feat(Bump version):

build(deps-dev): bump mkdocstrings from 0.27.0 to 0.28.1

.dockerignore

@@ -1,2 +1,45 @@
-.git/
+# Byte-compiled / optimized / DLL files
 __pycache__/
+*.py[cod]
+
+# Distribution / packaging
+build/
+dist/
+*.egg-info/
+
+# Virtual environments
+
+.venv/
+env/
+ENV/
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.cache
+nosetests.xml
+coverage.xml
+
+# PyInstaller
+*.spec
+
+# macOS specific files
+.DS_Store
+
+# Windows specific files
+Thumbs.db
+desktop.ini
+
+# Tools and editors
+.idea/
+.vscode/
+cmder/
+
+# Output directories
+Output/
+te/

.github/workflows/pre-commit.yml

@@ -0,0 +1,21 @@
+name: Pre-Commit Checks
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  pre-commit:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - name: Set up Python
+        uses: actions/setup-python@v4
+        with:
+          python-version: '3.11'
+      - name: Install pre-commit
+        run: pip install pre-commit
+      - name: Run pre-commit
+        run: pre-commit run --all-files

.gitignore

@@ -8,3 +8,11 @@ runs/
 logs/
 modal_agent.py
 sandbox.py
+site/
+agesec.toml
+.clinerules
+garak_rest.json
+2025.*.json
+inv/
+scripts/
+docx/

.pre-commit-config.yaml

@@ -43,6 +43,10 @@ repos:
     -   id: check-shebang-scripts-are-executable
     -   id: check-added-large-files
         args: ['--maxkb=100']
+    -   id: trailing-whitespace
+        types: [python]
+    -   id: end-of-file-fixer
+        types: [python]
 
   - repo: https://github.com/executablebooks/mdformat
     rev: 0.7.17
@@ -52,10 +56,10 @@ repos:
         entry: mdformat .
         language_version: python3.11
 
-  - repo: https://github.com/hadialqattan/pycln
-    rev: v2.4.0
-    hooks:
-      - id: pycln
+  # - repo: https://github.com/hadialqattan/pycln
+  #   rev: v2.4.0
+  #   hooks:
+  #     - id: pycln
 
   - repo: https://github.com/isidentical/teyit
     rev: 0.4.3

Dockerfile

@@ -1,18 +1,42 @@
-FROM python:3.11-slim
+# Build stage
+FROM python:3.11-slim as builder
 
 WORKDIR /app
 
+# Install system dependencies
 RUN apt-get update && apt-get install -y curl && rm -rf /var/lib/apt/lists/*
 
+# Install Poetry
 RUN curl -sSL https://install.python-poetry.org | python3 -
-
-# Ensure Poetry is available in PATH
 ENV PATH="/root/.local/bin:$PATH"
 RUN poetry self add "poetry-plugin-export"
 
+# Copy only dependency files to leverage Docker layer caching
 COPY pyproject.toml poetry.lock ./
 
+# Install dependencies
 RUN poetry export -f requirements.txt --without-hashes -o requirements.txt
 RUN pip install --no-cache-dir -r requirements.txt
 
+# Runtime stage
+FROM python:3.11-slim
+
+# Set environment variables
+ENV PYTHONDONTWRITEBYTECODE=1
+ENV PYTHONUNBUFFERED=1
+
+WORKDIR /app
+
+# Copy only the necessary files from the builder stage
+COPY --from=builder /usr/local/lib/python3.11/site-packages /usr/local/lib/python3.11/site-packages
+COPY --from=builder /usr/local/bin /usr/local/bin
+
+# Copy application code
 COPY . .
+
+# Health check
+HEALTHCHECK --interval=30s --timeout=30s --start-period=5s --retries=3 \
+    CMD curl -f http://localhost:8718/health || exit 1
+
+# Default command
+CMD ["python", "-m", "agentic_security"]

