listener-creator

작성자: anthropic

특정 조건(예: 상사로부터의 긴급 이메일, 보관할 뉴스레터, 패키지 추적)을 모니터링하고 실행하는 이벤트 기반 이메일 리스너를 생성합니다.

npx skills add https://github.com/anthropics/claude-agent-sdk-demos --skill listener-creator

Listener Creator

Creates TypeScript listener files that monitor email events and execute custom logic when conditions are met.

When to Use This Skill

Use this skill when the user wants to:

  • Get notifications about specific emails ("notify me when boss sends urgent emails")
  • Automatically handle certain emails ("auto-archive newsletters")
  • Monitor for patterns ("watch for package tracking emails")
  • Set up scheduled actions ("daily email summary at 9am")
  • Create custom email workflows

How Listeners Work

Listeners are TypeScript files in agent/custom_scripts/listeners/ that:

  1. Export a config object defining the event type and metadata
  2. Export a handler function that filters and processes events
  3. Use ListenerContext methods to perform actions (notify, archive, star, etc.)

The system automatically loads enabled listeners and executes them when matching events occur.

Creating a Listener

1. Understand the User's Intent

Parse the user's request to identify:

  • Event type: What triggers this listener? (email_received, email_sent, email_starred, email_archived, email_labeled, scheduled_time)
  • Filter conditions: What specific emails/events to match? (sender, subject keywords, time-based)
  • Actions: What should happen? (notify, archive, star, mark as read, add label)
  • Priority: How urgent is this? (high/normal/low)

2. Choose an Event Type

// Available event types:
- "email_received"  // Most common - new email arrives
- "email_sent"      // User sends an email
- "email_starred"   // Email is starred
- "email_archived"  // Email is archived
- "email_labeled"   // Label added to email
- "scheduled_time"  // Time-based (cron) - requires scheduler setup

3. Write the Listener File

Create a file in agent/custom_scripts/listeners/ with this structure:

import type { ListenerConfig, Email, ListenerContext } from "../types";

export const config: ListenerConfig = {
  id: "unique_listener_id",           // kebab-case, descriptive
  name: "Human Readable Name",         // For UI display
  description: "What this does",       // Optional but helpful
  enabled: true,                       // Start enabled
  event: "email_received"              // Event type
};

export async function handler(email: Email, context: ListenerContext): Promise<void> {
  // 1. Basic filter (identity/sender only)
  if (!email.from.includes("example@email.com")) return;

  // 2. Use AI for intelligent classification (PREFERRED over keyword matching)
  const analysis = await context.callAgent<{ isUrgent: boolean; reason: string }>({
    prompt: `Is this email urgent?\nSubject: ${email.subject}\nBody: ${email.body.substring(0, 500)}`,
    schema: {
      type: "object",
      properties: {
        isUrgent: { type: "boolean" },
        reason: { type: "string" }
      },
      required: ["isUrgent", "reason"]
    },
    model: "haiku"
  });

  if (!analysis.isUrgent) return;

  // 3. Perform actions via context methods
  await context.notify(`Urgent email: ${email.subject}\n${analysis.reason}`, {
    priority: "high"
  });

  await context.starEmail(email.messageId);
}

4. File Naming Convention

Use kebab-case matching the listener's purpose:

  • boss-urgent-watcher.ts
  • auto-archive-newsletters.ts
  • package-tracking.ts
  • daily-summary.ts

5. Available Context Methods

The ListenerContext provides these methods:

// Notifications
await context.notify(message, { priority: "high" | "normal" | "low" });

// Email actions
await context.archiveEmail(emailId);
await context.starEmail(emailId);
await context.unstarEmail(emailId);
await context.markAsRead(emailId);
await context.markAsUnread(emailId);
await context.addLabel(emailId, "label-name");
await context.removeLabel(emailId, "label-name");

// AI-powered analysis
const result = await context.callAgent<ResultType>({
  prompt: "Your prompt with email content",
  schema: {
    type: "object",
    properties: { field: { type: "string" } },
    required: ["field"]
  },
  model: "haiku" // or "sonnet" or "opus"
});

Recommended Approach: AI-Powered Classification

Default to using context.callAgent() for intelligent decision-making instead of hard-coded keyword lists. This provides better accuracy and adaptability.

// PREFERRED: AI-based urgency detection
const analysis = await context.callAgent<{ isUrgent: boolean; reason: string }>({
  prompt: `Analyze if this email is urgent:
Subject: ${email.subject}
Body: ${email.body.substring(0, 500)}

Is this email urgent or time-sensitive? Consider context, not just keywords.`,

  schema: {
    type: "object",
    properties: {
      isUrgent: { type: "boolean" },
      reason: { type: "string" }
    },
    required: ["isUrgent", "reason"]
  },
  model: "haiku" // Fast and cost-effective
});

if (analysis.isUrgent) {
  await context.notify(`Urgent: ${email.subject}\n${analysis.reason}`);
}

// AVOID: Hard-coded keyword lists (brittle and prone to false positives)
// const isUrgent = subject.includes("urgent") || subject.includes("asap");

Examples and Templates

Reference the template files for common patterns:

