Enumerations Classes

Smarter API Guardrail Manifest - enumerated datatypes.

Note

Experimental. The Guardrail was designed and coded by Claude Code (Anthropic’s Claude Opus 5.5), with Lawrence McDaniel as co-author. It is experimental, and will be documented.

class smarter.apps.guardrail.manifest.enum.SAMGuardrailAction(*values)

Bases: SmarterEnumAbstract

What a guardrail does when it triggers.

• log: record the event only.

• flag: record the event for review, and continue.

• redact: replace every match with replacement, and continue.

• transform: replace every match with replacement, which for the regex strategy may refer to groups, e.g. \1, and continue.

• block: stop, and return message instead of the LLM’s reply.

• escalate: record the event for human review, notify reviewers, and continue.

BLOCK = 'block'

ESCALATE = 'escalate'

FLAG = 'flag'

LOG = 'log'

REDACT = 'redact'

TRANSFORM = 'transform'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailCategory(*values)

Bases: SmarterEnumAbstract

The risk that a guardrail addresses.

It classifies findings, and does not change behavior.

COMPLIANCE = 'compliance'

CUSTOM = 'custom'

FORMATTING = 'formatting'

HALLUCINATION = 'hallucination'

JAILBREAK = 'jailbreak'

OFF_TOPIC = 'off_topic'

PII = 'pii'

PROMPT_INJECTION = 'prompt_injection'

SECRETS = 'secrets'

SELF_HARM = 'self_harm'

TOXICITY = 'toxicity'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailDetector(*values)

Bases: SmarterEnumAbstract

The built-in detectors of the detector strategy.

API_KEY = 'api_key'

AWS_ACCESS_KEY = 'aws_access_key'

CREDIT_CARD = 'credit_card'

EMAIL = 'email'

IBAN = 'iban'

IP_ADDRESS = 'ip_address'

JWT = 'jwt'

PASSWORD_ASSIGNMENT = 'password_assignment'

PHONE_NUMBER = 'phone_number'

PRIVATE_KEY = 'private_key'

US_SSN = 'us_ssn'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailMode(*values)

Bases: SmarterEnumAbstract

Whether a guardrail acts.

• enforce: it takes its action.

• monitor: it only records what its action would have been, to evaluate a new guardrail against real traffic before enforcing it.

ENFORCE = 'enforce'

MONITOR = 'monitor'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailModerationCategory(*values)

Bases: SmarterEnumAbstract

The categories of OpenAI’s moderation models.

HARASSMENT = 'harassment'

HARASSMENT_THREATENING = 'harassment/threatening'

HATE = 'hate'

HATE_THREATENING = 'hate/threatening'

ILLICIT = 'illicit'

ILLICIT_VIOLENT = 'illicit/violent'

SELF_HARM = 'self-harm'

SELF_HARM_INSTRUCTIONS = 'self-harm/instructions'

SELF_HARM_INTENT = 'self-harm/intent'

SEXUAL = 'sexual'

SEXUAL_MINORS = 'sexual/minors'

VIOLENCE = 'violence'

VIOLENCE_GRAPHIC = 'violence/graphic'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailStage(*values)

Bases: SmarterEnumAbstract

When a guardrail runs.

• input: on the user’s message, before it is sent to the LLM.

• output: on the LLM’s reply, before it is returned to the user.

• both: on both.

BOTH = 'both'

INPUT = 'input'

OUTPUT = 'output'

class smarter.apps.guardrail.manifest.enum.SAMGuardrailStrategy(*values)

Bases: SmarterEnumAbstract

How a guardrail detects what it guards against.

• regex: a regular expression.

• keyword: a list of words or phrases.

• detector: built-in, validated detectors of personal data and secrets, e.g. credit card numbers that pass the Luhn check.

• semantic: embedding similarity to reference examples.

• moderation: an LLM provider’s moderation model, e.g. OpenAI’s omni-moderation.

• llm_judge: an LLM that judges the text, with a prompt.

regex, keyword and detector are deterministic, fast and free. The others call an LLM provider, and add latency and cost to each prompt.

DETECTOR = 'detector'

KEYWORD = 'keyword'

LLM_JUDGE = 'llm_judge'

MODERATION = 'moderation'

REGEX = 'regex'

SEMANTIC = 'semantic'

classmethod locating()

Return the strategies that locate the text they match, as redact and transform require.

Return type:

list[str]

classmethod scored()

Return the strategies that produce a confidence score, and use threshold.

Return type:

list[str]