Composable guardrails for LLM agent workflows in R, with three layers of defense: input validation, code analysis, and output filtering.
Most guardrail products send your data to an external API. secureguard runs
entirely locally. It uses pattern-based prompt-injection detection, walks the
R AST to flag dangerous code before it executes, and scans output for PII and
secrets. You can use it on its own or plug it into
securer’s pre-execute hooks, and
securebench measures how well it performs.