dsh-model-switch 0.4.36 → 0.4.38

package/CHANGELOG.md

@@ -1,3 +1,24 @@
+## v0.4.38
+
+计划卡的「设置为目标并执行」现在会**真的顶掉**旧目标：宿主把 `GoalError` 经 RPC 网关中转成 `gateway/internal`，而插件只认 `GOAL_ALREADY_EXISTS`，于是 get→clear→create 的替换分支一直是死代码，点一次报一次「设置目标失败」。
+
+- **P0 root cause: the conflict was matched on a code the gateway never sends (`src/picker/plan-goal.ts`).** `@deepseek-ai/dsh-goal` 的 `create()` 对任何非 `complete` 的现存目标 `throw new GoalError('goal "…" already exists with phase "…"', 'GOAL_ALREADY_EXISTS')`；网关把它转成 `{ code: 'gateway/internal', message: <原文> }`。插件比较的是 `error.code === 'GOAL_ALREADY_EXISTS'`，永远不成立，替换流程从未执行——用户看到的就是 `设置目标失败：gateway/internal: goal "goal-32ce…" already exists with phase "paused"`（该会话目标卡在 paused）。已核对内核实现：`clear(agent, ref)` 只做 id+revision 的 compare-and-set，**不限制 phase**，所以 active/paused/blocked 都能被顶替；`create` 只放过 `complete`。
+- **Fix 1: recognise both refusal shapes.** `isAlreadyExistsError()` 先认域码，再用 `/already exists with phase/i` 识别网关转述的文本。只有这条消息才允许清旧目标——其它 `gateway/internal` 拒绝（如 agent 不存活）一律不碰现有目标，用例钉死。
+- **Fix 2: transport failures no longer escape.** `create` / `get` / `clear` 全部经 `guarded()`：被 reject 的调用转成 `{ code: 'GOAL_TRANSPORT', message }` 的拒绝信封，不再冒泡成未处理异常（否则卡片丢掉错误态、计划停在待审却没有解释）。
+- **Fix 3: readable messages.** `normalizeHostMessage()` 剥掉 `gateway/xxx:` 前缀，`failure()` 丢弃传输层码，卡片不会再出现双重前缀；失败信息在源头就干净，卡片无需改动。
+- **Tests.** `tests/picker/plan-goal.spec.ts` 新增 6 例：网关形状的 paused / blocked 替换、非冲突网关拒绝不清目标、create / get / clear 三处 reject。`pnpm run check` 全绿（56 测试文件 / **510** 测试 + pack 闸门）。
+
+## v0.4.37
+
+「能力路由」四张卡片现在说明自己的层级：摘要行是**默认（无组时）**的值，另起一行列出所有覆盖它的供应商配置组，改完组不再看起来像"没生效"。
+
+- **纯展示改动，解析行为不变。** `resolveCapabilityRoute` 仍是「组 ?? 默认」（`src/session-route.ts`）：压缩 / 自动审查 / 子代理是单次静态解析、不做失败回落，会话命名仍走既有候选链（组对 → 默认对 → 行自身对）。
+- **新增 `capabilityGroupPins(groups, capability)`（`src/session-route.ts`）**：按 provider 排序收集"固定了该能力"的组；组内空串或半对一律视为未固定，因此 `titleProvider: ""`（跟随会话模型）不会被误当成固定值。
+- **卡片新增一行 `note`（`src/client/ModelSwitchSettings.tsx`）**：子代理、自动审查、压缩、会话命名四处；≤3 个组时全列（`组覆盖：apigoto → apigoto · gpt-6-luna；…`），超过 3 个折叠为「… 等 N 个组」；无组覆盖时**不渲染**，与旧版输出一致。子代理行额外带上组里的 `subagentReasoningEffort`。
+- **文案**：`src/client/locales.ts` 新增 `groupOverridesLabel` / `groupOverridesMore`（zh + en）。
+- **测试**：`tests/capability-group-pins.spec.ts` 四例（provider 排序 / 子代理 effort / 空串跳过 / 无组表）。`pnpm run check` 绿（57 文件 / **508** 测试 + pack 闸门）。
+- **Lab 验证（正式版 0.2.1-rc.2，`DSH_HOME=~/.dsh-lab`，3082，released artifact）**：注入一个 `deepseek-account` 组后打开「设置 → 模型切换」，子代理/自动审查/压缩三张卡片出现 `组覆盖：deepseek-account → deepseek-account · deepseek-flash；无组时用上方默认值`，会话命名因组内留空**不出现**；随后把组里自动审查与子代理两项清空，重载后组覆盖行从 3 条降为 1 条（只剩压缩）。改动在运行时重组后立即反映，无需重启。
+
 ## v0.4.36
 
