dsh-llm-workbuddy 0.1.7 → 0.1.11

package/README.md

@@ -16,7 +16,7 @@
 1. **模型能力**：Web 界面的模型选择器（composer 模型菜单或 `/model` 命令）会多出
    一个 **WorkBuddy** 分组，模型（DeepSeek-V4、GLM-5.x、Kimi-K2.x、MiniMax-M3、
    Hy3、Hunyuan…）随账号可用列表实时同步，点一下即可切换；支持推理的模型还会
-   显示**推理等级**选择器（低 / 中 / 高）。
+   显示**推理等级**选择器（Low / Medium / High）。
 2. **Web 登录状态小组件**：在 Web GUI 右下角常驻一个状态胶囊，**实时显示登录/
    代理状态**，未登录时一键在新标签页打开 WorkBuddy 登录页，登录完成后自动变绿。
    无需再回到终端手动跑登录脚本。
@@ -280,8 +280,10 @@ DSH 的 `dsh.client` 机制只要求 `package.json` 里：
 
 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。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）。
 
 ---
 
@@ -302,9 +304,12 @@ DSH 的 `dsh.client` 机制只要求 `package.json` 里：
 
 ## 限制
 
-- 当前为纯文本适配器：图片输入会以 `UNSUPPORTED_CONTENT` 拒绝（后续可加）。
+- 图片输入：声明支持图片的模型（`inputModalities: ["text", "image"]`，如 deepseek-v4-pro、
+  glm-5.2、kimi-k2.x、hy3 等）可附带图片，适配器会通过 DSH 的附件服务把图片编码为
+  `data:<mime>;base64,<bytes>` 以 OpenAI `image_url` 格式透传给代理。若附件服务不可用
+  （headless 等无附件场景），图片输入会以 `UNSUPPORTED_CONTENT` 稳定报错。
 - 推理等级（reasoning effort）：支持推理的模型（如 DeepSeek-V4、GLM、Kimi、MiniMax、
-  Hy3 等）会显示推理等级下拉（低 / 中 / 高），默认值取平台默认强度。`reasoning_effort`
+  Hy3 等）会显示推理等级下拉（Low / Medium / High），默认值取平台默认强度。`reasoning_effort`
   会透传给代理；若某模型平台侧只接受平台默认、忽略该参数，则退化为平台默认强度，不影响出字。
 - 代理未运行时，模型请求会以 `TRANSPORT` 错误快速失败（连接被拒绝）；但状态
   小组件本身不依赖代理——代理挂了它仍能显示「代理未运行」并允许触发登录。

package/lib/index.js

@@ -75,19 +75,15 @@ const DISCOVERY_TIMEOUT_MS = 2_000;
 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-5.0", name: "GLM-5.0", contextWindow: 200_000, maxTokens: 48_000, reasoningEffort: "high" },
-  { id: "glm-4.7", name: "GLM-4.7", contextWindow: 200_000, maxTokens: 48_000, reasoningEffort: "high" },
-  { id: "glm-4.6", name: "GLM-4.6", contextWindow: 168_000, maxTokens: 32_000, reasoningEffort: "high" },
   { 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: "kimi-k2-thinking", name: "Kimi-K2-Thinking", contextWindow: 164_000, maxTokens: 32_000, reasoningEffort: "medium" },
   { id: "minimax-m3", name: "MiniMax-M3", contextWindow: 512_000, maxTokens: 128_000, inputModalities: ["text", "image"], reasoningEffort: "medium" },
-  { id: "minimax-m2.5", name: "MiniMax-M2.5", contextWindow: 200_000, maxTokens: 48_000, 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: "hunyuan-chat", name: "Hunyuan-Turbos", contextWindow: 200_000, maxTokens: 8_192 },
@@ -97,9 +93,9 @@ const DEFAULT_MODELS = [
 
 /** Selectable reasoning efforts exposed to the harness UI, in display order. */
 const REASONING_EFFORTS = [
-  { id: "low", name: "低" },
-  { id: "medium", name: "中" },
-  { id: "high", name: "高" },
+  { id: "low", name: "Low" },
+  { id: "medium", name: "Medium" },
+  { id: "high", name: "High" },
 ];
 
