llm-ai-security
Pass
Audited by Gen Agent Trust Hub on Jul 11, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: The skill contains several examples of direct and indirect prompt injection payloads (e.g., 'Ignore the instructions above', 'Repeat the message above'). These are explicitly categorized as reference materials for security researchers to test target applications and do not attempt to override the operational behavior of the agent using the skill.
- [REMOTE_CODE_EXECUTION]: The documentation discusses RCE vulnerabilities in AI tools and provides examples of malicious commands (e.g., 'rm -rf /') to illustrate how researchers can test for tool injection in target systems. It does not contain code that executes these commands on the user's host.
- [EXTERNAL_DOWNLOADS]: The skill references established open-source security tools including 'garak', 'PyRIT', and 'promptfoo', as well as the Microsoft security research repository. These are well-known industry resources appropriate for the skill's stated purpose.
- [DATA_EXFILTRATION]: Concepts such as ASCII smuggling and markdown image exfiltration are described as attack vectors to look for during an audit. The URLs used in these examples (e.g., 'attacker.com') are placeholders for educational purposes.
Audit Metadata