 「自动审查」不再在插件安装类重组后停在「完全权限」：Auto 选择改为**持续记录**，回放只消费真正恢复掉的会话。
@@ -6,7 +27,7 @@
 - **Fix 1: a continuous record that survives reordering.** Every session's `permission/preset` feed is wired to the record as the session is seen; a change that arrives while this row has released the registration is the official teardown flip (not a user choice) and keeps its id recorded, anything else removes it. Release-time `persistAutoSnapshot()` now merges the live read **into** that record instead of replacing it, so an empty or failed read can no longer destroy the selection.
 - **Fix 2: a replay consumes only what it restored (`src/auto-sessions-state.ts`).** `planAutoSnapshotReplay()` splits the recorded ids into the ones a live session claims and the rest; the rest stay in the file with their **original** `savedAt` — an early claim cannot eat the selection of sessions that load later, and the 15-minute replay window is not restarted behind the user's back. `flushAutoSnapshot()` removes the file only once every recorded id has been replayed.
 - **Tests.** `tests/auto-sessions-state.spec.ts` adds the timestamp-preserving decode, the merge semantics that pin the out-of-order teardown case, the empty-list no-consume case, and a repeated-recompose pass that narrows the record to empty. `pnpm run check` green (56 test files / **504** tests + pack gate).
-- **Lab verification (official 0.2.1-rc.2, `DSH_HOME=~/.dsh-lab`, port 3082) — and one gap it exposed, now closed.** The build was installed in both lab profiles and driven against the released `0.4.35` as the A/B: with a session on「自动审查」the plugin was removed and re-added, a new plugin row was inserted through `cordis.patch.yml`, and a bundle-list change forced a recompose. The record half of the fix ran end to end there — a teardown wrote `auto-sessions.json` with the exact session id, the record survived the plugin being absent, and the next mount's replay consumed it. That replay is what the lab caught being incomplete: a claim runs before the web Host has lazily loaded the session, so ids could sit in the file forever with nobody to consume them. The replay now also runs whenever a session is observed, on tool-call activity, and from a bounded 10-second sweep while the record holds ids — the file observed still pending after a plugin swap was consumed by the next session load. The flip itself (`auto` → `danger-full-access`) does not reproduce in the lab — even a full plugin removal leaves the stored preset name untouched there — so that scenario stays covered by the desktop session-log forensics above and the out-of-order teardown tests.
+- **Lab verification (official 0.2.1-rc.2, `DSH_HOME=~/.dsh-lab`, port 3082) — and one gap it exposed, now closed.** The build was installed in both lab profiles and driven against the released `0.4.35` as the A/B: with a session on「自动审查」the plugin was removed and re-added, a new plugin row was inserted through `cordis.patch.yml`, and a bundle-list change forced a recompose. The record half of the fix ran end to end there — a teardown wrote `auto-sessions.json` with the exact session id, the record survived the plugin being absent, and the next mount's replay consumed it. That replay is what the lab caught being incomplete: a claim runs before the web Host has lazily loaded the session, so ids could sit in the file forever with nobody to consume them. The replay now also runs whenever a session is observed, on tool-call activity, and from a bounded 10-second sweep while the record holds ids — the file observed still pending after a plugin swap was consumed by the next session load. The flip itself (`auto` → `danger-full-access`) does not reproduce in the lab — even a full plugin removal leaves the stored preset name untouched there — so that scenario stays covered by the desktop session-log forensics above and the out-of-order teardown tests. Open-item audit of the deletion path: the empty-list delete has **no second loss face** — every caller (`trackAutoSelection`'s explicit un-select, a fully consumed replay, a teardown merge, and the stale/malformed cleanup in `flushAutoSnapshot`) deletes only a record that already decodes to empty. The one theoretical residual is on the *recording* side, not the deletion side: a session that switched to Auto while never observed (listener not yet wired) plus a teardown whose live read fails at the same moment; the observation passes (claim, replay, tool calls, 10-second sweep) bound that window to one sweep interval. Closing it entirely needs a session-load event on the public `sessions` face, which does not exist — recorded for a later upstream proposal, per the repo's public-seam rule.
 
