action-creator

bởi anthropic

Tạo các mẫu hành động một cú nhấp chuột dành riêng cho người dùng, thực thi các thao tác email khi được nhấp trong giao diện trò chuyện. Sử dụng khi người dùng muốn các hành động có thể tái sử dụng cho…

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

Action Creator

Creates TypeScript action template files that define reusable, user-specific operations users can execute with one click in the chat interface.

When to Use This Skill

Use this skill when the user wants to:

  • Create reusable actions for their specific workflows ("I often need to send payment reminders to ACME Corp")
  • Set up one-click operations for their vendors/customers ("Forward bugs to engineering team")
  • Automate repetitive email tasks with their specific context ("Archive newsletters from TechCrunch/Morning Brew")
  • Build personalized email management tools for their business processes

Key difference from listeners: Actions are user-triggered (clicked in chat), while listeners are event-triggered (automatic).

How Actions Work

Actions are TypeScript files in agent/custom_scripts/actions/ that:

  1. Export a config object defining the template metadata and parameter schema
  2. Export a handler function that executes the operation with given parameters
  3. Use ActionContext methods to perform operations (email API, send emails, call AI, etc.)

The agent creates action instances during conversation by providing specific parameters to these templates, which appear as clickable buttons in the chat.

Creating an Action Template

1. Understand User-Specific Workflow

Parse the user's request to identify:

  • User context: Who are their specific vendors/customers/teams?
  • Operation: What specific action do they need? (send to ACME Corp, forward to engineering team, etc.)
  • Parameters: What varies per execution? (invoice number, priority level, days old)
  • Frequency: How often will they use this?

2. Write the Action Template File

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

import type { ActionTemplate, ActionContext, ActionResult } from "../types";

export const config: ActionTemplate = {
  id: "unique_action_id",                    // kebab-case, user-specific
  name: "Human Readable Name",                // For UI display
  description: "What this action does",       // Explain the operation
  icon: "📨",                                 // Optional emoji icon
  parameterSchema: {
    type: "object",
    properties: {
      paramName: {
        type: "string",                      // or "number", "boolean"
        description: "Parameter description",
        enum: ["option1", "option2"],        // Optional: restrict values
        default: "defaultValue"              // Optional: default value
      }
    },
    required: ["paramName"]                  // List required parameters
  }
};

export async function handler(
  params: Record<string, any>,
  context: ActionContext
): Promise<ActionResult> {
  const { paramName } = params;

  context.log(`Starting action: ${config.name}`);

  try {
    // 1. Perform operations using context methods
    // 2. Use AI for intelligent processing if needed
    // 3. Update emails, send emails, etc.

    context.notify("Action completed successfully", {
      type: "success",
      priority: "normal"
    });

    return {
      success: true,
      message: "Action completed successfully",
      data: { /* optional structured data */ },
      refreshInbox: true  // Optional: refresh inbox after action
    };
  } catch (error: any) {
    context.log(`Action failed: ${error}`, "error");
    return {
      success: false,
      message: `Failed: ${error.message}`
    };
  }
}

3. File Naming Convention

Use kebab-case that reflects the user-specific operation:

  • send-payment-reminder-to-acme.ts (not send-email.ts)
  • forward-bugs-to-engineering.ts (not forward-email.ts)
  • archive-newsletters-from-techcrunch.ts (not archive-emails.ts)
  • summarize-weekly-updates-from-ceo.ts (not summarize-emails.ts)

Important: Templates should be specific to the user's actual workflows, vendors, teams, and processes.

