Agentgate Clawhub

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...

Security Scan

The skill's requirements and runtime instructions are coherent with an API-gateway proxy for personal services; it asks only for an API URL and token and contains no surprising installs or file accesses.

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.

Assessment

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.

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: https://github.com/monteslu/agentgate Docs: https://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 https://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: pending → approved → executing → completed (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": "<base64 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