Readme.md

@@ -6,25 +6,30 @@
     The open-source Agentic LLM Vulnerability Scanner
     <br />
     <br />
-
-<p>
-<img alt="GitHub Contributors" src="https://img.shields.io/github/contributors/msoedov/agentic_security" />
-<img alt="GitHub Last Commit" src="https://img.shields.io/github/last-commit/msoedov/agentic_security" />
-<img alt="" src="https://img.shields.io/github/repo-size/msoedov/agentic_security" />
-<img alt="Downloads" src="https://static.pepy.tech/badge/agentic_security" />
-<img alt="GitHub Issues" src="https://img.shields.io/github/issues/msoedov/agentic_security" />
-<img alt="GitHub Pull Requests" src="https://img.shields.io/github/issues-pr/msoedov/agentic_security" />
-<img alt="Github License" src="https://img.shields.io/github/license/msoedov/agentic_security" />
 </p>
-  </p>
+
+<p align="center">
+  <a href="https://github.com/msoedov/agentic_security/commits/main">
+    <img alt="GitHub Last Commit" src="https://img.shields.io/github/last-commit/msoedov/agentic_security?style=for-the-badge&logo=git&labelColor=000000&logoColor=FFFFFF&label=Last Commit&color=6A35FF" />
+  </a>
+  <a href="https://github.com/msoedov/agentic_security">
+    <img alt="GitHub Repo Size" src="https://img.shields.io/github/repo-size/msoedov/agentic_security?style=for-the-badge&logo=database&labelColor=000000&logoColor=FFFFFF&label=Repo Size&color=yellow" />
+  </a>
+  </a>
+  <a href="https://github.com/msoedov/agentic_security/blob/master/LICENSE">
+    <img alt="GitHub License" src="https://img.shields.io/github/license/msoedov/agentic_security?style=for-the-badge&logo=codeigniter&labelColor=000000&logoColor=FFFFFF&label=License&color=FFCC19" />
+  </a>
+  <a href="https://discord.com/channels/1340010688764051499/1340010689309315247"><img alt="Join the community" src="https://img.shields.io/badge/Join%20the%20community-black.svg?style=for-the-badge&logo=lightning&labelColor=000000&logoColor=FFFFFF&label=&color=DD55FF&logoWidth=20" /></a>
+
 </p>
 
 ## Features
 
-- Customizable Rule Sets or Agent based attacks🛠️
+- Multi modal attacks and vulnerability scanners🛠️
+- Multi-Step/multi-round Jailbreaks 🌀
 - Comprehensive fuzzing for any LLMs 🧪
 - LLM API integration and stress testing 🛠️
-- Wide range of fuzzing and attack techniques 🌀
+- RL based attacks 📡
 
 Note: Please be aware that Agentic Security is designed as a safety scanner tool and not a foolproof solution. It cannot guarantee complete protection against all possible threats.
 
@@ -376,6 +381,10 @@ This sample GitHub Action is designed to perform automated security scans
 
 This setup ensures a continuous integration approach towards maintaining security in your projects.
 
+## Module Class
+
+The `Module` class is designed to manage prompt processing and interaction with external AI models and tools. It supports fetching, processing, and posting prompts asynchronously for model vulnerabilities. Check out [module.md](https://github.com/msoedov/agentic_security/blob/main/docs/module.md) for details.
+
 ## Documentation
 
 For more detailed information on how to use Agentic Security, including advanced features and customization options, please refer to the official documentation.

agentic_security/integrations/__init__.py

@@ -0,0 +1,12 @@
+import asyncio
+from typing import Protocol
+
+
+class IntegrationProto(Protocol):
+    def __init__(
+        self, prompt_groups: list, tools_inbox: asyncio.Queue, opts: dict = {}
+    ):
+        ...
+
+    async def apply(self) -> list:
+        ...

