opencode-voice-dictation 0.0.0-stage → 0.1.0

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Fran Carlos
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,3 +1,253 @@
-# Temporary Holding Version
+# Dictado por voz para OpenCode Desktop
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+Plugin de [OpenCode](https://opencode.ai) V2 que añade voz a la app de
+escritorio:
+
+- **Dictado**: pulsas la píldora, hablas y el texto aparece en el prompt.
+- **Lectura**: pulsa «Leer» y reproduce en voz alta la última respuesta.
+
+- **Dictado offline**: usa `System.Speech` de Windows, sin API keys ni internet.
+- **Sin portapapeles**: escribe con `composer.insert()` en la posición del
+  cursor, sin robar el foco ni simular `Ctrl+V`.
+- **Lectura local**: voces del sistema (`speechSynthesis`), con respaldo a
+  `System.Speech.Synthesis` si Electron no expone voces.
+- **Solo Windows**: la app declara `os: ["windows"]`; en otros sistemas aparece
+  como no disponible en vez de fallar.
+
+## Instalación rápida
+
+```bash
+opencode plugin add opencode-voice-dictation
+```
+
+Reinicia OpenCode Desktop. El plugin queda registrado en el `opencode.json`
+global del entorno donde ejecutes el comando.
+
+Actualizar:
+
+```bash
+opencode plugin update opencode-voice-dictation
+```
+
+También puedes instalarlo directamente desde el repositorio (el bundle `dist/`
+está commiteado):
+
+```bash
+opencode plugin add github:evious/opencode-voice-dictation
+```
+
+## Compatibilidad
+
+Probado con OpenCode Desktop **2.2.6** en Windows 11. La API de plugins GUI es
+experimental (`gui-experimental`) y puede cambiar entre versiones de OpenCode.
+
+## Cómo funciona
+
+```
+┌─ Ventana (Solid) ─────────────────────────────┐
+│ ComposerPill 🎤  →  ctx.main(Dictation).start │
+│         ↑                                     │
+│    evento "final"  →  composer.insert(texto)  │
+└───────────────┬───────────────────────────────┘
+                │ Ipc (proceso main de Electron)
+┌───────────────▼───────────────────────────────┐
+│ powershell.exe -File dictation-<hash>.ps1     │
+│   System.Speech + DictationGrammar            │
+│   una línea JSON por frase reconocida         │
+└───────────────────────────────────────────────┘
+```
+
+El motor vive en el proceso `main` de Electron porque necesita
+`node:child_process`. La ventana solo pinta el estado y escribe el texto.
+
+## Requisitos
+
+- OpenCode Desktop en **Windows 10/11**.
+- Node.js 22 o superior para compilar (solo en desarrollo).
+- Un **paquete de idioma de reconocimiento de voz** instalado:
+  Configuración → Hora e idioma → Idioma y región → (idioma) → Opciones →
+  Reconocimiento de voz. No basta con el paquete de visualización.
+- Micrófono predeterminado configurado.
+
+Comprobación rápida en PowerShell:
+
+```powershell
+Add-Type -AssemblyName System.Speech
+[System.Speech.Recognition.SpeechRecognitionEngine]::InstalledRecognizers() |
+  ForEach-Object { $_.Culture.Name }
+```
+
+## Compilar
+
+```bash
+npm install
+npm run build
+```
+
+Genera `dist/gui-experimental.cjs`. Para desarrollo, `npm run dev` reconstruye
+en cada cambio y OpenCode Desktop recarga el plugin solo.
+
+## Instalación manual (desarrollo)
+
+1. Copia al lado de Windows una carpeta con:
+
+   ```
+   voice-dictation/
+   ├── package.json
+   ├── dist/gui-experimental.cjs
+   └── src/index.ts
+   ```
+
+   `node_modules` no hace falta: la parte servidor solo importa tipos y el
+   bundle usa los módulos que la app ya proporciona.
+
+2. Añade la carpeta al `opencode.json` **global** de la app de escritorio
+   (`C:\Users\<usuario>\.config\opencode\opencode.json`):
+
+   ```jsonc
+   {
+     "plugins": ["C:\\Users\\<usuario>\\opencode-plugins\\voice-dictation"]
+   }
+   ```
+
+   La app de escritorio solo lee plugins GUI del `opencode.json` global, no de
+   `cli.json`.
+
+3. Abre **Ajustes → Extensiones → Plugins** y comprueba que aparece
+   `voice-dictation`. Si no aparece, revisa la ruta y los logs de la app.
+
+> Si compilas en WSL, copia la carpeta a una ruta de Windows. La app no puede
+> cargar el bundle desde una ruta de Linux.
+
+## Uso
+
+| Acción | Cómo |
+|---|---|
+| Empezar / parar dictado | Clic en la píldora 🎤 sobre el composer |
+| Atajo de dictado | `Ctrl+Shift+M` (configurable en Ajustes → Atajos) |
+| Comando de dictado | `/dictado` en el prompt |
+| Leer el último mensaje | Clic en la píldora «Leer», o `Ctrl+Shift+L` |
+| Comando de lectura | `/leer` en el prompt |
+| Ajustes | Ajustes → Dictado por voz, o paleta: "Ajustes de dictado por voz" |
+
+La píldora del micrófono muestra el estado: `Dictar`, `Iniciando…`,
+`Escuchando…` (icono en rojo) y `Reintentar dictado` si algo falla.
+
+La píldora «Leer» cambia a `Leyendo…`; un segundo clic detiene la lectura.
+Dictar detiene la lectura y leer detiene el dictado, para que el micrófono no
+capture la voz del altavoz.
+
+El texto se inserta en la posición del cursor. Si ya hay texto y no termina en
+espacio, se añade uno de separación (configurable).
+
+## Ajustes
+
+| Ajuste | Por defecto | Qué hace |
+|---|---|---|
+| Idioma | `es-ES` | Reconocedor instalado en Windows. Si no existe el exacto, usa el mismo idioma base. |
+| Confianza mínima | 0,4 | Descarta resultados con menor confianza para evitar ruido. |
+| Espacio antes del texto | Sí | Separa el dictado de lo que ya hay escrito. |
+| Enfocar el prompt al terminar | Sí | Mueve el cursor al prompt tras cada frase. |
+| Probar micrófono | — | Lista los idiomas instalados y comprueba el micrófono. |
+
+### Lectura en voz alta
+
+| Ajuste | Por defecto | Qué hace |
+|---|---|---|
+| Voz | Automática | Voz del sistema; automática elige una en el idioma de la app. |
+| Velocidad | 1× | 0,8×, 1×, 1,2× o 1,5×. |
+| Ignorar bloques de código | Sí | Lee solo la prosa; omite código, rutas y comandos. |
+| Probar voz | — | Lee una frase corta con la configuración actual. |
+
+La lectura limpia el Markdown antes de hablar: quita fences, backticks,
+negritas, encabezados, viñetas y convierte `[texto](url)` en `texto`. Trocea el
+texto por frases para poder detenerse rápido.
