fix(usage): keep the sidebar usage summary current while it is shown

The summary fetched reports once on mount and never again: no interval,
window-focus and reconnect refetches are off app-wide, and the daemon does
not push report changes. It only updated when the Usage modal or Settings
mounted the same query and refetched it. It now re-reads every minute while
visible; the daemon answers from its five-minute cache, so source APIs are
still hit at most once per cache window.

packages/app/e2e/browser/usage-sidebar-item.spec.ts

@@ -469,3 +469,27 @@ test("visible usage summary requests reports on load", async ({ page }) => {
   await expectPinnedUsage(page, ["31% 5h", "54% wk", "7% 5h", "12% wk"]);
   expect(usage.listRequests()).toHaveLength(1);
 });
+
+test("visible usage summary picks up changed reports without opening Usage", async ({ page }) => {
+  test.setTimeout(180_000);
+  const usage = await installUsageReportsFixture(page, {
+    lists: [() => claudeSessionAt(31), () => claudeSessionAt(80)],
+  });
+  await seedSidebarFooterPreferences(page, [{ key: "usage", visible: true }]);
+  await page.setViewportSize(WIDE);
+  await page.clock.install();
+  await gotoAppShell(page);
+  await expectPinnedUsage(page, ["31% 5h", "54% wk", "7% 5h", "12% wk"]);
+  expect(usage.listRequests()).toHaveLength(1);
+
+  await page.clock.fastForward(61_000);
+  await expectPinnedUsage(page, ["80% 5h", "54% wk", "7% 5h", "12% wk"]);
+  await expect(usageModal(page)).toHaveCount(0);
+});
+
+function claudeSessionAt(usedPct: number) {
+  const reports = claudeAndCodexReports();
+  const claude = reports[0]!;
+  if (claude.report.status === "available") claude.report.windows[0]!.usedPct = usedPct;
+  return reports;
+}

packages/app/src/usage/queries.ts

@@ -166,7 +166,8 @@ const NO_REPORTS: UsageReportEntry[] = [];
 
 /**
  * The reports of the sidebar's usage host, which is connected and reports usage; none until they
- * load, or without a host.
+ * load, or without a host. Re-read every minute while the summary is shown, since the daemon never
+ * pushes report changes.
  */
 export function useUsageHostReports(serverId: string | null): UsageReportEntry[] {
   const setting = useUsageSetting(serverId);
@@ -177,6 +178,7 @@ export function useUsageHostReports(serverId: string | null): UsageReportEntry[]
     queryFn: serverId ? () => listReports(queryClient, serverId) : skipToken,
     dataShape: "list",
     staleTimeMs: REPORTS_STALE_TIME_MS,
+    refetchInterval: REPORTS_STALE_TIME_MS,
   });
   return setting.enabled ? (query.data ?? NO_REPORTS) : NO_REPORTS;
 }

packages/server/scripts/dictation-generate-baseline.ts

@@ -33,9 +33,9 @@ async function main(): Promise<void> {
 
   const apiKey = requireEnv("OPENAI_API_KEY");
   const transcriptionModel = process.env.STT_MODEL ?? "gpt-4o-transcribe";
-  // Mirrors the daemon: no built-in prompt, only an explicit override. Whisper
-  // treats `prompt` as preceding transcript, so a default would bias the baseline.
-  const prompt = process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT?.trim() || undefined;
+  const prompt =
+    process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT ??
+    "Transcribe only what the speaker says. Do not add words. Preserve punctuation and casing. If the audio is silence or non-speech noise, return an empty transcript.";
 
   const openai = new OpenAI({ apiKey });
 
