Token导航 LogoToken导航

Agent Skill · 研究检索

build-product-integrations(构建产品集成)

build-product-integrations 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

build-product-integrations 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为接触敏感数据;许可证信息还不明确,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 40,402。

总安装 1,294周安装 55GitHub Stars 4下载 453核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:build-product-integrations(构建产品集成)
来源仓库:https://github.com/membranehq/agent-skills
仓库路径:skills/build-product-integrations
安装命令:
npx skills add https://github.com/membranehq/agent-skills --skill build-product-integrations
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/membranehq/agent-skills --skill build-product-integrations

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

通过 Membrane 代理构建第三方产品连接器。支持身份验证、API 映射和数据转换。可创建多步骤自动化业务流程。build-product-integrations 属于研究检索类 Skill,可作为该场景下的辅助能力补充。需注册 Membrane 服务获取会话 ID。操作复杂度取决于目标系统接口复杂度。

来源仓库内路径:skills/build-product-integrations

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/membranehq/agent-skills --skill build-product-…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要接触敏感数据,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

build-product-integrations 怎么安装?

在终端执行 npx skills add https://github.com/membranehq/agent-skills --skill build-product-integratio…。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

build-product-integrations 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

build-product-integrations 需要什么权限?

权限级别是接触敏感数据。该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

build-product-integrations 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

build-product-integrations 用的人多吗?

公开数据是累计安装 1,294 次,近一周安装 55 次,下载 453 次,GitHub 4 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

build-product-integrations 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。

先想清楚的地方

  • 需要接触敏感数据,装到常用环境前要先评估影响面。
  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 环境里有大量密钥、凭证,不方便给第三方 Skill 读取的人。
  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
build-product-integrations

构建产品集成 · 本页

1,2944接触敏感数据未明确有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

App Integration

Build apps that integrate with external services via Membrane. Uses the Membrane CLI.

Use this skill when you need to add third-party integrations to a product — connecting customers to external apps, running actions on their behalf, syncing data, or providing AI agent tools.

How Membrane Works

Membrane is an integration engine. You define what your app needs (which external apps, what operations, what data) and Membrane handles authentication, API calls, and data transformation.

Core Concepts

Workspace — your project in Membrane. Contains all integrations, connectors, and configuration for your product.

Customer — a user or organization in your product. Each customer gets their own set of connections and integration state. Identified by a unique ID you choose (e.g. your user/org ID).

Connector — a pre-built adapter for an external app (e.g. Slack, HubSpot). Handles authentication (OAuth, API keys), API client setup, and connection testing. Membrane has connectors for thousands of apps — if one doesn't exist, Membrane Agent can build it.

Integration — the relationship between your product and an external app. Defined in your workspace, it groups connectors, actions, data collections, and flows for a specific app.

Connection — an authenticated link between a customer and an external app. Created when a customer goes through the OAuth/auth flow. Contains credentials managed by Membrane.

Action — a single operation on a connected app (e.g. "Send message", "Create task", "List contacts"). Has typed inputSchema and outputSchema. Can be run via API or used as AI agent tools.

Data Collection — a consistent API for working with external app data like a database table. Supports list, find, search, create, update, delete operations.

Flow — multi-step async logic triggered by events, schedules, or API calls. Composed of trigger, function, and control nodes.

Element Layers

Membrane elements exist at three layers:

  1. Universal — work across multiple external apps (no specific integration). Example: a "Create Task" action that works with any task management app.
  2. Integration — specific to one external app (has integrationId). Example: a "Create Jira Issue" action.
  3. Connection — specific to a customer's connection (has connectionId). Can customize integration-level elements per customer.

Integration-level elements need implementations per integration. Connection-level elements can override/customize integration-level defaults.

