Token导航 LogoToken导航

Agent Skill · 研究检索

nexus(nexus 搜索)

nexus 用于查找、检索和筛选相关信息,适合在 OpenClaw 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。

nexus 是研究检索类 Agent Skill,可用宿主为 OpenClaw,通过ClawHub获取,权限为接触敏感数据;许可证为 MIT-0,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 20,405。

总安装 5,221周安装 222GitHub Stars 公开资料未说明下载 1,829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
MIT-0
版本
1.0.5

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:nexus(nexus 搜索)
来源仓库:https://github.com/karimsherifyehia/nexus
安装命令:
openclaw skills install nexus
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 OpenClaw 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。

ClawHubOpenClaw
openclaw skills install nexus

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。

与 Nexus 电商操作系统交互,管理多租户零售业务运营数据。适用于订单处理、库存同步与客户关系管理集成场景。需设置 NEXUS_API_KEY 环境变量并验证代理前缀 nxs_ 有效性。注意 API 速率限制、租户隔离策略与敏感操作日志记录要求。nexus 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

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 OpenClaw;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 openclaw skills install nexus;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

nexus 怎么安装?

在终端执行 openclaw skills install nexus。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

nexus 支持哪些 AI 工具?

公开资料标注支持 OpenClaw。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。

nexus 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。

nexus 是免费的吗?

来源标注的许可证是 MIT-0。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

nexus 用的人多吗?

公开数据是累计安装 5,221 次,近一周安装 222 次,下载 1,829 次。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。

nexus 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

  • 专注 OpenClaw 一种宿主,触发方式和使用习惯不用来回切换。
  • 许可证明确为 MIT-0,商用和分发边界好判断。

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。
  • 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 OpenClaw,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
nexus

nexus 搜索 · 本页

5,221—接触敏感数据MIT-0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

\---

name: nexus-agent description: Use Nexus as an AI agent through the hosted MCP server or direct Nexus API. Supports authenticated read operations by default and explicit-confirmation write operations for CRM, orders, inventory, messaging, shipping, internal chat, social publishing, invites, and conversation intelligence. homepage: https://nexus-docs.aiforstartups.io user-invocable: true disable-model-invocation: true metadata: {"openclaw":{"requires":{"env":\["NEXUS\_API\_KEY"]},"primaryEnv":"NEXUS\_API\_KEY","homepage":"https://nexus-docs.aiforstartups.io"}}


Nexus Agent Skill

Nexus is a multi-tenant operations OS for ecommerce and retail workflows. Use this skill to authenticate an AI agent, access the hosted Nexus MCP server, and fall back to direct HTTP API access when MCP is not sufficient.

This skill is intentionally user-invocable only and is designed to be read-first. Any write operation requires explicit user confirmation.

Required credential

This skill requires the environment variable:

  • NEXUS\_API\_KEY

NEXUS\_API\_KEY must be an agent API key, not a human JWT and not a short-lived agent JWT.

Credential handling rules

  • Read only NEXUS\_API\_KEY from the runtime environment.
  • If NEXUS\_API\_KEY is missing, ask the user or operator to configure it.
  • Do not read unrelated environment variables, local files, or stored secrets.
  • Do not hardcode live secrets into the skill.
  • Use the raw API key only with agent-auth.
  • After exchanging the API key for a short-lived JWT, use that JWT for MCP and direct API calls.

Optional runtime config for direct PostgREST access

When calling PostgREST directly, the runtime must also provide an anon key header such as:

  • VITE\_SUPABASE\_ANON\_KEY
  • NEXT\_PUBLIC\_SUPABASE\_ANON\_KEY
  • or an operator-provided equivalent runtime secret

Do not publish a live anon key inside distributed skills or public documentation.

Example OpenClaw configuration

{
  "skills": {
    "entries": {
      "nexus-agent": {
        "apiKey": {
          "source": "env",
          "provider": "default",
          "id": "NEXUS\_API\_KEY"
        }
      }
    }
  }
}

Safety and execution policy

Trusted domains only

Only call Nexus on these hosts:

  • api.nexus.aiforstartups.io
  • nexus-docs.aiforstartups.io

Do not send Nexus credentials or payloads to any other host.

Read-first behavior

Default to read operations whenever possible.

