feat: polish hotword overrides, audio archive, auto-update check

openless-all/app/src-tauri/src/commands.rs

@@ -436,8 +436,8 @@ pub async fn start_microphone_level_monitor(
         let level_handler: Arc<dyn Fn(f32) + Send + Sync> = Arc::new(move |level| {
             let _ = level_app.emit("microphone:level", serde_json::json!({ "level": level }));
         });
-        let (recorder, _runtime_errors) =
-            Recorder::start(microphone_device_name, consumer, level_handler)
+        let (recorder, _runtime_errors, _archive_active) =
+            Recorder::start(microphone_device_name, consumer, level_handler, None)
                 .map_err(|e| e.to_string())?;
         *state.lock() = Some(recorder);
         Ok(())
@@ -1115,6 +1115,59 @@ pub fn clear_history(coord: CoordinatorState<'_>) -> Result<(), String> {
     coord.history().clear().map_err(|e| e.to_string())
 }
 
+/// 读取某次会话的原始麦克风 wav 字节流。仅当用户开过
+/// `prefs.record_audio_for_debug` 并且这条 session 是开关打开后录的，才会有文件。
+/// 文件名规约：`<data_dir>/recordings/<session_id>.wav`，与 DictationSession.id 同名。
+///
+/// 路径校验：session_id **必须**严格匹配 UUID-v4 字面（36 字符 = 8-4-4-4-12 + 4 个 `-`，
+/// 内容仅 ASCII 十六进制 + `-`）。白名单胜过黑名单——绝对路径前缀、Windows ADS、
+/// 百分号编码、NUL 字节都不在合法字符集里，挡掉所有 Path::join 越界的可能。
+/// session_id 在仓库内由 `Uuid::new_v4()` 生成 (`dictation.rs:1531`)，前端只会回传
+/// 自己列出的合法 id，但 IPC = boundary，按 boundary 规则严格校验。
+///
+/// async fs：单条 5 分钟 wav 约 9.6MB，同步 `std::fs::read` 会阻塞 Tauri IPC 主循环。
+/// 改 `tokio::fs::read` 后让出线程给其它 IPC。
+#[tauri::command]
+pub async fn read_audio_recording(session_id: String) -> Result<Vec<u8>, String> {
+    if !is_valid_session_id(&session_id) {
+        return Err("invalid session id".into());
+    }
+    let path =
+        crate::persistence::recording_path_for_session(&session_id).map_err(|e| e.to_string())?;
+    if !path.exists() {
+        return Err("recording not found".into());
+    }
+    // TOCTOU 兜底：exists() 通过到 read 之间文件可能被 prune（条数 cap / retention
+    // 清理 / 用户手动删）。把 NotFound 标准化成跟 exists() 失败同样的错误字符串，
+    // 前端单条 'recording not found' catch 就能稳定隐藏按钮，不依赖本地化 OS 错误。
+    tokio::fs::read(&path).await.map_err(|e| {
+        if e.kind() == std::io::ErrorKind::NotFound {
+            "recording not found".into()
+        } else {
+            format!("read wav failed: {e}")
+        }
+    })
+}
+
+/// UUID-v4 字面校验：36 字符 + 5 段 `-` 分隔（8-4-4-4-12）+ 仅 ASCII 十六进制。
+fn is_valid_session_id(s: &str) -> bool {
+    if s.len() != 36 {
+        return false;
+    }
+    let bytes = s.as_bytes();
+    for (i, b) in bytes.iter().enumerate() {
+        let is_dash_position = matches!(i, 8 | 13 | 18 | 23);
+        if is_dash_position {
+            if *b != b'-' {
+                return false;
+            }
+        } else if !b.is_ascii_hexdigit() {
+            return false;
+        }
+    }
+    true
+}
+
 // ─────────────────────────── vocab ───────────────────────────
 
