← 返回 Skills 市场

Backend Developer

Name: Backend Developer
Author: jawadsadiq01

作者 JawadSadiq01 · GitHub ↗ · v1.0.0

cross-platform ⚠ suspicious

963

总下载

当前安装

版本数

在 OpenClaw 中安装

/install langchain-email-agent

功能描述

Compose and send emails using AI with human approval for sending, allowing edits or rejection before delivery.

使用说明 (SKILL.md)

EmailAgent README

Overview

The EmailAgent class provides an AI-powered email composition and sending capability using LangChain with OpenAI's GPT model. It includes human-in-the-loop middleware that requires approval before emails are sent.

Configuration

Environment Variables

Variable	Default	Description
`OPENAI_MODEL`	`gpt-4o-mini`	OpenAI model to use

Usage

import { EmailAgent } from './email.agent';
import { SendEmailDto } from '../dto/send-email.dto';

const agent = new EmailAgent();

const dto: SendEmailDto = {
  email: '[email protected]',
  name: 'John Doe',
  subject: 'Meeting Request',      // optional
  body: 'Initial email content',   // optional
  instructions: 'Keep it formal'   // optional
};

const result = await agent.sendEmail(dto);

Human-in-the-Loop Middleware

The agent uses humanInTheLoopMiddleware which interrupts execution on the EmailTool before sending emails. This allows for:

approve - Send the email as composed
edit - Modify the email before sending
reject - Cancel the email operation

The readEmailTool is excluded from interruption (false), allowing read operations to proceed without approval.

Parameters

Parameter	Type	Required	Description
`email`	string	Yes	Recipient email address
`name`	string	Yes	Recipient name
`subject`	string	No	Email subject line
`body`	string	No	Initial email body content
`instructions`	string	No	AI instructions for composing the email

Return Value

Returns the final message content from the agent as a string.

安全使用建议

This package claims to be an email-sending agent but the bundle is incomplete and metadata is sparse: key modules (EmailTool, DTOs, message templates) are missing and required credentials (OpenAI API key and likely SMTP/email provider credentials) are not declared. Before installing or enabling it: 1) ask the publisher for the missing files and a clear description of what credentials are required and why; 2) review the implementation of the send_email.tool and any email/SMTP integration to ensure it won't send messages without explicit approval; 3) require explicit environment-variable documentation (including OpenAI API key usage) and check where any credentials would be sent or stored; 4) prefer packages with a homepage, repository, or known publisher; 5) if you must test, run it in a restricted/sandboxed environment and do not provide real credentials until you’ve reviewed the missing code. The inconsistencies make this suspicious rather than clearly benign.

功能分析

Type: OpenClaw Skill Name: langchain-email-agent Version: 1.0.0 The skill is designed to compose and send emails using an AI agent. While it accepts user-provided instructions that could be a vector for prompt injection, the critical safeguard is the `humanInTheLoopMiddleware` applied to the `EmailTool` in `email.agent.ts`. This middleware explicitly requires human approval (approve, edit, or reject) before any email is sent, effectively preventing unauthorized email sending or data exfiltration via email, even if the AI agent is compromised by prompt injection. The `SKILL.md` clearly documents this human-in-the-loop process.

能力评估

⚠ Purpose & Capability

The skill bundle's name is 'Backend Developer' and no description/homepage is provided, but the README and code implement an EmailAgent. This naming/metadata mismatch is confusing. The code imports an EmailTool, DTOs, and message constants from relative paths that are not included in the package, so the included file alone cannot perform the stated email-sending function.

⚠ Instruction Scope

SKILL.md describes an EmailAgent with human-in-the-loop approval before sending, and the code enforces middleware interrupt for EmailTool. The instructions do not request unrelated data. However, the runtime instructions and code rely on external modules (send_email.tool, dto, messages) that are absent from the bundle, so actual behavior is undefined and could change depending on the missing implementations.

✓ Install Mechanism

There is no install specification (instruction-only + one code file). That minimizes installer-level risk because nothing is downloaded or extracted by the skill installer.

⚠ Credentials

SKILL.md documents an OPENAI_MODEL env var but the package manifest declares no required environment variables. The code constructs a ChatOpenAI instance which in practice requires OpenAI credentials (e.g., OPENAI_API_KEY) but that credential is not declared. Additionally, sending email normally requires SMTP/integration credentials (not declared). The skill asks for no credentials yet can't function without at least an OpenAI API key and likely email provider credentials — a mismatch that should be clarified.

✓ Persistence & Privilege

The skill does not request always:true and uses default autonomous invocation settings (disable-model-invocation: false). Autonomous invocation plus the ability to send emails is a potential risk if the skill actually has send capability, but on its own the persistence/privilege settings are normal.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install langchain-email-agent
安装完成后，直接呼叫该 Skill 的名称或使用 /langchain-email-agent 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

- Initial release of EmailAgent with AI-powered email composition and sending - Integrated human-in-the-loop middleware for approval, editing, or rejection of outgoing emails - Supports configuration of OpenAI model via environment variable (default: gpt-4o-mini) - Allows email composition customization with optional subject, body, and instructions fields - Read-only email operations proceed without interruption or approval

元数据

Slug langchain-email-agent

版本 1.0.0

许可证 —

累计安装 0

当前安装数 0

历史版本数 1

常见问题