X Openai Twilio Voice Outbound Calls Ad4ff44

---
name: x-openai-twilio-voice-outbound-calls-ad4ff44
description: 'Make outbound phone calls via Twilio''s Programmable Voice REST API. Covers the full voice platform: calls.create(), answering machine detection (AMD), conference-based agent bridging, call recording, status tracking, and SIP Trunking. Use this skill for outbound calls, sales dialers, or when asking what voice APIs are available.'
---

### 导出副本边界与技能别名

迁移、安装、修复和经验回写只作用于当前加载的目标宿主技能副本。原 Codex 技能/MCP、下载的源码和 `original-skills` 均为来源材料，不因执行本副本而写入；只有用户另行明确要求配置 Codex 产品本身时，才按该产品流程处理指定目标。脚本需要安装目录或状态目录时，显式提供目标路径；不得从 `CODEX_HOME` 隐式推断写入位置。

跨技能名称可能沿用源库别名。先查 [技能别名表](../../skill-aliases.json)，优先选择同来源、同插件的条目，再使用其 `id` 调用或读取对应入口；不得假定旧 `$skill`、`superpowers:skill` 名称在新宿主直接注册。不能确定目标时先核对，不替换真实产品命令或 Shell 环境变量。

## 跨宿主使用约定

本副本供支持 Agent Skills 的客户端使用。先检查当前宿主实际提供的工具、权限和已连接服务，再执行正文工作流。原文中的宿主工具标识是能力参考；只有实际发现该工具时才可调用，其他宿主应使用经核验的等价 MCP、CLI 或 API。没有等价能力的步骤必须报告不可用，不能虚构调用或完成状态。

技能资源以本技能目录为基准；输出写入当前项目中用户指定的位置。通过技能名称或自然语言触发，具体快捷命令由宿主决定。涉及真实产品、模型 API、协议字段和官方网址的名称保持原意。授权、审批、隐私和工具范围限制在迁移后仍然有效；不得为兼容而扩大权限。具体依赖与迁移说明见同目录 `PORTABILITY.md`。

## Overview

> **Agent safety:** Before placing an outbound call, always confirm the recipient number and intent with the user. Outbound calls are irreversible and may incur charges. For automated systems, implement TCPA compliance: obtain prior express consent, respect quiet hours (8 AM–9 PM recipient local time), and maintain a Do Not Call list.

Every outbound call requires a `from` Twilio number, a `to` recipient, and TwiML instructions that define what happens when the call is answered — either as a webhook URL or inline.

---

## Voice Platform Capabilities

Outbound calls go beyond basic `calls.create()`. Here's what you can build:

| Capability | How | When to use |
|-----------|-----|-------------|
| **Basic outbound call** | `calls.create()` with TwiML or webhook URL | Any outbound call — see Quickstart below |
| **Answering Machine Detection (AMD)** | `machineDetection` parameter on `calls.create()` | Sales dialers, call campaigns — filter voicemail from humans |
| **Conference-based agent bridging** | `<Dial><Conference>` in TwiML | Connect agents to live prospects with whisper, barge, hold |
| **SIP Trunking** | Elastic SIP Trunking | Bring your own carrier for outbound calls — cost reduction at scale |
| **Call Recording** | `record=True` on `calls.create()` or `<Record>` verb | Compliance, QA, training |
| **Voice Insights** | Automatic per-call metrics | Call quality monitoring — latency, jitter, packet loss, answer rates |
| **AI Voice Agents** | ConversationRelay + LLM | Real-time speech recognition → LLM → TTS for conversational AI |

For TwiML verbs (Say, Gather, Dial, Record, Conference, Pay), see `twilio-voice-twiml`.
For AI voice agents, see `twilio-voice-conversation-relay`.

