“agent-defense”

Prompt Injection Defense for Tools: Context Separation and Tainting

Defend against prompt injection in agent tools using context separation and data tainting. Prevent tool outputs from hijacking agent reasoning.