority, protocol state, or output cursors. A new layer would speculate about persistence and synchronization beyond the POC.
**Add a generic distributed-runtime abstraction or reconnect live handles.** Rejected because the existing capability seams carry the demonstrated contracts, while remote identity alone cannot reconstruct callbacks, pending promises, authority, protocol state, or output cursors. A new layer would speculate about persistence and synchronization beyond the demonstrated consumer contracts.
## Consequences
A remote execution provider implements only its shared sandbox owner plus filesystem and subprocess adapters. Bash, PTY, and LSP compose above them, so fixes to those capabilities remain provider-neutral.
A remote execution family supplies its shared connection owner and matching filesystem, subprocess and, for confined calls, sandbox providers. Bash, PTY, and LSP compose above them, so fixes to those capabilities remain provider-neutral.
The fundamental interfaces are wider, and a filesystem/subprocess pair must agree on one execution world. The added operations are limited to facts and lifecycle mechanics that current generic consumers require; model schemas, protocol framing, readiness policy, and presentation do not leak into the providers.
The local implementation absorbs `node-pty` and platform process inspection because it owns local terminal mechanics. On supported Linux hosts, the user-systemd scope retains descendants that call `setsid` or reparent, while process inspection continues to own foreground attribution and synchronous fallback evidence. Other hosts use the observational teardown: disposal sweeps descendants before and after terminating the top-level shell, waits for exact PID-identity-fenced descendants retained during foreground inspection, and retains Linux session members that survive top-level exit. macOS cannot enumerate a POSIX session after its leader exits, so a child that reparents between inspection snapshots remains an explicit local-provider limitation rather than a reason to move process mechanics back into the PTY consumer.
The E2B composition demonstrates that a shared sandbox owner plus filesystem and subprocess adapters are sufficient to move the mutable coding world off-host while leaving higher capabilities provider-neutral. Its POC limits remain explicit: the SDK retains complete command transport in host memory, remote startup cannot publish a PID synchronously, exact terminal stdin-wait and independent signal facts are unavailable, numeric PID/PGID operations are not identity-fenced, the initial environment probe cannot hide unknown sandbox-default secrets from already-running same-UID processes, and adapter artifacts remain until sandbox deletion. These are provider constraints, not justification for compatibility shims or more E2B packages.
Remote implementations must preserve the existing consumer contracts or reject unsupported operations explicitly. Transport-specific buffering, identity and disconnect limits belong to the provider; they do not justify duplicate Bash, PTY or LSP implementations.
@@ -16,6 +16,8 @@ Status: implemented
`ctx.fs` 与 `ctx.subprocess` 共同定义一个执行世界。共同挂载的提供方必须描述相同的路径命名空间、可执行文件、进程和终端会话;上层能力消费这两个接口,而不引用具体提供方。
前台 Bash 和 PowerShell 调用由同一执行器 deadline 覆盖异步 confinement 准备与进程执行。准备超时返回没有进程退出或信号事实的结果,晚到的 argv 不能启动进程;后台准备仍只跟随调用方取消。子类通过受保护的执行结果区分是否已调用 subprocess provider,只有发布后才附加 enforcement 事实。
文件系统接口负责其他能力需要的路径事实,同时不公开其不透明目标身份:规范化进程路径、规范化 `file:` URI 和包含关系。现有完整文本与流式文本操作仍归文件系统负责;协议消费方在消费流时执行各自的保留上限。
进程管理接口负责可执行文件查找与进程原语:以原始或收集模式 spawn 普通进程,以及 `spawnTerminal()`。普通句柄把 target identity 保持为私有事实:`.done` 报告 direct target,`terminate()` 与 `waitForExit()` 则控制并观察同一个由提供方管理的范围。[原生 containment 决策](2026-08-28-subprocess-native-containment.zh.md)负责本地 Linux scope、Windows Job 及其已声明的 fallback。终端操作是一项深层原语,其句柄负责文本 I/O、前台进程组、信号发送,以及一项须等待的 TERM→KILL 操作;该操作会结算所有在途句柄调用,并使提供方拥有的范围中每个成员完全停稳;观察型 fallback 只能把该范围限制为它仍可观察到的 identity。其信号只取消分配;句柄一经发布,便负责自身生命周期。提示符检测、空闲推断、scrollback、沙箱策略和所有者生命周期仍由 PTY 消费方负责。
@@ -26,19 +28,15 @@ Status: implemented
- `dsh-lsp-stdio` 通过 `ctx.fs` 读取源文件并验证包含关系,通过 `ctx.subprocess` 解析和启动语言服务器,并让由提供方负责的文件 URI 贯穿初始化与结果渲染。一个提供方生命周期信号会在资源释放期间中止文件系统与协议操作,包括取得队列所有权之前的工作区查找;其 JSON-RPC、池化、同步和规范化保持不变。
- `dsh-terminal-bash` 把持久 shell 语义映射到 `ctx.subprocess.spawnTerminal()`。本地 `node-pty` 与进程检查实现移入 `dsh-subprocess-local`;其他进程管理提供方则提供相同原语。`danger-full-access` 不需要 `ctx.sandbox`;受限模式要求同一执行世界中存在沙箱提供方,未挂载时会在 spawn 前失败。提供方开始写入时,系统会丢弃异步写入前检查期间收集的提示符与静默证据。取消会在在途写入结算期间保留发送预留,随后向前台进程组发送信号,因此延迟字节和该信号都无法落到后续发送;在途就绪检查无法释放该预留,写入被拒绝时也不会发送信号。绝对截止时间会在整个取消期间保持启用。信号发送失败会成为终结性传输失败。陈旧检查完成后,会针对当前发送恢复轮询。启动取消会立即开始终端回滚,而不等待停滞的就绪检查或信号发送调用。关闭操作会拒绝新的公开信号,并把由提供方管理的会话完全停稳委托给句柄上须等待的终止操作。
## E2B POC 边界
## 远程提供方的职责
可选启用的 E2B 实现在 `packages/e2b/` 下恰好只有三个提供方专用包:`dsh-e2b` 创建一个沙箱,并在超时或资源释放时将其删除;`dsh-fs-e2b` 实现 `ctx.fs`;`dsh-subprocess-e2b` 基于 E2B Commands、PTY 和远程 Linux 进程组实现 `ctx.subprocess`。两个适配器都从所有者取得唯一的 SDK 句柄,绝不创建私有沙箱。
[E2B 提供方移除决策](../simplification/2026-09-11-remove-e2b-providers.zh.md)取代本决策中的 E2B 实现。文件系统/子进程约定与异步终端约定继续适用于远程实现。[POSIX SSH 提供方](2026-09-11-posix-ssh-runtime.zh.md)通过一个已安装辅助程序与独立流通道实现这些约定。
E2B 负责可变文件系统、受管命令与 Bash 进程、终端分配与终端会话组、语言服务器进程与源文件读取,以及 `.dsh-e2b` 下的适配器私有文件。宿主负责 Cordis 与插件对象、agent loop(智能体循环)、agent 状态、会话状态与目标状态、会话日志与持久化、LLM(大语言模型)调用、提示词与工具、权限、skill(技能)、subagent 编排、PTY 缓冲区与就绪状态、LSP 协议状态,以及 E2B SDK/网络缓冲区。该叠加层既不上传,也不同步宿主工作区。
适配器只保留执行基底机制。文件系统规范化以严格的 base64 加 NUL 分帧穿过 SDK 已解码的命令传输;流式读取把字节上限留给消费方执行。进程管理命令输出与环境快照采用 ASCII/base64,避免 SDK 分片解码丢失字节;私有控制 shell 隔离 profile,后续启动会把已发现且名称呈凭据特征的环境变量置空。进程与终端清理使用远程进程组,并在结算前证明完全停稳。
沙箱状态有意保持短暂:超时与资源释放会删除远程文件和非托管状态。该 POC 不提供重新连接、pause/leave 保留、会话持久化后端、模板构建器、卷、快照、网络策略层、沙箱目录、工作区同步、持久远程句柄,也不会在其中运行整个 harness。
远程提供方负责可变文件、命令与终端进程、语言服务器进程,以及提供方私有运行时文件。宿主负责 Cordis 与插件对象、智能体循环、智能体/会话/目标状态、会话日志与持久化、模型传输、权限、技能、子智能体编排、终端就绪判断和 LSP 协议状态。移动执行位置不意味着工作区同步或持久远程句柄。
## 验证
聚焦的包测试套件锁定了沙箱生命周期、规范化路径分帧、文件系统元数据与原子版本、进程管理发布/回滚、终端文本 I/O 与会话清理、输出上限、取消、资源释放和不变式注册。一项受凭据门控的 Loader 组合通过源代码导入与构建后导出运行同一套三包提供方组合,其中包括 FS/Bash 可见性、恶意登录 profile、跨字节边界拆分的 UTF-8 输出、进程与终端清理、LSP 查询、宿主工作区隔离,以及最终沙箱删除。
本地文件系统、子进程、终端与 LSP 测试覆盖路径身份、可执行文件查找、受管清理、输出上限、取消与资源释放。终端消费方测试保留延迟的提供方写入、前台检查和信号发送,以便在无需远程服务时验证异步所有权。
## 考虑过的替代方案
@@ -60,14 +58,14 @@ E2B 负责可变文件系统、受管命令与 Bash 进程、终端分配与终
**只通过 shell 命令实现远程文件系统操作。** 不予采纳,因为这会丢弃现有文件工具已消费的结构化文件系统身份、错误、流式输出、版本保护和原子变更语义。
**新增通用分布式运行时抽象,或重新连接活跃句柄。** 不予采纳,因为现有能力 seam 已承载经证实的约定,而仅凭远程身份无法重建回调、待处理 promise、权限、协议状态或输出游标。新增一层只会推测 POC 边界之外的持久化与同步问题。
**新增通用分布式运行时抽象,或重新连接活跃句柄。** 不予采纳,因为现有能力 seam 已承载经证实的约定,而仅凭远程身份无法重建回调、待处理 promise、权限、协议状态或输出游标。新增一层只会推测 已验证消费方约定之外的持久化与同步问题。
## 后果
远程执行提供方只需实现共享沙箱所有者,以及文件系统与进程管理适配器。Bash、PTY 与 LSP 组合在这些适配器之上,因此对这些能力的修复仍与提供方无关。
远端执行家族提供共享连接管理器及配套文件系统、子进程提供方;受限调用还需要沙箱提供方。Bash、PTY 与 LSP 组合在这些适配器之上,因此对这些能力的修复仍与提供方无关。
基础接口更宽,一对文件系统/进程管理提供方必须在同一个执行世界上保持一致。新增操作仅限当前通用消费方所需的事实与生命周期机制;模型 schema、协议分帧、就绪策略和呈现不会渗入提供方。
本地实现承接 `node-pty` 和平台进程检查,因为它负责本地终端机制。在受支持的 Linux 宿主上,user-systemd scope 会保留调用 `setsid` 或发生 reparent 的后代,进程检查则继续负责前台归属与同步 fallback 证据。其他宿主使用观察型拆卸:dispose(资源释放)会在终止顶层 shell 前后清理后代进程,等待前台检查期间保留下来且受精确 PID 身份围栏保护的后代进程,并继续追踪在顶层进程退出后仍存活的 Linux 会话成员。macOS 无法在 POSIX 会话 leader 退出后枚举该会话,因此在两次检查快照之间重新设定父进程的子进程仍是明确的本地提供方限制,而不是把进程机制移回 PTY 消费方的理由。
E2B 组合证明,共享沙箱所有者加上文件系统与进程管理适配器,就足以在保持上层能力与提供方无关的同时,把可变编码世界移出宿主。其 POC 限制仍明确在案:SDK 会把完整命令传输内容保留在宿主内存中;远程启动无法同步发布 PID;无法获得精确的终端 stdin 等待状态与独立信号事实;基于数值 PID/PGID 的操作没有身份围栏;初始环境探测无法向已在运行的同 UID 进程隐藏未知的沙箱默认 secret;适配器产物会一直保留到沙箱删除。这些是提供方限制,不是引入兼容性 shim 或更多 E2B 包的理由。
远程实现必须保留现有消费方约定,或明确拒绝不支持的操作。传输特定的缓冲、身份与断线限制归提供方负责;这些限制不是复制 Bash、PTY 或 LSP 实现的理由。
@@ -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
@@ -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.
@@ -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。
@@ -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
@@ -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.
@@ -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。
@@ -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
@@ -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.
@@ -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。
@@ -1,6 +1,6 @@
# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/simplification/2026-09-06-agent-request-freeze-provenance.md
2026-09-06-agent-request-freeze-provenance.md: 1235a87ea549c6bbd9c53620017cb1d96f8e7cf7
2026-09-06-agent-request-freeze-provenance.zh.md: 357b25b0f252a9423c485b2cc119f1226ec16687
# pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-31-ptc-runtime-python-fd3-protocol.md
2026-07-31-ptc-runtime-python-fd3-protocol.md: 097bcfd55f333a0973084f378a6c97cb630a4bec
2026-07-31-ptc-runtime-python-fd3-protocol.zh.md: ac656e421bd822994c50c33d0dfaf65ae46aa817
@@ -1,16 +1,16 @@
# Agent Note: the code-runtime-python fd-3 frame protocol
# Agent Note: the ptc-runtime-python fd-3 frame protocol
Status: implemented
The CPython code runtime now lives at `packages/experimental/code-runtime-python` (private, npm name `@deepseek-ai/dsh-experimental-code-runtime-python`); promotion to a released package follows the experimental-packages decision.
The CPython PTC runtime lives at `packages/experimental/ptc-runtime-python` and publishes as `@deepseek-ai/dsh-experimental-ptc-runtime-python`; the [publication decision](../process/2026-09-12-publish-all-experimental-packages.md) preserves its experimental status.
English | [中文](2026-07-31-code-runtime-python-fd3-protocol.zh.md)
English | [中文](2026-07-31-ptc-runtime-python-fd3-protocol.zh.md)
## Problem
`@deepseek-ai/dsh-experimental-code-runtime-python` owns the wire protocol intended for a CPython code-runtime provider. Such a provider runs each model program in a fresh `python3 -I` subprocess and bridges binding calls and completion values over the child's fd 3. The host cannot trust that channel: model code has full access to fd 3 and can forge any frame, so every inbound frame is hostile input that the host must validate and rebuild before reading. The protocol also has to carry lossless JSON without the depth limit `JSON.stringify` and `json.dumps` impose, because the seam's `CodeJsonValue` is depth-unbounded.
`@deepseek-ai/dsh-experimental-ptc-runtime-python` owns the wire protocol intended for a CPython ptc-runtime provider. Such a provider runs each model program in a fresh `python3 -I` subprocess and bridges binding calls and completion values over the child's fd 3. The host cannot trust that channel: model code has full access to fd 3 and can forge any frame, so every inbound frame is hostile input that the host must validate and rebuild before reading. The protocol also has to carry lossless JSON without the depth limit `JSON.stringify` and `json.dumps` impose, because the seam's `PtcJsonValue` is depth-unbounded.
The private experimental package contains both the protocol and runtime implementation: `PythonCodeRuntime` (the plugin's default export), the `python3 -I` subprocess path, and the Python-side JSON codec all live in `@deepseek-ai/dsh-experimental-code-runtime-python`. The protocol builds on the [portable identifier seam](../../archived/architecture/2026-07-31-code-runtime-portable-identifier-seam.md).
The experimental package contains both the protocol and runtime implementation: `PythonPtcRuntime` (the plugin's default export), the `python3 -I` subprocess path, and the Python-side JSON codec all live in `@deepseek-ai/dsh-experimental-ptc-runtime-python`. The protocol builds on the [portable identifier seam](../../archived/architecture/2026-07-31-code-runtime-portable-identifier-seam.md).
## Decision
@@ -42,4 +42,4 @@ Frames are JSON-lines on fd 3, one object per line, leaving stdout/stderr free f
Bought: the fd-3 protocol and its hostile-input codec form a self-contained, fully unit-covered layer, with an executing guard against TypeScript/Python field-set drift. The runtime built on it (`bootstrap.py`) consumes the reviewed wire contract.
Cost: the package name denotes a Python runtime family and `src/index.ts` exports the full `PythonCodeRuntime` implementation, so the protocol vocabulary is only one part of the package surface. The mirror e2e compares field names and required/optional status across the two sides but not field types; comparing type declarations across TypeScript and Python has no mechanical equivalent, so review and the runtime's real-subprocess suite retain that responsibility.
Cost: the package name denotes a Python runtime family and `src/index.ts` exports the full `PythonPtcRuntime` implementation, so the protocol vocabulary is only one part of the package surface. The mirror e2e compares field names and required/optional status across the two sides but not field types; comparing type declarations across TypeScript and Python has no mechanical equivalent, so review and the runtime's real-subprocess suite retain that responsibility.
@@ -1,16 +1,16 @@
# Agent Note: the code-runtime-python fd-3 frame protocol
# Agent Note: the ptc-runtime-python fd-3 frame protocol
Status: implemented
CPython 代码运行时现在位于 `packages/experimental/code-runtime-python`(私有,npm 名 `@deepseek-ai/dsh-experimental-code-runtime-python`);提升为发布包遵循 experimental-packages 决策。
CPython PTC 运行时位于 `packages/experimental/ptc-runtime-python`,以 `@deepseek-ai/dsh-experimental-ptc-runtime-python` 名称发布;[发布决策](../process/2026-09-12-publish-all-experimental-packages.zh.md)保留其实验性状态。
[English](2026-07-31-code-runtime-python-fd3-protocol.md) | 中文
[English](2026-07-31-ptc-runtime-python-fd3-protocol.md) | 中文
## Problem
`@deepseek-ai/dsh-experimental-code-runtime-python` 负责供 CPython code-runtime 提供方使用的 wire protocol。这样的提供方会在全新的 `python3 -I` 子进程中运行每个模型程序,并通过子进程 fd 3 桥接 binding 调用与完成值。Host 不能信任这条通道:模型代码可以完全访问 fd 3 并伪造任意帧,因此 host 必须把每个入站帧视为敌意输入,先校验并重建后才能读取。协议还必须承载无深度限制的 lossless JSON,因为 seam 的 `CodeJsonValue` 深度无界,而 `JSON.stringify` 和 `json.dumps` 都有递归深度限制。
`@deepseek-ai/dsh-experimental-ptc-runtime-python` 负责供 CPython ptc-runtime 提供方使用的 wire protocol。这样的提供方会在全新的 `python3 -I` 子进程中运行每个模型程序,并通过子进程 fd 3 桥接 binding 调用与完成值。Host 不能信任这条通道:模型代码可以完全访问 fd 3 并伪造任意帧,因此 host 必须把每个入站帧视为敌意输入,先校验并重建后才能读取。协议还必须承载无深度限制的 lossless JSON,因为 seam 的 `PtcJsonValue` 深度无界,而 `JSON.stringify` 和 `json.dumps` 都有递归深度限制。
这个私有实验包同时包含协议与 runtime 实现:`PythonCodeRuntime`(插件的默认导出)、`python3 -I` 子进程路径与 Python 侧 JSON codec 都在 `@deepseek-ai/dsh-experimental-code-runtime-python` 中。协议建立在[可移植标识符 seam](../../archived/architecture/2026-07-31-code-runtime-portable-identifier-seam.md)之上。
这个实验包同时包含协议与 runtime 实现:`PythonPtcRuntime`(插件的默认导出)、`python3 -I` 子进程路径与 Python 侧 JSON codec 都在 `@deepseek-ai/dsh-experimental-ptc-runtime-python` 中。协议建立在[可移植标识符 seam](../../archived/architecture/2026-07-31-code-runtime-portable-identifier-seam.md)之上。
## Decision
@@ -42,4 +42,4 @@ CPython 代码运行时现在位于 `packages/experimental/code-runtime-python`
收获:fd-3 协议及其敌意输入 codec 构成自包含、unit 全覆盖的一层,并由执行中的 guard 防止 TypeScript/Python 字段集漂移。基于它构建的 runtime(`bootstrap.py`)消费经过评审的 wire contract。
代价:包名表示 Python runtime 家族,而 `src/index.ts` 导出完整的 `PythonCodeRuntime` 实现,协议 vocabulary 只是包表面的一部分。mirror e2e 会比较两侧字段名与必填/可选状态,但不比较字段类型;跨 TypeScript 与 Python 比较类型声明没有机械等价物,因此评审与 runtime 的真实子进程套件继续负责这项检查。
代价:包名表示 Python runtime 家族,而 `src/index.ts` 导出完整的 `PythonPtcRuntime` 实现,协议 vocabulary 只是包表面的一部分。mirror e2e 会比较两侧字段名与必填/可选状态,但不比较字段类型;跨 TypeScript 与 Python 比较类型声明没有机械等价物,因此评审与 runtime 的真实子进程套件继续负责这项检查。
@@ -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/2026-08-03-per-session-agent-presets.md
2026-08-03-per-session-agent-presets.md: 14d568689a662e3e7d1fedf26c22aea6faddd56b
2026-08-03-per-session-agent-presets.zh.md: 9b296fa7a1c2b9164d9e7a0ad676929179a64fef
2026-08-03-per-session-agent-presets.md: 47f2e28c2aac583ce4ca810157e9c7abe58256d5
2026-08-03-per-session-agent-presets.zh.md: 88b643ea710364f18ff7c73a8ca0e10de0fbb44c
@@ -27,13 +27,13 @@ The presets the deployment ships are the directories under `packages/preset/agen
Mounting is per-session by default. Measured cost for a twelve-row composition is ~3ms and ~600KB per session, so isolation is the cheaper default than any sharing scheme, and a preset authored by a user or by an agent then has the smallest possible blast radius. A preset that genuinely owns an expensive singleton opts into sharing with Cordis's own `isolate` vocabulary: a named realm label is process-global, so two subtrees naming the same label resolve one instance.
Which preset an unnamed session gets is a user setting (`agent-presets.default`) layered over the composition's own `default`, which becomes the `base`. Both layers are needed: the composition value is what a deployment ships and must keep working with no settings provider at all, and the setting is what a person changes without editing a `cordis.yml` they may not own.
The `agent-presets` user-settings namespace carries `modeSelectionEnabled` and `default`. `modeSelectionEnabled` defaults to `true`: the existing new-session picker remains present and an unnamed session resolves to the saved user `default`, or the composition's deployment `default` when none exists. The Web Settings toggle changes only that policy: disabling selection temporarily uses the deployment default, while re-enabling it restores the saved user `default`. This is a deliberate exception to the ordinary user-over-composition settings precedence established in #1539: hiding the chooser disables the user's mode-selection policy without deleting its saved value. The Host policy governs every later session whose caller omits a preset; explicitly named presets and existing sessions remain unchanged. The composition value also keeps the package working with no settings provider, while an enabled user override changes later sessions without editing a deployment-owned `cordis.yml`.
## Consequences
**The effective default is read per resolution, never snapshotted.** A cached value would need a `watch` subscription and a reload path to stay honest, and the resolved scope already re-reads a hot-reloaded document. Reading through is also what makes the boundary correct rather than merely cheap: the new value applies to the next session created, and every running session keeps the composition it was built from. That invariant is the same one the session log enforces from the other side — the header records the id a session was CREATED with and an `agent-preset/selected` event records any later blank-session switch, so a reader resolves the pair (`resolveSessionPreset`) and never the header alone: a resume rebuilds the composition its history was produced under rather than the deployment default at resume time, a cold transcript's presenters resolve in that composition's layer, and the gateway rejects an attempt to adopt a live session under a preset other than the one it currently runs. A snapshot would make the two disagree at exactly the moment the setting changes.
**The effective default is read per resolution, never snapshotted.** A cached value would need a `watch` subscription and a reload path to stay honest, and the resolved scope already re-reads a hot-reloaded document. The Host setting itself applies when an unnamed session is resolved afterwards. An explicit Web Settings action additionally routes its accepted effective default through the existing blank-session selection path only when the captured session id is still current and blank; it never recomposes a running session or rewrites that session's history. The session log enforces the same invariant from the other side — the header records the id a session was CREATED with and an `agent-preset/selected` event records any later blank-session switch, so a reader resolves the pair (`resolveSessionPreset`) and never the header alone: a resume rebuilds the composition its history was produced under rather than the deployment default at resume time, a cold transcript's presenters resolve in that composition's layer, and the gateway rejects an attempt to adopt a live session under a preset other than the one it currently runs. A snapshot would make the two disagree at exactly the moment the setting changes.
**A directly-plugged subtree is invisible to the boot audit.** It never links itself to an `Entry`, so it is absent from `ctx.loader.entries()` and `assertEntriesActivated` cannot see it. The mount audits its own rows instead, reading the tree through an `Include` subclass that publishes it.
**A directly-plugged subtree is invisible to the boot audit.** It never links itself to an `Entry`, so it is absent from `ctx.loader.entries()` and `auditStartupEntries` cannot see it. The mount reads its own rows through an `Include` subclass that publishes the tree and requires every enabled row to activate, independently of the [application startup policy](2026-09-09-consumer-owned-startup-strictness.md).
**A preset can only name a group because the app registers one.** Sharing a realm across rows is a `cordis:group` row, and a preset living outside this workspace — the authored ones under the Harness home, which is the point — cannot resolve `@cordisjs/plugin-group` by name: Node's upward `node_modules` walk never reaches the harness from there. `boot()` therefore registers `cordis:group` beside `cordis:include` as a loader builtin, so both load through the ambient module pipeline rather than through the included tree's own specifier resolution. Without it the `isolate` vocabulary above is expressible one row at a time only, and a provider could never be grouped with its consumers.
@@ -61,11 +61,11 @@ Which preset an unnamed session gets is a user setting (`agent-presets.default`)
**A preset's package names must resolve from the harness, not from the preset.** `EntryTree.import()` resolves a row against its own tree's `baseUrl`, which `Include` sets to the composition's directory. That is right for a relative specifier and fatal for a package name: a locally authored preset lives under the user's home, where Node's upward `node_modules` walk never reaches the installed harness, so every `@deepseek-ai/dsh-*` row fails to import and the whole preset is unmountable. The shipped presets hid this — they sit inside the install. The mount records the host composition's base before plugging the subtree and sends bare specifiers there, leaving relative paths resolving from the preset so its own files still travel with it. The real-composition test writing a preset into a temp root is what found it.
**The preset id is model-visible and must be logged.** It determines the tool set and prompt, so a resumed session has to restore the same composition; recording it is a session fact, not runtime state. It rides the session header beside `cwd`, and the summary carries it so a picker shows what a session actually runs rather than the deployment's current default.
**The preset id is model-visible and must be logged.** It determines the tool set and prompt, so a resumed session has to restore the same composition; recording it is a session fact, not runtime state. It rides the session header beside `cwd`, and the summary carries it so the client surfaces show what a session actually runs rather than the deployment's current default.
**A durable header field is not durable until the provider writes it.** `agentPreset` landed on `SessionHeader` with the right rationale and the JSONL provider omitted it; the derived query index also maps header fields explicitly, so a resumed Session came back with no preset and the surfaces that name it fell silent. `summarizeCold` had the same form — it hand-built the cold list row instead of reusing the shared projection. A field declared durable needs a test that crosses a real store, not only the type that declares it.
**The choice belongs to the screen where it still works.** The composer seat spent almost its whole life disabled, since the preset is fixed once a turn has run. It moved to the new-session screen beside the workspace picker, where the pick is *staged*: that screen precedes the session it applies to, and the stage lands when a session becomes current and is still blank — covering both the session a workspace connect creates and the blank one it reuses, which riding `sessions.create` would miss. It is spent on first use, matching the workspace picker beside it. What a running session runs is then a read-only label in its header: a control there would promise a switch the host refuses outright.
**The choice belongs to the screen where it still works.** The control lives on the new-session screen beside the workspace picker, is present under the default-on Host policy, and disappears only after `modeSelectionEnabled` is disabled. Its pick is *staged*: that screen precedes the session it applies to, and the stage lands when a session becomes current and is still blank — covering both the session a workspace connect creates and the blank one it reuses, which riding `sessions.create` would miss. It is spent on first use, matching the workspace picker beside it; hiding the picker discards a stage that has not reached a session and returns the current blank session to the deployment default through the same selection path. What a running or historical session runs remains a read-only label in its header: a control there would promise a switch the host refuses outright.