Examples of read operations:

  • list contacts, orders, inventory, conversations, call logs, invites, or internal conversations
  • get a single contact, order, shipment, post, or annotation
  • read MCP schema resources and organization info
  • run search and analytics reads

Write gate

Require explicit user confirmation before any write action, including:

  • creating or updating contacts
  • creating orders or updating order status
  • sending customer messages
  • sending internal team messages
  • creating, updating, publishing, or deleting social posts
  • inviting, resending, or cancelling organization invites
  • updating message annotations, AI tag definitions, prompts, or org AI context
  • initiating a phone call
  • creating an AWB or triggering fulfillment steps

If the current API key has only read scope, do not attempt write actions.

No autonomous onboarding or background writes

  • Do not autonomously register organizations, create accounts, or start onboarding flows.
  • Do not autonomously create orders, publish posts, send messages, or change records in the background.
  • If the user explicitly asks to onboard or enable a Nexus integration, ask only for the minimum information required for that step.

Authentication

AI agents do not use human user JWTs directly.

The correct flow is:

  1. A human org admin creates an agent API key with agent-api-key-create
  2. The agent exchanges that key for a short-lived JWT with agent-auth
  3. The agent uses the short-lived JWT for MCP or direct API calls

Step 1: human admin creates an API key

This is a human-admin operation.

POST https://api.nexus.aiforstartups.io/functions/v1/agent-api-key-create
Authorization: Bearer <human-admin-jwt>
Content-Type: application/json

{
  "name": "Bahig - Karim assistant",
  "scopes": \["read", "write"],
  "expires\_at": "2026-12-31T23:59:59Z"
}

Response:

{
  "api\_key": "nxs\_ak\_abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ12",
  "key\_prefix": "nxs\_ak\_ab12CD34",
  "organization\_id": "uuid",
  "name": "Bahig - Karim assistant",
  "scopes": \["read", "write"],
  "expires\_at": "2026-12-31T23:59:59.000Z",
  "message": "Store this API key now. It is only returned once."
}

Important:

  • the raw key is returned once only
  • Nexus stores only a bcrypt hash
  • keys can be revoked

Step 2: exchange API key for an agent JWT

POST https://api.nexus.aiforstartups.io/functions/v1/agent-auth
Content-Type: application/json

{"api\_key": "nxs\_ak\_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"}

Response:

{
  "access\_token": "eyJ...",
  "organization\_id": "uuid",
  "expires\_in": 3600
}

Use the access\_token as Authorization: Bearer <token> on all subsequent requests.

API key format

nxs\_ak\_ prefix + 48 random alphanumeric characters.

Scopes

  • read = read-only operations
  • write = read + create + update
  • admin = full access

MCP authentication

The hosted MCP endpoint accepts the agent JWT, not the raw API key:

POST https://api.nexus.aiforstartups.io/functions/v1/mcp-server
Authorization: Bearer <agent-jwt>

Base URLs

|Service|URL| |-|-| |Edge Functions|https://api.nexus.aiforstartups.io/functions/v1/<function-name>| |PostgREST (DB)|https://api.nexus.aiforstartups.io/rest/v1/<table>| |MCP|https://api.nexus.aiforstartups.io/functions/v1/mcp-server|

For direct PostgREST calls, include both:

  • Authorization: Bearer <agent-jwt>
  • apikey: <anon-key-from-secure-runtime-config>

Preferred integration path: MCP first

Prefer the hosted MCP server when possible. It exposes a stable, documented tool surface for common agent workflows.

MCP tools

  • nexus\_list\_contacts
  • nexus\_get\_contact
  • nexus\_create\_contact
  • nexus\_update\_contact
  • nexus\_list\_orders
  • nexus\_get\_order
  • nexus\_create\_order
  • nexus\_update\_order\_status
  • nexus\_list\_inventory
  • nexus\_check\_stock
  • nexus\_list\_conversations
  • nexus\_send\_message
  • nexus\_search
  • nexus\_sync\_social\_posts
  • nexus\_list\_social\_posts
  • nexus\_get\_social\_post
  • nexus\_create\_social\_post
  • nexus\_update\_social\_post
  • nexus\_publish\_social\_post
  • nexus\_delete\_social\_post
  • nexus\_list\_social\_accounts
  • nexus\_get\_social\_analytics
  • nexus\_get\_content\_calendar
  • nexus\_list\_organization\_invites
  • nexus\_invite\_organization\_member
  • nexus\_cancel\_organization\_invite
  • nexus\_resend\_organization\_invite
  • nexus\_list\_internal\_conversations
  • nexus\_get\_internal\_messages
  • nexus\_open\_internal\_dm
  • nexus\_send\_internal\_message
  • nexus\_list\_ai\_tag\_definitions
  • nexus\_list\_message\_annotations
  • nexus\_get\_message\_ai\_annotation
  • nexus\_get\_org\_ai\_annotation\_settings
  • nexus\_get\_customer\_conversation\_intelligence
  • nexus\_get\_conversation\_intelligence\_dashboard
  • nexus\_update\_message\_annotation
  • nexus\_update\_ai\_tag\_definition
  • nexus\_set\_org\_ai\_annotation\_prompt
  • nexus\_update\_org\_conversation\_intelligence\_context