+
+## Permisos
+
+El plugin es local y no hace peticiones de red. Para funcionar:
+
+- Escribe un script `dictation-<hash>.ps1` o `speak-<hash>.ps1` en
+  `%TEMP%\opencode-voice-dictation`.
+- Ejecuta `powershell.exe` con ese script.
+- El script abre el micrófono (`System.Speech`) y, al leer, el altavoz.
+
+Por eso Windows y antivirus como Kaspersky pueden pedir permiso la primera vez:
+es esperado. Elige **Permitir ahora** y marca **Recordar mi elección**. Si
+bloqueas el micrófono, verás el aviso «No hay ningún micrófono disponible».
+
+## Solución de problemas
+
+| Síntoma | Causa / solución |
+|---|---|
+| «El idioma de reconocimiento es-ES no está instalado» | Instala el paquete de **reconocimiento** de voz, no solo el de visualización. El aviso muestra los idiomas disponibles. |
+| «Windows no tiene ningún reconocedor de voz instalado» | Mismo caso: no hay ningún idioma de voz instalado. |
+| «No hay ningún micrófono disponible» | Revisa Configuración → Sistema → Sonido → Entrada y el micrófono predeterminado. |
+| El motor se detiene inesperadamente | Otro proceso puede estar usando el micrófono en modo exclusivo. Ejecuta la prueba de micrófono en Ajustes. |
+| La píldora no aparece | Comprueba que el plugin está en el `opencode.json` global, que la app corre en Windows y que el plugin figura como cargado en Ajustes → Extensiones → Plugins. |
+| No transcribe nada | Sube el nivel de entrada del micrófono y baja la confianza mínima a 0,2. |
+| «Windows no tiene voces instaladas para la lectura» | Instala una voz en Configuración → Hora e idioma → Voz, o revisa que `System.Speech` esté disponible. El plugin usa primero las voces de Chromium y cae a SAPI si no hay. |
+| La lectura se corta o no suena | Prueba «Probar voz» en Ajustes; si suena ahí, el problema es el dispositivo de salida de la app. |
+| «Abre una conversación para leer su último mensaje» | En Home o en un chat nuevo sin mensajes no hay respuesta que leer. |
+
+Los errores del plugin se registran en el log de OpenCode Desktop; los `toast`
+de la app muestran los más importantes.
+
+## Limitaciones
+
+- La API de plugins GUI es **experimental** (`gui-experimental`) y puede
+  cambiar entre versiones de OpenCode.
+- `System.Speech` no añade puntuación ni mayúsculas automáticamente y su
+  precisión es inferior a Whisper o al dictado nativo de Windows 11.
+- No hay resultados parciales: el texto aparece al terminar cada frase.
+- La lectura usa las voces del sistema; si Electron no las expone, el respaldo
+  por PowerShell usa las voces SAPI, con una latencia de arranque de ~0,5 s.
+- La lectura empieza con el mensaje tal como esté en ese momento: si el agente
+  sigue escribiendo, leerá la versión parcial.
+- Una sola dictación a la vez (hay un micrófono). Si empiezas en otra ventana,
+  la anterior se detiene.
+- Si cambias de sesión durante el dictado, el texto se descarta con un aviso
+  para no escribirlo en la sesión equivocada.
+
+El motor está aislado en `src/gui/engines/`: para cambiar a Whisper local o a
+una API en la nube basta con respetar el mismo contrato (`startEngine` y los
+eventos `ready` / `final` / `error`).
+
+## Desarrollo
+
+```
+src/
+├── index.ts                 Parte servidor: fija el id "voice-dictation"
+├── gui.tsx                  Entrada ./gui-experimental
+└── gui/
+    ├── definition.ts        os, stores, i18n es/en
+    ├── contract.ts          Ipc: dictado y lectura
+    ├── window.tsx           Píldoras, comandos e inserción de texto
+    ├── main.ts              Dictado y voz de respaldo
+    ├── speak.ts             Limpieza de Markdown, troceado y speechSynthesis
+    ├── settings.tsx         Página de ajustes
+    └── engines/
+        ├── script.ts        Script de PowerShell para el dictado
+        ├── system-speech.ts Spawn, parseo de JSON, diagnose, taskkill
+        ├── speak-script.ts  Script de PowerShell para la lectura
+        └── sapi-speech.ts   Voz de respaldo y listado de voces SAPI
+```
+
+`npm run typecheck` valida los tipos, `npm run build` genera el bundle CJS y
+`npm test` ejecuta el smoke test (`test/smoke.cjs`): simula el host de la app,
+un `powershell.exe` falso y la API de voz del navegador para comprobar dictado,
+lectura, limpieza de Markdown, cancelación, fallback y errores.
+
+## Publicar
+
+Mantenedores:
+
+```bash
+npm login
+npm version patch   # o minor / major
+npm publish
+git push --tags
+```
+
+El workflow de release publica en npm al empujar un tag `v*`. Configura el
+secreto `NPM_TOKEN` en GitHub, o usa Trusted Publishing con OIDC.