 ## v0.4.35
 

package/lib/client.js

@@ -268,7 +268,7 @@ window.__ModuleLoader__.load({
 			});
 		}
 		/** A non-blank string. Blank values behave exactly like an absent field. */
-		function present(value) {
+		function present$1(value) {
 			return typeof value === "string" && value.trim() !== "";
 		}
 		/** Normalize one raw entry to its owning provider and route view; undefined for non-object junk. */
@@ -278,10 +278,10 @@ window.__ModuleLoader__.load({
 			const group = {};
 			for (const field of ROUTE_FIELDS) {
 				const item = source[field];
-				if (present(item)) group[field] = item;
+				if (present$1(item)) group[field] = item;
 			}
 			return {
-				provider: present(source.provider) ? source.provider : "",
+				provider: present$1(source.provider) ? source.provider : "",
 				group
 			};
 		}
@@ -349,6 +349,72 @@ window.__ModuleLoader__.load({
 			return Object.entries(groups).map(([key, value]) => entryOf(key, value));
 		}
 		//#endregion
+		//#region src/session-route.ts
+		/** Flat deployment-default fields per capability, in group-storage field names. */
+		const CAPABILITY_FIELDS = Object.freeze({
+			compaction: {
+				provider: "compactProvider",
+				model: "compactModel"
+			},
+			autoReview: {
+				provider: "autoReviewProvider",
+				model: "autoReviewModel"
+			},
+			title: {
+				provider: "titleProvider",
+				model: "titleModel"
+			},
+			subagent: {
+				provider: "subagentProvider",
+				model: "subagentModel"
+			}
+		});
+		/** A non-blank string; blank values behave exactly like an absent field. */
+		function present(value) {
+			return typeof value === "string" && value.trim() !== "";
+		}
+		/** Read one capability's pair from a flat route record; undefined unless both halves are present. */
+		function pairOf(source, fields) {
+			if (source === void 0) return void 0;
+			const record = source;
+			const provider = record[fields.provider];
+			const model = record[fields.model];
+			if (!present(provider) || !present(model)) return void 0;
+			return {
+				provider,
+				model
+			};
+		}
+		/**
+		* Every provider group that overrides one capability, provider-sorted.
+		*
+		* Reads the stored table only and never attributes a session, so a caller can
+		* describe the layering ("this is the default; these groups override it")
+		* without claiming which layer a given session receives.
+		* @param groups - the normalized provider → group map.
+		* @param capability - capability whose group pins are collected.
+		* @returns The complete pins, ordered by provider id; empty when no group pins it.
+		*/
+		function capabilityGroupPins(groups, capability) {
+			const pins = [];
+			if (groups === void 0) return pins;
+			for (const provider of Object.keys(groups).sort()) {
+				const group = groups[provider];
+				const route = pairOf(group, CAPABILITY_FIELDS[capability]);
+				if (route === void 0) continue;
+				const effort = capability === "subagent" ? group?.subagentReasoningEffort : void 0;
+				pins.push(present(effort) ? {
+					provider,
+					route,
+					reasoningEffort: effort
+				} : {
+					provider,
+					route
+				});
+			}
+			return pins;
+		}
+		//#endregion
 		//#region src/client/plugin-rpc.ts
 		function callPluginRpc(rpc, method, endpoint, payload, signal) {
 			return rpc.call("/api", method, {
@@ -657,6 +723,21 @@ window.__ModuleLoader__.load({
 		function compact(...values) {
 			return values.filter((value) => value !== void 0 && value !== "").join(" · ");
 		}
+		/**
+		* The layering line under a capability card.
+		*
+		* The card's own summary is the deployment-wide default ("默认（无组时）"); a
+		* provider group overrides it for that provider's sessions, so name the pins
+		* instead of letting the default look like the effective route.
+		* @param t - settings locale lookup.
+		* @param pins - the capability's provider-group pins.
+		* @returns The line, or undefined when no group pins this capability.
+		*/
+		function groupNote(t, pins) {
+			if (pins.length === 0) return void 0;
+			const listed = pins.slice(0, 3).map((pin) => `${pin.provider} → ${pin.route.provider} · ${pin.route.model}${pin.reasoningEffort === void 0 ? "" : ` · ${pin.reasoningEffort}`}`).join("；");
+			return pins.length > 3 ? t("groupOverridesMore").replace("{listed}", listed).replace("{count}", String(pins.length)) : t("groupOverridesLabel").replace("{pins}", listed);
+		}
 		/** A complete pinned pair; either half blank leaves the capability at its default. */
 		function draftPair(provider, model) {
 			return provider !== void 0 && provider.trim() !== "" && model !== void 0 && model.trim() !== "" ? {
