@johpaz/hive-agents 0.0.35 → 0.0.37

package/packages/skills/src/bundled/voice/voice_assistant/SKILL.md

@@ -0,0 +1,174 @@
+---
+name: voice_assistant
+description: "Full voice-to-voice interaction: transcribe user speech, process request, and respond with synthesized speech"
+version: 1.0.0
+author: Hive Team
+icon: "🎙️🔊"
+category: voice
+permissions:
+  - voice_transcribe
+  - voice_speak
+dependencies: []
+tools: [voice_transcribe, voice_speak]
+
+# Structured skill fields
+triggers:
+  - "modo voz"
+  - "voice mode"
+  - "asistente de voz"
+  - "voice assistant"
+  - "hablá conmigo"
+  - "talk to me"
+  - "interacción por voz"
+  - "voice interaction"
+  - "respuesta hablada"
+  - "spoken response"
+  - "comando de voz"
+  - "voice command"
+  - "diálogo por voz"
+  - "voice dialogue"
+
+preferred_agents: []
+
+steps:
+  - step: 1
+    action: voice_transcribe
+    instruction: "Transcribe user's voice input to text"
+    params:
+      audio: "user audio input"
+    output: user_text
+
+  - step: 2
+    action: process_request
+    instruction: "Process transcribed text and generate appropriate response"
+    params:
+      text: "transcribed user request"
+    output: response_text
+
+  - step: 3
+    action: voice_speak
+    instruction: "Synthesize response text to speech"
+    params:
+      text: "response text"
+      voice_id: "configured voice"
+    output: response_audio
+
+  - step: 4
+    action: deliver_response
+    instruction: "Send synthesized speech back to user"
+    output: delivered
+
+rules:
+  - "Maintain conversation context across voice exchanges"
+  - "Keep responses concise for voice (avoid long paragraphs)"
+  - "Use natural, conversational tone optimized for speech"
+  - "Handle interruptions gracefully (if streaming supported)"
+  - "Indicate when processing (typing indicators, sounds)"
+  - "Fallback to text if voice fails at any step"
+
+output_format:
+  structure: audio_with_text
+  sections:
+    - "audio_response"
+    - "text_fallback"
+    - "conversation_context"
+  max_length: "Short, natural speech responses (30-60 seconds max)"
+
+examples:
+  - user_input: "[audio] ¿Cuál es el clima hoy?"
+    expected_behavior: "Transcribe → process weather query → respond with voice synthesis"
+
+  - user_input: "[audio] Contame un chiste"
+    expected_behavior: "Transcribe → generate joke → speak response with appropriate timing"
+
+  - user_input: "activá el modo voz"
+    expected_behavior: "Enable voice mode → all future responses include audio synthesis"
+---
+
+# Voice Assistant Skill
+
+## Cuándo se Activa
+
+Esta skill se activa para interacción completa voz a voz: el usuario habla, el asistente procesa y responde con voz.
+
+## Herramientas Disponibles
+
+| Tool | Qué hace | Cuándo usarla |
+|------|----------|---------------|
+| `voice_transcribe` | Audio → texto | Input del usuario |
+| `voice_speak` | Texto → audio | Respuesta del asistente |
+
+## Workflow
+
+### Voice-to-Voice
+```javascript
+// 1. Usuario habla
+const userAudio = receiveAudio()
+
+// 2. Transcribir
+const userText = voice_transcribe({
+  audio: userAudio,
+  language: "auto"
+})
+// → "¿Cuál es el clima hoy?"
+
+// 3. Procesar request
+// - Entender intención
+// - Ejecutar acción (ej. consultar API clima)
+// - Generar respuesta
+const responseText = "Hoy hay 25 grados y soleado en Buenos Aires"
+
+// 4. Sintetizar respuesta
+const responseAudio = voice_speak({
+  text: responseText,
+  voice_id: "eleven_flash_v2_5",
+  language: "es"
+})
+
+// 5. Enviar audio
+sendAudio(responseAudio)
+```
+
+## Casos de Uso
+
+| Caso | Flujo |
+|------|-------|
+| Pregunta simple | Transcribe → responde → sintetiza |
+| Comando | Transcribe → ejecuta → confirma por voz |
+| Diálogo | Mantener contexto entre exchanges |
+| Wake word | Escuchar "hey bee" → activar → procesar |
+
+## Configuración
+
+### Wake Word
+```json
+{
+  "voice_wake_word": "hey bee",
+  "voice_wake_enabled": true
+}
+```
+
+### Canal Voice
+```json
+{
+  "voice_enabled": true,
+  "tts_enabled": true,
+  "stt_provider": "groq-whisper",
+  "tts_provider": "elevenlabs",
+  "tts_voice_id": "eleven_flash_v2_5"
+}
+```
+
+## Mejores Prácticas
+
+- Respuestas cortas y naturales (<60s)
+- Mantener contexto conversacional
+- Indicadores de procesamiento
+- Fallback a texto si falla voz
+
+## Errores a Evitar
+
+- ❌ Respuestas muy largas para audio
+- ❌ Perder contexto entre exchanges
+- ❌ No indicar que está procesando
+- ❌ No tener fallback si falla TTS/STT

