Struct InjectionDetector 

pub struct InjectionDetector { /* private fields */ }

Detects prompt-injection attempts in LLM responses.

Two layers of defense:

1. Instruction-boundary enforcement — scans for patterns that indicate the LLM is trying to override its system instructions or produce unauthorized output.
2. Output schema validation — ensures structured output (tool calls, JSON responses) conforms to expected format.

Implementations

impl InjectionDetector

pub fn new() -> Self

Create a new injection detector with default settings

pub fn permissive() -> Self

Create a detector with all checks disabled (permissive)

pub fn with_instruction_boundary(self, enabled: bool) -> Self

Enable or disable instruction-boundary checking

pub fn with_output_schema(self, enabled: bool) -> Self

Enable or disable output schema checking

pub fn with_custom_pattern(self, pattern: &str) -> Self

Add a custom injection pattern

pub fn check(&self, content: &str) -> InjectionVerdict

Check LLM response content for injection attempts.

Returns InjectionVerdict::Clean if the output appears safe, or InjectionVerdict::Suspicious(reason) if injection is detected.

Trait Implementations

impl Clone for InjectionDetector

fn clone(&self) -> InjectionDetector

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for InjectionDetector

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for InjectionDetector

fn default() -> Self

Returns the “default value” for a type.