feat: enhance session sync and add voice input feature

- 增强会话同步：确保所有会话完全同步到新账号
  - 强制更新 model_provider 字段
  - 添加 session_index.jsonl 重建功能
  - 确保数据库 UPSERT 逻辑正确处理所有情况
- 新增语音输入功能（voice_cdp.js, voice_inject.js）
- 测试脚本和增强模块

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

electron/lib/router.js

@@ -304,6 +304,16 @@ function llmGroup() {
   return config.groups.find(usable) || null;
 }
 
+// 供「语音输入」转写用：优先当前激活且勾选 STT 的分组，否则取任一勾选 STT 的分组。
+// 需配齐 baseUrl+key 且勾了 stt；返回的 group 带 sttModel（如 speech-01 / whisper-1）。
+function sttGroup() {
+  load();
+  const ok = (g) => g && !g.isDefault && g.baseUrl && g.key && g.stt;
+  const act = config.groups.find((g) => g.id === config.activeId);
+  if (ok(act)) return act;
+  return config.groups.find(ok) || null;
+}
+
 // 拉取某个中转 API 的模型列表（试 /v1/models 再 /models）
 async function fetchModels(baseUrl, key) {
   if (!baseUrl) throw new Error("请先填 API 地址");
@@ -343,6 +353,64 @@ function writeCapture(name, data) {
   } catch {}
 }
 
+function requestTextForIntent(req) {
+  const parts = [];
+  const walk = (v) => {
+    if (v == null) return;
+    if (typeof v === "string") {
+      parts.push(v);
+      return;
+    }
+    if (Array.isArray(v)) {
+      for (const it of v) walk(it);
+      return;
+    }
+    if (typeof v === "object") {
+      for (const key of ["text", "content", "input", "output", "arguments"]) walk(v[key]);
+    }
+  };
+  walk(req && req.input);
+  return parts.join("\n");
+}
+
+function hasExplicitGenImageIntent(req) {
+  return /\$gen[\s_-]*image\b/i.test(requestTextForIntent(req));
+}
+
+function stripGenImageSkill(instructions) {
+  if (typeof instructions !== "string" || !instructions.includes("gen-image")) return instructions;
+  return instructions
+    .split(/\r?\n/)
+    .filter((line) => !/^\s*-\s+gen-image\s*:/.test(line))
+    .join("\n");
+}
+
+function rewriteResponsesBody(body, group) {
+  if (!body.length) return body;
+  let j;
+  try {
+    j = JSON.parse(body.toString("utf-8"));
+  } catch {
+    return body;
+  }
+  if (!j || typeof j !== "object") return body;
+  let changed = false;
+  if (group.model && "model" in j && j.model !== group.model) {
+    j.model = group.model;
+    changed = true;
+  }
+  // 普通聊天分组没有 image-generation 权限时，上游会因 gen-image skill 声明误判为生图请求。
+  // 只有用户显式输入 "$gen image" 才保留该 skill，让真正生图走脚本里的图片接口分组。
+  if (!hasExplicitGenImageIntent(j)) {
+    const nextInstructions = stripGenImageSkill(j.instructions);
+    if (nextInstructions !== j.instructions) {
+      j.instructions = nextInstructions;
+      changed = true;
+    }
+  }
+  return changed ? Buffer.from(JSON.stringify(j), "utf-8") : body;
+}
+
 function handle(req, res) {
   const g = activeGroup();
   if (!g) {
@@ -360,16 +428,7 @@ function handle(req, res) {
   req.on("end", () => {
     let body = Buffer.concat(chunks);
     const doCapture = /responses/.test(req.url || "");
-    // 分组指定了模型 → 覆盖请求体里的 model
-    if (g.model && body.length) {
-      try {
-        const j = JSON.parse(body.toString("utf-8"));
-        if (j && typeof j === "object" && "model" in j) {
-          j.model = g.model;
-          body = Buffer.from(JSON.stringify(j), "utf-8");
-        }
-      } catch {}
-    }
+    if (doCapture) body = rewriteResponsesBody(body, g);
     // —— chat 模式：Responses ⇄ Chat Completions 协议转换（DeepSeek/GLM/Kimi 等）——
     if (g.mode === "chat" && /responses/.test(req.url || "")) {
       let responsesReq = {};
@@ -577,6 +636,8 @@ function status() {
       hasKey: g.isDefault ? true : !!g.key,
       isDefault: !!g.isDefault,
       codexImage: !!g.codexImage,
+      stt: !!g.stt,
+      sttModel: g.sttModel || "",
     })),
   };
 }
@@ -880,5 +941,6 @@ module.exports = {
   groupById,
   usableGroups,
   llmGroup,
+  sttGroup,
   fetchModels,
 };