@@ -811,7 +892,7 @@ window.__ModuleLoader__.load({
 			});
 			return null;
 		}
-		function RouteCard({ title, summary, hint, icon, open, onToggle, disabled = false, badge, badgeWarn = false, trailing, expandable = true, children }) {
+		function RouteCard({ title, summary, hint, note, icon, open, onToggle, disabled = false, badge, badgeWarn = false, trailing, expandable = true, children }) {
 			const header = /* @__PURE__ */ (0, react_jsx_runtime.jsxs)("button", {
 				type: "button",
 				className: ModelSwitchSettings_module_css_default.routeHeader,
@@ -837,6 +918,10 @@ window.__ModuleLoader__.load({
 								className: ModelSwitchSettings_module_css_default.routeSummary,
 								children: summary
 							}),
+							note === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
+								className: ModelSwitchSettings_module_css_default.hint,
+								children: note
+							}),
 							hint === void 0 ? null : /* @__PURE__ */ (0, react_jsx_runtime.jsx)("span", {
 								className: ModelSwitchSettings_module_css_default.hint,
 								children: hint
@@ -1068,6 +1153,12 @@ window.__ModuleLoader__.load({
 				};
 			}, [loadSearchCapabilities]);
 			const providerGroups = normalizeProviderGroups(switchSettings.value?.providerGroups);
+			const groupPins = {
+				autoReview: capabilityGroupPins(providerGroups, "autoReview"),
+				compaction: capabilityGroupPins(providerGroups, "compaction"),
+				title: capabilityGroupPins(providerGroups, "title"),
+				subagent: capabilityGroupPins(providerGroups, "subagent")
+			};
 			/**
 			* Seed an unsaved group from the live routes.
 			*
@@ -1556,6 +1647,7 @@ window.__ModuleLoader__.load({
 							title: props.t("subagent"),
 							summary: subagentSummary,
 							hint: props.t("subagentHelp"),
+							note: groupNote(props.t, groupPins.subagent),
 							icon: "subagent",
 							open: open === "subagent" && subagentOn,
 							onToggle: () => {
@@ -1785,6 +1877,7 @@ window.__ModuleLoader__.load({
 								title: props.t("autoReview"),
 								summary: autoReviewDraft === void 0 ? props.t("loading") : autoReviewOn ? routeName(groups, autoReviewDraft) : props.t("autoReviewFollow"),
 								hint: props.t("autoReviewHelp"),
+								note: groupNote(props.t, groupPins.autoReview),
 								icon: "autoReview",
 								open: open === "autoReview" && autoReviewOn,
 								onToggle: () => {
@@ -1853,6 +1946,7 @@ window.__ModuleLoader__.load({
 								title: props.t("compaction"),
 								summary: compactionDraft === void 0 ? props.t("loading") : compactionOn ? routeName(groups, compactionDraft) : props.t("compactionFollow"),
 								hint: props.t("compactionHelp"),
+								note: groupNote(props.t, groupPins.compaction),
 								icon: "compaction",
 								open: open === "compaction" && compactionOn,
 								onToggle: () => {
@@ -1917,6 +2011,7 @@ window.__ModuleLoader__.load({
 								title: props.t("sessionTitle"),
 								summary: titleDraft === void 0 ? props.t("loading") : titleOn ? routeName(groups, titleDraft) : props.t("titleFollow"),
 								hint: props.t("titleHelp"),
+								note: groupNote(props.t, groupPins.title),
 								icon: "sessionTitle",
 								open: open === "sessionTitle" && titleOn,
 								onToggle: () => {
@@ -2307,6 +2402,8 @@ window.__ModuleLoader__.load({
 			providerGroupsBodyHelp: "每个供应商保存一组压缩、自动审查、会话命名与子代理路由。会话自身的供应商与该组一致时，这四项能力按组实时解析，未固定项回落到上方卡片里的全局设置。切换某个会话的模型不会写入任何全局设置，因此不同供应商的多个会话并行时互不影响，会话中切换供应商后下一次调用也会立即跟随新供应商的配置组。",
 			providerGroupPick: "选择供应商以编辑其配置组。",
 			providerGroupDelete: "删除该组",
+			groupOverridesLabel: "组覆盖：{pins}；无组时用上方默认值",
+			groupOverridesMore: "组覆盖：{listed} 等 {count} 个组；无组时用上方默认值",
 			saveTimeout: "保存超时，请重试",
 			saveRejected: "保存被拒绝，请重试",
 			"reason.central-subagent-routing": "Alpha.4 没有全局子代理启动路由接口。",
@@ -2367,6 +2464,8 @@ window.__ModuleLoader__.load({
 			providerGroupsBodyHelp: "Each provider stores one route group for compaction, auto review, session naming, and subagents. A session running on that provider resolves all four capabilities from the group and falls back to the cards above when a row is off. Switching a session’s model writes no global setting, so parallel sessions on different providers never affect each other, and mid-session provider switches take effect on the very next call.",
 			providerGroupPick: "Pick a provider to edit its group.",
 			providerGroupDelete: "Delete group",
+			groupOverridesLabel: "Group overrides: {pins}; without a group the default above applies",
+			groupOverridesMore: "Group overrides: {listed} and {count} groups; without a group the default above applies",