MCP resources

  • nexus://organization/info
  • nexus://schema/contacts
  • nexus://schema/orders
  • nexus://schema/inventory
  • nexus://schema/social
  • nexus://schema/conversation-intelligence

When to use MCP vs direct API

Use MCP when:

  • the client supports MCP
  • you want a compact tool surface
  • you want resource discovery and schema resources
  • you need social or content-calendar operations through a stable agent interface

Use direct API when:

  • the client does not support MCP
  • you need raw PostgREST querying
  • you are debugging a low-level integration
  • you need a workflow not yet covered by the MCP wrapper

Social MCP notes

  • Social tools require at least the starter plan.
  • Social write tools require write or admin scope.
  • Facebook and Instagram drafts and publishing are supported from MCP.
  • Instagram sync reads from the connected Instagram Business account and writes normalized rows into social\_posts.
  • twitter and linkedin remain reserved in request enums for forward compatibility and are not yet writable from this MCP surface.

Organization invites (MCP + agent-invite)

Agents can manage organization\_invites for their JWT organization only.

  • MCP tools: nexus\_list\_organization\_invites (read scope), nexus\_invite\_organization\_member, nexus\_cancel\_organization\_invite, nexus\_resend\_organization\_invite (write scope)
  • HTTP API: POST /functions/v1/agent-invite with Authorization: Bearer <agent-jwt> and JSON body { "action": "invite"|"cancel"|"resend"|"list", ... }

Requires BREVO\_KEY on Edge Functions.

Internal team chat (MCP)

Internal chat is separate from the customer omnichannel inbox (chats / messages). MCP tools talk to internal\_conversations and internal\_messages.

  • the agent API key must be linked to an agent\_accounts row (agent\_api\_keys.agent\_id)
  • nexus\_open\_internal\_dm: pass user\_id or user\_email of an active org member
  • nexus\_send\_internal\_message: sender\_id is null server-side; humans still send with their user id in the app
  • nexus\_list\_internal\_conversations and nexus\_get\_internal\_messages: only threads the agent created or has posted in

Core operations

Contacts (CRM)

journey\_stage\_id is a UUID foreign key to customer\_journey\_stages, not a free-text field.
# List contacts
GET /rest/v1/customers?organization\_id=eq.<org\_id>\&order=created\_at.desc\&limit=25
Authorization: Bearer <agent-jwt>
apikey: <anon-key>

# Search by phone (URL-encode + as %2B)
GET /rest/v1/customers?phone\_number=eq.%2B201234567890\&organization\_id=eq.<org\_id>

# Create contact — requires explicit confirmation and write scope
POST /rest/v1/customers
{"first\_name":"Ahmed","last\_name":"Hassan","phone\_number":"+201234567890","organization\_id":"<org\_id>"}

# Update tags — requires explicit confirmation and write scope
PATCH /rest/v1/customers?id=eq.<uuid>
{"tags":\["vip","repeat-buyer"]}

Messages (omnichannel inbox)

chats has no status column. Use is\_archived. messages sorts by timestamp, not created\_at.
# List open conversations
GET /rest/v1/chats?organization\_id=eq.<org\_id>\&is\_archived=eq.false\&order=last\_message\_time.desc

# Get messages in a conversation
GET /rest/v1/messages?chat\_id=eq.<chat\_id>\&order=timestamp.asc

# Send WhatsApp message — requires explicit confirmation and write scope
POST /functions/v1/whatsapp-send-message
{"chatId":"<chat\_id>","message":"Hello from Nexus AI agent","organizationId":"<org\_id>"}

# Send WhatsApp template — requires explicit confirmation and write scope
POST /functions/v1/whatsapp-send-template
{"phone\_number":"+201234567890","template\_name":"order\_shipped","language":"ar","components":\[]}

Orders

The status column is order\_status, not status.
# List orders
GET /rest/v1/orders?organization\_id=eq.<org\_id>\&order=created\_at.desc\&limit=50

# Filter by status — use order\_status
GET /rest/v1/orders?organization\_id=eq.<org\_id>\&order\_status=eq.pending

# Create order — requires explicit confirmation and write scope
POST /rest/v1/orders
{"customer\_id":"<uuid>","organization\_id":"<org\_id>","order\_status":"pending","total\_amount":450,"currency":"EGP"}