**A preset multiplies a cost the host was already paying: nothing disposes an agent.** Measured against the shipped compositions with `--expose-gc`, one live agent holds ~0.17 MB on `minimal` and ~1.31 MB on `standard`/`cordis`, mounting in ~38 ms and ~135 ms; the first agent of a process costs ~7 MB more as Node imports the modules, which every later mount then shares. Growth is strictly linear — 10, 30 and 50 agents give the same per-agent delta — and disposal reclaims essentially all of it (50 `standard` agents held 57.8 MB and returned it). So the object graph does not leak; the lifecycle does. `ApiSessionAgentController` discards the `AgentHandle` returned by the registry, `archiveSession` only edits the workspace registry, `AgentRegistry` has no eviction, and the sole disposal site in the host is the JSON-RPC server's own shutdown. A web host therefore retains every session it has touched, at ~1.3 MB each once presets are composed rather than ~0.2 MB before. Note that pruning the mount registry does not help here: it drops records whose fiber `uid` has cleared, and an agent that never dies never clears one.
@@ -27,14 +27,13 @@ Status: implemented
挂载默认按会话进行。实测一份十二行组装每会话约 3ms、约 600KB,因此隔离比任何共享方案都更划算;而由用户或 agent 写出的 preset 也因此拥有尽可能小的影响面。确实自带昂贵单例的 preset,可以用 Cordis 自身的 `isolate` 词汇显式选择共享:命名 realm 的 label 是进程级全局的,因此两棵子树只要写同一个 label 就解析到同一个实例。
未指名 preset 的会话拿到哪一个,是一项用户设置(`agent-presets.default`),叠在组装自身的 `default` 之上——后者成为 `base`。两层都需要:组装里的值是部署交付的东西,在完全没有 settings 提供方时也必须照常工作;而设置是让人不必去改一份可能并不属于自己的 `cordis.yml` 就能调整的东西。
`agent-presets` 用户设置命名空间同时携带 `modeSelectionEnabled` 与 `default`。`modeSelectionEnabled` 默认为 `true`:既有的新建会话选择器保持显示;未指名会话会解析到已保存的用户 `default`,尚未保存时则使用组装中 `default` 指定的部署默认值。Web 设置开关只改变该策略:关闭选择时临时使用部署默认值,再次开启时恢复已保存的用户 `default`。这是对 #1539 所确立“用户值覆盖组装值”这一普通 settings 优先级的有意例外:隐藏选择器会停用用户的模式选择策略,但不会删除其保存值。该 Host 策略适用于此后所有未显式指定 preset 的会话;显式指定及既有会话不受影响。组装值还使本包在没有 settings 提供方时照常工作;选择器开启后,用户可覆盖默认值来改变后续会话,而无需编辑部署所拥有的 `cordis.yml`。
## 后果
**有效默认值在每次解析时读取,绝不保存快照。** 缓存下来就需要一个 `watch` 订阅和一条重载路径才能保持诚实,而解析后的 scope 本来就会重读热重载过的文档。读穿也不只是省事,它让边界本身是对的:新值作用于**下一个新建的会话**,每个运行中的会话保持它被构建时的那份组装。这条不变量正是 session 日志从另一侧执行的同一条——header 记录会话**创建时**的 id,此后空白期的任何切换由 `agent-preset/selected` 事件记录,因此读取方解析的是两者之和(`resolveSessionPreset`)、绝不单看 header:恢复重建的是其历史所产出的那份组装而不是恢复时的部署默认值,冷读记录的 presenter 在那份组装的层里解析,网关也会拒绝把一个活着的会话收编到它当前运行的 preset 以外的 preset 之下。快照会让两者恰好在设置改变的那一刻各说各话。
**有效默认值在每次解析时读取,绝不保存快照。** 缓存下来就需要一个 `watch` 订阅和一条重载路径才能保持诚实,而解析后的 scope 本来就会重读热重载过的文档。Host 设置本身会在此后解析未指名会话时生效。Web Settings 中的明确操作还会把已接受的有效默认值送入既有的空白会话选择链路,但只在操作前捕获的会话 id 仍是当前空白会话时对齐;它绝不会重新组装运行中的会话,也不会改写该会话的历史。session 日志从另一侧执行同一条不变量——header 记录会话**创建时**的 id,此后空白期的任何切换由 `agent-preset/selected` 事件记录,因此读取方解析的是两者之和(`resolveSessionPreset`)、绝不单看 header:恢复重建的是其历史所产出的那份组装而不是恢复时的部署默认值,冷读记录的 presenter 在那份组装的层里解析,网关也会拒绝把一个活着的会话收编到它当前运行的 preset 以外的 preset 之下。快照会让两者恰好在设置改变的那一刻各说各话。
**直接挂载的子树对启动审计不可见。** 它不会把自己关联到 `Entry`,因此不在 `ctx.loader.entries()` 中,`assertEntriesActivated` 也看不到它。改由挂载过程自行校验各行,通过一个会公开自身 tree 的 `Include` 子类读取。
**直接挂载的子树对启动审计不可见。** 它不会把自己关联到 `Entry`,因此不在 `ctx.loader.entries()` 中,`auditStartupEntries` 也看不到它。挂载过程通过一个会公开自身 tree 的 `Include` 子类读取各行,并要求每个启用行都激活,不受[应用启动策略](2026-09-09-consumer-owned-startup-strictness.zh.md)影响。
**preset 能写出 group,是因为 app 注册了它。** 跨行共享 realm 就是一个 `cordis:group` 行,而住在本工作区之外的 preset——也就是 Harness home 下由人或 agent 创作的那些,正是这套设计的目的——无法按名字解析 `@cordisjs/plugin-group`:Node 向上查找 `node_modules` 的路径从那里永远走不到 harness。因此 `boot()` 把 `cordis:group` 与 `cordis:include` 并排注册为 loader builtin,两者都经由环境模块管线加载,而不依赖被包含树自身的说明符解析。没有它,上文那套 `isolate` 词汇就只能一行一行地表达,提供方也永远无法与它的消费方归入同一组。
@@ -62,11 +61,11 @@ Status: implemented
**preset 的包名必须从 harness 解析,而非从 preset 解析。** `EntryTree.import()` 按行所属树的 `baseUrl` 解析,而 `Include` 把它设为组装文件所在的目录。这对相对标识符是对的,对包名却是致命的:本地创作的 preset 位于用户主目录之下,Node 向上查找 `node_modules` 永远够不到已安装的 harness,因此每一个 `@deepseek-ai/dsh-*` 行都会导入失败,整个 preset 无法挂载。随部署提供的 preset 掩盖了这一点——它们本就在安装目录之内。挂载在插入子树之前先记录宿主组装的基址,并把裸标识符送往那里,同时让相对路径继续从 preset 解析,使它自带的文件仍随它一同迁移。发现它的正是那个把 preset 写入临时根目录的真实组装测试。
**preset id 对模型可见,必须写入日志。** 它决定工具集与提示词,因此被恢复的会话必须还原同一份组装;记录它属于会话事实,而非运行时状态。它与 `cwd` 并列写在会话头部,并由会话摘要携带,使选择器显示的是某个会话实际运行的 preset,而非部署当前的默认值。
**preset id 对模型可见,必须写入日志。** 它决定工具集与提示词,因此被恢复的会话必须还原同一份组装;记录它属于会话事实,而非运行时状态。它与 `cwd` 并列写在会话头部,并由会话摘要携带,使客户端界面显示某个会话实际运行的 preset,而非部署当前的默认值。
**持久化 header 字段在 provider 写入前都算不上持久。** `agentPreset` 带着正确理由落在 `SessionHeader` 上,而 JSONL provider 遗漏了它;派生 query index 也显式映射 header 字段,于是恢复后的 Session 没有 preset,所有据以命名它的 surface 随之失声。`summarizeCold` 是同一种形式——它手工拼装 cold list row,而没有复用共享 projection。声明为持久的字段,需要一个跨越真实 store 的测试,而不只是声明它的类型。
**这个选择属于它仍然可用的那个界面。** composer 座位几乎一生都处于禁用状态,因为一旦跑过一个轮次,preset 即固定。它移到了新建会话界面、工作区选择器旁边,选择在那里是**暂存**的:该界面先于它要应用到的会话存在,暂存值在某个会话成为当前会话且仍为空白时落地——这既覆盖工作区连接新建的会话,也覆盖它复用的那个空白会话,而搭 `sessions.create` 的便车会漏掉后者。它一经使用即被清空,与旁边的工作区选择器一致。至于运行中的会话在跑什么,则是其标题旁的一个只读标签:在那里放控件,等于承诺一次宿主会断然拒绝的切换。
**这个选择属于它仍然可用的那个界面。** 控件位于新建会话界面、工作区选择器旁边,在 Host 默认开启策略下直接显示,仅在 `modeSelectionEnabled` 关闭后隐藏。选择在那里是**暂存**的:该界面先于它要应用到的会话存在,暂存值在某个会话成为当前会话且仍为空白时落地——这既覆盖工作区连接新建的会话,也覆盖它复用的那个空白会话,而搭 `sessions.create` 的便车会漏掉后者。它一经使用即被清空,与旁边的工作区选择器一致;隐藏选择器会丢弃尚未到达会话的暂存选择,并通过同一条选择链路把当前空白会话带回部署默认值。至于运行中或历史会话在跑什么,仍由其标题旁的只读标签展示:在那里放控件,等于承诺一次宿主会断然拒绝的切换。
**preset 放大的是宿主本来就在付的代价:没有任何东西会 dispose 一个 agent。** 用 `--expose-gc` 对随附组装实测:一个存活的 agent 在 `minimal` 上约占 0.17 MB、在 `standard`/`cordis` 上约 1.31 MB,挂载耗时分别约 38 ms 与 135 ms;进程里第一个 agent 另需约 7 MB,那是 Node 首次 import 模块的一次性成本,此后每次挂载共享。增长严格线性——10、30、50 个的单个增量一致——且 dispose 后基本全额回收(50 个 `standard` 占住 57.8 MB,释放后全部归还)。所以对象图并不泄漏,缺的是生命周期。`ApiSessionAgentController` 会丢弃注册表返回的 `AgentHandle`,`archiveSession` 只改工作区注册表,`AgentRegistry` 没有驱逐机制,而宿主里唯一一处 dispose 是 JSON-RPC 服务器自身的关停。于是一个 web 宿主会留住它接触过的每一个会话,组装 preset 之后每个约 1.3 MB,而在此之前约 0.2 MB。注意:剪枝挂载注册表在这里没有用——它丢弃的是 fiber `uid` 已清空的记录,而永不死亡的 agent 永远不会清空它。
@@ -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/2026-08-09-client-conversation-node-assembly.md
2026-08-09-client-conversation-node-assembly.md: 842dfb0218478591f975c97064f101a35ea2f211
2026-08-09-client-conversation-node-assembly.zh.md: 6ab64c25957d33487461fe56e122faedb19f9421
2026-08-09-client-conversation-node-assembly.md: 43f9f0359822f7d7c5b52c09646fd3ffbf80f783
2026-08-09-client-conversation-node-assembly.zh.md: d7475f5ff8218419743448d455d5d5ffe3cc4091
@@ -263,7 +263,7 @@ Page size, record packing, the number of history loads, and RAF coalescing affec
| Message / `input-message` | Message ID | Append-surface `user/message` | None | Use source for a context message, or read the nearest next-step Inbox to distinguish user from steering |
| Request Prompt / `request-prompt` | Header Event seq | Each `request/header` | None | Read the preceding Request Prompt through Reader, retain the full prompt state, and classify system/tool changes |
| Assistant / `assistant-step` | `turn:step` | `step/start` | Live `assistant/live-chunk`, durable `assistant/message` or `assistant/attempt`, and same-step Retry | Aggregate blocks, usage, first-token time, settlement evidence, and retry-hidden state, then publish same-key Step data |
| Tool / `tool-call` | Root call ID | Root `tool/call` | Root result and Code Dispatch start/result | Aggregate the root, children, and parent Map; Dispatch Events route exactly through `rootCallId` |
| Tool / `tool-call` | Root call ID | Root `tool/call` | Root result and PTC dispatch start/result | Aggregate the root, children, and parent Map; Dispatch Events route exactly through `rootCallId` |
| Command / `command` | Command ID | `command/run` | `command/done` and compact lifecycle/checkpoint Events carrying a source command ID | Aggregate command outcome and manual-compaction evidence |
| Automatic Compaction / `compaction` | Compaction ID | `compaction/start` without a source command ID | Summary, end, and replacement checkpoint | Aggregate summary/checkpoint; sufficient checkpoint evidence supports fallback without a start |
| Retry / `model-retry` | Retry ID | Attempt 1 `llm/retry` | Later `llm/retry` and `llm/retry-started` | Aggregate one RetryId's attempts and scheduled/started state |
@@ -360,7 +360,7 @@ SessionEventLike window
Runtime tests pin Definition lifecycle registration, exact-ID append, update-before-start collection followed by forward replay after start, prepend identity, Reader window-gap repair, transitive dependencies, Location closure, Step→Turn data phase order, Location data replacement, publication cadence, illegal withdrawal, first-subscription activation, monotonic active targets, and per-target Builders.
Conversation tests cover every built-in Chat Definition, Assistant Step data, Turn Tail and Deliverables Turn data, Chat ordering and structural sharing, selector isolation, Assistant and Tool running-to-settled identity, nested Code Dispatch, steering, Compaction, Retry, interruption, load-older anchoring, and slot dispatch. Trajectory tests cover its independently registered Message, Assistant, Tool, Compaction, Request-header, and boundary Definitions together with the preserved stage-oriented view model.
Conversation tests cover every built-in Chat Definition, Assistant Step data, Turn Tail and Deliverables Turn data, Chat ordering and structural sharing, selector isolation, Assistant and Tool running-to-settled identity, nested PTC dispatch, steering, Compaction, Retry, interruption, load-older anchoring, and slot dispatch. Trajectory tests cover its independently registered Message, Assistant, Tool, Compaction, Request-header, and boundary Definitions together with the preserved stage-oriented view model.
Slot type/runtime tests pin required parent-provided common inject, the `hookContext` type, Hook isolation across Node contexts, stable factory/Hook identity, and the absence of business-renderer rerenders for unrelated Session publications. Existing entry-owned Observable Hook tests continue to pin the path that does not use a contextual factory.
@@ -263,7 +263,7 @@ Chat `order` 的结构性变化仍可能重排当前可见 key;纯 data 更新
| Message / `input-message` | message ID | append-surface `user/message` | 无 | 根据 source 生成 context message,或读取最近 next-step Inbox 判断 user/steering |
| Request Prompt / `request-prompt` | header Event seq | 每条 `request/header` | 无 | 通过 Reader 读取前一条 Request Prompt,保留完整 prompt 状态,并判定 system/tool 变化 |
| Assistant / `assistant-step` | `turn:step` | `step/start` | Live `assistant/live-chunk`、持久 `assistant/message` 或 `assistant/attempt`、同 step Retry | 聚合 block、usage、首 token 时间、settlement 证据与 retry-hidden state,再发布同 key Step data |
| Tool / `tool-call` | root call ID | root `tool/call` | root result、Code Dispatch start/result | 聚合 root、children 和 parent Map;Dispatch Event 用 `rootCallId` 精确路由 |
| Tool / `tool-call` | root call ID | root `tool/call` | root result、PTC dispatch start/result | 聚合 root、children 和 parent Map;Dispatch Event 用 `rootCallId` 精确路由 |
| Command / `command` | command ID | `command/run` | `command/done`、带 source command ID 的 compact lifecycle/checkpoint | 聚合 command outcome 和手动压缩证据 |
| Automatic Compaction / `compaction` | compaction ID | 无 source command ID 的 `compaction/start` | summary、end、replacement checkpoint | 聚合 summary/checkpoint;checkpoint 足够时可在缺 start 下 fallback |
| Retry / `model-retry` | retry ID | attempt 1 的 `llm/retry` | 后续 `llm/retry` 与 `llm/retry-started` | 聚合同一 RetryId 的 attempts 与 scheduled/started 状态 |
@@ -360,7 +360,7 @@ SessionEventLike window
Runtime tests 固定 Definition 生命周期注册、exact-ID append、update-before-start 收集与 start 后正序 replay、prepend identity、Reader window-gap 修复、传递依赖、Location closure、Step→Turn data phase order、Location data replacement、publication cadence、非法撤回、首次订阅 activation、单调 active target 和 per-target Builder。
Conversation tests 覆盖全部内建 Chat Definition、Assistant Step data、Turn Tail 与 Deliverables Turn data、Chat 排序和结构共享、selector isolation、Assistant/Tool running-to-settled identity、nested Code Dispatch、steering、Compaction、Retry、interruption、load-older anchoring 和 slot dispatch。Trajectory tests 则覆盖它独立注册的 Message、Assistant、Tool、Compaction、Request-header 与 boundary Definition,以及继续保留的 stage-oriented view model。
Conversation tests 覆盖全部内建 Chat Definition、Assistant Step data、Turn Tail 与 Deliverables Turn data、Chat 排序和结构共享、selector isolation、Assistant/Tool running-to-settled identity、nested PTC dispatch、steering、Compaction、Retry、interruption、load-older anchoring 和 slot dispatch。Trajectory tests 则覆盖它独立注册的 Message、Assistant、Tool、Compaction、Request-header 与 boundary Definition,以及继续保留的 stage-oriented view model。
Slot type/runtime tests 固定父注册必须提供声明的 common inject、`hookContext` 类型、不同 Node context 的 Hook 隔离、factory/Hook identity 稳定,以及无关 Session publication 不重渲染业务 renderer。原 entry-owned Observable Hook 测试继续固定未使用 contextual factory 的路径。
@@ -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
@@ -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.
@@ -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。
@@ -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-experimental-agent-teams-packages.md
2026-08-18-experimental-agent-teams-packages.md: 8ccbd690882cac0a4dc844d253656681e600fb74
2026-08-18-experimental-agent-teams-packages.zh.md: dd79d8f2171b545977b7b7e776463da0087724bc
2026-08-18-experimental-agent-teams-packages.md: 2a00b651e0073434a5a68df13b9716adcab5fccf
2026-08-18-experimental-agent-teams-packages.zh.md: df73ee5c05fd3a25bf843bee10b06535a1512783
@@ -8,13 +8,13 @@ English | [中文](2026-08-18-experimental-agent-teams-packages.zh.md)
Agent Teams needs the real Session log, subagent lifecycle, tools, examples, snapshots, and repository checks while its service and tool contracts continue to change. Users also need to install the complete Team composition from npm without building a source checkout.
Moving the packages into product-role groups would remove their experimental names and imply stable-package ownership. Publishing every package under `packages/experimental/` would instead expose unrelated internal prototypes. The release policy needs an explicit Agent Teams exception while preserving the private default.
Moving the packages into product-role groups would remove their experimental names and imply stable-package ownership. Publishing every package under `packages/experimental/` would instead expose unrelated internal prototypes. The release policy must let users install Agent Teams while keeping internal-only prototypes private.
## Decision
`packages/experimental/agent-team`, `packages/experimental/tool-agent-team`, `packages/experimental/agent-team-profile`, `packages/experimental/client-ui-agent-team`, and `packages/experimental/agent-team-web-profile` are public workspace packages. They retain their existing `@deepseek-ai/dsh-experimental-*` names and join the dsh release family. The [experimental package rules](../../../../packages/experimental/AGENTS.md) own the private default, this exception, and later promotion.
`packages/experimental/agent-team`, `packages/experimental/tool-agent-team`, `packages/experimental/agent-team-profile`, `packages/experimental/client-ui-agent-team`, and `packages/experimental/agent-team-web-profile` are public workspace packages. They retain their existing `@deepseek-ai/dsh-experimental-*` names and join the dsh release family. The [publication denylist decision](../process/2026-09-12-experimental-publication-denylist.md) owns the public default and private exceptions; the [experimental package rules](../../../../packages/experimental/AGENTS.md) own dependency isolation and later promotion.
The dsh pack and publish set and the local baseline publisher include exactly these five experimental package directories. Workspace constraints require them to omit `private`, set `publishConfig.access` to `public`, and keep the experimental npm prefix. Every other experimental package remains private and excluded from publication by default. Release packages and apps outside the experimental group, plus the Python runtime, cannot name experimental packages in `dependencies`, `optionalDependencies`, or `peerDependencies`; experimental packages may depend on release packages and each other.
The dsh pack and publish set and the local baseline publisher include these five Agent Teams directories and the [Cua Driver providers](2026-09-12-computer-use-provider-registration.md). Workspace constraints require them to omit `private`, set `publishConfig.access` to `public`, and keep the experimental npm prefix. Release packages and apps outside the experimental group, plus the Python runtime, cannot name experimental packages in `dependencies`, `optionalDependencies`, or `peerDependencies`; experimental packages may depend on release packages and each other.
The generic caller-reserved continuable child identity and selective direct-child drain remain in the stable Subagent service. They own Subagent identity and Activation lifecycle without importing or naming Agent Teams; the experimental Team service consumes them in the permitted direction.
@@ -38,4 +38,4 @@ Experimental status changes compatibility and support expectations, not publicat
Agent Teams publishes as five installable tarballs in the dsh release family without changing package names or enabling Team in a shipped profile. Public availability does not make the packages stable or supported by default, and stable release packages cannot take runtime dependencies on them.
The release family carries explicitly named experimental exceptions. Promotion still creates path and npm-name churn as specified by the experimental package rules.
The release family carries the experimental npm names. Promotion still creates path and npm-name churn as specified by the experimental package rules.
@@ -8,13 +8,13 @@ Status: implemented
Agent Teams 的服务与工具约定仍在变化,但它需要使用真实 Session 日志、subagent 生命周期、工具、示例、快照和仓库检查。用户还需要直接从 npm 安装完整 Team 组合,而无需构建源码 checkout。
把这些包移入产品职责组会移除实验性名称,并暗示稳定包 owner 已经就位。发布 `packages/experimental/` 下的所有包又会暴露无关的内部原型。发布策略需要为 Agent Teams 设置显式例外,同时保留默认私有原则。
把这些包移入产品职责组会移除实验性名称,并暗示稳定包 owner 已经就位。发布 `packages/experimental/` 下的所有包又会暴露无关的内部原型。发布策略必须让用户安装 Agent Teams,同时让内部专用原型保持私有。
## 决策
`packages/experimental/agent-team`、`packages/experimental/tool-agent-team`、`packages/experimental/agent-team-profile`、`packages/experimental/client-ui-agent-team` 与 `packages/experimental/agent-team-web-profile` 是公开 workspace 包。它们保留现有 `@deepseek-ai/dsh-experimental-*` 名称并加入 dsh 发布系列。[实验性包规则](../../../../packages/experimental/AGENTS.md)负责默认私有原则、本例外与后续 promotion。
`packages/experimental/agent-team`、`packages/experimental/tool-agent-team`、`packages/experimental/agent-team-profile`、`packages/experimental/client-ui-agent-team` 与 `packages/experimental/agent-team-web-profile` 是公开 workspace 包。它们保留现有 `@deepseek-ai/dsh-experimental-*` 名称并加入 dsh 发布系列。[发布拒绝列表决策](../process/2026-09-12-experimental-publication-denylist.zh.md)负责默认公开与私有例外;[实验性包规则](../../../../packages/experimental/AGENTS.md)负责依赖隔离与后续 promotion。
dsh pack 与 publish 集合以及本地 baseline 发布器只会纳入这五个实验性包目录。workspace 约束要求它们省略 `private`、设置 `publishConfig.access` 为 `public`,并保留实验性 npm 前缀。其他实验性包默认仍为私有且不发布。实验组外的发布包与 app 以及 Python runtime 不得通过 `dependencies`、`optionalDependencies` 或 `peerDependencies` 引用实验性包;实验性包可以依赖发布包和其他实验性包。
dsh 打包与发布集合以及本地基线发布器包含这五个 Agent Teams 目录和 [Cua Driver 提供方](2026-09-12-computer-use-provider-registration.zh.md)。workspace 约束要求它们省略 `private`、将 `publishConfig.access` 设为 `public`,并保留实验性 npm 前缀。实验组之外的发布包、应用和 Python 运行时不能在 `dependencies`、`optionalDependencies` 或 `peerDependencies` 中引用实验包;实验包可以依赖发布包和彼此。
通用的调用方预留 continuable child 身份和精确 direct-child drain 仍属于稳定 Subagent 服务。它们负责 Subagent 身份与 Activation 生命周期,不 import 或命名 Agent Teams;实验性 Team 服务沿允许的方向消费这些能力。
@@ -38,4 +38,4 @@ profile 安装通过自身 package manager 解析每个公开 bundle 及其依
Agent Teams 会作为 dsh 发布系列中的五个可安装 tarball 发布,同时保持包名不变,也不会在随附 profile 中启用 Team。公开可用不代表这些包稳定或默认受支持,稳定发布包也不能对其建立运行时依赖。
发布系列需要维护显式命名的实验性例外。promotion 仍会按照实验性包规则产生路径和 npm 名改动。
发布系列保留实验性 npm 名称。promotion 仍会按照实验性包规则产生路径和 npm 名改动。
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/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
@@ -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.
@@ -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。
@@ -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/2026-08-19-session-projection-mandatory-seam.md
2026-08-19-session-projection-mandatory-seam.md: 371faa254809649e93685a8f3ada3e640d5f391e
2026-08-19-session-projection-mandatory-seam.md: e1c6835311228fcaf7a22140f1db8da748603186
2026-08-19-session-projection-mandatory-seam.zh.md: 5d6a5169a556febae3f5c960b2d519c324d99863
@@ -25,7 +25,7 @@ The registry provides `stateOf(session, key)` for one typed host state and keeps
- **Default missing projected state.** This preserves more partial compositions but makes missing host state indistinguishable from a valid empty value. Rejected because official profiles mount the registry and configuration errors must fail explicitly.
- **Require every contributor at activation.** This makes the key set uniform but unnecessarily couples contribution lifecycle to service activation. Explicit first-access failure preserves the optional registration form without allowing silent degradation.
- **Use `snapshot()` for every read.** This keeps one method but computes unrelated wire views and encourages consumers to depend on batch transport data for host logic. Rejected in favor of typed single-key state reads.
- **Send full host values to clients.** This avoids separate view types but exposes provenance and policy knobs that no client consumes. Rejected in favor of explicit cropped views.
- **Send full host values to clients.** This avoids separate view types but exposes producer identities and policy knobs that no client consumes. Rejected in favor of explicit cropped views.
- **Broadcast registry additions and removals across Host and mux streams.** The streams have no shared ordering, so clients need tombstones, buffered frames, and baseline retries to reconcile them. Rejected because plugin-key churn does not justify a second synchronization protocol.
## Consequences
@@ -2,5 +2,5 @@
# side as of the last confirmed-consistent state. Both languages carry equal authority;
# after editing either side, bring the other along and re-record with:
# pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-08-21-deepseek-llm-api-request-extensions.md
2026-08-21-deepseek-llm-api-request-extensions.md: eadbe2a5f17de446c345120f9a6aeeeb531c43b4
2026-08-21-deepseek-llm-api-request-extensions.zh.md: f45210adbc075c30484754ce7f8c2b65b515be6c
2026-08-21-deepseek-llm-api-request-extensions.md: a0e38c1da5af32b0ba5e07df632ac4e25148419e
2026-08-21-deepseek-llm-api-request-extensions.zh.md: 763f418c7bd6029c08027cd5f4705bffd984b9ee
@@ -30,7 +30,7 @@ The `events` array contains complete canonical `SessionEvent` objects directly.
`@deepseek-ai/dsh-plugin-package-inventory-deepseek` owns the default-on `dsh_plugin_packages` field from the `llm` package family. It reads active non-group entries from the host Loader tree and, for a live requesting Agent, its standing preset tree. Node package resolution locates the owning manifest without requiring a `./package.json` export. Ordinary entries resolve from their owning tree, while a standing preset root mirrors its Loader's intentional harness-base override and nested includes retain their own bases. An anonymous nearest manifest marks a loose module; a named manifest must carry a version. Exact name/version pairs are deduplicated with deterministic ordering; simultaneously active versions remain separate.
Disabled, pending, failed, unloading, disposed, structural, loose non-package, ordinary dependency, programmatic child-fiber, and in-memory dynamic-plugin entries are outside this package inventory. This definition reports package-backed composition facts the runtime can prove instead of inventing provenance for arbitrary callbacks.
Disabled, pending, failed, unloading, disposed, structural, loose non-package, ordinary dependency, programmatic child-fiber, and in-memory dynamic-plugin entries are outside this package inventory. This definition reports package-backed composition facts the runtime can prove instead of inventing package identity for arbitrary callbacks.
