Skip to main content

guardrails.input_output_safety

Input/output safety and content guardrail capabilities.

PromptInjectionCapability Objects​

@dataclass
class PromptInjectionCapability(AbstractCapability[Any])

Prompt injection detector adapted from pydantic-ai-shields.

PiiDetectorCapability Objects​

@dataclass
class PiiDetectorCapability(AbstractCapability[Any])

PII detector for input prompts.

SecretRedactionCapability Objects​

@dataclass
class SecretRedactionCapability(AbstractCapability[Any])

Block secret leakage in model output.

BlockedKeywordsCapability Objects​

@dataclass
class BlockedKeywordsCapability(AbstractCapability[Any])

Keyword/regex blocklist for prompts.

NoRefusalsCapability Objects​

@dataclass
class NoRefusalsCapability(AbstractCapability[Any])

Block refusal-style model outputs.

DataScopeCapability Objects​

@dataclass
class DataScopeCapability(AbstractCapability[Any])

Simple data-scope enforcement on tool args/results.

ContentSafetyCapability Objects​

@dataclass
class ContentSafetyCapability(AbstractCapability[Any])

Detect prompt-injection patterns in tool output.