Overview

Developed by	jenisys
Date of development	2024-06-22
Validator type	rule-following
License	MIT
Input/Output	Output

Description

This guardrails validator provides support for cucumber-expressions, a simpler, more readable "regular expression" dialect with the following features:

• Supports built-in parameter_type(s) for common types, like: {int}, {float}, ...
• Supports user-defined parameter_type(s) with own regular expressions, type conversion and transformation
• Supports alternative text, like: apple/orange (matches: apple or orange)
• Support optional text, like: apple(s) (matches: apple, apples)

This validator is similar to regex_match, but often easier to use because:

• cucumber-expressions are often more readable (especially for people that are less familiar with regular expressions)
• less error-prone, because a readable placeholder(s) are used for the complex regular expression.

SEE ALSO:

• cucumber-expressions
• guardrails

Intended Use

Check if text follows a specified schema (described by this cucumber-expression).

Requirements

• Dependencies:

  • guardrails-ai>=0.5.0
  • cucumber-expressions>=17.1.0
  • rstr>=3.2.2

• Dev Dependencies:

  • pytest
  • pyright
  • ruff
  • codespell

Installation

$ guardrails hub install hub://guardrails/cucumber_expression_match

Usage Examples

Validating string output via Python

In this example, we apply the validator to a string output generated by an LLM.

# -- FILE: use_guardrails_cucumber_expression_match.py
from guardrails import Guard, OnFailAction
from guardrails.hub import CucumberExpressionMatch
from cucumber_expressions.parameter_type import ParameterType

# -- SETUP GUARD:
positive_number = ParameterType("positive_number", regexp=r"\d+", type=int)
guard = Guard().use(CucumberExpressionMatch,
    expression="I buy {positive_number} apple(s)/banana(s)/orange(s)",
    parameter_types=[positive_number],
    on_fail=OnFailAction.EXCEPTION
)

# -- VALIDATOR PASSES: Good cases
guard.validate("I buy 0 apples")    # Guardrail passes
guard.validate("I buy 1 apple")     # Guardrail passes
guard.validate("I buy 1 banana")    # Guardrail passes
guard.validate("I buy 2 bananas")   # Guardrail passes
guard.validate("I buy 1 orange")    # Guardrail passes
guard.validate("I buy 3 oranges")   # Guardrail passes

# -- VALIDATOR FAILS: Bad cases
try:
    guard.validate("I buy 2 melons")    # Guardrail fails: Unexpected fruit
    guard.validate("I buy -10 apples")  # Guardrail fails: Negative number
except Exception as e:
    print(e)