agentic_security/probe_actor/cost_module.py

@@ -0,0 +1,58 @@
+def calculate_cost(tokens: int, model: str = "deepseek-chat") -> float:
+    """Calculate API cost based on token count and model.
+
+    Args:
+        tokens (int): Number of tokens used
+        model (str): Model name to calculate cost for
+
+    Returns:
+        float: Cost in USD
+    """
+    # API pricing as of 2024-03-01
+    pricing = {
+        "deepseek-chat": {
+            "input": 0.0007 / 1000,  # $0.70 per million input tokens
+            "output": 0.0028 / 1000,  # $2.80 per million output tokens
+        },
+        "gpt-4-turbo": {
+            "input": 0.01 / 1000,  # $10 per million input tokens
+            "output": 0.03 / 1000,  # $30 per million output tokens
+        },
+        "gpt-4": {
+            "input": 0.03 / 1000,  # $30 per million input tokens
+            "output": 0.06 / 1000,  # $60 per million output tokens
+        },
+        "gpt-3.5-turbo": {
+            "input": 0.0015 / 1000,  # $1.50 per million input tokens
+            "output": 0.002 / 1000,  # $2.00 per million output tokens
+        },
+        "claude-3-opus": {
+            "input": 0.015 / 1000,  # $15 per million input tokens
+            "output": 0.075 / 1000,  # $75 per million output tokens
+        },
+        "claude-3-sonnet": {
+            "input": 0.003 / 1000,  # $3 per million input tokens
+            "output": 0.015 / 1000,  # $15 per million output tokens
+        },
+        "claude-3-haiku": {
+            "input": 0.00025 / 1000,  # $0.25 per million input tokens
+            "output": 0.00125 / 1000,  # $1.25 per million output tokens
+        },
+        "mistral-large": {
+            "input": 0.008 / 1000,  # $8 per million input tokens
+            "output": 0.024 / 1000,  # $24 per million output tokens
+        },
+        "mixtral-8x7b": {
+            "input": 0.002 / 1000,  # $2 per million input tokens
+            "output": 0.006 / 1000,  # $6 per million output tokens
+        },
+    }
+
+    if model not in pricing:
+        raise ValueError(f"Unknown model: {model}")
+
+    # For now, assume 1:1 input/output ratio
+    input_cost = tokens * pricing[model]["input"]
+    output_cost = tokens * pricing[model]["output"]
+
+    return round(input_cost + output_cost, 4)

agentic_security/probe_actor/fuzzer.py

@@ -10,6 +10,7 @@ from skopt.space import Real
 
 from agentic_security.http_spec import Modality
 from agentic_security.models.schemas import Scan, ScanResult
+from agentic_security.probe_actor.cost_module import calculate_cost
 from agentic_security.probe_actor.refusal import refusal_heuristic
 from agentic_security.probe_data import audio_generator, image_generator, msj_data
 from agentic_security.probe_data.data import prepare_prompts
@@ -38,8 +39,6 @@ def multi_modality_spec(llm_spec):
             return llm_spec
         case _:
             return llm_spec
-        # case _:
-        #     raise NotImplementedError(f"Modality {llm_spec.modality} not supported yet")
 
 
 async def process_prompt(
@@ -143,7 +142,7 @@ async def perform_single_shot_scan(
                     module_failures += 1
                 failure_rate = module_failures / max(processed_prompts, 1)
                 failure_rates.append(failure_rate)
-                cost = round(tokens * 1.5 / 1000_000, 2)
+                cost = calculate_cost(tokens)
 
                 yield ScanResult(
                     module=module.dataset_name,
@@ -274,7 +273,7 @@ async def perform_many_shot_scan(
 
                 failure_rate = module_failures / max(processed_prompts, 1)
                 failure_rates.append(failure_rate)
-                cost = round(tokens * 1.5 / 1000_000, 2)
+                cost = calculate_cost(tokens)
 
                 yield ScanResult(
                     module=module.dataset_name,

agentic_security/probe_actor/operator.py

@@ -1,30 +1,165 @@
+import asyncio
+from typing import Any
+
+from pydantic import BaseModel, Field
+from pydantic_ai import Agent, RunContext
+
+
+class AgentSpecification(BaseModel):
+    name: str | None = Field(None, description="Name of the LLM/agent")
+    version: str | None = Field(None, description="Version of the LLM/agent")
+    description: str | None = Field(None, description="Description of the LLM/agent")
+    capabilities: list[str] | None = Field(None, description="List of capabilities")
+    configuration: dict[str, Any] | None = Field(
+        None, description="Configuration settings"
+    )
+
+
+# Define the OperatorToolBox class
 class OperatorToolBox:
-    def __init__(self, llm_spec, datasets):
-        self.llm_spec = llm_spec
+    def __init__(self, spec: AgentSpecification, datasets: list[dict[str, Any]]):
+        self.spec = spec
         self.datasets = datasets
+        self.failures = []
 