package/dist/gui-experimental.cjs

@@ -0,0 +1,165 @@
+"use strict";Object.defineProperties(exports,{__esModule:{value:!0},[Symbol.toStringTag]:{value:"Module"}});const b=require("@opencode/plugin/gui"),d=require("effect"),A=require("node:child_process"),z=require("node:crypto"),R=require("node:fs"),D=require("node:os"),T=require("node:path"),m=require("solid-js/web"),N=require("solid-js"),X=b.Plugin.define({os:["windows"],stores:{prefs:b.Store.global(d.Schema.Struct({locale:d.Schema.String,minConfidence:d.Schema.Number,trailingSpace:d.Schema.Boolean,autoFocus:d.Schema.Boolean}),{locale:"es-ES",minConfidence:.4,trailingSpace:!0,autoFocus:!0}),speech:b.Store.global(d.Schema.Struct({voiceURI:d.Schema.String,rate:d.Schema.Number,skipCode:d.Schema.Boolean}),{voiceURI:"",rate:1,skipCode:!0})},i18n:{en:{"pill.idle":"Dictate","pill.starting":"Starting…","pill.listening":"Listening…","pill.error":"Retry dictation","pill.title":"Voice dictation","pill.stop":"Stop dictation","command.toggle":"Toggle voice dictation","command.toggle.description":"Transcribe what you say into the prompt","command.configure":"Voice dictation settings","error.title":"Voice dictation","error.language":"The speech language {locale} is not installed. Available: {available}.","error.none":"Windows has no speech recognizer installed. Settings › Time & language › Language › Speech.","error.mic":"No microphone is available for speech recognition.","error.system":"System.Speech is not available in this Windows installation.","error.exit":"The dictation engine stopped unexpectedly.","error.unknown":"Dictation failed ({code}).","toast.discarded":"Dictation discarded: the active session changed.","settings.title":"Voice dictation","settings.description":"Offline dictation with the Windows speech recognizer. The text is inserted into the prompt; it is never sent automatically.","settings.section":"Dictation","settings.language":"Language","settings.language.description":"A speech recognizer installed in Windows. Accents and punctuation depend on it.","settings.confidence":"Minimum confidence","settings.confidence.description":"Results below this threshold are ignored to avoid noise.","confidence.low":"Low (0.2)","confidence.normal":"Normal (0.4)","confidence.high":"High (0.6)","settings.space":"Add a space before the text","settings.space.description":"Separates the dictation from what is already in the prompt.","settings.focus":"Focus the prompt after dictating","settings.focus.description":"Moves the cursor to the prompt so you can review or send it.","settings.test.section":"Diagnostics","settings.test":"Test microphone","settings.test.description":"Checks installed recognizers and the default microphone.","settings.test.run":"Run test","settings.test.success":"Speech recognition ready: {locale}","settings.test.failed":"Not ready: {reason}","common.reason.no-recognizers":"no speech language installed","common.reason.no-microphone":"no microphone available","read.idle":"Read","read.speaking":"Reading…","read.stop":"Stop reading","read.title":"Read the last message aloud","command.read":"Read the last message aloud","command.read.description":"Speaks the last reply with a system voice","read.noSession":"Open a conversation to read its last message.","read.empty":"The last message has no text to read.","read.error":"The last message could not be read.","read.noVoices":"Windows has no voices installed for reading.","speech.section":"Read aloud","speech.voice":"Voice","speech.voice.description":"A system voice. Automatic picks one in the app language.","speech.voice.auto":"Automatic","speech.rate":"Speed","speech.rate.description":"How fast the text is spoken.","speech.rate.slow":"0.8×","speech.rate.normal":"1×","speech.rate.fast":"1.2×","speech.rate.faster":"1.5×","speech.skipCode":"Skip code blocks","speech.skipCode.description":"Read only the prose; code, paths and commands are left out.","speech.preview":"Test voice","speech.preview.description":"Speaks a short sample with the current voice and speed.","speech.preview.text":"Hello, this is a reading test."},es:{"pill.idle":"Dictar","pill.starting":"Iniciando…","pill.listening":"Escuchando…","pill.error":"Reintentar dictado","pill.title":"Dictado por voz","pill.stop":"Detener dictado","command.toggle":"Alternar dictado por voz","command.toggle.description":"Transcribe lo que dices en el prompt","command.configure":"Ajustes de dictado por voz","error.title":"Dictado por voz","error.language":"El idioma de reconocimiento {locale} no está instalado. Disponibles: {available}.","error.none":"Windows no tiene ningún reconocedor de voz instalado. Configuración › Hora e idioma › Idioma › Voz.","error.mic":"No hay ningún micrófono disponible para el reconocimiento de voz.","error.system":"System.Speech no está disponible en esta instalación de Windows.","error.exit":"El motor de dictado se detuvo inesperadamente.","error.unknown":"El dictado falló ({code}).","toast.discarded":"Dictado descartado: la sesión activa cambió.","settings.title":"Dictado por voz","settings.description":"Dictado offline con el reconocedor de voz de Windows. El texto se inserta en el prompt; nunca se envía solo.","settings.section":"Dictado","settings.language":"Idioma","settings.language.description":"Un reconocedor instalado en Windows. Los acentos y la puntuación dependen de él.","settings.confidence":"Confianza mínima","settings.confidence.description":"Los resultados por debajo de este umbral se ignoran para evitar ruido.","confidence.low":"Baja (0,2)","confidence.normal":"Normal (0,4)","confidence.high":"Alta (0,6)","settings.space":"Añadir un