新增: AI工具授权目录总开关(信任所有目录)

- path_auth判定加bypass旁路:开启时白名单外目录免逐次询问直接放行
- 黑名单凭据目录保护不随开关放行(安全硬边界保留)
- 设置页授权目录面板加开关(默认关,读取失败按关处理)
- 对话区开启时常驻放行提示横幅防静默裸奔,中英i18n齐

src/i18n/en/aiChat.ts

@@ -189,6 +189,8 @@ export default {
     dirAuthFailed: 'Authorization failed: {msg}',
     // M5:path deny second confirmation (deny is irreversible — tool returns Err, loop continues without retry; asymmetry with approve, prevents mis-clicks)
     dirAuthDenyConfirm: 'Deny authorization? The tool will fail immediately and will not be retried this round.',
+    // Persistent bypass banner shown at the top of the chat area while "trust all directories" is on
+    pathAuthBypassBanner: 'Authorization bypassed: AI can access files directly',
 
     // ── Input Augmentation layer: mention/skill injection feedback ──
     // search_files blind search denied by whitelist fallback (no auth dialog; prompt user to use @mention or absolute path)

src/i18n/en/settings.ts

@@ -192,6 +192,9 @@ export default {
     allowedDirsDup: 'This directory is already in the authorized list',
     allowedDirsAddFail: 'Failed to add authorized directory',
     allowedDirsRemoveFail: 'Failed to remove authorized directory',
+    // "Trust all directories" master switch (Settings - Security - AI Tool Allowed Dirs panel)
+    labelPathAuthBypass: 'Trust all directories',
+    descPathAuthBypass: 'When enabled, AI can directly access any unprotected directory without per-access prompts; protected (blacklisted) directories such as credential folders are still denied.',
 
     // ===== ScriptNode command execution safety =====
     panelScriptSafety: '🛡️ Command Execution Safety',