pyrit.prompt_converter

Prompt converters for transforming prompts before sending them to targets in red teaming workflows.

Converters are organized into categories: Text-to-Text (encoding, obfuscation, translation, variation), Audio (text-to-audio, audio-to-text, audio-to-audio), Image (text-to-image, image-to-image), Video (image-to-video), File (text-to-PDF/URL), Selective Converting (partial prompt transformation), and Human-in-the-Loop (interactive review). Converters can be stacked together to create complex transformation pipelines for testing AI system robustness.

Functions¶

get_converter_modalities¶

get_converter_modalities() → list[tuple[str, list[PromptDataType], list[PromptDataType]]]

Retrieve a list of all converter classes and their supported input/output modalities by reading the SUPPORTED_INPUT_TYPES and SUPPORTED_OUTPUT_TYPES class attributes.

Returns:

• list[tuple[str, list[PromptDataType], list[PromptDataType]]] — list[tuple[str, list[PromptDataType], list[PromptDataType]]]: A sorted list of tuples containing:

• Converter class name (str)

• List of supported input modalities (list[PromptDataType])

• List of supported output modalities (list[PromptDataType])

• list[tuple[str, list[PromptDataType], list[PromptDataType]]] — Sorted by input modality, then output modality, then converter name.

AddImageTextConverter¶

Bases: PromptConverter

Adds a string to an image and wraps the text into multiple lines if necessary.

This class is similar to :class:AddTextImageConverter except we pass in an image file path as an argument to the constructor as opposed to text.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by adding it as text to the image.

Returns:

• ConverterResult — The result containing path to the updated image.

Raises:

• ValueError — If the input type is not supported.

AddImageVideoConverter¶

Bases: PromptConverter

Adds an image to a video at a specified position.

Currently the image is placed in the whole video, not at a specific timepoint.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'image_path') → ConverterResult

Convert the given prompt (image) by adding it to a video.

Returns:

• ConverterResult — The result containing filename of the converted video.

Raises:

• ValueError — If the input type is not supported.

AddTextImageConverter¶

Bases: PromptConverter

Adds a string to an image and wraps the text into multiple lines if necessary.

This class is similar to :class:AddImageTextConverter except we pass in text as an argument to the constructor as opposed to an image file path.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'image_path') → ConverterResult

Convert the given prompt (image) by adding text to it.

Returns:

• ConverterResult — The result containing path to the updated image.

Raises:

• ValueError — If the input type is not supported.

AllWordsSelectionStrategy¶

Bases: WordSelectionStrategy

Selects all words (default strategy).

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select all words.

Returns:

• list[int] — List[int]: All word indices.

AnsiAttackConverter¶

Bases: PromptConverter

Generates prompts with ANSI codes to evaluate LLM behavior and system risks.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into an ANSI attack scenario.

Returns:

• ConverterResult — The result containing the generated ANSI scenario prompt.

Raises:

• ValueError — If the input type is not supported.

input_supported¶

input_supported(input_type: PromptDataType) → bool

Check if the input type is supported.

Returns:

• bool — True if the input type is supported, False otherwise.

output_supported¶

output_supported(output_type: PromptDataType) → bool

Check if the output type is supported.

Returns:

• bool — True if the output type is supported, False otherwise.

AsciiArtConverter¶

Bases: PromptConverter

Uses the art package to convert text into ASCII art.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into ASCII art.

Returns:

• ConverterResult — The result containing the ASCII art representation of the prompt.

Raises:

• ValueError — If the input type is not supported.

AsciiSmugglerConverter¶

Bases: SmugglerConverter

Implements encoding and decoding using Unicode Tags.

If ‘control’ is True, the encoded output is wrapped with: - U+E0001 (start control tag) - U+E007F (end control tag)

Replicates the functionality detailed in the following blog post: Rehberger, 2024

Constructor Parameters:

Methods:

decode_message¶

decode_message(message: str) → str

Decode a message encoded with Unicode Tags.

For each character in the Unicode Tags range, subtracts 0xE0000. Skips control tags if present.

Returns:

• str — The decoded message.

encode_message¶

encode_message(message: str) → tuple[str, str]

Encode the message using Unicode Tags.

Each ASCII printable character (0x20-0x7E) is mapped to a corresponding Unicode Tag (by adding 0xE0000). If control mode is enabled, wraps the output.

Returns:

• tuple[str, str] — Tuple[str, str]: A tuple with a summary of code points and the encoded message.

AskToDecodeConverter¶

Bases: PromptConverter

Wraps encoded text with prompts that ask a target to decode it.

This converter takes encoded text (e.g., Base64, ROT13, Morse code) and wraps it in various prompt templates that request decoding. The prompts can be generic (“Decode the following text:”) or encoding-specific (“Base64 encoded string:”). This is useful for testing whether AI systems will decode potentially harmful encoded content when explicitly asked.

Credit to Garak: garak/probes/encoding.py

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given encoded text by wrapping it with a decoding request prompt.

Returns:

• ConverterResult — The result containing the converted prompt.

Raises:

• ValueError — If the input type is not supported (only “text” is supported).

AtbashConverter¶

Bases: PromptConverter

‘Hello 123’ would encode to ‘Svool 876’.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt using the Atbash cipher.