espacio antes del texto","settings.space.description":"Separa el dictado de lo que ya hay en el prompt.","settings.focus":"Enfocar el prompt al terminar","settings.focus.description":"Mueve el cursor al prompt para revisar o enviar el texto.","settings.test.section":"Diagnóstico","settings.test":"Probar micrófono","settings.test.description":"Comprueba los reconocedores instalados y el micrófono predeterminado.","settings.test.run":"Ejecutar prueba","settings.test.success":"Reconocimiento de voz listo: {locale}","settings.test.failed":"No está listo: {reason}","common.reason.no-recognizers":"no hay ningún idioma de voz instalado","common.reason.no-microphone":"no hay micrófono disponible","read.idle":"Leer","read.speaking":"Leyendo…","read.stop":"Detener lectura","read.title":"Leer en voz alta el último mensaje","command.read":"Leer en voz alta el último mensaje","command.read.description":"Reproduce la última respuesta con una voz del sistema","read.noSession":"Abre una conversación para leer su último mensaje.","read.empty":"El último mensaje no tiene texto para leer.","read.error":"No se pudo leer el último mensaje.","read.noVoices":"Windows no tiene voces instaladas para la lectura.","speech.section":"Lectura en voz alta","speech.voice":"Voz","speech.voice.description":"Una voz del sistema. Automática elige una en el idioma de la app.","speech.voice.auto":"Automática","speech.rate":"Velocidad","speech.rate.description":"A qué ritmo se lee el texto.","speech.rate.slow":"0,8×","speech.rate.normal":"1×","speech.rate.fast":"1,2×","speech.rate.faster":"1,5×","speech.skipCode":"Ignorar bloques de código","speech.skipCode.description":"Lee solo la prosa; omite código, rutas y comandos.","speech.preview":"Probar voz","speech.preview.description":"Lee una frase corta con la voz y velocidad actuales.","speech.preview.text":"Hola, esta es una prueba de lectura."}}}),O=b.Ipc.define({id:"voice-dictation.dictation",state:d.Schema.Struct({status:d.Schema.Literals(["idle","starting","listening","error"]),error:d.Schema.String,locale:d.Schema.String}),methods:{start:{input:d.Schema.Struct({locale:d.Schema.String,minConfidence:d.Schema.Number})},stop:{},diagnose:{output:d.Schema.Struct({locales:d.Schema.Array(d.Schema.String),microphone:d.Schema.Boolean,selected:d.Schema.String,error:d.Schema.String})}},events:{final:d.Schema.Struct({text:d.Schema.String,confidence:d.Schema.Number}),error:d.Schema.Struct({code:d.Schema.String,detail:d.Schema.String,available:d.Schema.Array(d.Schema.String)})}}),V=b.Ipc.define({id:"voice-dictation.speech",methods:{speak:{input:d.Schema.Struct({text:d.Schema.String,rate:d.Schema.Number,voice:d.Schema.String})},stop:{},voices:{output:d.Schema.Array(d.Schema.String)}},events:{done:d.Schema.Struct({ok:d.Schema.Boolean})}}),F=String.raw`param(
+  [string]$Culture = "es-ES",
+  [double]$MinConfidence = 0.4,
+  [switch]$Diagnose
+)
+
+$ErrorActionPreference = "Stop"
+
+try {
+  $utf8 = New-Object System.Text.UTF8Encoding($false)
+  [Console]::OutputEncoding = $utf8
+} catch { }
+
+function Emit {
+  param([hashtable]$Payload)
+  [Console]::Out.WriteLine(($Payload | ConvertTo-Json -Compress))
+  [Console]::Out.Flush()
+}
+
+try {
+  Add-Type -AssemblyName System.Speech
+} catch {
+  Emit @{ type = "error"; code = "SYSTEM_SPEECH_UNAVAILABLE"; detail = "$($_.Exception.Message)"; available = @() }
+  exit 2
+}
+
+$installed = @()
+try {
+  foreach ($entry in [System.Speech.Recognition.SpeechRecognitionEngine]::InstalledRecognizers()) {
+    $installed += $entry.Culture.Name
+  }
+} catch { }
+
+$selected = ""
+foreach ($name in $installed) {
+  if ($name -eq $Culture) { $selected = $name; break }
+}
+if ($selected -eq "") {
+  $base = ($Culture -split "-")[0]
+  foreach ($name in $installed) {
+    if (($name -split "-")[0] -eq $base) { $selected = $name; break }
+  }
+}
+
+if ($Diagnose) {
+  $microphone = $false
+  $problem = ""
+  if ($installed.Count -gt 0) {
+    try {
+      $probeCulture = New-Object System.Globalization.CultureInfo($installed[0])
+      $probe = New-Object System.Speech.Recognition.SpeechRecognitionEngine($probeCulture)
+      $probe.SetInputToDefaultAudioDevice()
+      $microphone = $true
+      $probe.Dispose()
+    } catch {
+      $problem = "$($_.Exception.Message)"
+    }
+  } else {
+    $problem = "NO_RECOGNIZERS"
+  }
+  Emit @{ type = "diagnose"; locales = @($installed); microphone = $microphone; selected = $selected; error = $problem }
+  exit 0
+}
+
+if ($installed.Count -eq 0) {
+  Emit @{ type = "error"; code = "NO_RECOGNIZERS"; detail = ""; available = @() }
+  exit 3
+}
+
+if ($selected -eq "") {
+  Emit @{ type = "error"; code = "LANGUAGE_NOT_AVAILABLE"; detail = $Culture; available = @($installed) }
+  exit 4
+}
+
+try {
+  $cultureInfo = New-Object System.Globalization.CultureInfo($selected)
+  $recognizer = New-Object System.Speech.Recognition.SpeechRecognitionEngine($cultureInfo)
+  $recognizer.LoadGrammar((New-Object System.Speech.Recognition.DictationGrammar))
+  $recognizer.InitialSilenceTimeout = [TimeSpan]::FromSeconds(10)
+  $recognizer.BabbleTimeout = [TimeSpan]::FromSeconds(5)
+  $recognizer.EndSilenceTimeout = [TimeSpan]::FromSeconds(0.6)
+  $recognizer.EndSilenceTimeoutAmbiguous = [TimeSpan]::FromSeconds(1.2)
+  $recognizer.SetInputToDefaultAudioDevice()
+} catch {
+  Emit @{ type = "error"; code = "MICROPHONE_UNAVAILABLE"; detail = "$($_.Exception.Message)"; available = @($installed) }
+  exit 5
+}
+
+Emit @{ type = "ready"; locale = $selected; requested = $Culture }
+
+$failures = 0
+while ($true) {
+  try {
+    $result = $recognizer.Recognize([TimeSpan]::FromSeconds(5))
