← 返回 Skills 市场
monteslu

Agentgate Clawhub

作者 Luis Montes · GitHub ↗ · v1.0.4
cross-platform ✓ 安全检测通过
1338
总下载
2
收藏
7
当前安装
5
版本数
在 OpenClaw 中安装
/install agentgate
功能描述
API gateway for personal data with human-in-the-loop write approval. Connects agents to GitHub, Bluesky, Google Calendar, Home Assistant, and more — all thro...
使用说明 (SKILL.md)

agentgate

API gateway for AI agents to access personal data with human-in-the-loop write approval.

  • Reads (GET) execute immediately
  • Writes (POST/PUT/PATCH/DELETE) go through an approval queue
  • Bypass mode available for trusted agents (writes execute immediately)

GitHub: \x3Chttps://github.com/monteslu/agentgate> Docs: \x3Chttps://agentgate.org>

Setup

agentgate server runs on a separate machine from OpenClaw. This is by design — your agent should not have direct access to the server holding your credentials. Install and run agentgate on a different computer (or VPS/container on a different host). See \x3Chttps://agentgate.org> for setup instructions.

Once agentgate is running, configure these environment variables for your OpenClaw agent:

  • AGENT_GATE_URL — agentgate base URL (e.g., http://your-agentgate-host:3050)
  • AGENT_GATE_TOKEN — your agent's API key (create in the agentgate Admin UI → API Keys)

Authentication

All requests require the API key in the Authorization header:

Authorization: Bearer $AGENT_GATE_TOKEN

First Steps — Service Discovery

After connecting, discover what's available on your instance:

GET $AGENT_GATE_URL/api/agent_start_here
Authorization: Bearer $AGENT_GATE_TOKEN

Returns your agent's config, available services, accounts, and full API documentation.

Instance-Specific Skills

agentgate generates additional skills tailored to your instance with your specific accounts and endpoints. See the agentgate skills documentation for details on how to install and update them.

Supported Services

agentgate supports many services out of the box. Common ones include:

  • Code: GitHub, Jira
  • Social: Bluesky, Mastodon, LinkedIn
  • Search: Brave Search, Google Search
  • Personal: Google Calendar, YouTube, Fitbit
  • IoT: Home Assistant
  • Messaging: Twilio, Plivo

New services are added regularly. Check GET /api/agent_start_here for what's configured on your instance.

Reading Data

GET $AGENT_GATE_URL/api/{service}/{accountName}/{path}
Authorization: Bearer $AGENT_GATE_TOKEN

Example: GET $AGENT_GATE_URL/api/github/myaccount/repos/owner/repo

Writing Data

Writes go through the approval queue:

POST $AGENT_GATE_URL/api/queue/{service}/{accountName}/submit
Authorization: Bearer $AGENT_GATE_TOKEN
Content-Type: application/json

{
  "requests": [
    {
      "method": "POST",
      "path": "/the/api/path",
      "body": { "your": "payload" }
    }
  ],
  "comment": "Explain what you are doing and why"
}

Always include a clear comment explaining your intent. Include links to relevant resources.

Check write status

GET $AGENT_GATE_URL/api/queue/{service}/{accountName}/status/{id}
Authorization: Bearer $AGENT_GATE_TOKEN

Statuses: pendingapprovedexecutingcompleted (or rejected/failed/withdrawn)

Withdraw a pending request

DELETE $AGENT_GATE_URL/api/queue/{service}/{accountName}/status/{id}
Authorization: Bearer $AGENT_GATE_TOKEN
Content-Type: application/json

{ "reason": "No longer needed" }

Binary uploads

For binary data (images, files), set binaryBase64: true in the request body:

{
  "method": "POST",
  "path": "com.atproto.repo.uploadBlob",
  "binaryBase64": true,
  "headers": { "Content-Type": "image/jpeg" },
  "body": "\x3Cbase64 encoded data>"
}

Inter-Agent Messaging

Agents can message each other through agentgate for multi-agent coordination.

Send a message

POST $AGENT_GATE_URL/api/agents/message
Authorization: Bearer $AGENT_GATE_TOKEN
Content-Type: application/json

{ "to_agent": "agent_name", "message": "Hello!" }

Read messages

GET $AGENT_GATE_URL/api/agents/messages?unread=true
Authorization: Bearer $AGENT_GATE_TOKEN

Mark as read

POST $AGENT_GATE_URL/api/agents/messages/{id}/read
Authorization: Bearer $AGENT_GATE_TOKEN

Broadcast to all agents

POST $AGENT_GATE_URL/api/agents/broadcast
Authorization: Bearer $AGENT_GATE_TOKEN
Content-Type: application/json

{ "message": "Team announcement" }

Discover agents

GET $AGENT_GATE_URL/api/agents/messageable
Authorization: Bearer $AGENT_GATE_TOKEN

Messaging modes (configured by admin): off, supervised (requires approval), open (immediate delivery).

Mementos (Persistent Memory)

Store and retrieve notes across sessions using keyword tags.

