twilio-taskrouter-routing
$
npx mdskill add openai/plugins/twilio-taskrouter-routingRoute tasks to agents using Twilio TaskRouter's skills-based routing engine
- Solve multi-agent routing for contact centers, support queues, and AI escalations
- Uses Twilio Workers, Task Queues, Workflows, and Reservations APIs
- Matches tasks to workers based on skills, availability, and priority rules
- Delivers assignments via reservations that agents can accept or reject
SKILL.md
.github/skills/twilio-taskrouter-routingView on GitHub ↗
---
name: twilio-taskrouter-routing
description: >
Route tasks to agents using Twilio TaskRouter. Covers Workers, Task
Queues, Workflows, Reservations, skills-based routing, and common
gotchas (hyphen attributes, HAS operator, reservation cascade). Use this
skill for any multi-agent contact center, support queue, or AI agent
escalation routing.
---
## Overview
TaskRouter is Twilio's skills-based routing engine. Instead of building custom queuing logic, you define Workers (agents), Task Queues (groups), and Workflows (routing rules). TaskRouter matches incoming tasks to the best available worker.
```
Incoming Task → Workflow (routing rules) → Task Queue (skill match) → Worker (agent)
↓
Reservation
(accept/reject)
```
**Common mistake:** Developers reinvent TaskRouter in custom Node.js — don't. If you're building skills-based routing, queue management, or agent assignment, use TaskRouter.
---
## Prerequisites
- Twilio account — see `twilio-account-setup`
- `TWILIO_ACCOUNT_SID` and `TWILIO_AUTH_TOKEN` — see `twilio-iam-auth-setup`
- SDK: `pip install twilio` / `npm install twilio`
- For voice routing: a Twilio phone number with webhook configured — see `twilio-voice-twiml`
- For AI escalation: ConversationRelay with escalation tools — see `twilio-voice-conversation-relay`
---
## Quickstart
**Step 1 — Create a Workspace**
A Workspace is the top-level container for all TaskRouter resources.
**Python**
```python
import os
from twilio.rest import Client
client = Client(os.environ["TWILIO_ACCOUNT_SID"], os.environ["TWILIO_AUTH_TOKEN"])
workspace = client.taskrouter.v1.workspaces.create(
friendly_name="Support Center",
event_callback_url="https://yourapp.com/taskrouter-events"
)
workspace_sid = workspace.sid # WSxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
print(workspace_sid)
```
**Node.js**
```node
const twilio = require("twilio");
const client = twilio(process.env.TWILIO_ACCOUNT_SID, process.env.TWILIO_AUTH_TOKEN);
const workspace = await client.taskrouter.v1.workspaces.create({
friendlyName: "Support Center",
eventCallbackUrl: "https://yourapp.com/taskrouter-events",
});
const workspaceSid = workspace.sid;
```
**Step 2 — Create Activities (agent states)**
**Python**
```python
# Available — worker can receive tasks
available = client.taskrouter.v1.workspaces(workspace_sid).activities.create(
friendly_name="Available", available=True
)
# Offline — worker cannot receive tasks
offline = client.taskrouter.v1.workspaces(workspace_sid).activities.create(
friendly_name="Offline", available=False
)
# On a task — worker is busy
on_task = client.taskrouter.v1.workspaces(workspace_sid).activities.create(
friendly_name="On Task", available=False
)
```
**Step 3 — Create Workers (agents)**
> **Security:** Always use `json.dumps()` (Python) or `JSON.stringify()` (Node.js) to construct attribute payloads. String interpolation is vulnerable to JSON injection.
