ucu-mcp 0.6.6 → 0.6.7

package/CHANGELOG.md

@@ -5,6 +5,40 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.6.7] - 2026-06-18
+
+Claude Code 实机测试暴露的致命 bug：Chromium/Electron app 操控完全失败。
+
+### P0 Fixed — Chromium/Electron app 操控失败根因
+
+- **windowlist-helper 只返回当前屏幕可见窗口**（windowlist/main.swift）：`.optionOnScreenOnly` 导致其他 Space / 最小化 / 被遮挡的窗口不可见。Microsoft Edge（14 个窗口）完全不在列表里 → `focus_app("Microsoft Edge")` 找不到窗口 → 退到 tray fallback → 返回 `windowId:"tray"` → 所有后续操作发到错误 pid。**修复**：改为 `[.optionAll, .excludeDesktopElements]`（80 个窗口全可见），TS 层用 `isOnScreen` 过滤。
+- **selectWindowForApp 多窗口选择缺陷**（helpers.ts）：Chromium app 暴露多个 compositor 子窗口，`find` 第一个可能不是主浏览器窗口。**修复**：改为评分排序（isOnScreen +1000, has title +500, 最大面积），优先选可见有标题的大窗口。
+
+### P0 Fixed — 浏览器操作必要快捷键被屏蔽
+
+- **cmd+l 从 blocklist 移除**（guard.ts）：cmd+l = 浏览器聚焦地址栏（必要操作），不是锁屏（现代 macOS 用 ctrl+cmd+q）。屏蔽它让浏览器自动化完全不可能。
+- **cmd+w 从 blocklist 移除**（guard.ts）：cmd+w = 关闭标签页（常用操作）。
+- 新增 `ctrl+cmd+q`（现代 macOS 锁屏）到 blocklist。
+
+### 根因分析
+
+Claude Code 报告的 5 个连锁失败全部源于同一根因——windowlist 看不到 Edge 窗口：
+1. focus_app 返回 tray → pid 错误（tray helper 而非浏览器进程）
+2. press_key 到错误 pid → 假装成功
+3. type_text 无 target → 强制 HID → 打到前台终端
+4. find_element 返回空 → AX 树不可见（Chromium 正常，但 windowlist 问题是额外障碍）
+5. 截图只显示前台 app → 无法验证操作效果
+
+修复 1（windowlist .optionAll）解决了 1/2/3/5；修复 2（cmd+l/w）解决了 4 的连锁影响。
+
+### Tests
+
+- 330 passed（+2：cmd+l/cmd+w allow + ctrl+cmd+q block）。
+
+## [0.6.6] - 2026-06-18
+
+Skill 重写修复抢前台 + 菜单栏误用。
+
 ## [0.6.5] - 2026-06-18
 
 Code review P2 加固（3 批次）：安全面 + 输入正确性 + 日志卫生。

package/dist/src/platform/macos/helpers.js

@@ -58,6 +58,18 @@ export function appNameMatches(processName, requestedApp) {
 }
 export function selectWindowForApp(windows, requestedApp) {
     const requested = normalizeAppName(requestedApp);
-    return windows.find((window) => normalizeAppName(window.processName) === requested) ??
-        windows.find((window) => appNameMatches(window.processName, requestedApp));
+    const matched = windows.filter((window) => normalizeAppName(window.processName) === requested);
+    const pool = matched.length > 0 ? matched : windows.filter((window) => appNameMatches(window.processName, requestedApp));
+    if (pool.length === 0)
+        return undefined;
+    if (pool.length === 1)
+        return pool[0];
+    // Multiple windows for the same app (common for Chromium/Electron which expose
+    // many compositor sub-windows). Prefer: onScreen → has title → largest area.
+    const scored = pool.map((w) => ({
+        w,
+        score: (w.isOnScreen ? 1000 : 0) + (w.title ? 500 : 0) + (w.bounds.width * w.bounds.height),
+    }));
+    scored.sort((a, b) => b.score - a.score);
+    return scored[0].w;
 }

package/dist/src/safety/guard.js

@@ -10,15 +10,18 @@
 // Built-in blocked shortcuts
 // ---------------------------------------------------------------------------
 const DEFAULT_BLOCKED_KEYS = [
-    // macOS – app-level
-    "cmd+q",
+    // macOS – truly irreversible app/system actions
+    "cmd+q", // quit app (data loss risk)
     "cmd+shift+q", // log out（方向2 后字母 q 可解析，须显式拦截）
     "cmd+option+q", // log out variant
-    "cmd+w",
-    "cmd+l", // lock screen
+    // NOTE: cmd+l (lock screen / browser address bar) and cmd+w (close tab) were
+    // previously blocked, but they are essential for browser automation (cmd+l =
+    // focus address bar, cmd+w = close tab). Lock screen on modern macOS is
+    // ctrl+cmd+q, not cmd+l. Removed from blocklist v0.6.7.
     // macOS – system-level
     "cmd+option+esc", // Force-quit dialog
     "cmd+ctrl+power", // force restart
+    "ctrl+cmd+q", // lock screen (modern macOS)
     "cmd+option+power", // sleep
     // Windows / Linux
     "alt+f4",

package/native/windowlist/main.swift

@@ -28,7 +28,10 @@ struct Output: Encodable {
     let error: String?
 }
 
-let options: CGWindowListOption = .optionOnScreenOnly
+// .optionAll includes windows on all Spaces + minimized windows. We filter by
+// isOnScreen in the TS layer. Using .optionOnScreenOnly would miss windows on
+// other Spaces or minimized ones, causing focus_app to fail for any background app.
+let options: CGWindowListOption = [.optionAll, .excludeDesktopElements]
 guard let windowList = CGWindowListCopyWindowInfo(options, kCGNullWindowID) as? [[String: Any]] else {
     let out = Output(windows: [], error: "CGWindowListCopyWindowInfo returned nil")
     FileHandle.standardOutput.write(try! JSONEncoder().encode(out))

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "ucu-mcp",
-  "version": "0.6.6",
+  "version": "0.6.7",
   "description": "MCP server for Universal Computer Use — desktop automation for AI agents via Model Context Protocol",
   "type": "module",
   "bin": {