Fix Dependabot Python vulnerabilities (#77)

Author: davida-ps

.github/workflows/ci.yml

@@ -12,7 +12,7 @@ jobs:
     runs-on: ubuntu-latest
     strategy:
       matrix:
-        python-version: ['3.9', '3.10', '3.11']
+        python-version: ['3.10', '3.11']
 
     steps:
       - name: Checkout repository
@@ -49,3 +49,11 @@ jobs:
         run: |
           chmod +x ./build_package.sh
           ./build_package.sh
+
+      - name: Verify built wheel
+        run: |
+          python -m venv .package-test-venv
+          .package-test-venv/bin/python -m pip install --upgrade pip
+          .package-test-venv/bin/python -m pip install dist/*.whl
+          .package-test-venv/bin/python -m pip check
+          .package-test-venv/bin/prompt-security-fuzzer --list-providers

CHANGELOG.md

@@ -4,6 +4,15 @@ All notable changes to this project will be documented in this file.
 
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
+## [Unreleased]
+
+### Security
+- Upgraded LangChain, LangChain Core, python-dotenv, and related provider integrations to resolve current Dependabot advisories.
+
+### Changed
+- Minimum Python version raised from 3.9 to 3.10.
+- Updated fastparquet for NumPy 2 compatibility in the upgraded runtime dependency graph.
+
 ## [2.1.0] - 2026-02-16
 
 ### Added

CONTRIBUTING.md

@@ -13,7 +13,7 @@ By participating in this project, you agree to abide by its terms.
 ### Prerequisites
 
 Before you begin, ensure you have the following installed:
-- Python 3.7 or later
+- Python 3.10 or later
 - Git
 
 ### Setting Up Your Development Environment
@@ -29,8 +29,12 @@ cd ps-fuzz
 ### Set up a virtual environment
 
 ```bash
-python -m venv venv
+# Unix or macOS
+python3.10 -m venv venv
 source venv/bin/activate  # On Unix or macOS
+
+# Windows
+py -3.10 -m venv venv
 venv\Scripts\activate     # On Windows
 ```
 