Authentication Architecture

Membrane uses JWT tokens for API access:

  • Workspace Key — identifies your workspace (set as JWT iss claim)
  • Workspace Secret — used to sign tokens (never expose to frontend)
  • Customer Token — JWT containing workspace key + customer ID, generated on your backend, used by frontend SDKs and API calls

External app credentials (OAuth tokens, API keys) are managed by Membrane — your app never handles them directly.

Standard Workflow: Search → Use → Delegate

Always follow this pattern when working with Membrane:

  1. Search for what you need — use intent search to find actions, connectors, or integrations
  2. Use it if it exists — run the action, connect with the connector, etc.
  3. Delegate to Membrane Agent if it doesn't exist — create an agent session to build the missing element

Example: User wants to list Slack channels

# Step 1: Search for a "list channels" action
npx @membranehq/cli action list --connectionId <slack-conn-id> --intent "list channels" --limit 5 --json

# Step 2: If found, run it
npx @membranehq/cli action run <actionId> --connectionId <slack-conn-id> --json

# Step 3: If NOT found, delegate to Membrane Agent
npx @membranehq/cli agent-session create --agent action-building --message "Create an action to list Slack channels for connection <slack-conn-id>" --json
npx @membranehq/cli agent-session get <sessionId> --wait --json
# Then search again and run the newly created action

Example: User wants to add a new app (e.g. Notion)

# Step 1: Search for existing Notion connector
npx @membranehq/cli search notion --elementType connector --json

# Step 2: If found, create a connection
npx @membranehq/cli connect --connectorId <connectorId>

# Step 3: If NOT found, delegate to Membrane Agent
npx @membranehq/cli agent-session create --agent connection-building --message "Build a connector for Notion (https://notion.so)" --json
npx @membranehq/cli agent-session get <sessionId> --wait --json
# Then create a connection with the new connector

This pattern applies to everything in Membrane. When an element (action, connector, data collection, flow) doesn't exist, always delegate to Membrane Agent to build it — never try to build it yourself.

Authentication

For Development / CLI

Authenticate with the Membrane CLI for development and testing:

npx @membranehq/cli login --tenant

Alternatively, install globally (npm i -g @membranehq/cli@latest) and use membrane login --tenant.

Always use --tenant to get a tenant-scoped token — this authenticates on behalf of a specific tenant (workspace + customer), so you don't need to pass --workspaceKey and --tenantKey on every subsequent command.

This opens a browser or prints an authorization URL. The user authenticates in Membrane, then selects a workspace and tenant.

Credentials are stored in ~/.membrane/credentials.json.

Non-interactive Authentication

If browser login is not available (remote/headless), the CLI prints an authorization URL. Ask the user to open it and enter the code they see:

npx @membranehq/cli login complete <code>

For Production (Backend Token Generation)

In production, your backend generates JWT tokens per customer per request:

import jwt from 'jsonwebtoken'

function generateMembraneToken(customerId, customerName) {
  return jwt.sign(
    {
      id: customerId,
      name: customerName,
    },
    process.env.MEMBRANE_WORKSPACE_SECRET,
    {
      issuer: process.env.MEMBRANE_WORKSPACE_KEY,
      expiresIn: '2h',
    },
  )
}

Create an endpoint on your backend (e.g. /api/membrane-token) that returns this token. The frontend SDK calls this to authenticate.

Adding Integrations to Your App

There are three main ways to use Membrane in your app:

Option A: JavaScript SDK (Frontend)

npm install @membranehq/sdk
import { MembraneClient } from '@membranehq/sdk'

const membrane = new MembraneClient({
  fetchToken: async () => {
    const response = await fetch('/api/membrane-token')
    const { token } = await response.json()
    return token
  },
})

// List available integrations
const integrations = await membrane.integrations.find()

// Open connection UI (popup)
await membrane.ui.connect({ integrationKey: 'hubspot' })

// List customer's connections
const connections = await membrane.connections.find()

// Reconnect a disconnected connection
await membrane.ui.connect({ connectionId: 'conn_123' })

// Delete a connection
await membrane.connection('conn_123').archive()

