fix: harden extraction pipeline + update Codex Desktop v26.309.31024

- Fallback originator scanning: try/catch extractFromMainJs, scan all
.vite/build/*.js on failure instead of aborting
- Webview model discovery: scan webview/assets/*.js for additional
model IDs not found in main.js
- Fix bracket search in extractPrompts(): limit lastIndexOf("[") to
50-char window instead of unbounded backward search
- Prompt validation in savePrompt(): reject content <50 chars or
with >3 garbled lines to prevent corruption
- Fix corrupted title-generation.md (garbled lines 17-35)
- Update Codex Desktop to v26.309.31024 (build 962)

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

CHANGELOG.md

@@ -8,18 +8,34 @@
 
 ### Added
 
+- 双窗口配额显示：Dashboard 账号卡片同时展示主窗口（小时限制）和次窗口（周限制）的用量百分比、进度条和重置时间，后端 `secondary_window` 不再被忽略
 - 更新弹窗 + 自动重启：点击"有可用更新"弹出 Modal 显示 changelog，一键更新后服务器自动重启、前端自动刷新，零人工干预（git 模式 spawn 新进程、Docker/Electron 显示对应操作指引）
 - Model-aware 多计划账号路由：不同 plan（free/plus/business）的账号自动路由到各自支持的模型，business 账号可继续使用 gpt-5.4 等高端模型 (#57)
 - Structured Outputs 支持：`/v1/chat/completions` 支持 `response_format`（`json_object` / `json_schema`），Gemini 端点支持 `responseMimeType` + `responseSchema`，自动翻译为 Codex Responses API 的 `text.format`；`/v1/responses` 直通 `text` 字段
 
 ### Changed
 
+- 提取管道强化：`extract-fingerprint.ts` 新增 fallback 扫描（`.vite/build/*.js` 全文件回退）和 webview 模型发现（`webview/assets/*.js`），pattern 失败不再中断整个流程
+- 模型/别名自动添加降级为 semi-auto：后端已通过 `isCodexCompatibleId()` 自动合并新模型，`apply-update.ts` 不再自动写入 `models.yaml`（避免 `mutateYaml` 破坏 YAML 格式）
+- Codex Desktop 版本更新至 v26.309.31024 (build 962)
+
+### Fixed (pipeline)
+
+- Prompt 提取括号定位修复：`extractPrompts()` 的 `lastIndexOf("[")` 无限回溯导致匹配到无关 `[`，截取错误代码片段产出乱码；改为 50 字符窗口内搜索
+- Prompt 覆写安全校验：`savePrompt()` 和 `applyAutoChanges()` 新增内容验证（最小长度 50 字符、乱码行数 ≤3），拒绝将损坏数据写入 `config/prompts/`
+- `title-generation.md` 修复：还原因提取 bug 损坏的 title 生成 prompt（第 17-35 行乱码）
+
+### Changed (previous)
+
 - 模型目录大幅更新：后端移除 free 账号的 `gpt-5.4`、`gpt-5.3-codex` 全系列（plus 及以上仍可用），新旗舰模型为 `gpt-5.2-codex`（`codex` 别名指向此模型）
 - 新增模型：`gpt-5.2`、`gpt-5.1-codex`、`gpt-5.1`、`gpt-5-codex`、`gpt-5`、`gpt-oss-120b`、`gpt-oss-20b`、`gpt-5-codex-mini`
 - 模型目录从 23 个静态模型精简为 11 个（匹配后端实际返回）
 
 ### Fixed
 
+- 429 真实冷却时间：从 429 错误响应体解析 `resets_in_seconds` / `resets_at`，账号按后端实际冷却期（如 free 计划 5.5 天）标记限速，不再使用硬编码 60s 默认值 (#65)
+- 429 自动降级：收到 429 后自动尝试下一个可用账号，所有账号耗尽后才返回 429 给客户端 (#65)
+- 调度优先级优化：`least_used` 策略新增 `window_reset_at` 二级排序，配额窗口更早重置的账号优先使用 (#65)
 - JSON Schema `additionalProperties` 递归注入：`injectAdditionalProperties()` 递归注入 `additionalProperties: false` 到 JSON Schema 所有 object 节点，覆盖 `properties`、`patternProperties`、`$defs`/`definitions`、`items`、`prefixItems`、组合器（`oneOf`/`anyOf`/`allOf`）、条件（`if`/`then`/`else`），含循环检测；三个端点（OpenAI/Gemini/Responses passthrough）统一调用 (#64)
 - CONNECT tunnel header 解析：循环跳过中间 header block（CONNECT 200、100 Continue），修复代理模式下 tunnel 的 `HTTP/1.1 200` 被当作真实状态码导致上游 4xx 错误被掩盖为 502 的问题 (#64)
 - 上游 HTTP 状态码透传：非流式 collect 路径从错误消息提取真实 HTTP 状态码，不再硬编码 502；提取 `toErrorStatus()` 辅助函数统一 4 处 StatusCode 转换 (#64)