**Python**
```python
worker = client.taskrouter.v1.workspaces(workspace_sid).workers.create(
friendly_name="Alice",
attributes='{"skills": ["billing", "technical"], "languages": ["en", "es"], "department": "support"}'
)
```
**Node.js**
```node
const worker = await client.taskrouter.v1.workspaces(workspaceSid).workers.create({
friendlyName: "Alice",
attributes: JSON.stringify({
skills: ["billing", "technical"],
languages: ["en", "es"],
department: "support",
}),
});
```
**Step 4 — Create Task Queues**
**Python**
```python
# Billing queue — matches workers with "billing" skill
billing_queue = client.taskrouter.v1.workspaces(workspace_sid).task_queues.create(
friendly_name="Billing",
target_workers='skills HAS "billing"'
)
# Technical queue
tech_queue = client.taskrouter.v1.workspaces(workspace_sid).task_queues.create(
friendly_name="Technical",
target_workers='skills HAS "technical"'
)
# Catch-all queue
default_queue = client.taskrouter.v1.workspaces(workspace_sid).task_queues.create(
friendly_name="Default",
target_workers='1==1' # matches all workers
)
```
**Step 5 — Create a Workflow (routing rules)**
**Python**
```python
import json
workflow_config = {
"task_routing": {
"filters": [
{
"filter_friendly_name": "Billing",
"expression": "department == 'billing'",
"targets": [
{"queue": billing_queue.sid, "timeout": 120}
]
},
{
"filter_friendly_name": "Technical",
"expression": "department == 'technical'",
"targets": [
{"queue": tech_queue.sid, "timeout": 120}
]
}
],
"default_filter": {
"queue": default_queue.sid
}
}
}
workflow = client.taskrouter.v1.workspaces(workspace_sid).workflows.create(
friendly_name="Support Routing",
configuration=json.dumps(workflow_config),
assignment_callback_url="https://yourapp.com/assignment"
)
```
**Step 6 — Create a Task (from an incoming call)**
**Python**
```python
task = client.taskrouter.v1.workspaces(workspace_sid).tasks.create(
attributes='{"department": "billing", "caller": "+15558675310", "priority": 1}',
workflow_sid=workflow.sid
)
```
**Step 7 — Handle the Assignment Callback**
When TaskRouter finds a matching worker, it POSTs to your `assignment_callback_url`:
**Python (Flask)**
```python
@app.route("/assignment", methods=["POST"])
def assignment():
task_sid = request.form["TaskSid"]
worker_sid = request.form["WorkerSid"]
reservation_sid = request.form["ReservationSid"]
# Option A: Dequeue to the worker's phone
return jsonify({
"instruction": "dequeue",
"from": "+15551234567", # your Twilio number
"post_work_activity_sid": available_activity_sid
})
# Option B: Conference the caller and agent
# return jsonify({
# "instruction": "conference",
# "from": "+15551234567",
# "post_work_activity_sid": available_activity_sid
# })
```
**Node.js (Express)**
```node
app.post("/assignment", (req, res) => {
res.json({
instruction: "dequeue",
from: "+15551234567",
post_work_activity_sid: availableActivitySid,
});
});
```
---
## Key Patterns
### Skills-Based Routing
Match tasks to workers based on attributes:
| Worker expression | Matches |
|-------------------|---------|
| `skills HAS "billing"` | Workers whose `skills` array contains "billing" |
| `languages HAS "es"` | Spanish-speaking workers |
| `department == "support"` | Workers in support department |
| `experience > 5` | Workers with 5+ years experience |
| `skills HAS "billing" AND languages HAS "es"` | Spanish-speaking billing agents |
### Priority Routing
Tasks with higher priority are assigned first:
```python
# VIP customer — priority 10 (higher = first)
task = client.taskrouter.v1.workspaces(workspace_sid).tasks.create(
attributes='{"department": "billing", "priority": 10, "vip": true}',
workflow_sid=workflow.sid,
priority=10
)
```
### AI Agent Escalation
When an AI agent (via TAC) escalates to a human, create a TaskRouter task with the AI's context:
```python
# From your escalation webhook handler
def handle_escalation(escalation_data):
task = client.taskrouter.v1.workspaces(workspace_sid).tasks.create(
attributes=json.dumps({
"department": escalation_data["reason_code"],
"conversation_id": escalation_data["conversation_id"],
"profile_id": escalation_data["profile_id"],
"ai_summary": escalation_data["summary"],
"priority": 5
}),
workflow_sid=workflow.sid
)
```
The human agent receives the AI's conversation summary and customer profile.
