ai-security
Audited by Socket on Sep 19, 2026
3 alerts found:
SecurityAnomalyx2SUSPICIOUS. The skill is internally consistent as an AI security red-team guide, but it equips an AI agent with explicit offensive techniques for jailbreaks, prompt injection, system prompt extraction, and RAG poisoning. Install sources are mostly legitimate and do not look like malware delivery, yet the capability set is high risk for agent misuse and should not run with broad autonomous permissions against arbitrary targets.
The fragment is an adversarial-ML and security-control testing guide. It contains dual-use attack-generation logic and two active network-testing examples: a Scapy packet send and SQL-injection requests to a named remote endpoint. These create a meaningful operational security risk if executed against unauthorized systems, but the code does not show malware behaviors such as credential theft, persistence, exfiltration, or remote control. The risk assessment is limited because the surrounding file and execution context are not visible.
The fragment documents potentially unauthorized LLM/API model extraction and includes explicit rate-limit and detection-evasion guidance. It presents a meaningful abuse and security risk if executed against systems without authorization, but the visible portion contains no automatic execution, credential theft, persistence, destructive behavior, or conventional malware payload. Review surrounding code to determine whether these examples are exposed as runnable functionality or invoked during package installation or normal operation.