A high-performance safety layer that validates AI agent tool calls to prevent destructive actions and security vulnerabilities.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install vigil
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 vigil 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 Vigil?
Vigil is a drop-in safety layer designed to validate what AI agents actually do, rather than just monitoring what they say. As part of the Openclaw Skills ecosystem, it provides a robust defense mechanism against common vulnerabilities like SSRF, path traversal, and prompt injection. By integrating this skill, developers can ensure their agents operate within safe boundaries, catching malicious or accidental command execution in real-time.
This utility is specifically optimized for performance, featuring a tiny 12.3KB footprint and sub-2ms latency. It requires zero runtime dependencies, making it an ideal choice for developers building secure, autonomous agents that interact with shell environments, file systems, or external APIs. Using these Openclaw Skills ensures that your agentic workflows remain compliant with security best practices without sacrificing speed.
Vigil Use Cases
- Validating shell commands before execution in autonomous agents.
- Preventing Server-Side Request Forgery (SSRF) in agents with web access.
- Auditing agent behavior during complex multi-tool workflows.
- Protecting sensitive system files from unauthorized path traversal and access.
- Detecting and blocking data exfiltration attempts to external URLs.
- Shielding applications from prompt injection attacks that attempt to bypass system instructions.
How Vigil Works
- The AI agent generates a tool call request, including the specific tool name and associated parameters.
- Vigil intercepts the call before it reaches the execution environment or MCP server.
- The tool input is analyzed against 22 built-in security rules to identify dangerous patterns like SQL injection, destructive commands, or credential leaks.
- A result object is returned containing a decision (ALLOW, BLOCK, or ESCALATE), a detailed reason, and performance metrics.
- Based on the configured mode (warn, enforce, or log), the action is either permitted, blocked, or recorded for further audit.
Vigil Setup
To begin using this skill, install the necessary npm package:
npm install vigil-agent-safety
You can then initialize the guardrails within your application:
import { configure, checkAction } from 'vigil-agent-safety';
// Set the safety mode
configure({ mode: 'enforce' });
// Validate a tool call
const result = checkAction({
agent: 'my-agent',
tool: 'exec',
params: { command: 'ls -la' },
});
For CLI usage with Openclaw Skills, you can run checks directly:
npx vigil-agent-safety check --tool exec --params '{"command":"rm -rf /"}'
Vigil Data Schema & Taxonomy
Vigil processes tool call metadata and returns structured evaluation results. This allows developers to integrate safety checks into any logging or monitoring stack.
| Field | Type | Description |
|---|---|---|
| decision | String | The safety outcome: ALLOW, BLOCK, or ESCALATE. |
| reason | String | A descriptive explanation of why a specific rule was triggered. |
| latencyMs | Number | The processing time of the security check in milliseconds. |
| ruleId | String | The identifier for the specific security rule that was matched. |
Vigil Advanced Features
- Multi-mode operation allowing for silent logging, warnings, or strict enforcement of safety rules.
- Customizable policy templates including restrictive, moderate, and permissive levels to suit different agent environments.
- Support for identifying encoding-based attacks, such as base64-encoded malicious payloads or eval() calls.
- Native integration with Openclaw Skills workflows for securing command-line tool execution.
- Real-time escalation for sensitive data detection, including AWS keys, API tokens, and SSH identities.
SKILL.md
Loading
Related Openclaw Skills
Guitar Tab Finder
A streamlined tool for discovering high-quality guitar tabs and generating structured practice notes from song titles or URLs.
Stock Watcher
A professional stock market monitoring tool for managing personalized A-share watchlists with automated performance summaries.
Qwen Image Generation
A specialized AI skill for generating high-quality images using Alibaba Cloud's Qwen Image API with robust support for bilingual prompts.
Otaku Wiki (AniList Integration)
A high-performance AniList API wrapper for AI agents to query anime, characters, and staff data instantly.
Watch.dog Uptime Monitor
A powerful monitoring integration that enables AI agents to track website uptime, API health, and cron job status using the Watch.dog platform.
WeChat Moments 4-Grid Card Generator
Automatically transforms any topic into a complete WeChat Moments post featuring engaging copy and four professionally styled tutorial cards.
Featured*