## Deferred inventory caching
@@ -73,11 +73,11 @@ The receiver would also need to traverse the tagged tree, resolve paths into the
### Why not omit assistant chunks or overlapping event data?
About 98% of the measured v1 real-session events were `assistant/chunk`. Omitting them after reference encoding reduced the complete identity JSON by another 84.79% for late enable and 6.49% for steady state, but prevented lossless reconstruction and left message provenance dangling. V2 embeds compact streams in attempt settlements; `dsh_session_log` still sends every current canonical event whole and does not omit those embedded records. Fuzzy or normalized substitutions have the same reconstruction defect.
About 98% of the measured v1 real-session events were `assistant/chunk`. Omitting them after reference encoding reduced the complete identity JSON by another 84.79% for late enable and 6.49% for steady state, but prevented lossless reconstruction and left message source-event references dangling. V2 embeds compact streams in attempt settlements; `dsh_session_log` still sends every current canonical event whole and does not omit those embedded records. Fuzzy or normalized substitutions have the same reconstruction defect.
**Keep the upload cursor only in memory.** Rejected because a normal process restart would resend the entire Session. A canonical acceptance event makes restart recovery best-effort durable without another storage backend; the remaining crash window produces allowed duplicates.
**Inventory every live Cordis fiber.** Rejected because programmatic and in-memory fibers have no authoritative npm package provenance. Loader-backed host and preset entries provide exact resolvable package identity.
**Inventory every live Cordis fiber.** Rejected because programmatic and in-memory fibers have no authoritative npm package identity. Loader-backed host and preset entries provide exact resolvable package identity.
**Cache one process-global list or expire it on a TTL.** Rejected because one immutable list is incorrect for Loader lifecycle and per-Session presets, while a TTL permits stale metadata between expiry boundaries. The deferred epoch design invalidates on the authoritative active-state transition instead.
@@ -73,7 +73,7 @@ Status: implemented
### 为什么不省略 assistant 分片或重叠事件数据?
实测 v1 真实 Session event 中约 98% 为 `assistant/chunk`。在引用编码后省略它们,会让完整 identity JSON 在延迟启用场景进一步减少 84.79%,在稳态场景进一步减少 6.49%,但会阻止无损重建并让 message provenance 悬空。V2 把紧凑 stream 嵌入 attempt settlement;`dsh_session_log` 仍会完整发送每个当前规范 event,且不会省略这些嵌入式 record。模糊或规范化替换也有相同重建缺陷。
实测 v1 真实 Session event 中约 98% 为 `assistant/chunk`。在引用编码后省略它们,会让完整 identity JSON 在延迟启用场景进一步减少 84.79%,在稳态场景进一步减少 6.49%,但会阻止无损重建并让 message source-event reference 悬空。V2 把紧凑 stream 嵌入 attempt settlement;`dsh_session_log` 仍会完整发送每个当前规范 event,且不会省略这些嵌入式 record。模糊或规范化替换也有相同重建缺陷。
**只在内存中保留上传游标。** 已否决,因为普通进程重启会重发完整会话。权威接受事件让重启恢复获得尽力而为的持久性,无需另一存储后端;剩余崩溃窗口只会产生允许的重复。
@@ -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: 6b19dc881d572bfece345cbbd5eca688b2e05aab
2026-08-23-client-derived-tool-presentation.zh.md: 98ada31627398f317e4c41336af56e8c6cf0837e
2026-08-23-client-derived-tool-presentation.md: 703960cb24f90eb678fda381f929efb3f94bbd28
2026-08-23-client-derived-tool-presentation.zh.md: 7f161379553f3aca33c72c2edf93e114d27eb7f9
@@ -12,13 +12,13 @@ A `tool/result` does not repeat the tool name or arguments. Host-side result pre
Host projection would also duplicate structured data. Read, diff, search, and web results already persist bounded facts in `tool/result.data.meta`; another view object increases Remote payload size and Client decoding without adding durable meaning.
The Client already owns a complete tool-presentation entry point. `ui-chat` assembles `tool/call`, `tool/result`, and Code Dispatch events into stable `ToolCallBlock` values. `ui-tool` owns the recursive call tree, the `tool.call.toolview` keyed slot dispatched by tool name, the Generic fallback, card models, and details output. A business Client plugin can register a renderer for its own tool names.
The Client already owns a complete tool-presentation entry point. `ui-chat` assembles `tool/call`, `tool/result`, and PTC dispatch events into stable `ToolCallBlock` values. `ui-tool` owns the recursive call tree, the `tool.call.toolview` keyed slot dispatched by tool name, the Generic fallback, card models, and details output. A business Client plugin can register a renderer for its own tool names.
Splitting presentation between Host presenters and Client keyed renderers creates two interpretations of the same event. The keyed renderer is the Web extension point, so an intermediate Host view provides no independent Web capability.
`ToolDefinition.presentCall` and `presentResult` remain useful Host APIs even though ACP is automation-only and the repository has no production TUI consumer. Removing their definitions is a separate decision from keeping Session reads independent of presentation.
The required result is one raw Session journal and one Client presentation owner without visual degradation or incidental enhancement. Specialized cards, interactions, and Code Dispatch topology remain stable while the transport stops carrying transient views.
The required result is one raw Session journal and one Client presentation owner without visual degradation or incidental enhancement. Specialized cards, interactions, and PTC dispatch topology remain stable while the transport stops carrying transient views.
## Decision
@@ -26,7 +26,7 @@ The visual-equivalence requirements below exclude the separately approved [neste
The Session Remote journal sends only raw, validated, persistable Session events. `session.page` and `session.follow` do not parse tool arguments, query the Tools registry, restore a presenter scope, execute `presentCall` or `presentResult`, or construct or clone any tool view.
The Client Conversation layer continues to own tool call/result identity, pairing, lifecycle, Code Dispatch topology, and stable Chat Nodes. It does not interpret individual tool names or produce terminal, diff, read, search, or web component props.
The Client Conversation layer continues to own tool call/result identity, pairing, lifecycle, PTC dispatch topology, and stable Chat Nodes. It does not interpret individual tool names or produce terminal, diff, read, search, or web component props.
Client `ui-tool` continues to own card models and concrete renderers. Each card model directly reads the tool name, raw arguments, result content, error, durable metadata, Session cwd, and Host home from `ToolCallBlock`, and produces the same component props as the current page.
@@ -51,7 +51,7 @@ The Host `ToolDefinition.presentCall`, `ToolDefinition.presentResult`, `ToolCall
| Retained | the Session log format, Remote journal lifecycle, and Conversation identity/topology |
| Retained | the existing keyed slot, Generic fallback, and Chat, Details, and Trajectory structure |
| Forbidden | a new Client presenter service, parallel registry, or wire renderer id |
| Forbidden | new cards, visual redesign, interaction redesign, or Code Dispatch rich-card enhancements except the [nested terminal-card exception](../bug-fix/2026-09-05-nested-terminal-cards.md) |
| Forbidden | new cards, visual redesign, interaction redesign, or PTC dispatch rich-card enhancements except the [nested terminal-card exception](../bug-fix/2026-09-05-nested-terminal-cards.md) |
| Forbidden | compatibility dual-writing, version negotiation, or retention of the old `view` field |
## Terminology
@@ -96,7 +96,7 @@ The Host `ToolDefinition.presentCall`, `ToolDefinition.presentResult`, `ToolCall
1. The Client Session stores one contiguous raw event window.
2. `SessionEventSource` publishes `SessionEventEntry` values containing only events.
3. `ui-conversation` folds each event without a presentation companion.
4. The Chat and Trajectory Tool Definitions pair top-level calls and results by callId and assemble Code Dispatch subtrees.
4. The Chat and Trajectory Tool Definitions pair top-level calls and results by callId and assemble PTC dispatch subtrees.
5. `RunningToolCall` and `ToolResultNode` retain raw facts, metadata, and existing parent identity.
6. `ToolCallTree` dispatches `tool.call.toolview` by wire tool name.
7. `ui-tool` derives card component props from the block at the render site.
@@ -132,7 +132,7 @@ Session page/follow
Client SessionEventSource
-> Conversation Tool Definition
-> root call/result pairing + Code Dispatch topology
-> root call/result pairing + PTC dispatch topology
-> ToolCallBlock(name, argsRaw, content, error, meta)
-> tool.call.toolview keyed dispatch
-> Client card model
@@ -242,11 +242,11 @@ The Chat and Trajectory Tool Definitions read no views. They derive the followin
`ToolCallBlock` does not gain a generic `view`, `card`, `kind`, or `locations` field to replace the deleted fields. Concrete presentation remains the responsibility of `ui-tool` and keyed renderers.
### Root and Code Dispatch subcalls
### Root and PTC dispatch subcalls
Host presenter APIs describe top-level calls and results. Code Dispatch subcalls retain Generic, flattened presentation for the diff, read, search, and web models covered here; supported terminal calls use the same eligibility rules as roots.
Host presenter APIs describe top-level calls and results. PTC dispatch subcalls retain Generic, flattened presentation for the diff, read, search, and web models covered here; supported terminal calls use the same eligibility rules as roots.
Code Dispatch start and result events already carry `parentCallId`. Conversation preserves that existing fact on each child `ToolCallBlock`; root Session calls omit it. The diff, read, search, and web models accept only blocks without `parentCallId`; the terminal model and existing renderers that intentionally support nested calls accept child blocks.
PTC dispatch start and result events already carry `parentCallId`. Conversation preserves that existing fact on each child `ToolCallBlock`; root Session calls omit it. The diff, read, search, and web models accept only blocks without `parentCallId`; the terminal model and existing renderers that intentionally support nested calls accept child blocks.
Shared card models apply the same terminal eligibility and nonterminal child restrictions wherever a block renders, so no second presentation surface needs a placement field; the details panel that once delegated a selected block was removed with the right-hand details column ([decision](../feature/2026-09-04-right-sidebar-docking-infrastructure.md)).
@@ -308,7 +308,7 @@ The Client terminal model derives existing `TerminalBlock` props from the tool n
| settled persistent `bash`/`pwsh` | Generic flattened result, with no new exit card |
| foreground `terminal_send` | terminal prompt and output |
| background/error `terminal_send` | Generic result |
| Code Dispatch child | same terminal eligibility and fallback rules as a root call |
| PTC dispatch child | same terminal eligibility and fallback rules as a root call |
Standard shell results parse trailing `[exit code: N]` and `[killed by signal: X]` markers. A final recognized spill-policy notice selects Generic output instead: expandable in shell rows and raw in Details, because the exit marker may be displaced or omitted. A parsed marker is removed from the terminal body; timeout, sandbox denial, and markers without a pill remain in the body.
@@ -331,7 +331,7 @@ Standard and persistent providers sharing the same tool name are a special compa
| successful settled `write`/`edit` | applied contextual hunks from `meta.diffs` |
| settled `str_replace_editor` | Generic, because the tool defines no result presenter |
| write create or missing/malformed/empty applied metadata | current argument fallback |
| error, malformed arguments, edit with malformed metadata, or Code Dispatch child | Generic |
| error, malformed arguments, edit with malformed metadata, or PTC dispatch child | Generic |
Paths, `oldText:null`, `newText`, result-over-call diff precedence, the eight-line Chat limit, full-height Details presentation, and file-opening behavior remain unchanged.
@@ -339,7 +339,7 @@ Paths, `oldText:null`, `newText`, result-over-call diff precedence, the eight-li
A running `read` continues to show only the summary row. A successful settled `read` reads path, offset, lines, totalLines, and lang from result metadata and confirms that the result is one text block matching the read envelope.
Missing metadata, malformed fields, a mismatched result envelope, an error, a missing call head, or a Code Dispatch child all use Generic. Cwd-relative path labels, home abbreviation, syntax language, total line count, the eight-line Chat limit, and full-height Details presentation remain unchanged.
Missing metadata, malformed fields, a mismatched result envelope, an error, a missing call head, or a PTC dispatch child all use Generic. Cwd-relative path labels, home abbreviation, syntax language, total line count, the eight-line Chat limit, and full-height Details presentation remain unchanged.
The Client does not need to construct Host `ReadResultView.content`; Generic fallback can always read raw result content directly.
@@ -347,7 +347,7 @@ The Client does not need to construct Host `ReadResultView.content`; Generic fal
A running `grep` or `glob` continues to show only the argument summary. Successful results produce grouped matches or a path list from `meta.shape:'matches'` and `meta.shape:'paths'`, respectively.
The Client validates path, lineNumber, line, truncated, and total. Empty matches or paths form a valid card. Missing or malformed metadata, an unknown shape, an error, a missing call head, or a Code Dispatch child uses Generic.
The Client validates path, lineNumber, line, truncated, and total. Empty matches or paths form a valid card. Missing or malformed metadata, an unknown shape, an error, a missing call head, or a PTC dispatch child uses Generic.
When `truncated:true`, the card continues to show a recovery locator from raw result content. It does not show one when untruncated. The eight-line Chat limit, full-height Details presentation, and expansion behavior remain unchanged.
@@ -355,7 +355,7 @@ When `truncated:true`, the card continues to show a recovery locator from raw re
A running `web_search` or `web_fetch` continues to show only the summary row. A successful search builds the card from `meta.sources`, `meta.answer`, and `meta.truncated`; a successful fetch builds it from `meta.url`, `meta.statusCode`, and `meta.truncated`.
The Client validates every source's url, title, snippet, and publishedAt, and continues rendering only http/https URLs as links. Missing or malformed metadata, an error, a missing call head, or a Code Dispatch child uses Generic.
The Client validates every source's url, title, snippet, and publishedAt, and continues rendering only http/https URLs as links. Missing or malformed metadata, an error, a missing call head, or a PTC dispatch child uses Generic.
Search answer text, source ordering, label fallback, and truncation notice remain unchanged. The fetch final URL, status, truncation notice, and raw body below Details remain unchanged.
@@ -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
@@ -420,7 +420,7 @@ The fixture does not import Host tool packages to compute page presentation and
| grep/glob | current grouped/path card, truncation, and recovery |
| web_search/web_fetch | current source/summary card and raw body |
| Todo/Question/Skill/Cordis | current specialized rows |
| Code Dispatch subcall | terminal cards when eligible; diff/read/search/web remain Generic/flattened |
| PTC dispatch subcall | terminal cards when eligible; diff/read/search/web remain Generic/flattened |
| Chat and Details | identical card fields for the same call |
| Trajectory | current identity, tree, selection, and details |
| Deliverables | current successful-mutation chips and links |
@@ -478,7 +478,7 @@ This change does not promise to preserve differences expressed only through a Ho
- Conversation input and Tool blocks contain no view fields.
- Chat and Trajectory Tool Definitions read raw events.
- Event pairing, Context replay, trees, and target snapshots remain unchanged.
- Child Tool blocks preserve the existing Code Dispatch `parentCallId`; row and Details slot owner props add no separate placement field.
- Child Tool blocks preserve the existing PTC dispatch `parentCallId`; row and Details slot owner props add no separate placement field.
### UI Tool and Deliverables
@@ -515,7 +515,7 @@ This change does not promise to preserve differences expressed only through a Ho
- replace, prepend, and append accept entries without views.
- Chat and Trajectory root call/result pairing remains unchanged.
- The Code Dispatch tree remains unchanged.
- The PTC dispatch tree remains unchanged.
- Result-only fallback remains unchanged.
- A synthetic interruption result copies no view.
- Node identity across registry rebuild, older prepend, and live append remains unchanged.
@@ -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;
@@ -589,7 +589,7 @@ Changes to this decision use `dsh-pre-push-checks` to select commands for the fi
- Deliverables does not depend on render intent and preserves current paths.
- Text, components, expanded content, states, links, and ordering for all first-party top-level tools remain unchanged.
- Malformed, missing-metadata, error, orphan, and unknown-tool cases continue to fall back safely.
- Code Dispatch diff, read, search, and web subcalls remain Generic and flattened; terminal subcalls follow root eligibility.
- PTC dispatch diff, read, search, and web subcalls remain Generic and flattened; terminal subcalls follow root eligibility.
- Chat, Details, and Trajectory behavior remains unchanged.
- Existing Web browser expected outputs pass without refresh.
- Host presenter APIs, implementations, and direct tests remain unchanged.
@@ -634,7 +634,7 @@ An on-demand RPC would turn one page read into N network calls and would still r
### Allow presentation enhancements
Bundling richer Code Dispatch cards, missing-call-head inference, or other historical presentation enhancements with the ownership change would prevent snapshots from proving equivalence. This decision rejects that coupling; the [nested terminal-card exception](../bug-fix/2026-09-05-nested-terminal-cards.md) does not relax nonterminal child restrictions.
Bundling richer PTC dispatch cards, missing-call-head inference, or other historical presentation enhancements with the ownership change would prevent snapshots from proving equivalence. This decision rejects that coupling; the [nested terminal-card exception](../bug-fix/2026-09-05-nested-terminal-cards.md) does not relax nonterminal child restrictions.
### Accept temporary Generic degradation
@@ -703,7 +703,7 @@ This note preserves result metadata from the [canonical tool output contract](20
## Deferred
- A separate explicit decision may evaluate deleting Host presenters if they remain without production consumers; this decision does not prejudge it.
- Specialized diff, read, search, and web cards for Code Dispatch subcalls require a separate design and visible-snapshot updates; terminal calls are covered by the linked partial supersession.
- Specialized diff, read, search, and web cards for PTC dispatch subcalls require a separate design and visible-snapshot updates; terminal calls are covered by the linked partial supersession.
- A third-party mutation tool that joins Deliverables requires a new Client-owned contribution; this decision does not create a registry for an absent consumer.
- Distinct Client presentation for same-named providers first requires a stable, non-presentational identity; it must not restore per-page Host views.
- If Client card-model performance needs measurement, an immutable-block microbenchmark can be added; the shipped architecture already prohibits scanning the Session window.
@@ -12,13 +12,13 @@ Session 历史是持久 journal 接口,工具卡片属于 Client 展示。在
Host 投影还会重复结构化数据。read、diff、search 与 web 结果已在 `tool/result.data.meta` 中持久化有界事实;另一份 view 只增加 Remote payload 与 Client 解码成本,不增加持久语义。
Client 已经拥有完整的工具展示入口。`ui-chat` 将 `tool/call`、`tool/result` 与 Code Dispatch 事件组装成稳定的 `ToolCallBlock`;`ui-tool` 拥有递归调用树、按工具名称分发的 `tool.call.toolview` keyed slot、Generic fallback、卡片模型和 details output;业务 Client 插件可以为自己的工具名称注册 renderer。
Client 已经拥有完整的工具展示入口。`ui-chat` 将 `tool/call`、`tool/result` 与 PTC dispatch 事件组装成稳定的 `ToolCallBlock`;`ui-tool` 拥有递归调用树、按工具名称分发的 `tool.call.toolview` keyed slot、Generic fallback、卡片模型和 details output;业务 Client 插件可以为自己的工具名称注册 renderer。
Host presenter 与 Client keyed renderer 分担展示会形成对同一事件的两套解释。keyed renderer 是 Web 扩展点,因此中间 Host view 不提供独立 Web 能力。
`ToolDefinition.presentCall`/`presentResult` 仍是保留的 Host API;ACP 采用 automation-only 协议,仓库也没有生产 TUI consumer。是否删除这些定义与 Session 读取是否独立于展示是两个决定。
所需结果是一条原始 Session journal 和一个 Client 展示 owner,且不发生可见退化或顺带增强。专用卡片、交互和 Code Dispatch 拓扑保持稳定,transport 不再携带临时 view。
所需结果是一条原始 Session journal 和一个 Client 展示 owner,且不发生可见退化或顺带增强。专用卡片、交互和 PTC dispatch 拓扑保持稳定,transport 不再携带临时 view。
## Decision
@@ -26,7 +26,7 @@ Host presenter 与 Client keyed renderer 分担展示会形成对同一事件的
Session Remote journal 只下发原始、已验证、可持久化的 Session event。`session.page` 和 `session.follow` 不解析工具参数,不查询 Tools registry,不恢复 presenter scope,不执行 `presentCall`/`presentResult`,也不构造或克隆任何 tool view。
Client Conversation 层继续负责工具调用与结果的 identity、配对、生命周期、Code Dispatch 拓扑和稳定 Chat Node。它不解释具体工具名称,也不生成 terminal、diff、read、search 或 web 组件 props。
Client Conversation 层继续负责工具调用与结果的 identity、配对、生命周期、PTC dispatch 拓扑和稳定 Chat Node。它不解释具体工具名称,也不生成 terminal、diff、read、search 或 web 组件 props。
Client `ui-tool` 继续负责 card model 和具体 renderer。每个 card model 改为直接读取 `ToolCallBlock` 中的工具名称、原始参数、结果内容、错误、持久 metadata、Session cwd 与 Host home,并生成与现有页面相同的组件 props。
@@ -51,7 +51,7 @@ Host 的 `ToolDefinition.presentCall`、`ToolDefinition.presentResult`、`ToolCa
| 保留 | Session 日志格式、Remote journal 生命周期与 Conversation identity/topology |
| 保留 | 现有 keyed slot、Generic fallback、Chat、Details 与 Trajectory 结构 |
| 禁止 | 新 Client presenter service、平行 registry 或 wire renderer id |
| 禁止 | 新卡片、视觉改版、交互改版或 Code Dispatch rich-card 增强,[嵌套 terminal 卡片例外](../bug-fix/2026-09-05-nested-terminal-cards.zh.md)除外 |
| 禁止 | 新卡片、视觉改版、交互改版或 PTC dispatch rich-card 增强,[嵌套 terminal 卡片例外](../bug-fix/2026-09-05-nested-terminal-cards.zh.md)除外 |
| 禁止 | 为兼容保留双写、版本协商或旧 `view` 字段 |
## 术语
@@ -96,7 +96,7 @@ Host 的 `ToolDefinition.presentCall`、`ToolDefinition.presentResult`、`ToolCa
1. Client Session 保存一个连续 raw event window。
2. `SessionEventSource` 发布只含 event 的 `SessionEventEntry`。
3. `ui-conversation` 在没有 presentation companion 的情况下 fold 每个事件。
4. Chat 与 Trajectory Tool Definition 按 callId 配对顶层 call/result,并组装 Code Dispatch 子树。
4. Chat 与 Trajectory Tool Definition 按 callId 配对顶层 call/result,并组装 PTC dispatch 子树。
5. `RunningToolCall` 与 `ToolResultNode` 保存 raw facts、metadata 与既有 parent identity。
6. `ToolCallTree` 按 wire tool name 分发 `tool.call.toolview`。
7. `ui-tool` 在 render site 从 block 派生 card component props。
@@ -132,7 +132,7 @@ Session page/follow
Client SessionEventSource
-> Conversation Tool Definition
-> root call/result pairing + Code Dispatch topology
-> root call/result pairing + PTC dispatch topology
-> ToolCallBlock(name, argsRaw, content, error, meta)
-> tool.call.toolview keyed dispatch
-> Client card model
@@ -242,11 +242,11 @@ Chat 和 Trajectory 的 Tool Definition 都不读取 view,而从事件生成
`ToolCallBlock` 不新增通用 `view`、`card`、`kind` 或 `locations` 字段替代被删除字段。具体展示仍只属于 `ui-tool` 与 keyed renderer。
### Root 与 Code Dispatch 子调用
### Root 与 PTC dispatch 子调用
Host presenter API 描述顶层 call/result。本决定覆盖的 diff、read、search 和 web model 对 Code Dispatch 子调用保留 Generic/flattened 展示;受支持的 terminal 调用使用与根调用相同的适用规则。
Host presenter API 描述顶层 call/result。本决定覆盖的 diff、read、search 和 web model 对 PTC dispatch 子调用保留 Generic/flattened 展示;受支持的 terminal 调用使用与根调用相同的适用规则。
Code Dispatch start 与 result event 已经携带 `parentCallId`。Conversation 在每个 child `ToolCallBlock` 上保留这项现有事实,root Session call 则不携带它。diff、read、search 和 web model 只接受没有 `parentCallId` 的 block;terminal model 与原本有意支持嵌套调用的 renderer 接受 child block。
PTC dispatch start 与 result event 已经携带 `parentCallId`。Conversation 在每个 child `ToolCallBlock` 上保留这项现有事实,root Session call 则不携带它。diff、read、search 和 web model 只接受没有 `parentCallId` 的 block;terminal model 与原本有意支持嵌套调用的 renderer 接受 child block。
共享的 card model 在 block 渲染到哪里都施加同样的终端资格与非终端子调用限制,因此不需要第二个展示面带 placement 字段;曾经原样委托选中 block 的详情面板已随右侧详情列一并删除([决策](../feature/2026-09-04-right-sidebar-docking-infrastructure.zh.md))。
@@ -308,7 +308,7 @@ Client terminal model 从工具名称、调用参数、结果 content、error
| persistent `bash`/`pwsh` settled | Generic flattened result,不新增 exit card |
| `terminal_send` 前台 | terminal prompt 与 output |
| `terminal_send` background/error | Generic 结果 |
| Code Dispatch child | 与根调用相同的 terminal 适用规则与 fallback 规则 |
| PTC dispatch child | 与根调用相同的 terminal 适用规则与 fallback 规则 |
标准 shell 结果解析末尾 `[exit code: N]` 与 `[killed by signal: X]`。末尾已识别的 spill 策略提示会改用 Generic 输出:在 shell 行中可展开,在 Details 中显示原文,因为退出标记可能被移位或省略。已解析的 marker 从 terminal 正文移除;timeout、sandbox denial 与没有 pill 的 marker 留在正文。
@@ -331,7 +331,7 @@ TerminalBlock 的 ANSI、光标重放、宽字符、行数上限、展开、复
| settled `write`/`edit` success | 从 `meta.diffs` 生成 applied contextual hunks |
| settled `str_replace_editor` | Generic,因为该工具没有 result presenter |
| write create 或 applied metadata 缺失、畸形、为空 | 当前 args fallback |
| error、畸形 args、edit 的 metadata 畸形、Code Dispatch child | Generic |
| error、畸形 args、edit 的 metadata 畸形、PTC dispatch child | Generic |
路径、`oldText:null`、`newText`、结果覆盖调用时 diff、Chat 8 行上限、Details 全高显示和文件打开行为不变。
@@ -339,7 +339,7 @@ TerminalBlock 的 ANSI、光标重放、宽字符、行数上限、展开、复
running `read` 继续只有摘要行。成功 settled `read` 从 result meta 读取 path、offset、lines、totalLines 与 lang,并确认结果是单个文本块且符合 read envelope。
meta 缺失、字段畸形、result envelope 不匹配、error、缺失 call head 或 Code Dispatch child 都走 Generic。路径 label 的 cwd 相对化、home 缩写、语法语言、总行数、Chat 8 行上限与 Details 全高显示不变。
meta 缺失、字段畸形、result envelope 不匹配、error、缺失 call head 或 PTC dispatch child 都走 Generic。路径 label 的 cwd 相对化、home 缩写、语法语言、总行数、Chat 8 行上限与 Details 全高显示不变。
Client 不需要构造 Host `ReadResultView.content`;Generic fallback 始终可直接读取原始 result content。
@@ -347,7 +347,7 @@ Client 不需要构造 Host `ReadResultView.content`;Generic fallback 始终
running `grep`/`glob` 继续只有参数摘要。成功结果分别从 `meta.shape:'matches'` 与 `meta.shape:'paths'` 生成 grouped matches 或 path list。
Client 校验 path、lineNumber、line、truncated 与 total。空 matches/paths 是有效卡片;缺失/畸形 meta、未知 shape、error、缺失 call head 与 Code Dispatch child 走 Generic。
Client 校验 path、lineNumber、line、truncated 与 total。空 matches/paths 是有效卡片;缺失/畸形 meta、未知 shape、error、缺失 call head 与 PTC dispatch child 走 Generic。
`truncated:true` 时继续从原始 result content 显示 recovery locator;未截断时不显示。Chat 8 行上限、Details 全高显示和展开行为不变。
@@ -355,7 +355,7 @@ Client 校验 path、lineNumber、line、truncated 与 total。空 matches/paths
running `web_search`/`web_fetch` 继续只有摘要行。成功 search 从 `meta.sources`、`meta.answer`、`meta.truncated` 生成卡片;成功 fetch 从 `meta.url`、`meta.statusCode`、`meta.truncated` 生成卡片。
Client 校验每个 source 的 url、title、snippet 与 publishedAt,并继续只把 http/https URL 渲染为链接。meta 缺失或畸形、error、缺失 call head 与 Code Dispatch child 走 Generic。
Client 校验每个 source 的 url、title、snippet 与 publishedAt,并继续只把 http/https URL 渲染为链接。meta 缺失或畸形、error、缺失 call head 与 PTC dispatch child 走 Generic。
search 的 answer、来源顺序、label fallback 与截断提示不变;fetch 的最终 URL、状态、截断提示与 Details 下方原始正文不变。
@@ -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
@@ -420,7 +420,7 @@ fixture 不导入 Host 工具包来计算页面展示,也不保留 presenter
| grep/glob | 当前 grouped/path card、截断与 recovery |
| web_search/web_fetch | 当前来源/摘要 card 与原始正文 |
| Todo/Question/Skill/Cordis | 当前专用行 |
| Code Dispatch subcall | 满足条件时显示 terminal 卡片;diff/read/search/web 保持 Generic/flattened |
| PTC dispatch subcall | 满足条件时显示 terminal 卡片;diff/read/search/web 保持 Generic/flattened |
| Chat 与 Details | 同一调用使用相同 card fields |
| Trajectory | 当前 identity、树、选择和 details |
| Deliverables | 当前成功 mutation chips 与链接 |
@@ -478,7 +478,7 @@ Host registry 允许不同 scope 为同一 tool name 提供不同定义;Sessio
- Conversation input 与 Tool block 不含 view 字段。
- Chat/Trajectory Tool Definition 读取 raw event。
- event pairing、Context replay、树与 target snapshot 保持不变。
- child Tool block 保留现有 Code Dispatch `parentCallId`;row 与 Details slot owner props 都不增加独立 placement 字段。
- child Tool block 保留现有 PTC dispatch `parentCallId`;row 与 Details slot owner props 都不增加独立 placement 字段。
### UI Tool 与 Deliverables
@@ -515,7 +515,7 @@ Host registry 允许不同 scope 为同一 tool name 提供不同定义;Sessio
- replace、prepend 与 append 接受无 view entry。
- Chat 与 Trajectory root call/result 配对不变。
- Code Dispatch 树不变。
- PTC dispatch 树不变。
- result-only fallback 不变。
- interruption synthetic result 不复制 view。
- registry rebuild、older prepend 与 live append 的 Node identity 不变。
@@ -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;
@@ -589,7 +589,7 @@ Host registry 允许不同 scope 为同一 tool name 提供不同定义;Sessio
- Deliverables 不依赖 render intent 且保持当前 paths。
- 所有第一方顶层工具的文本、组件、展开内容、状态、链接与排序不变。
- malformed、missing-meta、error、orphan 与 unknown-tool 继续安全 fallback。
- Code Dispatch 的 diff、read、search 和 web 子调用保持 Generic/flattened;terminal 子调用遵循根调用适用规则。
- PTC dispatch 的 diff、read、search 和 web 子调用保持 Generic/flattened;terminal 子调用遵循根调用适用规则。
- Chat、Details 与 Trajectory 行为不变。
- 现有 Web browser expected 无需刷新即可通过。
- Host presenter API、实现与直接测试不变。
@@ -634,7 +634,7 @@ read 行结构、applied diff、search 分组、web sources 和有效 truncation
### 允许展示增强
将更丰富的 Code Dispatch 卡片、缺失 call head 的推断或其他历史展示增强与所有权变更捆绑,会使快照无法证明对等。本决定拒绝这种捆绑;[嵌套 terminal 卡片例外](../bug-fix/2026-09-05-nested-terminal-cards.zh.md)不放宽非 terminal 子调用限制。
将更丰富的 PTC dispatch 卡片、缺失 call head 的推断或其他历史展示增强与所有权变更捆绑,会使快照无法证明对等。本决定拒绝这种捆绑;[嵌套 terminal 卡片例外](../bug-fix/2026-09-05-nested-terminal-cards.zh.md)不放宽非 terminal 子调用限制。
### 接受临时 Generic 退化
@@ -703,7 +703,7 @@ optional `view` 的缺失是所有 consumer 共同遵守的预发布 wire 类型
## Deferred
- Host presenter 若长期没有生产消费者,可由另一项明确决策评估删除;本决定不预判。
- Code Dispatch 子调用的 diff、read、search 和 web 专用卡片仍需独立设计并更新可见快照;terminal 调用由链接的部分取代决策负责。
- PTC dispatch 子调用的 diff、read、search 和 web 专用卡片仍需独立设计并更新可见快照;terminal 调用由链接的部分取代决策负责。
- 第三方 mutation tool 若要加入 Deliverables,需新增 Client-owned 贡献;本决定不为尚无消费者的扩展性建 registry。
- 同名 provider 若要不同 Client 展示,需先定义稳定、非展示性的 identity;不得恢复按页 Host view。
- Client card model 若需量化性能,可以增加 immutable-block 微基准;已交付架构禁止扫描 Session window。
@@ -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/2026-08-25-electron-desktop-packaging-and-updates.md
2026-08-25-electron-desktop-packaging-and-updates.md: 6d22777c8913911dbb1d89c09de9e891636873c8
2026-08-25-electron-desktop-packaging-and-updates.zh.md: 5108492ba6f029847735f570aa77c2cb505f981c
2026-08-25-electron-desktop-packaging-and-updates.md: 4a4dfe8910905b3c35fbdfdcaedd34a556b532f3
2026-08-25-electron-desktop-packaging-and-updates.zh.md: 69877127578ae4d61643653403b736dbb5e7b1e6
@@ -4,19 +4,21 @@ Status: implemented
English | [中文](2026-08-25-electron-desktop-packaging-and-updates.zh.md)
Profile mutation and recovery follow the [in-place profile decision](2026-09-09-desktop-in-place-profile.md).