 #[tauri::command]
@@ -2225,9 +2278,9 @@ mod tests {
     use super::{
         active_asr_is_keyless_for_validation, active_foundry_model_from_prefs,
         asr_configured_for_provider, asr_transcriptions_url, fetch_provider_models,
-        is_gemini_base_url, llm_configured_for_provider, local_asr_release_plan_for_provider,
-        models_url, normalize_foundry_language_hint, parse_gemini_model_ids,
-        parse_latest_beta_from_atom, parse_model_ids, persist_settings,
+        is_gemini_base_url, is_valid_session_id, llm_configured_for_provider,
+        local_asr_release_plan_for_provider, models_url, normalize_foundry_language_hint,
+        parse_gemini_model_ids, parse_latest_beta_from_atom, parse_model_ids, persist_settings,
         release_foundry_runtime_if_inactive, validate_foundry_model_alias, ProviderConfig,
         SettingsWriter,
     };
@@ -2942,4 +2995,32 @@ mod tests {
         assert_eq!(models, vec!["m1".to_string(), "m2".to_string()]);
         server.join().unwrap();
     }
+
+    #[test]
+    fn is_valid_session_id_accepts_canonical_uuid_v4() {
+        // canonical UUID-v4 字面：8-4-4-4-12，全小写、全大写、混合都接受。
+        assert!(is_valid_session_id("550e8400-e29b-41d4-a716-446655440000"));
+        assert!(is_valid_session_id("550E8400-E29B-41D4-A716-446655440000"));
+        assert!(is_valid_session_id("Abc12345-6789-abcd-EF01-234567890abc"));
+    }
+
+    #[test]
+    fn is_valid_session_id_rejects_path_traversal_and_garbage() {
+        assert!(!is_valid_session_id(""));
+        assert!(!is_valid_session_id("../../etc/passwd"));
+        assert!(!is_valid_session_id("..\\..\\windows\\system32"));
+        // 长度对但含 `/`：dash 位置错或非 hex 字符都不通过
+        assert!(!is_valid_session_id("550e8400-e29b-41d4-a716-44665544/000"));
+        assert!(!is_valid_session_id("550e8400_e29b_41d4_a716_446655440000")); // 用 _ 代 -
+        // 非 hex 字符
+        assert!(!is_valid_session_id("550e8400-e29b-41d4-a716-44665544000g"));
+        // 长度不对（35 / 37）
+        assert!(!is_valid_session_id("550e8400-e29b-41d4-a716-44665544000"));
+        assert!(!is_valid_session_id("550e8400-e29b-41d4-a716-4466554400000"));
+        // NUL 字节
+        assert!(!is_valid_session_id("550e8400-e29b-41d4-a716-44665544\x00000"));
+        // 百分号编码与绝对路径
+        assert!(!is_valid_session_id("%2e%2e/recordings/x"));
+        assert!(!is_valid_session_id("/Users/attacker/secret.wav"));
+    }
 }