# Update order status — requires explicit confirmation and write scope
PATCH /rest/v1/orders?id=eq.<uuid>
{"order\_status":"confirmed"}

# Get order items (qty column, not quantity)
GET /rest/v1/order\_items?order\_id=eq.<order\_id>

Inventory

Inventory uses items plus stock\_balances. There is no inventory\_items table.
# List products
GET /rest/v1/items?organization\_id=eq.<org\_id>\&is\_active=eq.true\&order=updated\_at.desc

# Search by SKU
GET /rest/v1/items?organization\_id=eq.<org\_id>\&sku=eq.TSHIRT-M-BLK

# Low stock
GET /rest/v1/stock\_balances?organization\_id=eq.<org\_id>\&available\_quantity=lt.10

# Stock for one item across warehouses
GET /rest/v1/stock\_balances?item\_id=eq.<uuid>

Shipping

# Create AWB — requires explicit confirmation and write scope
POST /functions/v1/create-awb-bosta
{"order\_id":"<uuid>","pickup\_date":"2025-06-17"}

# List shipments
GET /rest/v1/awbs?organization\_id=eq.<org\_id>\&order=created\_at.desc

# Get tracking history
GET /rest/v1/awb\_status\_logs?awb\_id=eq.<awb\_id>\&order=timestamp.desc

VoIP

# Initiate call — requires explicit confirmation and write scope
POST /functions/v1/call-initiate
{"phone\_number":"+201234567890"}

# Get call logs for a contact
GET /rest/v1/call\_logs?customer\_id=eq.<uuid>\&order=call\_date.desc

Conversation intelligence and CS AI

Prefer the MCP tools for conversation intelligence. They map to the current schema and RPCs more safely than ad hoc direct queries.

Read-oriented MCP tools:

  • nexus\_list\_ai\_tag\_definitions
  • nexus\_list\_message\_annotations
  • nexus\_get\_message\_ai\_annotation
  • nexus\_get\_org\_ai\_annotation\_settings
  • nexus\_get\_customer\_conversation\_intelligence
  • nexus\_get\_conversation\_intelligence\_dashboard

Write-capable MCP tools that require explicit confirmation and write scope:

  • nexus\_update\_message\_annotation
  • nexus\_update\_ai\_tag\_definition
  • nexus\_set\_org\_ai\_annotation\_prompt
  • nexus\_update\_org\_conversation\_intelligence\_context

Pagination

Use either the Range header or offset and limit query parameters.

GET /rest/v1/customers
Range: 0-24

Response includes Content-Range: 0-24/1523.

Filtering

Common PostgREST operators: eq, neq, gt, gte, lt, like, in, is

# Multiple filters — note order\_status, not status
GET /rest/v1/orders?organization\_id=eq.<org\_id>\&order\_status=eq.shipped\&created\_at=gte.2025-01-01

# Null check
GET /rest/v1/orders?shopify\_order\_id=is.null

Error handling

|Code|Meaning|Action| |-|-|-| |401|Invalid or expired token|Re-authenticate via agent-auth| |403|Scope not allowed|Check API key scopes| |404|Record not found|Verify the UUID and org scope| |429|Rate limited|Retry after Retry-After seconds|

Deployed endpoints for agents

Agent bootstrap

  • POST /functions/v1/agent-api-key-create
  • POST /functions/v1/agent-auth
  • POST /functions/v1/mcp-server

Outbound messaging used by MCP

  • POST /functions/v1/whatsapp-send-message
  • POST /functions/v1/facebook-send-message
  • POST /functions/v1/instagram-send-message

Notes

  • agent-auth is rate limited to 10 attempts per minute per IP
  • agent JWTs expire after 3600 seconds
  • use the raw API key only with agent-auth

Support

If you encounter an error or undocumented behavior, contact:

  • To: info@aiforstartups.io
  • CC: karim.sherif@aiforstartups.io

Include the endpoint, the error message, and a short description of the attempted action.

Additional resources

  • Full API reference: api-reference.md
  • Hosted MCP docs: https://nexus-docs.aiforstartups.io/api/ai-agents-mcp
  • OpenAPI spec: https://nexus-docs.aiforstartups.io/openapi.yaml
  • Docs: https://nexus-docs.aiforstartups.io

平台分布

OpenClaw

83.34%
按下载量换算1,524

安全审计

VirusTotal

通过

ClawScan

可疑

Static analysis

通过

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