## Problem
DeepSeek Harness needs an Electron desktop application that reuses the Web UI, works without system Node.js or pnpm, installs dsh and desktop plugins through an application-bundled pnpm, and updates the complete desktop release through one user-facing flow.
The desktop application and an npm-installed dsh share the `.dsh` data root, but they may have different dsh and plugin versions. They must share supported product data without sharing executable packages, lockfiles, `node_modules`, plugin activation, or package-manager configuration.
The current GUI protocol binds the Web client and backend release. Independently versioning the Electron artifact and its pnpm-installed dsh would create unqualified shell, client, backend, and plugin combinations and make update availability ambiguous.
The current GUI protocol binds the Web client and backend release. Independently versioning the Electron artifact and its bundled dsh would create unqualified shell, client, backend, and plugin combinations and make update availability ambiguous.
## Decision
Ship a small Electron shell with a bundled upstream Node.js executable and pinned pnpm. Electron starts the private Desktop Host package as an isolated child process; that package composes the installed dsh backend and matching client graph. Fetch metadata and bounded raw request and response chunks travel over two versioned framed byte pipes, Node IPC is reserved for readiness, fatal failure, and shutdown, and Electron serves validated assets through `dsh-app://`; it opens no listening port. Each frame carries a fixed marker, type, monotonic stream id, payload length, and validated payload. Serialized writers honor pipe drain, readers pause globally when a request or response stream applies backpressure, cancellation closes the matching stream, and late response frames for a retired stream stay inert. The Connection plugin provides its carrier-neutral RPC and Fetch registries without requiring `webServer`, while Client Modules provides the exact advertised combo-bundle responses to the shell-owned carrier; Web compositions attach their optional HTTP routes for both. The renderer keeps the same Fetch, RPC, and Remote-stream formats, while the child carrier avoids Base64 expansion and V8 serialization compatibility between Electron and the bundled upstream Node.js. Electron closes its request-pipe writer after sending shutdown, releasing an in-flight Windows pipe read before it waits for child exit. This follows the Electron reservation in the [GUI layering and RPC protocol note](../../archived/architecture/2026-07-19-gui-layering-and-rpc-protocol.md).
Electron owns the reserved profile at `.dsh/profiles/desktop`. Its exact `@deepseek-ai/dsh` dependency supplies the backend and matching Web UI, while the matching private `@deepseek-ai/dsh-desktop-host` dependency supplies only the Electron child-process entry and composition overlay. The dsh release, private Host, and their first-party dependency closures use local npm tarballs packed from the same source build; the profile manifest lists every core package as a local `file:` dependency, and `pnpm-workspace.yaml` repeats the mapping as overrides. The Host remains outside the public CLI package and is never published to npm. Desktop plugins are additional registry npm dependencies and ordered `dsh.profile.bundles` entries in the same profile, and resolve from its one `node_modules`.
Electron owns the reserved profile at `.dsh/profiles/desktop`. The [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md) owns core resource storage, external plugin dependencies, shared package links, and profile reconciliation. The private Desktop Host remains outside the public CLI package and is never published to npm.
One Desktop release number identifies the Electron artifact and its exact `@deepseek-ai/dsh` and `@deepseek-ai/dsh-desktop-host` dependencies. A release cannot select a different core version at build or runtime. Updating dsh therefore requires a new Electron release even when shell code is unchanged.
@@ -26,9 +28,9 @@ The browser Web UI, dsh backend, existing `dsh plugin` CLI, user npm, and user p
| Owner | Responsibility |
|---|---|
| Electron shell | Window and child lifecycle, framed byte pipes, lifecycle IPC, custom protocol, reserved desktop profile, plugin GUI, update coordination, rollback |
| Electron shell | Window and child lifecycle, framed byte pipes, lifecycle IPC, custom protocol, reserved desktop profile, plugin GUI, update coordination |
| Bundled Node.js and pnpm | Execute dsh and install exact desktop-project dependencies without consulting user `PATH` or pnpm state |
| Desktop profile | One dependency graph, ordered bundle list, and `node_modules` for the desktop dsh package and desktop plugins |
| Desktop profile | External plugin dependencies, ordered enabled bundles, and shared links defined by the bundled-runtime decision |
| Private Desktop Host package | Electron-only child-process entry and composition overlay installed with dsh but excluded from the public CLI package and npm publication |
| Installed dsh package | Backend, matching Web UI, boot manifest, client bundles, and product behavior |
| Shared `.dsh` owners | Sessions, settings, credentials, workspaces, and storage, guarded by their existing locks and format versions |
@@ -41,10 +43,6 @@ The renderer uses `nodeIntegration: false`, `contextIsolation: true`, and `sandb
```text
~/.dsh/
desktop/
staging/<transaction-id>/profile/
rollback/profile/
pending.json
lock
pnpm/
store/
cache/
@@ -54,52 +52,54 @@ The renderer uses `nodeIntegration: false`, `contextIsolation: true`, and `sandb
desktop/
package.json
pnpm-lock.yaml
lock
desktop-packages-pending
pnpm-workspace.yaml
desktop-release.json
desktop-packages.json
desktop-packages/
desktop-runtime-state.json
node_modules/
sessions/
storages/
```
`.dsh/profiles/desktop` is the only active desktop profile. Its package manifest records the built-in and installed plugin bundle order; Electron alone mutates its dependencies, lockfile, and `node_modules`. Production startup rejects a bundle resolved outside this profile, including the CLI-maintained `.dsh/profiles/node_modules` fallback. Package content installed for the desktop profile uses `.dsh/desktop/pnpm/store`.
`.dsh/profiles/desktop` is the only active desktop profile. Its executable package ownership and allowed resolution directories follow the [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md). Plugin package content uses `.dsh/desktop/pnpm/store`.
## Installation and resolution
The installer never mutates the active profile in place. It copies profile metadata into a transaction staging directory and applies an exact dependency change with the bundled pnpm. Before testing staging, Electron stops the active backend; it starts and stops the staged backend alone, then restores the active backend before activation, so two Desktop backends never concurrently share `.dsh` state. Activation stops the backend again, persists each next `pending.json` phase before its corresponding filesystem move, moves the active profile to `rollback/profile`, moves staging into `.dsh/profiles/desktop`, and restarts. Recovery combines the write-ahead phase with the actual active, rollback, and staging directories so either write-to-move interruption retains or restores a complete profile.
Desktop stops the Host before modifying its profile in place. Package failures retain partial changes for explicit repair; profile mutation and package-retry ownership follow the [in-place decision](2026-09-09-desktop-in-place-profile.md).
The process-lifetime Electron lock is the authoritative Desktop owner. The package transaction lock is depth defense and records the process that can still mutate package state: Electron between package operations and the spawned pnpm PID while pnpm runs. The owner change is truncated, written, and synchronized through the already-open exclusive lock file. If Electron terminates during pnpm execution, a later process observes the live worker and refuses to start a competing store or staging transaction; after that worker exits, the stale PID can be recovered.
The process-lifetime Electron lock is the authoritative Desktop owner. The package transaction lock is depth defense and records the process that can still mutate package state: Electron between package operations and the spawned pnpm PID while pnpm runs. The owner change is truncated, written, and synchronized through the already-open exclusive lock file. If Electron terminates during pnpm execution, a later process observes the live worker and refuses to start a competing package transaction; after that worker exits, the stale PID can be recovered.
The packaged seed is an offline installation kit, not an executable dsh tree. It contains the release identity, initial desktop-project manifest, a descriptor and immutable tarballs for the union of the first-party package closures rooted at dsh and the private Desktop Host, lockfile, integrity inventory, and required store subset. Each `mac-arm64`, `mac-x64`, and `win-x64` build owns its packed packages, runtime, package set, seed, pnpm preparation state, unpacked application, update metadata, and final artifacts under `.desktop-build/targets/<target>`; only the immutable, checksum-verified Node.js download cache is shared. The release build requires the Electron package, root dsh package, and private Host package to have the same version, creates final npm tarballs from the official source build, locally packs the private Host, selects the reachable dsh, Host, and vendored packages plus the Landlock entry, and verifies that the Host tarball contains `lib/index.js` and `config/desktop.cordis.patch.yml`. The Host `files` manifest contains only that runtime entry and overlay, and the package is never published to npm. Public package tarballs remain the official `pnpm pack` results governed by each package's publication manifest; Desktop does not remove published declarations or otherwise create a second package-content policy. The seed manifest lists every selected package as a local direct dependency, automatic peer installation is disabled, and the workspace file overrides every selected first-party name to its local tarball. The target Node.js executes bundled pnpm, so pnpm's operating-system and CPU selection makes the materialized dependency graph and seed target-specific. Bundled pnpm disables its global virtual store, materializes external production dependencies from npm without lifecycle scripts, deletes `node_modules` and every temporary pnpm cache, config, and state directory, then performs a clean offline installation from the final store alone and checks the private Host entry and overlay. The build rejects any lockfile that resolves one of the local first-party names by registry version. Inventory generation follows removal of that second `node_modules` tree and temporary pnpm project registrations. Requiring both Host files before copying the package set and after offline installation prevents a release whose process entry loads but cannot compose its required overlay from reaching application signing.
The seed stores pnpm content in 16 deterministic uncompressed tar shards selected by normalized store path. Apple notarization inspects Mach-O code inside those archives, so macOS seed preparation stages every referenced Mach-O content-addressed object and runs at most four independent Developer ID signers concurrently with a secure timestamp and hardened runtime. A signer failure is observed only after every active signer exits and leaves the original CAS objects and package index unchanged. After all signers succeed, preparation writes each object at its new SHA-512 path and transactionally rewrites every base and side-effects file reference in pnpm's MessagePack SQLite index. A second offline installation proves that pnpm resolves the rewritten store; preparation then shards it, extracts the final archives, and verifies every embedded signature. Package paths and non-native bytes remain unchanged, and the seed retains bundled architecture variants because removing files would create a Desktop-specific package file set. Seed integrity covers the shard manifest and every archive before extraction. Startup validates archive paths, entry types, uniqueness, and counts, extracts every shard into a unique Desktop-owned staging directory, replaces matching immutable store files, and transactionally merges each pnpm store version's SQLite `package_index` into `.dsh/desktop/pnpm/store`. Seed records replace matching keys while records downloaded for Desktop plugins remain. An interrupted file merge may leave valid immutable cache content, but each SQLite merge is atomic, and profile installation and activation still require pnpm integrity and the complete health check.
Startup requires the packaged release identity to equal Electron's application version, then compares `.dsh/profiles/desktop/desktop-release.json` plus the installed dsh and Desktop Host packages with that release before launching the backend. It installs the new seed manifest and lockfile with `pnpm install --offline --frozen-lockfile --trust-lockfile` in staging. After Electron replacement, it restores every plugin bundle recorded in the active profile at its exact installed version through one offline pnpm add from the existing desktop store and metadata cache. The complete graph must pass the same health check before activation.
The plugin GUI performs registry npm-package operations equivalent to `pnpm add <package> --save-exact`, `pnpm remove <package>`, and exact-version update in staging. Every mutation retains the local core-package descriptor, tarballs, dsh and Desktop Host dependencies, and complete override map. Electron validates the installed package manifest and updates the profile's dependency and ordered bundle entries; no renderer request can choose the registry, install directory, lifecycle policy, or arbitrary pnpm flags.
The backend and Loader use `.dsh/profiles/desktop/package.json` as their profile manifest and npm resolution anchor. The shared profile loader composes its ordered bundle entries, then the private Desktop Host applies its packaged overlay. The Host, dsh, Cordis, desktop plugins, plugin dependencies, and peer dependencies resolve through the ordinary pnpm `node_modules` graph. A desktop plugin contributing `dsh.client` code enters the boot manifest only after the complete profile passes health checking.
Core materialization, first launch, plugin installation, and shared-module resolution follow the [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md). The actual Host composes enabled desktop plugins contributing `dsh.client` code when it starts.
## Updates and recovery
Electron update uses one `electron-updater` release stream and signed `electron-builder` artifacts. Its version is the Desktop release version; there is no independent dsh manifest, compatibility range, or dsh-only update operation. A foreground install waits for an in-flight background check rather than reusing its result as an install result. The update dialog downloads and installs the Electron artifact, then restarts into the new release.
Before the new release opens a window, startup reconciles dsh from its packaged seed while retaining installed desktop plugins. The health check covers dependency resolution, native modules, shell API compatibility, backend startup and shutdown, Web assets, and the client boot graph. An incompatible plugin blocks activation and leaves the previous project available for rollback. Startup fails visibly rather than launching a shell and dsh version that do not match.
The [immediate-window decision](2026-09-09-desktop-immediate-window-and-direct-start.md) owns the local loading page, direct Host startup, and recovery in the main window. Profile reconciliation follows the [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md).
`DSH_DESKTOP_AUTO_UPDATE_ENV` selects the test deployment by default or the production deployment for both the target-specific generic-provider URL and COS destination. Release automation supplies the test HTTPS origin through `DOWNLOAD_TEST_ORIGIN` and each deployment's bucket through `DOWNLOAD_TEST_COS_BUCKET` or `DOWNLOAD_PROD_COS_BUCKET`; keeping mutable test routing and COS storage identities out of source lets deployment infrastructure change without a code release, while the public production origin remains fixed. Packaging resolves only the public updater URL, disables electron-builder publishing, removes every COS credential field from its subprocess environment, and writes a completion record only after electron-builder and every signing or notarization hook succeeds. Target upload additionally requires the selected bucket, then requires the completion record, root dsh version, Desktop version, version-derived channel metadata, artifact names, sizes, and SHA-512 values to agree before it reads the selected credentials or sends data. It uploads immutable versioned updater payloads and any separate blockmaps before replacing the channel metadata emitted by electron-builder, and it never deletes historical objects. Stable versions use the `latest` metadata name; prereleases use the first semantic-version prerelease identifier. NSIS embeds its blockmap in the signed executable; the macOS ZIP carries a separate blockmap. Both let electron-updater download changed blocks when supported, while application replacement and the local pnpm staging transaction remain separate operations.
`DSH_DESKTOP_AUTO_UPDATE_ENV` selects the test deployment by default or the production deployment for both the target-specific generic-provider URL and COS destination. Release automation supplies the test HTTPS origin through `DOWNLOAD_TEST_ORIGIN` and each deployment's bucket through `DOWNLOAD_TEST_COS_BUCKET` or `DOWNLOAD_PROD_COS_BUCKET`; keeping mutable test routing and COS storage identities out of source lets deployment infrastructure change without a code release, while the public production origin remains fixed. Packaging resolves only the public updater URL, disables electron-builder publishing, removes every COS credential field from its subprocess environment, and writes a completion record only after electron-builder and every signing or notarization hook succeeds. Target upload additionally requires the selected bucket, then requires the completion record, root dsh version, Desktop version, version-derived channel metadata, artifact names, sizes, and SHA-512 values to agree before it reads the selected credentials or sends data. It uploads immutable versioned updater payloads and any separate blockmaps before replacing the channel metadata emitted by electron-builder, and it never deletes historical objects. Stable versions use the `latest` metadata name; prereleases use the first semantic-version prerelease identifier. NSIS embeds its blockmap in the signed executable; the macOS ZIP carries a separate blockmap. Both let electron-updater download changed blocks when supported, while application replacement and the local pnpm package operation remain separate operations.
## Security and release policy
Core dsh and the private Desktop Host come only from integrity-recorded local npm tarballs inside the signed Electron release; pnpm overrides prevent transitive core packages from falling back to a registry. Store archives are integrity-checked and fully validated in an isolated extraction directory before their files can enter writable package state. Plugin installation accepts registry package specs allowed by desktop policy but never raw pnpm commands. Exact versions, lockfile integrity, a reviewed `allowBuilds` set, user-only directory permissions, redacted diagnostics, and health checking are required before activation.
Core dsh and the private Desktop Host come only from the signed application resource tree. Plugin installation accepts registry package specs allowed by desktop policy but never raw pnpm commands. Exact versions, lockfile integrity, a reviewed `allowBuilds` set, and user-only directory permissions are required before activation.
Electron artifacts are signed; macOS artifacts are notarized. Release automation must supply the application ID, macOS Developer ID qualifier, expected Team ID, and one complete notarytool credential strategy through explicit environment variables. Configuration loading rejects missing or malformed identifiers and incomplete notarization credentials, while macOS packaging requires signing so certificate discovery cannot silently select another installed identity or emit an unsigned release. Seed preparation verifies the exact Authority and Team ID plus the timestamp and hardened-runtime flags on every embedded Mach-O file. An after-sign hook performs Apple's deep strict application verification and requires the same leaf Authority and Team ID before artifact creation continues. Electron-builder then notarizes and staples the application and signs the DMG. The DMG artifact-completion hook separately notarizes and staples every DMG before requiring the configured identity, a valid ticket, and Gatekeeper acceptance; the upload event runs only after that hook succeeds. DMG blockmaps are disabled because macOS updates consume the signed ZIP, and stapling would otherwise invalidate an already-generated DMG blockmap. The custom protocol serves the installed frontend distribution plus client files named by the active module graph and rejects traversal or access outside those roots. The plugin installer API is available only to the Electron-owned management GUI and is absent from the browser application and backend RPC.
Electron release artifacts are signed; macOS artifacts are notarized. Release automation must supply the application ID, macOS Developer ID qualifier, expected Team ID, and one complete notarytool credential strategy through explicit environment variables. Configuration loading rejects missing or malformed identifiers and incomplete notarization credentials, while macOS packaging requires signing so certificate discovery cannot silently select another installed identity or emit an unsigned release. Runtime preparation verifies the exact Authority and Team ID plus the timestamp and hardened-runtime flags on every embedded Mach-O file. An after-sign hook performs Apple's deep strict application verification and requires the same leaf Authority and Team ID before artifact creation continues. The fixed-target installer command uses [isolated App copies for parallel notarization](../process/2026-09-09-parallel-macos-notarization.md): the ZIP contains a stapled App, while the signed DMG carries the ticket covering its unstapled inner App. The DMG artifact-completion hook requires the configured identity, a valid ticket, and Gatekeeper acceptance. Both artifact lanes must succeed before the command promotes their outputs and writes the release completion record; directory-only commands still notarize and staple the App. DMG blockmaps are disabled because macOS updates consume the signed ZIP, and stapling would otherwise invalidate an already-generated DMG blockmap. The custom protocol serves the installed frontend distribution plus client files named by the active module graph and rejects traversal or access outside those roots. The plugin installer API is available only to the Electron-owned management GUI and is absent from the browser application and backend RPC.
Windows release packaging supplies the public EV leaf certificate named by `DSH_DESKTOP_WINDOWS_CER_FILE` to the configured SafeNet-compatible SignTool through `/f` and identifies its matching private key through the required `DSH_DESKTOP_WINDOWS_KEY_CONTAINER`. The certificate file remains outside source control, and the private key remains on the USB token. The electron-builder hook passes each artifact to the CRLF `windows-sign.cmd`, whose single SignTool invocation uses the SafeNet `/kc "[{{PIN}}]=container"` value and CSP, a SHA-256 file digest, and a DigiCert SHA-256 RFC 3161 timestamp. The hook never substitutes another SignTool and never retries a failed request. Package orchestration withholds every `DSH_DESKTOP_WINDOWS_*` field from build and seed-preparation children and passes only the certificate path, SignTool path, key container, and PIN into electron-builder. The signer supplies only validated signing fields in an otherwise scrubbed CMD environment; the CMD disables delayed expansion, clears those fields before SignTool starts, and preserves the PIN only in the required SignTool command line. Every surfaced diagnostic replaces the PIN, and only the dedicated build account and administrators may inspect the runner. The signer signs electron-builder's temporary NSIS bootstrap before enterprise Code Integrity evaluates that executable and clears a generated executable's certificate-table entry only when it points beyond the file before applying the final signature. Packaging fails before producing unsigned artifacts when the SignTool, certificate, container, PIN, token, or signature is unavailable. The custom protocol serves the installed frontend distribution plus client files named by the active module graph and rejects traversal or access outside those roots. The plugin installer API is available only to the Electron-owned management GUI and is absent from the browser application and backend RPC.
The [pinned osx-sign patch](../../../../patches/@[email protected]) uses `lstat` in both published module builds, so Framework file and directory aliases do not trigger duplicate signing. The patch remains necessary until the selected upstream release skips those aliases. PAK files are resources sealed by the enclosing bundle; individual signatures add serial timestamp requests without additional resource integrity. Desktop preserves all locale files and skips only their standalone signatures. Executable code retains Developer ID signatures, secure timestamps, and hardened runtime. The [signer traversal regression](../../../../apps/desktop/tests/macos-signing-walk.spec.ts) exercises the installed dependency with real Framework aliases; release qualification still requires strict application verification, notarization, and startup.
Packaged applications ignore development resource and project environment overrides. Only an unpackaged Electron process can replace the Node.js binary, pnpm entry, seed, or active project.
Windows release packaging supplies the public EV leaf certificate named by `DSH_DESKTOP_WINDOWS_CER_FILE` to the configured SafeNet-compatible SignTool through `/f` and identifies its matching private key through the required `DSH_DESKTOP_WINDOWS_KEY_CONTAINER`. The certificate file remains outside source control, and the private key remains on the USB token. The electron-builder hook passes each artifact to the CRLF `windows-sign.cmd`, whose single SignTool invocation uses the SafeNet `/kc "[{{PIN}}]=container"` value and CSP, a SHA-256 file digest, and a DigiCert SHA-256 RFC 3161 timestamp. The hook never substitutes another SignTool and never retries a failed request. Package orchestration withholds every `DSH_DESKTOP_WINDOWS_*` field from build and runtime-preparation children and passes only the certificate path, SignTool path, key container, and PIN into electron-builder. The signer supplies only validated signing fields in an otherwise scrubbed CMD environment; the CMD disables delayed expansion, clears those fields before SignTool starts, and preserves the PIN only in the required SignTool command line. Every surfaced diagnostic replaces the PIN, and only the dedicated build account and administrators may inspect the runner. The signer signs electron-builder's temporary NSIS bootstrap before enterprise Code Integrity evaluates that executable and clears a generated executable's certificate-table entry only when it points beyond the file before applying the final signature. Packaging fails before producing unsigned artifacts when the SignTool, certificate, container, PIN, token, or signature is unavailable. The custom protocol serves the installed frontend distribution plus client files named by the active module graph and rejects traversal or access outside those roots. The plugin installer API is available only to the Electron-owned management GUI and is absent from the browser application and backend RPC.
The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compressed and 120–165 MB installed before the seed store subset. Architecture-specific builds must report actual component-level size deltas.
Windows package invocations force `ELECTRON_BUILDER_7Z_FILTER=BCJ`. The bundled 7-Zip 24.09 encoder automatically selects ARM64 filters for ARM64 PE files, but the NSIS decoder from `nsis-resources-3.4.1` omits those entries during extraction. A native extraction probe with the actual NSIS plugin loses both `node-pty` ARM64 binaries under automatic filtering and restores both byte-for-byte with BCJ. Keeping a compatible filter preserves dependency contents and runtime integrity instead of removing architecture-specific files or weakening verification.
