dsh-auto-model-routing 0.5.1 → 0.5.3

package/CHANGELOG.md

@@ -1,5 +1,14 @@
 # Changelog
 
+## 0.5.3
+
+- Replace both README demo GIFs with representative PNG screenshots (an applied AUTO decision with its distribution; decision service settings with a connection test), because the DSH plugin store does not show GIFs by default. No runtime behavior changes.
+
+## 0.5.2
+
+- README follows the standard plugin layout: demos side by side, then the architecture diagram, then the introduction; a closing section links the ohmydsh source, the MIT license and the issue tracker.
+- The release workflow now also creates the GitHub Release for each version and attaches the exact npm tarball. No runtime behavior changes.
+
 ## 0.5.1
 
 - Replace both retired four-tier GIFs with high-resolution recordings of the current AUTO interaction, applied decision distribution and settings/strategy UI, using isolated mock services and synthetic tasks.

package/README.md

@@ -3,13 +3,13 @@
 English · [简体中文](README.zh.md)
 
 <p align="center">
-  <img src="docs/auto-routing.gif" alt="AUTO: switch modes, apply effort and inspect the decision distribution" width="49%">
-  <img src="docs/settings.gif" alt="Settings: decision service, connection test, defaults and strategy details" width="49%">
+  <img src="docs/auto-routing.png" alt="AUTO applied a higher effort to the main session; the panel shows the returned decision distribution" width="49%">
+  <img src="docs/settings.png" alt="Settings: decision service, compatible endpoint and a successful connection test" width="49%">
 </p>
 
-*Left: AUTO interaction and applied decision. Right: current settings and strategy details. Recorded from an isolated DSH host with local mock Jev/model responses and synthetic tasks.*
+*Left: an applied AUTO decision with its effort distribution. Right: decision service settings with a successful connection test. Captured from an isolated DSH host with local mock Jev/model responses and synthetic tasks.*
 
-[Repository](https://github.com/prgrmrwy/dsh-auto-model-routing) · [Releases](https://github.com/prgrmrwy/dsh-auto-model-routing/releases) · [npm](https://www.npmjs.com/package/dsh-auto-model-routing)
+![AUTO architecture — independent judgments, validated requests and allocation feedback](docs/architecture-v2.en.svg)
 
 <!-- problem -->
 **Add AUTO mode to DSH: automatic effort for the main agent, automatic model + effort for sub-agents.** Stop adjusting reasoning investment every turn or choosing a model for every delegated task.
@@ -93,12 +93,11 @@ AUTO applies the selected effort to the current main request and awaits official
 - **Local state:** configuration is under `$DSH_HOME/plugins/dsh-auto-model-routing/`; mode, child routes and feedback are under `$DSH_HOME/state/auto-model-routing/`. Feedback stores resource metadata, IDs, the strategy family, returned distributions and timestamps—not task/reply text. Request event IDs allow later usage correlation; costs are not fabricated when unavailable. See [feedback and evaluation](docs/feedback-evaluation.md).
 - **No automatic learning.** Human allocation feedback is observational. It does not change routing or prove quality/cost benefits. Old messages without decision bindings are not backfilled; old records with missing evaluation fields remain unknown.
 - **Upgrade to 0.5.0:** there is one AUTO strategy and no version setting. Existing sessions keep their stored mode; new chats still default to MANUAL unless configured otherwise. Retired `policy` / `mainEfforts` / `childRoutes` keys in the plugin configuration are ignored, shown as no longer effective in Settings, and removed on the next save. Child decisions and feedback recorded by 0.4.x stay readable. To go back, pin the published `dsh-auto-model-routing@0.4.2` (see [CHANGELOG](CHANGELOG.md) for the data boundary).
+
 ## Architecture
 <!-- section: architecture -->
 
-![AUTO architecture — independent judgments, validated requests and allocation feedback](docs/architecture-v2.en.svg)
-
-Main failures disable AUTO; child failures stop dispatch. Explicit model/effort constraints are preserved. The diagram shows the AUTO paths; MANUAL makes no new judgments and fork is unsupported. Feedback does not automatically alter routing.
+The diagram at the top shows the AUTO paths. Main failures disable AUTO; child failures stop dispatch. Explicit model/effort constraints are preserved. MANUAL makes no new judgments and fork is unsupported. Feedback does not automatically alter routing.
 
 Entry: [index.ts](src/index.ts). Enforcement: [policy-host.ts](src/policy-host.ts). Actual rubric: [rubric.json](src/rubric.json). Family strategies (supported models, transports, effort starting points, capability slots): [strategies](src/strategies/index.ts); routing: [routing.ts](src/routing.ts). UI: [client](src/client/index.ts). Coverage and limitations (historical records): [development verification](docs/policy-v2-development.md), [real Magpie acceptance](docs/magpie-live-acceptance.md).
 
@@ -112,9 +111,7 @@ npm run build
 
 Tests use the official AgentLoop with fake LLM streams and mock/loopback Jev responses, not paid model calls. To observe AUTO decisions and their actual token usage, ask the agent to use the `auto-model-routing-report` skill (on demand only; it runs the read-only `lib/usage-report.js`). It is observational, not a savings measurement. Evaluation schema and deduplication rules are in [feedback-evaluation.md](docs/feedback-evaluation.md).
 
