Help us improve
Share bugs, ideas, or general feedback.
From twilio-developer-kit
Record Twilio voice calls correctly. Covers the critical distinction between Record verb (voicemail) and Dial record (call recording), dual-channel for QA, mid-call pause for PCI, Conference recording, and the ConversationRelay workaround. Use this skill whenever you need to capture call audio for compliance, QA, or analytics.
npx claudepluginhub robinebers/converted-plugins --plugin twilio-developer-kitHow this skill is triggered — by the user, by Claude, or both
Slash command
/twilio-developer-kit:twilio-call-recordingsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Twilio offers multiple recording methods. Choosing the wrong one is the **#1 developer mistake** in voice — using `<Record>` when you mean `<Dial record>` produces voicemail behavior instead of call recording.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Provides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Executes ctx7 CLI to fetch up-to-date library documentation, manage AI coding skills (install/search/generate/remove/suggest), and configure Context7 MCP. Useful for current API refs, skill handling, or agent setup.
Share bugs, ideas, or general feedback.
Twilio offers multiple recording methods. Choosing the wrong one is the #1 developer mistake in voice — using <Record> when you mean <Dial record> produces voicemail behavior instead of call recording.
| Method | What it does | Use when |
|---|---|---|
<Record> verb | Records the CALLER only (voicemail-style) | Leaving a message, capturing input |
<Dial record> | Records BOTH parties on a call | Call recording for two-party calls |
<Start><Recording> | Starts a recording alongside other verbs | ConversationRelay, multi-verb flows |
Conference record | Records the conference mix | Multi-party calls |
| Recordings REST API | Programmatic control mid-call | Pause during payment (PCI) |
twilio-account-setupTWILIO_ACCOUNT_SID and TWILIO_AUTH_TOKEN — see twilio-iam-auth-setuppip install twilio / npm install twiliotwilio-compliance-trafficUse <Dial record> — NOT <Record>.
Python (Flask)
from flask import Flask, request
from twilio.twiml.voice_response import VoiceResponse
app = Flask(__name__)
@app.route("/voice", methods=["POST"])
def incoming_call():
response = VoiceResponse()
response.say("This call may be recorded for quality assurance.")
dial = response.dial(
record="record-from-answer-dual", # dual-channel: agent on one, caller on other
recording_status_callback="https://yourapp.com/recording-status"
)
dial.number("+15558675310") # agent's phone
return str(response)
Node.js (Express)
app.post("/voice", (req, res) => {
const response = new VoiceResponse();
response.say("This call may be recorded for quality assurance.");
const dial = response.dial({
record: "record-from-answer-dual",
recordingStatusCallback: "https://yourapp.com/recording-status",
});
dial.number("+15558675310");
res.type("text/xml").send(response.toString());
});
Security: Validate
X-Twilio-Signatureon recording callbacks in production. Without validation, attackers could POST fake recording URLs to your endpoint.
Python (Flask)
@app.route("/recording-status", methods=["POST"])
def recording_status():
recording_sid = request.form["RecordingSid"]
recording_url = request.form["RecordingUrl"]
call_sid = request.form["CallSid"]
status = request.form["RecordingStatus"] # "completed", "failed"
duration = request.form.get("RecordingDuration", 0)
if status == "completed":
# Store recording reference
save_recording(call_sid, recording_sid, recording_url, duration)
return "", 200
<Dial record>| Mode | What's recorded | Use case |
|---|---|---|
record-from-answer | Single channel, both parties mixed | Simple recording |
record-from-answer-dual | Dual channel — caller on left, agent on right | QA (separate agent/caller audio) |
record-from-ringing | Records from ring, not answer | Capture ring time + full call |
record-from-ringing-dual | Dual channel from ring | QA with ring time |
Always use dual for QA and analytics. Dual-channel lets speech analytics tools (like Conversation Intelligence) distinguish agent from caller.
Record multi-party calls via the Conference:
Python
response = VoiceResponse()
dial = response.dial()
dial.conference(
"support-room-123",
record="record-from-start", # Records from when conference starts
recording_status_callback="https://yourapp.com/conf-recording-status"
)
Note: Conference recording captures the main audio mix. Coach/whisper audio is NOT included. See twilio-conference-calls.