openless-all/app/src-tauri/src/coordinator.rs

@@ -116,6 +116,11 @@ struct Inner {
     #[cfg(target_os = "windows")]
     foundry_local_runtime: Arc<FoundryLocalRuntime>,
     recorder: Mutex<Option<SessionResource<Recorder>>>,
+    /// 当前 dictation / QA session 的 wav 归档是否真的被写到磁盘上。
+    /// 由 Recorder::start 返回值 (archive_active) 写入；history.append 路径读取，
+    /// 决定 DictationSession.has_audio_recording 字段。比单纯读 prefs.record_audio_for_debug
+    /// 更准确：用户开了开关但路径无法创建（权限 / 磁盘满）也算 false。
+    audio_archive_active: AtomicBool,
     recording_mute: Mutex<SharedRecordingMuteState>,
     hotkey: Mutex<Option<HotkeyMonitor>>,
     hotkey_status: Mutex<HotkeyStatus>,
@@ -203,6 +208,7 @@ impl Coordinator {
                     state: Mutex::new(SessionState::default()),
                     asr: Mutex::new(None),
                     recorder: Mutex::new(None),
+                    audio_archive_active: AtomicBool::new(false),
                     recording_mute: Mutex::new(SharedRecordingMuteState::new()),
                     hotkey: Mutex::new(None),
                     hotkey_status: Mutex::new(HotkeyStatus::default()),
@@ -252,6 +258,7 @@ impl Coordinator {
                 state: Mutex::new(SessionState::default()),
                 asr: Mutex::new(None),
                 recorder: Mutex::new(None),
+                audio_archive_active: AtomicBool::new(false),
                 recording_mute: Mutex::new(SharedRecordingMuteState::new()),
                 hotkey: Mutex::new(None),
                 hotkey_status: Mutex::new(HotkeyStatus::default()),
@@ -2584,8 +2591,15 @@ async fn begin_qa_session(inner: &Arc<Inner>) -> Result<(), String> {
     let microphone_device_name = selected_microphone_device_name(inner);
     stop_microphone_preview_monitor(inner, "QA recorder");
     acquire_recording_mute(inner, "qa").await;
-    match Recorder::start(microphone_device_name, consumer, level_handler) {
-        Ok((rec, runtime_errors)) => {
+    // QA 默认不留痕（qa_save_history 默认 false），录音文件归档也跟着不开。
+    // 调试 QA 麦克风请用主听写路径。
+    match Recorder::start(microphone_device_name, consumer, level_handler, None) {
+        Ok((rec, runtime_errors, archive_active)) => {
+            // QA 路径不写 dictation 的 history，但仍把 archive 状态归零，避免 dictation
+            // 接力时读到上一个 QA session 的过期值。
+            inner
+                .audio_archive_active
+                .store(archive_active, std::sync::atomic::Ordering::Relaxed);
             *inner.qa_recorder.lock() = Some(rec);
             // QA 也跟主听写一样监听 cpal runtime error。设备中途消失 / panic 时
             // 不能让 QA 永远卡在 Recording 没反馈。详见 issue #168。
@@ -2853,11 +2867,14 @@ async fn end_qa_session(inner: &Arc<Inner>) -> Result<(), String> {
             error_code: Some("qaSession".to_string()),
             duration_ms: Some(raw.duration_ms),
             dictionary_entry_count: None,
+            has_audio_recording: None,
         };
-        if let Err(e) = inner
-            .history
-            .append_with_retention(session, inner.prefs.get().history_retention_days)
-        {
+        let prefs_snapshot = inner.prefs.get();
+        if let Err(e) = inner.history.append_with_retention(
+            session,
+            prefs_snapshot.history_retention_days,
+            prefs_snapshot.history_max_entries,
+        ) {
             log::error!("[coord] QA history append failed: {e}");
         }
     }

