Skip to main content

detect_injection

Function detect_injection 

Source
pub fn detect_injection(text: &str) -> InjectionScore
Expand description

Detect prompt-injection signals in user text.

Iterates the weighted rule set; for each match the rule’s weight is added to a running sum and its label is recorded. The aggregate score is the sum clamped to [0.0, 1.0] (sum.min(1.0)), so a single typical signal (weight 0.40.5) already scores at or above 0.4, two signals saturate to 1.0, and it is exactly 0.0 when nothing matches. Matched labels are deduplicated, preserving first-seen order.