Yabbie Net for Openclaw

A high-performance safety proxy that intercepts and validates AI tool calls in real-time to prevent malicious or accidental actions.

devlines
v0.2.0
Mar 30, 2026
0
780
0

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.

npx clawhub@latest install yabbie-net

2. Manual Installation

Copy the skill folder to one of these locations

Global
~/.openclaw/skills/
Workspace
<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.

Help me install yabbie-net using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).

Prefer to download?

Get the raw skill files in a ZIP archive.

What is Yabbie Net?

Yabbie Net acts as a robust security layer for developers deploying Openclaw Skills. It functions as a Model Context Protocol (MCP) proxy that sits between an AI agent and its toolset, catching unsafe actions before they can be executed. By providing a transparent interception mechanism, it ensures that agentic workflows remain within defined safety boundaries without requiring expensive hardware or complex infrastructure.

This security skill is essential for protecting sensitive environments from prompt injection vulnerabilities and unauthorized data exfiltration. It offers three distinct tiers of protection, ranging from instant local pattern matching to sophisticated AI-driven intent validation, making it one of the most versatile safety tools for modern Openclaw Skills.

Yabbie Net Use Cases

  • Restricting AI agents from reading sensitive system files like .env, secrets, or SSH keys.
  • Blocking the execution of high-risk tools such as shell commands in unprivileged environments.
  • Implementing rate limits on tool calls to prevent runaway costs or system abuse.
  • Requiring human approval for irreversible actions like database deletions or production deployments.

How Yabbie Net Works

  1. The AI agent sends a tool execution request via the standard MCP stdio channel.
  2. Yabbie Net intercepts the JSON-RPC message before it reaches the intended MCP server.
  3. The proxy evaluates the tool name and arguments against Tier 1 deterministic rules, such as path deny lists.
  4. If Tier 2 is active, the skill sends a truncated summary to an AI judge to verify the intent against the stated task context.
  5. If the action is flagged as uncertain or irreversible, Tier 3 triggers a human escalation request via the terminal.
  6. Approved calls are forwarded to the original server, while blocked calls return a safe error message to the agent.

Yabbie Net Setup

To begin securing your Openclaw Skills, install the package via npm:

npm install -g [email protected]

Configure your openclaw.json to wrap an existing MCP server with the Yabbie Net proxy:

{
  "mcpServers": {
    "filesystem": {
      "command": "npx",
      "args": ["[email protected]", "--verbose", "--", "npx", "-y", "@modelcontextprotocol/server-filesystem", "/tmp"]
    }
  }
}

Yabbie Net Data Schema & Taxonomy

Yabbie Net manages security configurations and audit trails locally, ensuring privacy for your Openclaw Skills integrations.

Component Description Format
yabbie.yaml Main configuration file defining security tiers and deny lists. YAML
.yabbie/audit.jsonl A local, append-only log of every intercepted tool call and verdict. JSONL
taskContext A user-defined string describing the agent's safe operating scope. String

Yabbie Net Advanced Features

  • Triple-tier defense architecture combining local rules, AI intent judging, and human-in-the-loop verification.
  • Privacy-first AI judge support using local LLMs via Ollama or remote models via Anthropic.
  • Intelligent data masking that replaces large file bodies with byte counts to prevent data leakage during AI auditing.
  • Real-time monitoring tools including a CLI tailing command for live audit logs and performance metrics.
  • Seamless integration as a transparent proxy, requiring zero code changes to your existing Openclaw Skills.

SKILL.md


Loading

Related Openclaw Skills

METADATA

Github Stars: 0
forks: 0

Featured*