-    def get_spec(self):
-        return self.llm_spec
+    def get_spec(self) -> AgentSpecification:
+        return self.spec
 
-    def get_datasets(self):
+    def get_datasets(self) -> list[dict[str, Any]]:
         return self.datasets
 
-    def validate(self):
-        # Validate the tool box
-        pass
+    def validate(self) -> bool:
+        # Validate the tool box based on the specification
+        if not self.spec.name or not self.spec.version:
+            self.failures.append("Invalid specification: Name or version is missing.")
+            return False
+        if not self.datasets:
+            self.failures.append("No datasets provided.")
+            return False
+        return True
 
-    def stop(self):
+    def stop(self) -> None:
         # Stop the tool box
-        pass
+        print("Stopping the toolbox...")
 
-    def run(self):
+    def run(self) -> None:
         # Run the tool box
-        pass
+        print("Running the toolbox...")
 
-    def get_results(self):
+    def get_results(self) -> list[dict[str, Any]]:
         # Get the results
-        pass
+        return self.datasets
 
-    def get_failures(self):
+    def get_failures(self) -> list[str]:
         # Handle failure
-        pass
+        return self.failures
+
+    def run_operation(self, operation: str) -> str:
+        # Run an operation based on the specification
+        if operation not in ["dataset1", "dataset2", "dataset3"]:
+            self.failures.append(f"Operation '{operation}' failed: Dataset not found.")
+            return f"Operation '{operation}' failed: Dataset not found."
+        return f"Operation '{operation}' executed successfully."
+
+
+# Initialize OperatorToolBox with AgentSpecification
+spec = AgentSpecification(
+    name="GPT-4",
+    version="4.0",
+    description="A powerful language model",
+    capabilities=["text-generation", "question-answering"],
+    configuration={"max_tokens": 100},
+)
+
+# dataset_manager_agent.py
+
+
+# Initialize OperatorToolBox
+toolbox = OperatorToolBox(spec=spec, datasets=["dataset1", "dataset2", "dataset3"])
+
+# Define the agent with OperatorToolBox as its dependency
+dataset_manager_agent = Agent(
+    model="gpt-4",
+    deps_type=OperatorToolBox,
+    result_type=str,  # The agent will return string results
+    system_prompt="You can validate the toolbox, run operations, and retrieve results or failures.",
+)
+
+
+@dataset_manager_agent.tool
+async def validate_toolbox(ctx: RunContext[OperatorToolBox]) -> str:
+    """Validate the OperatorToolBox."""
+    is_valid = ctx.deps.validate()
+    if is_valid:
+        return "ToolBox validation successful."
+    else:
+        return "ToolBox validation failed."
+
+
+@dataset_manager_agent.tool
+async def execute_operation(ctx: RunContext[OperatorToolBox], operation: str) -> str:
+    """Execute an operation on a dataset."""
+    result = ctx.deps.run_operation(operation)
+    return result
+
+
+@dataset_manager_agent.tool
+async def retrieve_results(ctx: RunContext[OperatorToolBox]) -> str:
+    """Retrieve the results of operations."""
+    results = ctx.deps.get_results()
+    if results:
+        formatted_results = "\n".join([f"{op}: {res}" for op, res in results.items()])
+        return f"Operation Results:\n{formatted_results}"
+    else:
+        return "No operations have been executed yet."
+
+
+@dataset_manager_agent.tool
+async def retrieve_failures(ctx: RunContext[OperatorToolBox]) -> str:
+    """Retrieve the list of failures."""
+    failures = ctx.deps.get_failures()
+    if failures:
+        formatted_failures = "\n".join(failures)
+        return f"Failures:\n{formatted_failures}"
+    else:
+        return "No failures recorded."
+
+
+# Synchronous run example
+def run_dataset_manager_agent_sync():
+    prompts = [
+        "Validate the toolbox.",
+        "Execute operation on 'dataset2'.",
+        "Execute operation on 'dataset4'.",  # This should fail
+        "Retrieve the results.",
+        "Retrieve any failures.",
+    ]
+
+    for prompt in prompts:
+        result = dataset_manager_agent.run_sync(prompt, deps=toolbox)
+        print(f"Prompt: {prompt}")
+        print(f"Response: {result.data}\n")
+
+
+# Asynchronous run example
+async def run_dataset_manager_agent_async():
+    prompts = [
+        "Validate the toolbox.",
+        "Execute operation on 'dataset2'.",
+        "Execute operation on 'dataset4'.",  # This should fail
+        "Retrieve the results.",
+        "Retrieve any failures.",
+    ]
+
+    for prompt in prompts:
+        result = await dataset_manager_agent.run(prompt, deps=toolbox)
+        print(f"Prompt: {prompt}")
+        print(f"Response: {result.data}\n")
+
+
+if __name__ == "__main__":
+    # Run synchronous example
+    run_dataset_manager_agent_sync()
+
+    # Run asynchronous example
+    asyncio.run(run_dataset_manager_agent_async())