Local Windows installation testing uses an explicit `--unsigned` package invocation with the same build and runtime preparation. It strips certificate inputs, isolates artifacts in `unsigned-artifacts`, and omits updater configuration and the release completion record. The regular package command explicitly selects signed mode even when its parent environment requests unsigned mode. This separation permits installation diagnosis without an EV token while preventing local test output from qualifying for release upload.
NSIS extracts into its private `7z-out` directory before copying files into the application directory. Its default exit cleanup can overlap the backend's file reads after Finish launches the application. The [installer hook](../../../../apps/desktop/scripts/installer.nsh) removes only that extraction directory during `customInstall`, before interactive and silent launch branches. It preserves package archives, plugin DLLs, rollback directories, registers, and error status; the [native cleanup smoke](../../../../apps/desktop/tests/fixtures/installer-cleanup-smoke.nsi) checks these constraints. Moving cleanup into installation does not remove filesystem work, so total installation time and Finish-to-window time require separate measurements.
Direct `Nsis7z::Extract` into the application directory is not enabled. A native [locked-file probe](../../../../apps/desktop/tests/fixtures/installer-write-failure-smoke.nsi) leaves an old locked file beside a new asset while reporting no error; the staged `CopyFiles` operation sets the error flag for the same failed replacement. An unchanged 737,557,488-byte payload on Windows took 172.625 seconds through extraction, copying, and cleanup versus 28.031 seconds through direct extraction, but that single sample per path does not justify losing failure detection. The clocks exclude registry changes, old-version removal, and post-extraction verification; system caches were not cleared. Desktop-specific payload filtering reduces the copied file set while retaining the installer's replacement-error handling. This handling is not a promise of complete installation rollback.
Packaged applications ignore development resource and project environment overrides. Only an unpackaged Electron process can replace the Node.js binary, pnpm entry, dsh resources, or active project.
The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compressed and 120–165 MB installed before the dsh production tree. Architecture-specific builds must report actual component-level size deltas.
## Implementation
@@ -107,10 +107,10 @@ The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compr
|---|---|
| Shell | `apps/desktop` owns Electron windows, restricted preloads, the custom protocol, child lifecycle, project transactions, the plugin GUI, update coordination, and electron-builder configuration. |
| Installed runtime | Private `@deepseek-ai/dsh-desktop-host` boots the portless desktop composition from the active project and streams API and asset responses over validated framed byte pipes. |
| Package state | The release seed and every later mutation run through bundled Node.js and pnpm with desktop-owned store, config, cache, state, and home paths; core packages resolve from release tarballs while plugins resolve from the fixed npm registry. |
| Qualification | macOS packaging requires the configured company identity and notary credentials, verifies every native seed object after final archive extraction, verifies the completed application signature, and requires notarization plus Gatekeeper acceptance for both the application and DMG. Windows packaging requires the configured public certificate, SafeNet private-key container, Token Password, and SignTool, and verifies every produced signature. Update hosting, previous-version installed-artifact tests, and platform GUI recordings remain release-environment gates. |
| Package state | Bundled Node.js executes immutable core resources; bundled pnpm modifies only the external plugin graph in the Desktop profile. |
| Qualification | macOS packaging requires the configured company identity and notary credentials, verifies every native runtime file before inventory generation, verifies the completed application signature, and requires notarization plus Gatekeeper acceptance for both the application and DMG. Windows packaging requires the configured public certificate, SafeNet private-key container, Token Password, and SignTool, and verifies every produced signature. Update hosting, previous-version installed-artifact tests, and platform GUI recordings remain release-environment gates. |
`dev:desktop` builds the current workspace, projects the built CLI and private Desktop Host packages plus their dependency links into a disposable project, uses an isolated Harness home, opens the Main, Renderer, and Host debuggers, and starts unpackaged Electron without preparing release resources. Package mutation is disabled in this mode because its linked dependency graph is not a pnpm-installed desktop project. Fixed macOS arm64, macOS x64, and Windows x64 package commands pass one target through runtime preparation, seed installation, and electron-builder; each also has an unpacked-directory variant for release-path verification before installer generation.
`dev:desktop` builds the current workspace, projects the built CLI and private Desktop Host packages plus their dependency links into a disposable project, uses an isolated Harness home, opens the Main, Renderer, and Host debuggers, and starts unpackaged Electron without preparing release resources. Package mutation is disabled in this mode because its linked dependency graph is not a pnpm-installed desktop project. Fixed macOS arm64, macOS x64, and Windows x64 package commands pass one target through runtime preparation, dsh preparation, and electron-builder; each also has an unpacked-directory variant for release-path verification before installer generation.
## Alternatives considered
@@ -122,11 +122,11 @@ The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compr
**Reuse the existing CLI or browser plugin installer.** That crosses the desktop authorization and release scope and can use the user's package-manager state. Desktop package mutation remains exclusively Electron-owned.
**Let the desktop profile use CLI-managed packages or plugins.** Either product could change the other's dependency graph, Cordis version, plugin version, or native module. The desktop profile therefore owns a complete `node_modules` and rejects bundle resolution through the CLI profile fallback.
**Let the desktop profile use CLI-managed packages or plugins.** Either product could change the other’s dependency graph, Cordis version, plugin version, or native module. Desktop rejects package resolution through the CLI profile fallback.
**Install dsh and plugins into separate desktop projects.** This creates a second resolution anchor and peer-dependency fallback. One ordinary npm project already provides the required installation and resolution model.
**Separate core and plugin resolution without shared package ownership.** That permits duplicate host modules and uncontrolled peer fallback. The [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md) supplies explicit links and dependency validation for the separate resource and profile directories.
**Remove non-target Mach-O files from registry packages.** Architecture pruning saves a small amount of seed space, but packages can deliberately ship several architecture variants and callers can observe their installed file set. Signing every shipped Mach-O object satisfies notarization without inventing a Desktop-specific package layout.
**Remove non-target Mach-O files from registry packages.** Architecture pruning saves a small amount of runtime space, but packages can deliberately ship several architecture variants and callers can observe their installed file set. Signing every shipped Mach-O object satisfies notarization without inventing a Desktop-specific package layout.
**Export the Windows EV private key in a PFX file.** The externally supplied public leaf certificate lets SignTool construct the signature while `/csp` and `/kc` locate the hardware key. The EV private key remains non-exportable on the token.
@@ -136,15 +136,15 @@ The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compr
## Consequences
- A clean offline machine with no system Node.js or pnpm installs the seed into `.dsh/profiles/desktop` and starts a working dsh session.
- The signed application inventories a fixed small set of seed store shards instead of every pnpm cache file; every Mach-O object inside the macOS shards has the release Developer ID, secure timestamp, and hardened runtime, every Windows artifact has the configured hardware-backed EV signature, and the installed private store retains the ordinary pnpm layout.
- `.dsh/profiles/desktop/node_modules` contains and resolves the desktop dsh package and every GUI-installed desktop plugin.
- A clean offline machine without system Node.js or pnpm starts bundled dsh without installing core dependencies.
- The signed application inventories final runtime files; every macOS native file has the release Developer ID, secure timestamp, and hardened runtime, and every Windows artifact has the configured hardware-backed EV signature.
- `.dsh/profiles/desktop/node_modules` resolves shared host links and every GUI-installed desktop plugin.
- Every desktop pnpm operation uses the bundled executable and `.dsh/desktop/pnpm/store`; none reads user `PATH`, config, store, or profile `node_modules`.
- The Electron-only GUI installs, removes, and updates ordinary npm plugin packages without exposing raw pnpm arguments.
- The backend and browser application cannot mutate desktop packages.
- npm/CLI dsh and Electron never resolve or install plugins from each other's `node_modules`.
- The active backend and Web UI report the same dsh version and a compatible shell API before the product window opens.
- Failed installation, health checking, or update leaves the current profile usable or restores `rollback/profile` after restart.
- The active backend and Web UI report the same dsh version and a compatible shell API before the product UI loads.
- Package or Host failures retain partial profile changes and expose recovery controls; no automatic profile rollback is promised.
- One Desktop version binds Electron and dsh; every dsh update arrives through one Electron update dialog and one user-visible restart.
- Shared `.dsh` data rejects incompatible readers before migration or mutation.
- No loopback listener is opened, and the sandboxed renderer cannot access arbitrary filesystem or Electron APIs.
@@ -157,10 +157,10 @@ The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compr
| Decision | Recommendation |
|---|---|
| First launch | Bundle an offline seed store subset and install it through pnpm |
| Desktop profile | One Electron-owned reserved profile containing exact dsh and plugin dependencies |
| First launch | Check bundled release metadata and create profile links without installing core dependencies |
| Desktop profile | One Electron-owned reserved profile for external plugins and shared package links |
| Plugin management | Electron-only GUI and package service; no CLI, backend, or browser installation path |
| Activation | Staging project, complete health check, journaled directory replacement, one rollback copy |
| Activation | In-place package changes followed by actual Host startup |
| Initial platforms | macOS arm64/x64 and Windows x64; Linux has no supported release target |
| Update behavior | Background check, explicit confirmation before differential download and restart, startup dsh reconciliation |
@@ -168,10 +168,14 @@ The bundled upstream Node.js and pnpm are expected to add about 35–50 MB compr
Plugin lifecycle scripts execute third-party code. The allowed registry, package policy, exact versions, integrity, `allowBuilds`, and diagnostics require security review before GUI installation ships.
Updating the bound dsh can invalidate plugin peer dependencies or native modules. pnpm resolution and full-project health checking must reject the staged project before replacing the active one.
Updating the bound dsh can invalidate plugin peer dependencies or native modules. Reconciliation validates changed dependencies and rebuilds native packages; failures require explicit repair through the recovery UI.
An npm-installed dsh and desktop dsh may have different versions while sharing durable data. Each shared owner must enforce its format version and process lock before reading, migrating, or writing.
Directory replacement differs across operating systems and can be interrupted. The activation journal and installed-artifact fault tests must prove recovery at every filesystem move.
Interrupted package operations retain a pending marker. Installed-artifact tests must verify that a later launch retries the locked installation and approved native builds.
Code signing, notarization, and update hosting require production release infrastructure. Repository tests alone cannot complete that qualification.
## Related proposals
The [bundled-runtime decision](2026-09-08-desktop-bundled-runtime-and-external-plugins.md) supersedes this note’s offline runtime preparation and single-project package ownership. Release identity, signing, portless transport, process ownership, and Electron-only package authorization remain owned here.
@@ -4,19 +4,21 @@ Status: implemented
[English](2026-08-25-electron-desktop-packaging-and-updates.md) | 中文
profile 修改与恢复遵循[直接修改 profile 决策](2026-09-09-desktop-in-place-profile.zh.md)。
## 问题
DeepSeek Harness 需要一个复用 Web UI 的 Electron 桌面应用。该应用无需系统 Node.js 或 pnpm 即可工作,通过应用内置 pnpm 安装 dsh 与桌面插件,并通过一个面向用户的流程更新完整桌面发布。
桌面应用与通过 npm 安装的 dsh 共享 `.dsh` 数据根目录,但两者可能使用不同的 dsh 与插件版本。它们必须共享受支持的产品数据,同时不得共享可执行包、lockfile、`node_modules`、插件激活状态或包管理器配置。
当前 GUI 协议绑定 Web 客户端与后端版本。Electron 产物与其中通过 pnpm 安装的 dsh 如果独立定版本,就会产生未经验证的壳、客户端、后端与插件组合,也无法明确判断更新是否可用。
当前 GUI 协议绑定 Web 客户端与后端版本。Electron 产物与其中通过 内置 dsh 如果独立定版本,就会产生未经验证的壳、客户端、后端与插件组合,也无法明确判断更新是否可用。
## 决策
交付一个小型 Electron 壳,其中内置上游 Node.js 可执行文件和固定版本的 pnpm。Electron 把私有 Desktop Host 包作为隔离子进程启动;该包组合已安装的 dsh 后端与匹配的客户端图。Fetch 元数据及有界的原始请求与响应分块通过两条带版本的分帧字节管道传递,Node IPC 只承载就绪、致命失败和关闭,Electron 通过 `dsh-app://` 提供经过验证的资源;它不会打开监听端口。每个帧都包含固定标记、类型、单调 stream id、负载长度和经过验证的负载。串行 writer 遵守 pipe drain,请求或响应 stream 施加背压时 reader 会全局暂停,取消会关闭匹配的 stream,已退役 stream 的迟到响应帧保持无效。Connection 插件无需 `webServer` 即可提供与载体无关的 RPC 与 Fetch 注册表,Client Modules 则向 shell-owned carrier 提供与广告内容完全一致的组合 bundle 响应;Web 组合为两者挂载可选 HTTP route。渲染进程保留相同的 Fetch、RPC 与 Remote-stream 格式,子进程载体则避免 Base64 膨胀,也不依赖 Electron 与内置上游 Node.js 之间的 V8 序列化兼容性。发送 shutdown 后,Electron 会关闭自己持有的请求管道写端,以便在等待子进程退出前释放 Windows 上仍在进行的管道读取。该设计沿用 [GUI 分层与 RPC 协议 Agent Note](../../archived/architecture/2026-07-19-gui-layering-and-rpc-protocol.md)中的 Electron 预留。
Electron 拥有保留 profile `.dsh/profiles/desktop`。其中精确的 `@deepseek-ai/dsh` 依赖提供后端与匹配的 Web UI,匹配的私有 `@deepseek-ai/dsh-desktop-host` 依赖则只提供 Electron 子进程入口与组合 overlay。dsh 发布、私有 Host 及其第一方依赖闭包使用同一次源码构建生成的本地 npm tarball;profile manifest 把每个核心包列为本地 `file:` 依赖,`pnpm-workspace.yaml` 再通过 overrides 重复该映射。Host 不进入公共 CLI 包,也不会发布到 npm。桌面插件既是同一 profile 中来自 registry 的其他 npm 依赖,也是有序的 `dsh.profile.bundles` 条目,并从该 profile 唯一的 `node_modules` 解析。
Electron 拥有 `.dsh/profiles/desktop` 保留 profile。[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)负责核心资源存储、外部插件依赖、共享包链接和 profile 协调。私有 Desktop Host 保持独立于公共 CLI 包,且不会发布到 npm。
一个 Desktop 发布号同时标识 Electron 产物及其精确的 `@deepseek-ai/dsh` 与 `@deepseek-ai/dsh-desktop-host` 依赖。发布不能在构建或运行时选择不同的核心版本。因此,即使壳代码没有变化,更新 dsh 也必须产生新的 Electron 发布。
@@ -26,9 +28,9 @@ Electron 拥有保留 profile `.dsh/profiles/desktop`。其中精确的 `@deepse
| Owner | 职责 |
|---|---|
| Electron 壳 | 窗口与子进程生命周期、分帧字节管道、生命周期 IPC、自定义协议、保留 desktop profile、插件 GUI、更新协调、回滚 |
| Electron 壳 | 窗口与子进程生命周期、分帧字节管道、生命周期 IPC、自定义协议、保留 desktop profile、插件 GUI、更新协调 |
| 内置 Node.js 与 pnpm | 执行 dsh 并安装桌面项目的精确依赖,不读取用户 `PATH` 或 pnpm 状态 |
| Desktop profile | 为桌面 dsh 包与桌面插件提供一个依赖图、有序 bundle 列表和一个 `node_modules` |
| Desktop profile | 由内置运行时决策定义的外部插件依赖、已启用 bundle 顺序和共享链接 |
| 私有 Desktop Host 包 | 与 dsh 一起安装、但不进入公共 CLI 包或 npm 发布的 Electron 专用子进程入口与组合 overlay |
| 已安装 dsh 包 | 后端、匹配的 Web UI、启动 manifest、客户端包和产品行为 |
| 共享 `.dsh` owner | 会话、设置、凭据、工作区和存储,由其现有锁与格式版本保护 |
@@ -41,10 +43,6 @@ Electron 拥有保留 profile `.dsh/profiles/desktop`。其中精确的 `@deepse
```text
~/.dsh/
desktop/
staging/<transaction-id>/profile/
rollback/profile/
pending.json
lock
pnpm/
store/
cache/
@@ -54,50 +52,52 @@ Electron 拥有保留 profile `.dsh/profiles/desktop`。其中精确的 `@deepse
desktop/
package.json
pnpm-lock.yaml
lock
desktop-packages-pending
pnpm-workspace.yaml
desktop-release.json
desktop-packages.json
desktop-packages/
desktop-runtime-state.json
node_modules/
sessions/
storages/
```
`.dsh/profiles/desktop` 是唯一活跃的 desktop profile。其 package manifest 记录内置与已安装插件 bundle 的顺序;只有 Electron 可以修改它的依赖、lockfile 和 `node_modules`。生产启动会拒绝解析到该 profile 之外的 bundle,包括 CLI 维护的 `.dsh/profiles/node_modules` fallback。desktop profile 安装的所有包内容都使用 `.dsh/desktop/pnpm/store`。
`.dsh/profiles/desktop` 是唯一活动桌面 profile。其可执行包归属及允许解析的目录遵循[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)。插件包内容使用 `.dsh/desktop/pnpm/store`。
## 安装与解析
安装器绝不原地修改活跃 profile。它把 profile 元数据复制到事务暂存目录,并使用内置 pnpm 应用精确依赖变更。测试 staging 前,Electron 会停止活跃后端;它单独启动并停止 staging 后端,再在激活前恢复活跃后端,因此两个 Desktop 后端绝不会并发共享 `.dsh` 状态。激活过程再次停止后端,在对应目录移动前先持久化 `pending.json` 的每个下一阶段,把活跃 profile 移到 `rollback/profile`,把暂存 profile 移到 `.dsh/profiles/desktop`,然后重启。恢复过程会结合预写阶段与真实的 active、rollback 和 staging 目录,因此任一个写入与移动间隙中断后仍会保留或恢复一个完整 profile。
Desktop 在直接修改 profile 前停止 Host。包操作失败后保留部分修改,供显式修复;profile 修改和包重试的职责遵循[直接修改决策](2026-09-09-desktop-in-place-profile.zh.md)。
进程生命周期 Electron 锁是 Desktop 的权威 owner。包事务锁用于纵深防御,并记录仍能修改包状态的进程:包操作之间记录 Electron,pnpm 运行期间记录已生成的 pnpm PID。Owner 变更通过已经打开的排他锁文件完成截断、写入与同步。如果 Electron 在 pnpm 执行期间终止,后续进程会发现仍存活的 worker,并拒绝启动并发的 store 或 staging 事务;该 worker 退出后,陈旧 PID 才可以恢复。
进程生命周期 Electron 锁是 Desktop 的权威 owner。包事务锁用于纵深防御,并记录仍能修改包状态的进程:包操作之间记录 Electron,pnpm 运行期间记录已生成的 pnpm PID。Owner 变更通过已经打开的排他锁文件完成截断、写入与同步。如果 Electron 在 pnpm 执行期间终止,后续进程会发现仍存活的 worker,并拒绝启动并发的 包事务;该 worker 退出后,陈旧 PID 才可以恢复。
打包 seed 是离线安装包,而不是可执行 dsh 目录。它包含发布身份、初始桌面项目 manifest、分别以 dsh 和私有 Desktop Host 为根的第一方包闭包之并集的描述文件及不可变 tarball、lockfile、完整性清单和所需 store 子集。每个 `mac-arm64`、`mac-x64` 和 `win-x64` 构建都在 `.desktop-build/targets/<target>` 下持有自己的打包输入、运行时、包集合、seed、pnpm 准备状态、未打包应用、更新元数据和最终产物;只有不可变且经过校验和验证的 Node.js 下载缓存会被共享。发布构建要求 Electron 包、根 dsh 包与私有 Host 包使用相同版本,从正式源码构建生成最终 npm tarball,在本地打包私有 Host,选择可达的 dsh、Host 与 vendored 包以及 Landlock 入口,并验证 Host tarball 中包含 `lib/index.js` 与 `config/desktop.cordis.patch.yml`。Host 的 `files` manifest 只包含该运行入口与 overlay,并且该包不会发布到 npm。公共包 tarball 仍是由各包发布 manifest 控制的正式 `pnpm pack` 结果;Desktop 不删除已发布的声明文件,也不建立第二套包内容策略。seed manifest 把每个选中的包列为本地直接依赖,关闭 peer dependency 自动安装,workspace 文件再把每个选中的第一方包 override 到对应本地 tarball。目标 Node.js 执行内置 pnpm,因此 pnpm 的操作系统和 CPU 选择会使物化的依赖图与 seed 成为目标专用内容。内置 pnpm 关闭全局 virtual store,在禁用生命周期脚本的情况下从 npm 物化外部生产依赖,删除 `node_modules` 以及所有临时 pnpm cache、config 和 state 目录,然后只使用最终 store 执行一次干净的离线安装,并检查私有 Host 的入口与 overlay。构建会拒绝任何通过 registry 版本解析本地第一方包名的 lockfile。生成清单前会删除第二次生成的 `node_modules` 和临时 pnpm 项目注册。在复制 package set 前与离线安装后都要求这两个 Host 文件,可防止进程入口能够加载、却无法组合所需 overlay 的发布进入应用签名阶段。
种子根据规范化 store 路径,把 pnpm 内容放入 16 个确定性的未压缩 tar 分片。Apple 公证会检查这些归档内的 Mach-O 代码,因此 macOS seed 会 staging 每个被引用的内容寻址 Mach-O 对象,最多并发四个独立的 Developer ID 签名进程,并带上安全时间戳与 hardened runtime。任一签名失败后,准备过程会等待已启动的签名进程全部退出,原始 CAS 对象与包索引保持不变。所有签名成功后,准备过程把每个对象写到新的 SHA-512 路径,并以事务方式重写 pnpm MessagePack SQLite 索引内全部基础文件和 side-effects 文件引用。第二次离线安装证明 pnpm 可以解析重写后的 store;准备过程随后完成分片、解包最终归档并验证每个内嵌签名。包路径和非原生字节保持不变;种子保留包内附带的架构变体,因为删除文件会创建 Desktop 专属的包文件集。种子完整性覆盖分片 manifest 和解包前的每个归档。启动时验证归档路径、条目类型、唯一性和数量,把所有分片解包到唯一且由 Desktop 拥有的 staging 目录,替换匹配的不可变 store 文件,并以事务方式把各 pnpm store 版本的 SQLite `package_index` 合并进 `.dsh/desktop/pnpm/store`。Seed 记录替换匹配的键,为 Desktop 插件下载的记录继续保留。中断的文件合并可能留下有效的不可变缓存内容,但每次 SQLite 合并都是原子的,profile 安装与激活仍必须通过 pnpm 完整性与完整健康检查。
启动过程先要求安装包内的发布身份等于 Electron 应用版本,再在启动后端前比较 `.dsh/profiles/desktop/desktop-release.json`、已安装 dsh 包、已安装 Desktop Host 包与该发布版本。它在 staging 中通过 `pnpm install --offline --frozen-lockfile --trust-lockfile` 安装新的 seed manifest 与 lockfile。Electron 替换后,启动过程再通过一次离线 pnpm add,从桌面端现有 store 与元数据缓存恢复活跃 profile 记录的每个插件 bundle 精确版本。完整依赖图必须通过同一套健康检查才能激活。
插件 GUI 执行等价于 `pnpm add <package> --save-exact`、`pnpm remove <package>` 和精确版本更新的 registry npm 包操作。每次修改都保留本地核心包描述文件、tarball、dsh 与 Desktop Host 依赖和完整 override 映射。Electron 验证已安装包 manifest,并更新 profile 的依赖与有序 bundle 条目;任何渲染进程请求都不能选择 registry、安装目录、生命周期策略或任意 pnpm flag。
后端与 Loader 把 `.dsh/profiles/desktop/package.json` 作为 profile manifest 和 npm 解析锚点。公共 profile loader 先组合其中的有序 bundle 条目,再由私有 Desktop Host 应用其打包的 overlay。Host、dsh、Cordis、桌面插件、插件依赖和 peer dependency 均通过普通 pnpm `node_modules` 图解析。贡献 `dsh.client` 代码的桌面插件只有在完整 profile 通过健康检查后才进入启动 manifest。
核心物化、首次启动、插件安装和共享模块解析遵循[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)。实际 Host 启动时会组合已启用且提供 `dsh.client` 代码的桌面插件。
## 更新与恢复
Electron 更新只使用一个 `electron-updater` 发布流和签名 `electron-builder` 产物。该版本就是 Desktop 发布版本;不存在独立 dsh manifest、兼容范围或仅更新 dsh 的操作。前台安装会等待正在进行的后台检查,而不会把检查结果复用成安装结果。更新弹窗下载并安装 Electron 产物,然后重启进入新发布。
新发布在打开窗口前从安装包种子校准 dsh,同时保留已安装桌面插件。健康检查覆盖依赖解析、原生模块、壳 API 兼容性、后端启停、Web 资源和客户端启动图。不兼容插件会阻止激活,并保留上一个项目用于回滚。启动过程会明确失败,而不会运行版本不匹配的壳与 dsh。