-Development stays in the source monorepo; this repository is its release mirror. The package builds and loads independently using its declared dependencies, bundled skill and official DSH services; no monorepo checkout or monorepo instruction file is required. External design references are optional background, not runtime inputs.
-
-The [current specification](https://github.com/prgrmrwy/ohmydsh/blob/main/openspec/specs/auto-model-routing/spec.md) lives in the source monorepo. The bundled [delegation skill](skills/subagent-model-routing/SKILL.md) supplies guidance; enforcement belongs to the official host hooks, not the prompt. No measured task-quality or token-savings claim is made.
+The package builds and loads independently using its declared dependencies, bundled skill and official DSH services; no monorepo checkout or monorepo instruction file is required. External design references are optional background, not runtime inputs. The bundled [delegation skill](skills/subagent-model-routing/SKILL.md) supplies guidance; enforcement belongs to the official host hooks, not the prompt. No measured task-quality or token-savings claim is made.
 
 ## Removal
 <!-- section: removal -->
@@ -124,3 +121,9 @@ dsh plugin --profile web remove dsh-auto-model-routing
 ```
 
 Restart DSH and refresh the page. The plugin's skill/tools and request hooks disappear; settings and logs remain. Without the plugin, child replay is no longer enforced and host restoration uses the original creation configuration. In [ohmydsh](https://github.com/prgrmrwy/ohmydsh), disable the local entry and rebuild instead.
+
+## Source, license and contributing
+
+This plugin is developed in [ohmydsh](https://github.com/prgrmrwy/ohmydsh) (`packages/auto-model-routing`); [this repository](https://github.com/prgrmrwy/dsh-auto-model-routing) is its release mirror, published automatically to [npm](https://www.npmjs.com/package/dsh-auto-model-routing) and [GitHub Releases](https://github.com/prgrmrwy/dsh-auto-model-routing/releases). The [current specification](https://github.com/prgrmrwy/ohmydsh/blob/main/openspec/specs/auto-model-routing/spec.md) lives there too.
+
+[MIT](LICENSE). Issues and pull requests are welcome at [prgrmrwy/dsh-auto-model-routing](https://github.com/prgrmrwy/dsh-auto-model-routing/issues) — including support for model families beyond Claude and GPT.

package/README.zh.md

@@ -3,13 +3,13 @@
 [English](README.md) · 简体中文
 
 <p align="center">
-  <img src="docs/auto-routing.gif" alt="AUTO：切换模式、应用推理投入并查看决策分布" width="49%">
-  <img src="docs/settings.gif" alt="设置：决策服务、连接测试、默认值与策略说明" width="49%">
+  <img src="docs/auto-routing.png" alt="AUTO 已为主会话应用更高推理投入，面板显示返回的决策分布" width="49%">
+  <img src="docs/settings.png" alt="设置：决策服务、兼容端点与连接测试成功" width="49%">
 </p>
 
-*左：AUTO 交互与已应用的决策；右：当前设置页与策略说明。录制来自隔离 DSH 宿主，使用本地模拟 Jev/模型响应与合成任务。*
+*左：已应用的 AUTO 决策与投入分布；右：决策服务设置与连接测试成功。截图来自隔离 DSH 宿主，使用本地模拟 Jev/模型响应与合成任务。*
 
-[插件仓库](https://github.com/prgrmrwy/dsh-auto-model-routing) · [版本发布](https://github.com/prgrmrwy/dsh-auto-model-routing/releases) · [npm](https://www.npmjs.com/package/dsh-auto-model-routing)
+![AUTO 架构：独立判断、请求校验与分配反馈](docs/architecture-v2.zh.svg)
 
 <!-- problem -->
 **为 DSH 增加 AUTO 模式：主 Agent 自动调 effort，子 Agent 自动选 model + effort。** 不必每轮手动选择推理投入，也不必为每个子任务手动挑模型。
@@ -97,9 +97,7 @@ AUTO 将选定 effort 应用于本次主请求；effort 改变时，等待官方
 ## 架构
 <!-- section: architecture -->
 
-![AUTO 架构：独立判断、请求校验与分配反馈](docs/architecture-v2.zh.svg)
-
-主判断失败停用 AUTO，子失败停止派发；保留显式 model/effort。图示为 AUTO 路径，MANUAL 不做新判断，fork 暂不支持。反馈不会自动改变路由。
+顶部的架构图为 AUTO 路径。主判断失败停用 AUTO，子失败停止派发；保留显式 model/effort。MANUAL 不做新判断，fork 暂不支持。反馈不会自动改变路由。
 
 入口：[index.ts](src/index.ts)。执行接入：[policy-host.ts](src/policy-host.ts)。实际判据：[rubric.json](src/rubric.json)。家族策略（支持的模型、transport、effort 起点、能力槽）：[strategies](src/strategies/index.ts)；路由计算：[routing.ts](src/routing.ts)。界面：[client](src/client/index.ts)。覆盖边界（历史记录）：[开发验证](docs/policy-v2-development.md)、[真实 Magpie 验收](docs/magpie-live-acceptance.md)。
 
@@ -113,9 +111,7 @@ npm run build
 
 测试使用官方 AgentLoop、假 LLM 流和 mock/回环 Jev，不产生真实模型费用。如需观察 AUTO 决策及实际 token 用量，请明确让 agent 使用 `auto-model-routing-report` skill（仅按需调用，运行只读的 `lib/usage-report.js`）；它是观察统计，不衡量节省；新数据结构与去重口径见[feedback-evaluation.md](docs/feedback-evaluation.md)。
 
-继续在主仓开发，本仓库仅作发布镜像。插件使用自身声明的依赖、包内 skill 和官方 DSH 服务即可独立构建与加载，不要求检出主仓，也不要求安装主仓环境指令。外部设计链接仅供背景阅读，不是运行输入。
-
-[当前规范](https://github.com/prgrmrwy/ohmydsh/blob/main/openspec/specs/auto-model-routing/spec.md)在源码 monorepo 中维护。包内[委派 skill](skills/subagent-model-routing/SKILL.md)提供指引，执行保障来自官方宿主钩子，而不是提示词。暂无完整任务质量或 token 节省的量化结论。
+插件使用自身声明的依赖、包内 skill 和官方 DSH 服务即可独立构建与加载，不要求检出主仓，也不要求安装主仓环境指令。外部设计链接仅供背景阅读，不是运行输入。包内[委派 skill](skills/subagent-model-routing/SKILL.md)提供指引，执行保障来自官方宿主钩子，而不是提示词。暂无完整任务质量或 token 节省的量化结论。
 
 ## 移除
 <!-- section: removal -->
@@ -125,3 +121,9 @@ dsh plugin --profile web remove dsh-auto-model-routing
 ```
 