Option B: React SDK (Frontend)

npm install @membranehq/react
import { MembraneProvider, useIntegrations, useConnections, useMembrane } from '@membranehq/react'

function App() {
  return (
    <MembraneProvider
      fetchToken={async () => {
        const res = await fetch('/api/membrane-token')
        const { token } = await res.json()
        return token
      }}
    >
      <IntegrationsPage />
    </MembraneProvider>
  )
}

function IntegrationsPage() {
  const { items: integrations, loading } = useIntegrations()
  const { items: connections } = useConnections()
  const membrane = useMembrane()

  const handleConnect = (integrationKey) => {
    membrane.ui.connect({ integrationKey })
  }

  // ... render integrations list with connect buttons
}

Option C: REST API (Backend)

Use the Membrane API directly from your backend.

Base URL: https://api.getmembrane.com (or https://api.integration.app) Auth: Authorization: Bearer <token>

# List integrations
GET /integrations

# List connections
GET /connections

# Create connection request (returns auth URL)
POST /connection-requests
{"connectorId": "<id>"}

# Check connection request status
GET /connection-requests/<id>

# List actions for a connection
GET /actions?connectionId=<id>

# Run an action
POST /actions/<id>/run?connectionId=<cid>
{"input": {"channel": "#general", "text": "Hello!"}}

Connection UI

Membrane provides a pre-built connection UI that handles OAuth flows and credential collection:

https://ui.integration.app/embed/integrations/{INTEGRATION_KEY}/connect?token={TOKEN}

Use in an iframe or redirect. Optional query params:

  • redirectUri — redirect back after connection
  • allowMultipleConnections=1 — allow multiple connections per integration
  • name — pre-set connection name

For reconnecting disconnected connections:

https://ui.integration.app/embed/connections/{CONNECTION_ID}/refresh?token={TOKEN}

Running Actions

Actions are operations on connected apps. Find and run them:

Using CLI

# Search actions by intent
npx @membranehq/cli action list --connectionId abc123 --intent "send a message" --limit 10 --json

# Run an action
npx @membranehq/cli action run <actionId> --connectionId abc123 --input '{"channel": "#general", "text": "Hello!"}' --json

Using API

# Search actions
GET /actions?connectionId=abc123&intent=send+a+message&limit=10

# Run action
POST /actions/<actionId>/run?connectionId=abc123
{"input": {"channel": "#general", "text": "Hello!"}}

Each action has:

  • id — unique identifier
  • name, description — what it does
  • inputSchema — JSON Schema of accepted parameters
  • outputSchema — JSON Schema of the return value

The result is in the output field of the response.

AI Agent Tools

Membrane actions can serve as tools for AI agents. Two approaches:

Static Tools (known toolset)

Use when you know which tools the agent needs before the session starts.

  1. List connections: GET /connections
  2. Get actions for connection: GET /actions?connectionId=<id>
  3. Map to your agent's tool format: const tools = actions.items.map((action) => ({id: action.id, name: action.name, description: action.description, inputSchema: action.inputSchema,}))
  4. When agent calls a tool: POST /actions/<id>/run?connectionId=<cid> with {"input": {...}}

Dynamic Tools (runtime discovery)

Use when tools depend on user intent or conversation context.

Search actions by natural language intent:

GET /actions?connectionId=abc123&intent=send+a+message

This uses semantic search to find the best matching actions.

MCP Server

Membrane provides an official MCP server for AI agents:

https://mcp.integration.app/sse?token={TOKEN}&productKey={PRODUCT_KEY}

Works with Claude, Cursor, and any MCP-compatible agent.

Working with Membrane Agent

Membrane Agent is an AI agent that builds integration elements (connectors, actions, data collections, flows) for you. You should not build these yourself — delegate to Membrane Agent instead.