Returns:

• ConverterResult — The result containing the encoded prompt.

Raises:

• ValueError — If the input type is not supported.

AudioEchoConverter¶

Bases: PromptConverter

Adds an echo effect to an audio file.

The echo is created by mixing a delayed, attenuated copy of the signal back into the original. The delay and decay parameters control the timing and loudness of the echo respectively. Sample rate, bit depth, and channel count are preserved.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file by adding an echo effect.

Returns:

• ConverterResult — The result containing the converted audio file path.

Raises:

• ValueError — If the input type is not supported.

• Exception — If there is an error during the conversion process.

AudioFrequencyConverter¶

Bases: PromptConverter

Shifts the frequency of an audio file by a specified value. By default, it will shift it above the human hearing range (=20 kHz).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file by shifting its frequency.

Returns:

• ConverterResult — The result containing the audio file path.

Raises:

• ValueError — If the input type is not supported.

• Exception — If there is an error during the conversion process.

AudioSpeedConverter¶

Bases: PromptConverter

Changes the playback speed of an audio file without altering pitch or other audio characteristics.

A speed_factor > 1.0 speeds up the audio (shorter duration), while a speed_factor < 1.0 slows it down (longer duration). The converter resamples the audio signal using interpolation so that the sample rate, bit depth, and number of channels remain unchanged.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file by changing its playback speed.

The audio is resampled via interpolation so that the output has a different number of samples (and therefore a different duration) while keeping the original sample rate. This preserves the pitch and tonal qualities of the audio.

Returns:

• ConverterResult — The result containing the converted audio file path.

Raises:

• ValueError — If the input type is not supported.

• Exception — If there is an error during the conversion process.

AudioVolumeConverter¶

Bases: PromptConverter

Changes the volume of an audio file by scaling the amplitude.

A volume_factor > 1.0 increases the volume (louder), while a volume_factor < 1.0 decreases it (quieter). A volume_factor of 1.0 leaves the audio unchanged. The converter scales all audio samples by the given factor and clips the result to the valid range for the original data type. Sample rate, bit depth, and number of channels are preserved.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file by changing its volume.

The audio samples are scaled by the volume factor. For integer audio formats the result is clipped to prevent overflow.

Returns:

• ConverterResult — The result containing the converted audio file path.

Raises:

• ValueError — If the input type is not supported.

• Exception — If there is an error during the conversion process.

AudioWhiteNoiseConverter¶

Bases: PromptConverter

Adds white noise to an audio file.

White noise is generated and mixed into the original signal at a level controlled by the noise_scale parameter. The output preserves the original sample rate, bit depth, channel count, and number of samples.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file by adding white noise.

Returns:

• ConverterResult — The result containing the converted audio file path.

Raises:

• ValueError — If the input type is not supported.

• Exception — If there is an error during the conversion process.

AzureSpeechAudioToTextConverter¶

Bases: PromptConverter

Transcribes a .wav audio file into text using Azure AI Speech service.

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/speech-to-text

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'audio_path') → ConverterResult

Convert the given audio file into its text representation.

Returns:

• ConverterResult — The result containing the transcribed text.

Raises:

• ValueError — If the input type is not supported or if the provided file is not a .wav file.

recognize_audio¶

recognize_audio(audio_bytes: bytes) → str

Recognize audio file and return transcribed text.

Returns:

• str — Transcribed text.

Raises:

• ModuleNotFoundError — If the azure.cognitiveservices.speech module is not installed.

stop_cb¶

stop_cb(evt: Any, recognizer: Any) → None

Stop continuous recognition upon receiving an event ‘evt’.

Raises:

• ModuleNotFoundError — If the azure.cognitiveservices.speech module is not installed.

transcript_cb¶

transcript_cb(evt: Any, transcript: list[str]) → None

Append transcribed text upon receiving a “recognized” event.

AzureSpeechTextToAudioConverter¶

Bases: PromptConverter

Generates a wave file from a text prompt using Azure AI Speech service.

https://learn.microsoft.com/en-us/azure/ai-services/speech-service/text-to-speech

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given text prompt into its audio representation.

Returns:

• ConverterResult — The result containing the audio file path.

Raises:

• ModuleNotFoundError — If the azure.cognitiveservices.speech module is not installed.

• RuntimeError — If there is an error during the speech synthesis process.

• ValueError — If the input type is not supported or if the prompt is empty.

Base2048Converter¶

Bases: PromptConverter

Converter that encodes text to base2048 format.

This converter takes input text and converts it to base2048 encoding, which uses 2048 different Unicode characters to represent binary data. This can be useful for obfuscating text or testing how systems handle encoded Unicode content.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to base2048 encoding.

Returns:

• ConverterResult — The converted text representation of the original prompt in base2048.

Raises:

• ValueError — If the input type is not supported.

Base64Converter¶

Bases: PromptConverter

Converter that encodes text to base64 format.

This converter takes input text and converts it to base64 encoding, which can be useful for obfuscating text or testing how systems handle encoded content.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to base64 encoding.

Returns:

• ConverterResult — The converted text representation of the original prompt in base64.

Raises:

• ValueError — If the input type is not supported.

BinAsciiConverter¶

Bases: WordLevelConverter

Converts text to various binary-to-ASCII encodings.

