@bojackduy/opencode-voice 0.11.1 → 0.11.3

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/lib/stt.js CHANGED
@@ -5,6 +5,7 @@ import path from "node:path";
5
5
  import os from "node:os";
6
6
  import { spawn, execSync } from "node:child_process";
7
7
  import { getActiveSessionTitle, getRecentConversationContext } from "./session.js";
8
+ import { sttLog } from "./stt-log.js";
8
9
 
9
10
  let sttApiEndpoint = null;
10
11
  let sttApiModel = null;
@@ -337,10 +338,24 @@ export const STREAMING_MODE_DEFAULTS = {
337
338
  cadenceMs: 1000,
338
339
  };
339
340
 
340
- export function resolveSttMode(opts) {
341
+ export function resolveSttMode(opts, kv = null) {
342
+ // kv wins when set: the v2 host has been observed delivering only a
343
+ // subset of file-plugin options, so a persisted choice must override.
344
+ try {
345
+ const saved = kv?.get("stt.mode");
346
+ if (saved === "streaming" || saved === "batch") return saved;
347
+ } catch {}
341
348
  return opts?.sttMode === "streaming" ? "streaming" : "batch";
342
349
  }
343
350
 
351
+ export function resolveSttAutoSubmit(opts, kv = null) {
352
+ try {
353
+ const saved = kv?.get("stt.autoSubmit");
354
+ if (saved === true || saved === false) return saved;
355
+ } catch {}
356
+ return opts?.sttAutoSubmit === true;
357
+ }
358
+
344
359
  export function resolveStreamTimings(opts) {
345
360
  const windowMs =
346
361
  Number(opts?.sttStreamWindowMs) > 0
@@ -427,6 +442,14 @@ const WAVE_LEN = 16;
427
442
  const WAVE_INTERVAL = 35;
428
443
  const WAVE_DURATION = 70;
429
444
  let useWave = true;
445
+ // v2's toast API has no update/replace: every call stacks as a new
446
+ // notification, so the v1 polling pattern becomes a toast storm. v2 shows
447
+ // live state in the slot preview instead and uses one-shot toasts.
448
+ let toastPolling = true;
449
+
450
+ export function __setToastPolling(v) {
451
+ toastPolling = !!v;
452
+ }
430
453
 
431
454
  export function __setRecordingToastFn(fn) {
432
455
  recordingToastFn = fn;
@@ -478,6 +501,7 @@ export function showProcessingToast(message) {
478
501
  variant: "info",
479
502
  duration: PROCESSING_TOAST_DURATION,
480
503
  });
504
+ if (!toastPolling) return;
481
505
  processingToastTimer = setInterval(() => {
482
506
  if (processingToastFn) {
483
507
  processingToastFn({
@@ -579,7 +603,9 @@ export function showStreamingToast(message, { slowAfterMs } = {}) {
579
603
  streamingLastPartialMs = null;
580
604
  if (Number(slowAfterMs) > 0) streamingSlowAfterMs = Math.floor(Number(slowAfterMs));
581
605
  if (!streamingToastFn) return;
606
+ sttLog("toast.streaming", `base=${streamingToastBase} polling=${toastPolling}`);
582
607
  emitStreamingToast();
608
+ if (!toastPolling) return;
583
609
  streamingToastTimer = setInterval(() => {
584
610
  emitStreamingToast();
585
611
  }, STREAMING_TOAST_INTERVAL);
@@ -749,6 +775,63 @@ export function getWaveString() {
749
775
  }
750
776
  }
751
777
 