 重启 DSH 并刷新页面。插件的 skill、工具和请求钩子撤销，配置与日志保留。关闭插件后不再执行子路由恢复，宿主使用原始创建配置。在 [ohmydsh](https://github.com/prgrmrwy/ohmydsh) 中则禁用 local 条目并重新构建。
+
+## 来源、许可与参与
+
+本插件在 [ohmydsh](https://github.com/prgrmrwy/ohmydsh)（`packages/auto-model-routing`）中持续开发，[本仓库](https://github.com/prgrmrwy/dsh-auto-model-routing)是发布镜像，自动发布到 [npm](https://www.npmjs.com/package/dsh-auto-model-routing) 与 [GitHub Releases](https://github.com/prgrmrwy/dsh-auto-model-routing/releases)。[当前规范](https://github.com/prgrmrwy/ohmydsh/blob/main/openspec/specs/auto-model-routing/spec.md)也在主仓维护。
+
+采用 [MIT](LICENSE) 许可。欢迎到 [prgrmrwy/dsh-auto-model-routing](https://github.com/prgrmrwy/dsh-auto-model-routing/issues) 提 issue 和 PR，包括支持 Claude、GPT 以外的模型家族。

package/docs/SCREENSHOTS.md

@@ -2,5 +2,5 @@
 
 | file | kind | source | date | reviewer | verdict | note |
 | --- | --- | --- | --- | --- | --- | --- |
-| auto-routing.gif | screenshot | isolated DSH_HOME under /tmp, port 3199; current plugin UI with local loopback mock Jev and model streaming responses; Playwright/CDP lossless 2x screenshots, global-palette GIF | 2026-10-28 | lead | pass (reviewed source frames: synthetic task/reply, complete conversation and decision panel; no real credentials or user data) | 1840x1600, 207702 bytes; MANUAL → AUTO, same-model medium → high, applied decision with returned distribution, back to MANUAL |
-| settings.gif | screenshot | same isolated host; lossless 2x screenshots box-filtered to 1137x1070 | 2026-10-28 | lead | pass (reviewed source frames: full settings dialog, empty key input, synthetic loopback endpoint; no real credentials) | 385416 bytes; decision service menu, successful mock connection test, default mode, effort cap, save and expand current strategy details; replaces the retired four-tier demo |
+| auto-routing.png | screenshot | frame 5 of the 2026-10-28 recording (isolated DSH_HOME under /tmp, port 3199; current plugin UI with local loopback mock Jev and model streaming responses; Playwright/CDP lossless 2x screenshots), exported losslessly from the global-palette GIF | 2026-10-29 | lead | pass (reviewed frame: synthetic task/reply and decision panel; no real credentials or user data) | 1840x1600, 245191 bytes; applied same-model medium → high decision with the returned effort distribution; replaces auto-routing.gif because the plugin store does not show GIFs by default |
+| settings.png | screenshot | frame 3 of the same 2026-10-28 recording (lossless 2x screenshots box-filtered to 1137x1070), exported losslessly from the GIF | 2026-10-29 | lead | pass (reviewed frame: settings dialog, empty key input, synthetic loopback endpoint; no real credentials) | 198719 bytes; decision service, decision model, compatible endpoint and a successful mock connection test; replaces settings.gif |