4. Available Context Methods

The ActionContext provides these capabilities:

// Email API operations
const emails = await context.emailAPI.getInbox({ limit: 30, includeRead: false });
const results = await context.emailAPI.searchEmails({ from: "sender@example.com" });
const results = await context.emailAPI.searchWithGmailQuery("from:sender after:2024/01/01");
const emails = await context.emailAPI.getEmailsByIds(["id1", "id2"]);
const email = await context.emailAPI.getEmailById("email-id");

// Direct email operations
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");

// Send emails
const result = await context.sendEmail({
  to: "recipient@example.com",
  subject: "Email subject",
  body: "Email body content",
  cc: "cc@example.com",           // Optional
  bcc: "bcc@example.com",          // Optional
  replyTo: "reply@example.com"     // Optional
});

// AI-powered processing
const analysis = await context.callAgent<ResultType>({
  prompt: "Analyze this email and extract key info...",
  systemPrompt: "You are an expert at...",  // Optional
  tools: ["Read", "WebSearch"],              // Optional
  maxTokens: 2000                            // Optional
});

// Session messaging (inject into chat)
context.addUserMessage("User said this");
context.addAssistantMessage("Assistant responds");
context.addSystemMessage("System notification");

// Notifications
context.notify("Operation completed", {
  priority: "high" | "normal" | "low",
  type: "info" | "success" | "warning" | "error"
});

// External API access
const response = await context.fetch("https://api.example.com/data");
const data = await response.json();

// Logging (visible in server logs)
context.log("Info message", "info");
context.log("Warning message", "warn");
context.log("Error message", "error");

Action Result

Always return an ActionResult object:

return {
  success: true,                          // Required: boolean
  message: "Human-readable result",       // Required: string
  data: { key: "value" },                // Optional: structured data
  suggestedActions: [],                   // Optional: follow-up actions
  refreshInbox: true                      // Optional: refresh inbox
};

Examples and Templates

Reference the template files for common patterns:

Best Practices

  1. User-Specific Templates: Create templates tailored to user's actual vendors, customers, teams, and processes
  2. Descriptive Naming: Use specific names that reflect the operation (not generic like "send-email")
  3. Rich Parameter Schemas: Define clear parameter types with descriptions
  4. AI-Powered: Use context.callAgent() for intelligent processing
  5. Error Handling: Always wrap operations in try-catch and return meaningful errors
  6. Clear Messages: Return human-readable success/failure messages
  7. Idempotency: Design handlers to be safely re-runnable when possible
  8. Logging: Use context.log() for debugging and audit trail

Parameter Schema Guidelines

Define parameters using JSON Schema:

parameterSchema: {
  type: "object",
  properties: {
    // String parameter
    emailId: {
      type: "string",
      description: "Email ID to process"
    },

    // Number parameter with default
    daysOld: {
      type: "number",
      description: "Number of days old",
      default: 30
    },

    // Enum parameter (dropdown)
    priority: {
      type: "string",
      description: "Priority level",
      enum: ["P0 - Critical", "P1 - High", "P2 - Medium", "P3 - Low"]
    },

    // Boolean parameter
    sendNotification: {
      type: "boolean",
      description: "Send notification when complete"
    }
  },
  required: ["emailId", "priority"]  // List required params
}

Creating the File

When the user requests an action template:

  1. Clarify user-specific context:

    • Who are their vendors/customers/teams?
    • What are their specific workflows?
    • What parameters vary per execution?
  2. Write the TypeScript file in agent/custom_scripts/actions/

  3. Use Write tool to create the file with:

    • Proper imports from "../types"
    • User-specific config (not generic)
    • Parameter schema with all required fields
    • Handler with error handling
    • Clear success/failure messages
  4. Test parameters: Ensure all required parameters are defined

  5. Confirm with user that the action matches their workflow

Common Patterns

1. Send Email to Specific Recipient

User-specific → Compose email with template → Send → Return result

const body = `Hi ${recipientName},
Your invoice ${invoiceNumber} for ${amount} is ${daysPastDue} days past due...`;

await context.sendEmail({
  to: "accounts.payable@acmecorp.com",
  subject: `Payment Reminder: Invoice ${invoiceNumber}`,
  body
});

2. Bulk Email Operation

Search emails → Filter → Apply operation to each → Return count

const emails = await context.emailAPI.searchWithGmailQuery(query);
for (const email of emails) {
  await context.archiveEmail(email.messageId);
}
return { success: true, message: `Archived ${emails.length} emails` };

3. AI-Powered Email Processing

Get email → Call AI to analyze → Use AI result → Take action → Return summary

const email = await context.emailAPI.getEmailById(emailId);
const analysis = await context.callAgent({
  prompt: `Analyze this bug report: ${email.body}...`,
  maxTokens: 1000
});
await context.sendEmail({ to: "engineering@company.com", ... });

