Struct TrustPolicy 

pub struct TrustPolicy {
    pub open: String,
    pub close: String,
    pub screen: Option<Box<dyn InjectionScreen>>,
}

How untrusted content is fenced when assembling a prompt.

Fields

open: String

Opening/closing delimiters for the untrusted region.

close: String

screen: Option<Box<dyn InjectionScreen>>

Optional screen applied to each untrusted span.

Implementations

impl TrustPolicy

pub fn with_screen(self, screen: Box<dyn InjectionScreen>) -> Self

pub fn assemble(&self, parts: &[TrustedContent]) -> AssembledPrompt

Assemble content: trusted spans pass through verbatim; untrusted spans are fenced in a delimited, clearly-labelled region (so the model treats them as data, not instructions). Untrusted spans are run through the screen if configured.

Trait Implementations

impl Default for TrustPolicy

fn default() -> Self

Returns the “default value” for a type.