package/lib/client/index.d.ts

@@ -7,6 +7,12 @@ interface AutoSupport {
     model?: string;
 }
 interface View {
+    childNotice?: {
+        id: number;
+        childId: string;
+        reason: string;
+        at: number;
+    };
     mode: 'AUTO' | 'MANUAL';
     depth?: string;
     effort?: string;
@@ -38,6 +44,6 @@ export declare function RoutingChip({ sessionId, rpc, initialView, initiallyOpen
     onMouseLeave: () => void;
     onFocus: () => void;
     onBlur: () => void;
-}, HTMLSpanElement>;
+}, HTMLSpanElement> | null;
 export declare function apply(ctx: Context): void;
 export {};

package/lib/client/index.js

@@ -39,11 +39,13 @@ const CSS = `
 function SwapIcon() {
     return h('svg', { className: 'amr-icon amr-swap', width: 14, height: 14, viewBox: '0 0 16 16', fill: 'none', stroke: 'currentColor', strokeWidth: 1.4, strokeLinecap: 'round', strokeLinejoin: 'round', 'aria-hidden': true }, h('path', { d: 'M3 5.5h9.5M10 3l2.5 2.5L10 8' }), h('path', { d: 'M13 10.5H3.5M6 8l-2.5 2.5L6 13' }));
 }
+const noticeText = (notice) => `最近子任务未路由：${notice.reason}（按 DSH 默认运行）`;
 export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false, initialToast, subscribeSelection }) {
     const [view, setView] = useState(initialView);
     const [open, setOpen] = useState(initiallyOpen);
     const [busy, setBusy] = useState(false);
     const [error, setError] = useState('');
+    const [notApplicable, setNotApplicable] = useState(false);
     const timer = useRef(undefined);
     const [toast, setToast] = useState(initialToast);
     // The composer hides the right-side controls while the agent runs (when Jev decides), so the hint is
@@ -79,6 +81,25 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
         remember();
         setToast(next);
     }, [view, sessionId]);
+    const seenNotice = useRef(undefined);
+    const noticePrimed = useRef(false);
+    useEffect(() => {
+        if (!view || viewSession.current !== sessionId)
+            return;
+        const notice = view.childNotice;
+        if (!noticePrimed.current) {
+            noticePrimed.current = true;
+            seenNotice.current = notice?.id;
+            return;
+        }
+        if (!notice || notice.id === seenNotice.current)
+            return;
+        seenNotice.current = notice.id;
+        if (!view.decisionToast)
+            return;
+        remember();
+        setToast({ id: -notice.id, lines: [noticeText(notice)] });
+    }, [view, sessionId]);
     // Hiding depends only on the toast shown, so later state changes cannot cancel it.
     useEffect(() => {
         if (!toast)
@@ -111,19 +132,34 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
     useEffect(() => {
         let active = true;
         primed.current = false;
+        noticePrimed.current = false;
         seen.current = undefined;
         viewSession.current = undefined;
         setToast(undefined);
-        const refresh = () => { void callRouting(rpc, sessionId).then(value => { if (active) {
-            viewSession.current = sessionId;
-            setView(value);
-            setError('');
-            ensureSubscription();
-        } }, reason => { if (active)
-            setError(String(reason)); }); };
+        let poll;
+        const refresh = () => {
+            void callRouting(rpc, sessionId).then(value => {
+                if (!active)
+                    return;
+                // Child sessions have no AUTO chip; their origin never changes, so stop polling for good.
+                if (value?.applicable === false) {
+                    setNotApplicable(true);
+                    clearInterval(poll);
+                    subscription.current?.stop();
+                    subscription.current = undefined;
+                    return;
+                }
+                viewSession.current = sessionId;
+                setView(value);
+                setError('');
+                ensureSubscription();
+            }, reason => { if (active)
+                setError(String(reason)); });
+        };
         refreshRef.current = refresh;
+        setNotApplicable(false);
         refresh();
-        const poll = setInterval(refresh, 3000);
+        poll = setInterval(refresh, 3000);
         return () => {
             active = false;
             clearInterval(poll);
@@ -180,8 +216,11 @@ export function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false
             h('p', { key: 'depth', className: 'amr-muted' }, `AUTO · ${view?.depth ?? '尚无判断'}；主模型不变，推理投入按请求应用；子任务自动选配。`),
             off && h('p', { key: 'disabled', role: 'status', className: 'amr-error' }, `Jev 停用（${off.category}）：${hint}。${new Date(off.at).toLocaleString()}。点击 AUTO 重试。`),
             h('p', { key: 'data', className: 'amr-muted' }, 'AUTO 会把脱敏任务文本发送给设置中所选的决策服务判断。'),
+            view?.childNotice && h('p', { key: 'notice', role: 'status', className: 'amr-muted' }, noticeText(view.childNotice)),
             error && h('p', { key: 'error', role: 'alert', className: 'amr-error' }, error),
         ];
+    if (notApplicable)
+        return null;
     return h('span', { ref: (el) => { rootRef.current = el; remember(); }, className: 'amr-root', onMouseEnter: on('enter'), onMouseLeave: on('leave'), onFocus: on('focus'), onBlur: on('blur') }, h('style', null, CSS), h('button', {
         type: 'button', className: 'amr-trigger', disabled: busy || !view, onClick: () => { if (locked)
             return; void toggle(); },

