dsh-connect-trae 2.0.1 → 2.0.3

package/CHANGELOG.md

@@ -1,5 +1,31 @@
 # Changelog
 
+## 2.0.3 (2026-09-16)
+
+### Fixes
+
+- **修复内置兜底目录里存在「不可调用模型」的问题**——插件在首次实时刷新落地前会先提供一份内置兜底模型列表，且这份列表**故意不经过「死 config_name」过滤**（过滤依赖实时目录，会把兜底本身误删）。因此它的 id 必须在源头就是正确的 `llm_utils_chat` config_name，而它此前不是：
+  - `auto` 不是任何 Trae 的 config_name，选中后会以 `4001` 失败——已移除。该条目会在「尚未刷新目录」时（未登录 / 网络抖动 / 卡片未刷新）被真正提供给用户，因此并非理论问题。
+  - `DeepSeek-V4-Flash` / `DeepSeek-V4-Pro` 缺少 `-Official` 后缀，不是真实 wire id——已更正为 `DeepSeek-V4-Flash-Official` / `DeepSeek-V4-Pro-Official`（依据 2026-09-15 实测目录，`docs/DS41_CALLABILITY.md`）。
+  - 新增回归测试锁定三条不变量：兜底 id 必须是真实 config_name（显式禁 `auto` 与裸名）、两个区域的兜底都不得出现 TraeCode 独占模型、`mergeTraeModelSources` 确实丢弃无 wire 匹配的目录行。
+
+### Docs
+
+- **README 新增「模型覆盖范围」一节（中英同步）**——说明插件提供的是 Trae SOLO 通道的模型，而 TraeCode 通道的 4 个模型（`deepseek-v4.1-flash`、`glm-5.3-flash`、`kimi-k2.8-preview`、`qwen3.8-flash`）尚未开放到 SOLO 通道，因此暂不支持，**需要等官方开放到 SOLO 后才能支持**；并区分了 `GLM-5.3`（支持）与 `glm-5.3-flash`（不支持）。
+- 新增 TraeCode / agent-task 通道与 TraeCLI 的可行性取证（`docs/CODEC_CHANNEL_FEASIBILITY.md`、`docs/TRAECLI_FEASIBILITY.md`、`docs/TRAE_ECOSYSTEM_RESEARCH.md`）：四条路径逐一实测均不通，且阻塞原因互相独立（协议 / 配额 / 同机 / 登录态与 agent 归属），据此确认这批模型属上游授权边界而非插件可绕过的技术问题。
+
+## 2.0.2 (2026-09-15)
+
+### Fixes
+
+- **修复「Trae 目录里有、插件列表里没有」的整类问题**（issue #7：GLM-5.3 缺失）——根因是**模型调用的 `function` 选错了**，不是目录读取有误：
+  - Trae 把可调用目录**分散在多个 SOLO 模式 function 下**，且一个模型**只能通过列出它的那个 function 调用**。插件此前固定用 `solo_work_lite` 取目录与发请求，而 `glm-5.3` 只存在于 **`solo_work_remote`**：用前者调用会被上游拒绝（`4001 param is invalid`），用后者则正常流式返回（2026-09-15 受控实测，同一份请求体，仅 function 不同）。
+  - **目录发现改为多 function 并集**：CN 依序请求 `solo_work_remote` → `solo_work_lite`，国际请求 `solo_agent` → `solo_work_remote` → `solo_work_lite`；同一个 config 由**先列出的 function 拥有**（顺序即优先级）。某个 function 失败不再影响其它 function 的目录。
+  - **聊天请求按模型回放其来源 function**：每个模型随目录记下 `wireFunction`（并纳入设置 schema 持久化，重启后仍生效），调用时写回请求体——`glm-5.3` 因而会走 `solo_work_remote`。
+  - 顺带确认：CN 的两个登录（Trae CN IDE 与 TRAE SOLO CN）在同一 function 下拿到**完全相同**的列表，因此该修复对两者一致生效；国际版覆盖最全的仍是 `solo_agent`。
+  - 新增 4 个回归测试（多 function 并集、来源 function 标记、优先级归属、显式 function 覆盖默认值），230/230 全绿。
+- **目录拉取失败时重试一次**——Trae 的目录端点存在间歇性 401/限流（实测同一凭证数秒前 200、随后 401）。启动时若恰好撞上，live 目录会拉取失败并**静默回退到上次保存的快照**，从而丢掉自那次保存以来的所有新模型（本版要放出的 glm-5.3 就在其中，实测复现）。现在合并骨架（remote 目录）失败会先重试一次（800ms 间隔）再降级。
+
 ## 2.0.1 (2026-09-15)
 
 ### Fixes

package/README.en.md

@@ -10,6 +10,7 @@
   <a href="README.md">中文</a> ·
   <a href="#install">Install</a> ·
   <a href="#how-it-works">How it works</a> ·
+  <a href="#model-coverage">Model coverage</a> ·
   <a href="CHANGELOG.md">Changelog</a> ·
   <a href="https://github.com/dingminhua/dsh-connect-trae/issues">Issues</a>
 </p>
@@ -56,6 +57,25 @@ Usage overview hits the read-only `https://api.trae.cn/trae/api/v2/pay/*` and `/
 