agentic_security/probe_actor/refusal.py

@@ -1,3 +1,5 @@
+from abc import ABC, abstractmethod
+
 from agentic_security.refusal_classifier import RefusalClassifier
 
 classifier = RefusalClassifier()
@@ -35,23 +37,81 @@ REFUSAL_MARKS = [
 ]
 
 
-def check_refusal(response: str, refusal_phrases: list = REFUSAL_MARKS) -> bool:
-    """Check if the response contains any phrases indicating refusal to answer.
+class RefusalClassifierPlugin(ABC):
+    """Base class for refusal classifier plugins."""
 
-    Args:
-    - response (str): The response from the language model.
-    - refusal_phrases (list): A list of phrases indicating refusal to answer.
+    @abstractmethod
+    def is_refusal(self, response: str) -> bool:
+        """Check if the response contains a refusal.
 
-    Returns:
-    - bool: True if the response contains a refusal phrase, False otherwise.
-    """
-    return any(phrase in response for phrase in refusal_phrases)
+        Args:
+            response (str): The response from the language model.
+
+        Returns:
+            bool: True if the response contains a refusal, False otherwise.
+        """
+        pass
 
 
-# TODO: add plugin system for multiple classifiers of the output
+class DefaultRefusalClassifier(RefusalClassifierPlugin):
+    """Default refusal classifier using a list of refusal phrases."""
+
+    def __init__(self, refusal_phrases: list[str] = REFUSAL_MARKS):
+        self.refusal_phrases = refusal_phrases
+
+    def is_refusal(self, response: str) -> bool:
+        """Check if the response contains any phrases indicating refusal to answer.
+
+        Args:
+            response (str): The response from the language model.
+
+        Returns:
+            bool: True if the response contains a refusal phrase, False otherwise.
+        """
+        return any(phrase in response for phrase in self.refusal_phrases)
+
+
+class RefusalClassifierManager:
+    """Manager for refusal classifier plugins."""
+
+    def __init__(self):
+        self.plugins: dict[str, RefusalClassifierPlugin] = {}
+
+    def register_plugin(self, name: str, plugin: RefusalClassifierPlugin):
+        """Register a refusal classifier plugin.
+
+        Args:
+            name (str): The name of the plugin.
+            plugin (RefusalClassifierPlugin): The plugin instance.
+        """
+        self.plugins[name] = plugin
+
+    def is_refusal(self, response: str) -> bool:
+        """Check if the response contains a refusal using all registered plugins.
+
+        Args:
+            response (str): The response from the language model.
+
+        Returns:
+            bool: True if any plugin detects a refusal, False otherwise.
+        """
+        return any(plugin.is_refusal(response) for plugin in self.plugins.values())
+
+
+# Initialize the plugin manager and register the default plugin
+refusal_classifier_manager = RefusalClassifierManager()
+refusal_classifier_manager.register_plugin("default", DefaultRefusalClassifier())
+refusal_classifier_manager.register_plugin("ml_classifier", classifier)
 
 
 def refusal_heuristic(request_json):
-    # TODO: improve this heuristic
+    """Check if the request contains a refusal using the plugin system.
+
+    Args:
+        request_json: The request to check.
+
+    Returns:
+        bool: True if the request contains a refusal, False otherwise.
+    """
     request = str(request_json)
-    return check_refusal(request) or classifier.is_refusal(request)
+    return refusal_classifier_manager.is_refusal(request)