openless-all/app/src-tauri/src/coordinator/dictation.rs

@@ -785,8 +785,31 @@ pub(super) async fn start_recorder_for_starting(
     let microphone_device_name = selected_microphone_device_name(inner);
     stop_microphone_preview_monitor(inner, "dictation recorder");
     acquire_recording_mute(inner, "dictation").await;
-    match Recorder::start(microphone_device_name, consumer, level_handler) {
-        Ok((rec, runtime_errors)) => {
+    let audio_archive_path = if inner.prefs.get().record_audio_for_debug {
+        // 用 coordinator 的 SessionId 作为文件名，跟 history 那条记录 id 对齐（见
+        // 下游 polish 收尾时 `history_session_id = current_session_id.to_string()`）。
+        // 顺手把超龄 / 超量录音清理一下，避免 debug 开关常开时磁盘膨胀。
+        let prefs = inner.prefs.get();
+        let _ = crate::persistence::prune_recordings(
+            prefs.history_retention_days,
+            prefs.audio_recording_max_entries,
+        );
+        crate::persistence::recording_path_for_session(&session_id.to_string()).ok()
+    } else {
+        None
+    };
+    match Recorder::start(
+        microphone_device_name,
+        consumer,
+        level_handler,
+        audio_archive_path,
+    ) {
+        Ok((rec, runtime_errors, archive_active)) => {
+            // 把 archive 实际创建状态存到 Inner，让 history 写入路径（含 empty-transcript
+            // 失败分支）读真实情况，而不是 prefs 开关。修 pr_agent "Wrong Flag" 反馈。
+            inner
+                .audio_archive_active
+                .store(archive_active, std::sync::atomic::Ordering::Relaxed);
             store_recorder_for_session(inner, session_id, rec);
             spawn_recorder_error_monitor(inner, runtime_errors);
             // 不在这里 emit Recording capsule。
@@ -1230,11 +1253,17 @@ pub(super) async fn end_session(inner: &Arc<Inner>) -> Result<(), String> {
             error_code: Some("emptyTranscript".to_string()),
             duration_ms: Some(raw.duration_ms),
             dictionary_entry_count: Some(enabled_phrases(inner).len() as u32),
+            // empty-transcript（ASR 没识别到任何文字）也保留 wav 标记——这是用户最想
+            // 通过原始录音定位"是不是麦克风太小声 / ASR 模型问题"的场景。修 pr_agent
+            // "Missing Audio" 反馈。
+            has_audio_recording: Some(inner.audio_archive_active.load(Ordering::Relaxed)),
         };
-        if let Err(e) = inner
-            .history
-            .append_with_retention(session, inner.prefs.get().history_retention_days)
-        {
+        let prefs_snapshot = inner.prefs.get();
+        if let Err(e) = inner.history.append_with_retention(
+            session,
+            prefs_snapshot.history_retention_days,
+            prefs_snapshot.history_max_entries,
+        ) {
             log::error!("[coord] history append failed: {e}");
         }
         emit_capsule(
@@ -1507,8 +1536,11 @@ pub(super) async fn end_session(inner: &Arc<Inner>) -> Result<(), String> {
     .map(str::to_string);
     let tsf_required_insert_failed = error_code.as_deref() == Some("windowsImeTsfRequired");
 
-    let history_session_id = Uuid::new_v4().to_string();
+    // 与 coordinator 内部 SessionId 对齐：方便 recorder 旁路写盘的 `<session_id>.wav`
+    // 跟 history 这条 DictationSession.id 同名，前端凭 id 就能找到对应录音文件。
+    let history_session_id = current_session_id.to_string();
     let history_created_at = Utc::now().to_rfc3339();
+    let prefs_snapshot = inner.prefs.get();
     let session = DictationSession {
         id: history_session_id.clone(),
         created_at: history_created_at.clone(),
@@ -1523,11 +1555,15 @@ pub(super) async fn end_session(inner: &Arc<Inner>) -> Result<(), String> {
         // 历史详情页的"X 个热词"显示：用本次实际命中次数（每个匹配实例算一次），
         // 比"启用词条总数"更能反映本段口述命中了多少。u64 → u32 截断对单段听写足够。
         dictionary_entry_count: Some(total_hits.min(u32::MAX as u64) as u32),
+        // 用 begin_session 时 Recorder::start 返回的实际写盘状态，而不是 prefs 开关——
+        // 开关打开但路径创建失败时这里是 false，避免前端渲染播放按钮后端 404。
+        has_audio_recording: Some(inner.audio_archive_active.load(Ordering::Relaxed)),
     };
-    if let Err(e) = inner
-        .history
-        .append_with_retention(session, inner.prefs.get().history_retention_days)
-    {
+    if let Err(e) = inner.history.append_with_retention(
+        session,
+        prefs_snapshot.history_retention_days,
+        prefs_snapshot.history_max_entries,
+    ) {
         log::error!("[coord] history append failed: {e}");
     }
     let done_message = if tsf_required_insert_failed {

openless-all/app/src-tauri/src/lib.rs

@@ -283,6 +283,7 @@ pub fn run() {
             commands::list_history,
             commands::delete_history_entry,
             commands::clear_history,
+            commands::read_audio_recording,
             commands::list_vocab,
             commands::add_vocab,
             commands::remove_vocab,