🛡️ Universal AI defense framework protecting agents from prompt injection attacks | 67% attack reduction | Based on academic research

Policy-driven guardrails for tool-using agents: sanitize untrusted tool output, redact runtime-provided sensitive terms, and optionally use PromptGuard for prompt-injection risk escalation, with policy linting, structured reporting, and eval scripts.

PromptGuard Cursor plugin -- LLM security rules, skills, commands, MCP server, and agent for AI-powered editors