### Workflow with Timeout Escalation
Route to specialized queue first, then overflow to general:
```python
workflow_config = {
"task_routing": {
"filters": [
{
"filter_friendly_name": "Billing Specialist First",
"expression": "department == 'billing'",
"targets": [
{"queue": billing_queue.sid, "timeout": 60}, # Try billing queue for 60s
{"queue": default_queue.sid, "timeout": 120} # Overflow to general
]
}
],
"default_filter": {
"queue": default_queue.sid
}
}
}
```
### Worker Activity Management
```python
# Set worker to available
client.taskrouter.v1.workspaces(workspace_sid) \
.workers(worker_sid) \
.update(activity_sid=available_activity_sid)
# Get real-time worker statistics
stats = client.taskrouter.v1.workspaces(workspace_sid) \
.workers \
.statistics() \
.fetch()
print(f"Available: {stats.realtime['total_available_workers']}")
```
---
## Scale Guidance
| Agents | Architecture | Notes |
|--------|-------------|-------|
| < 10 | Single workflow, one queue per skill | No Flex needed — agents use phone |
| 10-50 | Multi-queue workflows, skills-based routing | Flex recommended for desktop |
| 50+ | Multi-tier workflows, priority routing, real-time monitoring | Full Flex + supervisor tools |
---
## Gotchas
### 1. Hyphens in Attribute Names Break Silently
```python
# WRONG — hyphens in attribute keys break workflow expressions
worker = client.taskrouter.v1.workspaces(workspace_sid).workers.create(
friendly_name="Alice",
attributes='{"skill-level": 5}' # hyphen breaks expression evaluation
)
# RIGHT — use underscores or camelCase
worker = client.taskrouter.v1.workspaces(workspace_sid).workers.create(
friendly_name="Alice",
attributes='{"skill_level": 5}'
)
```
No error — the expression silently fails to match.
### 2. HAS Operator on Non-Array Attributes
```python
# WRONG — "billing" is a string, not an array. HAS silently matches nothing.
target_workers = 'department HAS "billing"'
# RIGHT — use == for string attributes
target_workers = 'department == "billing"'
# RIGHT — use HAS only for arrays
target_workers = 'skills HAS "billing"' # skills: ["billing", "technical"]
```
Tasks sit in queue forever with no error.
### 3. Reservation Timeout Cascade
When a reservation times out:
1. Worker moves to the timeout Activity (often "Offline")
2. Fewer workers available → other reservations also time out
3. Positive feedback loop → entire queue backs up
**Fix:** Set the timeout Activity to a short-duration state, not "Offline". Or implement a reservation timeout handler that keeps the worker available:
```python
@app.route("/taskrouter-events", methods=["POST"])
def taskrouter_event():
event_type = request.form["EventType"]
if event_type == "reservation.timeout":
worker_sid = request.form["WorkerSid"]
# Keep worker available instead of moving to offline
client.taskrouter.v1.workspaces(workspace_sid) \
.workers(worker_sid) \
.update(activity_sid=available_activity_sid)
return "", 200
```
### 4. Activity Available Flag
Updating an Activity's `available` flag returns 200 OK but may not change the value if workers are currently in that activity. Create new activities instead of modifying existing ones.
---
## CANNOT
- **Hyphens in attribute names break expressions** — `skill-level` is treated as subtraction (`skill` minus `level`). Error 20001. Always use underscores: `skill_level`.
- **`HAS` on non-array silently matches nothing** — `department HAS "billing"` on a string attribute is accepted at creation but never matches. Tasks sit in queue forever with no error.
- **Expression validation is syntactic only** — Queue creation validates parse but NOT worker matching. Semantically wrong expressions create successfully with zero matching workers.
