dsh-llm-workbuddy 0.1.14 → 0.1.15

package/CHANGELOG.md

@@ -4,6 +4,35 @@
 每次发版请同步 `package.json` 的 `version` 并打一个 `git tag`（如 `v0.1.13`），
 在 GitHub 创建 Release 时本文件即为更新说明来源。
 
+## [0.1.15] - 2026-09-10
+
+### Fixed
+- **刷新模型看不到新上架模型**：内置目录此前同时充当白名单，代理
+  `/v1/models` 里凡不在目录中的模型一律被丢弃，所以平台新上架的模型
+  （如 `deepseek-v4.1-flash`）刷新多少次都不会出现。现改为**黑名单**
+  `RETIRED_MODEL_IDS`：只隐藏实测不可用（上游 `service info not found`）的
+  旧模型 id，其余照单显示。顺带放出 7 个别名仍在服务、却被旧白名单误藏的模型
+  （`minimax-m2.7`、`glm-5.0-turbo`、`hy3-preview`、`deepseek-v3-1-lkeap`、
+  `deepseek-v3-0324-lkeap`、`deepseek-r1-0528-lkeap`、`hunyuan-2.0-instruct`）。
+- 内置目录新增 `deepseek-v4.1-flash`，并在代理不可达时作为兜底条目。
+- 🔍 诊断的「模型出字」误报修复：探针用 `hy3` + `max_tokens: 4` 发一次请求，
+  而推理模型会把这 4 个 token 全花在思考上、`content` 返回空串，于是
+  "代理正常"被误判成「代理返回了空响应」。现承认"有计费 completion token +
+  有 choice"即为正常。
+
+### Added
+- **模型名后显示积分消耗倍率**：读取代理 `/v1/models` 的 `credits` 字段
+  （`"x0.06 credits"`）并拼成 `Deepseek-V4-Flash ×0.06`；实时值优先，内置目录
+  `credits` 兜底。平台未声明倍率的模型只显示模型名——不会用 0 或猜测值冒充。
+- 内置目录补齐各模型 `credits`（取自 WorkBuddy 应用 `product.json`）。
+
+### Changed（代理侧，`.workbuddy-src` / workbuddy2api）
+- `/v1/models` 透出 `credits` 字段（`model_to_codex_format`），倍率不再由各客户端
+  各自维护。
+- 官方应用 `product.json` 与本地 `models_config.json` 合并时倍率单独回填：官方
+  条目为空则沿用本地倍率，避免官方合并把倍率抹平。
+- `models_config.json` 补充 `deepseek-v4.1-flash`（1M 上下文 / 50k 输出）。
+
 ## [0.1.14] - 2026-08-28
 
 ### Changed
@@ -50,6 +79,7 @@
 ---
 
 <!-- 历史版本锚点（便于生成 Release 时对比区间） -->
+[0.1.15]: https://github.com/zdk119746/dsh-llm-workbuddy/compare/v0.1.14...v0.1.15
 [0.1.14]: https://github.com/zdk119746/dsh-llm-workbuddy/compare/v0.1.13...v0.1.14
 [0.1.13]: https://github.com/zdk119746/dsh-llm-workbuddy/compare/v0.1.12...v0.1.13
 [0.1.12]: https://github.com/zdk119746/dsh-llm-workbuddy/compare/v0.1.11...v0.1.12

