dsh-session-guard 4.1.5 → 4.1.15

package/CHANGELOG.ja.md

@@ -11,6 +11,18 @@
 - [Changelog на русском](./CHANGELOG.ru.md)
 - [Changelog en español](./CHANGELOG.es.md)
 
+## 4.1.14 — 2026-10-08
+
+### 削除
+
+## 4.1.13 — 2026-10-08
+
+### 修正 — ≤0.1.5 でのブリッジ記述子スケルトン（5セルゲートで発覚）
+
+- **≤0.1.5 ホストの describe `value` はフィールドスケルトン**（実測：`{enabled:{},…}`、実値は `base`＋`user` 層）。旧世代で `descriptor.value` をそのまま消費すると `officialProviders` が `{}` になり `?? []` を素通りして `ListRow` の `values.join` が throw、`dsh-family.tab` スロット全体が殉葬していた。`descriptorPlainValue()` がスケルトンを検出して `{...base, ...user}` を再構築。
+- **カード字段消毒（多層防御）**：スロットは一つの描画エラーで全族タブが死ぬため、`ListRow` 入力を `toList()` で無毒化、`拦截方式` セレクトは非 `error` 値を `hold` にフォールバック。
+- 5セルファームゲートで検証：0.1.0/0.1.5 はブリッジ経由、0.1.7 は configForms 経由でシード値 `321123` の読み返しを確認。
+
 ## 4.0.0 — 2026-09-29
 
 ### Changed

package/CHANGELOG.ko.md

@@ -11,6 +11,18 @@
 - [Changelog на русском](./CHANGELOG.ru.md)
 - [Changelog en español](./CHANGELOG.es.md)
 
+## 4.1.14 — 2026-10-08
+
+### 제거
+
+## 4.1.13 — 2026-10-08
+
+### 수정 — ≤0.1.5 브리지 디스크립터 스켈레톤(5셀 게이트 발견)
+
+- **≤0.1.5 호스트의 describe `value`는 필드 스켈레톤**(실측: `{enabled:{},…}`, 실값은 `base`+`user` 레이어). 구세대에서 `descriptor.value`를 그대로 쓰면 `officialProviders`가 `{}`가 되어 `?? []`를 통과하고 `ListRow`의 `values.join`이 throw, `dsh-family.tab` 슬롯 전체가 순직.`descriptorPlainValue()`가 스켈레톤을 감지해 `{...base, ...user}`로 재구성.
+- **카드 필드 소독(다층 방어)**: 슬롯은 렌더 오류 하나로 전체 탭이 죽으므로 `ListRow` 입력을 `toList()`로 무독화, `拦截方式` 셀렉트는 non-`error` 값을 `hold`로 폴백.
+- 5셀 팜 게이트 검증: 0.1.0/0.1.5는 브리지 경유, 0.1.7은 configForms 경유로 시드값 `321123` 재판독 확인.
+
 ## 4.0.0 — 2026-09-29
 
 ### Changed

package/CHANGELOG.md

@@ -11,6 +11,92 @@ All notable changes to `dsh-session-guard` are recorded here. Versions follow se
 - [Changelog на русском](./CHANGELOG.ru.md)
 - [Changelog en español](./CHANGELOG.es.md)
 