Critical: record:true on the REST API call is silently ignored with ConversationRelay. No error. No recording.
Correct approach: Use <Start><Recording> in TwiML before <Connect>:
Python
@app.route("/voice", methods=["POST"])
def voice():
response = VoiceResponse()
response.say("This call may be recorded.")
# Start recording BEFORE connecting ConversationRelay
start = Start()
start.recording(
recording_status_callback="https://yourapp.com/recording-status",
recording_status_callback_event="completed"
)
response.append(start)
# Now connect ConversationRelay
connect = Connect()
connect.conversation_relay(url="wss://yourapp.com/ws/voice")
response.append(connect)
return str(response)
Node.js
app.post("/voice", (req, res) => {
const response = new VoiceResponse();
response.say("This call may be recorded.");
const start = response.start();
start.recording({
recordingStatusCallback: "https://yourapp.com/recording-status",
recordingStatusCallbackEvent: "completed",
});
const connect = response.connect();
connect.conversationRelay({ url: "wss://yourapp.com/ws/voice" });
res.type("text/xml").send(response.toString());
});
Pause recording when a customer provides payment information:
Python
def pause_recording_for_payment(call_sid, recording_sid):
"""Pause recording during credit card capture."""
client.calls(call_sid).recordings(recording_sid).update(
status="paused"
)
def resume_recording(call_sid, recording_sid):
"""Resume recording after payment processed."""
client.calls(call_sid).recordings(recording_sid).update(
status="in-progress"
)
Node.js
async function pauseForPayment(callSid, recordingSid) {
await client.calls(callSid).recordings(recordingSid).update({
status: "paused",
});
}
async function resumeRecording(callSid, recordingSid) {
await client.calls(callSid).recordings(recordingSid).update({
status: "in-progress",
});
}
PCI DSS: Never record card numbers. Use Twilio's <Pay> verb when possible. If collecting verbally, pause recording for the duration. PCI Mode is IRREVERSIBLE and account-wide — use a sub-account if only some calls need PCI.
Python
# List recordings for a specific call
recordings = client.recordings.list(call_sid=call_sid)
for recording in recordings:
print(f"SID: {recording.sid}")
print(f"Duration: {recording.duration}s")
print(f"URL: https://api.twilio.com{recording.uri.replace('.json', '.mp3')}")
# Download a recording
import requests as req
audio = req.get(
f"https://api.twilio.com/2010-04-01/Accounts/{account_sid}/Recordings/{recording_sid}.mp3",
auth=(account_sid, auth_token)
)
with open("recording.mp3", "wb") as f:
f.write(audio.content)
# Delete a recording (GDPR right to deletion)
client.recordings(recording_sid).delete()
| Feature | Default | Notes |
|---|---|---|
| Storage location | Twilio cloud | Can configure external storage (S3, GCS) |
| Retention | Indefinite | Delete manually via API or set auto-delete policy |
| Formats | WAV (default), MP3 | Request MP3 by appending .mp3 to URL |
| Encryption | At rest | Additional encryption with PCI Mode |
| Symptom | Cause | Fix |
|---|---|---|
| Recording captures only caller (no agent) | Used <Record> verb instead of <Dial record> | Switch to <Dial record="record-from-answer"> |
| No recording at all | Used REST API record:true with ConversationRelay | Use <Start><Recording> in TwiML |
| Recording is empty / silent | Webhook endpoint unreachable, recording never started | Check StatusCallback URL reachability |
| Recording has both parties on same channel | Used record-from-answer (mono) | Use record-from-answer-dual for separate channels |
| Coach audio missing from conference recording | Expected behavior — coach audio isn't in the mix | Record coach's call leg separately |
recordingTrack has no observable effect via TwiML — The <Start><Recording> TwiML parameter recordingTrack does not isolate tracks. Use the Recordings REST API with recordingTrack for actual track isolation.record:true is silently ignored ("not eligible for recording"). Must use <Start><Recording> before <Connect> in TwiML.update with status="paused" or status="in-progress").Record=true defaults to mono. Must specify recordingChannels: 'dual'.<Record> verb for call recording — <Record> captures the caller only (voicemail-style). Use <Dial record> or <Start><Recording> for call recording.twilio-conference-callstwilio-taskrouter-routingtwilio-compliance-traffictwilio-debugging-observability