FEAT: Converter Registry Initializer

.pyrit_conf_example

@@ -23,6 +23,7 @@ memory_db_type: sqlite
 #
 # Available initializers:
 # - target: Registers available prompt targets into the TargetRegistry
+# - converter: Registers the curated core converter presets
 # - scorer: Registers pre-configured scorers into the ScorerRegistry
 # - technique: Registers attack techniques into the AttackTechniqueRegistry
 # - load_default_datasets: Optionally preloads all registered datasets into memory
@@ -49,6 +50,7 @@ initializers:
       tags:
         - default
         - scorer
+  - name: converter
   - name: scorer
   - name: technique
   # Optional full preload/cache warming for offline or shared environments.

MANIFEST.in

@@ -5,6 +5,7 @@ recursive-include pyrit *.prompt
 recursive-include pyrit *.yaml
 recursive-include pyrit *.pdf
 recursive-include pyrit *.png
+recursive-include pyrit *.jpg
 recursive-include pyrit *.wav
 recursive-include pyrit *.mp4
 recursive-include pyrit *.md

doc/getting_started/pyrit_conf.md

@@ -103,9 +103,19 @@ Most users should enable the following initializers. These are what the `.pyrit_
 | Initializer | What It Registers | When You Need It |
 | --- | --- | --- |
 | `target` | Prompt targets (OpenAI, Azure, AML, etc.) into the `TargetRegistry` | Recommended for `pyrit_scan` and registry-based workflows |
+| `converter` | Curated core converter presets | Recommended when attaching registered converters to techniques |
 | `scorer` | Scorers (refusal, content safety, harm-category, Likert, etc.) into the `ScorerRegistry` | Recommended for automated scoring and `pyrit_scan` evaluations |
 | `technique` | Attack techniques into the `AttackTechniqueRegistry` | Recommended for scenarios that select registered techniques |
 