+    $failures = 0
+    if ($result -ne $null -and $result.Text -ne "" -and $result.Confidence -ge $MinConfidence) {
+      Emit @{ type = "final"; text = $result.Text; confidence = [math]::Round($result.Confidence, 3); locale = $selected }
+    }
+  } catch {
+    $failures++
+    if ($failures -ge 5) {
+      Emit @{ type = "error"; code = "RECOGNITION_FAILED"; detail = "$($_.Exception.Message)"; available = @() }
+      exit 6
+    }
+    Start-Sleep -Milliseconds 250
+  }
+}
+`,J="powershell.exe";function U(){const e=T.join(D.tmpdir(),"opencode-voice-dictation");R.mkdirSync(e,{recursive:!0});const t=z.createHash("sha1").update(F).digest("hex").slice(0,12),i=T.join(e,`dictation-${t}.ps1`);return R.writeFileSync(i,F,"utf8"),i}function j(e){return A.spawn(J,e,{windowsHide:!0,stdio:["ignore","pipe","pipe"]})}function B(e){const t=e.pid;if(t===void 0){e.kill();return}A.spawn("taskkill",["/PID",String(t),"/T","/F"],{windowsHide:!0,stdio:"ignore"}).on("error",()=>e.kill())}function K(e,t){const i=j(["-NoProfile","-NonInteractive","-ExecutionPolicy","Bypass","-File",U(),"-Culture",e.locale,"-MinConfidence",String(e.minConfidence)]);let c=!1,r="",s="";const g=o=>{switch(o.type){case"ready":t.onReady(o.locale??e.locale);break;case"final":o.text&&t.onFinal(o.text,o.confidence??0);break;case"error":t.onError(o.code??"UNKNOWN",o.detail??"",o.available??[]);break}};return i.stdout?.setEncoding("utf8"),i.stdout?.on("data",o=>{s+=o;let a=s.indexOf(`
+`);for(;a>=0;){const l=s.slice(0,a).replace(/\r$/,"").trim();if(s=s.slice(a+1),l)try{g(JSON.parse(l))}catch{}a=s.indexOf(`
+`)}}),i.stderr?.setEncoding("utf8"),i.stderr?.on("data",o=>{r=(r+o).slice(-2e3)}),i.on("error",o=>{c||t.onError("ENGINE_START_FAILED",o.message,[])}),i.on("close",o=>{c||t.onExit(o,r.trim().slice(-500))}),{stop(){c=!0,B(i)}}}function Z(){return new Promise((e,t)=>{const i=j(["-NoProfile","-NonInteractive","-ExecutionPolicy","Bypass","-File",U(),"-Diagnose"]);let c="",r="",s=!1;const g=(a,l)=>{s||(s=!0,clearTimeout(o),a?t(a):e(l))},o=setTimeout(()=>{B(i),g(new Error("diagnose timed out"))},2e4);i.stdout?.setEncoding("utf8"),i.stdout?.on("data",a=>{c+=a}),i.stderr?.setEncoding("utf8"),i.stderr?.on("data",a=>{r+=a}),i.on("error",a=>g(a)),i.on("close",()=>{const a=c.split(/\r?\n/).map(l=>l.trim()).filter(l=>l.startsWith("{")).at(-1);if(!a){g(new Error(r.trim()||"no output from PowerShell"));return}try{const l=JSON.parse(a);g(void 0,{locales:l.locales??[],microphone:l.microphone===!0,selected:l.selected??"",error:l.error??""})}catch(l){g(l instanceof Error?l:new Error(String(l)))}})})}const P=String.raw`param(
+  [string]$TextFile = "",
+  [int]$Rate = 0,
+  [string]$Voice = "",
+  [switch]$ListVoices
+)
+
+$ErrorActionPreference = "Stop"
+
+try {
+  $utf8 = New-Object System.Text.UTF8Encoding($false)
+  [Console]::OutputEncoding = $utf8
+} catch { }
+
+try {
+  Add-Type -AssemblyName System.Speech
+} catch {
+  exit 2
+}
+
+try {
+  $synth = New-Object System.Speech.Synthesis.SpeechSynthesizer
+} catch {
+  exit 3
+}
+
+if ($ListVoices) {
+  try {
+    foreach ($entry in $synth.GetInstalledVoices()) {
+      if ($entry.Enabled) {
+        [Console]::Out.WriteLine($entry.VoiceInfo.Name)
+      }
+    }
+  } catch { }
+  $synth.Dispose()
+  exit 0
+}
+
+if ($Rate -lt -10) { $Rate = -10 }
+if ($Rate -gt 10) { $Rate = 10 }
+
+try { $synth.Rate = $Rate } catch { }
+
+if ($Voice -ne "") {
+  try { $synth.SelectVoice($Voice) } catch { }
+}
+
+try {
+  $text = [System.IO.File]::ReadAllText($TextFile, [System.Text.Encoding]::UTF8)
+  $synth.Speak($text)
+} catch { }
+
+$synth.Dispose()
+`,Y="powershell.exe";function G(){const e=T.join(D.tmpdir(),"opencode-voice-dictation");return R.mkdirSync(e,{recursive:!0}),e}function M(){const e=z.createHash("sha1").update(P).digest("hex").slice(0,12),t=T.join(G(),`speak-${e}.ps1`);return R.writeFileSync(t,P,"utf8"),t}function Q(e){const t=T.join(G(),`speech-${z.randomUUID()}.txt`);return R.writeFileSync(t,e,"utf8"),t}function q(e){return A.spawn(Y,e,{windowsHide:!0,stdio:["ignore","pipe","pipe"]})}function H(e){const t=e.pid;if(t===void 0){e.kill();return}A.spawn("taskkill",["/PID",String(t),"/T","/F"],{windowsHide:!0,stdio:"ignore"}).on("error",()=>e.kill())}function x(e){return Math.max(-10,Math.min(10,Math.round((e-1)*5)))}function ee(e,t){const i=Q(e.text),c=q(["-NoProfile","-NonInteractive","-ExecutionPolicy","Bypass","-File",M(),"-TextFile",i,"-Rate",String(x(e.rate)),"-Voice",e.voice]);let r=!1;const s=()=>{try{R.rmSync(i,{force:!0})}catch{}};return c.on("error",()=>{s(),r||t(!1)}),c.on("close",g=>{s(),r||t(g===0)}),{stop(){r=!0,H(c),s()}}}function te(){return new Promise(e=>{const t=q(["-NoProfile","-NonInteractive","-ExecutionPolicy","Bypass","-File",M(),"-ListVoices"]);let i="",c=!1;const r=g=>{c||(c=!0,clearTimeout(s),e(g))},s=setTimeout(()=>{H(t),r([])},1e4);t.stdout?.setEncoding("utf8"),t.stdout?.on("data",g=>{i+=g}),t.on("error",()=>r([])),t.on("close",()=>{const g=i.split(/\r?\n/).map(o=>o.trim()).filter(o=>o.length>0);r(g)})})}const ie=e=>{let t="idle",i="",c="",r,s=!1;const g=()=>{r=void 0,s=!1},o=e.provide(O,{state:()=>({status:t,error:i,locale:c}),start(S,w){r&&(s=!0,r.stop(),g()),t="starting",i="",c=S.locale,o.changed();const p=w.window;try{r=K({locale:S.locale,minConfidence:S.minConfidence},{onReady(u){t="listening",c=u,o.changed()},onFinal(u,h){o.emit("final",{text:u,confidence:h},p)},onError(u,h,C){t="error",i=u,o.changed(),o.emit("error",{code:u,detail:h,available:C},p)},onExit(u,h){if(s){g(),t="idle",i="",c="",o.changed();return}g(),t!=="error"&&(t="error",i=u===null?"ENGINE_EXIT":`ENGINE_EXIT_${u}`,o.changed(),o.emit("error",{code:"ENGINE_EXIT",detail:h||(u===null?"":String(u)),available:[]},p))}})}catch(u){throw t="error",i="ENGINE_START_FAILED",o.changed(),u}},stop(){r&&(s=!0,r.stop(),g()),t="idle",i="",c="",o.changed()},diagnose(){return Z()}});let a;const l=e.provide(V,{speak(S){a?.stop(),a=ee(S,w=>{a=void 0,l.emit("done",{ok:w})})},stop(){a&&(a.stop(),a=void 0),l.emit("done",{ok:!0})},voices(){return te()}});e.onCleanup(()=>{r&&(s=!0,r.stop(),g()),a?.stop(),a=void 0})},ne=250;function oe(e,t){let i=e.replace(/\r\n/g,`
+`);return t?i=i.replace(/```[\s\S]*?```/g," "):i=i.replace(/```[^\n]*\n([\s\S]*?)```/g,"$1"),i=i.replace(/`([^`]+)`/g,"$1"),i=i.replace(/!\[[^\]]*\]\([^)]*\)/g," "),i=i.replace(/\[([^\]]+)\]\([^)]*\)/g,"$1"),i=i.replace(/^\s*#{1,6}\s+/gm,""),i=i.replace(/^\s*>\s?/gm,""),i=i.replace(/^\s*[-*+]\s+/gm,""),i=i.replace(/^\s*\d+\.\s+/gm,""),i=i.replace(/(\*\*|__)(.*?)\1/g,"$2"),i=i.replace(/(\*|_)(.*?)\1/g,"$2"),i=i.replace(/^\s*\|.*\|\s*$/gm," "),i=i.replace(/\n{2,}/g,". "),i=i.replace(/\n/g," "),i=i.replace(/\s+/g," "),i=i.replace(/\s+([.,;:!?])/g,"$1"),i=i.replace(/([.,;:!?]){2,}/g,"$1"),i.trim()}function re(e,t=ne){const i=e.trim();if(!i)return[];const c=i.split(new RegExp("(?<=[.!?…])\\s+")),r=[];let s="";const g=o=>{let a="";for(const l of o.split(/\s+/))l&&(a&&a.length+1+l.length>t?(r.push(a),a=l):a=a?`${a} ${l}`:l);return a};for(const o of c){if(o.length>t){s&&(r.push(s),s=""),s=g(o);continue}s&&s.length+1+o.length>t?(r.push(s),s=o):s=s?`${s} ${o}`:o}return s&&r.push(s),r}function se(e,t){const i=e.stores.speech,[c,r]=N.createSignal(!1),[s,g]=N.createSignal([]),[o,a]=N.createSignal("none");let l=0,S=[];const w=()=>typeof globalThis.speechSynthesis<"u"?globalThis.speechSynthesis:void 0,p=()=>{const v=w()?.getVoices()??[];if(v.length===0)return;const f=i.value.voiceURI;if(f){const $=v.find(I=>I.voiceURI===f);if($)return $}const y=e.locale.locale().split("-")[0]?.toLowerCase()??"es";return v.find($=>$.lang.toLowerCase().startsWith(y))??v[0]},u=async()=>{const v=w()?.getVoices()??[];if(v.length>0){g(v.map(f=>({id:f.voiceURI,label:`${f.name} — ${f.lang}`}))),a("web");return}try{const f=await t.voices();if(f.length>0){g(f.map(y=>({id:y,label:y}))),a("sapi");return}}catch{}a("none")},h=()=>{l++,S=[],w()?.cancel(),r(!1),t.stop().catch(()=>{})},C=async v=>{const f=oe(v,i.value.skipCode);if(!f)return!1;if(h(),await u(),o()==="web"){const y=w();if(!y)return!1;const $=p(),I=i.value.rate,W=l;S=re(f),r(!0);const L=()=>{if(W!==l)return;const _=S.shift();if(!_){r(!1);return}const k=new SpeechSynthesisUtterance(_);$&&(k.voice=$),$?.lang&&(k.lang=$.lang),k.rate=I,k.onend=L,k.onerror=L,y.speak(k)};return L(),!0}if(o()==="sapi"){r(!0);try{await t.speak({text:f,rate:i.value.rate,voice:i.value.voiceURI})}catch(y){throw r(!1),y}return!0}throw r(!1),new Error(e.t("read.noVoices"))},n=async()=>{const v=e.sessions.current();if(!v)return"";const y=(await v.server.client.message.list({sessionID:v.id,type:"assistant",order:"desc",limit:1})).data[0];return!y||y.type!=="assistant"?"":y.content.filter($=>$.type==="text").map($=>$.text).join(`
+`)};return t.on("done",()=>r(!1)),w()?.addEventListener?.("voiceschanged",()=>{u()}),u(),{speaking:c,backend:o,voices:s,refreshVoices:u,lastAssistantText:n,speak:C,cancel:h,preview:()=>C(e.t("speech.preview.text"))}}const ae=["es-ES","es-MX","es-AR","es-CO","pt-BR","pt-PT","en-US","en-GB","fr-FR","de-DE","it-IT"];function ce(e){const t=b.usePlugin(),i=t.ui,c=t.stores.prefs,r=t.stores.speech,s=t.main(O),[g,o]=N.createSignal([]),[a,l]=N.createSignal(!1);N.onMount(()=>{s.diagnose().then(n=>{n.locales.length>0&&o(n.locales)}).catch(()=>{}),e.speaker.refreshVoices()});const S=()=>{const n=g(),E=c.value.locale;return(n.includes(E)?n:[E,...n]).map(f=>({value:f,label:f}))},w=()=>[{value:"",label:t.t("speech.voice.auto")},...e.speaker.voices().map(n=>({value:n.id,label:n.label}))],p=n=>{c.update(E=>{n.locale!==void 0&&(E.locale=n.locale),n.minConfidence!==void 0&&(E.minConfidence=n.minConfidence),n.trailingSpace!==void 0&&(E.trailingSpace=n.trailingSpace),n.autoFocus!==void 0&&(E.autoFocus=n.autoFocus)})},u=n=>{r.update(E=>{n.voiceURI!==void 0&&(E.voiceURI=n.voiceURI),n.rate!==void 0&&(E.rate=n.rate),n.skipCode!==void 0&&(E.skipCode=n.skipCode)})},h=async()=>{try{await e.speaker.preview()}catch(n){t.toast.show({variant:"error",title:t.t("read.title"),description:n instanceof Error?n.message:t.t("read.error")})}},C=async()=>{l(!0);try{const n=await s.diagnose();if(n.locales.length===0){t.toast.show({variant:"error",title:t.t("error.title"),description:t.t("settings.test.failed",{reason:t.t("common.reason.no-recognizers")}),persistent:!0});return}if(!n.microphone){t.toast.show({variant:"error",title:t.t("error.title"),description:t.t("settings.test.failed",{reason:t.t("common.reason.no-microphone")}),persistent:!0});return}n.locales.length>0&&!g().length&&o(n.locales),t.toast.show({variant:"success",title:t.t("error.title"),description:t.t("settings.test.success",{locale:n.selected||n.locales[0]||c.value.locale})})}catch(n){t.toast.show({variant:"error",title:t.t("error.title"),description:n instanceof Error?n.message:String(n)})}finally{l(!1)}};return m.createComponent(i.SettingsLayout,{get title(){return t.t("settings.title")},get description(){return t.t("settings.description")},get children(){return[m.createComponent(i.SettingsSection,{get title(){return t.t("settings.section")},get children(){return[m.createComponent(i.SettingsRow,{id:"language",get title(){return t.t("settings.language")},get description(){return t.t("settings.language.description")},get children(){return m.createComponent(i.Select,{get