FEAT Add WordDocConverter by mhadica · Pull Request #1368 · Azure/PyRIT

mhadica · 2026-02-12T18:33:50Z

Description

Implemented a Word document file converter to support indirect prompt injection scenarios, addressing FEAT File Converter: Word Document (#424).

Feature: WordDoc_Converter
Added WordDoc_Converter in pyrit.prompt_converter.word_doc_converter as a PromptConverter that transforms text prompts into .docx files.
Exposed via pyrit.prompt_converter.init for discovery through get_converter_modalities.

Supported Modes
New Document Generation
Creates a simple .docx with optionally templated prompt content in a single paragraph.
Placeholder-Based Injection
Replaces a literal placeholder in an existing .docx (within paragraph runs) with rendered content while preserving formatting.

Templating and Safety
Uses SeedPrompt.render_template_value for templating.
Does not render Jinja2 templates from arbitrary .docx content.
Injected content is plain text output from SeedPrompt.

Formatting Constraints
Placeholders must be fully contained within a single run.
If spanning multiple runs, replacement is skipped and a warning is logged to avoid formatting issues.

Tests
Added unit tests for:
Creating a new .docx from a plain text prompt.
Injecting content when the placeholder is within a single run.
Verifying no replacement and warning when placeholders span multiple runs.

Documentation
Updated documentation to describe:
WordDoc_Converter
Supported modes
Placeholder requirements
Use of SeedPrompt for templating

mhadica · 2026-02-12T18:37:41Z

@microsoft-github-policy-service agree

Copilot

Pull request overview

Adds a new prompt converter that outputs Word .docx documents to support indirect prompt-injection scenarios, and wires it into the prompt-converter discovery surface.

Changes:

Introduces word_doc_converter.py implementing .docx generation and placeholder-based injection for existing documents.
Exposes the new converter via pyrit.prompt_converter.__init__ so it participates in converter discovery.
Adds python-docx as a project dependency.

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 8 comments.

File	Description
`pyrit/prompt_converter/word_doc_converter.py`	New Word `.docx` converter with new-doc generation and placeholder injection logic, plus identifier + serialization.
`pyrit/prompt_converter/__init__.py`	Exports the new converter for discovery via `__all__` / `get_converter_modalities`.
`pyproject.toml`	Adds `python-docx` runtime dependency needed by the converter.

Copilot · 2026-02-16T12:33:37Z

pyrit/prompt_converter/word_doc_converter.py

+    async def _serialize_docx(self, docx_bytes: bytes) -> DataTypeSerializer:
+        """Serialize the generated document using a data serializer."""
+        extension = "docx"
+
+        serializer = data_serializer_factory(
+            category="prompt-memory-entries",
+            data_type="binary_path",
+            extension=extension,
+        )
+        await serializer.save_data(docx_bytes)
+        return serializer


_serialize_docx is an async method but its name does not end with _async. Rename it (and its call site) to follow the project convention for async method naming.

Copilot · 2026-02-16T12:33:37Z

pyrit/prompt_converter/word_doc_converter.py

+        The placeholder must appear fully inside a single run; if it only exists
+        across multiple runs, it will not be replaced.
+        """
+        assert self._injection_config.existing_docx is not None


Avoid using assert for required runtime checks. In optimized runs (python -O) assertions are stripped, which could allow existing_docx to be None here and lead to a less clear failure. Prefer an explicit check with a clear exception (or rely on the earlier constructor validation and remove the assertion entirely).

Suggested change

assert self._injection_config.existing_docx is not None

if self._injection_config.existing_docx is None:

raise ValueError(

"Cannot inject into an existing Word document because no 'existing_docx' "

"path was provided in the injection configuration."

)

Copilot · 2026-02-16T12:33:38Z

pyrit/prompt_converter/word_doc_converter.py

+    async def convert_async(self, *, 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.
+
+        Args:
+            prompt: The prompt or dynamic data used to generate the content.
+            input_type: The type of input data. Must be ``"text"``.
+
+        Returns:
+            ConverterResult: Contains the path to the generated `.docx` file in
+            ``output_text`` and ``output_type="binary_path"``.
+
+        Raises:
+            ValueError: If the input type is not supported.
+        """
+        if not self.input_supported(input_type):
+            raise ValueError("Input type not supported")
+
+        content = self._prepare_content(prompt)
+
+        if self._injection_config.existing_docx:
+            doc_bytes = self._inject_into_existing_docx(content)
+        else:
+            doc_bytes = self._generate_new_docx(content)
+
+        serializer = await self._serialize_docx(doc_bytes)
+
+        return ConverterResult(output_text=serializer.value, output_type="binary_path")


This PR introduces a new converter with non-trivial behavior (new doc generation + placeholder injection across runs), but there are no corresponding unit tests under tests/unit/converter/ (the repo has extensive converter test coverage, e.g., test_pdf_converter.py). Add unit tests for the main success and failure paths to prevent regressions.