Supports hex, quoted-printable, and UUencode formats.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a word using the specified encoding function.

Returns:

• str — The encoded word.

Raises:

• ValueError — If an unsupported encoding function is encountered.

join_words¶

join_words(words: list[str]) → str

Join words appropriately based on the encoding type and selection mode.

Returns:

• str — The joined string.

BinaryConverter¶

Bases: WordLevelConverter

Transforms input text into its binary representation with configurable bits per character (8, 16, or 32).

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

join_words¶

join_words(words: list[str]) → str

Join the converted words with the binary representation of a space.

Returns:

• str — The final joined string with spaces in binary format.

validate_input¶

validate_input(prompt: str) → None

Check if bits_per_char is sufficient for the characters in the prompt.

Raises:

• ValueError — If bits_per_char is too small to represent any character in the prompt.

BrailleConverter¶

Bases: PromptConverter

Converts text into Braille Unicode representation.

This converter transforms standard text into Braille patterns using Unicode Braille characters (U+2800 to U+28FF). It supports lowercase and uppercase letters, numbers, common punctuation, and spaces. Uppercase letters are prefixed with the Braille capitalization indicator.

The Braille mapping is based on the implementation from Garak: garak/probes/encoding.py

Note: This converter is useful for testing how AI systems handle Braille-encoded text, which can be used to obfuscate potentially harmful content.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given text into Braille Unicode representation.

Returns:

• ConverterResult — The text converted to Braille Unicode characters.

Raises:

• ValueError — If the input type is not supported (only “text” is supported).

CaesarConverter¶

Bases: PromptConverter

Encodes text using the Caesar cipher with a specified offset.

Using offset=1, ‘Hello 123’ would encode to ‘Ifmmp 234’, as each character would shift by 1. Shifts for digits 0-9 only work if the offset is less than 10, if the offset is equal to or greater than 10, any numeric values will not be shifted.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt using the Caesar cipher.

Returns:

• ConverterResult — The result containing the converted prompt and its type.

Raises:

• ValueError — If the input type is not supported.

CharSwapConverter¶

Bases: WordLevelConverter

Applies character swapping to words in the prompt to test adversarial textual robustness.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

CharacterSpaceConverter¶

Bases: PromptConverter

Spaces out the input prompt and removes specified punctuations.

For more information on the bypass strategy, refer to Robust Intelligence, 2024.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by removing punctuation and spacing out characters.

Returns:

• ConverterResult — The result containing the converted text.

Raises:

• ValueError — If the input type is not supported.

CodeChameleonConverter¶

Bases: PromptConverter

Code Chameleon Converter Lv et al., 2024.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by applying the specified encryption function.

Returns:

• ConverterResult — The result containing the converted prompt.

Raises:

• ValueError — If the input type is not supported.

ColloquialWordswapConverter¶

Bases: PromptConverter

Converts text by replacing words with regional colloquial alternatives.

Supports loading substitutions from YAML files (e.g., Singaporean, Filipino, Indian) or accepting a custom substitution dictionary. Defaults to Singaporean substitutions.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by replacing words with regional colloquial alternatives.

Returns:

• ConverterResult — The result containing the converted prompt.

Raises:

• ValueError — If the input type is not supported.

ConverterResult¶

The result of a prompt conversion, containing the converted output and its type.

DenylistConverter¶

Bases: LLMGenericTextConverter

Replaces forbidden words or phrases in a prompt with synonyms using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by removing any words or phrases that are in the denylist, replacing them with synonymous words.

Returns:

• ConverterResult — The result containing the modified prompt.

DiacriticConverter¶

Bases: PromptConverter

Applies diacritics to specified characters in a string.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by applying diacritics to specified characters.

Returns:

• ConverterResult — The result containing the modified text.

Raises:

• ValueError — If the input type is not supported.

EcojiConverter¶

Bases: PromptConverter

Converter that encodes text using Ecoji encoding.

Ecoji is an encoding scheme that represents binary data using emojis. See https://ecoji.io/ for more details.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to Ecoji encoding.

Returns:

• ConverterResult — The result containing the Ecoji-encoded text.

Raises:

• ValueError — If the input type is not supported.

EmojiConverter¶

Bases: WordLevelConverter

Converts English text to randomly chosen circle or square character emojis.

Inspired by src/utils.ts

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

FirstLetterConverter¶

Bases: WordLevelConverter

Replaces each word of the prompt with its first letter (or digit). Whitespace and words that do not contain any letter or digit are ignored.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

join_words¶

join_words(words: list[str]) → str

Join the converted words using the specified letter separator.

Returns:

• str — The joined string of converted words.

FlipConverter¶

Bases: PromptConverter

Flips the input text prompt. For example, “hello me” would be converted to “em olleh”.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by reversing the text.

Returns:

• ConverterResult — The converted text representation of the original prompt with characters reversed.

Raises:

• ValueError — If the input type is not supported.

HumanInTheLoopConverter¶

Bases: PromptConverter

Allows review of each prompt sent to a target before sending it.

Users can choose to send the prompt as is, modify the prompt, or run the prompt through one of the passed-in converters before sending it.

.. deprecated:: This converter is deprecated and will be removed in v0.13.0. Use the React-based GUI (CoPyRIT) instead.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by allowing user interaction before sending it to a target.