config/default.yaml

@@ -3,8 +3,8 @@ api:
   timeout_seconds: 60
 client:
   originator: Codex Desktop
-  app_version: 26.305.950
-  build_number: "863"
+  app_version: 26.309.31024
+  build_number: "962"
   platform: darwin
   arch: arm64
   chromium_version: "144"
@@ -34,5 +34,4 @@ tls:
   curl_binary: auto
   impersonate_profile: chrome144
   proxy_url: null
-  # 当代理不支持 HTTP/2 时启用此选项（如遇到 SETTINGS frame 错误）
   force_http11: true

config/prompts/title-generation.md

@@ -14,22 +14,14 @@ Generate a single-line title that captures the question or core change requested
 - Do not use punctuation at the end.
 - Output the title as plain text with no surrounding quotes or backticks.
 - Use precise, non-redundant language.
-s locale (e.g., "Fix bug" -> "Corrige el error" in Spanish-ES), but leave code terms in English unless a widely adopted translation exists.`,"- If the user provides a title explicitly, reuse it (translated if needed) and skip generation logic.",
-Fix
-Add
-Find
-Locate
-Count
-,"- Do NOT respond to the user, answer questions, or attempt to solve the problem; just write a title that can represent the user
-,
-,
-,'- User: 
- -> Add dark-mode support','- User: 
- (de-DE) -> Login-Fehler 500 beheben','- User: 
- (fr-FR) -> Refactoriser composant sidebar','- User: 
- -> Troubleshoot login bug','- User: 
- -> Locate foo_bar',`- User: 
- -> Calculate 2+2`,
-,
-,
-,
+- Translate fixed phrases into the user's locale (e.g., "Fix bug" -> "Corrige el error" in Spanish-ES), but leave code terms in English unless a widely adopted translation exists.
+- If the user provides a title explicitly, reuse it (translated if needed) and skip generation logic.
+- Do NOT respond to the user, answer questions, or attempt to solve the problem; just write a title that can represent the user's query.
+
+Examples:
+- User: "Can we add dark-mode support to the settings page?" -> Add dark-mode support
+- User: "Fehlerbehebung: Beim Anmelden erscheint 500." (de-DE) -> Login-Fehler 500 beheben
+- User: "Refactoriser le composant sidebar pour réduire le code dupliqué." (fr-FR) -> Refactoriser composant sidebar
+- User: "How do I fix our login bug?" -> Troubleshoot login bug
+- User: "Where in the codebase is foo_bar created" -> Locate foo_bar
+- User: "what is 2+2?" -> Calculate 2+2

scripts/extract-fingerprint.ts

@@ -188,6 +188,17 @@ function extractFromMainJs(
   };
 }
 
+/**
+ * Find the nearest `[` bracket within maxDistance chars before the given position.
+ * Prevents unbounded `lastIndexOf("[")` from matching a wrong bracket thousands of chars away.
+ */
+function findNearbyBracket(content: string, position: number, maxDistance = 50): number {
+  const searchStart = Math.max(0, position - maxDistance);
+  const slice = content.slice(searchStart, position);
+  const idx = slice.lastIndexOf("[");
+  return idx !== -1 ? searchStart + idx : -1;
+}
+
 /**
  * Step B (continued): Extract system prompts from main.js
  */
@@ -227,8 +238,8 @@ function extractPrompts(content: string): {
     // Find the enclosing array end: ].join(
     const joinIdx = content.indexOf("].join(", titleStart);
     if (joinIdx !== -1) {
-      // Extract the array content between [ and ]
-      const bracketStart = content.lastIndexOf("[", titleStart);
+      // Find the opening [ within 50 chars before the marker (not unbounded lastIndexOf)
+      const bracketStart = findNearbyBracket(content, titleStart);
       if (bracketStart !== -1) {
         const arrayContent = content.slice(bracketStart + 1, joinIdx);
         // Parse string literals from the array
@@ -244,7 +255,7 @@ function extractPrompts(content: string): {
   if (prStart !== -1) {
     const joinIdx = content.indexOf("].join(", prStart);
     if (joinIdx !== -1) {
-      const bracketStart = content.lastIndexOf("[", prStart);
+      const bracketStart = findNearbyBracket(content, prStart);
       if (bracketStart !== -1) {
         const arrayContent = content.slice(bracketStart + 1, joinIdx);
         prGeneration = parseStringArray(arrayContent);
@@ -318,6 +329,17 @@ function savePrompt(name: string, content: string | null): { hash: string | null
   const sanitized = sanitizePrompt(content);
   if (!sanitized) return { hash: null, path: null };
 