778
+ // ---- Streaming waveform (live transcript + wave) ----
779
+ // The batch meter samples a WAV file on disk; streaming keeps PCM in memory,
780
+ // so the controller exposes its newest tail instead. Same thresholds and
781
+ // noise-floor gating as the batch meter, with slightly slower decay because
782
+ // this samples at ~80 ms rather than ~35 ms.
783
+
784
+ const STREAM_WAVE_BYTES = 2560; // ~80 ms of 16 kHz mono 16-bit PCM
785
+ let streamWavePrev = [];
786
+ let streamNoiseFloor = null;
787
+
788
+ export function __resetStreamWave() {
789
+ streamWavePrev = [];
790
+ streamNoiseFloor = null;
791
+ }
792
+
793
+ export function waveLevelsFromPcm(pcm) {
794
+ if (!pcm || pcm.length < 2) return Array(WAVE_LEN).fill(0);
795
+ const chunkBytes = Math.max(2, Math.floor(pcm.length / WAVE_LEN));
796
+ const raw = [];
797
+ for (let column = 0; column < WAVE_LEN; column++) {
798
+ const offset = column * chunkBytes;
799
+ if (offset >= pcm.length) {
800
+ raw.push(0);
801
+ continue;
802
+ }
803
+ const chunk = pcm.subarray(offset, Math.min(offset + chunkBytes, pcm.length));
804
+ const samples = Math.floor(chunk.length / 2);
805
+ if (samples === 0) {
806
+ raw.push(0);
807
+ continue;
808
+ }
809
+ let sum = 0;
810
+ for (let i = 0; i + 1 < chunk.length; i += 2) {
811
+ const sample = chunk.readInt16LE(i) / 32768;
812
+ sum += sample * sample;
813
+ }
814
+ raw.push(Math.sqrt(sum / samples));
815
+ }
816
+ const floor = Math.min(...raw);
817
+ if (streamNoiseFloor === null) streamNoiseFloor = floor;
818
+ else streamNoiseFloor = Math.min(streamNoiseFloor, floor) * 0.995 + floor * 0.005;
819
+ const levels = raw.map((rms) => {
820
+ const effective = Math.max(0, rms - streamNoiseFloor * 1.15);
821
+ if (effective < 0.005) return 0;
822
+ return rmsToLevel(effective * 1.6);
823
+ });
824
+ if (streamWavePrev.length !== WAVE_LEN) streamWavePrev = Array(WAVE_LEN).fill(0);
825
+ const smoothed = levels.map((level, index) => {
826
+ const prev = streamWavePrev[index] || 0;
827
+ if (level > prev) return level;
828
+ const decayed = Math.floor(prev * 0.35);
829
+ return level > decayed ? level : decayed;
830
+ });
831
+ streamWavePrev = smoothed;
832
+ return smoothed;
833
+ }
834
+
752
835
  let currentRecordKeybind = "<leader>[";
753
836
  export function __setRecordKeybind(kb) {
754
837
  if (kb) currentRecordKeybind = kb.replace(/^<leader>/, "leader+");
@@ -767,10 +850,14 @@ export function __setStopHint(hint) {
767
850
  savedRecordKeybind = null;
768
851
  }
769
852
  }
770
- export function buildRecordingMessage() {
853
+ function recordingModeSuffix() {
771
854
  // When the conversation loop overrides the stop hint, the toast says so
772
855
  // explicitly - otherwise users wonder why it names a key they did not press.
773
- const mode = savedRecordKeybind !== null ? " (conversation)" : "";
856
+ return savedRecordKeybind !== null ? " (conversation)" : "";
857
+ }
858
+
859
+ export function buildRecordingMessage() {
860
+ const mode = recordingModeSuffix();
774
861
  if (!useWave || recordingStartMs === null)
775
862
  return `${RECORDING_TOAST_BASE}${mode} · ${currentRecordKeybind} to stop`;
776
863
  const elapsed = formatElapsed(Date.now() - recordingStartMs);
@@ -781,10 +868,19 @@ export function buildRecordingMessage() {
781
868
  export function showRecordingToast() {
782
869
  clearRecordingToast();
783
870
  if (!recordingToastFn) return;
871
+ sttLog("toast.recording", `polling=${toastPolling} wave=${useWave}`);
784
872
  recordingStartMs = Date.now();
785
873
  prevWaveLevels = Array(WAVE_LEN).fill(0);
786
874
  noiseFloorRms = null;
787
875
  calibrationEndMs = Date.now() + 120;
876
+ if (!toastPolling) {
877
+ recordingToastFn({
878
+ message: `${RECORDING_TOAST_BASE}${recordingModeSuffix()} · ${currentRecordKeybind} to stop`,
879
+ variant: "info",
880
+ duration: RECORDING_TOAST_DURATION,
881
+ });
882
+ return;
883
+ }
788
884
  const interval = useWave ? WAVE_INTERVAL : RECORDING_TOAST_INTERVAL;
789
885
  const duration = useWave ? WAVE_DURATION : RECORDING_TOAST_DURATION;
790
886
  recordingToastFn({
@@ -1376,9 +1472,11 @@ async function appendTranscription(client, renderer, text, submit, isCurrent) {
1376
1472
  return true;
1377
1473
  }
1378
1474
 
1379
- // Only the submit-only adapter consumes this fence; v1 request bodies
1380
- // remain unchanged. Re-check after an asynchronous home session create.
1381
- const fence = client.tui.submitOnly ? { isCurrent } : {};
1475
+ // Only the v2 adapter consumes this fence; v1 request bodies remain
1476
+ // unchanged. Re-check after an asynchronous home session create. The
1477
+ // `submit` flag lets the adapter distinguish an explicit submit from a
1478
+ // plain append with no editable field focused (append must not send).
1479
+ const fence = client.tui?.v2 ? { isCurrent, submit } : {};
1382
1480
  let appendResult = await client.tui.appendPrompt({ body: { text }, ...fence });
1383
1481
 
1384
1482
  if (appendResult?.error?.data?.message === "Expected object, got undefined") {
@@ -1391,7 +1489,9 @@ async function appendTranscription(client, renderer, text, submit, isCurrent) {
1391
1489
  );
1392
1490
  }
1393
1491
 
1394
- if (submit) {
1492
+ if (appendResult?.retained) return appendResult;
1493
+
1494
+ if (submit && !appendResult?.submitted) {
1395
1495
  await client.tui.submitPrompt();
1396
1496
  }
1397
1497
  return appendResult;
@@ -1515,15 +1615,15 @@ async function appendTurnText(client, renderer, toast, text, logger) {
1515
1615
  );
1516
1616
  clearProcessingToast();
1517
1617
  logger?.log("STT", `Turn appended chars=${text.length}`, "debug");
1518
- // Submit-only hosts (OpenCode v2: no append-to-composer API) submit via
1519
- // the client shim, which marks the result — so the toast must not claim
1520
- // the text was "added to prompt" for editing there. v1 results carry no
1618
+ // The v2 adapter retains the transcript when no editable field has
1619
+ // focus — it never sends an unrequested prompt. v1 results carry no
1521
1620
  // marker and keep the original toast.
1621
+ const retained = appendResult?.retained === true;
1522
1622
  toast(
1523
- appendResult?.submitted === true
1524
- ? "Transcription submitted as-is (submit-only mode)"
1623
+ retained
1624
+ ? "No text field focused - transcription kept (not sent)"
1525
1625
  : "Transcription added to prompt",
1526
- "success",
1626
+ retained ? "warning" : "success",
1527
1627
  );
1528
1628
  return { text };
1529
1629
  } catch (err) {
@@ -1601,6 +1701,17 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1601
1701
  const client = api.client;
1602
1702
  const renderer = opts?.focusMode === "primary" ? null : api.renderer;
1603
1703
  const preview = api.voicePreview;
1704
+ __setToastPolling(!preview);
1705
+ const bootMode = resolveSttMode(opts, kv);
1706
+ let modeSource = "default-batch";
1707
+ try {
1708
+ if (kv?.get("stt.mode") === bootMode) modeSource = "kv";
1709
+ else if (opts?.sttMode === bootMode) modeSource = "options";
1710
+ } catch {}
1711
+ sttLog(
1712
+ "register",
1713
+ `sttMode=${bootMode} source=${modeSource} preview=${preview ? "yes" : "no"} toastPolling=${!preview} optKeys=${Object.keys(opts || {}).join(",")}`,
1714
+ );
1604
1715
  let streamGeneration = 0;
1605
1716
  let streamDisposed = false;
1606
1717
  let finalizingController = null;
@@ -1692,7 +1803,30 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1692
1803
  }
1693
1804
 