- **Activity `available` flag is silently immutable** — Updating returns 200 OK but does not change the value. Must delete and recreate the Activity.
- **`multiTaskEnabled` cannot be reverted to false** — Once enabled on a Workspace, cannot be disabled. One-way door.
- **Reservation timeout moves worker to timeout Activity** — Worker automatically moved to Offline. Must manually set back. This cascades: fewer available workers → more timeouts → queue collapse. See Gotcha #3.
- **Workflow target timeout auto-cancels tasks** — When all targets exhaust timeouts, task is canceled. Always include a `default_filter` as catch-all.
- **Worker `friendlyName` is case-insensitive unique** — "alice" collides with "Alice".
- **`workflowSid` is required for task creation** — API does not auto-select a default Workflow.
- **Cannot update task status and attributes in same request** — Must be two separate API calls.
- **Assignment callback must respond in 5 seconds** — If both primary and fallback URLs fail, reservation is canceled.
- **Tasks auto-cancel after 1,000 rejections** — If a task cycles through 1,000 reservation rejections, it is automatically canceled.
- **`page` query param not supported** — Use `PageToken` for pagination. `page` returns error 40153.
- **Cannot use malformed JSON in worker attributes** — Silently breaks matching with no error
- **Cannot use regex in workflow expressions** — Only supports ==, !=, <, >, HAS, IN, CONTAINS, AND, OR, NOT
- **Cannot exceed 50,000 Workers per Workspace** — Hard limit
- **Cannot exceed 250 Task Queues per Workspace** — Hard limit
- **Cannot delay reservation callback response beyond 15 seconds** — Timeout results in reservation failure
---
## Next Steps
- **Conference for transfers:** `twilio-conference-calls`
- **Call recording:** `twilio-call-recordings`
- **AI agent voice integration:** `twilio-voice-conversation-relay`
- **Voice IVR before routing:** `twilio-voice-twiml`
More from openai/plugins
- accessibility-and-inclusive-visualizationMake data visualizations accessible and inclusive. Use when the user needs chart or diagram accessibility guidance, text alternatives for complex visuals, color and contrast review, keyboard support, reduced-motion behavior for animation or parallax, or an accessibility QA workflow for exported figures, UML-like diagrams, and dashboards.
- agent-browserBrowser automation CLI for AI agents. Use when the user needs to interact with websites, verify dev server output, test web apps, navigate pages, fill forms, click buttons, take screenshots, extract data, or automate any browser task. Also triggers when a dev server starts so you can verify it visually.
- agent-browser-verifyAutomated browser verification for dev servers. Triggers when a dev server starts to run a visual gut-check with agent-browser — verifies the page loads, checks for console errors, validates key UI elements, and reports pass/fail before continuing.
- agents-sdkBuild AI agents on Cloudflare Workers using the Agents SDK. Load when creating stateful agents, durable workflows, real-time WebSocket apps, scheduled tasks, MCP servers, or chat applications. Covers Agent class, state management, callable RPC, Workflows integration, and React hooks. Biases towards retrieval from Cloudflare docs over pre-trained knowledge.
- ai-elementsAI Elements component library guidance — pre-built React components for AI interfaces built on shadcn/ui. Use when building chat UIs, message displays, tool call rendering, streaming responses, reasoning panels, or any AI-native interface with the AI SDK.
- ai-gatewayVercel AI Gateway expert guidance. Use when configuring model routing, provider failover, cost tracking, or managing multiple AI providers through a unified API.
- ai-generation-persistenceAI generation persistence patterns — unique IDs, addressable URLs, database storage, and cost tracking for every LLM generation
- ai-sdkVercel AI SDK expert guidance. Use when building AI-powered features — chat interfaces, text generation, structured output, tool calling, agents, MCP integration, streaming, embeddings, reranking, image generation, or working with any LLM provider.
- aiq-deploy|
- aiq-research|