package/lib/client.js

@@ -312,11 +312,6 @@ window.__ModuleLoader__.load({
 			id: "claude",
 			label: "Claude",
 			modelPrefixes: ["claude-"],
-			transports: {
-				claude: "",
-				"dsh-magpie-connect": "claude/",
-				"dsh-magpie-connector": "claude/"
-			},
 			effortStarts: {
 				"claude-opus-5-5": "medium",
 				"claude-sonnet-5-5": "medium",
@@ -335,11 +330,6 @@ window.__ModuleLoader__.load({
 			id: "codex",
 			label: "GPT",
 			modelPrefixes: ["gpt-"],
-			transports: {
-				codex: "",
-				"dsh-magpie-connect": "codex/",
-				"dsh-magpie-connector": "codex/"
-			},
 			effortStarts: {
 				"gpt-6.1-sol": "medium",
 				"gpt-6-luna": "high",
@@ -364,6 +354,14 @@ window.__ModuleLoader__.load({
 			"xhigh"
 		];
 		const bare = (model) => model.slice(model.lastIndexOf("/") + 1);
+		const prefixOf = (model) => model.slice(0, Math.max(0, model.lastIndexOf("/")));
+		const KEYS = /* @__PURE__ */ new Set([
+			"id",
+			"label",
+			"modelPrefixes",
+			"effortStarts",
+			"childModels"
+		]);
 		const overlaps = (a, b) => a.startsWith(b) || b.startsWith(a);
 		/** Validates the whole list up front; any conflict throws and no partial registry exists. */
 		function createRegistry(input) {
@@ -371,9 +369,9 @@ window.__ModuleLoader__.load({
 			const ids = /* @__PURE__ */ new Set();
 			const starts = /* @__PURE__ */ new Map();
 			const prefixes = [];
-			const childPrefixes = /* @__PURE__ */ new Map();
 			for (const s of strategies) {
 				if (!s.id) throw new Error("策略缺少 id");
+				for (const key of Object.keys(s)) if (!KEYS.has(key)) throw new Error(`策略 ${s.id} 含未知键：${key}`);
 				if (ids.has(s.id)) throw new Error(`策略 id 重复：${s.id}`);
 				ids.add(s.id);
 				for (const prefix of s.modelPrefixes) {
@@ -391,18 +389,10 @@ window.__ModuleLoader__.load({
 					if (starts.has(model)) throw new Error(`起点键重复：${model}（${starts.get(model)} 与 ${s.id}）`);
 					starts.set(model, s.id);
 				}
-				for (const capability of CAPABILITIES) if (typeof s.childModels?.[capability] !== "string" || !s.childModels[capability]) throw new Error(`策略 ${s.id} 缺少能力槽：${capability}`);
-				for (const [provider, prefix] of Object.entries(s.transports)) {
-					const existing = childPrefixes.get(provider) ?? [];
-					const clash = existing.find((p) => p.prefix === "" || prefix === "" ? true : overlaps(p.prefix, prefix));
-					if (clash) {
-						if (clash.prefix === "" || prefix === "") throw new Error(`provider ${provider} 的空前缀只能属于一个策略（${clash.id} 与 ${s.id} 冲突）`);
-						throw new Error(`provider ${provider} 下子模型前缀重叠：${prefix}（${s.id}）与 ${clash.prefix}（${clash.id}）`);
-					}
-					childPrefixes.set(provider, [...existing, {
-						prefix,
-						id: s.id
-					}]);
+				for (const capability of CAPABILITIES) {
+					const slot = s.childModels?.[capability];
+					if (typeof slot !== "string" || !slot) throw new Error(`策略 ${s.id} 缺少能力槽：${capability}`);
+					if (slot.includes("/")) throw new Error(`策略 ${s.id} 的能力槽不能含 /：${slot}`);
 				}
 			}
 			const byId = new Map(strategies.map((s) => [s.id, s]));
@@ -413,23 +403,17 @@ window.__ModuleLoader__.load({
 					const name = bare(model).toLowerCase();
 					return strategies.find((s) => s.modelPrefixes.some((p) => name.startsWith(p.toLowerCase())));
 				},
-				familyOfChild: (provider, model) => {
-					const entries = childPrefixes.get(provider) ?? [];
-					const hit = entries.find((e) => e.prefix !== "" && model.startsWith(e.prefix)) ?? (entries.length === 1 && entries[0].prefix === "" ? entries[0] : void 0);
-					return hit ? byId.get(hit.id) : void 0;
-				},
-				childModel: (familyId, provider, capability) => {
-					const s = byId.get(familyId);
-					const prefix = s?.transports[provider];
-					return s && prefix !== void 0 ? `${prefix}${s.childModels[capability]}` : void 0;
+				resolveSlot: (familyId, capability, parentModel, models) => {
+					const slot = byId.get(familyId)?.childModels[capability]?.toLowerCase();
+					if (!slot) return void 0;
+					const hits = models.filter((m) => bare(m).toLowerCase() === slot && prefixOf(m) === prefixOf(parentModel));