package/packages/skills/src/bundled/voice/voice_input/SKILL.md

@@ -0,0 +1,146 @@
+---
+name: voice_input
+description: "Transcribe audio input to text using STT (Speech-to-Text) providers like Groq Whisper or OpenAI Whisper"
+version: 1.0.0
+author: Hive Team
+icon: "🎙️"
+category: voice
+permissions:
+  - voice_transcribe
+dependencies: []
+tools: [voice_transcribe]
+
+# Structured skill fields
+triggers:
+  - "transcribí este audio"
+  - "transcribe audio"
+  - "convertí voz a texto"
+  - "convert voice to text"
+  - "qué dice el audio"
+  - "what does audio say"
+  - "escuchá esto"
+  - "listen to this"
+  - "audio a texto"
+  - "audio to text"
+  - "reconocimiento de voz"
+  - "speech recognition"
+  - "nota de voz"
+  - "voice note"
+
+preferred_agents: []
+
+steps:
+  - step: 1
+    action: receive_audio
+    instruction: "Receive audio file or stream from user"
+    params:
+      source: "file upload | voice message | stream"
+    output: audio_data
+
+  - step: 2
+    action: voice_transcribe
+    instruction: "Send audio to STT provider for transcription"
+    params:
+      audio: "audio buffer or file path"
+      language: "auto-detect or specified"
+    output: transcription
+
+  - step: 3
+    action: synthesize
+    instruction: "Format transcription with punctuation and structure"
+    output: formatted_text
+
+  - step: 4
+    action: deliver_result
+    instruction: "Return transcription to user"
+    output: delivered
+
+rules:
+  - "Auto-detect language when not specified"
+  - "Add punctuation and capitalization for readability"
+  - "Indicate speaker changes if multi-speaker audio"
+  - "Mark unclear segments with [inaudible] or [unclear]"
+  - "Preserve original language unless translation requested"
+  - "Handle background noise gracefully — note if affects quality"
+
+output_format:
+  structure: markdown
+  sections:
+    - "transcription"
+    - "language_detected"
+    - "duration"
+    - "confidence"
+    - "notes"
+  max_length: "Full transcription with timestamps if long"
+
+examples:
+  - user_input: "[envía audio] transcribí esto"
+    expected_behavior: "voice_transcribe → return text transcription with punctuation"
+
+  - user_input: "convertí esta nota de voz a texto"
+    expected_behavior: "voice_transcribe → return formatted transcription"
+
+  - user_input: "qué dice este audio"
+    expected_behavior: "voice_transcribe → summarize if very long, full transcription if short"
+---
+
+# Voice Input Skill
+
+## Cuándo se Activa
+
+Esta skill se activa cuando el usuario envía audio y necesita transcripción a texto: notas de voz, grabaciones, comandos de voz.
+
+## Herramientas Disponibles
+
+| Tool | Qué hace | Cuándo usarla |
+|------|----------|---------------|
+| `voice_transcribe` | Convierte audio → texto | Transcripción de cualquier audio |
+
+## Workflow
+
+### Transcripción
+```javascript
+// 1. Recibir audio
+// - File upload
+// - Voice message (Telegram, WhatsApp)
+// - Stream en vivo
+
+// 2. Transcribir
+const result = voice_transcribe({
+  audio: audioBuffer,
+  language: "es"  // o "auto" para detectar
+})
+
+// 3. Formatear
+// - Agregar puntuación
+// - Capitalizar
+// - Marcar speakers si hay múltiples
+
+// 4. Entregar resultado
+```
+
+## Proveedores STT Soportados
+
+| Provider | Modelos | Idiomas |
+|----------|---------|---------|
+| Groq | whisper-large-v3, turbo | Multi |
+| OpenAI | whisper-1 | Multi |
+
+## Configuración por Canal
+
+Cada canal puede configurar su proveedor STT preferido:
+- `stt_provider`: "groq-whisper" | "openai-whisper"
+
+## Mejores Prácticas
+
+- Detectar idioma automáticamente
+- Agregar puntuación para legibilidad
+- Marcar segmentos inaudibles
+- Preservar idioma original
+
+## Errores a Evitar
+
+- ❌ Traducir sin pedir (mantener idioma)
+- ❌ Omitir puntuación
+- ❌ No indicar baja confianza
+- ❌ Ignorar ruido de fondo que afecta calidad