4. Email Forwarding with Enhancement

Get email → AI analysis → Compose enhanced forward → Send → Label original

const email = await context.emailAPI.getEmailById(emailId);
const analysis = await context.callAgent({ /* analyze */ });
await context.sendEmail({
  to: "team@company.com",
  subject: `[${priority}] ${email.subject}`,
  body: `AI Analysis:\n${analysis}\n\nOriginal:\n${email.body}`
});
await context.addLabel(emailId, "FORWARDED");

Type Imports

Always import types from the correct location:

import type { ActionTemplate, ActionContext, ActionResult } from "../types";

// ActionTemplate: Template metadata and parameter schema
// ActionContext: Runtime context with all capabilities
// ActionResult: Return type for handler function

How Users Trigger Actions

After you create an action template:

  1. Agent discovers template: During conversation, agent reads available actions
  2. Agent creates instance: Agent provides specific parameters for user's situation
  3. User sees button: Action instance appears as clickable button in chat
  4. User clicks: Action executes with pre-filled parameters
  5. Result appears: Success/failure message shown in chat

Example flow:

User: "I need to follow up on the ACME invoice"
Agent: [searches emails, finds Invoice #2024-001 is 15 days overdue]
Agent: Creates action instance with parameters:
  {
    templateId: "send_payment_reminder_acme",
    params: { invoiceNumber: "INV-2024-001", amount: "$5,000", daysPastDue: 15 }
  }
User: [sees button "Send payment reminder to ACME Corp for Invoice #2024-001"]
User: [clicks button]
Action: Executes, sends email, returns "Payment reminder sent to ACME Corp"

Reference

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

  • Complete type definitions
  • ActionsManager implementation
  • WebSocket protocol
  • Frontend integration
  • Advanced examples
  • Logging and audit trail

Thêm skills từ anthropic

access
anthropic
Quản lý quyền truy cập kênh Discord — phê duyệt ghép cặp, chỉnh sửa danh sách cho phép, đặt chính sách DM/nhóm. Sử dụng khi người dùng yêu cầu ghép cặp, phê duyệt ai đó, kiểm tra ai được phép,…
official
session-report
anthropic
Tạo báo cáo HTML có thể khám phá về việc sử dụng phiên Claude Code (token, bộ nhớ đệm, tác nhân phụ, kỹ năng, lời nhắc tốn kém) từ bản ghi ~/.claude/projects.
official
build-mcp-server
anthropic
Kỹ năng này nên được sử dụng khi người dùng yêu cầu "xây dựng một máy chủ MCP", "tạo một MCP", "tạo tích hợp MCP", "bọc một API cho Claude", "cung cấp công cụ cho…
official
cookbook-audit
anthropic
Kiểm tra một notebook trong Anthropic Cookbook dựa trên tiêu chí đánh giá. Sử dụng khi có yêu cầu xem xét hoặc kiểm tra notebook.
official
handle-complaint
anthropic
Xử lý khiếu nại của khách hàng từ đầu đến cuối — lấy ngữ cảnh, soạn phản hồi và đề xuất giải pháp vận hành. Chấp nhận email hoặc ID ticket tùy chọn…
official
use-case-triage
anthropic
Nhanh chóng xác định xem một hoạt động xử lý có cần PIA, DPIA bắt buộc theo GDPR hay có thể tiến hành — phát hiện xung đột chính sách quyền riêng tư và định tuyến đến đúng…
official
board-minutes
anthropic
Soạn thảo biên bản cuộc họp hội đồng hoặc ủy ban theo định dạng của bạn. Tự động phát hiện các cuộc họp hội đồng và ủy ban sắp tới từ lịch của bạn, yêu cầu chương trình họp và…
official
renewal-tracker
anthropic
Hiển thị các hợp đồng có thời hạn hủy sắp đến và cảnh báo trước khi cửa sổ thông báo đóng, làm việc từ sổ đăng ký gia hạn được duy trì. Sử dụng khi người dùng yêu cầu…
official