+					return hits.length === 1 ? hits[0] : void 0;
 				},
 				effortStart: (model) => {
 					const name = bare(model);
 					for (const s of strategies) if (Object.hasOwn(s.effortStarts, name)) return s.effortStarts[name];
 				},
-				hasProvider: (provider) => childPrefixes.has(provider),
 				get: (familyId) => byId.get(familyId),
-				providers: () => [...childPrefixes.keys()],
 				ids: () => strategies.map((s) => s.id),
 				labels: () => strategies.map((s) => s.label)
 			});
@@ -842,11 +826,13 @@ window.__ModuleLoader__.load({
 				"aria-hidden": true
 			}, (0, react.createElement)("path", { d: "M3 5.5h9.5M10 3l2.5 2.5L10 8" }), (0, react.createElement)("path", { d: "M13 10.5H3.5M6 8l-2.5 2.5L6 13" }));
 		}
+		const noticeText = (notice) => `最近子任务未路由：${notice.reason}（按 DSH 默认运行）`;
 		function RoutingChip({ sessionId, rpc, initialView, initiallyOpen = false, initialToast, subscribeSelection }) {
 			const [view, setView] = (0, react.useState)(initialView);
 			const [open, setOpen] = (0, react.useState)(initiallyOpen);
 			const [busy, setBusy] = (0, react.useState)(false);
 			const [error, setError] = (0, react.useState)("");
+			const [notApplicable, setNotApplicable] = (0, react.useState)(false);
 			const timer = (0, react.useRef)(void 0);
 			const [toast, setToast] = (0, react.useState)(initialToast);
 			const rootRef = (0, react.useRef)(null);
@@ -876,6 +862,25 @@ window.__ModuleLoader__.load({
 				remember();
 				setToast(next);
 			}, [view, sessionId]);
+			const seenNotice = (0, react.useRef)(void 0);
+			const noticePrimed = (0, react.useRef)(false);
+			(0, react.useEffect)(() => {
+				if (!view || viewSession.current !== sessionId) return;
+				const notice = view.childNotice;
+				if (!noticePrimed.current) {
+					noticePrimed.current = true;
+					seenNotice.current = notice?.id;
+					return;
+				}
+				if (!notice || notice.id === seenNotice.current) return;
+				seenNotice.current = notice.id;
+				if (!view.decisionToast) return;
+				remember();
+				setToast({
+					id: -notice.id,
+					lines: [noticeText(notice)]
+				});
+			}, [view, sessionId]);
 			(0, react.useEffect)(() => {
 				if (!toast) return;
 				const hide = setTimeout(() => setToast(void 0), TOAST_MS);
@@ -910,24 +915,33 @@ window.__ModuleLoader__.load({
 			(0, react.useEffect)(() => {
 				let active = true;
 				primed.current = false;
+				noticePrimed.current = false;
 				seen.current = void 0;
 				viewSession.current = void 0;
 				setToast(void 0);
+				let poll;
 				const refresh = () => {
 					callRouting(rpc, sessionId).then((value) => {
-						if (active) {
-							viewSession.current = sessionId;
-							setView(value);
-							setError("");
-							ensureSubscription();
+						if (!active) return;
+						if (value?.applicable === false) {
+							setNotApplicable(true);
+							clearInterval(poll);
+							subscription.current?.stop();
+							subscription.current = void 0;
+							return;
 						}
+						viewSession.current = sessionId;
+						setView(value);
+						setError("");
+						ensureSubscription();
 					}, (reason) => {
 						if (active) setError(String(reason));
 					});
 				};
 				refreshRef.current = refresh;
+				setNotApplicable(false);
 				refresh();
-				const poll = setInterval(refresh, 3e3);
+				poll = setInterval(refresh, 3e3);
 				return () => {
 					active = false;
 					clearInterval(poll);
@@ -1006,12 +1020,18 @@ window.__ModuleLoader__.load({
 					key: "data",
 					className: "amr-muted"
 				}, "AUTO 会把脱敏任务文本发送给设置中所选的决策服务判断。"),
+				view?.childNotice && (0, react.createElement)("p", {
+					key: "notice",
+					role: "status",
+					className: "amr-muted"
+				}, noticeText(view.childNotice)),
 				error && (0, react.createElement)("p", {
 					key: "error",
 					role: "alert",
 					className: "amr-error"
 				}, error)
 			];
+			if (notApplicable) return null;
 			return (0, react.createElement)("span", {
 				ref: (el) => {
 					rootRef.current = el;

