com um clique
create-agent
Create a new voice agent with YAML configuration and prompt template
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.
Menu
Create a new voice agent with YAML configuration and prompt template
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 | create-agent |
| description | Create a new voice agent with YAML configuration and prompt template |
Create a new agent in apps/artagent/backend/registries/agentstore/.
agentstore/
└── {agent_name}/
├── agent.yaml # Agent configuration
└── prompt.jinja # System prompt template
name: AgentName
description: Brief description of agent's purpose
handoff:
trigger: handoff_agent_name # Tool name other agents use to transfer here
greeting: |
Hi{{ ' ' + caller_name if caller_name else '' }}, I'm {{ agent_name | default('your assistant') }}.
return_greeting: |
Welcome back{{ ', ' + caller_name if caller_name else '' }}. How else can I help?
voice:
name: en-US-AlloyTurboMultilingualNeural
type: azure-standard
rate: "-4%"
voicelive_model:
deployment_id: gpt-realtime
temperature: 0.7
max_tokens: 2048
cascade_model:
deployment_id: gpt-4o
temperature: 0.8
max_tokens: 2048
session:
modalities: [TEXT, AUDIO]
input_audio_format: PCM16
output_audio_format: PCM16
input_audio_transcription_settings:
model: gpt-4o-transcribe
language: en-US
turn_detection:
type: azure_semantic_vad
threshold: 0.5
silence_duration_ms: 720
tool_choice: auto
tools:
- tool_name_1
- tool_name_2
- handoff_concierge # Return to main agent
prompts:
path: prompt.jinja
You are {{ agent_name | default('an assistant') }} at {{ institution_name | default('our company') }}.
## Your Role
[Describe the agent's specific responsibilities]
## Guidelines
- Be concise and clear
- [Add specific behavioral guidelines]
## Available Tools
{% for tool in tools %}
- {{ tool }}
{% endfor %}
## Current Context
{% if caller_name %}Customer: {{ caller_name }}{% endif %}
{% if client_id %}Client ID: {{ client_id }}{% endif %}
apps/artagent/backend/registries/agentstore/{agent_name}/agent.yaml with configurationprompt.jinja with system promptregistries/toolstore/ or create new onesUnifiedAgent in registries/agentstore/base.py - base dataclassHandoffConfig, VoiceConfig, ModelConfig - configuration dataclassesCheck existing tools in registries/toolstore/:
auth.py - Authentication toolsbanking/ - Banking operationsfraud.py - Fraud detectionhandoffs.py - Agent transfer tools