Module safety 

Safety utilities for AI-native applications.

Provides secret masking, output sanitization, and error classification to prevent PII leaks and injection attacks.

use llm_kernel::safety::{mask_secrets, sanitize_output};

let masked = mask_secrets("key=sk-abc123def456 token=Bearer xyz789");
assert!(masked.contains("****"));

let clean = sanitize_output("hello\u{202E}evil\u{FFF9}tags");
assert!(!clean.contains('\u{202E}'));

Re-exports

pub use classify::FailureCategory;

pub use classify::classify_failure;

pub use injection::InjectionScore;

pub use injection::detect_injection;

pub use sanitize::mask_secrets;

pub use sanitize::sanitize_output;

pub use sanitize::strip_ansi;

Modules

classify

Error classification via regex rules.

injection

Prompt-injection detection via weighted regex rules. Prompt-injection detection via weighted regex rules.

sanitize

Output sanitization — secret masking and control character removal.