twilio-voice-outbound-calls — Codex skill: what it does
Guides agents in placing outbound phone calls using Twilio Programmable Voice REST API with AMD and TwiML webhooks.
npx skills add openai/plugins --skill twilio-voice-outbound-calls -a codexWhat twilio-voice-outbound-calls does
The skill provides patterns for initiating outbound voice calls using calls.create() in Python and Node.js. It explains how to pass inline TwiML or route call handling to webhook servers built with Flask or Express. You can track call states from queued to completed and trigger call recordings with status callbacks.
Answering Machine Detection (AMD) is covered for dialer workflows, including the instant Enable mode and DetectMessageEnd for leaving messages after the tone. Trial accounts can only place calls to verified phone numbers. Running these calls requires the official SDK plus the TWILIO_ACCOUNT_SID and TWILIO_AUTH_TOKEN environment variables.
Because outbound calls incur billing fees and cannot be undone, the agent must ask for explicit user confirmation of the target number before dialing.
What it helps with
- trigger outbound phone calls using Python or Node.js Twilio SDKs
- differentiate between human answers and voicemail using AMD
- serve dynamic TwiML instructions via Flask or Express webhooks
- enable call recording and handle status tracking callbacks
Example prompts for the agent
- Write a Python script to initiate an outbound call using calls.create() and inline TwiML
- Set up an Express webhook to handle Twilio VoiceResponse with Gather
- Configure answering machine detection with async AMD status callbacks in Node.js
How to install twilio-voice-outbound-calls
1Skills CLI (Vercel)
npx skills add openai/plugins --skill twilio-voice-outbound-calls -a codex2Manually: put the files in place
.agents/skills/twilio-voice-outbound-calls/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add openai/plugins --skill twilio-voice-outbound-calls -a claude-code2Manually: put the files in place
~/.claude/skills/twilio-voice-outbound-calls/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/twilio-voice-outbound-calls/
Manually: put the files in place
.cursor/skills/twilio-voice-outbound-calls/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install openai/plugins twilio-voice-outbound-calls2Manually: put the files in place
.github/skills/twilio-voice-outbound-calls/SKILL.mdPlatform CLI
gemini skills install https://github.com/openai/plugins --path plugins/twilio-developer-kit/skills/twilio-voice-outbound-callsManually: put the files in place
.devin/skills/twilio-voice-outbound-calls/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: Codex skills: how to install, where they live, how to call them
Questions about twilio-voice-outbound-calls
How to install twilio-voice-outbound-calls in Codex?
Run the command npx skills add openai/plugins --skill twilio-voice-outbound-calls -a codex. The skill file will be placed in .agents/skills/twilio-voice-outbound-calls/SKILL.md.
What does twilio-voice-outbound-calls do?
It equips the AI coding agent with implementation patterns for Twilio Programmable Voice. It guides call initiation, status webhooks, answering machine detection, and call recording.
What is required before placing outbound calls with this skill?
You need a Twilio account with a voice-enabled number and configured TWILIO_ACCOUNT_SID and TWILIO_AUTH_TOKEN environment variables. Trial accounts can only dial verified phone numbers.
Source text
SKILL.md · MIT
---
name: twilio-voice-outbound-calls
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.
---
## 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`
© Twilio · MIT · openai/plugins