A specialized security skill that enables AI agents to recognize, audit, and mitigate malicious instruction patterns in external skill registries.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install security-awareness-skill
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 security-awareness-skill 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 Agent Security Patterns?
The Agent Security Patterns skill is a critical defense mechanism designed for autonomous AI systems that consume and execute third-party code. Based on the Snyk ToxicSkills research, which revealed that 13.4% of public skills contain critical security issues, this skill provides a logic-based framework for agents to identify threats like prompt injection and obfuscated malware. It transitions the agent from a passive command executor into a security-aware reasoning system.
By implementing this skill, developers ensure their agents can detect harmful behaviors such as instruction overrides, hidden external fetches, and unauthorized credential requests. This is essential for anyone building robust environments using Openclaw Skills where trust and safety are paramount.
Agent Security Patterns Use Cases
- Auditing third-party skills from public registries before execution to prevent system compromise.
- Identifying obfuscated payloads or encoded scripts that hide malicious intent.
- Verifying that external fetch requests in a skill are legitimate and necessary for the task.
- Protecting sensitive environment variables and API keys from exfiltration through malicious Openclaw Skills.
How Agent Security Patterns Works
- The agent analyzes the target SKILL.md file against six primary malicious patterns, including obfuscation and purpose mismatch.
- It applies a 7-question decision framework to evaluate the transparency and necessity of every instruction.
- The system performs a Transparency, Purpose, and Permission test to validate the skill's integrity.
- If a risk is identified, the agent pauses the workflow and generates a detailed security report for the user.
- The agent offers safer alternatives or requests explicit user verification before proceeding with potentially dangerous Openclaw Skills.
Agent Security Patterns Setup
To integrate security awareness into your agent, include this skill in your agent's active skill directory:
# Clone the security patterns repository
git clone https://github.com/jisokuor/agent-security-patterns.git
# Enable the skill in your local configuration
openclaw install ./agent-security-patterns
Agent Security Patterns Data Schema & Taxonomy
| Component | Type | Description |
|---|---|---|
| Pattern Guide | Markdown | Defines 6 common attack vectors used in malicious Openclaw Skills. |
| Decision Framework | Checklist | The 7-question logic check used by the agent for reasoning. |
| Validation Tests | Framework | The Transparency, Purpose, and Permission tests for skill auditing. |
| Response Logic | Templates | Pre-defined communication patterns for alerting users to security risks. |
Agent Security Patterns Advanced Features
- Recognition of Pattern 1: Override Instructions that attempt to bypass system safety prompts.
- Automated detection of Pattern 2: Obfuscated Payloads including Base64 and Unicode encoding.
- External Fetch auditing to prevent 'curl | bash' style attacks in autonomous environments.
- Purpose Mismatch detection to catch 'bait-and-switch' skills that claim to be one thing but do another.
- Interactive response framework that allows for sandboxing or limited-permission execution of Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
SJ-IRAC Trademark Refusal Reasoning Engine
A law-firm-grade reasoning engine that transforms CNIPA refusal notices into examiner-ready IRAC briefs and verified evidence chains.
Jiang IRAC Trademark Opposition Engine
A law-firm-grade CNIPA trademark opposition and invalidation engine that transforms case materials into examiner-ready legal arguments using IRAC and SJ-6 evidence chains.
AgentGuilds
An on-chain AI labor marketplace on Monad that enables users to create missions and hire autonomous agents for specialized tasks without managing private keys.
TrendRadar Agent
An AI-powered agent designed to detect and classify web trend signals for multiple keywords.
MongoDB Document Storage
A powerful MongoDB interface for AI agents to perform persistent document storage and complex data operations via a simple CLI.
Qwen3-TTS Local Inference
A high-performance local Python engine for generating speech, designing voices, and cloning audio using Qwen3-TTS without external API dependencies.
Featured*