@@ -128,7 +132,7 @@ from ..attack_config import AttackConfig
 from ..test_base import TestBase, StatusUpdate
 from ..attack_registry import register_test
 from typing import Generator
-from pkg_resources import resource_filename
+from .utils import packaged_attack_data_path
 import logging
 logger = logging.getLogger(__name__)
 ```
@@ -208,4 +212,3 @@ Here are some helpful resources to get you started with best practices for contr
 - [GitHub Flow](https://guides.github.com/introduction/flow/) - An introduction to the GitHub workflow, which explains branches, pull requests, and more.
 - [Writing Good Commit Messages](https://chris.beams.io/posts/git-commit/) - A guide on how to write clear and concise commit messages, which are crucial for following the changes in a project.
 - [Python Coding Style](https://pep8.org/) - Guidelines for writing clean and understandable Python code.
-

README.md

@@ -83,6 +83,7 @@ Table of Contents
 1. Install the Fuzzer package
    <a id="using-pip"></a>
    #### Using pip install
+   Requires Python 3.10 or newer.
    ```zsh
    pip install prompt-security-fuzzer
    ```
@@ -327,4 +328,3 @@ Turn this into a community project! We want this to be useful to everyone buildi
 Interested in contributing to the development of our tools? Great! For a guide on making your first contribution, please see our [Contributing Guide](https://github.com/prompt-security/ps-fuzz/blob/main/CONTRIBUTING.md#get-started-with-your-first-contribution-adding-a-new-test). This section offers a straightforward introduction to adding new tests.
 
 For ideas on what tests to add, check out the issues tab in our GitHub repository. Look for issues labeled `new-test` and `good-first-issue`, which are perfect starting points for new contributors.
-

claude.md

@@ -2,7 +2,7 @@
 
 ## Python Environment
 
-This project requires Python >= 3.9 (tested with 3.9, 3.10, 3.11).
+This project requires Python >= 3.10 (tested with 3.10, 3.11).
 
 ### Setup with uv
 

ps_fuzz/attacks/base64_injection.py

@@ -4,10 +4,9 @@
 from ..test_base import TestBase, StatusUpdate
 from ..attack_registry import register_test
 from ..util import contains_refusal_keywords
-# from .utils import evaluate_response
+from .utils import packaged_attack_data_path
 from typing import Generator
 import pandas as pd
-from pkg_resources import resource_filename # for loading attack data file packaged with the library
 import logging
 import base64
 import sys
@@ -24,8 +23,8 @@ def __init__(self, client_config: ClientConfig, attack_config: AttackConfig):
         )
 
     def run(self)  -> Generator[StatusUpdate, None, None]:
-        dataset_filename = resource_filename('ps_fuzz', 'attack_data/prompt_injections_for_base64.parquet')
-        data = pd.read_parquet(dataset_filename, engine='fastparquet')
+        with packaged_attack_data_path('prompt_injections_for_base64.parquet') as dataset_filename:
+            data = pd.read_parquet(dataset_filename, engine='fastparquet')
         rows_count = data.shape[0]
 
         if rows_count > self.attack_config.attack_prompts_count:

ps_fuzz/attacks/custom_benchmark.py

@@ -8,7 +8,6 @@
 from typing import Generator
 import pandas as pd
 import os
-from pkg_resources import resource_filename # for loading attack data file packaged with the library
 import logging
 logger = logging.getLogger(__name__)
 

ps_fuzz/attacks/harmful_behavior.py

@@ -4,10 +4,9 @@
 from ..test_base import TestBase, StatusUpdate
 from ..attack_registry import register_test
 # from ..util import contains_refusal_keywords
-from .utils import evaluate_response
+from .utils import evaluate_response, packaged_attack_data_path
 from typing import Generator
 import pandas as pd
-from pkg_resources import resource_filename # for loading attack data file packaged with the library
 import logging
 import sys
 logger = logging.getLogger(__name__)
@@ -23,9 +22,9 @@ def __init__(self, client_config: ClientConfig, attack_config: AttackConfig):
         )
 
     def run(self) -> Generator[StatusUpdate, None, None]:
-        dataset_filename = resource_filename('ps_fuzz', 'attack_data/harmful_behavior.csv')
-        logger.info(f"Dataset filename: {dataset_filename}")
-        data = pd.read_csv(dataset_filename)
+        with packaged_attack_data_path('harmful_behavior.csv') as dataset_filename:
+            logger.info(f"Dataset filename: {dataset_filename}")
+            data = pd.read_csv(dataset_filename)
         rows_count = data.shape[0]
 
         if rows_count > self.attack_config.attack_prompts_count:

ps_fuzz/attacks/rag_poisoning.py

@@ -33,45 +33,54 @@ def _suppress_loggers(logger_names):
             logging.getLogger(name).setLevel(lvl)
 
 # Check dependencies availability
-DEPENDENCIES_AVAILABLE = True
 MISSING_PACKAGES = []
 
 suppress_names = []
 try:
-    from langchain_community.vectorstores import Chroma
+    from langchain_chroma import Chroma
     suppress_names = ["chromadb"]
-
 except ImportError:
-    DEPENDENCIES_AVAILABLE = False
     MISSING_PACKAGES.append("chromadb")
+    Chroma = None
+
+try:
+    from langchain_openai import OpenAIEmbeddings
+except ImportError:
+    MISSING_PACKAGES.append("langchain-openai")
+    OpenAIEmbeddings = None
 
 try:
-    from langchain_community.embeddings import OpenAIEmbeddings, OllamaEmbeddings
+    from langchain_ollama import OllamaEmbeddings
 except ImportError:
-    DEPENDENCIES_AVAILABLE = False