 > See `docs/IMPLEMENTATION_PLAN.md`, `docs/SOLO_ROUTE_DECISION.md`, `docs/USAGE_API_RESEARCH.md`.
 
+## Model coverage
+
+This plugin serves models from Trae's **SOLO channel** (`DeepSeek-V4-Flash-Official`, `DeepSeek-V4-Pro-Official`, `GLM-5.3`, `GLM-5.2`, `Kimi-K3`, `MiniMax-M3`, `Qwen3.8-Max`, `Doubao-Seed-*`, …), for both the domestic and international sides.
+
+The following four models currently come from the **TraeCode (Trae IDE) channel** and have **not** been opened up to the SOLO channel, so this plugin does **not** support them yet:
+
+| Model |
+| --- |
+| `deepseek-v4.1-flash` |
+| `glm-5.3-flash` |
+| `kimi-k2.8-preview` |
+| `qwen3.8-flash` |
+
+These four are reachable only inside the Trae IDE client; there is no callable channel for them on the plugin side. **They can only be supported once Trae officially opens them up to the SOLO channel.**
+
+> Note the distinction: **`GLM-5.3` is supported** (it goes through the SOLO channel), but **`glm-5.3-flash` is not** — they are different models.
+
+To use these four models today, use the **Trae IDE itself**.
+
 ## Install
 
 ```sh

package/README.md

@@ -10,6 +10,7 @@
   <a href="README.en.md">English</a> ·
   <a href="#安装">安装</a> ·
   <a href="#工作原理">工作原理</a> ·
+  <a href="#模型覆盖范围">模型覆盖范围</a> ·
   <a href="CHANGELOG.md">更新日志</a> ·
   <a href="https://github.com/dingminhua/dsh-connect-trae/issues">问题反馈</a>
 </p>
@@ -56,6 +57,25 @@ DSH PiAiAdapter（每个 provider 一套）
 
 > 详见 `docs/IMPLEMENTATION_PLAN.md`、`docs/SOLO_ROUTE_DECISION.md`、`docs/USAGE_API_RESEARCH.md`。
 
+## 模型覆盖范围
+
+插件提供的是 Trae **SOLO 通道**的模型（`DeepSeek-V4-Flash-Official`、`DeepSeek-V4-Pro-Official`、`GLM-5.3`、`GLM-5.2`、`Kimi-K3`、`MiniMax-M3`、`Qwen3.8-Max`、`Doubao-Seed-*` 等），国内版与国际版均走这条通道。
+
+以下 4 个模型目前来自 **TraeCode（Trae IDE）通道**，尚未开放到 SOLO 通道，因此本插件**暂不支持**：
+
+| 模型 |
+| --- |
+| `deepseek-v4.1-flash` |
+| `glm-5.3-flash` |
+| `kimi-k2.8-preview` |
+| `qwen3.8-flash` |
+
+这 4 个模型只在 Trae IDE 客户端内可用，插件侧没有可调用的通道。**需要等官方把它们开放到 SOLO 通道后，插件才能支持。**
+
+> 注意区分：**`GLM-5.3` 是支持的**（走 SOLO 通道），但 **`glm-5.3-flash` 不支持**——两者是不同的模型。
+
+如果你现在就要用这 4 个模型，请直接在 **Trae IDE 本体**中使用。
+
 ## 安装
 
 推荐使用 DSH 插件命令安装 npm 已发布版本：

package/lib/index.d.ts

@@ -109,8 +109,24 @@ interface TraeModelInfo {
   maxContextWindow?: number;
   /** The `config_name` `llm_utils_chat` accepts; absent means `id` is already the wire id. */
   wireConfigName?: string;
+  /** The directory function this model must be called through (see {@link TraeWireModel.function}). */
+  wireFunction?: string;
 }
-/** Bootstrap catalog: identity only where current Trae metadata has not been fetched yet. */
+/**
+ * Bootstrap catalog: identity only where current Trae metadata has not been
+ * fetched yet.
+ *
+ * Every id here must be a `config_name` that `llm_utils_chat` actually accepts,
+ * because this list is served verbatim before the first live refresh lands and
+ * it is NOT filtered against the wire map (the wire map can only confirm ids it
+ * happens to know, so filtering would delete the safety net exactly when it is
+ * needed — see `fallbackModels` in index.ts). The current CN roster was
+ * verified against the live remote directory on 2026-09-15
+ * (docs/DS41_CALLABILITY.md): `DeepSeek-V4-Flash-Official` and
+ * `DeepSeek-V4-Pro-Official` carry the `-Official` suffix, and there is no
+ * `auto` config_name — an `auto` row was previously served here and would have
+ * failed on selection.
+ */
 declare const FALLBACK_TRAE_MODELS: readonly TraeModelInfo[];
 /**
  * Bootstrap catalog for the international (ai) region, captured from the
@@ -147,6 +163,14 @@ interface TraeWireModel {
   contextWindow?: number;
   maxTokens?: number;
   reasoning?: TraeReasoningCapability;
+  /**
+   * The `get_detail_param` function this config_name came from. Trae splits its
+   * callable roster across several functions (SOLO modes), and a model is only
+   * callable through the function that actually lists it — glm-5.3 answers
+   * `solo_work_remote` but rejects `solo_work_lite` with 4001. The chat call
+   * therefore has to replay the directory's own function.
+   */
+  function?: string;
 }
 /**
  * Merge the two Trae model sources into one authoritative catalog.
@@ -164,8 +188,8 @@ interface TraeWireModel {
  *
  * Joining is two-tier, in priority order:
  *  1. `wire.id` (the `config_name`) equals the remote id — the model's display
- *     id is already its wire id (the common case: glm-5.2, DeepSeek-V4-Flash,
- *     kimi-k3, …).
+ *     id is already its wire id (the common case: glm-5.2,
+ *     DeepSeek-V4-Flash-Official, kimi-k3, …).
  *  2. `wire.name` (the `display_name`) equals the remote display name — for
  *     models whose display id differs from the wire id across Trae versions.
  * When the matched wire `config_name` differs from the remote id it is recorded
@@ -996,13 +1020,15 @@ declare function decodeTraeEvent(event: SseEvent): TraeStreamEvent;
 declare const TRAE_SOLO_FUNCTION = "solo_work_lite";
 declare const TRAE_SOLO_CHAT_PATH = "/api/agent/v3/llm_utils_chat";
 declare const TRAE_SOLO_MODELS_PATH = "/api/ide/v1/get_detail_param";
-declare function prepareSoloBody(source: string, defaultModel?: string): string;
+declare function prepareSoloBody(source: string, defaultModel?: string, functionName?: string): string;
 interface TraeSoloModel {
   id: string;
   name: string;
   contextWindow?: number;
   maxTokens?: number;
   reasoning?: TraeReasoningCapability;
+  /** The directory function that listed this config (replayed when calling it). */
+  function?: string;
 }
 interface TraeSoloClientOptions {
   credential(): Promise<TraeCredential>;
@@ -1015,15 +1041,39 @@ declare class TraeSoloUpstreamClient {
   private readonly options;
   private readonly fetchImpl;
   constructor(options: TraeSoloClientOptions);
+  /**
+   * Read the callable roster for this credential's region.
+   *
+   * Every function in {@link TRAE_DIRECTORY_FUNCTIONS} is asked, in order, and
+   * their answers are unioned: the first function to list a `config_name` owns
+   * it. Trae splits its roster across SOLO modes, and a model is only callable
+   * through the function that lists it (glm-5.3 exists solely under
+   * `solo_work_remote` on the CN gateway). Asking one function therefore
+   * silently hides models that the other one serves. The remote directory
+   * remains the merge skeleton, so agent-internal entries (search_agent_*,
+   * paygo variants) never surface even though they appear here.
+   */
   fetchModels(signal?: AbortSignal): Promise<TraeSoloModel[]>;
-  chatStream(bodyJson: string, signal?: AbortSignal): Promise<TraeChatResult>;
+  /** Merge one function's config list into the shared catalogue (first wins). */
+  private collectModels;
+  chatStream(bodyJson: string, signal?: AbortSignal, functionName?: string): Promise<TraeChatResult>;
 }
 //#endregion
 //#region src/solo-bridge.d.ts
 /** Convert Trae's named SSE events into OpenAI chat-completion SSE chunks. */
 declare function bridgeTraeSoloStream(response: Response, model: string): Response;
-/** Resolves a display model id to the real llm_utils_chat config_name. */
-type TraeWireResolver = (displayId: string) => string | undefined;
+/**
+ * One resolved wire target: the `config_name` `llm_utils_chat` accepts, plus
+ * the directory function that listed it. Trae's roster is split across several
+ * SOLO-mode functions and a model is only callable through the one listing it
+ * (glm-5.3 answers only under `solo_work_remote`), so the chat call replays it.
+ */
+interface TraeWireTarget {
+  configName: string;
+  function?: string;
+}
+/** Resolves a display model id to its wire target (config_name + function). */
+type TraeWireResolver = (displayId: string) => TraeWireTarget | undefined;
 /** Native SOLO client wrapper used by the loopback OpenAI adapter. */
 declare class TraeSoloBridge implements TraeUpstreamClient {
   private readonly upstream;

package/lib/index.js

@@ -15,18 +15,28 @@ import { Readable } from "node:stream";
 import { execFile } from "node:child_process";
 import { promisify } from "node:util";
 //#region src/catalog.ts
-/** Bootstrap catalog: identity only where current Trae metadata has not been fetched yet. */
+/**
+* Bootstrap catalog: identity only where current Trae metadata has not been
+* fetched yet.
+*
+* Every id here must be a `config_name` that `llm_utils_chat` actually accepts,
+* because this list is served verbatim before the first live refresh lands and
+* it is NOT filtered against the wire map (the wire map can only confirm ids it
+* happens to know, so filtering would delete the safety net exactly when it is
+* needed — see `fallbackModels` in index.ts). The current CN roster was
+* verified against the live remote directory on 2026-09-15
+* (docs/DS41_CALLABILITY.md): `DeepSeek-V4-Flash-Official` and
+* `DeepSeek-V4-Pro-Official` carry the `-Official` suffix, and there is no
+* `auto` config_name — an `auto` row was previously served here and would have
+* failed on selection.
+*/
 const FALLBACK_TRAE_MODELS = [
 	{
-		id: "auto",
-		name: "Auto"
-	},
-	{
-		id: "DeepSeek-V4-Flash",
+		id: "DeepSeek-V4-Flash-Official",
 		name: "DeepSeek-V4-Flash"
 	},
 	{
-		id: "DeepSeek-V4-Pro",
+		id: "DeepSeek-V4-Pro-Official",
 		name: "DeepSeek-V4-Pro"
 	},
 	{
@@ -126,8 +136,8 @@ function displayKey(name) {
 *
 * Joining is two-tier, in priority order:
 *  1. `wire.id` (the `config_name`) equals the remote id — the model's display
-*     id is already its wire id (the common case: glm-5.2, DeepSeek-V4-Flash,
-*     kimi-k3, …).
+*     id is already its wire id (the common case: glm-5.2,
+*     DeepSeek-V4-Flash-Official, kimi-k3, …).
 *  2. `wire.name` (the `display_name`) equals the remote display name — for
 *     models whose display id differs from the wire id across Trae versions.
 * When the matched wire `config_name` differs from the remote id it is recorded
@@ -156,7 +166,8 @@ function mergeTraeModelSources(remote, wire) {
 				reasoning: model.reasoning,
 				reasoningEfforts: Object.fromEntries(model.reasoning.supported.map((effort) => [effort, effort === "low" ? "light" : effort === "xhigh" ? "extra_high" : "high"]))
 			},
-			...wireModel.id !== "" && wireModel.id !== model.id ? { wireConfigName: wireModel.id } : {}
+			...wireModel.id !== "" && wireModel.id !== model.id ? { wireConfigName: wireModel.id } : {},
+			...wireModel.function === void 0 ? {} : { wireFunction: wireModel.function }
 		});
 	}
 	return result;
@@ -1818,6 +1829,25 @@ function applyReasoningEffort(body, effort, capability) {
 //#endregion
 //#region src/solo.ts
 const TRAE_SOLO_FUNCTION = "solo_work_lite";
+/**
+* Directory functions to union per region, in priority order.
+*
+* Trae spreads its callable roster across several SOLO-mode functions, and a
+* model is only usable through the one that lists it: `glm-5.3` is absent from
+* `solo_work_lite` but present in `solo_work_remote` (verified 2026-09-15 —
+* calling it through the former answers `4001 param is invalid`, through the
+* latter streams normally). Rather than betting on a single function, the
+* directory unions them; the first function to provide a config wins, so the
+* order below decides which wire name a model is called with.
+*/
+const TRAE_DIRECTORY_FUNCTIONS = {
+	cn: ["solo_work_remote", TRAE_SOLO_FUNCTION],
+	ai: [
+		"solo_agent",
+		"solo_work_remote",
+		TRAE_SOLO_FUNCTION
+	]
+};
 const TRAE_SOLO_CHAT_PATH = "/api/agent/v3/llm_utils_chat";
 const TRAE_SOLO_MODELS_PATH = "/api/ide/v1/get_detail_param";
 function classify(status) {
@@ -1831,14 +1861,14 @@ function classify(status) {
 function finitePositive$1(value) {
 	return typeof value === "number" && Number.isFinite(value) && value > 0 ? value : void 0;
 }
-function prepareSoloBody(source, defaultModel = "glm-5.2") {
+function prepareSoloBody(source, defaultModel = "glm-5.2", functionName) {
 	const input = JSON.parse(source);
 	const model = typeof input["model"] === "string" && input["model"].trim() !== "" ? input["model"].trim() : defaultModel;
 	const body = {
 		...Array.isArray(input["messages"]) ? { messages: input["messages"] } : {},
 		model,
 		config_name: model,
-		function: TRAE_SOLO_FUNCTION,
+		function: typeof input["function"] === "string" && input["function"] !== "" ? input["function"] : functionName ?? "solo_work_lite",
 		stream: true,
 		...Array.isArray(input["tools"]) ? { tools: input["tools"] } : {},
 		...typeof input["reasoning_effort"] === "string" ? { reasoning_effort: input["reasoning_effort"] } : {}
@@ -1880,32 +1910,60 @@ var TraeSoloUpstreamClient = class {
 		this.options = options;
 		this.fetchImpl = options.fetchImpl ?? fetch;
 	}
+	/**
+	* Read the callable roster for this credential's region.
+	*
+	* Every function in {@link TRAE_DIRECTORY_FUNCTIONS} is asked, in order, and
+	* their answers are unioned: the first function to list a `config_name` owns
+	* it. Trae splits its roster across SOLO modes, and a model is only callable
+	* through the function that lists it (glm-5.3 exists solely under
+	* `solo_work_remote` on the CN gateway). Asking one function therefore
+	* silently hides models that the other one serves. The remote directory
+	* remains the merge skeleton, so agent-internal entries (search_agent_*,
+	* paygo variants) never surface even though they appear here.
