What it does
The screening extension asks whether retrieved text is trying to instruct the AI. A high score adds a fixed warning before the next model call; errors fail open.
Agent tooling
deer-flow
DeerFlow includes opt-in Jev extensions for text classification, prompt-injection screening, and pruning old tool results.
Bookmark: deer-flow
Keep this in your collection.
Leave a noteWhat would you try with this?
: deer-flow
Only you can see your notes.
Screenshot unavailable. Open the experiment ↗
How you can use it
Your developer can adapt this screening tool to check outside text. The extension sends text to the System One service. The service checks if that text contains hidden instructions. This helps the app spot attempts to trick it.
If the service marks the text as suspicious, the app adds a warning. This happens before the main model reads the prompt. The app still runs normally if the connection fails. This means dangerous text might slip through during an outage.