---

## Prerequisites

- Twilio account with a voice-capable phone number
  — New to Twilio? See `twilio-account-setup` for signup, getting a number, and trial limitations
  — Trial accounts can only call verified numbers
- Environment variables:
  - `TWILIO_ACCOUNT_SID`
  - `TWILIO_AUTH_TOKEN`
  — See `twilio-iam-auth-setup` for credential setup and best practices
- SDK: `pip install twilio` / `npm install twilio`

---

## Quickstart

**Python**
```python
import os
from twilio.rest import Client

client = Client(os.environ["TWILIO_ACCOUNT_SID"], os.environ["TWILIO_AUTH_TOKEN"])

call = client.calls.create(
    from_="+15017122661",    # Your Twilio number (E.164)
    to="+15558675310",       # Recipient (E.164)
    twiml="<Response><Say>Your order has shipped. Goodbye.</Say></Response>"
)

print(call.sid)     # CAxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
print(call.status)  # queued | ringing | in-progress | completed | failed
```

**Node.js**
```node
const twilio = require("twilio");
const client = twilio(process.env.TWILIO_ACCOUNT_SID, process.env.TWILIO_AUTH_TOKEN);

const call = await client.calls.create({
    from: "+15017122661",
    to: "+15558675310",
    twiml: "<Response><Say>Your order has shipped. Goodbye.</Say></Response>",
});

console.log(call.sid);
console.log(call.status);
```

---

## Key Patterns

### Use a Webhook URL for Dynamic Call Handling

Pass a `url` instead of inline `twiml` — Twilio POSTs to your server when the call connects and executes the TwiML you return.

**Python**
```python
call = client.calls.create(
    from_="+15017122661",
    to="+15558675310",
    url="https://yourapp.com/twiml/welcome"
)
```

**Node.js**
```node
const call = await client.calls.create({
    from: "+15017122661",
    to: "+15558675310",
    url: "https://yourapp.com/twiml/welcome",
});
```

**Python (Flask) — TwiML webhook handler**
```python
from flask import Flask
from twilio.twiml.voice_response import VoiceResponse

app = Flask(__name__)

@app.route("/twiml/welcome", methods=["POST"])
def welcome():
    response = VoiceResponse()
    response.say("Hello! Press 1 to hear your account balance.")
    response.gather(num_digits=1, action="/twiml/handle-input")
    return str(response)
```

**Node.js (Express) — TwiML webhook handler**
```node
const { VoiceResponse } = require("twilio").twiml;

app.post("/twiml/welcome", (req, res) => {
    const response = new VoiceResponse();
    response.say("Hello! Press 1 to hear your account balance.");
    response.gather({ numDigits: 1, action: "/twiml/handle-input" });
    res.type("text/xml").send(response.toString());
});
```

For all TwiML verbs (Say, Gather, Dial, Record, Conference), see `twilio-voice-twiml`.

### Track Call Status

**Python**
```python
call = client.calls.create(
    from_="+15017122661",
    to="+15558675310",
    url="https://yourapp.com/twiml/welcome",
    status_callback="https://yourapp.com/call-status",
    status_callback_method="POST"
)
```

**Node.js**
```node
const call = await client.calls.create({
    from: "+15017122661",
    to: "+15558675310",
    url: "https://yourapp.com/twiml/welcome",
    statusCallback: "https://yourapp.com/call-status",
    statusCallbackMethod: "POST",
});
```

Status transitions: `queued → ringing → in-progress → completed` (or `failed`/`busy`/`no-answer`).

### Record a Call

**Python**
```python
call = client.calls.create(
    from_="+15017122661",
    to="+15558675310",
    url="https://yourapp.com/twiml/welcome",
    record=True,
    recording_status_callback="https://yourapp.com/recording-ready"
)
```

**Node.js**
```node
const call = await client.calls.create({
    from: "+15017122661",
    to: "+15558675310",
    url: "https://yourapp.com/twiml/welcome",
    record: true,
    recordingStatusCallback: "https://yourapp.com/recording-ready",
});
```

Recordings accessible at `https://api.twilio.com/2010-04-01/Accounts/{AccountSid}/Recordings`.

