threat-model

What This Skill Does

The threat-model skill acts as an automated security analyst within your OpenClaw agent, enabling rigorous proactive assessment of systems and architectures. By utilizing the proven STRIDE methodology (Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, and Elevation of Privilege), this skill identifies potential vulnerabilities before they can be exploited. It provides a structured framework for analyzing custom risks specific to AI agents, such as prompt injection, unauthorized role changes, and crypto-wallet security, ensuring your agent environment remains resilient against both external and internal threats.

Installation

To integrate this security analysis tool into your agent, run the following command in your terminal: clawhub install openclaw/skills/skills/sa9saq/threat-model

Use Cases

This skill is designed for developers and security-conscious users who want to build secure autonomous agents. Use it during the design phase to identify potential attack surfaces in your agent's infrastructure. It is particularly valuable for agents dealing with financial assets or sensitive API keys, as it helps define the necessary guardrails and audit logging strategies required to maintain system integrity. Furthermore, it can be used for tabletop exercises, simulating attack paths to verify that your current defensive measures, such as rate limiting and authentication controls, are functioning as intended.

Example Prompts

1. "Perform a STRIDE analysis on my current agent architecture to identify critical security gaps."
2. "Create an attack tree for a hypothetical scenario where an attacker attempts to exfiltrate my system prompt via indirect prompt injection."
3. "Evaluate the risk level of my API exposure using the risk matrix, considering a 'Likely' occurrence probability and 'High' impact."

Tips & Limitations

To maximize the effectiveness of this skill, ensure you provide the agent with accurate architectural details, as the quality of the model depends on the specificity of the input context. Remember that this skill identifies risks theoretically; it does not perform live penetration testing. For real-world hardening, always combine the findings from this skill with standard security practices like secret rotation, network-level firewalls, and rigorous input validation at the code level. Avoid inputting sensitive credentials or actual production secrets when defining your architecture for analysis.