agentic_security/probe_actor/test_refusal.py

@@ -1,13 +1,13 @@
-from agentic_security.probe_actor.refusal import check_refusal
+from agentic_security.probe_actor.refusal import DefaultRefusalClassifier
 
 
 class TestCheckRefusal:
     # The function correctly identifies a refusal phrase in the response.
     def test_identify_refusal_phrase(self):
         response = "I'm sorry, but I cannot provide that information."
-        assert check_refusal(response)
+        assert DefaultRefusalClassifier().is_refusal(response)
 
     # The response is an empty string.
     def test_empty_response(self):
         response = ""
-        assert not check_refusal(response)
+        assert not DefaultRefusalClassifier().is_refusal(response)

agentic_security/probe_data/__init__.py

@@ -408,6 +408,21 @@ REGISTRY = REGISTRY_V0 + [
         },
         "modality": "text",
     },
+    {
+        "dataset_name": "Reinforcement Learning Optimization",
+        "num_prompts": 0,
+        "tokens": 0,
+        "approx_cost": 0.0,
+        "source": "Cloud hosted model",
+        "selected": False,
+        "url": "",
+        "dynamic": True,
+        "opts": {
+            "port": 8718,
+            "modules": ["encoding"],
+        },
+        "modality": "text",
+    },
     {
         "dataset_name": "InspectAI",
         "num_prompts": 0,

agentic_security/probe_data/audio_generator.py

@@ -52,11 +52,37 @@ def generate_audio_mac_wav(prompt: str) -> bytes:
     return audio_bytes
 
 
+def generate_audio_cross_platform(prompt: str) -> bytes:
+    """
+    Generate an audio file from the provided prompt using gTTS for cross-platform support.
+
+    Parameters:
+        prompt (str): Text to convert into audio.
+
+    Returns:
+        bytes: The audio data in MP3 format.
+    """
+    from gtts import gTTS  # Import gTTS for cross-platform support
+
+    tts = gTTS(text=prompt, lang="en")
+    temp_mp3_path = f"temp_audio_{uuid.uuid4().hex}.mp3"
+    tts.save(temp_mp3_path)
+
+    try:
+        with open(temp_mp3_path, "rb") as f:
+            audio_bytes = f.read()
+    finally:
+        if os.path.exists(temp_mp3_path):
+            os.remove(temp_mp3_path)
+
+    return audio_bytes
+
+
 @cache_to_disk()
 def generate_audioform(prompt: str) -> bytes:
     """
     Generate an audio file from the provided prompt in WAV format.
-    Uses macOS 'say' command if the operating system is macOS.
+    Uses macOS 'say' command if the operating system is macOS, otherwise uses gTTS.
 
     Parameters:
         prompt (str): Text to convert into audio.
@@ -67,9 +93,11 @@ def generate_audioform(prompt: str) -> bytes:
     current_os = platform.system()
     if current_os == "Darwin":  # macOS
         return generate_audio_mac_wav(prompt)
+    elif current_os in ["Windows", "Linux"]:
+        return generate_audio_cross_platform(prompt)
     else:
         raise NotImplementedError(
-            "Audio generation is only supported on macOS for now."
+            "Audio generation is only supported on macOS, Windows, and Linux for now."
         )
 
 

agentic_security/probe_data/data.py

@@ -16,6 +16,7 @@ from agentic_security.probe_data.modules import (
     fine_tuned,
     garak_tool,
     inspect_ai_tool,
+    rl_model,
 )
 
 
@@ -265,6 +266,11 @@ def prepare_prompts(dataset_names, budget, tools_inbox=None, options=[]):
             garak_tool.Module(group, tools_inbox=tools_inbox, opts=opts).apply(),
             lazy=True,
         ),
+        "Reinforcement Learning Optimization": lambda opts: dataset_from_iterator(
+            "Reinforcement Learning Optimization",
+            rl_model.Module(group, tools_inbox=tools_inbox, opts=opts).apply(),
+            lazy=True,
+        ),
         "InspectAI": lambda opts: dataset_from_iterator(
             "InspectAI",
             inspect_ai_tool.Module(group, tools_inbox=tools_inbox).apply(),

agentic_security/probe_data/image_generator.py

@@ -38,12 +38,13 @@ def generate_image_dataset(
 
 
 @cache_to_disk()
-def generate_image(prompt: str) -> bytes:
+def generate_image(prompt: str, variant: int = 0) -> bytes:
     """
     Generate an image based on the provided prompt and return it as bytes.
 
