Last released Oct 2, 2026
Runtime guardrails for AI coding agents: blocks dangerous commands, keeps secrets out of the model, and catches unverified 'tests pass' claims. Claude Code hooks, no dependencies.