 /**
@@ -169,14 +165,32 @@ function serializeAssistant(message) {
 /**
  * Serialize the harness conversation into OpenAI chat-completions wire
  * messages. `tool-result` blocks become standalone `{role: 'tool'}` messages;
- * image content is rejected (the initial version is text-only).
+ * image blocks are read through the durable attachment service and emitted as
+ * OpenAI `image_url` parts (`data:<mime>;base64,<bytes>`), which the
+ * workbuddy2api proxy passes through to the upstream platform. When the
+ * attachment service is unavailable, image input degrades to the stable
+ * `UNSUPPORTED_CONTENT` error.
  */
-function serializeMessages(messages) {
+async function serializeMessages(messages, attachments, signal) {
+  const refs = new Map();
+  for (const message of messages) collectImageRefs(message.content, refs);
+  const requestImages = new Map();
+  if (refs.size > 0) {
+    if (attachments === undefined) {
+      throw new LlmError("WorkBuddy image input requires the durable attachment service", "UNSUPPORTED_CONTENT");
+    }
+    // Keep images small enough for the upstream platform: the proxy relays the
+    // base64 body verbatim, and oversized payloads are dropped/ignored upstream
+    // (models reply "no image attached"). Mirror the official adapter budgets.
+    const policy = { maxPixels: 2048 * 2048, maxBytes: 1024 * 1024 };
+    const ordered = [...refs.values()];
+    const prepared = await Promise.all(ordered.map((ref) => attachments.readImageRequest(ref, policy, signal)));
+    for (let index = 0; index < ordered.length; index += 1) {
+      requestImages.set(ordered[index].attachmentId, prepared[index]);
+    }
+  }
   const wire = [];
   for (const message of messages) {
-    if (message.content.some((block) => block.type === "image")) {
-      throw new LlmError("The WorkBuddy adapter does not support image content yet.", "UNSUPPORTED_CONTENT");
-    }
     if (message.role === "system") {
       wire.push({ role: "system", content: flattenText(message.content) });
       continue;
@@ -187,7 +201,26 @@ function serializeMessages(messages) {
     }
     const toolResults = message.content.filter((block) => block.type === "tool-result");
     const text = flattenText(message.content);
-    if (text.length > 0 || toolResults.length === 0) wire.push({ role: "user", content: text });
+    const images = message.content.filter((block) => block.type === "image");
+    if (text.length > 0 || toolResults.length === 0) {
+      if (images.length > 0) {
+        const parts = [];
+        if (text.length > 0) parts.push({ type: "text", text });
+        for (const block of images) {
+          const version = requestImages.get(block.attachment.attachmentId);
+          if (version === undefined) {
+            throw new LlmError("WorkBuddy image input missing attachment bytes", "UNSUPPORTED_CONTENT");
+          }
+          parts.push({
+            type: "image_url",
+            image_url: { url: `data:${version.mediaType};base64,${Buffer.from(version.data).toString("base64")}` },
+          });
+        }
+        wire.push({ role: "user", content: parts });
+      } else {
+        wire.push({ role: "user", content: text });
+      }
+    }
     for (const result of toolResults) {
       wire.push({ role: "tool", tool_call_id: result.toolCallId, content: flattenText(result.content) || "(no output)" });
     }
@@ -195,11 +228,19 @@ function serializeMessages(messages) {
   return wire;
 }
 
+/** Collect image attachment refs from a content block list (recursing into tool results). */
+function collectImageRefs(blocks, refs) {
+  for (const block of blocks) {
+    if (block.type === "image") refs.set(block.attachment.attachmentId, block.attachment);
+    else if (block.type === "tool-result") collectImageRefs(block.content, refs);
+  }
+}
+
 /** Build the full wire request. Always streaming with usage reporting on. */
-function serializeRequest(options) {
+async function serializeRequest(options, attachments, signal) {
   const messages = [];
   if (options.system !== undefined) messages.push({ role: "system", content: options.system });
-  messages.push(...serializeMessages(options.messages));
+  messages.push(...await serializeMessages(options.messages, attachments, signal));
   const tools = options.tools?.map((tool) => ({
     type: "function",
     function: { name: tool.name, description: tool.description, parameters: tool.parameters },
@@ -511,6 +552,11 @@ export class WorkBuddyAdapter extends LlmAdapter {
         if (seen.has(entry.id)) continue;
         seen.add(entry.id);
         const catalog = byId.get(entry.id);
+        // Only keep models the static catalog declares. The proxy announces
+        // retired models (glm-4.6v, hunyuan-image-v3.0, ...) that upstream
+        // rejects with `service info not found`; the catalog doubles as the
+        // whitelist so those never reach the UI.
+        if (catalog === undefined) continue;
         const effort = liveReasoningEffort(entry, catalog?.reasoningEffort);
         const reasoning = modelReasoningInfo(effort);
         merged.push({
@@ -588,7 +634,8 @@ export class WorkBuddyAdapter extends LlmAdapter {
   }
 
   async *request(options, signal, connection, onComment) {
-    const body = serializeRequest(options);
+    const attachments = this.config.resolveAttachments?.();
+    const body = await serializeRequest(options, attachments, signal);
     const headers = {
       "content-type": "application/json",
       "accept": "text/event-stream",
@@ -1053,7 +1100,10 @@ export function apply(ctx, config) {
     }
   };
   options();
-  const adapter = new WorkBuddyAdapter({ options });
+  const adapter = new WorkBuddyAdapter({
+    options,
+    resolveAttachments: () => ctx.get("attachments"),
+  });
   ctx.llm.registerConfigurableProviders([{
     provider: PROVIDER,
     displayName: "WorkBuddy",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dsh-llm-workbuddy",
-  "version": "0.1.7",
+  "version": "0.1.11",
   "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",