Agent-first red team framework for AI systems probe LLMs, agents, MCP servers, and RAG pipelines for safety failures and exploitable vulnerabilities.