@@ -43,7 +43,7 @@ async function main(): Promise<void> {
     file: await import("node:fs").then((fs) => fs.createReadStream(audioPath)),
     language: "en",
     model: transcriptionModel,
-    ...(prompt ? { prompt } : {}),
+    prompt,
     response_format: "json",
     // Aim for determinism.
     temperature: 0,

packages/server/src/server/dictation/dictation-stream-manager.test.ts

@@ -57,13 +57,11 @@ class FakeRealtimeSession extends EventEmitter implements StreamingTranscription
 class FakeSttProvider implements SpeechToTextProvider {
   public readonly id = "fake";
   public lastLanguage?: string;
-  public lastPrompt?: string;
   constructor(private readonly session: FakeRealtimeSession) {}
   createSession(
     params: Parameters<SpeechToTextProvider["createSession"]>[0],
   ): StreamingTranscriptionSession {
     this.lastLanguage = params.language;
-    this.lastPrompt = params.prompt;
     return this.session;
   }
 }
@@ -596,58 +594,3 @@ it("closes every dictation stream when one provider cleanup fails", async () =>
   expect(second.closed).toBe(true);
   expect(manager.hasDemand).toBe(false);
 });
-
-describe("DictationStreamManager (transcription prompt)", () => {
-  const originalPromptEnv = process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT;
-
-  afterEach(() => {
-    if (originalPromptEnv === undefined) {
-      delete process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT;
-    } else {
-      process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT = originalPromptEnv;
-    }
-  });
-
-  const startDictation = async (
-    options: { language?: string } = {},
-  ): Promise<{ provider: FakeSttProvider; manager: DictationStreamManager }> => {
-    const session = new FakeRealtimeSession();
-    const provider = new FakeSttProvider(session);
-    const manager = new DictationStreamManager({
-      logger: pino({ level: "silent" }),
-      emit: () => {},
-      sessionId: "prompt",
-      stt: provider,
-      ...(options.language ? { language: options.language } : {}),
-    });
-    await manager.handleStart("d1", "audio/pcm;rate=24000;bits=16");
-    manager.cleanupAll();
-    return { provider, manager };
-  };
-
-  it("sends no prompt by default", async () => {
-    delete process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT;
-    const { provider } = await startDictation();
-    expect(provider.lastPrompt).toBeUndefined();
-  });
-
-  it("sends no prompt for non-English dictation by default", async () => {
-    delete process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT;
-    const { provider } = await startDictation({ language: "zh" });
-    expect(provider.lastLanguage).toBe("zh");
-    expect(provider.lastPrompt).toBeUndefined();
-  });
-
-  it("passes an explicit prompt override through", async () => {
-    process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT =
-      "  Vocabulary: Paseo, Dokploy, Kubernetes.  ";
-    const { provider } = await startDictation();
-    expect(provider.lastPrompt).toBe("Vocabulary: Paseo, Dokploy, Kubernetes.");
-  });
-
-  it.each(["", "   "])("treats a blank override (%j) as no prompt", async (value) => {
-    process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT = value;
-    const { provider } = await startDictation();
-    expect(provider.lastPrompt).toBeUndefined();
-  });
-});

packages/server/src/server/dictation/dictation-stream-manager.ts

@@ -38,12 +38,6 @@ function parseNonNegativeNumber(value: string | undefined): number | null {
   return parsed;
 }
 
-/** Unset, empty, and whitespace-only overrides all mean "send no prompt". */
-function readTranscriptionPromptOverride(env: NodeJS.ProcessEnv): string | undefined {
-  const value = env.PASEO_DICTATION_TRANSCRIPTION_PROMPT?.trim();
-  return value ? value : undefined;
-}
-
 function convertPCMToWavBuffer(
   pcmBuffer: Buffer,
   sampleRate: number,
@@ -191,16 +185,16 @@ export class DictationStreamManager {
       return;
     }
 
-    // Whisper uses the prompt as transcript context, not silence instructions.
-    // Keep context opt-in so it does not bias speech or introduce unspoken text.
-    const transcriptionPrompt = readTranscriptionPromptOverride(process.env);
+    const transcriptionPrompt =
+      process.env.PASEO_DICTATION_TRANSCRIPTION_PROMPT ??
+      "Transcribe only what the speaker says. Do not add words. Preserve punctuation and casing. If the audio is silence or non-speech noise, return an empty transcript.";
 
     let stt: ReturnType<SpeechToTextProvider["createSession"]>;
     try {
       stt = sttProvider.createSession({
         logger: this.logger.child({ dictationId }),
         language: this.language,
-        ...(transcriptionPrompt ? { prompt: transcriptionPrompt } : {}),
+        prompt: transcriptionPrompt,
       });
     } catch (error) {
       const message = error instanceof Error ? error.message : String(error);