When to Use Membrane Agent

  • A connector for an external app doesn't exist yet
  • An action you need isn't available
  • You need a data collection, flow, or other integration element
  • You need to customize how an integration works

Creating Agent Sessions

Use the CLI to create agent sessions:

npx @membranehq/cli agent-session create --agent <agentName> --message "<description>" --json

Available agent types:

  • connection-building — builds connectors for external apps
  • action-building — builds actions for connected apps
  • membrane — general-purpose Membrane agent (can build any element)

Agent Session Workflow

  1. Create a session with a clear description of what you need: npx @membranehq/cli agent-session create --agent action-building --message "Create an action to send a message in a Slack channel for connection abc123" --json
  2. Poll until complete — the agent works asynchronously: npx @membranehq/cli agent-session get <sessionId> --wait --json

- state: "busy" — still working, poll again - state: "idle" — done with current request - status: "completed" — session finished - summary — description of what was done (available when idle)

  1. Send follow-ups if needed: npx @membranehq/cli agent-session send <sessionId> --message "Also add support for thread replies" --json
  2. Abort if something goes wrong: npx @membranehq/cli agent-session abort <sessionId> --json

What to Delegate vs. What to Build

Delegate to Membrane Agent:

  • Connectors (authentication, API clients, connection testing)
  • Actions (API mappings, input/output schemas, implementation logic)
  • Data Collections (field schemas, CRUD operations, pagination)
  • Flows (multi-step automation, event handling, scheduling)
  • External Events (webhook subscriptions, event parsing)
  • Field Mappings (data transformation between apps)

Build yourself:

  • Backend token generation endpoint
  • Frontend UI for displaying integrations and connections
  • Agent tool mapping (converting Membrane actions to your LLM's tool format)
  • Business logic that uses Membrane actions/data
  • Custom UI components for your integration experience

Tips for Effective Agent Prompts

When creating agent sessions, provide:

  • Connection ID — always include when building actions (for connection abc123)
  • App name and URL — when building connectors (Build a connector for Notion (https://notion.so))
  • Specific operation — describe exactly what the action should do (Create an action to list all tasks assigned to the current user)
  • Input/output expectations — what parameters should be accepted and what should be returned

Building a Connector via Agent

If no connector exists for an app:

# Search first
npx @membranehq/cli search notion --elementType connector --json

# If not found, build one
npx @membranehq/cli agent-session create --agent connection-building --message "Build a connector for Notion (https://notion.so)" --json

# Poll for completion
npx @membranehq/cli agent-session get <sessionId> --wait --json

# After built, create a connection
npx @membranehq/cli connect --connectorId <connectorId>

Building an Action via Agent

If the action you need doesn't exist:

# Search first
npx @membranehq/cli action list --connectionId abc123 --intent "create a task" --limit 10 --json

# If not found, build one
npx @membranehq/cli agent-session create --agent action-building --message "Create an action to create a task with title, description, and assignee for connection abc123" --json

# Poll for completion
npx @membranehq/cli agent-session get <sessionId> --wait --json

# Search again to get the action ID
npx @membranehq/cli action list --connectionId abc123 --intent "create a task" --limit 10 --json

Troubleshooting

Connection Issues

Check connection status:

npx @membranehq/cli connection get <connectionId> --json

Key fields:

  • disconnected: true — credentials expired or revoked, needs reconnection
  • error — details about what went wrong

Reconnect:

npx @membranehq/cli connect --connectionId <connectionId>

Action Errors

When an action fails, the response includes error details. Common issues:

  • Missing required input — check inputSchema for required fields
  • Connection disconnected — reconnect before retrying
  • Rate limiting — the external app is rate-limiting requests, retry after delay
  • Permission denied — the connection's OAuth scope doesn't cover this operation

Finding the Right Action

If action list --intent doesn't return what you need:

  1. Try different phrasings of your intent
  2. List all actions without intent filter: npx @membranehq/cli action list --connectionId abc123 --json
  3. If the action doesn't exist, build it via Membrane Agent (see above)

CLI Reference

All commands support --json for structured JSON output. Add --workspaceKey <key> and --tenantKey <key> to override defaults.