package/README.md

@@ -284,16 +284,29 @@ DSH 的 `dsh.client` 机制只要求 `package.json` 里：
     # discovery: true                   # 实时拉取代理的 /v1/models（30s 缓存）
     # models: [...]                     # 静态目录（代理不可达时的兜底）
     #   每个条目可带 reasoningEffort: 'low'|'medium'|'high'（默认推理等级）
+    #   以及 credits: 'x0.06 credits'（显示为模型名后的 ×0.06）
     # loginScript: ''                   # 登录脚本绝对/相对路径；默认用包内 login_workbuddy.py
     # sessionFile: ~/.codebuddy-session.json  # 会话文件路径；默认同上
 ```
 
 模型列表默认取插件内置目录；代理可达时改为实时拉取 `/v1/models`（支持
 `{"models": [...]}` / `{"data": [...]}` 两种返回），未列出的模型 id 仍可原样
-传递。**内置目录充当白名单**：代理返回的已下架/不可用模型（如 glm-4.6v、
-glm-5.0、minimax-m2.5 等，上游返回 `service info not found`）会被过滤，不显示
-在 UI 中。每个模型的推理等级（`reasoningEffort`）默认来自内置目录，代理可达时
-优先沿用目录值（代理 `/v1/models` 当前只上报 `High`，不区分模型）。
+传递。**内置目录不再是白名单**：平台新上架的模型刷新后即出现在选择器里，
+无需等插件发版；只有已下架/不可用的模型 id（`RETIRED_MODEL_IDS`，上游返回
+`service info not found`）会被隐藏。每个模型的推理等级（`reasoningEffort`）
+默认来自内置目录，代理可达时优先沿用目录值（代理 `/v1/models` 当前只上报
+`High`，不区分模型）。
+
+**积分倍率**：模型名后面会拼上平台声明的积分倍率，例如
+`Deepseek-V4-Flash ×0.06`。倍率来自代理 `/v1/models` 的 `credits` 字段
+（`"x0.06 credits"`，平台写在 WorkBuddy 应用 `product.json` /
+`models_config.json` 里），实时值优先、内置目录兜底；平台没声明的模型
+（如个别新模型尚未写入 `product.json`）只显示模型名，不会猜一个数字。
+
+**刷新模型**：⚙️ 设置 → 🔄 刷新模型 会清掉插件的 30s 发现缓存并重新拉取
+代理列表。注意代理自身的模型清单来自本机 WorkBuddy 应用的 `product.json`
+与 `models_config.json`——两者都没有的模型，刷新也不会出现，需先更新应用或
+把模型补进代理配置。
 
 ---
 

package/lib/index.js

@@ -77,26 +77,53 @@ const DISCOVERY_TIMEOUT_MS = 2_000;
  * CodeBuddy/WorkBuddy platform (from workbuddy2api's models_config.json).
  * It is advisory and replaced by the live `/v1/models` answer whenever the
  * proxy is reachable; entries not announced by the proxy are still listed.
+ *
+ * `credits` is the platform's declared multiplier (`"x0.06 credits"`), shown
+ * after the model name. The live answer wins when it carries one — this copy
+ * only covers the proxy-down / discovery-off fallback.
  */
 const DEFAULT_MODELS = [
-  { id: "deepseek-v4-pro", name: "Deepseek-V4-Pro", contextWindow: 1_000_000, maxTokens: 50_000, inputModalities: ["text", "image"], reasoningEffort: "high" },
-  { id: "deepseek-v4-flash", name: "Deepseek-V4-Flash", contextWindow: 1_000_000, maxTokens: 50_000, inputModalities: ["text", "image"], reasoningEffort: "high" },
-  { id: "deepseek-v3-2-volc", name: "DeepSeek-V3.2", contextWindow: 96_000, maxTokens: 32_000, reasoningEffort: "medium" },
-  { id: "glm-5.2", name: "GLM-5.2", contextWindow: 1_000_000, maxTokens: 48_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "glm-5.1", name: "GLM-5.1", contextWindow: 200_000, maxTokens: 48_000, reasoningEffort: "medium" },
-  { id: "glm-5v-turbo", name: "GLM-5v-Turbo", contextWindow: 200_000, maxTokens: 64_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "kimi-k3-1", name: "Kimi-K3", contextWindow: 1_000_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "kimi-k2.7", name: "Kimi-K2.7-Code", contextWindow: 256_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "kimi-k2.6", name: "Kimi-K2.6", contextWindow: 256_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "kimi-k2.5", name: "Kimi-K2.5", contextWindow: 164_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "minimax-m3", name: "MiniMax-M3", contextWindow: 512_000, maxTokens: 128_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "hy3", name: "Hy3", contextWindow: 192_000, maxTokens: 64_000, inputModalities: ["text", "image"], reasoningEffort: "high" },
-  { id: "hunyuan-2.0-thinking", name: "Hunyuan-2.0-Thinking", contextWindow: 128_000, maxTokens: 24_000, reasoningEffort: "medium" },
+  { id: "deepseek-v4.1-flash", name: "Deepseek-V4.1-Flash", contextWindow: 1_000_000, maxTokens: 50_000, inputModalities: ["text", "image"], reasoningEffort: "high" },
+  { id: "deepseek-v4-pro", name: "Deepseek-V4-Pro", contextWindow: 1_000_000, maxTokens: 50_000, inputModalities: ["text", "image"], reasoningEffort: "high", credits: "x0.16 credits" },
+  { id: "deepseek-v4-flash", name: "Deepseek-V4-Flash", contextWindow: 1_000_000, maxTokens: 50_000, inputModalities: ["text", "image"], reasoningEffort: "high", credits: "x0.06 credits" },
+  { id: "deepseek-v3-2-volc", name: "DeepSeek-V3.2", contextWindow: 96_000, maxTokens: 32_000, reasoningEffort: "medium", credits: "x0.29 credits" },
+  { id: "glm-5.2", name: "GLM-5.2", contextWindow: 1_000_000, maxTokens: 48_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.79 credits" },
+  { id: "glm-5.1", name: "GLM-5.1", contextWindow: 200_000, maxTokens: 48_000, reasoningEffort: "medium", credits: "x0.79 credits" },
+  { id: "glm-5v-turbo", name: "GLM-5v-Turbo", contextWindow: 200_000, maxTokens: 64_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.95 credits" },
+  { id: "kimi-k3-1", name: "Kimi-K3", contextWindow: 1_000_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x1.62 credits" },
+  { id: "kimi-k2.7", name: "Kimi-K2.7-Code", contextWindow: 256_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.57 credits" },
+  { id: "kimi-k2.6", name: "Kimi-K2.6", contextWindow: 256_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.52 credits" },
+  { id: "kimi-k2.5", name: "Kimi-K2.5", contextWindow: 164_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.45 credits" },
+  { id: "minimax-m3", name: "MiniMax-M3", contextWindow: 512_000, maxTokens: 128_000, inputModalities: ["text", "image"], reasoningEffort: "medium", credits: "x0.25 credits" },
+  { id: "hy3", name: "Hy3", contextWindow: 192_000, maxTokens: 64_000, inputModalities: ["text", "image"], reasoningEffort: "high", credits: "x0.00 credits" },
+  { id: "hunyuan-2.0-thinking", name: "Hunyuan-2.0-Thinking", contextWindow: 128_000, maxTokens: 24_000, reasoningEffort: "medium", credits: "x0.04 credits" },
   { id: "hunyuan-chat", name: "Hunyuan-Turbos", contextWindow: 200_000, maxTokens: 8_192 },
   { id: "auto", name: "Auto", contextWindow: 168_000, maxTokens: 32_000, inputModalities: ["text", "image"], reasoningEffort: "high" },
-  { id: "default", name: "Default", contextWindow: 200_000, maxTokens: 24_000 },
+  { id: "default", name: "Default", contextWindow: 200_000, maxTokens: 24_000, credits: "x2.00 credits" },
 ];
 
+/**
+ * Models the proxy still announces but that the upstream platform rejects with
+ * `service info not found` (each id re-probed 2026-09-10 against the live
+ * endpoint). The catalog is no longer a whitelist — a model the platform
+ * launches shows up after a refresh without a plugin update — so dead ids are
+ * hidden here by name instead.
+ */
+const RETIRED_MODEL_IDS = new Set([
+  "minimax-m2.5",
+  "glm-5.0",
+  "glm-4.7",
+  "glm-4.6",
+  "glm-4.6v",
+  "kimi-k2-thinking",
+  "deepseek-v3-1",
+  "deepseek-v3-1-volc",
+  "kimi-k2-instruct-taiji",
+  "completion-gf",
+  "default-1.1",
+  "default-1.2",
+]);
+
 /** Selectable reasoning efforts exposed to the harness UI, in display order. */
 const REASONING_EFFORTS = [
   { id: "low", name: "Low" },
@@ -156,6 +183,26 @@ function liveInputModalities(entry) {
   return valid.length > 0 ? [...new Set(valid)] : ["text"];
 }
 
+/**
+ * Parse the platform's credit declaration into the `×0.06` suffix appended to
+ * a model name. The platform writes these as `"x0.06 credits"` (product.json /
+ * models_config.json), and the proxy relays the string verbatim; a bare number
+ * or an already-compact `×1.62` also parses. Anything without a recognizable
+ * number yields `undefined`, so callers omit the suffix rather than guess —
+ * an unknown multiplier must not be rendered as 0.
+ */
+function creditsMultiplier(value) {
+  if (typeof value !== "string") return undefined;
+  const match = /^[x×*]?\s*(\d+(?:\.\d+)?)\s*(?:credits?)?$/i.exec(value.trim());
+  return match === null ? undefined : `×${match[1]}`;
+}
+
+/** Append the credit multiplier to a model's display name, when declared. */
+function withCredits(name, credits) {
+  const multiplier = creditsMultiplier(credits);
+  return multiplier === undefined ? name : `${name} ${multiplier}`;
+}
+
 
 // #region serialize
 
@@ -580,7 +627,7 @@ function modelInfo(provider, model) {
   return {
     provider,
     id: model.id,
-    name: model.name ?? model.id,
+    name: withCredits(model.name ?? model.id, model.credits),
     ...(model.description === undefined ? {} : { description: model.description }),
     inputModalities: model.inputModalities ?? ["text"],
     ...(reasoning === undefined ? {} : { reasoning }),
@@ -673,6 +720,10 @@ export class WorkBuddyAdapter extends LlmAdapter {
           description: typeof entry.description === "string" && entry.description.length > 0
             ? entry.description
             : undefined,
+          // Platform-declared credit multiplier, relayed by the proxy.
+          credits: typeof entry.credits === "string" && entry.credits.trim().length > 0
+            ? entry.credits.trim()
+            : undefined,
           modalities: liveInputModalities(entry),
           contextWindow: Number.isInteger(entry.context_window) && entry.context_window > 0
             ? entry.context_window
@@ -689,19 +740,23 @@ export class WorkBuddyAdapter extends LlmAdapter {
         if (seen.has(entry.id)) continue;
         seen.add(entry.id);
         const catalog = byId.get(entry.id);