 			saveTimeout: "The save timed out; try again",
 			saveRejected: "The save was rejected; try again",
 			"reason.central-subagent-routing": "Alpha.4 exposes no global Subagent start-routing seam.",
@@ -3884,15 +3983,68 @@ window.__ModuleLoader__.load({
 		}
 		//#endregion
 		//#region src/picker/plan-goal.ts
+		/**
+		* Strip the gateway's transport prefix from a Host message.
+		*
+		* The goal service throws `GoalError`, and the RPC gateway relays a thrown
+		* error as `code: "gateway/internal"` with the original text still in
+		* `message`. Showing `gateway/internal: goal "…" already exists…` to a human
+		* buries the only sentence that matters, so the prefix is removed here.
+		*/
+		function normalizeHostMessage(message) {
+			return message.replace(/^\s*gateway\/[a-z0-9-]+\s*:\s*/i, "").trim();
+		}
+		/**
+		* Whether a refused goal call means "this session already has a goal".
+		*
+		* Two shapes reach here. A Host that returns the domain code says
+		* `GOAL_ALREADY_EXISTS`; the gateway instead relays the thrown `GoalError` as
+		* `gateway/internal` and keeps only the text, so the message itself has to be
+		* recognised. The domain code is checked first and the text only decides the
+		* message shape the Host is known to produce — a refusal for any other reason
+		* must never clear the user's current goal.
+		*/
+		function isAlreadyExistsError(error) {
+			if (error.code === "GOAL_ALREADY_EXISTS") return true;
+			return /already exists with phase/i.test(error.message);
+		}
+		/** Codes that describe the transport, not the goal domain. */
+		function isTransportCode(code) {
+			return code === "GOAL_TRANSPORT" || code.startsWith("gateway/");
+		}
 		function failure(result, fallback) {
 			if (result.ok) return { message: fallback };
 			const { code, message } = result.error;
-			return {
+			const text = normalizeHostMessage(message);
+			const wording = text === "" ? fallback : text;
+			return isTransportCode(code) ? { message: wording } : {
 				code,
-				message: message.trim() === "" ? fallback : message
+				message: wording
 			};
 		}
 		/**
+		* Run one remote call, turning a rejected promise into a refused envelope.
+		*
+		* Generated remotes normally resolve `{ ok: false }`, but a transport failure
+		* (or an older Host that rejects instead of answering) must not escape as an
+		* unhandled rejection: the card would lose its error state and the plan would
+		* stay pending with no explanation.
+		*/
+		async function guarded(call, fallback) {
+			try {
+				return await call();
+			} catch (error) {
+				const text = normalizeHostMessage(error instanceof Error ? error.message : String(error));
+				return {
+					ok: false,
+					error: {
+						code: "GOAL_TRANSPORT",
+						message: text === "" ? fallback : text
+					}
+				};
+			}
+		}
+		/**
 		* Address the plan document exactly as the official opener does.
 		*
 		* A logged plan (`callId` present) resolves from `dsh-resource://plan/...`
@@ -3935,26 +4087,26 @@ window.__ModuleLoader__.load({
 				message: "goal objective must be a non-empty string"
 			};
 			const objective = `持续推进当前plan直到完成（plan 文档：${planDocumentAddress(sessionId, ref)}）`;
-			const created = await remote.create(sessionId, { objective });
+			const created = await guarded(() => remote.create(sessionId, { objective }), "the goal could not be created");
 			if (created.ok) return { kind: "created" };
-			if (created.error.code !== "GOAL_ALREADY_EXISTS") return {
+			if (!isAlreadyExistsError(created.error)) return {
 				kind: "failed",
 				...failure(created, "the goal could not be created")
 			};
-			const current = await remote.get(sessionId);
+			const current = await guarded(() => remote.get(sessionId), "the current goal could not be read for replacement");
 			if (!current.ok || current.value === void 0 || current.value === null) return {
 				kind: "failed",
 				...failure(current, "the current goal could not be read for replacement")
 			};
-			const cleared = await remote.clear(sessionId, {
+			const cleared = await guarded(() => remote.clear(sessionId, {
 				id: current.value.id,
 				revision: current.value.revision
-			});
+			}), "the current goal could not be cleared");
 			if (!cleared.ok) return {
 				kind: "failed",
 				...failure(cleared, "the current goal could not be cleared")
 			};
-			const retry = await remote.create(sessionId, { objective });
+			const retry = await guarded(() => remote.create(sessionId, { objective }), "the goal could not be created");
 			if (retry.ok) return { kind: "replaced" };
 			return {
 				kind: "failed",