package/packages/skills/src/bundled/voice/voice_output/SKILL.md

@@ -0,0 +1,151 @@
+---
+name: voice_output
+description: "Convert text to synthesized speech using TTS (Text-to-Speech) providers like ElevenLabs, OpenAI TTS, or Gemini TTS"
+version: 1.0.0
+author: Hive Team
+icon: "🔊"
+category: voice
+permissions:
+  - voice_speak
+dependencies: []
+tools: [voice_speak]
+
+# Structured skill fields
+triggers:
+  - "leé esto en voz alta"
+  - "read this aloud"
+  - "convertí a voz"
+  - "convert to speech"
+  - "hablá este texto"
+  - "speak this text"
+  - "texto a voz"
+  - "text to speech"
+  - "generá audio"
+  - "generate audio"
+  - "síntesis de voz"
+  - "voice synthesis"
+  - "escuchá la respuesta"
+  - "listen to response"
+
+preferred_agents: []
+
+steps:
+  - step: 1
+    action: receive_text
+    instruction: "Receive text content to convert to speech"
+    params:
+      text: "text to synthesize"
+      language: "language code (es, en, etc.)"
+    output: text_content
+
+  - step: 2
+    action: optimize_for_speech
+    instruction: "Preprocess text for natural speech (expand abbreviations, handle special chars)"
+    output: optimized_text
+
+  - step: 3
+    action: voice_speak
+    instruction: "Send text to TTS provider for synthesis"
+    params:
+      text: "optimized text"
+      voice_id: "configured voice ID"
+      language: "language code"
+    output: audio_buffer
+
+  - step: 4
+    action: deliver_audio
+    instruction: "Return synthesized audio to user via channel"
+    output: delivered
+
+rules:
+  - "Use configured TTS provider for channel (tts_provider config)"
+  - "Respect user's voice preference if set (tts_voice_id)"
+  - "Preprocess text: expand numbers, dates, abbreviations for natural speech"
+  - "Handle SSML tags if present for prosody control"
+  - "Split long text into chunks if exceeds provider limits"
+  - "Cache frequently spoken responses to reduce API calls"
+
+output_format:
+  structure: audio
+  sections:
+    - "audio_file"
+    - "duration"
+    - "voice_used"
+    - "language"
+  max_length: "Audio attachment or file path"
+
+examples:
+  - user_input: "leé esta respuesta en voz alta"
+    expected_behavior: "voice_speak → return audio file with synthesized speech"
+
+  - user_input: "convertí este texto a audio"
+    expected_behavior: "voice_speak → generate and return audio file"
+
+  - user_input: "respondeme por voz"
+    expected_behavior: "Generate response text → voice_speak → send audio"
+---
+
+# Voice Output Skill
+
+## Cuándo se Activa
+
+Esta skill se activa cuando el usuario necesita convertir texto a voz: leer respuestas, generar audio, síntesis de voz.
+
+## Herramientas Disponibles
+
+| Tool | Qué hace | Cuándo usarla |
+|------|----------|---------------|
+| `voice_speak` | Convierte texto → audio | Síntesis de voz |
+
+## Workflow
+
+### Text-to-Speech
+```javascript
+// 1. Recibir texto
+const text = "Hola, ¿cómo estás?"
+
+// 2. Preprocesar
+// - Expandir números: "5" → "cinco"
+// - Expandir fechas: "01/01" → "primero de enero"
+// - Expandir abbreviaturas: "Dr." → "Doctor"
+
+// 3. Sintetizar
+const audio = voice_speak({
+  text: optimizedText,
+  voice_id: "eleven_flash_v2_5",  // o configured voice
+  language: "es"
+})
+
+// 4. Entregar audio
+// - Enviar como archivo
+// - Streaming si el canal lo soporta
+```
+
+## Proveedores TTS Soportados
+
+| Provider | Modelos | Voces |
+|----------|---------|-------|
+| ElevenLabs | Flash V2.5, Turbo V2.5, Multilingual V2, V3 | 1000+ |
+| OpenAI | tts-1, tts-1-hd, gpt-4o-mini-tts | 6+ |
+| Gemini | 2.5 Flash TTS, 2.5 Pro TTS | Multi |
+| Qwen | Qwen TTS Flash, Instruct | Multi |
+
+## Configuración por Canal
+
+Cada canal configura su proveedor TTS:
+- `tts_provider`: "elevenlabs" | "openai-tts" | "gemini-tts"
+- `tts_voice_id`: ID específico de voz (ej. ElevenLabs voice ID)
+
+## Mejores Prácticas
+
+- Preprocesar texto para naturalidad
+- Usar voz configurada por usuario
+- Cachear respuestas frecuentes
+- Split de textos largos
+
+## Errores a Evitar
+
+- ❌ Enviar texto crudo sin preprocesar
+- ❌ Ignorar preferencia de voz
+- ❌ No manejar límites de longitud
+- ❌ No cachear (costo API)

