The application formats retrieved passages, tool output, or user data into a designated prompt section. Clear delimiters, provenance, and instruction hierarchy help the model distinguish evidence from commands embedded in untrusted content.
Context injection adds selected external information to a model request so generation can use it.
The application formats retrieved passages, tool output, or user data into a designated prompt section. Clear delimiters, provenance, and instruction hierarchy help the model distinguish evidence from commands embedded in untrusted content.