ClawGuard is an LLM-powered security guardrail that evaluates and intercepts risky tool executions to protect your agent environment.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install clawguard
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install clawguard 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 ClawGuard?
ClawGuard is a sophisticated security plugin designed specifically for the ecosystem of Openclaw Skills. It operates as an LLM-as-a-Judge, providing a critical layer of defense by analyzing tool calls before they are executed. By leveraging the intelligence of large language models, it can understand the intent behind a command and determine if the operation poses a threat to the system.
This plugin is indispensable for developers building autonomous agents with Openclaw Skills who require robust safety protocols. Whether you are running agents in production or experimenting with complex workflows, ClawGuard ensures that risky operations—such as unauthorized file access or destructive system commands—are identified and optionally blocked before any damage occurs.
ClawGuard Use Cases
- Preventing autonomous agents from executing potentially harmful or destructive shell commands.
- Implementing a zero-trust architecture for tool execution within Openclaw Skills.
- Auditing tool usage and intent for compliance and security monitoring in enterprise environments.
- Setting up a safe sandbox environment where high-risk tool calls are logged for review without being executed.
How ClawGuard Works
- The plugin intercepts the workflow of Openclaw Skills by registering a before_tool_call hook.
- When a tool call is initiated, the plugin captures the tool's JSON payload and the surrounding session context.
- This data is transmitted to an LLM provider to perform a security evaluation based on the provided context.
- The LLM judge assigns a risk level ranging from none to critical.
- Based on your specific configuration, ClawGuard will either allow the tool to run, log the risk, or terminate the call to prevent execution.
ClawGuard Setup
To get started with ClawGuard for Openclaw Skills, ensure your gateway is configured correctly:
# Enable the chat completions endpoint required for the judge
openclaw config set gateway.http.endpoints.chatCompletions.enabled true
# Install the plugin via the CLI
openclaw plugins install @capsulesecurity/clawguard
# Restart your gateway. If using Docker, use the force-recreate flag:
docker compose up -d --force-recreate openclaw-gateway
ClawGuard Data Schema & Taxonomy
ClawGuard allows for granular control over security settings for Openclaw Skills through its configuration schema:
| Option | Type | Default | Description |
|---|---|---|---|
securityCheckEnabled |
Boolean | true |
Activates the LLM-as-a-Judge evaluation logic. |
blockOnRisk |
Boolean | true |
Automatically blocks tool calls rated as high or critical risk. |
logToolCalls |
Boolean | true |
Logs the full JSON of every tool call to the gateway logs for auditing. |
maxContextWords |
Integer | 2000 |
The word limit for the session context sent to the LLM judge. |
timeoutMs |
Integer | 15000 |
The maximum time allowed for a security evaluation before timing out. |
ClawGuard Advanced Features
- Log-only mode: Disable blockOnRisk to monitor agent behavior and fine-tune security prompts without interrupting workflows.
- Model Agnostic: Use any model configured within your Openclaw Skills environment as the security judge.
- Real-time Metrics: Integrated metrics collection to monitor the frequency and severity of blocked tool calls.
- Contextual Awareness: Evaluates tool calls based on the specific conversation history to detect prompt injection or social engineering attempts.
SKILL.md
Loading
Related Openclaw Skills
Giraffe Guard
A comprehensive security auditing tool designed to scan Openclaw Skills for malicious code, supply chain vulnerabilities, and sensitive data leaks.
issuefinder-tool
A professional automotive diagnostic tool for Openclaw Skills that automates vehicle log retrieval from the cloud and provides intelligent fault analysis.
Boot KPI Analyzer
An automated diagnostic tool for analyzing the startup and shutdown performance metrics of SS4 vehicle head units.
Proxmox VE Automation
A comprehensive automation skill for managing Proxmox VE virtual machines, containers, and cluster resources through a robust REST API interface.
Context Compression
A technical framework for managing long-running agent sessions through structured summarization and tokens-per-task optimization.
Error Guard
A proactive error management and circuit-breaking system designed to prevent AI agent failures through real-time monitoring and automated recovery.
Featured*