Copilot · 2026-02-16T12:33:38Z

pyrit/prompt_converter/word_doc_converter.py

+                final content before injection. If provided, ``prompt`` should
+                be a dict-like object (or string representation) whose keys map
+                to the template parameters.


The docstring for prompt_template mentions that prompt can be a “dict-like object”, but PromptConverter.convert_async is typed/documented to accept prompt: str and other converters treat the templated case as a string that can be parsed into a dict. Consider updating this documentation to avoid implying that non-string prompts are supported by the converter interface.

Suggested change

final content before injection. If provided, ``prompt`` should

be a dict-like object (or string representation) whose keys map

to the template parameters.

final content before injection. If provided, ``prompt`` passed

to ``convert_async`` must be a string whose contents can be

interpreted as the template parameters (for example, a

JSON-encoded or other parseable mapping of keys to values).

Copilot · 2026-02-16T12:33:38Z

pyrit/prompt_converter/word_doc_converter.py

+    """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.
+
+    Security note:
+        This converter does **not** render Jinja2 templates from arbitrary
+        ``.docx`` content. Templating is handled via ``SeedPrompt`` (if provided),
+        and only the already-rendered text is injected into the document. This
+        avoids executing untrusted Jinja2 templates from document bodies.
+    """


PR description mentions updated docs and added unit tests for this converter, but the changeset here only adds the converter, an __init__ export, and a dependency. If docs/tests are expected, they appear to be missing from this PR.

Copilot · 2026-02-16T12:33:38Z

pyrit/prompt_converter/__init__.py

+from pyrit.prompt_converter.word_doc_converter import WordDoc_Converter
 from pyrit.prompt_converter.zalgo_converter import ZalgoConverter


WordDoc_Converter export/import name likely needs to change if the converter class is renamed to match existing converter naming conventions (no underscores in class names).

Copilot · 2026-02-16T12:33:39Z

pyproject.toml

    "python-dotenv>=1.0.1",
    "pypdf>=6.6.2",
    "reportlab>=4.4.4",
+    "python-docx>=1.1.0",


Keep dependencies roughly sorted for maintainability; python-docx is currently placed after reportlab (R) but should be grouped with the other py* / python-* dependencies (near python-dotenv / pypdf).

Suggested change

"python-dotenv>=1.0.1",

"pypdf>=6.6.2",

"reportlab>=4.4.4",

"python-docx>=1.1.0",

"pypdf>=6.6.2",

"python-docx>=1.1.0",

"python-dotenv>=1.0.1",

"reportlab>=4.4.4",

Copilot · 2026-02-16T12:33:39Z

pyrit/prompt_converter/word_doc_converter.py

+    placeholder: str
+
+
+class WordDoc_Converter(PromptConverter):


The class name WordDoc_Converter is inconsistent with the naming pattern used by other converters (e.g., PDFConverter, UrlConverter) and will surface in discovery APIs (e.g., get_converter_modalities). Consider renaming to WordDocConverter and updating the corresponding import/export in pyrit.prompt_converter.__init__.

Suggested change

class WordDoc_Converter(PromptConverter):

class WordDocConverter(PromptConverter):

FEAT Add WordDocConverter

f6d0a59

Merge branch 'main' into main

4edfced

romanlutz requested a review from Copilot February 16, 2026 12:27

romanlutz self-assigned this Feb 16, 2026

Copilot started reviewing on behalf of romanlutz February 16, 2026 12:27 View session

Copilot AI reviewed Feb 16, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

FEAT Add WordDocConverter#1368

FEAT Add WordDocConverter#1368
mhadica wants to merge 2 commits intoAzure:mainfrom
mhadica:main

mhadica commented Feb 12, 2026

Uh oh!

mhadica commented Feb 12, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Copilot AI Feb 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

-        assert self._injection_config.existing_docx is not None
+        if self._injection_config.existing_docx is None:
+            raise ValueError(
+                "Cannot inject into an existing Word document because no 'existing_docx' "
+                "path was provided in the injection configuration."
+            )

-                final content before injection. If provided, ``prompt`` should
-                be a dict-like object (or string representation) whose keys map
-                to the template parameters.
+                final content before injection. If provided, ``prompt`` passed
+                to ``convert_async`` must be a string whose contents can be
+                interpreted as the template parameters (for example, a
+                JSON-encoded or other parseable mapping of keys to values).

		from pyrit.prompt_converter.word_doc_converter import WordDoc_Converter
		from pyrit.prompt_converter.zalgo_converter import ZalgoConverter

	class WordDoc_Converter(PromptConverter):
	class WordDocConverter(PromptConverter):

Conversation

mhadica commented Feb 12, 2026

Uh oh!

mhadica commented Feb 12, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants