Menu
Add a new voice handler or feature to the voice module
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
Read-only — assemble the wider runtime picture of a deployed voice app from azd deployment artifacts and Azure Monitor (Application Insights / Log Analytics) via Azure MCP or az CLI, then render it as KQL, call timelines, latency waterfalls, and mermaid diagrams
Service catalog and guided onboarding for the azd deployment. USE WHEN the user wants to discover, install, set up, or be walked through the deployable components (Azure OpenAI/AI Foundry, Speech, ACS/telephony, Cosmos DB, Redis, Container Apps, Key Vault, App Config, CardAPI MCP), asks "what gets deployed", "what services does this use", "help me onboard", "set up the deployment", "guide me through azd up", "which components do I need", or wants to enable optional pieces (phone number, EasyAuth, data seeding). Acts as the entry point an agent hooks into to assess current state, present the catalog, and onboard each component. DO NOT USE FOR: deep azd hook/flow internals or model-availability checks (use deployment-guide); runtime failure diagnosis (use troubleshoot); telemetry/log analysis (use observability-insights).
Agent-first, read-only diagnosis of the voice pipeline (deploy, telephony, STT, LLM, TTS, state) — gather evidence via Azure MCP / azd artifacts / CLI, probe the user for missing details, and recommend fixes without changing anything
Require relevant tests and documentation updates for any code or config change, and report what was run.
Create or update evaluation scenarios for the tests/evaluation framework, including session-based scenarios and A/B comparisons
Guide azd-based deployments, including where azure.yaml and azd hook scripts live, the current deployment flow, troubleshooting docs, and regional/model availability checks for Azure OpenAI
| name | add-voice-handler |
| description | Add a new voice handler or feature to the voice module |
Add new voice features to apps/artagent/backend/voice/.
"""
Voice Feature Module
====================
Brief description of the voice feature.
"""
from __future__ import annotations
from typing import TYPE_CHECKING
from apps.artagent.backend.voice.shared.context import VoiceSessionContext, TransportType
from apps.artagent.backend.voice.shared.handoff_service import HandoffService
from apps.artagent.backend.voice.shared.metrics_factory import LazyMeter, build_session_attributes
from utils.ml_logging import get_logger
if TYPE_CHECKING:
from azure.cognitiveservices.speech import SpeechConfig
logger = get_logger(__name__)
# Lazy metrics
_meter = LazyMeter("voice.my_feature", version="1.0.0")
_latency = _meter.histogram(
name="voice.my_feature.latency",
description="Feature latency",
unit="ms",
)
async def handle_my_feature(
context: VoiceSessionContext,
data: bytes,
) -> None:
"""
Handle voice feature.
Args:
context: Voice session context (use instead of websocket.state)
data: Input data to process
"""
# Transport-aware processing
if context.transport_type == TransportType.BROWSER:
sample_rate = 48000
elif context.transport_type == TransportType.ACS:
sample_rate = 16000
else:
sample_rate = 24000 # VoiceLive
# Process and record metrics
attrs = build_session_attributes(context.session_id)
_latency.record(latency_ms, attributes=attrs)
voice/ or appropriate subdirectoryVoiceSessionContext instead of websocket.statevoice. prefixfrom apps.artagent.backend.voice.shared.context import VoiceSessionContext
context = VoiceSessionContext.from_websocket(websocket)
session_id = context.session_id
transport = context.transport_type
from apps.artagent.backend.voice.tts import TTSPlayback
tts = TTSPlayback(context)
await tts.speak(text) # Auto-routes to browser/ACS/VoiceLive
from apps.artagent.backend.voice.shared.handoff_service import HandoffService
handoff_service = HandoffService(
scenario_name=scenario_name,
handoff_map=handoff_map,
agents=agents,
memo_manager=memo_manager,
)
resolution = handoff_service.resolve_handoff(from_agent, to_agent)
from typing import TYPE_CHECKING
if TYPE_CHECKING:
from azure.cognitiveservices.speech import SpeechConfig
def create_speech_config() -> "SpeechConfig":
from azure.cognitiveservices.speech import SpeechConfig
return SpeechConfig(...)
VoiceSessionContext instead of websocket.stateTTSPlayback for audio outputLazyMeter pattern for metricsvoice. prefixSee voice-module.instructions.md for full patterns and contracts.