-    MISSING_PACKAGES.append("langchain-community (embeddings)")
+    MISSING_PACKAGES.append("langchain-ollama")
+    OllamaEmbeddings = None
 
 try:
     from langchain_core.documents import Document
 except ImportError:
-    DEPENDENCIES_AVAILABLE = False
-    MISSING_PACKAGES.append("langchain (schema)")
+    MISSING_PACKAGES.append("langchain-core")
+    Document = None
+
+DEPENDENCIES_AVAILABLE = Chroma is not None and Document is not None
 
 # Create dummy classes for when dependencies are not available
-if not DEPENDENCIES_AVAILABLE:
+if Document is None:
     class Document:
         def __init__(self, page_content, metadata):
             self.page_content = page_content
             self.metadata = metadata
-    
+
+if Chroma is None:
     class Chroma:
         def __init__(self, *args, **kwargs):
             pass
-    
+
+if OpenAIEmbeddings is None:
     class OpenAIEmbeddings:
         def __init__(self, *args, **kwargs):
             pass
-    
+
+if OllamaEmbeddings is None:
     class OllamaEmbeddings:
         def __init__(self, *args, **kwargs):
             pass
@@ -126,6 +135,8 @@ def _get_embedding_function(self):
             model = config.embedding_model
 
         if provider == 'open_ai':
+            if 'langchain-openai' in MISSING_PACKAGES:
+                raise ImportError("OpenAI embeddings require the langchain-openai package")
             # Get embedding-specific base URL if configured
             base_url = None
             if isinstance(config, dict):
@@ -141,6 +152,8 @@ def _get_embedding_function(self):
             return OpenAIEmbeddings(**kwargs)
 
         elif provider == 'ollama':
+            if 'langchain-ollama' in MISSING_PACKAGES:
+                raise ImportError("Ollama embeddings require the langchain-ollama package")
             # Get embedding-specific base URL if configured
             base_url = None
             if isinstance(config, dict):
@@ -255,11 +268,16 @@ def _setup_poisoned_vector_database(self):
         with _suppress_loggers(suppress_names):
             self.vectorstore.add_documents([poisoned_doc])
 
-        # Persist the database (ChromaDB 0.4.0+ auto-persists with persist_directory)
-        try:
-            self.vectorstore.persist()
-        except AttributeError:
-            pass  # ChromaDB 0.4.0+ auto-persists when using persist_directory
+        # Chroma persists automatically when persist_directory is configured.
+
+    def _retrieve_relevant_documents(self, query: str):
+        """Use the Runnable API required by modern LangChain retrievers."""
+        retriever = self.vectorstore.as_retriever(
+            search_type="similarity",
+            search_kwargs={"k": 3}
+        )
+        with _suppress_loggers(suppress_names):
+            return retriever.invoke(query)
 
     def _cleanup(self):
         """Clean up temporary resources"""
@@ -275,7 +293,7 @@ def run(self) -> Generator[StatusUpdate, None, None]:
         # Preflight check: Skip if dependencies not available
         if not DEPENDENCIES_AVAILABLE:
             missing_packages_str = ", ".join(MISSING_PACKAGES) if MISSING_PACKAGES else "chromadb"
-            error_msg = f"Required dependencies not available: {missing_packages_str}. Install with: pip install chromadb (or uv pip install chromadb) or install with RAG extras: pip install .[rag]"
+            error_msg = f"Required dependencies not available: {missing_packages_str}. Reinstall prompt-security-fuzzer with its default dependencies."
             logger.warning(f"RAG poisoning attack skipped: {error_msg}")
             self.status.report_skipped("", error_msg)
             yield StatusUpdate(self.client_config, self.test_name, self.status, "Skipped", 1, 1)
@@ -311,13 +329,7 @@ def run(self) -> Generator[StatusUpdate, None, None]:
 
                 try:
                     # Retrieve relevant documents
-                    retriever = self.vectorstore.as_retriever(
-                        search_type="similarity",
-                        search_kwargs={"k": 3}
-                    )
-                    # suppress telemetry during retrieval
-                    with _suppress_loggers(suppress_names):
-                        relevant_docs = retriever.get_relevant_documents(query)
+                    relevant_docs = self._retrieve_relevant_documents(query)
 
                     # Create context from retrieved documents
                     context = "\n\n".join([doc.page_content for doc in relevant_docs])
@@ -394,4 +406,4 @@ def run(self) -> Generator[StatusUpdate, None, None]:
                 yield StatusUpdate(self.client_config, self.test_name, self.status, "Error", 1, 1)
         finally:
             # Always cleanup
-            self._cleanup()
+            self._cleanup()

ps_fuzz/attacks/utils.py

@@ -1,4 +1,19 @@
+from contextlib import contextmanager
+from importlib.resources import as_file, files
+from pathlib import Path
+from typing import Iterator
+
 from ..chat_clients import ChatSession
+
+
+@contextmanager
+def packaged_attack_data_path(filename: str) -> Iterator[Path]:
+    """Yield a filesystem path for packaged attack data on every installer type."""
+    resource = files('ps_fuzz').joinpath('attack_data', filename)
+    with as_file(resource) as path:
+        yield path
+
+
 def evaluate_response(client,response,expected_response):
     """
     Evaluate the response from the model against the expected response.