Returns:

• ConverterResult — The result containing the modified prompt.

Raises:

• ValueError — If no converters are provided and the user chooses to run a converter.

ImageCompressionConverter¶

Bases: PromptConverter

Compresses images to reduce file size while preserving visual quality.

This converter supports multiple compression strategies across JPEG, PNG, and WEBP formats, each with format-specific optimization settings. It can maintain the original image format or convert between formats as needed.

When converting images with transparency (alpha channel) to JPEG format, the converter automatically composites the transparent areas onto a solid background color.

Supported input types: File paths to any image that PIL can open (or URLs pointing to such images): https://pillow.readthedocs.io/en/stable/handbook/image-file-formats.html#fully-supported-formats

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'image_path') → ConverterResult

Convert the given prompt (image) by compressing it.

Returns:

• ConverterResult — The result containing path to the compressed image.

Raises:

• ValueError — If the input type is not supported.

IndexSelectionStrategy¶

Bases: TextSelectionStrategy

Selects text based on absolute character indices.

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select a range based on absolute character indices.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index).

InsertPunctuationConverter¶

Bases: PromptConverter

Inserts punctuation into a prompt to test robustness.

Punctuation insertion: inserting single punctuations in string.punctuation. Words in a prompt: a word does not contain any punctuation and space. “a1b2c3” is a word; “a1 2” are 2 words; “a1,b,3” are 3 words.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text', punctuation_list: Optional[list[str]] = None) → ConverterResult

Convert the given prompt by inserting punctuation.

Returns:

• ConverterResult — The result containing an iteration of modified prompts.

Raises:

• ValueError — If the input type is not supported.

JsonStringConverter¶

Bases: PromptConverter

Converts a string to a JSON-safe format using json.dumps().

This converter is useful when a string needs to be embedded within a JSON payload, such as when sending prompts to HTTP targets that expect JSON-formatted requests. The converter properly escapes special characters like quotes, newlines, backslashes, and unicode characters.

The output is the escaped string content without the surrounding quotes that json.dumps() adds, making it ready to be inserted into a JSON string field.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to a JSON-safe string.

Returns:

• ConverterResult — The result containing the JSON-escaped string.

Raises:

• ValueError — If the input type is not supported.

KeywordSelectionStrategy¶

Bases: TextSelectionStrategy

Selects text around a keyword with optional context.

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select the range around the first occurrence of the keyword.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index) including context, or (0, 0) if keyword not found.

LLMGenericTextConverter¶

Bases: PromptConverter

Represents a generic LLM converter that expects text to be transformed (e.g. no JSON parsing or format).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt using an LLM via the specified converter target.

Returns:

• ConverterResult — The result containing the converted output and its type.

Raises:

• ValueError — If the input type is not supported.

LeetspeakConverter¶

Bases: WordLevelConverter

Converts a string to a leetspeak version.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

MaliciousQuestionGeneratorConverter¶

Bases: LLMGenericTextConverter

Generates malicious questions using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the input prompt into malicious questions.

Returns:

• ConverterResult — The result of the conversion.

input_supported¶

input_supported(input_type: PromptDataType) → bool

Check if the input type is supported.

Returns:

• bool — True if the input type is “text”, False otherwise.

output_supported¶

output_supported(output_type: PromptDataType) → bool

Check if the output type is supported.

Returns:

• bool — True if the output type is “text”, False otherwise.

MathObfuscationConverter¶

Bases: PromptConverter

Convert text into character-level algebraic identities.

This converter encodes each character of the input text into an equation of the form X = nX - (n - 1)X, where n is a randomly chosen integer greater than or equal to 2. This creates a deterministic, reversible obfuscation of the original input.

The transformation follows these rules:

• Each non-space character becomes one algebraic line.

• Space characters are represented as blank output lines.

• Newline characters are preserved as blank output lines.

An inline hint is added after the first equation, and a suffix instruction is appended to prompt the model to decode the content.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert text into algebraic obfuscated form.

Each non-space character in the input string is transformed into a corresponding algebraic identity. Space characters are represented as blank output lines, preserving word boundaries. Newline characters are preserved as block breaks in the output.

Returns:

• ConverterResult — An instance containing the obfuscated text and output format.

Raises:

• ValueError — If input_type is not "text".

MathPromptConverter¶

Bases: LLMGenericTextConverter

Converts natural language instructions into symbolic mathematics problems using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into a mathematical problem format.

Returns:

• ConverterResult — The result containing the mathematical representation and real-world example.

MorseConverter¶

Bases: PromptConverter

Encodes prompts using morse code.

Uses ‘-’ and ‘.’ characters, with ’ ’ to separate characters and ‘/’ to separate words. Invalid or unsupported characters are replaced with an error sequence ‘........’.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to morse code.

Returns:

• ConverterResult — The result containing the morse code representation of the prompt.

Raises:

• ValueError — If the input type is not supported.

NatoConverter¶

Bases: PromptConverter

Converts text into NATO phonetic alphabet representation.

This converter transforms standard text into NATO phonetic alphabet format, where each letter is replaced with its corresponding NATO phonetic code word (e.g., “A” becomes “Alfa”, “B” becomes “Bravo”). Only alphabetic characters are converted; non-alphabetic characters are ignored.

The NATO phonetic alphabet is the most widely used spelling alphabet, designed to improve clarity of voice communication. This converter can be used to test how AI systems handle phonetically encoded text, which can be used to obfuscate potentially harmful content.

Reference: NATO phonetic alphabet

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given text into NATO phonetic alphabet representation.

Returns:

• ConverterResult — The text converted to NATO phonetic alphabet format.

Raises:

• ValueError — If the input type is not supported (only “text” is supported).

NegationTrapConverter¶

Bases: PromptConverter

Converts a prompt into a negation-based logical trap. This technique exploits LLM reasoning patterns by asking the model to confirm or deny a wrong answer, potentially causing it to reveal the correct one.

This technique was discovered during CTF red teaming exercises where targets would leak information when asked to evaluate incorrect statements.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the prompt into a negation trap.

This technique works by presenting an obviously wrong answer and asking the target to correct it, which may cause it to reveal protected information.

Returns:

• ConverterResult — The prompt converted to a negation trap.

Raises:

• ValueError — If the input type is not supported.

input_supported¶

input_supported(input_type: PromptDataType) → bool

Check if the input type is supported.

Returns:

• bool — True if the input type is supported, False otherwise.

output_supported¶

output_supported(output_type: PromptDataType) → bool

Check if the output type is supported.

Returns:

• bool — True if the output type is supported, False otherwise.

NoiseConverter¶

Bases: LLMGenericTextConverter

Injects noise errors into a conversation using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

PDFConverter¶

Bases: PromptConverter

Converts a text prompt into a PDF file.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into a PDF.

If a template is provided, it injects the prompt into the template, otherwise, it generates a simple PDF with the prompt as the content. Further it can modify existing PDFs.

Returns:

• ConverterResult — The result containing the full file path to the generated PDF.

Raises:

• ValueError — If the input type is not supported.

PersuasionConverter¶

Bases: PromptConverter

Rephrases prompts using a variety of persuasion techniques.

Based on Zeng et al., 2024.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt using the persuasion technique specified during initialization.

Returns:

• ConverterResult — The result containing the converted prompt text.

Raises:

• ValueError — If the input type is not supported.

send_persuasion_prompt_async¶

send_persuasion_prompt_async(request: Message) → str

Send the prompt to the converter target and process the response.

Returns:

• str — The converted prompt text extracted from the response.

Raises:

• InvalidJsonException — If the response is not valid JSON or missing expected keys.

PositionSelectionStrategy¶

Bases: TextSelectionStrategy

Selects text based on proportional start and end positions.

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select a range based on the relative position in the text.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index).

PromptConverter¶

Bases: Identifiable

Base class for converters that transform prompts into a different representation or format.

Concrete subclasses must declare their supported input and output modalities using class attributes:

• SUPPORTED_INPUT_TYPES: tuple of PromptDataType values that the converter accepts

• SUPPORTED_OUTPUT_TYPES: tuple of PromptDataType values that the converter produces

These attributes are enforced at class definition time for all non-abstract subclasses.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into the target format supported by the converter.

Returns:

• ConverterResult — The result containing the converted output and its type.

convert_tokens_async¶

convert_tokens_async(prompt: str, input_type: PromptDataType = 'text', start_token: str = '⟪', end_token: str = '⟫') → ConverterResult

Convert substrings within a prompt that are enclosed by specified start and end tokens. If there are no tokens present, the entire prompt is converted.

Returns:

• ConverterResult — The prompt with specified substrings converted.

Raises:

• ValueError — If the input is inconsistent.

input_supported¶

input_supported(input_type: PromptDataType) → bool

Check if the input type is supported by the converter.

Returns:

• bool — True if the input type is supported, False otherwise.

output_supported¶

output_supported(output_type: PromptDataType) → bool

Check if the output type is supported by the converter.

Returns:

• bool — True if the output type is supported, False otherwise.

ProportionSelectionStrategy¶

Bases: TextSelectionStrategy

Selects a proportion of text anchored to a specific position (start, end, middle, or random).

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select a proportion of text based on the anchor position.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index).

QRCodeConverter¶

Bases: PromptConverter

Converts a text string to a QR code image.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt to a QR code image.

Returns:

• ConverterResult — The result containing filename of the converted QR code image.

Raises:

• ValueError — If the input type is not supported or if the prompt is empty.

ROT13Converter¶

Bases: WordLevelConverter

Encodes prompts using the ROT13 cipher.

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

RandomCapitalLettersConverter¶

Bases: PromptConverter

Takes a prompt and randomly capitalizes it by a percentage of the total characters.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by randomly capitalizing a percentage of its characters.

Returns:

• ConverterResult — The result containing the converted text.

Raises:

• ValueError — If the input type is not supported.

generate_random_positions¶

generate_random_positions(total_length: int, set_number: int) → list[int]

Generate a list of unique random positions within the range of total_length.

Returns:

• list[int] — A list of unique random positions.

Raises:

• ValueError — If set_number is greater than total_length.

is_percentage¶

is_percentage(input_string: float) → bool

Check if the input string is a valid percentage between 1 and 100.

Returns:

• bool — True if the input string is a valid percentage, False otherwise.

string_to_upper_case_by_percentage¶

string_to_upper_case_by_percentage(percentage: float, prompt: str) → str

Convert a string by randomly capitalizing a percentage of its characters.

Returns:

• str — The converted string with randomly capitalized characters.

Raises:

• ValueError — If the percentage is not between 1 and 100.

RandomTranslationConverter¶

Bases: LLMGenericTextConverter, WordLevelConverter

Translates each individual word in a prompt to a random language using an LLM.

An existing PromptChatTarget is used to perform the translation (like Azure OpenAI).

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into the target format supported by the converter.

Returns:

• ConverterResult — The result containing the converted output and its type.

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

RangeSelectionStrategy¶

Bases: TextSelectionStrategy

Selects text based on proportional start and end positions.

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select a range based on proportional positions.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index).

RegexSelectionStrategy¶

Bases: TextSelectionStrategy

Selects text based on the first regex match.

Constructor Parameters:

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select the range of the first regex match.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index) of the first match, or (0, 0) if no match found.

RepeatTokenConverter¶

Bases: PromptConverter

Repeats a specified token a specified number of times in addition to a given prompt.

Based on: https://dropbox.tech/machine-learning/bye-bye-bye-evolution-of-repeated-token-attacks-on-chatgpt-models

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by repeating the specified token a specified number of times.

Returns:

• ConverterResult — The result containing the modified prompt with repeated tokens.

Raises:

• ValueError — If the input type is not supported.

ScientificTranslationConverter¶

Bases: LLMGenericTextConverter

Uses an LLM to transform simple or direct prompts into scientifically-framed versions using technical terminology, chemical notation, or academic phrasing. This can be useful for red-teaming scenarios to test whether safety filters can be bypassed through scientific translation.

Constructor Parameters:

SearchReplaceConverter¶

Bases: PromptConverter

Converts a string by replacing chosen phrase with a new phrase of choice.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by replacing the specified pattern with a random choice from the replacement list.

Returns:

• ConverterResult — The result containing the converted text.

Raises:

• ValueError — If the input type is not supported.

SelectiveTextConverter¶

Bases: PromptConverter

A wrapper converter that applies another converter to selected portions of text.

This converter supports multiple selection strategies:

• Character-level: Selects a contiguous character range (e.g., IndexSelectionStrategy, RegexSelectionStrategy)

• Word-level: Selects specific words (e.g., WordIndexSelectionStrategy, WordPositionSelectionStrategy)

• Token-based: Auto-detects and converts text between ⟪⟫ tokens (TokenSelectionStrategy)

Most use cases will use word-level strategies for more intuitive selection.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert selected portions of the prompt using the wrapped converter.

Returns:

• ConverterResult — The result containing the converted output and its type.

Raises:

• ValueError — If the input type is not “text”.

SneakyBitsSmugglerConverter¶

Bases: SmugglerConverter

Encodes and decodes text using a bit-level approach.

Constructor Parameters:

Methods:

decode_message¶

decode_message(message: str) → str

Decode the message encoded using Sneaky Bits mode.

The method filters out only the valid invisible characters (self.zero_char and self.one_char), groups them into 8-bit chunks, reconstructs each byte, and finally decodes the byte sequence using UTF-8.

Returns:

• str — The decoded original message.

encode_message¶

encode_message(message: str) → tuple[str, str]

Encode the message using Sneaky Bits mode.

The message is first converted to its UTF-8 byte sequence. Then each byte is represented as 8 bits, with each bit replaced by an invisible character (self.zero_char for 0 and self.one_char for 1).

Returns:

• str — Tuple[str, str]: A tuple where the first element is a bit summary (empty in this implementation)

• str — and the second element is the encoded message containing the invisible bits.

StringJoinConverter¶

Bases: WordLevelConverter

Converts text by joining its characters with the specified join value.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

SuffixAppendConverter¶

Bases: PromptConverter

Appends a specified suffix to the prompt. E.g. with a suffix !!!, it converts a prompt of test to test !!!.

See https://github.com/Azure/PyRIT/tree/main/pyrit/auxiliary_attacks/gcg for adversarial suffix generation.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by appending the specified suffix.

Returns:

• ConverterResult — The prompt with the suffix appended.

Raises:

• ValueError — If the input type is not supported.

SuperscriptConverter¶

Bases: WordLevelConverter

Converts text to superscript.

This converter leaves characters that do not have a superscript equivalent unchanged.

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

TemplateSegmentConverter¶

Bases: PromptConverter

Uses a template to randomly split a prompt into segments defined by the template.

This converter is a generalized version of this: https://adversa.ai/blog/universal-llm-jailbreak-chatgpt-gpt-4-bard-bing-anthropic-and-beyond/

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by splitting it into random segments and using them to fill the template parameters. The prompt is split into N segments (where N is the number of template parameters) at random word boundaries. Each segment is then used to fill the corresponding template parameter.

Returns:

• ConverterResult — The result containing the template filled with prompt segments.

Raises:

• ValueError — If the input type is not supported.

TenseConverter¶

Bases: LLMGenericTextConverter

Converts a conversation to a different tense using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

TextJailbreakConverter¶

Bases: PromptConverter

Uses a jailbreak template to create a prompt.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt using the jailbreak template.

Returns:

• ConverterResult — The result containing the converted output and its type.

Raises:

• ValueError — If the input type is not supported.

TextSelectionStrategy¶

Bases: abc.ABC

Base class for text selection strategies used by SelectiveTextConverter and WordLevelConverter. Defines how to select a region of text or words for conversion.

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Select a range of characters in the text to be converted.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index) representing the character range. The range is inclusive of start_index and exclusive of end_index.

TokenSelectionStrategy¶

Bases: TextSelectionStrategy

A special selection strategy that signals SelectiveTextConverter to auto-detect and convert text between start/end tokens (e.g., ⟪ and ⟫).

This strategy is used when chaining converters with preserve_tokens=True. Instead of programmatically selecting text, it relies on tokens already present in the text from a previous converter.

Methods:

select_range¶

select_range(text: str) → tuple[int, int]

Do not use this method for TokenSelectionStrategy. SelectiveTextConverter handles token detection separately.

Returns:

• tuple[int, int] — tuple[int, int]: Always returns (0, 0) as this strategy uses token detection instead.

ToneConverter¶

Bases: LLMGenericTextConverter

Converts a conversation to a different tone using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Constructor Parameters:

ToxicSentenceGeneratorConverter¶

Bases: LLMGenericTextConverter

Generates toxic sentence starters using an LLM.

An existing PromptChatTarget is used to perform the conversion (like Azure OpenAI).

Based on Project Moonshot’s attack module that generates toxic sentences to test LLM safety guardrails: attack-modules/toxic_sentence_generator.py

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into a toxic sentence starter.

Returns:

• ConverterResult — The conversion result containing the toxic sentence starter.

input_supported¶

input_supported(input_type: PromptDataType) → bool

Check if the input type is supported.

Returns:

• bool — True if the input type is supported, False otherwise.

output_supported¶

output_supported(output_type: PromptDataType) → bool

Check if the output type is supported.

Returns:

• bool — True if the output type is supported, False otherwise.

TranslationConverter¶

Bases: PromptConverter

Translates prompts into different languages using an LLM.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by translating it using the converter target.

Returns:

• ConverterResult — The result containing the generated version of the prompt.

Raises:

• ValueError — If the input type is not supported.

TransparencyAttackConverter¶

Bases: PromptConverter

Currently, only JPEG images are supported as input. Output images will always be saved as PNG with transparency.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'image_path') → ConverterResult

Convert the given prompt by blending an attack image (potentially harmful) with a benign image. Uses the Novel Image Blending Algorithm from McKee & Noever, 2024.

Returns:

• ConverterResult — The result containing path to the manipulated image with transparency.

Raises:

• ValueError — If the input type is not supported or if the prompt is invalid.

UnicodeConfusableConverter¶

Bases: PromptConverter

Applies substitutions to words in the prompt to test adversarial textual robustness by replacing characters with visually similar ones.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by applying confusable substitutions. This leads to a prompt that looks similar, but is actually different (e.g., replacing a Latin ‘a’ with a Cyrillic ‘а’).

Returns:

• ConverterResult — The result containing the prompt with confusable substitutions applied.

Raises:

• ValueError — If the input type is not supported.

UnicodeReplacementConverter¶

Bases: WordLevelConverter

Converts a prompt to its unicode representation.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

join_words¶

join_words(words: list[str]) → str

Join a list of words into a single string, optionally encoding spaces as unicode.

Returns:

• str — The joined string.

UnicodeSubstitutionConverter¶

Bases: PromptConverter

Encodes the prompt using any unicode starting point.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by encoding it using any unicode starting point. Default is to use invisible flag emoji characters.

Returns:

• ConverterResult — The result containing the converted output and its type.

Raises:

• ValueError — If the input type is not supported.

UrlConverter¶

Bases: PromptConverter

Converts a prompt to a URL-encoded string.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into a URL-encoded string.

Returns:

• ConverterResult — The result containing the URL-encoded prompt.

Raises:

• ValueError — If the input type is not supported.

VariationConverter¶

Bases: PromptConverter

Generates variations of the input prompts using the converter target.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by generating variations of it using the converter target.

Returns:

• ConverterResult — The result containing the generated variations.

Raises:

• ValueError — If the input type is not supported.

send_variation_prompt_async¶

send_variation_prompt_async(request: Message) → str

Send the message to the converter target and retrieve the response.

Returns:

• str — The response message from the converter target.

Raises:

• InvalidJsonException — If the response is not valid JSON or does not contain the expected keys.

VariationSelectorSmugglerConverter¶

Bases: SmugglerConverter

Extension: In addition to embedding into a base character, we also support appending invisible variation selectors directly to visible text—enabling mixed visible and hidden content within a single string.

Constructor Parameters:

Methods:

decode_message¶

decode_message(message: str) → str

Decode a message encoded using Unicode variation selectors. The decoder scans the string for variation selectors, ignoring any visible separator.

Returns:

• str — The decoded message.

decode_visible_hidden¶

decode_visible_hidden(combined: str) → tuple[str, str]

Extract the visible text and decodes the hidden text from a combined string.

It searches for the first occurrence of the base character (self.utf8_base_char) and treats everything from that point on as the hidden payload.