public-docs/voice.md

@@ -142,19 +142,8 @@ Keys and base URLs follow the fallback rule above.
 - `PASEO_VOICE_LOCAL_STT_MODEL`, `PASEO_VOICE_LOCAL_TTS_MODEL`, local voice STT/TTS model IDs
 - `PASEO_DICTATION_LANGUAGE`, dictation STT language
 - `PASEO_VOICE_LANGUAGE`, voice mode STT language; falls back to `PASEO_DICTATION_LANGUAGE` when unset
-- `PASEO_DICTATION_TRANSCRIPTION_PROMPT`, optional OpenAI dictation transcription prompt (including compatible servers). Unset by default; empty or whitespace-only is treated as unset
 - `PASEO_VOICE_LOCAL_TTS_SPEAKER_ID`, `PASEO_VOICE_LOCAL_TTS_SPEED`, optional local voice TTS tuning
 
-## Transcription Prompt
-
-With the OpenAI provider, including compatible servers, Paseo sends no transcription prompt by default. Whisper-family models use `prompt` as transcript context, so instruction text can distort the transcript or bias its language. Omitting it does not prevent every hallucination on silence or noise.
-
-For OpenAI dictation, set `PASEO_DICTATION_TRANSCRIPTION_PROMPT` to send a prompt explicitly, for example to bias toward project vocabulary. An empty or whitespace-only value means no prompt. Local providers ignore this setting, and voice mode never sends a prompt.
-
-```bash
-PASEO_DICTATION_TRANSCRIPTION_PROMPT="Vocabulary: Paseo, TypeScript."
-```
-
 ## Operational Notes
 
 Voice mode can launch and control agents. Treat voice prompts with the same care as direct agent instructions, especially when specifying working directories or destructive operations.

public-docs/voice.zh-CN.md

@@ -120,19 +120,8 @@ Paseo 在配置的 OpenAI base URL 下使用以下路径：
 - `PASEO_VOICE_LOCAL_STT_MODEL`、`PASEO_VOICE_LOCAL_TTS_MODEL`：本地语音 STT/TTS 模型 ID
 - `PASEO_DICTATION_LANGUAGE`：口述 STT 语言（仅限 OpenAI STT；本地 Parakeet 会忽略）
 - `PASEO_VOICE_LANGUAGE`：语音模式 STT 语言；未设置时退回到 `PASEO_DICTATION_LANGUAGE`（仅限 OpenAI STT；本地 Parakeet 会忽略）
-- `PASEO_DICTATION_TRANSCRIPTION_PROMPT`：可选的 OpenAI 口述转录提示词（包括兼容服务器）。默认不设置；空值或仅含空白字符的值视为未设置
 - `PASEO_VOICE_LOCAL_TTS_SPEAKER_ID`、`PASEO_VOICE_LOCAL_TTS_SPEED`：可选的本地语音 TTS 调优
 
-## 转录提示词 {#transcription-prompt}
-
-使用 OpenAI Provider（包括兼容服务器）进行口述时，Paseo 默认不发送转录提示词。Whisper 系列模型把 `prompt` 当作转录上下文，因此指令文本可能使转录失真或影响输出语言。不发送提示词也无法消除静音或噪声中的所有幻觉。
-
-对于 OpenAI 口述，可以设置 `PASEO_DICTATION_TRANSCRIPTION_PROMPT` 来显式发送提示词，例如提示项目术语。空值或仅含空白字符的值表示不发送提示词。本地 Provider 会忽略此设置，语音模式也不会发送提示词。
-
-```bash
-PASEO_DICTATION_TRANSCRIPTION_PROMPT="Vocabulary: Paseo, TypeScript."
-```
-
 ## 使用须知 {#operational-notes}
 
 语音模式可以启动和控制 Agent。请像对待直接的 Agent 指令一样谨慎对待语音提示词，尤其是在指定工作目录或执行破坏性操作时。