[立即显示窗口决策](2026-09-09-desktop-immediate-window-and-direct-start.zh.md)负责本地加载页、直接启动 Host 和主窗口恢复。profile 协调遵循[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)。
`DSH_DESKTOP_AUTO_UPDATE_ENV` 默认为测试部署,也可以选择生产部署,并同时决定目标专用的 generic-provider URL 与 COS 目标。发布自动化通过 `DOWNLOAD_TEST_ORIGIN` 提供测试 HTTPS origin,并通过 `DOWNLOAD_TEST_COS_BUCKET` 或 `DOWNLOAD_PROD_COS_BUCKET` 提供各部署的 bucket;可变的测试路由与 COS 存储身份不写入源码,部署基础设施变更时无需发布新代码,而公开的生产 origin 仍固定。打包只解析公开更新 URL、禁止 electron-builder 发布、从子进程环境中删除每个 COS 凭据字段,并且只有在 electron-builder 以及每个签名或公证 hook 成功后才写入完成记录。目标上传还必须提供所选 bucket,随后会先要求完成记录、根 dsh 版本、Desktop 版本、根据版本得出的频道元数据、产物名称、大小与 SHA-512 全部一致,再读取所选凭据或发送数据。它先上传不可变且带版本的更新载荷与所有独立 blockmap,最后替换 electron-builder 生成的频道元数据,并且不会删除历史对象。稳定版本使用 `latest` 元数据名称,预发布版本则使用语义化版本的第一个预发布标识符。NSIS 把 blockmap 嵌入已签名的可执行文件,macOS ZIP 则使用独立 blockmap;两者都让 electron-updater 在平台支持时只下载变化的数据块,而应用替换与本地 pnpm staging 事务仍是两个独立操作。
`DSH_DESKTOP_AUTO_UPDATE_ENV` 默认为测试部署,也可以选择生产部署,并同时决定目标专用的 generic-provider URL 与 COS 目标。发布自动化通过 `DOWNLOAD_TEST_ORIGIN` 提供测试 HTTPS origin,并通过 `DOWNLOAD_TEST_COS_BUCKET` 或 `DOWNLOAD_PROD_COS_BUCKET` 提供各部署的 bucket;可变的测试路由与 COS 存储身份不写入源码,部署基础设施变更时无需发布新代码,而公开的生产 origin 仍固定。打包只解析公开更新 URL、禁止 electron-builder 发布、从子进程环境中删除每个 COS 凭据字段,并且只有在 electron-builder 以及每个签名或公证 hook 成功后才写入完成记录。目标上传还必须提供所选 bucket,随后会先要求完成记录、根 dsh 版本、Desktop 版本、根据版本得出的频道元数据、产物名称、大小与 SHA-512 全部一致,再读取所选凭据或发送数据。它先上传不可变且带版本的更新载荷与所有独立 blockmap,最后替换 electron-builder 生成的频道元数据,并且不会删除历史对象。稳定版本使用 `latest` 元数据名称,预发布版本则使用语义化版本的第一个预发布标识符。NSIS 把 blockmap 嵌入已签名的可执行文件,macOS ZIP 则使用独立 blockmap;两者都让 electron-updater 在平台支持时只下载变化的数据块,而应用替换与本地 pnpm 包操作仍是两个独立操作。
## 安全与发布策略
核心 dsh 与私有 Desktop Host 只能来自签名 Electron 发布内经过完整性记录的本地 npm tarball;pnpm overrides 防止传递核心包回退到 registry。Store 归档经过完整性检查,并在隔离的解包目录中完成全部验证,归档文件随后才能进入可写包状态。插件安装接受桌面策略允许的 registry 包 spec,但绝不接受原始 pnpm 命令。激活前必须具备精确版本、lockfile 完整性、经过评审的 `allowBuilds` 集合、仅限用户的目录权限、遮盖后的诊断和健康检查。
核心 dsh 和私有 Desktop Host 只来自签名应用的资源树。插件安装接受桌面策略允许的 registry 包规格,不接受原始 pnpm 命令。激活前要求精确版本、锁文件完整性、经过审查的 `allowBuilds` 集合和仅限用户访问的目录权限。
Electron 产物必须签名;macOS 产物必须公证。发布自动化必须通过明确的环境变量提供应用 ID、macOS Developer ID 限定名、预期 Team ID 与一套完整的 notarytool 凭据。配置加载会拒绝缺失或格式错误的标识符和不完整的公证凭据,macOS 打包还会强制签名,避免证书发现过程静默选择其他已安装身份或生成未签名发布。Seed 准备会验证每个内嵌 Mach-O 文件的精确 Authority 与 Team ID,以及时间戳和 hardened-runtime 标记。签名后钩子会执行 Apple 的深度严格应用验证,并要求同一叶证书 Authority 与 Team ID 完全匹配,验证通过后才继续生成产物。Electron-builder 随后公证应用并钉票、签署 DMG。DMG 的 artifact-completion hook 会单独公证每个 DMG 并钉票,再要求其使用配置的身份、具备有效票据并通过 Gatekeeper;只有该 hook 成功,上传事件才会执行。macOS 更新使用签名 ZIP,因此 DMG 不生成 blockmap;否则钉票会让已经生成的 DMG blockmap 失效。自定义协议提供已安装的前端分发目录和活跃模块图点名的客户端文件,并拒绝路径穿越或访问这些根目录之外的内容。插件安装器 API 只对 Electron 拥有的管理 GUI 可用,不存在于浏览器应用或后端 RPC 中。
Electron 发布产物必须签名;macOS 产物必须公证。发布自动化必须通过明确的环境变量提供应用 ID、macOS Developer ID 限定名、预期 Team ID 与一套完整的 notarytool 凭据。配置加载会拒绝缺失或格式错误的标识符和不完整的公证凭据,macOS 打包还会强制签名,避免证书发现过程静默选择其他已安装身份或生成未签名发布。运行时准备会验证每个内嵌 Mach-O 文件的精确 Authority 与 Team ID,以及时间戳和 hardened-runtime 标记。签名后钩子会执行 Apple 的深度严格应用验证,并要求同一叶证书 Authority 与 Team ID 完全匹配,验证通过后才继续生成产物。固定目标安装包命令使用[隔离的 App 副本并行公证](../process/2026-09-09-parallel-macos-notarization.zh.md):ZIP 包含已钉票的 App,签名 DMG 则携带覆盖其中未钉票 App 的票据。DMG 的 artifact-completion hook 要求其使用配置的身份、具备有效票据并通过 Gatekeeper。只有两条产物流都成功,命令才会移入其输出并写入发布完成记录;仅生成目录的命令仍会公证 App 并钉票。macOS 更新使用签名 ZIP,因此 DMG 不生成 blockmap;否则钉票会让已经生成的 DMG blockmap 失效。自定义协议提供已安装的前端分发目录和活跃模块图点名的客户端文件,并拒绝路径穿越或访问这些根目录之外的内容。插件安装器 API 只对 Electron 拥有的管理 GUI 可用,不存在于浏览器应用或后端 RPC 中。
Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool 提供 `DSH_DESKTOP_WINDOWS_CER_FILE` 指定的公开 EV 叶证书,并通过必需的 `DSH_DESKTOP_WINDOWS_KEY_CONTAINER` 标识匹配的私钥。证书文件保留在源码仓库之外,私钥仍留在 USB Token 上。electron-builder hook 把每个产物交给采用 CRLF 的 `windows-sign.cmd`;该 CMD 只调用一次 SignTool,并指定 SafeNet `/kc "[{{PIN}}]=容器"` 值与 CSP、SHA-256 文件摘要和 DigiCert SHA-256 RFC 3161 时间戳。hook 不会改用其他 SignTool,也不会重试失败的请求。打包编排不会把任何 `DSH_DESKTOP_WINDOWS_*` 字段传给构建与 seed 准备子进程,只会把证书路径、SignTool 路径、密钥容器和 PIN 传入 electron-builder。签名器在已清理的 CMD 环境中只提供经过校验的签名字段;CMD 会禁用延迟展开,在 SignTool 启动前清除这些字段,并仅在 SignTool 必需的命令行中保留 PIN。所有对外诊断都会替换 PIN,而且只能允许专用构建账号和管理员检查该 runner。签名器会在企业 Code Integrity 检查 electron-builder 的临时 NSIS bootstrap 前先为该可执行文件签名;对于生成的可执行文件,只有证书表条目指向文件末尾之外时,才会在最终签名前清除该条目。SignTool、证书、容器、PIN、Token 或签名不可用时,打包会在产生未签名产物前失败。自定义协议提供已安装的前端分发目录和活跃模块图点名的客户端文件,并拒绝路径穿越或访问这些根目录之外的内容。插件安装器 API 只对 Electron 持有的管理 GUI 可用,不存在于浏览器应用或后端 RPC 中。
[固定版本的 osx-sign 补丁](../../../../patches/@[email protected])在两种已发布模块构建中使用 `lstat`,因此 Framework 的文件和目录别名不会触发重复签名。选定的上游版本能够跳过这些别名前,仍需保留该补丁。PAK 文件由外层 bundle 签名记录完整性;逐个签名会增加串行时间戳请求,但不会增加资源完整性保护。Desktop 保留全部语言文件,只跳过其单独签名。可执行代码仍使用 Developer ID 签名、安全时间戳和 hardened runtime。[签名器遍历回归测试](../../../../apps/desktop/tests/macos-signing-walk.spec.ts)使用真实 Framework 别名执行已安装依赖;发布验收仍要求严格应用验证、公证和启动。
打包应用会忽略开发资源和项目环境变量覆盖。只有未打包的 Electron 进程可以替换 Node.js 可执行文件、pnpm 入口、seed 或活跃项目。
Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool 提供 `DSH_DESKTOP_WINDOWS_CER_FILE` 指定的公开 EV 叶证书,并通过必需的 `DSH_DESKTOP_WINDOWS_KEY_CONTAINER` 标识匹配的私钥。证书文件保留在源码仓库之外,私钥仍留在 USB Token 上。electron-builder hook 把每个产物交给采用 CRLF 的 `windows-sign.cmd`;该 CMD 只调用一次 SignTool,并指定 SafeNet `/kc "[{{PIN}}]=容器"` 值与 CSP、SHA-256 文件摘要和 DigiCert SHA-256 RFC 3161 时间戳。hook 不会改用其他 SignTool,也不会重试失败的请求。打包编排不会把任何 `DSH_DESKTOP_WINDOWS_*` 字段传给构建与 运行时准备子进程,只会把证书路径、SignTool 路径、密钥容器和 PIN 传入 electron-builder。签名器在已清理的 CMD 环境中只提供经过校验的签名字段;CMD 会禁用延迟展开,在 SignTool 启动前清除这些字段,并仅在 SignTool 必需的命令行中保留 PIN。所有对外诊断都会替换 PIN,而且只能允许专用构建账号和管理员检查该 runner。签名器会在企业 Code Integrity 检查 electron-builder 的临时 NSIS bootstrap 前先为该可执行文件签名;对于生成的可执行文件,只有证书表条目指向文件末尾之外时,才会在最终签名前清除该条目。SignTool、证书、容器、PIN、Token 或签名不可用时,打包会在产生未签名产物前失败。自定义协议提供已安装的前端分发目录和活跃模块图点名的客户端文件,并拒绝路径穿越或访问这些根目录之外的内容。插件安装器 API 只对 Electron 持有的管理 GUI 可用,不存在于浏览器应用或后端 RPC 中。
Windows 打包调用强制设置 `ELECTRON_BUILDER_7Z_FILTER=BCJ`。内置的 7-Zip 24.09 编码器会为 ARM64 PE 文件自动选择 ARM64 过滤器,但 `nsis-resources-3.4.1` 中的 NSIS 解码器会在解压时遗漏这些条目。使用实际 NSIS 插件的原生解压验证表明,自动过滤会丢失两个 `node-pty` ARM64 二进制文件,而 BCJ 可以逐字节还原二者。使用兼容的过滤器能够保留依赖内容与运行时完整性,无需删除特定架构的文件或削弱校验。
本地 Windows 安装测试使用显式的 `--unsigned` 打包调用,并执行相同的构建和运行时准备。它清除证书输入,将产物隔离到 `unsigned-artifacts`,并省略更新器配置和发布完成记录。即使父进程环境请求未签名模式,常规打包命令也会显式选择签名模式。这样既能在没有 EV Token 时诊断安装问题,也能防止本地测试产物通过发布上传校验。
NSIS 先解压到私有的 `7z-out` 目录,再把文件复制到应用目录。Finish 启动应用后,默认退出清理可能与后端的文件读取重叠。[安装器 hook](../../../../apps/desktop/scripts/installer.nsh) 在 `customInstall` 阶段仅删除该解压目录,早于交互和静默启动分支。它保留包归档、插件 DLL、回滚目录、寄存器和错误状态;[原生清理 smoke](../../../../apps/desktop/tests/fixtures/installer-cleanup-smoke.nsi) 检查这些约束。把清理移入安装阶段并不会减少文件系统工作,因此必须分别测量安装总耗时与点击 Finish 到窗口出现的耗时。
安装器不开启直接向应用目录执行 `Nsis7z::Extract`。原生[文件占用探针](../../../../apps/desktop/tests/fixtures/installer-write-failure-smoke.nsi)会在未报错的情况下留下被占用的旧文件和新资源;暂存后执行的 `CopyFiles` 在相同替换失败时会设置错误标志。Windows 上同一份 737,557,488 字节载荷经过解压、复制和清理耗时 172.625 秒,直接解压耗时 28.031 秒,但每条路径的单次样本不足以支持放弃失败检测。计时不包括注册表修改、旧版删除及解压后的验证,也没有清空系统缓存。桌面专用载荷过滤减少需要复制的文件,同时保留安装器的替换错误处理。这种处理并不承诺完整的安装回滚。
打包应用会忽略开发资源和项目环境变量覆盖。只有未打包的 Electron 进程可以替换 Node.js 可执行文件、pnpm 入口、dsh 资源 或活跃项目。
在种子 store 子集之外,内置上游 Node.js 与 pnpm 预计增加约 35–50 MB 压缩体积和 120–165 MB 安装体积。分架构构建必须报告实际组件级体积增量。
@@ -107,10 +107,10 @@ Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool
|---|---|
| 壳 | `apps/desktop` 负责 Electron 窗口、受限 preload、自定义协议、子进程生命周期、项目事务、插件 GUI、更新协调和 electron-builder 配置。 |
| 已安装运行时 | 私有 `@deepseek-ai/dsh-desktop-host` 从活跃项目启动无端口桌面组合,并通过经过验证的分帧字节管道流式传输 API 与资源响应。 |
| 包状态 | 发布种子和后续每次修改都通过内置 Node.js 与 pnpm 执行,并使用桌面端拥有的 store、config、cache、state 和 home 路径;核心包从发布 tarball 解析,插件从固定 npm registry 解析。 |
| 资格验证 | macOS 打包要求已配置的公司身份与公证凭据可用,在解包最终归档后验证每个原生 seed 对象,验证完整应用签名,并要求应用和 DMG 都完成公证且通过 Gatekeeper。Windows 打包要求已配置的公开证书、SafeNet 私钥容器、Token Password 与 SignTool,并验证生成的每个签名。更新托管、跨上一版本的已安装产物测试和各平台 GUI 录制仍是发布环境门槛。 |
| 包状态 | 内置 Node.js 执行不可变核心资源;内置 pnpm 只修改 Desktop profile 中的外部插件依赖图。 |
| 资格验证 | macOS 打包要求已配置的公司身份与公证凭据可用,在生成清单前验证每个原生运行时文件,验证完整应用签名,并要求应用和 DMG 都完成公证且通过 Gatekeeper。Windows 打包要求已配置的公开证书、SafeNet 私钥容器、Token Password 与 SignTool,并验证生成的每个签名。更新托管、跨上一版本的已安装产物测试和各平台 GUI 录制仍是发布环境门槛。 |
`dev:desktop` 会构建当前 workspace,把已构建 CLI 包、私有 Desktop Host 包及其依赖链接投影为一次性项目,使用隔离的 Harness home,打开 Main、Renderer 和 Host 调试器,并在不准备发布资源的情况下启动未打包 Electron。该模式的链接依赖图不是由 pnpm 安装的桌面项目,因此会禁用包修改。固定的 macOS arm64、macOS x64 与 Windows x64 打包命令会把同一目标传给运行时准备、seed 安装和 electron-builder;每条命令还提供未封装安装器的变体,用于在生成安装器前验证发布路径。
`dev:desktop` 会构建当前 workspace,把已构建 CLI 包、私有 Desktop Host 包及其依赖链接投影为一次性项目,使用隔离的 Harness home,打开 Main、Renderer 和 Host 调试器,并在不准备发布资源的情况下启动未打包 Electron。该模式的链接依赖图不是由 pnpm 安装的桌面项目,因此会禁用包修改。固定的 macOS arm64、macOS x64 与 Windows x64 打包命令会把同一目标传给运行时准备、dsh 准备和 electron-builder;每条命令还提供未封装安装器的变体,用于在生成安装器前验证发布路径。
## 考虑过的替代方案
@@ -122,11 +122,11 @@ Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool
**复用现有 CLI 或浏览器插件安装器。** 这会跨越桌面授权与发布 scope,并可能使用用户的包管理器状态。桌面包修改完全由 Electron 拥有。
**让 desktop profile 使用 CLI 管理的包或插件。** 任一产品都可能改变另一方的依赖图、Cordis 版本、插件版本或原生模块。因此 desktop profile 持有完整 `node_modules`,并拒绝通过 CLI profile fallback 解析 bundle。
**让桌面 profile 使用 CLI 管理的包或插件。** 任一产品都可能改变另一产品的依赖图、Cordis 版本、插件版本或原生模块。Desktop 拒绝通过 CLI profile 回退目录解析包。
**把 dsh 与插件安装到不同桌面项目。** 这会产生第二解析锚点和 peer dependency 回退。一个普通 npm 项目已经提供所需安装与解析模型。
**分离核心与插件解析,却不明确共享包归属。** 这会允许宿主模块重复以及不可控的 peer 回退。[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)为分离的资源和 profile 目录提供明确链接及依赖验证。
**从 registry 包删除非目标 Mach-O 文件。** 架构裁剪可以节省少量 seed 空间,但包可能有意附带多个架构变体,调用方也可以观察安装后的文件集。签署每个实际携带的 Mach-O 对象,无需发明 Desktop 专属包布局就能满足公证要求。
**从 registry 包删除非目标 Mach-O 文件。** 架构裁剪可以节省少量 运行时空间,但包可能有意附带多个架构变体,调用方也可以观察安装后的文件集。签署每个实际携带的 Mach-O 对象,无需发明 Desktop 专属包布局就能满足公证要求。
**把 Windows EV 私钥导出到 PFX 文件。** 外部提供的公开叶证书让 SignTool 构造签名,`/csp` 与 `/kc` 则定位硬件密钥。EV 私钥保持不可导出,并留在 Token 上。
@@ -136,15 +136,15 @@ Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool
## 结果
- 没有系统 Node.js 或 pnpm 的干净离线机器把种子安装进 `.dsh/profiles/desktop`,并启动可工作的 dsh 会话。
- 已签名应用记录固定少量的 seed store 分片,而不是记录每个 pnpm 缓存文件;macOS 分片内每个 Mach-O 对象都带有发布 Developer ID、安全时间戳与 hardened runtime,每个 Windows 产物都带有配置的硬件支持 EV 签名,安装后的私有 store 仍保持普通 pnpm 布局。
- `.dsh/profiles/desktop/node_modules` 包含并解析桌面 dsh 包和每个 GUI 安装的桌面插件。
- 没有系统 Node.js 或 pnpm 的干净离线机器能够启动内置 dsh,无需安装核心依赖。
- 签名应用记录最终运行时文件清单;每个 macOS 原生文件都具有发布 Developer ID、安全时间戳和 hardened runtime,每个 Windows 产物都具有配置的硬件 EV 签名。
- `.dsh/profiles/desktop/node_modules` 能解析共享宿主链接和每个通过 GUI 安装的桌面插件。
- 每个桌面 pnpm 操作都使用内置可执行文件和 `.dsh/desktop/pnpm/store`;不读取用户 `PATH`、配置、store 或 profile `node_modules`。
- Electron-only GUI 安装、删除和更新普通 npm 插件包,而不暴露原始 pnpm 参数。
- 后端与浏览器应用不能修改桌面包。
- npm/CLI dsh 与 Electron 绝不从对方的 `node_modules` 解析或安装插件。
- 在产品窗口打开前,活跃后端与 Web UI 报告相同 dsh 版本和兼容壳 API。
- 安装、健康检查或更新失败后,当前 profile 仍然可用,或在重启后恢复 `rollback/profile`。
- 在产品 UI 加载前,活跃后端与 Web UI 报告相同 dsh 版本和兼容壳 API。
- 包操作或 Host 失败后保留部分 profile 修改,并提供恢复控件;不承诺自动回滚 profile。
- 一个 Desktop 版本绑定 Electron 与 dsh;每次 dsh 更新都通过一个 Electron 更新弹窗交付,并产生一次用户可见的重启。
- 共享 `.dsh` 数据在迁移或修改前拒绝不兼容的读取方。
- 不打开回环监听端口,沙箱渲染进程不能访问任意文件系统或 Electron API。
@@ -157,10 +157,10 @@ Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool
| 决策 | 建议 |
|---|---|
| 首次启动 | 打包离线种子 store 子集,并通过 pnpm 安装 |
| Desktop profile | 一个包含精确 dsh 与插件依赖、由 Electron 持有的保留 profile |
| 首次启动 | 检查内置发布元数据并创建 profile 链接,不安装核心依赖 |
| 桌面 profile | 一个由 Electron 拥有的保留 profile,保存外部插件和共享包链接 |
| 插件管理 | Electron-only GUI 与包服务;没有 CLI、后端或浏览器安装路径 |
| 激活 | 暂存项目、完整健康检查、记录式目录替换和一个回滚副本 |
| 激活 | 直接修改包后启动实际 Host |
| 初始平台 | macOS arm64/x64 与 Windows x64;Linux 尚无受支持的发布目标 |
| 更新行为 | 后台检查,差分下载与重启前显式确认,启动时校准 dsh |
@@ -168,10 +168,14 @@ Windows 发布打包通过 `/f` 向已配置且与 SafeNet 兼容的 SignTool
插件生命周期脚本会执行第三方代码。在 GUI 安装功能交付前,获准 registry、包策略、精确版本、完整性、`allowBuilds` 和诊断都需要安全评审。
更新绑定的 dsh 可能使插件 peer dependency 或原生模块失效。pnpm 解析与完整项目健康检查必须在替换活跃项目前拒绝暂存项目。
更新绑定的 dsh 可能使插件 peer dependency 或原生模块失效。校准会验证变化的依赖并重建原生包;失败后需要通过恢复 UI 显式修复。
通过 npm 安装的 dsh 与桌面 dsh 可能在共享持久化数据时使用不同版本。每个共享 owner 都必须在读取、迁移或写入前执行格式版本与进程锁。
不同操作系统的目录替换行为不同,而且可能中断。激活记录与已安装产物故障测试必须证明每次文件系统移动都可以恢复。
中断的包操作保留未完成标记。已安装产物测试必须验证后续启动会重试锁定依赖的安装和获准的原生构建。
代码签名、公证和更新托管需要生产发布基础设施。只运行仓库测试不能完成这些认证。
## 相关提案
[内置运行时决策](2026-09-08-desktop-bundled-runtime-and-external-plugins.zh.md)取代本记录的离线 运行时准备与单项目包归属决策。发布身份、签名、无端口传输、进程归属及仅限 Electron 的包授权仍由本记录负责。
@@ -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-adjacent-agent-steer-messaging.md
2026-08-27-adjacent-agent-steer-messaging.md: 7bd97d5c992acac240a28d61d27e35e1f1058ae3
2026-08-27-adjacent-agent-steer-messaging.md: 47d9bd05c24871a14f78ebfa0afb8c13a6c2754d
2026-08-27-adjacent-agent-steer-messaging.zh.md: 1fd0fc39b937287ac34a2862c7cb9cde90256c09
@@ -6,7 +6,7 @@ English | [中文](2026-08-27-adjacent-agent-steer-messaging.zh.md)
## Problem
Continuable Agents originally used direction-specific model controls. A parent called `send_message({ subagent_id, message })`, which delegated to a FIFO `followup` service operation. A child instead received a child-scoped `report({ output })` tool, a `tool:report` system-prompt section, and deployment-selected quiet or waking delivery. The tools described one adjacent-Agent operation through different schemas, service paths, provenance, and scheduling.
Continuable Agents originally used direction-specific model controls. A parent called `send_message({ subagent_id, message })`, which delegated to a FIFO `followup` service operation. A child instead received a child-scoped `report({ output })` tool, a `tool:report` system-prompt section, and deployment-selected quiet or waking delivery. The tools described one adjacent-Agent operation through different schemas, service paths, source attribution, and scheduling.
A continuable child owns its own Session, so its parent does not automatically receive the child's transcript, tool output, or reasoning. The return path must therefore remain explicit and repeatable: a child may send progress before it finishes, remain available after sending, or fail before it can cooperate. Turning every final assistant message into an implicit result would conflate turn completion with model-selected communication and would not cover abnormal endings.
@@ -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: 67927a9d1404e0b14c6e420cc5bea462be5c87ad
2026-08-27-outbound-proxy-policy.zh.md: cf0b203a061522450a2a1b0c337a060b0c387684
2026-08-27-outbound-proxy-policy.md: e588ff751fd18762fbe5e24aa17f9bbbc4ff8deb
2026-08-27-outbound-proxy-policy.zh.md: 57faea256120f5f76ecc4dab1bd8709003672fd0
@@ -6,9 +6,9 @@ English | [中文](2026-08-27-outbound-proxy-policy.zh.md)
## Problem
Node's built-in `fetch` ignores `HTTP_PROXY` and `HTTPS_PROXY`. Every other tool a developer runs — curl, git, npm, pip — honours them, so a user behind a proxy exports the variables once and expects everything to follow. The harness did not: `setGlobalDispatcher`, `ProxyAgent`, and `EnvHttpProxyAgent` appeared zero times across `packages/` and `apps/`, so the model request, every web search, `web_fetch`, MCP over HTTP, the OTLP exporter, and the E2B SDK all connected directly, silently, with no diagnostic anywhere.
Node's built-in `fetch` ignores `HTTP_PROXY` and `HTTPS_PROXY`. Every other tool a developer runs — curl, git, npm, pip — honours them, so a user behind a proxy exports the variables once and expects everything to follow. The harness did not: `setGlobalDispatcher`, `ProxyAgent`, and `EnvHttpProxyAgent` appeared zero times across `packages/` and `apps/`, so the model request, every web search, `web_fetch`, MCP over HTTP, and the OTLP exporter all connected directly, silently, with no diagnostic anywhere.
The repository had briefly had an answer and lost it without noticing. PR #971 set `NODE_USE_ENV_PROXY=1` in `bin/dsh`; eleven days later `bbb1b1cc38 cleanup: remove managed source installer` deleted that launcher wholesale, taking the flag with it. What survived was one sentence in `apps/cli/reference/README.md` telling the reader to set a variable that nothing consumed any more.
The repository had briefly had an answer and lost it without noticing. PR #971 set `NODE_USE_ENV_PROXY=1` in `bin/dsh`; eleven days later the “cleanup: remove managed source installer” change deleted that launcher wholesale, taking the flag with it. What survived was one sentence in `apps/cli/reference/README.md` telling the reader to set a variable that nothing consumed any more.
That sentence could not have worked anyway, for three measured reasons. `NODE_USE_ENV_PROXY` samples the environment at process start, while `loadLayeredEnv()` merges the `.env` layers afterwards, so a proxy declared in `$DSH_HOME/.env` is invisible to it. It reaches Node 24.0+ and, on the 22 line, only 22.21+ — while `engines` admits `^22.19.0`, where the variable does not exist and setting it warns about nothing. And it does not reach `web-fetch-http` at all: that provider passes its own `dispatcher` to `fetch`, and an explicit dispatcher overrides the global one whatever the flag says.
@@ -24,9 +24,7 @@ An earlier revision put it in a new `net/` package group, reasoning that dependi
The plugin that revision shipped is gone with it. It let a composition declare the policy in `cordis.yml`, but no shipped bundle mounted it, so the launcher's path was the only reachable one — and its `Config` was the sole supplier of a configuration branch nothing else could reach.
**Four functions, because the call sites converged rather than the package growing an export each.** An earlier revision exported six: a dispatcher factory, a `node:http` agent factory, a proxy-URL lookup, a policy accessor, an installer, and a child-environment builder. Each existed for one SDK's transport, which is how a transport-policy package turns into a catalogue of other packages' constraints. Review asked whether the call sites could converge instead; they could, and each removal took a whole shape with it. Telemetry stopped being routed at all, retiring the `node:http` factory. `web-fetch-http` builds its own pinning agent under an annotated exemption, retiring the dispatcher factory. E2B reads `route.proxy`, retiring the proxy-URL lookup.
What remains is `installProxyFromEnvironment`, `proxyRouteFor`, `proxyEnvironmentForChild`, and `clearedProxyEnv` — one per way a caller can need the policy, none per SDK. Installation absorbed resolution and diagnostic reporting, which no caller needed apart: a resolved policy that is not installed routes nothing.
**One operation per caller need.** `installProxyFromEnvironment`, `proxyRouteFor`, `proxyEnvironmentForChild`, and `clearedProxyEnv` cover installation, per-request routing, child inheritance and fixture isolation. SDK-specific factories would expose individual transport constraints through the shared API. `web-fetch-http` consumes the resolved route when constructing its pinning transport; the OTLP exporter remains direct. Installation includes resolution and diagnostic reporting because callers need one operation that resolves and installs routing.
`proxyRouteFor` also closes a defect the old accessor made expressible. `web-fetch-http` read the policy to decide whether to pin, then read it again to build a transport; an unmount between the two returned a direct, unpinned agent for a URL the first read had cleared as proxied. A route carries both, so the branch and the request cannot disagree. Its dispatcher is the process-wide one, closed rather than destroyed on disposal, so a request already in flight when a policy is unmounted still finishes.
@@ -42,19 +40,19 @@ 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 code 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.
This accepts a documented seam. Such a context matches bypass entries by Node's rules, which differ from this package's in separators and IPv4-range support, and the flag exists only on Node 22.21+ and 24+.
**Two SDKs do not reach `globalThis.fetch`, and reading their code said otherwise.** The audit first classified the OTLP exporter and the E2B SDK as covered, on a grep that found `globalThis.fetch` in `@opentelemetry/otlp-exporter-base`. That match is the *browser* transport; on Node the delegate selects `http-exporter-transport`, which posts through `node:http` — where a global dispatcher does not reach. E2B is a second shape again: it builds its own undici `Agent`/`ProxyAgent` and takes a `proxy` URL that it never reads from the environment. Both were measured direct. E2B is handed `route.proxy` from `proxyRouteFor`, the same call `web-fetch-http` makes. Telemetry is deliberately left direct, and that exclusion is the more interesting half.
**SDK transports need independent verification.** The OTLP exporter selects `http-exporter-transport` on Node, which posts through `node:http` and bypasses the global fetch dispatcher; a `globalThis.fetch` reference in its browser implementation does not prove Node routing. The [E2B removal](../simplification/2026-09-11-remove-e2b-providers.md) retires a second SDK transport integration without changing this requirement.
**Telemetry stays direct on purpose.** Routing it needs one of two things, and both cost more than the channel is worth. An `http.Agent` reads the environment through `proxyEnv`, which arrived in Node 22.21 and 24.5 — inside the engines range, so 22.19, 22.20, and 24.0–24.4 would stay direct regardless, and the proxy package would have to keep a `createNodeHttpAgent` export for a path that works on some runtimes. Replacing the transport with the SDK's `fetch` delegate covers every runtime, but that delegate has no compression, and the shipped `base` bundle enables gzip: a realistic OTLP batch measures 6.4x smaller with it. An attempt that refused `exporter.compression` instead broke every test that boots the shipped bundle, and one that gzipped at the serializer worked but put transport code in a telemetry plugin to keep it working.