package/lib/index.js

@@ -11,6 +11,7 @@ import { policyHost } from './policy-host.js';
 import { PolicyStore } from './policy-store.js';
 import { currentMainSelection, mainSupported } from './support.js';
 import { createRegistry, registry as builtin } from './strategies/index.js';
+const SCOPE_NOTE = '按官方工具 ALS 上下文、同父与 startup 绑定；不能区分同窗口插件旁支，旁支可能一并受管。';
 export const name = 'dsh-auto-model-routing';
 /** Child decision records. The historical directory name is kept on purpose: records are not migrated. */
 export const CHILD_DECISIONS_DIR = 'state/auto-model-routing/policy-v2';
@@ -19,7 +20,7 @@ export function apply(ctx, config = {}) {
     const home = config.home ?? process.env.DSH_HOME ?? resolve(homedir(), '.dsh');
     const registry = config.strategies ? createRegistry(config.strategies) : builtin;
     const settings = new SettingsStore(resolve(home, 'plugins/dsh-auto-model-routing/config.json'));
-    const hostSettings = settingsHost(ctx, settings, registry);
+    const hostSettings = settingsHost(ctx, settings);
     const jev = new JevClient({ get: hostSettings.credential }, config.jevBase, config.jevTimeoutMs ?? (() => settings.read().settings.jevTimeoutMs), config.jevBase ? undefined : () => settings.read().settings.decisionService);
     const stateFiles = new FileStateStore(resolve(home, 'state/auto-model-routing/sessions'));
     const state = new RoutingState(stateFiles);
@@ -27,7 +28,7 @@ export function apply(ctx, config = {}) {
     // Official writer is available only while the session controller is mounted.
     let selectModel;
     const policy = policyHost(ctx, { feedback, jev, state, selectModel: async (request) => { if (!selectModel)
-            throw new Error('session controller is unavailable'); await selectModel(request); }, store: new PolicyStore(resolve(home, CHILD_DECISIONS_DIR), registry), settings: () => settings.read().settings, catalog: () => hostSettings.catalog(), registry });
+            throw new Error('session controller is unavailable'); await selectModel(request); }, store: new PolicyStore(resolve(home, CHILD_DECISIONS_DIR), registry), settings: () => settings.read().settings, catalog: provider => hostSettings.catalog(provider), registry });
     const toastOn = () => settings.read().settings.decisionToast;
     ctx.effect(() => ctx.tools.register(routeSubagentTool(jev, { agents: ctx.agents, state, readSettings: () => settings.read().settings, catalog: hostSettings.catalog, registry })));
     ctx.on('agent/created', ({ agent, source }) => {
@@ -57,6 +58,57 @@ export function apply(ctx, config = {}) {
         child.effect(() => () => { selectModel = undefined; });
     });
     ctx.inject(['connection', 'sessionController'], child => {
+        const modeRequest = async (operation, payload) => {
+            const resolveAgent = async (id) => {
+                const result = await child.sessionController.resolveAgent(id);
+                return 'agent' in result ? ref(result.agent) : undefined;
+            };
+            // Child sessions have no AUTO chip: answer before any support check, lock or decoration.
+            if (operation === 'state' && typeof payload?.sessionId === 'string') {
+                const target = await resolveAgent(payload.sessionId);
+                if (target?.origin === 'subagent')
+                    return Response.json({ applicable: false });
+            }
+            let manualDecision;
+            if (operation === 'setMode' && typeof payload?.sessionId === 'string') {
+                const id = payload.sessionId, mode = payload.mode;
+                if ((mode === 'AUTO' || mode === 'MANUAL') && state.read(id).mode !== mode && await resolveAgent(id)) {
+                    const decisionId = policy.appliedDecision(id);
+                    if (mode === 'MANUAL' && decisionId !== undefined)
+                        manualDecision = { sessionId: id, decisionId };
+                    policy.invalidate(id);
+                }
+            }
+            const response = await endpoint(operation, payload, resolveAgent, state);
+            if (!response.ok)
+                return response;
+            if (manualDecision) {
+                try {
+                    feedback.fact(manualDecision.sessionId, manualDecision.decisionId, 'manual-switch');
+                }
+                catch {
+                    ctx.logger.warn('auto-model-routing: could not record MANUAL switch');
+                }
+            }
+            const sessionId = payload.sessionId;
+            const resolved = await child.sessionController.resolveAgent(sessionId).catch(() => undefined);
+            const selection = resolved && 'agent' in resolved ? currentMainSelection(ctx, resolved.agent) : undefined;
+            const supported = mainSupported(selection, registry);
+            if (!supported && state.read(sessionId).mode === 'AUTO') {
+                state.setMode(sessionId, 'MANUAL');
+                policy.invalidate(sessionId);
+            }
+            const { previous: _previous, ...persisted } = await response.json();
+            return Response.json({ ...persisted, mode: state.read(sessionId).mode, ...policy.mainStatus(sessionId), ...(policy.childNotice(sessionId) ? { childNotice: policy.childNotice(sessionId) } : {}), ...(toastOn() ? { decisionToast: true } : {}), autoSupport: { supported, ...(selection ? { provider: selection.provider, model: selection.model } : {}) } });
+        };
+        child.provide('autoModelRouting', { call: async (operation, payload) => {
+                if (operation !== 'state' && operation !== 'setMode')
+                    throw new Error('Unsupported routing operation');
+                const response = await modeRequest(operation, payload);
+                if (!response.ok)
+                    throw new Error('Routing mode request rejected');
+                return response.json();
+            } });
         for (const operation of ['state', 'setMode', 'label', 'observe', 'settings', 'saveSettings', 'setKey', 'testDecision', 'childRouting']) {
             child.effect(() => child.connection.fetch.register({
                 path: `/api/auto-model-routing.${operation}`,
@@ -77,54 +129,15 @@ export function apply(ctx, config = {}) {
                         const result = await child.sessionController.resolveAgent(id);
                         return 'agent' in result ? ref(result.agent) : undefined;
                     };
-                    const resolveSessionAgent = async (id) => {
-                        const result = await child.sessionController.resolveAgent(id);
-                        return 'agent' in result ? result.agent : undefined;
-                    };
                     const payload = enveloped ? wire.payload : body;
-                    let manualDecision;
-                    if (operation === 'setMode' && typeof payload?.sessionId === 'string') {
-                        const id = payload.sessionId;
-                        const mode = payload.mode;
-                        if ((mode === 'AUTO' || mode === 'MANUAL') && state.read(id).mode !== mode && await resolveAgent(id)) {
-                            const decisionId = policy.appliedDecision(id);
-                            if (mode === 'MANUAL' && decisionId !== undefined)
-                                manualDecision = { sessionId: id, decisionId };
-                            policy.invalidate(id);
-                        }
-                    }
                     const response = operation === 'childRouting'
                         ? (typeof payload?.sessionId === 'string'
-                            ? Response.json(policy.status(payload.sessionId) ?? { status: 'unsupported', reason: '没有已绑定的子任务决策' })
+                            ? Response.json((record => record ? { ...record, scopeNote: SCOPE_NOTE } : { status: 'unsupported', reason: '没有已绑定的子任务决策' })(policy.status(payload.sessionId)))
                             : Response.json({ error: 'invalid request' }, { status: 400 }))
                         : operation === 'settings' || operation === 'saveSettings' || operation === 'setKey' || operation === 'testDecision'
                             ? await hostSettings.endpoint(operation, payload)
                             : operation === 'observe' ? await observeEndpoint(payload, resolveAgent, feedback, () => ctx.logger.warn('auto-model-routing: could not record exposure'))
-                                : operation === 'label' ? await labelEndpoint(payload, resolveAgent, feedback) : await endpoint(operation, payload, resolveAgent, state);
-                    if (manualDecision && response.ok) {
-                        try {
-                            feedback.fact(manualDecision.sessionId, manualDecision.decisionId, 'manual-switch');
-                        }
-                        catch {
-                            ctx.logger.warn('auto-model-routing: could not record MANUAL switch');
-                        }
-                    }
-                    if ((operation === 'state' || operation === 'setMode') && response.ok) {
-                        const sessionId = payload.sessionId;
-                        const agent = await resolveSessionAgent(sessionId).catch(() => undefined);
-                        const selection = agent ? currentMainSelection(ctx, agent) : undefined;
-                        const supported = mainSupported(selection, registry);
-                        // Leaving AUTO is persisted: a pick outside the registered families must not keep judging turns.
-                        if (!supported && state.read(sessionId).mode === 'AUTO') {
-                            state.setMode(sessionId, 'MANUAL');
-                            policy.invalidate(sessionId);
-                        }
-                        const value = await response.json();
-                        const main = policy.mainStatus(sessionId);
-                        const { previous: _previous, ...persisted } = value;
-                        const result = { ...persisted, mode: state.read(sessionId).mode, ...main, ...(toastOn() ? { decisionToast: true } : {}), autoSupport: { supported, ...(selection ? { provider: selection.provider, model: selection.model } : {}) } };
-                        return Response.json(enveloped ? { type: 'server-response', rpcId: wire.rpcId, result: { ok: true, value: result } } : result);
-                    }
+                                : operation === 'label' ? await labelEndpoint(payload, resolveAgent, feedback) : await modeRequest(operation, payload);
                     if (!enveloped)
                         return response;
                     const value = await response.json();