-        // The static catalog doubles as a whitelist. The proxy still announces
-        // retired/legacy models (deepseek-v3-1, glm-4.6, glm-4.6v, kimi-k2,
-        // kimi-k2-thinking, minimax-m2.5, hunyuan-image-v3.0, ...) that the
-        // upstream rejects with `service info not found`. Drop any live entry
-        // that is not in the catalog so those never reach the UI; the catalog
-        // is the curated set that matches what the official client shows.
-        if (catalog === undefined) continue;
+        // Retired ids are hidden by name (RETIRED_MODEL_IDS) rather than by
+        // absence from the catalog: the proxy still announces models upstream
+        // rejects, but a model the platform launches must appear after a
+        // refresh without waiting for a new plugin release. The catalog is now
+        // only a metadata overlay plus a fallback for unannounced entries.
+        if (RETIRED_MODEL_IDS.has(entry.id)) continue;
         const effort = liveReasoningEffort(entry, catalog?.reasoningEffort);
         const reasoning = modelReasoningInfo(effort);
         merged.push({
           provider,
           id: entry.id,
-          name: entry.name ?? catalog?.name ?? entry.id,
+          // The proxy's multiplier tracks the platform; the catalog's is a
+          // snapshot, so the live value wins and the catalog only fills gaps.
+          name: withCredits(
+            entry.name ?? catalog?.name ?? entry.id,
+            entry.credits ?? catalog?.credits,
+          ),
           ...(catalog?.description !== undefined || entry.description !== undefined
             ? { description: catalog?.description ?? entry.description }
             : {}),
@@ -843,6 +898,9 @@ const catalogModel = z.object({
   maxTokens: z.number().step(1).min(1),
   inputModalities: z.array(z.union(MODEL_MODALITIES)).min(1).default(["text"]),
   reasoningEffort: z.string(),
+  // Platform-declared credit multiplier, e.g. "x0.06 credits"; displayed after
+  // the model name as "×0.06".
+  credits: z.string(),
 });
 
 export const Config = z.object({
@@ -890,6 +948,10 @@ function resolveModels(models) {
     if (reasoningEffort !== undefined && !REASONING_EFFORT_IDS.has(reasoningEffort)) {
       throw new Error(`dsh-llm-workbuddy: catalog model "${model.id}" reasoningEffort must be one of low|medium|high`);
     }
+    const credits = typeof model.credits === "string" ? model.credits.trim() : undefined;
+    if (credits !== undefined && credits.length > 0 && creditsMultiplier(credits) === undefined) {
+      throw new Error(`dsh-llm-workbuddy: catalog model "${model.id}" credits must read like "x0.06 credits"`);
+    }
     return {
       id: model.id,
       ...(model.name === undefined ? {} : { name: model.name }),
@@ -898,6 +960,7 @@ function resolveModels(models) {
       ...(model.maxTokens === undefined ? {} : { maxTokens: model.maxTokens }),
       inputModalities: [...inputModalities],
       ...(reasoningEffort === undefined ? {} : { reasoningEffort }),
+      ...(credits === undefined || credits.length === 0 ? {} : { credits }),
     };
   });
 }
@@ -1234,10 +1297,19 @@ async function probeChatWorking(baseURL) {
       return { chatWorking: false, chatError: `HTTP ${response.status}${text ? `: ${text.slice(0, 200)}` : ""}` };
     }
     const body = await response.json().catch(() => null);
-    const content = body?.choices?.[0]?.message?.content;
+    const choice = body?.choices?.[0];
+    const content = choice?.message?.content;
     if (typeof content === "string" && content.length > 0) {
       return { chatWorking: true };
     }
+    // A reasoning model spends the whole 4-token cap on thinking and answers
+    // with `content: ""` + `finish_reason: "length"` — that is a healthy proxy,
+    // not an empty response (probed against hy3, which is all-thinking at this
+    // cap). Any billed completion token proves the upstream round trip worked.
+    const completionTokens = body?.usage?.completion_tokens;
+    if (choice !== undefined && typeof completionTokens === "number" && completionTokens > 0) {
+      return { chatWorking: true };
+    }
     return { chatWorking: false, chatError: "代理返回了空响应" };
   } catch (error) {
     return { chatWorking: false, chatError: String(error?.message ?? error) };

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-llm-workbuddy",
-  "version": "0.1.14",
+  "version": "0.1.15",
   "description": "WorkBuddy (via the local workbuddy2api proxy) LLM provider adapter for DeepSeek Harness, with a Web login-status widget",
   "type": "module",
   "main": "lib/index.js",