package/packages/skills/src/bundled/web/browser_automate/SKILL.md

@@ -0,0 +1,120 @@
+---
+name: browser_automate
+description: "Automate web workflows with navigation, clicks, form filling, and visual verification"
+version: 1.0.0
+author: Hive Team
+icon: "🤖"
+category: web
+permissions:
+  - browser_control
+dependencies: []
+tools: [browser_navigate, browser_click, browser_type, browser_screenshot]
+
+# Structured skill fields
+triggers:
+  - "automatizá el navegador"
+  - "automate browser"
+  - "completá el formulario"
+  - "fill form"
+  - "hacé clic en"
+  - "click on"
+  - "iniciá sesión"
+  - "login"
+  - "registrate"
+  - "sign up"
+  - "interactuá con la web"
+  - "interact with website"
+  - "flujo web"
+  - "web workflow"
+
+preferred_agents: []
+
+steps:
+  - step: 1
+    action: browser_navigate
+    instruction: "Navigate to target URL and wait for page to fully load"
+    params:
+      url: "target URL"
+    output: page_loaded
+
+  - step: 2
+    action: browser_click
+    instruction: "Click on elements (buttons, links) to navigate or trigger actions"
+    params:
+      selector: "CSS selector for element"
+    output: click_result
+
+  - step: 3
+    action: browser_type
+    instruction: "Type text into form fields (inputs, textareas)"
+    params:
+      selector: "CSS selector for input"
+      text: "text to type"
+    output: type_result
+
+  - step: 4
+    action: browser_screenshot
+    instruction: "Take screenshot to verify state after interactions"
+    output: verification_screenshot
+
+rules:
+  - "Wait for page to fully load after each navigation or significant interaction"
+  - "Use specific, stable CSS selectors (IDs preferred over classes)"
+  - "Take screenshots after critical steps for verification"
+  - "Handle errors gracefully — element may not exist or be interactable"
+  - "For multi-step flows, verify state after each step before proceeding"
+
+output_format:
+  structure: markdown
+  sections:
+    - "workflow_steps"
+    - "final_state"
+    - "screenshot_path"
+  max_length: "Summary of automation flow"
+
+examples:
+  - user_input: "automatizá el login en example.com"
+    expected_behavior: "Navigate → type credentials → click submit → screenshot verification"
+
+  - user_input: "completá el formulario de contacto"
+    expected_behavior: "Navigate → type name/email/message → click submit → verify submission"
+
+  - user_input: "hacé clic en todos los enlaces del menú"
+    expected_behavior: "Navigate → click each menu item → screenshot each page"
+---
+
+# Browser Automate Skill
+
+## Cuándo se Activa
+
+Esta skill se activa para automatizar flujos de interacción con aplicaciones web: logins, formularios, navegación programática.
+
+## Herramientas Disponibles
+
+| Tool | Qué hace | Cuándo usarla |
+|------|----------|---------------|
+| `browser_navigate` | Navega a URL | Inicio de flujo |
+| `browser_click` | Click en elementos | Botones, enlaces, triggers |
+| `browser_type` | Escribe en inputs | Formularios, búsquedas |
+| `browser_screenshot` | Captura estado | Verificación visual |
+
+## Workflow Típico
+
+1. **Navegar** → URL inicial
+2. **Interactuar** → click/type según flujo
+3. **Verificar** → screenshot después de acciones críticas
+4. **Repetir** → para flujos multi-paso
+
+## Mejores Prácticas
+
+- Selectores estables (IDs > classes > XPath)
+- Esperar carga después de navegación
+- Verificar estado visual con screenshots
+- Manejar errores de elementos no encontrados
+
+## Errores a Evitar
+
+- ❌ Selectores frágiles que cambian
+- ❌ No esperar carga de página
+- ❌ Ignorar errores de elementos
+- ❌ No verificar estado después de acciones