Returns:

• tuple[str, str] — Tuple[str, str]: A tuple with the visible text and the decoded hidden text.

encode_message¶

encode_message(message: str) → tuple[str, str]

Encode the message using Unicode variation selectors.

The message is converted to UTF-8 bytes, and each byte is mapped to a variation selector: - 0x00-0x0F => U+FE00 to U+FE0F. - 0x10-0xFF => U+E0100 to U+E01EF.

If embed_in_base is True, the payload is embedded directly into the base character; otherwise, a visible separator (a space) is inserted between the base and payload.

Returns:

• tuple[str, str] — Tuple[str, str]: A tuple containing a summary of the code points and the encoded string.

encode_visible_hidden¶

encode_visible_hidden(visible: str, hidden: str) → tuple[str, str]

Combine visible text with hidden text by encoding the hidden text using variation_selector_smuggler mode.

The hidden payload is generated as a composite using the current embedding setting and then appended to the visible text.

Returns:

• tuple[str, str] — Tuple[str, str]: A tuple containing a summary and the combined text.

WordDocConverter¶

Bases: PromptConverter

Convert a text prompt into a Word (.docx) document.

This converter supports two main modes:

1. New document generation If no existing document is provided, the converter creates a simple .docx containing the rendered prompt content in a single paragraph.

2. Placeholder-based injection into an existing document If an existing_docx is provided, the converter searches for a literal placeholder string (for example {{INJECTION_PLACEHOLDER}}) in the document’s paragraphs. When the placeholder is found fully inside a single run, it is replaced with the rendered prompt content while preserving the rest of the paragraph and its formatting.

   .. important:: Placeholders must be fully contained within a single run. If a placeholder spans multiple runs (for example due to mixed formatting), this converter will not replace it. This limitation is intentional to avoid collapsing mixed formatting or rewriting complex run structures.

Constructor Parameters:

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt into a Word document (.docx).

If prompt_template is provided, the prompt is first used to render the template via SeedPrompt.render_template_value. Otherwise, the raw prompt string is used as the content.

• When existing_docx is set, this content is injected into the document by replacing the configured placeholder string.

• When no existing_docx is provided, a new document with a single paragraph containing the content is created.

Returns:

• ConverterResult — Contains the path to the generated .docx file in

• ConverterResult — output_text and output_type="binary_path".

Raises:

• ValueError — If the input type is not supported.

WordIndexSelectionStrategy¶

Bases: WordSelectionStrategy

Selects words based on their indices in the word list.

Constructor Parameters:

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select words at the specified indices.

Returns:

• list[int] — List[int]: The list of valid indices.

Raises:

• ValueError — If any indices are out of range.

WordKeywordSelectionStrategy¶

Bases: WordSelectionStrategy

Selects words that match specific keywords.

Constructor Parameters:

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select words that match the keywords.

Returns:

• list[int] — List[int]: The list of indices where keywords were found.

WordPositionSelectionStrategy¶

Bases: WordSelectionStrategy

Selects words based on proportional start and end positions.

Constructor Parameters:

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select words based on the relative position.

Returns:

• list[int] — List[int]: The list of indices in the specified position range.

WordProportionSelectionStrategy¶

Bases: WordSelectionStrategy

Selects a random proportion of words.

Constructor Parameters:

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select a random proportion of words.

Returns:

• list[int] — List[int]: The list of randomly selected indices.

WordRegexSelectionStrategy¶

Bases: WordSelectionStrategy

Selects words that match a regex pattern.

Constructor Parameters:

Methods:

select_words¶

select_words(words: list[str]) → list[int]

Select words that match the regex pattern.

Returns:

• list[int] — List[int]: The list of indices where words matched the pattern.

WordSelectionStrategy¶

Bases: TextSelectionStrategy

Base class for word-level selection strategies.

Word selection strategies work by splitting text into words and selecting specific word indices. They provide a select_words() method and implement select_range() by converting word selections to character ranges.

Methods:

select_range¶

select_range(text: str, word_separator: str = ' ') → tuple[int, int]

Select a character range by first selecting words, then converting to character positions.

This implementation splits the text by word_separator, gets selected word indices, then calculates the character range that spans those words.

Returns:

• tuple[int, int] — tuple[int, int]: A tuple of (start_index, end_index) representing the character range that encompasses all selected words.

select_words¶

select_words(words: list[str]) → list[int]

Select word indices to be converted.

Returns:

• list[int] — List[int]: A list of indices representing which words should be converted.

ZalgoConverter¶

Bases: WordLevelConverter

Converts text into cursed Zalgo text using combining Unicode marks.

Constructor Parameters:

Methods:

convert_word_async¶

convert_word_async(word: str) → str

Convert a single word into the target format supported by the converter.

Returns:

• str — The converted word.

validate_input¶

validate_input(prompt: str) → None

Validate the input prompt before conversion.

ZeroWidthConverter¶

Bases: PromptConverter

Injects zero-width spaces between characters in the provided text to bypass content safety mechanisms.

Methods:

convert_async¶

convert_async(prompt: str, input_type: PromptDataType = 'text') → ConverterResult

Convert the given prompt by injecting zero-width spaces between each character.

Returns:

• ConverterResult — The result containing the modified prompt.

Raises:

• ValueError — If the input type is not supported.