+## 4.1.14 — 2026-10-08
+
+### Removed
+
+## 4.1.13 — 2026-10-08
+
+### Fixed — bridge descriptor skeleton on ≤0.1.5 (five-cell gate catch, 2026-10-08)
+
+- **Host describe `value` is a field skeleton on ≤0.1.5** (0.1.0/0.1.5 empirical):
+  the old host generations serialize `descriptor.value` as `{enabled:{},…,officialProviders:{}}`
+  (one empty object per schema field) while the real merged config lives in
+  `descriptor.base` (combined defaults) + `descriptor.user` (user layer).
+  `BridgeDocHandle` consumed `value` verbatim, so `officialProviders` arrived as
+  `{}` — truthy non-array, invisible to `?? []` — and `ListRow`'s
+  `values.join` threw, crashing the whole `dsh-family.tab` slot (guard card +
+  IT card + TL nav all died together). `descriptorPlainValue()` now detects the
+  skeleton (object whose every own value is an empty plain object — impossible
+  for guard's scalar/array config) and rebuilds the effective value as
+  `{...base, ...user}` (shallow merge is exact for guard's flat config).
+- **Card-field sanitization (defense in depth)**: any card render error kills
+  the entire family tab slot on these hosts, so `ListRow` inputs are now
+  sanitized at the call sites (`toList()` — non-arrays become `[]`) and the
+  `拦截方式` select falls back to `hold` for any non-`error` value. A malformed
+  descriptor may degrade one field, never the tab.
+- Verified on the five-cell farm gate: describe replays on 0.1.0/0.1.5 now
+  yield plain merged values; the seeded `stepGateTimeoutMs: 321123` reads back
+  through the bridge on ≤0.1.5 and natively (configForms) on 0.1.7.
+
+## 4.1.12 — 2026-10-08
+
+### Added — settings webServer bridge (M4/T20-b): dual-track card data plane
+
+- **Top-level card registration refactor (prerequisite)**: the three settings
+  faces (`dsh-family.tab`, `plugins.bundle.config`, `settings.plugin.item`,
+  `shell.overlay`) were registered inside the `onScope` callback of the dual
+  scoped-inject waist — on ≤0.1.5 that callback never fires (settingsScope fiber
+  never resolves; 0.1.0/0.1.1 lack both services), so the cards never registered
+  and a bridge would have had no consumer. All registrations now run at apply
+  top level with a `scopeRef` live reference; the callback degrades to a pure
+  data arm (perm-gate `client/index.ts` L166-257 shape).
+- **Server bridge** (`src/settings-bridge.js`): `describe`/`mutate` endpoints at
+  `/api/session-guard/settings/*` (POST-only, `{ok,value|error}` envelope,
+  redacted describe filtered to the plugin ns, ops forwarded to the host
+  settings service with `expectedRevision` optimistic concurrency, 256 KB cap).
+  Registration is fail-open (error log, never blocks plugin load).
+- **Client bridge handle** (`src/client/settings-bridge.ts`): `BridgeDocHandle`
+  implements the same `SettingsScope` face as the native handles
+  (getSnapshot/subscribe/set/unset); describe misses retry within a bounded
+  budget (settings service resolving later than card mount) before landing on
+  `unavailable`. Card faces read `scopeRef.scope ?? bridgeScope()` — native
+  first, bridge fallback (perm-gate T13b dual-track). **Generation split in the
+  data arm (five-cell empirical, 2026-10-08 night)**: on 0.1.2/0.1.5 the
+  `settingsScope` handle is write-alive but read-dead — `set()` persists to
+  settings.yaml while `getSnapshot().value` stays `undefined`, so that arm
+  deliberately leaves `scopeRef` empty and the card falls to the bridge (bridge
+  mutate lands in the same host service; describe heals the read path).
+
+### Corrected — 4.1.11 publication note
+
+- The 4.1.11 entry said "none of 4.1.5-4.1.10 was published to npm"; **4.1.5
+  was published** (npm registry confirms `4.1.5` in versions, and `latest`
+  pointed at it until this release line). 4.1.6-4.1.10 indeed never shipped.
+  The earlier claim came from reading the npmmirror mirror instead of the
+  authoritative registry (registry-pin discipline, B2+).
+
+## 4.1.11 — 2026-10-08
+
+### Fixed — family-section takeover rebuilt (4.1.5 → 4.1.11 arc; none of 4.1.5-4.1.10 was published to npm)
+
+- **4.1.5 "half backfill" root cause**: the takeover body read `props.hooks.tabs`,
+  but the host's `bindInjectHooks` strips an inject face's `hooks` key into
+  `use<Tab>` props on every generation — the component threw, the slot abdicated,
+  and the takeover body never rendered anywhere. It now reads the `useTabs`
+  compartment first, with a raw-hooks `useState`/`useEffect` fallback.
+- **Apply-time inject registration (4.1.10)**: the slots core fires inject
+  factories only on declaration events (= the settings panel opening), so the
+  grace-delayed registration of 4.1.7/4.1.8 waited for a declaration that never
+  came. The `settings.section` inject now registers at apply time.
+- **Deferred yield-election (4.1.11)**: ≤0.1.7 shells give every same-id section
+  entry its own nav row — the old "priority-10 shadow stays invisible" assumption
+  produced a double 起子插件设置 row on all five farm cells. The factory now waits
+  500 ms after the declaration batch and stands down when a `dsh-family` entry is
+  already seated (thinking-levels head); it takes over only when TL is absent.
+  Verified single-row on all five farm cells (0.1.0 → 0.1.7), pairing with
+  dsh-thinking-levels 4.2.0-beta.26's symmetric stand-down.
+
 ## 4.0.0 — 2026-09-29
 
 ### Changed

package/dsh.plugin.json

@@ -1,16 +1,16 @@
-{
-  "id": "dsh-external/dsh-session-guard",
-  "version": "4.0.0",
-  "main": "./src/index.js",
-  "description": "高峰自动会话门：step 级门控 + 周末模式 + 高峰自动暂停 + 官方源二维判定 + 会话级冻结 + 后端自动重试。",
-  "engines": {
-    "dsh": ">=0.2.0-rc.1 <0.2.1-0"
-  },
-  "contributes": {
-    "tools": [],
-    "skills": []
-  },
-  "client": {
-    "main": "./lib/client.js"
-  }
-}
+{
+  "id": "dsh-external/dsh-session-guard",
+  "version": "4.1.14",
+  "main": "./src/index.js",
+  "description": "高峰自动会话门：step 级门控 + 周末模式 + 高峰自动暂停 + 官方源二维判定 + 会话级冻结 + 后端自动重试。",
+  "engines": {
+    "dsh": ">=0.2.0-rc.1 <0.2.1-0"
+  },
+  "contributes": {
+    "tools": [],
+    "skills": []
+  },
+  "client": {
+    "main": "./lib/client.js"
+  }
+}