package/lib/types/client/locales.d.ts

@@ -49,6 +49,8 @@ export declare const zh: {
     readonly providerGroupsBodyHelp: "每个供应商保存一组压缩、自动审查、会话命名与子代理路由。会话自身的供应商与该组一致时，这四项能力按组实时解析，未固定项回落到上方卡片里的全局设置。切换某个会话的模型不会写入任何全局设置，因此不同供应商的多个会话并行时互不影响，会话中切换供应商后下一次调用也会立即跟随新供应商的配置组。";
     readonly providerGroupPick: "选择供应商以编辑其配置组。";
     readonly providerGroupDelete: "删除该组";
+    readonly groupOverridesLabel: "组覆盖：{pins}；无组时用上方默认值";
+    readonly groupOverridesMore: "组覆盖：{listed} 等 {count} 个组；无组时用上方默认值";
     readonly saveTimeout: "保存超时，请重试";
     readonly saveRejected: "保存被拒绝，请重试";
     readonly 'reason.central-subagent-routing': "Alpha.4 没有全局子代理启动路由接口。";

package/lib/types/session-route.d.ts

@@ -77,6 +77,28 @@ export declare function subagentRouteOf(settings: RouteBearingSettings | undefin
  * @returns The complete group pair, or undefined when the group pins nothing here.
  */
 export declare function groupRoute(settings: RouteBearingSettings | undefined, provider: string | undefined, capability: CapabilityId): CapabilityRoute | undefined;
+/**
+ * One provider group's pin for one capability, as the settings page shows it.
+ */
+export interface CapabilityGroupPin {
+    /** The provider whose sessions this group applies to. */
+    readonly provider: string;
+    /** The pinned route. */
+    readonly route: CapabilityRoute;
+    /** A subagent spawn's effort; only meaningful for `subagent`. */
+    readonly reasoningEffort?: string;
+}
+/**
+ * Every provider group that overrides one capability, provider-sorted.
+ *
+ * Reads the stored table only and never attributes a session, so a caller can
+ * describe the layering ("this is the default; these groups override it")
+ * without claiming which layer a given session receives.
+ * @param groups - the normalized provider → group map.
+ * @param capability - capability whose group pins are collected.
+ * @returns The complete pins, ordered by provider id; empty when no group pins it.
+ */
+export declare function capabilityGroupPins(groups: ProviderGroups | undefined, capability: CapabilityId): readonly CapabilityGroupPin[];
 /**
  * The route one capability should use for one session.
  *

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-model-switch",
-  "version": "0.4.36",
+  "version": "0.4.38",
   "description": "Explicit model routing plus composer/Plan picker for DeepSeek Harness",
   "repository": {
     "type": "git",