Weighed against that, telemetry is the one outbound channel whose loss costs the user nothing: no tool, no model request, and no session depends on it, and an export that cannot connect is already dropped silently. A user behind a mandatory proxy is left exactly where they were before this change rather than regressed. `egress.spec.ts` now asserts the exclusion — an SDK upgrade that moved the exporter onto `fetch` would start routing telemetry through a proxy silently, and that case is what makes it visible.
**Every call site carries an egress test, because reading the code was not enough.** `egress.spec.ts` in each owning package drives that site's real code path at an unresolvable `.invalid` host through a fake proxy and asserts the proxy saw the request. Nine of them cover the search backends, pi-ai discovery, MCP over HTTP, E2B, a spawned child Node, a worker thread, and telemetry's exclusion. The gate below cannot see inside a dependency; these can, and they are what turns "an SDK changed its transport" from a silent regression into a failing test.
**Every call site carries an egress test.** Each owning package's `egress.spec.ts` drives its actual transport through a fake proxy and checks the observed route. These tests cover search backends, pi-ai discovery, MCP over HTTP, child Node processes, worker threads and telemetry's direct-route exception. They detect dependency transport changes that a static call-site check cannot observe.
**A gate keeps the defect from returning.** `verify-no-bare-dispatcher` parses the TypeScript AST — `scripts/AGENTS.md` requires syntax-aware discovery, and a line-wise regex missed both the `{ dispatcher }` shorthand this repository already uses and a `new Alias(...)` behind a renamed import. It rejects an undici agent construction and an explicit `dispatcher` option outside the owning package. `proxyRouteFor(url)` is the sanctioned replacement, and the one call site that genuinely owns its transport — `web-fetch-http`, pinning a request to addresses it validated — says so with a `proxy-exempt:` comment. The rule exists because `web-fetch-http`'s original `new Agent` was entirely reasonable when it was written — proxying simply did not exist yet, and nothing would have caught it.
@@ -72,7 +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 the `code-runtime` worker the proxy too.** Rejected. Model-authored programs run there with no ambient environment at all — a stronger containment than the scrubbed environment spawned commands get — and a proxy URL may carry credentials. Handing model code a credentialed URL to reach the network is the wrong trade; the exclusion is recorded in that package's limitations.
**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
@@ -6,9 +6,9 @@ Status: implemented
## Problem
Node 内置的 `fetch` 会忽略 `HTTP_PROXY` 与 `HTTPS_PROXY`。开发者运行的其他工具——curl、git、npm、pip——都遵循它们,所以代理后面的用户导出一次变量就期待一切随之生效。Harness 并没有:`setGlobalDispatcher`、`ProxyAgent` 与 `EnvHttpProxyAgent` 在 `packages/` 与 `apps/` 中出现次数为零,因此模型请求、每次 web 搜索、`web_fetch`、走 HTTP 的 MCP、OTLP 导出器与 E2B SDK 全部直连,且是静默的,任何地方都没有诊断。
Node 内置的 `fetch` 会忽略 `HTTP_PROXY` 与 `HTTPS_PROXY`。开发者运行的其他工具——curl、git、npm、pip——都遵循它们,所以代理后面的用户导出一次变量就期待一切随之生效。Harness 并没有:`setGlobalDispatcher`、`ProxyAgent` 与 `EnvHttpProxyAgent` 在 `packages/` 与 `apps/` 中出现次数为零,因此模型请求、每次 web 搜索、`web_fetch`、走 HTTP 的 MCP 与 OTLP 导出器全部直连,且是静默的,任何地方都没有诊断。
仓库曾短暂拥有过答案,又在无人察觉时弄丢了。PR #971 在 `bin/dsh` 里设置了 `NODE_USE_ENV_PROXY=1`;十一天后 `bbb1b1cc38 cleanup: remove managed source installer` 整体删除了那个启动器,把该标志一并带走。留下的只有 `apps/cli/reference/README.md` 里的一句话,让读者去设置一个已经无人消费的变量。
仓库曾短暂拥有过答案,又在无人察觉时弄丢了。PR #971 在 `bin/dsh` 里设置了 `NODE_USE_ENV_PROXY=1`;十一天后的“cleanup: remove managed source installer”改动整体删除了那个启动器,把该标志一并带走。留下的只有 `apps/cli/reference/README.md` 里的一句话,让读者去设置一个已经无人消费的变量。
即便照做,那句话也不可能生效,原因有三条且都经过实测。`NODE_USE_ENV_PROXY` 在进程启动时对环境取快照,而 `loadLayeredEnv()` 是在之后才合并 `.env` 层,因此写在 `$DSH_HOME/.env` 中的代理对它不可见。它只覆盖 Node 24.0+,在 22 线上只覆盖 22.21+——而 `engines` 允许 `^22.19.0`,那里根本没有这个变量,设置了也不会有任何警告。它也完全触及不到 `web-fetch-http`:该提供方向 `fetch` 传入自己的 `dispatcher`,而显式 dispatcher 无论标志如何都会覆盖全局的那个。
@@ -24,9 +24,7 @@ Node 内置的 `fetch` 会忽略 `HTTP_PROXY` 与 `HTTPS_PROXY`。开发者运
那次修订一并引入的插件也随之删除。它让某个组合可以把策略写进 `cordis.yml`,但没有任何随附 bundle 挂载它,因此启动器那条路径是唯一可达的——而它的 `Config` 是那条配置分支唯一的供给方,别处无从到达。
**四个函数——收敛的是调用方,而不是让本包为每个 SDK 各加一个导出。** 早先一版导出六个:dispatcher 工厂、`node:http` agent 工厂、代理 URL 查询、策略访问器、安装器与子进程环境构造器。每一个都为某个 SDK 的传输而存在,而这正是一个传输策略包退化成「别的包的约束目录」的过程。Review 问能不能反过来让调用方收敛;能,而且每删掉一个导出都带走了一整种写法。遥测不再被路由,`node:http` agent 工厂随之退场。`web-fetch-http` 在带注释的豁免下自建 pin agent,dispatcher 工厂随之退场。E2B 读 `route.proxy`,代理 URL 查询随之退场。
剩下的是 `installProxyFromEnvironment`、`proxyRouteFor`、`proxyEnvironmentForChild` 与 `clearedProxyEnv`——按「调用方需要策略的方式」各一个,而不是按 SDK 各一个。安装吸收了解析与诊断上报,因为没有调用方需要把它们分开:解析出来却不安装的策略什么也路由不了。
**每种调用需求对应一项操作。** `installProxyFromEnvironment`、`proxyRouteFor`、`proxyEnvironmentForChild` 与 `clearedProxyEnv` 分别负责安装、逐请求路由、子进程继承和 fixture 隔离。特定于 SDK 的工厂会通过共享 API 暴露各自的传输约束。`web-fetch-http` 在构造地址固定传输时使用已解析路由;OTLP 导出器保持直连。安装包含解析与诊断上报,因为调用方需要一项同时解析并安装路由的操作。
`proxyRouteFor` 还堵掉了旧访问器让人写得出来的一个缺陷。`web-fetch-http` 先读策略决定是否 pin,再读一次去构造传输;两次读取之间发生卸载,就会为第一次读取已判定走代理的 URL 返回一个直连且未 pin 的 agent。路由把两者一起交出,分支与请求便无从分歧。它携带的是进程级 dispatcher,dispose 时是 close 而非 destroy,因此策略被卸载时已经发出的请求仍会跑完。
@@ -42,19 +40,19 @@ 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 code 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。
这接受了一处已记录的接缝。此类上下文按 Node 自己的规则匹配绕过条目,其分隔符与 IPv4 区间支持与本包不同,且该标志仅存在于 Node 22.21+ 与 24+。
**有两个 SDK 并不落到 `globalThis.fetch`,而读代码给出的答案是相反的。** 审计最初把 OTLP 导出器与 E2B SDK 判为已覆盖,依据是在 `@opentelemetry/otlp-exporter-base` 里 grep 到了 `globalThis.fetch`。那处命中属于**浏览器**传输;在 Node 上 delegate 选择的是 `http-exporter-transport`,它通过 `node:http` 投递——那里全局 dispatcher 触及不到。E2B 又是另一种形态:它自建 undici `Agent`/`ProxyAgent`,并接受一个自己从不从环境读取的 `proxy` URL。两者都实测为直连。E2B 接收 `proxyRouteFor` 给出的 `route.proxy`,与 `web-fetch-http` 调的是同一个函数。遥测则被有意保留为直连,而这个排除项才是更值得说的一半。
**SDK 传输需要独立验证。** OTLP 导出器在 Node 上选择 `http-exporter-transport`,通过 `node:http` 投递并绕过全局 fetch dispatcher;其浏览器实现中的 `globalThis.fetch` 引用不能证明 Node 路由。[E2B 移除决策](../simplification/2026-09-11-remove-e2b-providers.zh.md)撤下另一项 SDK 传输集成,但不改变这一要求。
**遥测的直连是有意为之。** 要让它走代理只有两条路,代价都超过这条通道本身的价值。`http.Agent` 通过 `proxyEnv` 读取环境,而该选项自 Node 22.21 与 24.5 才有——落在 engines 范围之内,因此 22.19、22.20 与 24.0–24.4 无论如何仍是直连,而代理包还得为一条只在部分运行时生效的路径保留 `createNodeHttpAgent` 导出。改用 SDK 的 `fetch` delegate 替换传输可以覆盖所有运行时,但该 delegate 没有压缩能力,而随附的 `base` bundle 启用了 gzip:实测一批真实规模的 OTLP 数据启用后体积只有 1/6.4。曾有一版转而在加载期拒绝 `exporter.compression`,结果凡是启动随附 bundle 的测试全部失败;另一版在 serializer 处 gzip 确实能跑通,但代价是把传输层代码塞进了遥测插件。
与之相比,遥测是唯一一条丢失了对用户毫无代价的出网通道:没有任何工具、模型请求或会话依赖它,而连不上的导出本就被静默丢弃。处在强制代理后的用户,只是停留在本次改动之前的状态,而不是被弄坏。`egress.spec.ts` 现在断言这一排除——若某次 SDK 升级把导出器挪到 `fetch` 上,遥测就会开始静默走代理,而该用例正是让这件事暴露出来的东西。
**每个出网点都配一份出网测试,因为读代码不够。** 各所属包中的 `egress.spec.ts` 驱动该点的真实代码路径,目标是无法解析的 `.invalid` 主机,穿过一个假代理,并断言代理确实收到了请求。九份测试覆盖搜索后端、pi-ai 发现、走 HTTP 的 MCP、E2B、派生的子 Node、worker 线程,以及遥测的排除。下面那条门禁看不进依赖内部;这些能,它们把「某个 SDK 换了传输」从静默回归变成失败的测试。
**每个出网点都配有出网测试。** 各所属包中的 `egress.spec.ts` 通过假代理驱动实际传输,并检查观察到的路由。这些测试覆盖搜索后端、pi-ai 发现、走 HTTP 的 MCP、子 Node 进程、worker 线程和遥测的直连例外。它们可发现静态调用点检查无法观察到的依赖传输变化。
**用门禁防止该缺陷复现。** `verify-no-bare-dispatcher` 解析 TypeScript AST——`scripts/AGENTS.md` 要求 source-ownership 门禁使用语法感知发现,而逐行正则漏掉了本仓库已在使用的 `{ dispatcher }` 简写,以及重命名导入后的 `new Alias(...)`。它在所属包之外拒绝 undici agent 构造与显式 `dispatcher` 选项。`proxyRouteFor(url)` 是受支持的替代;唯一一处确实自有传输的调用点——`web-fetch-http`,它把请求钉在已校验的地址上——用 `proxy-exempt:` 注释说明。这条规则之所以存在,是因为 `web-fetch-http` 里原本那行 `new Agent` 在写下时完全合理——那时根本还没有代理这回事,也没有任何机制会拦下它。
@@ -72,7 +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。
**也把代理给 `code-runtime` worker。** 被否决。模型编写的程序在那里运行时完全没有环境变量——这比派生命令得到的 scrubbed 环境更严——而代理 URL 可能携带凭据。把带凭据的 URL 交给模型代码去访问网络是错误的取舍;该排除已记入那个包的限制清单。
**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
@@ -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/2026-08-28-subprocess-native-containment.md
2026-08-28-subprocess-native-containment.md: 0c03884bba67dab6e2e38f96ce2e874ed62ba00f
2026-08-28-subprocess-native-containment.zh.md: b7850e1c4fee06dfeb1a05c968d132de5994686f
2026-08-28-subprocess-native-containment.md: 1b0fd162be77356001bcd9224bb2ee2889e6d6c0
2026-08-28-subprocess-native-containment.zh.md: f33643f98a6ea6a90c0780432cbe11a35a868562
@@ -22,7 +22,7 @@ The first eligible Linux ordinary or PTY call in one runtime deeply checks the e
The parent creates one 0700 directory with a complete 0600 `launch-request.json` containing the final target cwd and environment. The private `DSH_SUBPROCESS_RUNNER` value locates that request while the runner starts from the provider cwd and a bootstrap-safe environment. `systemd-run --user --scope --quiet --collect --expand-environment=no` registers its process in the scope, then the one-shot bootstrap removes and validates the request, changes to the target cwd, restores the complete target environment, resolves a bare executable with the target PATH rules, clears `FD_CLOEXEC` on fd 0 through fd 2, and calls libc `execve()` with the original argv. The bootstrap becomes the target in place and preserves its inherited stdio; it does not remain as a supervisor.
Request consumption or a manager observation of a loaded unit establishes scope ownership. Unit absence before either fact remains unresolved while the direct launcher is running. If that launcher exits while the request remains unconsumed, the direct result rejects with the startup failure while range observation records that the scope never existed and resolves the empty-range wait. The parent checks this unresolved interval every 50 milliseconds; after establishment, state queries back off exponentially to the existing 5-second systemctl bound. Each query reads both `LoadState` and `ActiveState`: loaded `inactive` or `failed`, or an established unit becoming `not-found`/`inactive` or otherwise collected away, proves the range empty. `active`, `activating`, `reloading`, and `deactivating` remain nonterminal. Unknown or malformed combinations and unreadable manager results reject `waitForExit()` instead of claiming quiescence. `terminate()` wakes a sleeping observer for an immediate recheck, and settlement cancels the losing backoff sleep. A strict sibling `startup-error.json` carries only request/bootstrap or target pre-exec failure, and the parent removes this spawn's private paths at observable lifecycle completion.
Request consumption or a manager observation of a loaded unit establishes scope ownership. Unit absence before either fact remains unresolved while the direct launcher is running. If that launcher exits while the request remains unconsumed, the direct result rejects with startup failure unless its observed signal matches a termination requested while the launcher was running. A matching signal preserves the actual exit outcome for both ordinary and PTY launches; a recorded startup error always takes precedence. Range observation independently resolves an empty range when the launcher has exited and the unit is absent. The parent checks this unresolved interval every 50 milliseconds; after establishment, state queries back off exponentially to the existing 5-second systemctl bound. Each query reads `LoadState`, `ActiveState`, and `TasksCurrent`: loaded `inactive` or `failed`, or an established unit becoming `not-found`/`inactive` or otherwise collected away, proves the range empty. `active`, `activating`, `reloading`, and `deactivating` remain nonterminal, except that once termination was requested and either the launcher has exited or the bootstrap has consumed its request, a still-active unit reporting no processes is an empty range. Request consumption proves the payload entered the scope, so zero members establishes quiescence even before Node receives the direct-process exit notification. Without request consumption, launcher exit remains necessary: the manager ends a scope only on the populated-to-empty transition, so a payload killed before it entered the cgroup never triggers one. That conclusion stops the leftover unit so transient units cannot accumulate, while an unreported or `[not set]` process count stays unknown and keeps waiting. Unknown or malformed combinations and unreadable manager results reject `waitForExit()` instead of claiming quiescence. `terminate()` wakes a sleeping observer for an immediate recheck, and settlement cancels the losing backoff sleep. A strict sibling `startup-error.json` carries only request/bootstrap or target pre-exec failure, and the parent removes this spawn's private paths at observable lifecycle completion.
The ordinary target result still comes from the same child process. The PTY path uses the same request and bootstrap without a resident runner, so the `node-pty` PID, process group, session leader, controlling terminal, foreground `inputWaiting`, `/dev/tty`, readiness, and direct terminal outcome retain their existing meanings while scope membership covers `setsid` and reparented descendants.
@@ -54,11 +54,12 @@ This note owns the current native-containment mechanism. It partially updates th
## Verification
- Provider and Linux protocol suites pin synchronous NUL rejection before launch side effects, strict request/error decoding, target cwd and complete environment restoration, private-variable collision, symlink-sensitive PATH traversal with preserved argv, close-on-exec removal for inherited stdio, pre-exec error ownership, failed-deep-probe retry plus successful-deep-probe caching with per-call manager checks, the three scope-establishment states including an exited launcher with an unconsumed request, `LoadState`/`ActiveState` parsing, `reloading`, terminate wake-up with losing-delay cancellation, bounded established-scope backoff, and exactly-once PTY managed-owner cleanup.
- Provider and Linux protocol suites pin synchronous NUL rejection before launch side effects, strict request/error decoding, target cwd and complete environment restoration, private-variable collision, symlink-sensitive PATH traversal with preserved argv, close-on-exec removal for inherited stdio, pre-exec error ownership, failed-deep-probe retry plus successful-deep-probe caching with per-call manager checks, the three scope-establishment states including requested versus unexpected exits with an unconsumed request, `LoadState`/`ActiveState`/`TasksCurrent` parsing, releasing a leftover scope left active with no processes beside the live-client, unterminated, and unset-count cases, `reloading`, terminate wake-up with losing-delay cancellation, bounded established-scope backoff, and exactly-once PTY managed-owner cleanup.
- Windows protocol and Win32 suites pin exactly two result branches, numeric-only target exits, ordinary-error start cancellation with raw parent-local reasons, the reduced `name`/`message`/`code`/`syscall`/`path` error record, the fixed `2`/`3`/`267` to `ENOENT`, `740` to `EACCES`, `5` to `EPERM`, `193` to `EFTYPE`, and remaining-code to `UNKNOWN` mapping, start delivery after runner spawn, empty-range settlement after pre-spawn failure, explicit ordinally sorted target environment blocks with `=C:` preservation and double-NUL termination, `uv_get_osfhandle()` carrier mapping and unsigned invalid-sentinel rejection, the null-device ignored-stdin carrier and piped non-ignored stdin, result-send and IPC-disconnect failures, direct-result latching before stdio settlement, active-process quiescence, and unique handle cleanup.
- A keyless [`bash-startup-timeout`](../../../../snapshots/session/bash-startup-timeout/snapshot.yml) Session snapshot pins the model-facing timeout result. A Linux user-systemd fixture holds the launch request unconsumed at an input barrier and verifies cancellation plus range settlement.
- Real Linux user-systemd tests run one ordinary and one `node-pty` `setsid`/reparent scenario through the production entry. They prove scope signalling and collection, bare executable lookup, escaped-descendant termination, range settlement, and unchanged PTY PID, session, controlling-terminal, foreground-input, `/dev/tty`, readiness, and startup-failure semantics.
- Native Windows tests prove suspended creation, Job assignment before resume, inherited stdio, default descendant inheritance, direct result, termination, active-process zero, abnormal/disconnected runner cleanup, kill-on-close, and synchronous host-exit termination. Source, built, and Python packaged smokes enter the same runner core.
- Public seam types, local and E2B providers, LSP and subagent consumers, shell fixtures, READMEs, the Cordis catalog, and the keyless subprocess API snapshot contain no ordinary PID; terminal PID remains.
- Public seam types, local provider, LSP and subagent consumers, shell fixtures, READMEs, the Cordis catalog, and the keyless subprocess API snapshot contain no ordinary PID; terminal PID remains.
## Alternatives considered
@@ -76,6 +77,8 @@ This note owns the current native-containment mechanism. It partially updates th
**Recover a failed native launch by replaying the command.** Rejected because an ambiguous failure may occur after target execution and replay can therefore execute the command twice.
**Infer startup failure from an unconsumed request alone.** Rejected because abort, timeout, or disposal can terminate the launcher before request consumption. Matching an observed signal to an owner-issued termination preserves cancellation without hiding an unrelated bootstrap exit or recorded pre-exec failure.
## Consequences
Supported Linux ordinary and PTY launches and Windows ordinary launches retain descendants through process-group escape and direct-parent exit, while direct target results remain independent from range quiescence. The cost is a per-spawn Linux manager check and scope/request or Windows runner/IPC/Job lifecycle, plus explicit failure when the selected owner cannot prove settlement.
@@ -22,7 +22,7 @@ detached POSIX 进程组、Windows direct-parent 遍历与 PTY 后代扫描只
parent 创建一个 0700 目录,其中的完整 0600 `launch-request.json` 保存最终 target cwd 与环境。私有 `DSH_SUBPROCESS_RUNNER` 值负责定位该 request,runner 则从 provider cwd 与 bootstrap-safe 环境启动。`systemd-run --user --scope --quiet --collect --expand-environment=no` 先把自身进程注册到 scope,再由 one-shot bootstrap 删除并校验 request、切换到 target cwd、恢复完整 target 环境、按 target PATH 规则解析裸可执行文件、清除 fd 0 至 fd 2 的 `FD_CLOEXEC`,并使用原始 argv 调用 libc `execve()`。bootstrap 会原地成为 target 并保留继承的 stdio,不作为常驻 supervisor。
request 被消费或 manager 已观察到 loaded unit 都能建立 scope ownership。在这两项事实出现前,只要 direct launcher 仍在运行,unit absence 就保持未决。如果 launcher 退出时 request 仍未消费,direct result 会以 startup failure reject,而 range observation 会记录 scope 从未存在,并成功结算 empty-range wait。parent 每 50 毫秒检查一次这段未决区间;建立后,状态查询按指数增长间隔退避,最多达到既有的 5 秒 systemctl 上限。每次查询同时读取 `LoadState` 与 `ActiveState`:loaded `inactive` 或 `failed`,以及已经建立的 unit 变为 `not-found`/`inactive` 或被 collect 卸载,都能证明 range 为空。`active`、`activating`、`reloading` 与 `deactivating` 仍是非终态。未知或 malformed 组合以及不可读的 manager 结果会使 `waitForExit()` reject,而不是宣称完全停稳。`terminate()` 会唤醒正在休眠的 observer 立即复查,结算时会取消未胜出的退避 sleep。严格的同目录 `startup-error.json` 只承载 request/bootstrap 或 target pre-exec failure,parent 会在可观察生命周期完成时移除本次 spawn 的私有路径。
request 被消费或 manager 已观察到 loaded unit 都能建立 scope ownership。在这两项事实出现前,只要 direct launcher 仍在运行,unit absence 就保持未决。如果 launcher 退出时 request 仍未消费,direct result 会以 startup failure reject,除非实际观察到的信号匹配 launcher 仍在运行时请求的终止信号。普通进程与 PTY 进程遇到匹配信号时都会保留实际退出结果;已记录的 startup error 始终优先。launcher 已退出且 unit 不存在时,range observation 会独立结算 empty-range wait。parent 每 50 毫秒检查一次这段未决区间;建立后,状态查询按指数增长间隔退避,最多达到既有的 5 秒 systemctl 上限。每次查询读取 `LoadState`、`ActiveState` 与 `TasksCurrent`:loaded `inactive` 或 `failed`,以及已经建立的 unit 变为 `not-found`/`inactive` 或被 collect 卸载,都能证明 range 为空。`active`、`activating`、`reloading` 与 `deactivating` 仍是非终态;例外是:一旦请求过终止,且 launcher 已经退出或 bootstrap 已消费其 request,报告没有任何进程却仍 active 的 unit 就是空 range。request 消费证明 payload 已进入 scope,因此即使 Node 尚未收到直接进程的退出通知,零成员也能证明完全停稳。request 未消费时仍须等待 launcher 退出——manager 只在观测到 populated→empty 转变时才结束 scope,因此进入 cgroup 前就被杀死的 payload 永远不会触发该转变。该判定会 stop 掉这个遗留 unit,使 transient unit 不会累积;未上报或为 `[not set]` 的进程数仍视为未知并继续等待。未知或 malformed 组合以及不可读的 manager 结果会使 `waitForExit()` reject,而不是宣称完全停稳。`terminate()` 会唤醒正在休眠的 observer 立即复查,结算时会取消未胜出的退避 sleep。严格的同目录 `startup-error.json` 只承载 request/bootstrap 或 target pre-exec failure,parent 会在可观察生命周期完成时移除本次 spawn 的私有路径。
普通 target result 仍来自同一个 child process。PTY 路径复用同一 request 与 bootstrap,但不增加常驻 runner,因此 `node-pty` PID、进程组、session leader、控制终端、前台 `inputWaiting`、`/dev/tty`、readiness 与 direct terminal outcome 保留既有含义,同时 scope membership 覆盖 `setsid` 与 reparent 后代。
@@ -54,11 +54,12 @@ selector 是 per-spawn locator 或 sentinel,不是凭据或持久格式。Linu
## Verification
- provider 与 Linux 协议测试套件固定同步 NUL 拒绝发生在启动副作用之前、严格 request/error 解码、target cwd 与完整环境恢复、私有变量碰撞、保留 argv 且对 symlink 敏感的 PATH 遍历、为继承 stdio 清除 close-on-exec、pre-exec error ownership、失败深度 probe 重试与成功深度 probe 缓存及逐调用 manager 检查、三种 scope 建立状态(包括 launcher 退出且 request 未消费)、`LoadState`/`ActiveState` 解析、`reloading`、带未胜出 delay 取消的 terminate wake-up、建立后有上限的退避,以及 PTY managed-owner 恰好一次 cleanup。
- provider 与 Linux 协议测试套件固定同步 NUL 拒绝发生在启动副作用之前、严格 request/error 解码、target cwd 与完整环境恢复、私有变量碰撞、保留 argv 且对 symlink 敏感的 PATH 遍历、为继承 stdio 清除 close-on-exec、pre-exec error ownership、失败深度 probe 重试与成功深度 probe 缓存及逐调用 manager 检查、三种 scope 建立状态(包括 request 未消费时的请求终止与意外退出)、`LoadState`/`ActiveState`/`TasksCurrent` 解析、释放被留在 active 且没有任何进程的遗留 scope(连同 client 仍存活、未请求终止与进程数未上报三种情形)、`reloading`、带未胜出 delay 取消的 terminate wake-up、建立后有上限的退避,以及 PTY managed-owner 恰好一次 cleanup。
- Windows 协议与 Win32 测试套件固定恰好两个 result 分支、只含数字的 target exit、使用普通 error 的 start cancellation 与 parent 原样保留的本地 reason、缩减到 `name`/`message`/`code`/`syscall`/`path` 的 error record、固定的 `2`/`3`/`267` 到 `ENOENT`、`740` 到 `EACCES`、`5` 到 `EPERM`、`193` 到 `EFTYPE` 及其余 code 到 `UNKNOWN` 的映射、runner spawn 后才发送 start、spawn 前 failure 的 empty-range settlement、按序数显式排序的 target 环境块及 `=C:` 保留和双 NUL 结尾、`uv_get_osfhandle()` carrier 映射与 unsigned invalid sentinel 拒绝、null-device ignored-stdin carrier 与非 ignore stdin pipe、result-send 与 IPC-disconnect failure、stdio settlement 前的 direct-result 锁存、active-process 完全停稳,以及唯一 handle cleanup。
- 无需密钥的 [`bash-startup-timeout`](../../../../snapshots/session/bash-startup-timeout/snapshot.yml) Session 快照固定模型可见的超时结果。Linux user-systemd fixture 通过输入屏障保持启动请求未消费,并验证取消与 range settlement。
- 真实 Linux user-systemd 测试会分别通过生产入口运行一条普通命令与一条 `node-pty` `setsid`/reparent 场景。它们证明 scope signalling 与 collection、裸可执行文件查找、逃逸后代终止、range settlement,以及不变的 PTY PID、session、控制终端、前台输入、`/dev/tty`、readiness 与 startup-failure 语义。
- native Windows 测试证明 suspended creation、resume 前 Job assignment、继承 stdio、默认后代继承、direct result、termination、active-process zero、异常/disconnected runner cleanup、kill-on-close 与同步 host-exit termination。source、built 与 Python packaged 冒烟测试进入同一 runner core。
- 公共 seam 类型、local 与 E2B provider、LSP 与 subagent 消费方、shell fixture、README、Cordis catalog 与 keyless subprocess API snapshot 都不包含普通 PID;terminal PID 保留。
- 公共 seam 类型、local provider、LSP 与 subagent 消费方、shell fixture、README、Cordis catalog 与 keyless subprocess API snapshot 都不包含普通 PID;terminal PID 保留。
## Alternatives considered
@@ -76,6 +77,8 @@ selector 是 per-spawn locator 或 sentinel,不是凭据或持久格式。Linu
**在 native launch 失败后重放命令。**不予采用,因为含糊 failure 可能发生在 target 已经执行之后,重放因此可能把命令执行两次。
**仅从未消费的 request 推断启动失败。** 拒绝,因为 abort、timeout 或 dispose 可能在 request 消费前终止 launcher。将实际观察到的信号与 owner 请求的终止信号匹配,既能保留取消结果,也不会隐藏无关的 bootstrap 退出或已记录的 pre-exec failure。
## Consequences
受支持的 Linux 普通与 PTY 启动、Windows 普通启动会在后代逃离进程组或 direct parent 退出后继续拥有它们,同时 direct target result 与 range 完全停稳保持独立。代价是每次 spawn 都需要一次 Linux manager 检查与 scope/request,或一套 Windows runner/IPC/Job 生命周期,而且所选 owner 无法证明 settlement 时会显式失败。
@@ -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/2026-08-30-retain-ignorable-external-session-events.md
2026-08-30-retain-ignorable-external-session-events.md: 1fe3a6d99a16daa6ad88f6717baa18f18e8c7355
2026-08-30-retain-ignorable-external-session-events.zh.md: 632b7b418252c2b299162f3e00a4d41169169509
2026-08-30-retain-ignorable-external-session-events.md: e085e429a6ad49cf742694cec7dda05770a5b8a1
2026-08-30-retain-ignorable-external-session-events.zh.md: 1fe4f81441503287961b993eae444fa4385f4803
@@ -6,7 +6,7 @@ English | [中文](2026-08-30-retain-ignorable-external-session-events.zh.md)
## Problem
The session event envelope carries `ignorable?: true` so a reader can accept an unrecognized informational event without treating every vocabulary addition as a new session format. [PR #3087](https://github.com/deepseek-harness/deepseek-harness/pull/3087) removed the field after finding no first-party producer and made every unknown event required-on-read.
