ClawGuard Shield is a robust multi-layered security framework designed to detect and block malicious prompt injections and intent manipulation in AI agent workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install clawguard-shield
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-shield 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 Shield?
ClawGuard Shield (CG-SD) v3 is an advanced security layer specifically built for the Openclaw Skills ecosystem. It provides proactive protection against a wide array of adversarial attacks targeting Large Language Models, including role hijacking, jailbreak attempts, and sophisticated obfuscation techniques. By acting as a firewall for AI interactions, it ensures that agents remain within their intended operational scope and safety boundaries.
This skill is indispensable for developers building production-grade AI applications where user-generated input could potentially compromise the system. With its enhanced v3 features, ClawGuard Shield goes beyond simple pattern matching to perform intent integrity validation and multi-factor risk scoring, making it a cornerstone for secure AI agent development.
ClawGuard Shield Use Cases
- Detecting hidden commands within Base64, Hex, or Unicode encoded inputs.
- Preventing role-play attacks where users attempt to impersonate administrators or override AI personas.
- Blocking known jailbreak patterns like DAN (Do Anything Now) and their obfuscated variants.
- Safeguarding system instructions from chain hijacking attempts like "ignore all previous rules."
- Monitoring for intent drift to ensure user requests align with the original task and safety constraints.
How ClawGuard Shield Works
- The input string is first scanned for encoded payloads, including Base64, Hex, and invisible zero-width characters.
- A role hijacking check identifies attempts to manipulate the AI persona using patterns like "act as root" or "you are now."
- The skill runs a comprehensive jailbreak detection sweep, looking for both direct and obfuscated variants of known exploit signatures.
- Instruction integrity is verified by checking for override keywords that attempt to invalidate previous command chains.
- The system performs an intent drift analysis to verify if the current request is consistent with established task boundaries.
- A final risk report is generated, providing a score and recommending automated actions such as blocking, sanitizing, or logging the input.
ClawGuard Shield Setup
To integrate ClawGuard Shield into your project, follow these standard installation steps for Openclaw Skills:
# Install the skill via the Openclaw CLI
openclaw install clawguard-shield
# Initialize the security configuration in your agent logic
clawguard init --version v3
Ensure that you call the validation function at the very beginning of your input processing pipeline to capture malicious intent before it reaches the core LLM.
ClawGuard Shield Data Schema & Taxonomy
ClawGuard Shield organizes security metadata into a structured report format for easy consumption by other Openclaw Skills:
| Attribute | Type | Description |
|---|---|---|
risk_score |
Integer | A score from 0-100 indicating the total threat level. |
risk_level |
String | Classification: SAFE, LOW_RISK, MEDIUM_RISK, HIGH_RISK, or CRITICAL. |
threats_found |
Array | A detailed list of detected patterns, including their location and severity. |
action_taken |
String | The automated response performed (e.g., REJECT, SANITIZE, or ALLOW). |
metadata |
Object | Contains technical details like encoding types detected or specific jailbreak variants identified. |
ClawGuard Shield Advanced Features
- Zero-Width Character Detection: Identifies and strips invisible characters used to hide malicious instructions from human reviewers.
- RTL Override Protection: Detects Right-to-Left override characters that attempt to manipulate the visual order of text to bypass filters.
- Obfuscation Variants Database: A frequently updated library of leetspeak and symbol-substitution patterns used in modern jailbreak attempts.
- Intent Integrity Validation: A deep analysis feature that compares user input against the conversation context to find subtle task tampering.
- Customizable Risk Matrix: Allows developers to adjust severity levels and response thresholds based on specific application requirements.
SKILL.md
Loading
Related Openclaw Skills
Tradfri Lights
A local integration for controlling IKEA TRÅDFRI lighting systems through a direct gateway connection using Openclaw Skills.
LightRAG Knowledge Base
A persistent, shared knowledge graph that allows multiple AI agents to store and query interconnected facts and historical data.
guoshun-bid-writer
An AI-driven technical bidding assistant that automates the creation of professional bidding documents, technical proposals, and qualification checks.
AgentVerif
A security and trust layer for AI agents providing OWASP scanning, cryptographic signing, and tamper detection.
Worst Travel Experience Challenge Agent
A humorous AI agent that intentionally plans the worst possible travel itineraries using real-time data for viral social media content.
TurboQuant+
A high-performance KV cache compression extension for llama.cpp designed to maximize context windows and model size on Apple Silicon.
Featured*