---

## Answering Machine Detection (AMD)

Detect whether a human or voicemail answers before connecting an agent or leaving a message. Two modes:

| Mode | Behavior | Best for |
|------|----------|----------|
| `Enable` | Returns result immediately when human/machine first detected | **Sales dialers** — connect agent to humans ASAP |
| `DetectMessageEnd` | Waits for voicemail greeting to finish (beep/silence) | **Leaving voicemail** — wait for beep, then play/record message |

**Python — Sales dialer (connect agents to live answers only)**
```python
call = client.calls.create(
    from_="+15017122661",
    to="+15558675310",
    url="https://yourapp.com/handle-answer",
    machine_detection="Enable",  # Immediate detection — use for live-agent connect
    async_amd=True,              # Non-blocking — call connects while AMD analyzes
    async_amd_status_callback="https://yourapp.com/amd-result",
    async_amd_status_callback_method="POST"
)
```

**Node.js**
```javascript
const call = await client.calls.create({
    from: "+15017122661",
    to: "+15558675310",
    url: "https://yourapp.com/handle-answer",
    machineDetection: "Enable",
    asyncAmd: true,
    asyncAmdStatusCallback: "https://yourapp.com/amd-result",
    asyncAmdStatusCallbackMethod: "POST",
});
```

**AMD webhook delivers `AnsweredBy` parameter:**

| Value | Meaning | Action |
|-------|---------|--------|
| `human` | Live person detected | Connect to agent |
| `machine_start` | Machine detected, greeting still playing | Hang up or wait |
| `machine_end_beep` | Voicemail beep heard | Leave message |
| `machine_end_silence` | Silence after greeting | Leave message |
| `fax` | Fax tone detected | Hang up |
| `unknown` | Could not determine | Treat as human or retry |

**Conference-based agent bridging** (production pattern for sales dialers):

```python
# In /handle-answer webhook — when AMD says "human":
response = VoiceResponse()
dial = response.dial()
dial.conference(
    f"Sales-{call_sid}",
    start_conference_on_enter=False,  # Prospect waits for agent
    end_conference_on_exit=True
)

# Separately, dial agent into same conference:
client.calls.create(
    from_="+15017122661",
    to=agent_number,
    twiml=f'<Response><Dial><Conference startConferenceOnEnter="true">Sales-{call_sid}</Conference></Dial></Response>'
)
```

This pattern gives you call control (whisper to agent before connecting, supervisor barge-in, hold) that direct `<Dial>` does not.

**Cost**: AMD adds ~$0.0075 per call to standard voice pricing.

---

## Response Fields

| Field | Description |
|-------|-------------|
| `sid` | Call identifier (`CA...`) |
| `status` | `queued`, `ringing`, `in-progress`, `completed`, `failed`, `busy`, `no-answer` |
| `duration` | Length in seconds (after completion) |
| `price` | Cost (populated after completion) |

---

## Common Errors

| Code | Meaning | Fix |
|------|---------|-----|
| 21211 | Invalid `to` number | Validate E.164 format |
| 13224 | Invalid TwiML at webhook URL | Validate TwiML response from your server |
| 13225 | TwiML URL returned non-200 status | Fix your webhook endpoint |

---

## CANNOT

- **Cannot use a private TwiML URL** — Must be publicly accessible. Use ngrok for local development only; deploy to a cloud provider for production.
- **Cannot exceed ~4,096 characters in inline `twiml` parameter** — Use a TwiML URL for longer responses
- **Cannot call unverified numbers on trial accounts** — Upgrade to paid or verify recipient numbers first
- **AMD accuracy is ~85-90%** — Expect some false positives/negatives. Tune `machineDetectionSpeechThreshold` (default 2400ms) based on your results.
- **AMD adds latency** — `Enable` mode returns results faster but may misclassify short greetings. `DetectMessageEnd` is more accurate but adds seconds before your app can act.
- **Cannot use AMD with SIP Trunking** — AMD is only available on calls made via the Calls API

---

## Next Steps

- **TwiML verb reference (Say, Gather, Dial, Record, Conference, Pay):** `twilio-voice-twiml`
- **AI voice agents with real-time speech/LLM:** `twilio-voice-conversation-relay`
- **Improve answer rates (Branded Calling, STIR/SHAKEN):** `twilio-numbers-senders`