+The `converter` initializer registers a curated set of text, LLM, audio, and image converter presets. The
+LLM-backed presets use the registered `adversarial_chat` target and are skipped when that target is not available.
+The Azure Speech preset uses the standard Azure Speech environment variables and is skipped when they are not set.
+Use `pyrit_scan list-converters` to list the presets that were registered.
+
+Parameterized presets include past and future tense, professional and sarcastic tone, Spanish translation,
+whitespace replacement with underscores, the `jailbreak_1.yaml` text jailbreak template, the packaged blank
+canvas for text-to-image conversion, and the packaged benign cake image for the transparency attack.
+
 ```{note}
 **Execution order follows listing order.** Initializers execute in the order they appear in the config. Ensure dependencies are satisfied — for example, list `target` before `scorer` since scorers need targets to be registered first.
 ```
@@ -123,6 +133,7 @@ initializers:
       tags:
         - default
         - scorer
+  - name: converter
   - name: scorer
   - name: technique
 ```
@@ -404,6 +415,7 @@ initializers:
       tags:
         - default
         - scorer
+  - name: converter
   - name: scorer
   - name: technique
   # Optional full preload/cache warming; scenarios fetch requested datasets on demand.

frontend/src/components/Configuration/Configuration.test.tsx

@@ -26,6 +26,9 @@ jest.mock('@/services/api', () => ({
 const mockedConfigurationApi = jest.mocked(configurationApi)
 const mockedInitializersApi = jest.mocked(initializersApi)
 
+// Fluent UI dialogs can render slowly in JSDOM under full test load.
+jest.setTimeout(60_000)
+
 function renderPage(): void {
   render(
     <FluentProvider theme={webLightTheme}>
@@ -204,8 +207,17 @@ describe('Configuration', () => {
       { selector: 'label' },
     )).toBeInTheDocument()
     await user.click(screen.getByRole('button', { name: 'Add initializer' }))
-    const dialog = screen.getByRole('dialog', { name: 'Add custom initializer' })
-    await user.type(within(dialog).getByRole('textbox', { name: /Initializer name/ }), 'new_custom')
+    const dialog = await screen.findByRole(
+      'dialog',
+      { name: 'Add custom initializer' },
+      { timeout: 15_000 },
+    )
+    const nameInput = await within(dialog).findByRole(
+      'textbox',
+      { name: /Initializer name/ },
+      { timeout: 15_000 },
+    )
+    await user.type(nameInput, 'new_custom')
     fireEvent.change(within(dialog).getByRole('textbox', { name: 'Python source' }), {
       target: { value: 'class NewCustom: pass' },
     })

pyrit/setup/initializers/__init__.py

@@ -10,6 +10,7 @@
 
 if TYPE_CHECKING:
     from pyrit.models.parameter import Parameter
+    from pyrit.setup.initializers.converters import ConverterInitializer
     from pyrit.setup.initializers.load_default_datasets import LoadDefaultDatasets
     from pyrit.setup.initializers.preload_scenario_metadata import PreloadScenarioMetadata
     from pyrit.setup.initializers.refresh_datasets import RefreshDatasets
@@ -21,6 +22,7 @@
 _LAZY_EXPORTS: dict[str, str | tuple[str, str | None]] = {
     "Parameter": "pyrit.models.parameter",
     "PyRITInitializer": "pyrit.setup.pyrit_initializer",
+    "ConverterInitializer": "pyrit.setup.initializers.converters",
     "TechniqueInitializer": "pyrit.setup.initializers.techniques",
     "ScorerInitializer": "pyrit.setup.initializers.scorers",
     "TargetInitializer": "pyrit.setup.initializers.targets",

pyrit/setup/initializers/converters.py

@@ -0,0 +1,187 @@
+# Copyright (c) Microsoft Corporation.
+# Licensed under the MIT license.
+
+"""Initializer for registering the core converter presets."""
+
+import asyncio
+import logging
+from collections.abc import Callable
+from dataclasses import dataclass, field
+from typing import TYPE_CHECKING, Any, ClassVar
+
+from pyrit.common.path import DATASETS_PATH
+from pyrit.registry import ConverterRegistry
+from pyrit.setup.pyrit_initializer import PyRITInitializer
+
+if TYPE_CHECKING:
+    from pyrit.converter import Converter
+
+logger = logging.getLogger(__name__)
+
+
+def _get_text_jailbreak_args() -> dict[str, Any]:
+    """
+    Build the constructor arguments for the text jailbreak preset.
+
+    Returns:
+        dict[str, Any]: Constructor arguments containing the jailbreak template.
+    """
+    from pyrit.datasets import TextJailBreak
+
+    return {"jailbreak_template": TextJailBreak(template_file_name="jailbreak_1.yaml")}
+
+
+@dataclass(frozen=True)
+class ConverterConfig:
+    """Configuration for a converter preset."""
+
+    registry_name: str
+    converter_type: str
+    constructor_args: dict[str, Any] = field(default_factory=dict)
+    constructor_args_factory: Callable[[], dict[str, Any]] | None = None
+
+    def get_constructor_args(self) -> dict[str, Any]:
+        """
+        Get a new constructor argument dictionary for this preset.
+
+        Returns:
+            dict[str, Any]: A copy of the configured arguments with deferred arguments included.
+        """
+        args = dict(self.constructor_args)
+        if self.constructor_args_factory:
+            args.update(self.constructor_args_factory())
+        return args
+
+
+class ConverterInitializer(PyRITInitializer):
+    """
+    Register the curated core converter presets into the ConverterRegistry.
+
+    Each preset explicitly declares its constructor arguments and target references.
+    Custom initializers can use the same pattern to register additional or parameterized
+    converter presets.
+    """
+
+    CONFIGS: ClassVar[tuple[ConverterConfig, ...]] = (
+        ConverterConfig(registry_name="base64", converter_type="Base64Converter"),
+        ConverterConfig(registry_name="binary", converter_type="BinaryConverter"),
+        ConverterConfig(registry_name="char_swap", converter_type="CharSwapConverter"),
+        ConverterConfig(registry_name="ecoji", converter_type="EcojiConverter"),