+  // Validate: reject suspiciously short or garbled content
+  if (sanitized.length < 50) {
+    console.warn(`[extract] Prompt "${name}" too short (${sanitized.length} chars), skipping save`);
+    return { hash: null, path: null };
+  }
+  const garbageLines = sanitized.split("\n").filter((l) => /^[,`'"]\s*$/.test(l.trim()));
+  if (garbageLines.length > 3) {
+    console.warn(`[extract] Prompt "${name}" has ${garbageLines.length} garbled lines, skipping save`);
+    return { hash: null, path: null };
+  }
+
   mkdirSync(PROMPTS_DIR, { recursive: true });
   const filePath = join(PROMPTS_DIR, `${name}.md`);
   writeFileSync(filePath, sanitized);
@@ -433,7 +455,47 @@ async function main() {
   if (mainJs) {
     console.log(`[extract] main.js loaded (${mainJs.split("\n").length} lines)`);
 
-    mainJsResults = extractFromMainJs(mainJs, patterns.main_js);
+    try {
+      mainJsResults = extractFromMainJs(mainJs, patterns.main_js);
+    } catch (err) {
+      console.warn(`[extract] Primary extraction failed: ${(err as Error).message}`);
+      console.log("[extract] Scanning all .vite/build/*.js for fallback...");
+
+      const buildDir = join(asarRoot, ".vite/build");
+      if (existsSync(buildDir)) {
+        const jsFiles = readdirSync(buildDir).filter((f) => f.endsWith(".js"));
+        for (const file of jsFiles) {
+          const content = readFileSync(join(buildDir, file), "utf-8");
+          const origPattern = patterns.main_js.originator;
+          if (origPattern?.pattern) {
+            const m = content.match(new RegExp(origPattern.pattern));
+            if (m) {
+              mainJsResults.originator = m[origPattern.group ?? 0] ?? m[0];
+              console.log(`[extract] Originator found in fallback file: ${file}`);
+              break;
+            }
+          }
+        }
+      }
+
+      // Re-extract non-critical fields from mainJs
+      const apiPattern = patterns.main_js.api_base_url;
+      if (apiPattern?.pattern) {
+        const m = mainJs.match(new RegExp(apiPattern.pattern));
+        if (m) mainJsResults.apiBaseUrl = m[0];
+      }
+      const modelPattern = patterns.main_js.models;
+      if (modelPattern?.pattern) {
+        const re = new RegExp(modelPattern.pattern, "g");
+        const groupIdx = modelPattern.group ?? 0;
+        const modelSet = new Set<string>();
+        for (const m of mainJs.matchAll(re)) {
+          modelSet.add(m[groupIdx] ?? m[0]);
+        }
+        mainJsResults.models = [...modelSet].sort();
+      }
+    }
+
     console.log(`  API base URL:  ${mainJsResults.apiBaseUrl}`);
     console.log(`  originator:    ${mainJsResults.originator}`);
     console.log(`  models:        ${mainJsResults.models.join(", ")}`);
@@ -448,6 +510,31 @@ async function main() {
     console.log(`  automation-response: ${promptResults.automationResponse ? "found" : "NOT FOUND"}`);
   }
 
+  // Scan webview assets for additional model IDs
+  const webviewAssetsDir = join(asarRoot, "webview/assets");
+  if (existsSync(webviewAssetsDir)) {
+    console.log("[extract] Scanning webview assets for additional models...");
+    const modelPattern = patterns.main_js.models;
+    if (modelPattern?.pattern) {
+      const webviewFiles = readdirSync(webviewAssetsDir).filter((f) => f.endsWith(".js"));
+      const webviewModels = new Set<string>();
+      for (const file of webviewFiles) {
+        const content = readFileSync(join(webviewAssetsDir, file), "utf-8");
+        const re = new RegExp(modelPattern.pattern, "g");
+        const groupIdx = modelPattern.group ?? 0;
+        for (const m of content.matchAll(re)) {
+          webviewModels.add(m[groupIdx] ?? m[0]);
+        }
+      }
+      const existingModels = new Set(mainJsResults.models);
+      const newFromWebview = [...webviewModels].filter((m) => !existingModels.has(m));
+      if (newFromWebview.length > 0) {
+        console.log(`[extract] Webview: ${newFromWebview.length} additional models: ${newFromWebview.join(", ")}`);
+        mainJsResults.models = [...mainJsResults.models, ...newFromWebview].sort();
+      }
+    }
+  }
+
   // Save extracted prompts
   const dc = savePrompt("desktop-context", promptResults.desktopContext);
   const tg = savePrompt("title-generation", promptResults.titleGeneration);