Best Practices

  1. Prefer AI Classification: Use context.callAgent() instead of hard-coded keyword lists for intelligent decision-making
  2. Filter Early: Return early if email doesn't match basic criteria (like sender)
  3. Clear IDs: Use descriptive, unique listener IDs
  4. Error Handling: Wrap context method calls in try-catch when appropriate
  5. Performance: Use "haiku" model for fast AI classification (< 1 second typical)
  6. Notify Wisely: Only notify when truly important
  7. Avoid Hard-Coded Lists: Let AI determine urgency, importance, or categories instead of keyword matching

Type Imports

Always import types from the correct location:

import type { ListenerConfig, Email, ListenerContext } from "../types";

// For scheduled listeners:
import type { ListenerConfig, ListenerContext } from "../types";

// For labeled event:
import type { ListenerConfig, Email, ListenerContext } from "../types";

Common Patterns

AI-Powered (PREFERRED)

Basic filter (sender/type) → Call AI agent for intelligent classification → Act on AI result → Notify if important

This is the recommended approach for most listeners as it:

  • Avoids brittle keyword matching
  • Adapts to nuanced language and context
  • Makes better decisions about urgency and categorization
  • Reduces false positives

Simple Notification (Use sparingly)

Basic filter (sender only) → Notify → Optional star/label

Only use this when: The trigger is purely identity-based (e.g., "notify me about ALL emails from X")

Auto-Archive

Basic filter → Archive → Mark as read → Optional notify

Scheduled

Run at specific time → Query emails → Analyze → Send summary

Creating the File

When the user requests a listener:

  1. Ask clarifying questions if the intent is unclear:

    • Who is the sender? What keywords?
    • What action should happen?
    • How urgent is this?
  2. Choose the right event type (usually email_received)

  3. Write the TypeScript file in agent/custom_scripts/listeners/

  4. Use Write tool to create the file with:

    • Proper imports
    • Descriptive config
    • Handler with early filtering
    • Appropriate context method calls
  5. Return listener reference in markdown format using [listener:filename.ts] notation (e.g., [listener:boss-urgent-watcher.ts]) for easy parsing and linking in the UI

  6. Confirm with user that the listener matches their intent

Output Format Example

When presenting a created listener to the user, use this format:

Created listener: [listener:boss-urgent-watcher.ts]

This listener will:
- Monitor emails from boss@company.com
- Use AI to detect urgent emails (not just keywords)
- Send high-priority notifications for truly urgent emails
- Star emails that require immediate action

When to Use AI vs Simple Filtering

Use AI (context.callAgent()) when:

  • Detecting urgency, importance, or sentiment
  • Classifying email content or intent
  • Extracting structured data from email bodies
  • Making nuanced decisions based on context
  • Any logic that involves "understanding" the email content

Use simple filtering when:

  • Checking exact sender/recipient
  • Basic pattern matching on email fields (e.g., "from specific domain")
  • Identity-based triggers (e.g., "all emails from X person")

Default to AI unless the filter is purely identity-based.

Scheduled Listeners

For time-based actions (daily summaries, weekly reports):

export const config: ListenerConfig = {
  id: "daily_summary",
  name: "Daily Email Summary",
  enabled: true,
  event: "scheduled_time"
  // Note: Cron schedule configured separately in scheduler
};

export async function handler(
  data: { timestamp: Date },
  context: ListenerContext
): Promise<void> {
  // Your scheduled logic here
  await context.notify("Good morning! Your daily summary...");
}

Note: Scheduled listeners require cron scheduler configuration outside the listener file.

Reference

Full specification: See project root LISTENERS_SPEC.md for complete details on:

  • All event types
  • Complete type definitions
  • ListenersManager implementation
  • Advanced examples
  • Error handling patterns

anthropic의 다른 스킬

analyzing-financial-statements
anthropic
이 스킬은 재무제표 데이터로부터 투자 분석을 위한 주요 재무 비율과 지표를 계산합니다.
applying-brand-guidelines
anthropic
이 스킬은 생성된 모든 문서에 일관된 기업 브랜딩과 스타일(색상, 글꼴, 레이아웃, 메시징 포함)을 적용합니다.
creating-financial-models
anthropic
이 스킬은 DCF 분석, 민감도 테스트, 몬테카를로 시뮬레이션, 시나리오 플래닝을 포함한 고급 재무 모델링 제품군을 투자…에 제공합니다.
board-minutes
anthropic
이사회 또는 위원회 회의록을 사내 형식으로 작성합니다. 캘린더에서 예정된 이사회 및 위원회 회의를 자동으로 감지하고, 안건을 요청한 후…
crm-cleanup
anthropic
HubSpot에서 오래된 거래, 중복 연락처, 누락된 필드를 스캔한 후 소유자가 승인한 항목을 수정합니다. 선택적 범위 인수를 받아 거래, 연락처 등을 지정할 수 있습니다.
redshift-api
anthropic
Amazon Redshift에 대해 SQL 실행 — 명령문 제출, 상태 폴링, 결과 페이지 탐색, 데이터베이스/스키마/테이블 탐색. 사용자가 원할 때마다 이 기능을 사용하세요…
ticket-deflector
anthropic
고객이 전달한 이메일이나 티켓을 읽고, PayPal에서 주문/환불 상태를 가져오며, HubSpot에서 계정 내역을 조회한 후, 소유자의 어조에 맞춰 답변을 작성합니다.
reg-feed-watcher
anthropic
규제 피드를 지금 확인하고, 마지막 확인 이후 새로 추가된 내용을 사용자의 중요도 기준에 따라 필터링하여 보고합니다. 사용자가 "피드 확인해 줘"라고 말할 때 사용하세요.