+        ConverterConfig(registry_name="insert_punctuation", converter_type="InsertPunctuationConverter"),
+        ConverterConfig(registry_name="leetspeak", converter_type="LeetspeakConverter"),
+        ConverterConfig(registry_name="rot13", converter_type="ROT13Converter"),
+        ConverterConfig(
+            registry_name="search_replace",
+            converter_type="SearchReplaceConverter",
+            constructor_args={"pattern": r"\s+", "replace": "_"},
+        ),
+        ConverterConfig(registry_name="string_join", converter_type="StringJoinConverter"),
+        ConverterConfig(
+            registry_name="text_jailbreak",
+            converter_type="TextJailbreakConverter",
+            constructor_args_factory=_get_text_jailbreak_args,
+        ),
+        ConverterConfig(registry_name="zalgo", converter_type="ZalgoConverter"),
+        ConverterConfig(
+            registry_name="malicious_question_generator",
+            converter_type="MaliciousQuestionGeneratorConverter",
+            constructor_args={"converter_target": "adversarial_chat"},
+        ),
+        ConverterConfig(
+            registry_name="math_prompt",
+            converter_type="MathPromptConverter",
+            constructor_args={"converter_target": "adversarial_chat"},
+        ),
+        ConverterConfig(
+            registry_name="noise",
+            converter_type="NoiseConverter",
+            constructor_args={"converter_target": "adversarial_chat"},
+        ),
+        ConverterConfig(
+            registry_name="tense_future",
+            converter_type="TenseConverter",
+            constructor_args={"converter_target": "adversarial_chat", "tense": "future"},
+        ),
+        ConverterConfig(
+            registry_name="tense_past",
+            converter_type="TenseConverter",
+            constructor_args={"converter_target": "adversarial_chat", "tense": "past"},
+        ),
+        ConverterConfig(
+            registry_name="tone_professional",
+            converter_type="ToneConverter",
+            constructor_args={"converter_target": "adversarial_chat", "tone": "professional"},
+        ),
+        ConverterConfig(
+            registry_name="tone_sarcastic",
+            converter_type="ToneConverter",
+            constructor_args={"converter_target": "adversarial_chat", "tone": "sarcastic"},
+        ),
+        ConverterConfig(
+            registry_name="translation_spanish",
+            converter_type="TranslationConverter",
+            constructor_args={"converter_target": "adversarial_chat", "language": "Spanish"},
+        ),
+        ConverterConfig(
+            registry_name="variation",
+            converter_type="VariationConverter",
+            constructor_args={"converter_target": "adversarial_chat"},
+        ),
+        ConverterConfig(
+            registry_name="add_image_text",
+            converter_type="AddImageTextConverter",
+            constructor_args={
+                "img_to_add": str(DATASETS_PATH / "seed_datasets" / "local" / "examples" / "blank_canvas.png")
+            },
+        ),
+        ConverterConfig(
+            registry_name="add_text_image",
+            converter_type="AddTextImageConverter",
+            constructor_args={"text_to_add": "PyRIT"},
+        ),
+        ConverterConfig(
+            registry_name="azure_speech_audio_to_text",
+            converter_type="AzureSpeechAudioToTextConverter",
+        ),
+        ConverterConfig(registry_name="image_color_saturation", converter_type="ImageColorSaturationConverter"),
+        ConverterConfig(registry_name="image_compression", converter_type="ImageCompressionConverter"),
+        ConverterConfig(registry_name="image_rotation", converter_type="ImageRotationConverter"),
+        ConverterConfig(registry_name="qr_code", converter_type="QRCodeConverter"),
+        ConverterConfig(
+            registry_name="transparency_attack",
+            converter_type="TransparencyAttackConverter",
+            constructor_args={
+                "benign_image_path": DATASETS_PATH / "seed_datasets" / "local" / "examples" / "benign_cake_question.jpg"
+            },
+        ),
+    )
+
+    async def initialize_async(self) -> None:
+        """Create and register the core converter presets."""
+        converter_registry = ConverterRegistry.get_registry_singleton()
+
+        for config in self.CONFIGS:
+            try:
+                converter = await asyncio.to_thread(
+                    self._create_converter,
+                    converter_registry=converter_registry,
+                    config=config,
+                )
+                converter_registry.instances.register(converter, name=config.registry_name)
+                logger.info("Registered converter: %s", config.registry_name)
+            except (FileNotFoundError, KeyError, TypeError, ValueError) as ex:
+                logger.warning("Skipping converter '%s': %s", config.registry_name, ex)
+
+    def _create_converter(
+        self,
+        *,
+        converter_registry: ConverterRegistry,
+        config: ConverterConfig,
+    ) -> "Converter":
+        """
+        Create one configured converter outside the event-loop thread.
+
+        Returns:
+            Converter: The configured converter instance.
+        """
+        return converter_registry.create_instance(config.converter_type, **config.get_constructor_args())

tests/unit/setup/test_configuration_loader.py

@@ -272,7 +272,7 @@ def test_example_uses_on_demand_dataset_fetching(self) -> None:
         initializer_names = [initializer.name for initializer in config._initializer_configs]
         example_text = example_path.read_text(encoding="utf-8")
 
-        assert initializer_names == ["target", "scorer", "technique"]
+        assert initializer_names == ["target", "converter", "scorer", "technique"]
         assert "# - name: load_default_datasets" in example_text
         assert "several minutes" in example_text
         assert "provider credentials" in example_text