llm-attacks-security

SUSPICIOUS: the skill is an offensive LLM attack guide and explicitly covers jailbreaks, prompt extraction, and sensitive data extraction. It also instructs the agent to ingest mutable third-party markdown from a personal GitHub repo, creating indirect prompt-injection risk if the agent can edit files based on that content. No malware or credential theft is directly shown, but the capability set is high-risk and not merely a benign documentation helper.