Classify instruction-mode changes for prompt-boundary risk before coding agents switch behavior or trust new directions.
CladisAI — Systematic threat classification for AI.