label(){return t.t("settings.language")},get value(){return c.value.locale},get options(){return m.memo(()=>S().length>0)()?S():ae.map(n=>({value:n,label:n}))},onChange:n=>p({locale:n})})}}),m.createComponent(i.SettingsRow,{id:"confidence",get title(){return t.t("settings.confidence")},get description(){return t.t("settings.confidence.description")},get children(){return m.createComponent(i.Select,{get label(){return t.t("settings.confidence")},get value(){return String(c.value.minConfidence)},get options(){return[{value:"0.2",label:t.t("confidence.low")},{value:"0.4",label:t.t("confidence.normal")},{value:"0.6",label:t.t("confidence.high")}]},onChange:n=>p({minConfidence:Number(n)||.4})})}}),m.createComponent(i.SettingsRow,{id:"space",get title(){return t.t("settings.space")},get description(){return t.t("settings.space.description")},get children(){return m.createComponent(i.Switch,{get label(){return t.t("settings.space")},hideLabel:!0,get checked(){return c.value.trailingSpace},onChange:n=>p({trailingSpace:n})})}}),m.createComponent(i.SettingsRow,{id:"focus",get title(){return t.t("settings.focus")},get description(){return t.t("settings.focus.description")},get children(){return m.createComponent(i.Switch,{get label(){return t.t("settings.focus")},hideLabel:!0,get checked(){return c.value.autoFocus},onChange:n=>p({autoFocus:n})})}})]}}),m.createComponent(i.SettingsSection,{get title(){return t.t("speech.section")},get children(){return[m.createComponent(i.SettingsRow,{id:"voice",get title(){return t.t("speech.voice")},get description(){return t.t("speech.voice.description")},get children(){return m.createComponent(i.Select,{get label(){return t.t("speech.voice")},get value(){return r.value.voiceURI},get options(){return w()},onChange:n=>u({voiceURI:n})})}}),m.createComponent(i.SettingsRow,{id:"rate",get title(){return t.t("speech.rate")},get description(){return t.t("speech.rate.description")},get children(){return m.createComponent(i.Select,{get label(){return t.t("speech.rate")},get value(){return String(r.value.rate)},get options(){return[{value:"0.8",label:t.t("speech.rate.slow")},{value:"1",label:t.t("speech.rate.normal")},{value:"1.2",label:t.t("speech.rate.fast")},{value:"1.5",label:t.t("speech.rate.faster")}]},onChange:n=>u({rate:Number(n)||1})})}}),m.createComponent(i.SettingsRow,{id:"skip-code",get title(){return t.t("speech.skipCode")},get description(){return t.t("speech.skipCode.description")},get children(){return m.createComponent(i.Switch,{get label(){return t.t("speech.skipCode")},hideLabel:!0,get checked(){return r.value.skipCode},onChange:n=>u({skipCode:n})})}}),m.createComponent(i.SettingsRow,{id:"preview",get title(){return t.t("speech.preview")},get description(){return t.t("speech.preview.description")},get children(){return m.createComponent(i.Button,{onClick:()=>{h()},get children(){return t.t("read.idle")}})}})]}}),m.createComponent(i.SettingsSection,{get title(){return t.t("settings.test.section")},get children(){return m.createComponent(i.SettingsRow,{id:"test",get title(){return t.t("settings.test")},get description(){return t.t("settings.test.description")},get children(){return m.createComponent(i.Button,{get busy(){return a()},onClick:()=>{C()},get children(){return t.t("settings.test.run")}})}})}})]}})}const le=e=>{const t=e.main(O),i=e.main(V),c=se(e,i),r=e.stores.prefs;let s;const g=(p,u,h)=>{switch(p){case"LANGUAGE_NOT_AVAILABLE":return e.t("error.language",{locale:u||r.value.locale,available:h.length>0?h.join(", "):"—"});case"NO_RECOGNIZERS":return e.t("error.none");case"MICROPHONE_UNAVAILABLE":return e.t("error.mic");case"SYSTEM_SPEECH_UNAVAILABLE":return e.t("error.system");case"ENGINE_EXIT":return e.t("error.exit");default:return e.t("error.unknown",{code:p})}},o=()=>{c.cancel(),s=e.sessions.current()?.key,t.start({locale:r.value.locale,minConfidence:r.value.minConfidence}).catch(p=>{e.toast.show({variant:"error",title:e.t("error.title"),description:p instanceof Error?p.message:String(p)})})},a=()=>{t.stop().catch(()=>{})},l=()=>{const p=t.state()?.status??"idle";p==="starting"||p==="listening"?a():o()},S=()=>{if(c.speaking()){c.cancel();return}if(!e.sessions.current()){e.toast.show({variant:"error",title:e.t("read.title"),description:e.t("read.noSession")});return}const p=t.state()?.status??"idle";(p==="starting"||p==="listening")&&t.stop().catch(()=>{}),(async()=>{try{const u=await c.lastAssistantText();if(!u.trim()){e.toast.show({variant:"default",title:e.t("read.title"),description:e.t("read.empty")});return}await c.speak(u)||e.toast.show({variant:"default",title:e.t("read.title"),description:e.t("read.empty")})}catch(u){e.toast.show({variant:"error",title:e.t("read.title"),description:u instanceof Error?u.message:e.t("read.error")})}})()};t.on("final",p=>{const u=e.sessions.current()?.key;if(s&&u&&u!==s){e.toast.show({variant:"error",title:e.t("error.title"),description:e.t("toast.discarded")});return}const h=e.screen.current()?.composer;if(!h)return;const C=h.text(),n=r.value.trailingSpace&&C.length>0&&!/\s$/.test(C)?" ":"";h.insert(n+p.text),r.value.autoFocus&&h.focus()}),t.on("error",p=>{e.toast.show({variant:"error",title:e.t("error.title"),description:g(p.code,p.detail,p.available),persistent:p.code==="LANGUAGE_NOT_AVAILABLE"||p.code==="NO_RECOGNIZERS",duration:8e3})});const w=()=>{if(e.app.platform!=="desktop")return;const p=t.state()?.status??"idle";return p==="starting"?{label:e.t("pill.starting"),icon:"outline-wave-audio",title:e.t("pill.title"),run:l}:p==="listening"?{label:e.t("pill.listening"),icon:"outline-wave-audio",tone:"danger",title:e.t("pill.stop"),run:l}:p==="error"?