+	*/
 	async fetchModels(signal) {
 		const [credential, identity] = await Promise.all([this.options.credential(), this.options.identity()]);
 		const region = regionOfCredential(credential);
 		const base = this.options.baseUrl ?? REGION_GATEWAYS[region].chat;
-		const directoryFunction = region === "ai" ? "solo_agent" : TRAE_SOLO_FUNCTION;
-		const response = await this.fetchImpl(traeEndpoint(base, TRAE_SOLO_MODELS_PATH), {
-			method: "POST",
-			headers: {
-				...buildTraeCnHeaders(credential, identity),
-				Accept: "application/json"
-			},
-			body: JSON.stringify({
-				function: directoryFunction,
-				config_names: null,
-				need_prompt: false,
-				current_config_info: null,
-				poly_prompt: true,
-				mode_type: null,
-				agent_type: null
-			}),
-			signal: signal ?? AbortSignal.timeout(3e4)
-		});
-		if (!response.ok) throw new Error(`Trae SOLO models returned HTTP ${response.status}`);
-		const document = await response.json();
-		const list = Array.isArray(document["config_info_list"]) ? document["config_info_list"] : [];
-		const models = [];
+		const headers = {
+			...buildTraeCnHeaders(credential, identity),
+			Accept: "application/json"
+		};
+		const byId = /* @__PURE__ */ new Map();
+		const failures = [];
+		for (const directoryFunction of TRAE_DIRECTORY_FUNCTIONS[region]) {
+			let list;
+			try {
+				const response = await this.fetchImpl(traeEndpoint(base, TRAE_SOLO_MODELS_PATH), {
+					method: "POST",
+					headers,
+					body: JSON.stringify({
+						function: directoryFunction,
+						config_names: null,
+						need_prompt: false,
+						current_config_info: null,
+						poly_prompt: true,
+						mode_type: null,
+						agent_type: null
+					}),
+					signal: signal ?? AbortSignal.timeout(3e4)
+				});
+				if (!response.ok) throw new Error(`HTTP ${response.status}`);
+				const document = await response.json();
+				list = Array.isArray(document["config_info_list"]) ? document["config_info_list"] : [];
+			} catch (error) {
+				failures.push(`${directoryFunction}: ${String(error).slice(0, 80)}`);
+				continue;
+			}
+			this.collectModels(list, directoryFunction, byId);
+		}
+		const models = [...byId.values()];
+		if (models.length === 0) throw new Error(`Trae SOLO models response contained no models (${failures.join("; ") || "empty directory"})`);
+		return models;
+	}
+	/** Merge one function's config list into the shared catalogue (first wins). */
+	collectModels(list, directoryFunction, byId) {
 		for (const raw of list) {
 			if (typeof raw !== "object" || raw === null) continue;
 			const config = raw;
@@ -1923,21 +1981,21 @@ var TraeSoloUpstreamClient = class {
 				...config,
 				...detail
 			});
-			models.push({
+			if (byId.has(id)) continue;
+			byId.set(id, {
 				id,
 				name: typeof display["display_name"] === "string" && display["display_name"] !== "" ? display["display_name"] : id,
 				...contextWindow === void 0 ? {} : { contextWindow },
 				...maxTokens === void 0 ? {} : { maxTokens },
-				...reasoning === void 0 ? {} : { reasoning }
+				...reasoning === void 0 ? {} : { reasoning },
+				function: directoryFunction
 			});
 		}
-		if (models.length === 0) throw new Error("Trae SOLO models response contained no models");
-		return models;
 	}
-	async chatStream(bodyJson, signal) {
+	async chatStream(bodyJson, signal, functionName) {
 		let prepared;
 		try {
-			prepared = prepareSoloBody(bodyJson);
+			prepared = prepareSoloBody(bodyJson, void 0, functionName);
 		} catch {
 			return {
 				ok: false,
@@ -2230,11 +2288,17 @@ var TraeSoloBridge = class {
 			const input = JSON.parse(bodyJson);
 			if (typeof input["model"] === "string" && input["model"] !== "") model = input["model"];
 			const entry = this.catalog?.current().find((item) => item.id === model);
-			const wireModel = entry?.wireConfigName ?? this.wireResolver?.(model) ?? this.wireResolver?.(entry?.name ?? "") ?? model;