1694
1805
  // ---- Streaming dictation session (stage 2, local-only, no LLM) ----
1695
- const sttMode = resolveSttMode(opts);
1806
+ // Mode/auto-submit resolve live from kv first: options seed kv on first
1807
+ // boot, then the /stt-mode and /stt-autosubmit commands own the runtime
1808
+ // value. This survives hosts that deliver partial plugin options.
1809
+ const seedKvFromOpts = () => {
1810
+ try {
1811
+ if (
1812
+ (opts?.sttMode === "streaming" || opts?.sttMode === "batch") &&
1813
+ kv.get("stt.mode") == null
1814
+ ) {
1815
+ kv.set("stt.mode", opts.sttMode);
1816
+ }
1817
+ } catch {}
1818
+ try {
1819
+ if (opts?.sttAutoSubmit !== undefined && kv.get("stt.autoSubmit") == null) {
1820
+ kv.set("stt.autoSubmit", opts.sttAutoSubmit === true);
1821
+ }
1822
+ } catch {}
1823
+ };
1824
+ seedKvFromOpts();
1825
+ const currentSttMode = () => resolveSttMode(opts, kv);
1826
+ const sttAutoSubmitEnabled = () => resolveSttAutoSubmit(opts, kv);
1827
+ // Boot-time snapshot for the mode log line below; behavior branches call
1828
+ // currentSttMode() live so /stt-mode takes effect without a restart.
1829
+ const sttMode = currentSttMode();
1696
1830
  const streamTimings = resolveStreamTimings(opts);
1697
1831
  logger?.log(
1698
1832
  "STT",
@@ -1793,6 +1927,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1793
1927
  });
1794
1928
  if (preview) await preview.begin();
1795
1929
  else editor.begin(target);
1930
+ sttLog("stream.begin", `preview=${preview ? "yes" : "no"}`);
1796
1931
  if (streamDisposed || myGen !== streamGeneration) {
1797
1932
  preview?.clear();
1798
1933
  return false;
@@ -1806,8 +1941,13 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1806
1941
  try {
1807
1942
  ed.applyPartial(p);
1808
1943
  } catch {}
1944
+ sttLog(
1945
+ "stream.partial",
1946
+ `stable=${(p?.stableText || "").length} tentative=${(p?.tentativeText || "").length}`,
1947
+ );
1809
1948
  },
1810
1949
  onStateChange: (s) => {
1950
+ sttLog("stream.state", s);
1811
1951
  try {
1812
1952
  if (s === "streaming") {
1813
1953
  updateStreamingToast("● Streaming dictation — listening");
@@ -1861,6 +2001,21 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1861
2001
  return false;
1862
2002
  }
1863
2003
  __setStreamingActive(true);
2004
+ sttLog("stream.started", `reusedWarm=${reusedWarm}`);
2005
+ // Live waveform rides the v2 preview slot next to the live transcript.
2006
+ // v1 has no persistent preview surface; its batch wave is unchanged.
2007
+ if (preview) {
2008
+ __resetStreamWave();
2009
+ let pcmLogCount = 0;
2010
+ preview.startWave?.(() => {
2011
+ const pcm = controller.peekPcm?.(STREAM_WAVE_BYTES);
2012
+ if (pcmLogCount < 3 && pcm) {
2013
+ pcmLogCount += 1;
2014
+ sttLog("stream.pcm", `bytes=${pcm.length}`);
2015
+ }
2016
+ return pcm ? waveLevelsFromPcm(pcm) : null;
2017
+ });
2018
+ }
1864
2019
  // Sticky, never silent: covers the cold model load ("Loading…") until
1865
2020
  // the controller reports streaming, then live capture with elapsed
1866
2021
  // time. Partial preview toasts continue alongside via the editor.
@@ -1871,6 +2026,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1871
2026
  return true;
1872
2027
  } catch (err) {
1873
2028
  preview?.clear();
2029
+ sttLog("stream.start-failed", err?.message || String(err));
1874
2030
  logger?.log("STT", `Streaming start failed: ${err?.message || err}`, "error");
1875
2031
  try {
1876
2032
  clearStreamingToast();
@@ -1907,6 +2063,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1907
2063
  // is at fault, on model/language change, or on plugin unload.
1908
2064
  async function finalizeStreamingDictation({ submit = false } = {}) {
1909
2065
  if (!streamingDictationActive || !streamController) return null;
2066
+ sttLog("stream.finalize", `submit=${submit}`);
1910
2067
  const controller = streamController;
1911
2068
  const editor = streamEditor;
1912
2069
  const target = streamTarget;
@@ -1984,18 +2141,9 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
1984
2141
  retainWarmStreamController(controller);
1985
2142
  return { text: "" };
1986
2143
  }
1987
- if (preview) {
1988
- // v2 has a visible preview but no composer append. Finalize submits
1989
- // once through the same home/create/cancellation fence as batch STT.
1990
- processing = true;
1991
- const result = await submitTurnText(client, null, toast, text, logger);
1992
- if (!streamDisposed && myGen === streamGeneration) retainWarmStreamController(controller);
1993
- else await controller.dispose?.();
1994
- return { ...result, submitted: !result.error };
1995
- }
1996
2144
  if (fin?.inserted) {
1997
2145
  logger?.log("STT", `Streaming finalized via owned range chars=${text.length}`, "debug");
1998
- if (submit || autoSubmit) {
2146
+ if (submit || autoSubmitNow()) {
1999
2147
  if (target && typeof target.submit === "function") {
2000
2148
  try {
2001
2149
  target.submit();
@@ -2009,6 +2157,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2009
2157
  } else {
2010
2158
  toast("Streaming transcription added", "success");
2011
2159
  }
2160
+ if (preview) processing = false;
2012
2161
  retainWarmStreamController(controller);
2013
2162
  return { text, inserted: true };
2014
2163
  }
@@ -2029,11 +2178,12 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2029
2178
  if (!inserted) {
2030
2179
  logger?.log("STT", "Streaming finalize: no editable field, transcript retained", "warn");
2031
2180
  toast("No editable field - dictated text kept (not inserted elsewhere)", "warning");
2181
+ if (preview) processing = false;
2032
2182
  retainWarmStreamController(controller);
2033
2183
  return { text, inserted: false };
2034
2184
  }
2035
2185
  logger?.log("STT", `Streaming finalized chars=${text.length}`, "debug");
2036
- if (submit || autoSubmit) {
2186
+ if (submit || autoSubmitNow()) {
2037
2187
  if (target && typeof target.submit === "function") {
2038
2188
  try {
2039
2189
  target.submit();
@@ -2047,6 +2197,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2047
2197
  } else {
2048
2198
  toast("Streaming transcription added", "success");
2049
2199
  }
2200
+ if (preview) processing = false;
2050
2201
  retainWarmStreamController(controller);
2051
2202
  return { text, inserted: true };
2052
2203
  }
@@ -2057,6 +2208,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2057
2208
  // next dictation (user cancel is per-dictation, never a server fault).
2058
2209
  async function cancelStreamingDictation() {
2059
2210
  streamGeneration += 1;
2211
+ sttLog("stream.cancel");
2060
2212
  preview?.clear();
2061
2213
  if (preview && processing) cancelRecording(logger);
2062
2214
  if (finalizingController) {
@@ -2136,7 +2288,12 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2136
2288
  cancel: () => cancelRecording(logger),
2137
2289
  discard: () => discardTurn(),
2138
2290
  transcribeTurn: () => transcribeTurn(kv, complete, client, api, toast, systemPrompt, logger),
2139
- submitTurnText: (text) => submitTurnText(client, renderer, toast, text, logger),
2291
+ // Conversation turns are hands-free submissions, not dictation: on v2
2292
+ // they must not press Enter in whatever field happens to have focus (a
2293
+ // permission dialog could own it), so they always go through the
2294
+ // session.prompt submit path. v1 keeps its focused-insert + submit.
2295
+ submitTurnText: (text) =>
2296
+ submitTurnText(client, client.tui?.v2 ? null : renderer, toast, text, logger),
2140
2297
  startStreaming: startStreamingDictation,
2141
2298
  finalizeStreaming: finalizeStreamingDictation,
2142
2299
  cancelStreaming: cancelStreamingDictation,
@@ -2144,7 +2301,9 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2144
2301
 
2145
2302
  // One-shot auto-submit: speak, press again, and the text submits itself
2146
2303
  // without landing in the box first. Off by default (append-then-edit).
2147
- const autoSubmit = opts?.sttAutoSubmit === true;
2304
+ // Live reader: /stt-autosubmit flips the kv value mid-session.
2305
+ const autoSubmit = sttAutoSubmitEnabled();
2306
+ const autoSubmitNow = () => sttAutoSubmitEnabled();
2148
2307
 
2149
2308
  const commands = [
2150
2309
  {
@@ -2164,7 +2323,8 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2164
2323
  onSelect() {
2165
2324
  // Streaming mode: toggle streaming record/finalize. Batch default
2166
2325
  // below is behavior-identical to before.
2167
- if (sttMode === "streaming") {
2326
+ sttLog("toggle", `sttMode=${currentSttMode()} streamingActive=${streamingDictationActive}`);
2327
+ if (currentSttMode() === "streaming") {
2168
2328
  if (streamingDictationActive) {
2169
2329
  void finalizeStreamingDictation({ submit: false });
2170
2330
  return;
@@ -2231,7 +2391,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2231
2391
  onSelect() {
2232
2392
  // Streaming mode: finalize then submit ONLY via the captured
2233
2393
  // target's own submit(). Never falls through to primary-chat submit.
2234
- if (sttMode === "streaming") {
2394
+ if (currentSttMode() === "streaming") {
2235
2395
  if (!streamingDictationActive) {
2236
2396
  toast("No streaming dictation in progress", "warning");
2237
2397
  return;
@@ -2283,7 +2443,7 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2283
2443
  onSelect() {
2284
2444
  // Streaming mode: cancel and remove ONLY plugin-owned dictation
2285
2445
  // text, never user-typed text.
2286
- if (sttMode === "streaming" || streamingDictationActive) {
2446
+ if (currentSttMode() === "streaming" || streamingDictationActive) {
2287
2447
  if (streamingDictationActive) {
2288
2448
  void cancelStreamingDictation();
2289
2449
  return;
@@ -2313,6 +2473,57 @@ export function registerSTT(api, kv, complete, prompts, opts, logger, deps = {})
2313
2473
  }
2314
2474
  },
2315
2475
  },
2476
+ {
2477
+ title: "STT: select dictation mode",
2478
+ value: "stt.mode",
2479
+ category: "opencode-voice",
2480
+ description: "Choose one-shot batch or live streaming dictation (saved, survives restarts)",
2481
+ slash: { name: "stt-mode" },
2482
+ onSelect() {
2483
+ const current = currentSttMode();
2484
+ api.ui.dialog.replace(() =>
2485
+ api.ui.DialogSelect({
2486
+ title: "Select dictation mode",
2487
+ current,
2488
+ options: [
2489
+ {
2490
+ title: "Streaming (live transcript + wave, local whisper, no LLM)",
2491
+ value: "streaming",
2492
+ onSelect() {
2493
+ kv.set("stt.mode", "streaming");
2494
+ sttLog("mode.set", "streaming via /stt-mode");
2495
+ toast("Dictation mode: streaming");
2496
+ api.ui.dialog.clear();
2497
+ },
2498
+ },
2499
+ {
2500
+ title: "Batch (record, then transcribe)",
2501
+ value: "batch",
2502
+ onSelect() {
2503
+ kv.set("stt.mode", "batch");
2504
+ sttLog("mode.set", "batch via /stt-mode");
2505
+ toast("Dictation mode: batch");
2506
+ api.ui.dialog.clear();
2507
+ },
2508
+ },
2509
+ ],
2510
+ }),
2511
+ );
2512
+ },
2513
+ },
2514
+ {
2515
+ title: "STT: toggle auto-submit",
2516
+ value: "stt.autosubmit",
2517
+ category: "opencode-voice",
2518
+ description: "Toggle submitting the transcript automatically after dictation (saved)",
2519
+ slash: { name: "stt-autosubmit" },
2520
+ onSelect() {
2521
+ const next = !sttAutoSubmitEnabled();
2522
+ kv.set("stt.autoSubmit", next);
2523
+ sttLog("mode.set", `autoSubmit=${next} via /stt-autosubmit`);
2524
+ toast(`STT auto-submit: ${next ? "on" : "off"}`);
2525
+ },
2526
+ },
2316
2527
  {
2317
2528
  title: sttApiEndpoint ? "STT: select model (API)" : "STT: select model",
2318
2529
  value: "stt.model",