package/packages/skills/src/bundled/web/browser_scrape/SKILL.md

@@ -0,0 +1,109 @@
+---
+name: browser_scrape
+description: "Navigate to web pages and capture rendered content including screenshots for dynamic sites"
+version: 1.0.0
+author: Hive Team
+icon: "📸"
+category: web
+permissions:
+  - browser_control
+dependencies: []
+tools: [browser_navigate, browser_screenshot, web_fetch]
+
+# Structured skill fields
+triggers:
+  - "capturá el contenido"
+  - "scrape content"
+  - "obtené la página renderizada"
+  - "get rendered page"
+  - "sitios dinámicos"
+  - "dynamic sites"
+  - "web con javascript"
+  - "javascript websites"
+  - "tomá screenshot y contenido"
+  - "screenshot and content"
+
+preferred_agents: []
+
+steps:
+  - step: 1
+    action: browser_navigate
+    instruction: "Navigate to URL and wait for full page load including JavaScript rendering"
+    params:
+      url: "target URL"
+    output: page_loaded
+
+  - step: 2
+    action: browser_screenshot
+    instruction: "Take screenshot to capture visual state of rendered page"
+    output: screenshot
+
+  - step: 3
+    action: web_fetch
+    instruction: "Extract text content from rendered page as markdown"
+    output: extracted_content
+
+  - step: 4
+    action: synthesize
+    instruction: "Combine screenshot and text content for comprehensive capture"
+    output: scraped_data
+
+rules:
+  - "Wait for full page load including JavaScript-rendered content"
+  - "Take screenshot before extracting text to capture initial state"
+  - "For infinite scroll pages, scroll down and capture multiple screenshots"
+  - "Respect website terms of service — no aggressive scraping"
+  - "Store both screenshot and text for complete record"
+
+output_format:
+  structure: markdown
+  sections:
+    - "url"
+    - "screenshot_path"
+    - "extracted_content"
+    - "timestamp"
+  max_length: "Full content extraction"
+
+examples:
+  - user_input: "capturá el contenido de https://example.com/dashboard"
+    expected_behavior: "browser_navigate → wait for JS render → browser_screenshot → browser_fetch → return both"
+
+  - user_input: "obtené la página renderizada de la app"
+    expected_behavior: "Navigate → wait for SPA to load → screenshot + fetch content"
+
+  - user_input: "scrapeá este sitio con javascript"
+    expected_behavior: "Full browser render → capture visual and text content"
+---
+
+# Browser Scrape Skill
+
+## Cuándo se Activa
+
+Esta skill se activa para sitios web dinámicos que requieren JavaScript rendering, donde el contenido no está disponible en HTML estático.
+
+## Herramientas Disponibles
+
+| Tool | Qué hace | Cuándo usarla |
+|------|----------|---------------|
+| `browser_navigate` | Navega y renderiza página completa | Sitios con JavaScript/SPA |
+| `browser_screenshot` | Captura estado visual | Evidencia de contenido renderizado |
+| `web_fetch` | Extrae texto como markdown | Contenido textual de página renderizada |
+
+## Workflow
+
+1. **Navegar** → `browser_navigate({ url })` + esperar renderizado JS
+2. **Capturar visual** → `browser_screenshot()`
+3. **Extraer texto** → `web_fetch()`
+4. **Combinar** → screenshot + texto para scrape completo
+
+## Mejores Prácticas
+
+- Esperar renderizado completo de JavaScript
+- Para infinite scroll: hacer scroll y múltiples screenshots
+- Capturar antes y después de interacciones si es dinámico
+
+## Errores a Evitar
+
+- ❌ No esperar renderizado JavaScript
+- ❌ Solo capturar HTML estático para sitios SPA
+- ❌ Ignorar términos de servicio del sitio