-			if (wireModel !== input["model"]) {
-				input["model"] = wireModel;
-				prepared = JSON.stringify(input);
-			}
+			const fromCatalog = entry?.wireConfigName === void 0 ? void 0 : {
+				configName: entry.wireConfigName,
+				...entry.wireFunction === void 0 ? {} : { function: entry.wireFunction }
+			};
+			const fromResolver = this.wireResolver?.(model) ?? this.wireResolver?.(entry?.name ?? "");
+			const target = fromCatalog ?? fromResolver;
+			const wireModel = target?.configName ?? model;
+			const wireFunction = target?.function ?? entry?.wireFunction;
+			if (wireModel !== input["model"]) input["model"] = wireModel;
+			if (wireFunction !== void 0 && input["function"] !== wireFunction) input["function"] = wireFunction;
+			if (wireModel !== JSON.parse(bodyJson)["model"] || wireFunction !== void 0) prepared = JSON.stringify(input);
 			if (typeof input["reasoning_effort"] === "string") {
 				const efforts = entry?.reasoningEfforts;
 				const requested = input["reasoning_effort"];
@@ -3467,7 +3531,8 @@ const modelConfig = z.object({
 	contextWindow: z.number().step(1).min(1),
 	maxTokens: z.number().step(1).min(1),
 	input: z.array(z.union(["text", "image"])),
-	creditMultiplier: z.number()
+	creditMultiplier: z.number(),
+	wireFunction: z.string()
 });
 const regionStateConfig = z.object({
 	lastCatalog: z.array(modelConfig).default([]),
@@ -3553,6 +3618,25 @@ function apply(ctx, config) {
 		if (explicit !== void 0) return explicit;
 		return legacyAccountRegion === region ? value.accountId : void 0;
 	};
+	/**
+	* Retry a directory read once. The gateways answer an intermittent 401 for a
+	* credential that works moments later (observed 2026-09-15), and a failed
+	* skeleton read would otherwise drop the runtime catalog back to the saved
+	* snapshot — hiding every model added since that save (issue #7's glm-5.3
+	* among them).
+	*/
+	async function withDirectoryRetry(read) {
+		try {
+			return await read();
+		} catch (first) {
+			await new Promise((resolve) => setTimeout(resolve, 800));
+			try {
+				return await read();
+			} catch {
+				throw first;
+			}
+		}
+	}
 	const stacks = {};
 	for (const region of REGION_KEYS) {
 		const catalog = new TraeCatalog(region);
@@ -3654,8 +3738,7 @@ function apply(ctx, config) {
 			delegating,
 			usageClient: new TraeUsageClient({ credential: () => store.resolve() }),
 			discoverModels: async (signal) => {
-				const [remote, wireModels] = await Promise.all([remoteCatalog.fetchModels(signal), solo.fetchModels(signal)]);
-				const merged = mergeTraeModelSources(remote, wireModels);
+				const merged = mergeTraeModelSources(await withDirectoryRetry(() => remoteCatalog.fetchModels(signal)), await solo.fetchModels(signal));
 				wire.callableKeys.clear();
 				for (const model of merged) {
 					wire.callableKeys.add(model.id.trim().toLowerCase());
@@ -3665,8 +3748,12 @@ function apply(ctx, config) {
 				wire.byId.clear();
 				wire.byName.clear();
 				for (const model of merged) if (model.wireConfigName !== void 0) {
-					wire.byId.set(model.id, model.wireConfigName);
-					wire.byName.set(model.name.trim().toLowerCase(), model.wireConfigName);
+					const target = {
+						configName: model.wireConfigName,
+						...model.wireFunction === void 0 ? {} : { function: model.wireFunction }
+					};
+					wire.byId.set(model.id, target);
+					wire.byName.set(model.name.trim().toLowerCase(), target);
 				}
 				return merged;
 			},

package/package.json

@@ -2,7 +2,7 @@
   "name": "dsh-connect-trae",
   "displayName": "DSH Connect Trae",
   "description": "Connect locally signed-in Trae models to DeepSeek Harness, plus a read-only usage/credits summary.",
-  "version": "2.0.1",
+  "version": "2.0.3",
   "author": "dingminhua",
   "copyright": "Copyright (c) 2026 LaoDing",
   "license": "MIT",