connection

npx @membranehq/cli connection list [--json]                # List all connections
npx @membranehq/cli connection get <id> [--json]           # Get a connection by ID

connect (interactive)

npx @membranehq/cli connect --connectorId <id>                    # Create connection via browser OAuth
npx @membranehq/cli connect --connectionId <id>                   # Reconnect existing connection

Opens a browser for authentication. Use --non-interactive to print the URL instead.

connection-request

npx @membranehq/cli connection-request create [options] [--json]   # Create connection request
npx @membranehq/cli connection-request get <requestId> [--json]    # Check request status

Options: --connectorId <id>, --integrationId <id>, --integrationKey <key>, --connectionId <id> (reconnect), --name <name>

Status values: pending, success, error, cancelled

action

npx @membranehq/cli action list [--connectionId <id>] [--intent <text>] [--limit <n>] [--json]   # List/search actions
npx @membranehq/cli action run <actionId> --connectionId <id> [--input <json>] [--json]           # Run an action

search

npx @membranehq/cli search <query> [--elementType <type>] [--limit <n>] [--json]   # Search connectors, integrations, etc.

agent-session

npx @membranehq/cli agent-session create --agent <agentName> --message <text> [--json]   # Create session
npx @membranehq/cli agent-session list [--json]                                          # List sessions
npx @membranehq/cli agent-session get <id> [--wait] [--timeout <seconds>] [--json]       # Get status
npx @membranehq/cli agent-session send <id> --message <text> [--json]                    # Send follow-up
npx @membranehq/cli agent-session abort <id> [--json]                                    # Abort session
npx @membranehq/cli agent-session messages <id> [--json]                                 # Get messages

Fallback: Raw API

If the CLI is not available, make direct API requests.

Base URL: https://api.getmembrane.com Auth header: Authorization: Bearer $MEMBRANE_TOKEN

CLI CommandAPI Equivalent
connection list --jsonGET /connections
connection get <id> --jsonGET /connections/:id
search <q> --jsonGET /search?q=<q>
connection-request create --connectorId <id> --jsonPOST /connection-requests with {"connectorId": "<id>"}
connection-request get <id> --jsonGET /connection-requests/:id
action list --connectionId <id> --intent <text> --jsonGET /actions?connectionId=<id>&intent=<text>
action run <id> --connectionId <cid> --input <json> --jsonPOST /actions/:id/run?connectionId=<cid> with {"input": <json>}
agent-session create --message <text> --jsonPOST /agent/sessions with {"prompt": "<text>"}
agent-session get <id> --wait --jsonGET /agent/sessions/:id?wait=true
agent-session send <id> --message <text> --jsonPOST /agent/sessions/:id/message with {"input": "<text>"}
agent-session abort <id> --jsonPOST /agent/sessions/:id/interrupt

External Endpoints

All requests go to the Membrane API. No other external services are contacted directly by this skill.

EndpointData Sent
https://api.getmembrane.com/*Auth credentials, connection parameters, action inputs, agent prompts

Security & Privacy

  • All data is sent to the Membrane API over HTTPS.
  • CLI credentials are stored locally in ~/.membrane/ with restricted file permissions.
  • Connection authentication (OAuth, API keys) is handled by Membrane — credentials for external apps are stored by the Membrane service, not locally.
  • Action inputs and outputs pass through the Membrane API to the connected external app.
  • Never expose your Workspace Secret to the frontend. Token generation must happen on your backend.

By using this skill, data is sent to Membrane. Only install if you trust Membrane with access to your connected apps.

平台分布

Codex

37.89%
按下载量换算172

Claude

28.07%
按下载量换算127

Cursor

18.84%
按下载量换算85

Gemini CLI

8.5%
按下载量换算39

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

接触敏感数据

该 Skill 可能接触密钥、Token、环境变量或敏感配置,应进入高风险复核队列,默认不自动发布。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills