CrucibleIR.Reliability.Guardrail (CrucibleIR v0.2.1)

Configuration for LLM guardrails and security.

Controls prompt injection detection, PII protection, and other safety measures for LLM applications.

Fields

:profiles - Security profiles to use (default: [:default])
:prompt_injection_detection - Detect prompt injection attempts
:jailbreak_detection - Detect jailbreak attempts
:pii_detection - Detect personally identifiable information
:pii_redaction - Redact detected PII
:content_moderation - Moderate content for safety
:fail_on_detection - Whether to fail when threats detected
:options - Additional guardrail options

Security Profiles

:default - Standard security measures
:strict - High security, may have false positives
:moderate - Balanced security and usability
:permissive - Minimal restrictions

Summary

Types

profile()

t()

Types

profile()

@type profile() :: :default | :strict | :moderate | :permissive | atom()

t()

@type t() :: %CrucibleIR.Reliability.Guardrail{
  content_moderation: boolean() | nil,
  fail_on_detection: boolean() | nil,
  jailbreak_detection: boolean() | nil,
  options: map() | nil,
  pii_detection: boolean() | nil,
  pii_redaction: boolean() | nil,
  profiles: [profile()],
  prompt_injection_detection: boolean() | nil
}