The session event envelope carries `ignorable?: true` so a reader can accept an unrecognized informational event without treating every vocabulary addition as a new session format. PR #3087 removed the field after finding no first-party producer and made every unknown event required-on-read.
That producer inventory did not cover a third-party plugin that currently depends on the field. Without `ignorable`, a first-party reader rejects a stored session containing the plugin's informational event because the event is outside the repository-generated `KNOWN_SESSION_EVENT_TYPES`. The plugin has no replacement registration or versioning mechanism, so deleting the field before a replacement exists breaks a current external consumer.
@@ -6,7 +6,7 @@ Status: implemented
## 问题
会话事件信封包含 `ignorable?: true`,读取器因此可以接受不认识的信息性事件,而不必把每次词汇增加都视为新的会话格式。[PR #3087](https://github.com/deepseek-harness/deepseek-harness/pull/3087) 在没有发现第一方生产方后删除了该字段,并把每个未知事件都改为读取必需项。
会话事件信封包含 `ignorable?: true`,读取器因此可以接受不认识的信息性事件,而不必把每次词汇增加都视为新的会话格式。PR #3087 在没有发现第一方生产方后删除了该字段,并把每个未知事件都改为读取必需项。
该生产方清单没有覆盖当前依赖此字段的一个第三方插件。没有 `ignorable` 时,第一方读取器会拒绝包含该插件信息性事件的已存会话,因为该事件不在仓库生成的 `KNOWN_SESSION_EVENT_TYPES` 中。插件没有可替代的注册或版本机制,因此在替代机制存在前删除该字段会破坏当前外部消费方。
@@ -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/2026-09-01-v2-embedded-assistant-streams.md
2026-09-01-v2-embedded-assistant-streams.md: 98207749028e2182c5e60073fc07985688ecfa30
2026-09-01-v2-embedded-assistant-streams.zh.md: 90dff5a385cf83071ec52a2fc2b57a893107b861
2026-09-01-v2-embedded-assistant-streams.md: 219879600f8435a6ecc9593661dd3f463ed1158c
2026-09-01-v2-embedded-assistant-streams.zh.md: d38c180b1f1de689620f2e0533d2b02f4a71e466
@@ -23,7 +23,7 @@ Session format v2 has no top-level `assistant/chunk` event. Each model attempt c
`AssistantStreamAccumulator` snapshots each chunk once. Consecutive text, reasoning, or tool-argument deltas for the same block become one compact run with its first timestamp, exact timestamp gaps, and one array member per original delta. Every other chunk remains a timestamped raw record. `expandAssistantStream()` strictly validates and reconstructs the exact timed sequence; compaction never joins delta boundaries.
The migration publication verifier and frozen v2 fixture validator require the embedded stream to reproduce a non-empty `assistant/message`'s content, usage, and replay state. An empty stream remains valid for a migrated legacy message that had no source chunks. Ordinary Session restoration validates the settlement fields needed by the runtime without expanding every historical stream; consumers that expand a compact stream validate its records when they read it. `assistant/message` cannot carry obsolete chunk `sourceEventSeqs`; ordinary user and tool surface provenance remains available.
The migration publication verifier and frozen v2 fixture validator require the embedded stream to reproduce a non-empty `assistant/message`'s content, usage, and replay state. An empty stream remains valid for a migrated legacy message that had no source chunks. Ordinary Session restoration validates the settlement fields needed by the runtime without expanding every historical stream; consumers that expand a compact stream validate its records when they read it. `assistant/message` cannot carry obsolete chunk `sourceEventSeqs`; ordinary user and tool source-event references remain available.
### Live presentation and durable replay
@@ -35,9 +35,9 @@ The Client event source passes durable settlements through unchanged. The Chat a
### Released v1 to v2 migration
The adjacent migration validates the complete frozen v1 artifact, groups chunks by turn, step, terminal boundary, and exact message provenance, and then substitutes one settlement per attempt. A successful group's chunks move into its message. An unclaimed group becomes `assistant/attempt` at the last consumed chunk's position. Unrelated interleaved events retain their relative order, and survivors receive dense v2 sequence numbers. The edge compacts embedded streams through the runtime `AssistantStreamAccumulator` from `dsh-llm` instead of a frozen copy, because that package owns the v2 stream encoding. The isolated publication verifier expands and re-assembles the written stream through `expandAssistantStream()` and `BlockAssembler`, then checks each migrated `assistant/message` against it before publication. A later format that changes the stream encoding must freeze copies of these helpers into this edge.
The adjacent migration validates the complete frozen v1 artifact, groups chunks by turn, step, terminal boundary, and exact message chunk references, and then substitutes one settlement per attempt. A successful group's chunks move into its message. An unclaimed group becomes `assistant/attempt` at the last consumed chunk's position. Unrelated interleaved events retain their relative order, and survivors receive dense v2 sequence numbers. The edge compacts embedded streams through the runtime `AssistantStreamAccumulator` from `dsh-llm` instead of a frozen copy, because that package owns the v2 stream encoding. The isolated publication verifier expands and re-assembles the written stream through `expandAssistantStream()` and `BlockAssembler`, then checks each migrated `assistant/message` against it before publication. A later format that changes the stream encoding must freeze copies of these helpers into this edge.
The edge remaps the finite declared reference inventory: envelope provenance, surface replacement endpoints, command source events, compaction ranges and shadowed lists, and title message lists. The model-visible text of a validated `session/title-llm-request` remains byte-identical in the source sequence namespace while its `messageSeqs` field moves to the v2 namespace; target validation therefore does not reconstruct that text from remapped sequences. A reference to a consumed chunk refuses migration; it is never redirected to a settlement with different meaning. The edge also refuses an inherited cut that splits an attempt.
The edge remaps the finite declared reference inventory: envelope source-event references, surface replacement endpoints, command source events, compaction ranges and shadowed lists, and title message lists. The model-visible text of a validated `session/title-llm-request` remains byte-identical in the source sequence namespace while its `messageSeqs` field moves to the v2 namespace; target validation therefore does not reconstruct that text from remapped sequences. A reference to a consumed chunk refuses migration; it is never redirected to a settlement with different meaning. The edge also refuses an inherited cut that splits an attempt.
The v2 physical header requires `isSeeded` and stores no numeric cut. A seeded artifact marks its exact cut with `session/end-seed { inherited: true }`; decoding derives the cut from the last tagged marker. The v2 codec writes one durable event per physical row, range-encodes only `sourceEventSeqs`, and validates physical envelopes without freezing ordinary event vocabulary or payload additions. The v1-to-v2 target validator separately freezes the released-v2 inventory, while current restoration uses the installed Session vocabulary. Frozen v0 and v1 codecs retain packed-row decoding for their immutable historical generations.
@@ -49,7 +49,7 @@ Generation selection and publication follow the [released Session migration deci
## Verification
The compact-stream tests pin exact accumulation and expansion for text, reasoning, tool arguments, raw chunks, timestamp gaps, malformed records, and detached snapshots. The v1-to-v2 tests cover successful and failed attempts, interleaving, dense sequence and reference remapping, source-sequence title framing, seed-cut insertion and split refusal, strict source and target validation, one-row v2 encoding, backend-compatible provenance ranges, raw and Zstandard publication, and no-write current reads.
The compact-stream tests pin exact accumulation and expansion for text, reasoning, tool arguments, raw chunks, timestamp gaps, malformed records, and detached snapshots. The v1-to-v2 tests cover successful and failed attempts, interleaving, dense sequence and reference remapping, source-sequence title framing, seed-cut insertion and split refusal, strict source and target validation, one-row v2 encoding, backend-compatible source-event ranges, raw and Zstandard publication, and no-write current reads.
The pre-merge performance acceptance measured static catalog-routing overhead against direct released-v2 restoration of the same already parsed physical rows across three runs, 100 warmup pairs, and 600 measured pairs; it did not compare v1 with v2 or time backend I/O. Every pooled median and p95 regression stayed within the 5% budget, with a worst p95 regression of 3.150%.
@@ -23,7 +23,7 @@ Session format v2 没有顶层 `assistant/chunk` 事件。每个模型 attempt
`AssistantStreamAccumulator` 对每个 chunk 只快照一次。同一 block 的连续 text、reasoning 或 tool argument delta 会变成一个紧凑 run,包含首个时间戳、精确时间戳间隔和每个原始 delta 对应的一个数组成员。其他 chunk 保留为带时间戳的 raw record。`expandAssistantStream()` 会严格校验并重建精确的带时间序列;压缩绝不会合并 delta 边界。
Migration publication verifier 与冻结的 v2 fixture validator 要求嵌入式 stream 能复现非空 `assistant/message` 的 content、usage 与 replay state。对于没有源 chunk 的已迁移旧 message,空 stream 仍然有效。普通 Session restore 只校验 runtime 直接依赖的 settlement 字段,不展开全部历史 stream;需要展开 compact stream 的 consumer 会在读取时校验 record。`assistant/message` 不能携带已停用的 chunk `sourceEventSeqs`;普通 user 与 tool surface provenance 保持可用。
Migration publication verifier 与冻结的 v2 fixture validator 要求嵌入式 stream 能复现非空 `assistant/message` 的 content、usage 与 replay state。对于没有源 chunk 的已迁移旧 message,空 stream 仍然有效。普通 Session restore 只校验 runtime 直接依赖的 settlement 字段,不展开全部历史 stream;需要展开 compact stream 的 consumer 会在读取时校验 record。`assistant/message` 不能携带已停用的 chunk `sourceEventSeqs`;普通 user 与 tool source-event reference 保持可用。
### 实时呈现与持久回放
@@ -35,9 +35,9 @@ Client event source 原样传递持久 settlement。Chat 与 Trajectory 的 Assi
### 已发布 v1 到 v2 迁移
相邻迁移会校验完整的冻结 v1 产物,按 turn、step、terminal boundary 与精确 message provenance 对 chunk 分组,再为每个 attempt 替换一个 settlement。成功分组的 chunk 移入其 message。未被认领的分组会在最后一个被消费 chunk 的位置变成 `assistant/attempt`。无关的交错事件保持相对顺序,存活事件获得密集 v2 序号。该迁移边通过 `dsh-llm` 运行时的 `AssistantStreamAccumulator` 压缩嵌入 stream,而不持有冻结副本,因为该包拥有 v2 stream 编码。隔离的 publication verifier 通过 `expandAssistantStream()` 与 `BlockAssembler` 展开并重组写入后的 stream,并在发布前检查每个迁移后的 `assistant/message` 是否与其一致。日后若某个格式改变 stream 编码,必须把这些 helper 的冻结副本纳入本迁移边。
相邻迁移会校验完整的冻结 v1 产物,按 turn、step、terminal boundary 与精确 message chunk reference 对 chunk 分组,再为每个 attempt 替换一个 settlement。成功分组的 chunk 移入其 message。未被认领的分组会在最后一个被消费 chunk 的位置变成 `assistant/attempt`。无关的交错事件保持相对顺序,存活事件获得密集 v2 序号。该迁移边通过 `dsh-llm` 运行时的 `AssistantStreamAccumulator` 压缩嵌入 stream,而不持有冻结副本,因为该包拥有 v2 stream 编码。隔离的 publication verifier 通过 `expandAssistantStream()` 与 `BlockAssembler` 展开并重组写入后的 stream,并在发布前检查每个迁移后的 `assistant/message` 是否与其一致。日后若某个格式改变 stream 编码,必须把这些 helper 的冻结副本纳入本迁移边。
该迁移边会重映射有限的已声明引用清单:信封 provenance、surface replacement 端点、command source event、compaction range 与 shadowed list,以及 title message list。经过校验的 `session/title-llm-request` 模型可见文本会在源序号命名空间中保持逐字节不变,而它的 `messageSeqs` 字段会迁移到 v2 命名空间;因此目标校验不会根据重映射后的序号重建该文本。指向被消费 chunk 的引用会使迁移失败;它绝不会被重定向到含义不同的 settlement。该迁移边也会拒绝切开 attempt 的继承切点。
该迁移边会重映射有限的已声明引用清单:信封 source-event reference、surface replacement 端点、command source event、compaction range 与 shadowed list,以及 title message list。经过校验的 `session/title-llm-request` 模型可见文本会在源序号命名空间中保持逐字节不变,而它的 `messageSeqs` 字段会迁移到 v2 命名空间;因此目标校验不会根据重映射后的序号重建该文本。指向被消费 chunk 的引用会使迁移失败;它绝不会被重定向到含义不同的 settlement。该迁移边也会拒绝切开 attempt 的继承切点。
v2 物理 header 要求 `isSeeded`,且不存储数值切点。带 seed 的产物用 `session/end-seed { inherited: true }` 标记其精确切点;解码从最后一个 tagged marker 推导切点。v2 编解码器为每个持久事件写一条物理行,只对 `sourceEventSeqs` 做范围编码,并在不冻结普通事件词汇或 payload 新增项的前提下校验物理 envelope。v1-to-v2 target validator 会另行冻结 released-v2 清单,current restoration 则使用 installed Session 词汇。冻结的 v0 与 v1 编解码器继续为不可变历史 generation 解码 packed row。
@@ -49,7 +49,7 @@ Generation 选择与发布遵循[已发布 Session 迁移决策](2026-08-31-rele
## 验证
紧凑 stream 测试固定 text、reasoning、tool argument、raw chunk、时间戳间隔、格式错误 record 与分离 snapshot 的精确累积和展开。v1 到 v2 测试覆盖成功与失败 attempt、交错、密集序号与引用重映射、源序号 title framing、seed 切点插入与切分拒绝、严格源与目标校验、每行一个事件的 v2 编码、与 backend 兼容的 provenance range、原始与 Zstandard 发布,以及无写入的当前读取。
紧凑 stream 测试固定 text、reasoning、tool argument、raw chunk、时间戳间隔、格式错误 record 与分离 snapshot 的精确累积和展开。v1 到 v2 测试覆盖成功与失败 attempt、交错、密集序号与引用重映射、源序号 title framing、seed 切点插入与切分拒绝、严格源与目标校验、每行一个事件的 v2 编码、与 backend 兼容的 source-event range、原始与 Zstandard 发布,以及无写入的当前读取。
合并前的 performance acceptance 在三轮、100 组 warmup pair 与 600 组 measured pair 下,针对同一批已经解析的物理 row,把静态 catalog routing 与直接 released-v2 restoration 比较;它不比较 v1 与 v2,也不计入 backend I/O。每个 pooled median 与 p95 regression 都保持在 5% 预算以内,最差 p95 regression 为 3.150%。
@@ -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/2026-09-05-workspace-files-service.md
2026-09-05-workspace-files-service.md: b2dbdfc99388d8c2f18991a7704599d2d95f070b
2026-09-05-workspace-files-service.zh.md: d860d220af50a24a6e0f40b640d0a8fb534c4741
2026-09-05-workspace-files-service.md: 40157508a83abc61245845a1dbe4f1749e175563
2026-09-05-workspace-files-service.zh.md: 51de0ad5b644f27e30557ea6b11ba62b8ef9a715
@@ -95,7 +95,7 @@ abstract readByteRange(target: FsTarget, range: { offset: number; length: number
It returns the bytes at `[offset, offset + length)`, shorter when the file ends inside the window and empty when `offset` lies at or past the end. The window is the bound: a backend transfers at most `length` bytes beyond the prefix it skips to reach `offset` and never buffers the whole file, so the caller's cap on `length` is the guard against unbounded buffering, sitting beside `readBytes`'s bound rather than replacing it. The parameter order follows `readText`, `streamText`, and `listDir` — target, then the operation's own arguments, then an optional signal — rather than `readBytes`'s signal-in-the-middle form, which is the one exception in the class. Both `offset` and `length` are non-negative integers by precondition; the seam is a typed same-process boundary and validates nothing, and the Remote method validates at the wire.
`fs-local` opens `createReadStream(targetKey, { start: offset, end: offset + length - 1 })` after the same regular-file stat as its other reads, returning an empty array for `length` 0 without opening a stream; `fs-sandbox` extends `LocalFileSystem` and inherits it. `fs-e2b` has an SDK that streams only from a file's start, so it skips `offset` bytes, copies `length` into the window, and cancels the stream the moment the window is full, transferring no more than the window beyond the skipped prefix; a stream that ends first is left to close. The four test doubles that extend `FileSystem` implement the method too.
`fs-local` opens `createReadStream(targetKey, { start: offset, end: offset + length - 1 })` after the same regular-file stat as its other reads, returning an empty array for `length` 0 without opening a stream; `fs-sandbox` extends `LocalFileSystem` and inherits it. The four test doubles that extend `FileSystem` implement the method too.
### The Client `file` provider
@@ -133,7 +133,7 @@ The [resource model](2026-09-05-client-resource-model.md) owns `ctx.resources`,
- Workspace file access belongs to the Host/Client faces of `api/workspace-files`; the Session Controller carries neither implementation, and compiler and runtime entries stay separate. Header-only Session scope lets ordinary, subagent, live, and cold Sessions resolve their own relative paths without an Agent lifecycle or parent fallback.
- A file of any size opens: text by line page, anything by byte window, each costing one page or window of memory on the Host; complete reads instead enforce `maxFileBytes`; the cost is that a consumer assembles pages itself and that a single line above `maxBytes` has no page at all, because pages are cut by lines.
- Every filesystem provider now offers a windowed raw read. `fs-e2b` pays for it by transferring the skipped prefix, since its SDK cannot seek; `fs-local` seeks.
- Every filesystem provider offers windowed raw reads; `fs-local` seeks directly to the requested offset.
- Paths on the wire are canonical: `absolutePath` and change frames spell a file with symlinks resolved. A follower binds to successful `stat.absolutePath`, so another spelling of the same file — a workspace root reached through a symlink — uses that canonical change key.
- Change frames report the agent's own operations only. A file edited by the user's editor, a shell, or a subprocess raises no frame; an agent merely reading a file that something else changed does raise one, because the read observes a new version.
- File-kind inspection precedes backend reads, and `list` reports kind before an outside position. A page's `version` may be one write behind its content, and a stalled `changes` consumer grows Host memory because a generation's queue is unbounded; each is a known trade-off recorded in the package README.
@@ -142,7 +142,7 @@ The [resource model](2026-09-05-client-resource-model.md) owns `ctx.resources`,
## Testing
Host specs in `packages/api/workspace-files/tests` exercise header-only scope resolution for live and cold subagent Sessions, the deployment fallback, missing identities, and lookup disposal; the paged read (whole file, nested path, empty file, multi-byte UTF-8, the line window's edges, defaults and refused limits, carriage returns kept); the byte window (defaults, a middle window with more following, tail windows exact and short, past-end and empty files, NUL and invalid UTF-8 round-tripping through base64, version parity with `stat`, the cap as `too-large`, bad ranges, a window of a file far above the cap, and `eof` inferred without a size); `stat`; outside-workspace reads and backend refusals; `list` with containment, truncation, symlink children, and `not-directory`; and the `changes` stream driven by `fs/observed` and filtered by root. Client specs cover the provider's frames, the change feed, unsupported addresses, and registration and disposal. `fs/fs`, `fs-local`, and `fs-e2b` specs pin `readByteRange`; `dsh-util-workspace-path` specs pin the file-address grammar. The connection fixture serves `stat`, paged `read`, `list`, and an opt-in `changes` frame for the web e2e suite.
Host specs in `packages/api/workspace-files/tests` exercise header-only scope resolution for live and cold subagent Sessions, the deployment fallback, missing identities, and lookup disposal; the paged read (whole file, nested path, empty file, multi-byte UTF-8, the line window's edges, defaults and refused limits, carriage returns kept); the byte window (defaults, a middle window with more following, tail windows exact and short, past-end and empty files, NUL and invalid UTF-8 round-tripping through base64, version parity with `stat`, the cap as `too-large`, bad ranges, a window of a file far above the cap, and `eof` inferred without a size); `stat`; outside-workspace reads and backend refusals; `list` with containment, truncation, symlink children, and `not-directory`; and the `changes` stream driven by `fs/observed` and filtered by root. Client specs cover the provider's frames, the change feed, unsupported addresses, and registration and disposal. `fs/fs` and `fs-local` specs pin `readByteRange`; `dsh-util-workspace-path` specs pin the file-address grammar. The connection fixture serves `stat`, paged `read`, `list`, and an opt-in `changes` frame for the web e2e suite.
## Deferred
@@ -95,7 +95,7 @@ abstract readByteRange(target: FsTarget, range: { offset: number; length: number
它返回 `[offset, offset + length)` 处的字节,文件在窗内结束则变短,`offset` 位于或越过末尾则为空。窗口即界:后端最多传输为到达 `offset` 而跳过的前缀之外的 `length` 字节,从不缓冲整个文件,因此调用方对 `length` 的上限就是防无界缓冲的守卫,与 `readBytes` 的界并列而非取代它。参数顺序遵循 `readText`、`streamText` 与 `listDir`——先目标,再操作自己的参数,最后可选 signal——而不是 `readBytes` 把 signal 放中间的形式,那是该类中唯一的例外。`offset` 与 `length` 按前置条件都是非负整数;seam 是类型化的同进程边界,不做任何校验,由 Remote 方法在线路处校验。
`fs-local` 在与其他读取相同的普通文件 stat 之后打开 `createReadStream(targetKey, { start: offset, end: offset + length - 1 })`,对 `length` 为 0 直接返回空数组而不开流;`fs-sandbox` 继承 `LocalFileSystem`,随之继承该方法。`fs-e2b` 的 SDK 只能从文件开头开始流式读取,于是它跳过 `offset` 字节、把 `length` 字节拷入窗口,并在窗口填满的那一刻取消流,除跳过的前缀外传输量不超过窗口;先行结束的流则任其关闭。继承 `FileSystem` 的四个测试替身也实现了该方法。
`fs-local` 在与其他读取相同的普通文件 stat 之后打开 `createReadStream(targetKey, { start: offset, end: offset + length - 1 })`,对 `length` 为 0 直接返回空数组而不开流;`fs-sandbox` 继承 `LocalFileSystem`,随之继承该方法。继承 `FileSystem` 的四个测试替身也实现了该方法。
### Client `file` 提供者
@@ -133,7 +133,7 @@ Client 导出向 `ctx.resources` 注册一个 `ResourceProvider<'file'>`,存
- 工作区文件访问由 `api/workspace-files` 的 Host/Client 两面共同承担;Session Controller 不携带其中任何实现,两面的编译与运行时入口保持独立。header-only Session scope 让普通、subagent、live 与 cold Session 都能解析自己的相对路径,不需要 Agent 生命周期,也不回退父 Session。
- 任意大小的文件都能打开:文本按行页、任何文件按字节窗口,在 Host 上各自只花一页或一窗内存;全文读取则受 `maxFileBytes` 约束;代价是消费者自己拼装页面,且单行超过 `maxBytes` 的行没有任何页,因为页按行切。
- 每个文件系统提供者现在都提供开窗的原始读取。`fs-e2b` 为此付出传输被跳过前缀的代价,因为其 SDK 不能 seek;`fs-local` 能 seek。
- 每个文件系统提供者都提供开窗的原始读取;`fs-local` 直接定位到所请求的偏移量。
- 线路上的路径是规范的:`absolutePath` 与变更帧以符号链接已解析的拼法命名文件。跟随者绑定到成功的 `stat.absolutePath`,因此同一文件的另一种拼法——经符号链接到达的工作区根——也使用该规范变更键。
- 变更帧只报告 agent 自己的操作。用户编辑器、shell 或子进程改动的文件不产生帧;agent 仅仅读取一个被别处改动的文件却会产生帧,因为读取观察到了新版本。
- 文件类型检查先于后端读取,`list` 也先报种类再报根外位置。页的 `version` 可能落后内容一次写入,停滞的 `changes` 消费者会让 Host 内存增长,因为一代流的队列无界;每一条都是包 README 记录在册的已知取舍。
@@ -142,7 +142,7 @@ Client 导出向 `ctx.resources` 注册一个 `ResourceProvider<'file'>`,存
## Testing
`packages/api/workspace-files/tests` 中的 Host spec 覆盖 live 与 cold subagent Session 的 header-only scope 解析、部署 fallback、缺失身份与 lookup 释放;分页读取(整文件、嵌套路径、空文件、多字节 UTF-8、行窗口边界、缺省与拒绝的 limit、保留回车);字节窗口(缺省、中段与尾窗、越界与空文件、base64 往返、版本、上限、坏范围以及无大小时的 `eof`);`stat`;工作区外读取及后端拒绝;`list` 的包含、截断、符号链接与 `not-directory`;以及由 `fs/observed` 驱动并按根过滤的 `changes`。Client spec 覆盖提供者帧、变更流、不支持地址及注册与释放。`fs/fs`、`fs-local` 与 `fs-e2b` spec 钉住 `readByteRange`;`dsh-util-workspace-path` spec 钉住文件地址语法。connection fixture 为 web e2e 套件提供 `stat`、分页 `read`、`list` 与一帧可选启用的 `changes`。
`packages/api/workspace-files/tests` 中的 Host spec 覆盖 live 与 cold subagent Session 的 header-only scope 解析、部署 fallback、缺失身份与 lookup 释放;分页读取(整文件、嵌套路径、空文件、多字节 UTF-8、行窗口边界、缺省与拒绝的 limit、保留回车);字节窗口(缺省、中段与尾窗、越界与空文件、base64 往返、版本、上限、坏范围以及无大小时的 `eof`);`stat`;工作区外读取及后端拒绝;`list` 的包含、截断、符号链接与 `not-directory`;以及由 `fs/observed` 驱动并按根过滤的 `changes`。Client spec 覆盖提供者帧、变更流、不支持地址及注册与释放。`fs/fs` 与 `fs-local` spec 钉住 `readByteRange`;`dsh-util-workspace-path` spec 钉住文件地址语法。connection fixture 为 web e2e 套件提供 `stat`、分页 `read`、`list` 与一帧可选启用的 `changes`。
## Deferred
@@ -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
@@ -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.
@@ -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。
Loaded 100 of 3429 files, more files were not shown because too many files have changed in this diff. Show more