Store a memento

POST $AGENT_GATE_URL/api/agents/memento
Authorization: Bearer $AGENT_GATE_TOKEN
Content-Type: application/json

{ "content": "Important info to remember", "keywords": ["project", "notes"] }

Search by keyword

GET $AGENT_GATE_URL/api/agents/memento/search?keywords=project&limit=10
Authorization: Bearer $AGENT_GATE_TOKEN

Fetch full content by IDs

GET $AGENT_GATE_URL/api/agents/memento/42,38
Authorization: Bearer $AGENT_GATE_TOKEN

List your keywords

GET $AGENT_GATE_URL/api/agents/memento/keywords
Authorization: Bearer $AGENT_GATE_TOKEN

Important Notes

  • Always include clear comments on write requests
  • Be patient with writes — approval requires human action
  • Use GET /api/agent_start_here to discover available services
  • See agentgate docs for instance-specific skill setup
安全使用建议
This skill is internally consistent, but it gives any agent that has AGENT_GATE_TOKEN the ability to read your connected services and submit writes for approval. Before installing: (1) ensure AGENT_GATE_URL points to a legitimate, HTTPS-protected agentgate instance you control; (2) run the agentgate server on a separate host as recommended; (3) create a scoped API key for the agent with minimal permissions and keep it secret; (4) verify the human-in-the-loop approval workflow is enabled and that 'bypass/trusted agent' modes are not enabled for this key unless you fully trust the agent; (5) review agentgate.org / your instance docs and admin UI to confirm how requests are logged/audited and how to rotate/revoke keys if needed.
功能分析
Type: OpenClaw Skill Name: agentgate Version: 1.0.4 The skill describes how an AI agent can interact with the `agentgate` API gateway, which is designed to provide controlled access to personal data with human-in-the-loop write approval. The `SKILL.md` documentation explicitly states that the `agentgate` server runs on a separate machine from OpenClaw and emphasizes safety controls for write operations. The instructions only detail making HTTP requests to the specified `AGENT_GATE_URL` using an API token, without any evidence of malicious execution, data exfiltration to unauthorized endpoints, or prompt injection designed to subvert the agent's core directives.
能力评估
Purpose & Capability
Name/description (API gateway with human approval) matches the declared env vars (AGENT_GATE_URL, AGENT_GATE_TOKEN) and the SKILL.md examples (HTTP GET/POST to the gateway). Required credentials are exactly what an agent would need to call a gateway service.
Instruction Scope
SKILL.md instructs the agent to make HTTP calls to the configured AGENT_GATE_URL and to include the AGENT_GATE_TOKEN in Authorization headers. It does not instruct the agent to read unrelated files, other env vars, or system paths, nor to send data to unexpected endpoints outside the declared gateway.
Install Mechanism
This is instruction-only with no install spec and no code on disk; there is no download or package install behavior to review, which minimizes install-time risk.
Credentials
The skill requires two env vars (base URL and API token) which are proportionate and expected for a remote API gateway. No extraneous secrets or unrelated credentials are requested.
Persistence & Privilege
The skill is not marked always:true and has no install footprint. disable-model-invocation is false (the platform default) — this is normal and not a standalone concern. The SKILL.md warns the server should run on a separate host, which is a safer deployment pattern.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install agentgate
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /agentgate 触发
  4. 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
v1.0.4
Remove fetch-and-write skill install instructions; point to docs instead
v1.0.3
Consistent auth headers on all examples; proper formatting for messaging and mementos sections; fix binary upload example
v1.0.2
Remove npm install step; clarify agentgate server runs on separate machine; replace curl|node with safe GET approach; add skill update instructions
v1.0.1
Replace curl|node with safe GET approach; add skill update instructions
v1.0.0
Initial release of agentgate — an API gateway for personal data with human-in-the-loop safety. - Provides secure, unified API access to services including GitHub, Bluesky, Google Calendar, Home Assistant, and more. - Immediate access for read requests; write requests require human approval (with optional bypass for trusted agents). - Supports dynamic per-instance skill generation based on configured accounts and services. - Built-in inter-agent messaging and persistent memory (mementos) features. - Simple authentication via bearer token; easy install/setup instructions included.
元数据
Slug agentgate
版本 1.0.4
许可证
累计安装 8
当前安装数 7
历史版本数 5
常见问题

Agentgate Clawhub 是什么?

API gateway for personal data with human-in-the-loop write approval. Connects agents to GitHub, Bluesky, Google Calendar, Home Assistant, and more — all thro... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 1338 次。

如何安装 Agentgate Clawhub?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install agentgate」即可一键安装,无需额外配置。

Agentgate Clawhub 是免费的吗?

是的,Agentgate Clawhub 完全免费(开源免费),可自由下载、安装和使用。

Agentgate Clawhub 支持哪些平台?

Agentgate Clawhub 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Agentgate Clawhub?

由 Luis Montes(@monteslu)开发并维护,当前版本 v1.0.4。

💬 留言讨论