Secure browser-agent workflows with prompt-boundary checks before agents act on page text, tool results, or external instructions.
CladisAI — Systematic threat classification for AI.