{label:e.t("pill.error"),icon:"outline-wave-audio",tone:"warning",title:e.t("pill.title"),run:l}:{label:e.t("pill.idle"),icon:"outline-wave-audio",title:e.t("pill.title"),run:l}};e.add(b.ComposerPill,()=>({id:"dictate",show:()=>w()})),e.add(b.ComposerPill,()=>({id:"read",order:1,show:()=>{if(e.app.platform==="desktop")return c.speaking()?{label:e.t("read.speaking"),icon:"outline-audio",tone:"success",title:e.t("read.stop"),run:S}:{label:e.t("read.idle"),icon:"outline-audio",title:e.t("read.title"),run:S}}})),e.add(b.Command,()=>({id:"toggle",title:e.t("command.toggle"),description:e.t("command.toggle.description"),keywords:"voice dictation microphone speech dictado voz microfono",group:e.t("pill.title"),section:"prompt",bind:"mod+shift+m",slash:{name:"dictado"},editable:!0,enabled:e.app.platform==="desktop",run:l})),e.add(b.Command,()=>({id:"configure",title:e.t("command.configure"),keywords:"voice dictation microphone speech dictado voz microfono",group:e.t("pill.title"),enabled:e.app.platform==="desktop",run:()=>e.layout.settings(e.id)})),e.add(b.Command,()=>({id:"read",title:e.t("command.read"),description:e.t("command.read.description"),keywords:"read aloud speech voice leer voz altavoz lectura",group:e.t("pill.title"),section:"prompt",bind:"mod+shift+l",slash:{name:"leer"},editable:!0,enabled:e.app.platform==="desktop",run:S})),e.add(b.SettingsPage,{id:e.id,icon:"outline-audio",get title(){return e.t("settings.title")},get entries(){return[{id:"language",title:e.t("settings.language"),keywords:"voice language idioma voz"},{id:"confidence",title:e.t("settings.confidence"),keywords:"voice confidence confianza"},{id:"test",title:e.t("settings.test"),keywords:"voice microphone microphone test microfono"},{id:"voice",title:e.t("speech.voice"),keywords:"read aloud voice voz lectura altavoz"},{id:"rate",title:e.t("speech.rate"),keywords:"read aloud speed velocidad lectura"},{id:"skip-code",title:e.t("speech.skipCode"),keywords:"read aloud code codigo lectura"},{id:"preview",title:e.t("speech.preview"),keywords:"read aloud test voice probar voz"}]},render:()=>m.createComponent(ce,{speaker:c})})},de=b.Plugin.define({...X,setup:le,main:ie});exports.default=de;

package/package.json

@@ -1,6 +1,55 @@
 {
   "name": "opencode-voice-dictation",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "0.1.0",
+  "description": "Offline voice dictation and read-aloud for OpenCode Desktop on Windows",
+  "license": "MIT",
+  "author": "Fran Carlos <francarlos803@gmail.com>",
+  "homepage": "https://github.com/evious/opencode-voice-dictation#readme",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/evious/opencode-voice-dictation.git"
+  },
+  "bugs": {
+    "url": "https://github.com/evious/opencode-voice-dictation/issues"
+  },
+  "keywords": [
+    "opencode",
+    "opencode-plugin",
+    "voice",
+    "dictation",
+    "speech-to-text",
+    "text-to-speech",
+    "windows",
+    "offline"
+  ],
+  "type": "module",
+  "exports": {
+    ".": "./src/index.ts",
+    "./gui-experimental": "./dist/gui-experimental.cjs"
+  },
+  "files": [
+    "src",
+    "dist",
+    "README.md",
+    "LICENSE"
+  ],
+  "publishConfig": {
+    "access": "public"
+  },
+  "scripts": {
+    "build": "vite build",
+    "dev": "vite build --watch",
+    "test": "node test/smoke.cjs",
+    "typecheck": "tsc --noEmit",
+    "prepublishOnly": "npm run typecheck && npm run build && npm test"
+  },
+  "devDependencies": {
+    "@opencode/plugin": "2.2.6",
+    "effect": "4.1.0",
+    "@types/node": "^24.0.0",
+    "solid-js": "^1.9.17",
+    "typescript": "^5.9.0",
+    "vite": "^7.3.0",
+    "vite-plugin-solid": "^2.11.0"
+  }
+}

package/src/gui/contract.ts

@@ -0,0 +1,77 @@
+import { Schema } from "effect"
+import { Ipc } from "@opencode/plugin/gui"
+
+/**
+ * Bridge between the window setup and the Electron main process.
+ *
+ * The engine lives in `main` because it needs `node:child_process`; the window
+ * only paints status and inserts the transcript.
+ */
+export const Dictation = Ipc.define({
+  id: "voice-dictation.dictation",
+  state: Schema.Struct({
+    status: Schema.Literals(["idle", "starting", "listening", "error"]),
+    error: Schema.String,
+    locale: Schema.String,
+  }),
+  methods: {
+    start: {
+      input: Schema.Struct({
+        locale: Schema.String,
+        minConfidence: Schema.Number,
+      }),
+    },
+    stop: {},
+    diagnose: {
+      output: Schema.Struct({
+        locales: Schema.Array(Schema.String),
+        microphone: Schema.Boolean,
+        selected: Schema.String,
+        error: Schema.String,
+      }),
+    },
+  },
+  events: {
+    final: Schema.Struct({
+      text: Schema.String,
+      confidence: Schema.Number,
+    }),
+    error: Schema.Struct({
+      code: Schema.String,
+      detail: Schema.String,
+      available: Schema.Array(Schema.String),
+    }),
+  },
+})
+
+export interface DiagnoseResult {
+  locales: string[]
+  microphone: boolean
+  selected: string
+  error: string
+}
+
+/**
+ * Text-to-speech bridge. The window speaks with `speechSynthesis` when the
+ * Electron build exposes voices, and falls back to this main-process engine
+ * (`System.Speech.Synthesis`) when it does not.
+ */
+export const Speech = Ipc.define({
+  id: "voice-dictation.speech",
+  methods: {
+    speak: {
+      input: Schema.Struct({
+        text: Schema.String,
+        rate: Schema.Number,
+        voice: Schema.String,
+      }),
+    },
+    stop: {},
+    voices: {
+      output: Schema.Array(Schema.String),
+    },
+  },
+  events: {
+    done: Schema.Struct({ ok: Schema.Boolean }),
+  },
+})