ctf-ai-ml

SUSPICIOUS: the skill is internally consistent as an offensive CTF AI/ML attack guide, but that stated purpose is itself a high-risk offensive security capability for an AI agent. Install trust is relatively benign aside from unpinned pip packages; the primary risk comes from explicit instructions for prompt injection, system-prompt extraction, tool abuse, sensitive file access, and data exfiltration against targets.

An automated exploitation harness targets a chat API endpoint by sending crafted prompts to induce disclosure of protected information (system prompts, flags) and to trigger or influence tool-mediated privileged actions on the server side. It emphasizes prompt-injection style payloads and printing of returned sensitive data, representing a high security risk for supply chains unless confined to controlled environments.