package/lib/policy-host.d.ts

@@ -17,16 +17,22 @@ export declare function policyHost(ctx: Context, deps: {
     state: RoutingState;
     store: PolicyStore;
     settings: () => Settings;
-    catalog: () => Promise<RoutingCatalog>;
+    catalog: (provider?: string) => Promise<RoutingCatalog>;
     registry?: Registry;
 }): {
     appliedDecision: (id: string) => number | undefined;
     invalidate: (id: string) => void;
+    childNotice: (rootId: string) => {
+        id: number;
+        childId: string;
+        reason: string;
+        at: number;
+    } | undefined;
     status: (id: string) => {
         childId: string;
         parentId: string;
         rootId: string;
-        status: "failed" | "recommended" | "unsupported" | "needs-context" | "applied";
+        status: "failed" | "recommended" | "needs-context" | "unsupported" | "applied";
         path: "ordinary" | "fork" | "teammate";
         explicit: {
             provider: string;
@@ -34,6 +40,11 @@ export declare function policyHost(ctx: Context, deps: {
             reasoningEffort?: string | undefined;
         };
         family?: string | undefined;
+        parentSelection?: {
+            provider: string;
+            model?: string | undefined;
+            reasoningEffort?: string | undefined;
+        } | undefined;
         decision?: {
             id: number;
             source: "main" | "subagent";