governclaw-middleware（治理爪中间件）

GovernClaw Middleware

This skill provides governed wrappers for sensitive operations. It acts as a policy enforcement layer between agents and external systems.

When to Use This Skill

You MUST use governed tools from this skill instead of raw tools when:

• Calling external HTTP APIs (governedHttp instead of http)
• Running shell commands (governedShell - future)
• Reading/writing files (governedFile - future)
• Controlling a browser (governedBrowser - future)

How It Works

1. You call a governed tool (e.g., governedHttp)
2. The skill sends your request metadata to GovernClaw for policy evaluation
3. GovernClaw returns allow or block with a reason
4. If allowed: the underlying operation executes and returns results
5. If blocked: the operation is cancelled and you receive a block reason

Available Tools

governedHttp

Makes HTTP requests through the GovernClaw policy engine.

Parameters:

• method (string): HTTP method - "GET", "POST", "PUT", "DELETE"
• url (string): Target URL
• body (object, optional): Request body for POST/PUT
• headers (object, optional): Custom headers

Returns:

• On success: The HTTP response from the target
• On block: { ok: false, blocked: true, reason: "..." }

Example:

const result = await context.tools.governclawMiddleware.governedHttp({
  method: "GET",
  url: "https://api.example.com/data"
});

if (result.blocked) {
  // Handle policy block
  console.log("Blocked:", result.reason);
}

Configuration

Set the GovernClaw service URL in your environment:

export GOVERNCLAW_URL="http://127.0.0.1:8000"

Or in openclaw.json:

{
  "skills": {
    "governclaw-middleware": {
      "env": {
        "GOVERNCLAW_URL": "http://127.0.0.1:8000"
      }
    }
  }
}

Governance Context

The skill automatically forwards these context fields to GovernClaw:

• parent_id: The session ID (who owns the request)
• child_id: The agent ID (who is making the request)
• source: Where the request originated (agent, control, cron, etc.)
• channel: The channel ID (if applicable)
• node_id: The node ID (if applicable)
• skill: Always "governclaw-middleware"

Error Handling

Always check for blocked in responses:

const response = await context.tools.governclawMiddleware.governedHttp({...});

if (!response.ok && response.blocked) {
  // Policy violation - do not retry
  return { error: response.reason };
}

if (!response.ok) {
  // Network or other error - may retry
  return { error: "Request failed" };
}

// Success
return response.data;

Policy Modes

GovernClaw supports three governance modes:

• playground: Log-only, actions always allowed
• governed: Default mode, enforce policies
• strict: Block on any uncertainty

The skill defaults to governed mode. Future versions may allow per-request mode overrides.