     Parameters:
         prompt (str): Text to display on the generated image.
+        variant (int): The variant style of the image.
 
     Returns:
         bytes: The image data in JPG format.
@@ -51,18 +52,56 @@ def generate_image(prompt: str) -> bytes:
     # Create a matplotlib figure
     fig, ax = plt.subplots(figsize=(6, 4))
 
-    # Customize the plot (background color, text, etc.)
-    ax.set_facecolor("lightblue")
-    ax.text(
-        0.5,
-        0.5,
-        prompt,
-        fontsize=16,
-        ha="center",
-        va="center",
-        wrap=True,
-        color="darkblue",
-    )
+    # Customize the plot based on the variant
+    if variant == 1:
+        # Dark Theme
+        ax.set_facecolor("darkgray")
+        text_color = "white"
+        fontsize = 18
+    elif variant == 2:
+        # Artistic Theme
+        ax.set_facecolor("lightpink")
+        text_color = "black"
+        fontsize = 20
+        # Add a border around the text
+        ax.text(
+            0.5,
+            0.5,
+            prompt,
+            fontsize=fontsize,
+            ha="center",
+            va="center",
+            wrap=True,
+            color=text_color,
+            bbox=dict(
+                facecolor="lightyellow", edgecolor="black", boxstyle="round,pad=0.5"
+            ),
+        )
+    elif variant == 3:
+        # Minimalist Theme
+        ax.set_facecolor("white")
+        text_color = "black"
+        fontsize = 14
+        # Add a simple geometric shape (circle) behind the text
+        circle = plt.Circle((0.5, 0.5), 0.3, color="lightblue", fill=True)
+        ax.add_artist(circle)
+    else:
+        # Default Theme
+        ax.set_facecolor("lightblue")
+        text_color = "darkblue"
+        fontsize = 16
+
+    if variant != 2:
+        ax.text(
+            0.5,
+            0.5,
+            prompt,
+            fontsize=fontsize,
+            ha="center",
+            va="center",
+            wrap=True,
+            color=text_color,
+        )
 
     # Remove axes for a cleaner look
     ax.axis("off")

agentic_security/probe_data/modules/rl_model.py

@@ -0,0 +1,247 @@
+import asyncio
+import os
+import random
+import uuid as U
+from abc import ABC, abstractmethod
+from collections import deque
+from typing import Deque
+
+import numpy as np
+import requests
+from loguru import logger
+
+AUTH_TOKEN: str = os.getenv("AS_TOKEN", "gh0-5f4a8ed2-37c6-4bd7-a0cf-7070eae8115b")
+
+
+class PromptSelectionInterface(ABC):
+    """Abstract base class for prompt selection strategies."""
+
+    @abstractmethod
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        """Selects the next prompt based on current state and guard result."""
+        pass
+
+    @abstractmethod
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        """Selects the next prompts based on current state and guard result."""
+        pass
+
+    @abstractmethod
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        """Updates internal rewards based on the outcome of the last selected prompt."""
+        pass
+
+
+class RandomPromptSelector(PromptSelectionInterface):
+    """Random prompt selector with cycle prevention using history."""
+
+    def __init__(self, prompts: list[str], history_size: int = 300):
+        if not prompts:
+            raise ValueError("Prompts list cannot be empty")
+        self.prompts = prompts
+        self.history: Deque[str] = deque(maxlen=history_size)
+
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        return [self.select_next_prompt(current_prompt, passed_guard)]
+
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        self.history.append(current_prompt)
+        available = [p for p in self.prompts if p not in self.history]
+
+        if not available:
+            available = self.prompts
+            self.history.clear()
+
+        return random.choice(available)
+
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        pass  # No learning in random selection
+
+
+class CloudRLPromptSelector(PromptSelectionInterface):
+    """Cloud-based reinforcement learning prompt selector with fallback."""
+
+    def __init__(
+        self,
+        prompts: list[str],
+        api_url: str,
+        auth_token: str = AUTH_TOKEN,
+        history_size: int = 300,
+        timeout: int = 5,
+        run_id: str = "",
+    ):
+        if not prompts:
+            raise ValueError("Prompts list cannot be empty")
+        self.prompts = prompts
+        self.api_url = api_url
+        self.headers = {"Authorization": f"Bearer {auth_token}"}
+        self.timeout = timeout
+        self.run_id = run_id or U.uuid4().hex
+
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        return self.select_next_prompts(current_prompt, passed_guard)[0]
+
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> str:
+        try:
+            response = requests.post(
+                f"{self.api_url}/rl-model/select-next-prompt",
+                json={
+                    "run_id": U.uuid4().hex,
+                    "current_prompt": current_prompt,
+                    "passed_guard": passed_guard,
+                },
+                headers=self.headers,
+                timeout=self.timeout,
+            )
+            response.raise_for_status()
+            return response.json().get("next_prompts", [])
+        except requests.exceptions.RequestException as e:
+            logger.error(f"Cloud request failed: {e}")
+            return [self._fallback_selection()]
+
+    def _fallback_selection(self) -> str:
+        return random.choice(self.prompts)
+
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        ...
+
+
+class QLearningPromptSelector(PromptSelectionInterface):
+    """Q-Learning based prompt selector with exploration/exploitation tradeoff."""
+
+    def __init__(
+        self,
+        prompts: list[str],
+        learning_rate: float = 0.1,
+        discount_factor: float = 0.9,
+        initial_exploration: float = 1.0,
+        exploration_decay: float = 0.995,
+        min_exploration: float = 0.01,
+        history_size: int = 300,
+    ):
+        if not prompts:
+            raise ValueError("Prompts list cannot be empty")
+
+        self.prompts = prompts
+        self.learning_rate = learning_rate
+        self.discount_factor = discount_factor
+        self.exploration_rate = initial_exploration
+        self.exploration_decay = exploration_decay
+        self.min_exploration = min_exploration
+        self.history: Deque[str] = deque(maxlen=history_size)
+
+        # Initialize Q-table with small random values
+        self.q_table: dict[str, dict[str, float]] = {
+            state: {
+                action: np.random.uniform(0, 0.1)
+                for action in prompts
+                if action != state
+            }
+            for state in prompts
+        }
+
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        return [self.select_next_prompt(current_prompt, passed_guard)]
+
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        self.history.append(current_prompt)
+        available = [a for a in self.prompts if a not in self.history]
+
+        if not available:
+            available = self.prompts
+            self.history.clear()
+
+        # Exploration-exploitation tradeoff
+        if np.random.random() < self.exploration_rate:
+            selected = random.choice(available)
+        else:
+            q_values = {a: self.q_table[current_prompt][a] for a in available}
+            selected = max(q_values, key=q_values.get)  # type: ignore
+
+        # Decay exploration rate
+        self.exploration_rate = max(
+            self.min_exploration, self.exploration_rate * self.exploration_decay
+        )
+        return selected
+
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        if (
+            previous_prompt not in self.q_table
+            or current_prompt not in self.q_table[previous_prompt]
+        ):
+            return
+
+        # Calculate temporal difference error
+        max_future_q = max(self.q_table[current_prompt].values(), default=0.0)
+        td_target = reward + self.discount_factor * max_future_q
+        td_error = td_target - self.q_table[previous_prompt][current_prompt]
+
+        # Update Q-value
+        self.q_table[previous_prompt][current_prompt] += self.learning_rate * td_error
+
+
+class Module:
+    def __init__(
+        self, prompt_groups: list[str], tools_inbox: asyncio.Queue, opts: dict = {}
+    ):
+        self.tools_inbox = tools_inbox
+        self.opts = opts
+        self.prompt_groups = prompt_groups
+        self.max_prompts = self.opts.get("max_prompts", 10)  # Default max M prompts
+        self.run_id = U.uuid4().hex
+        self.batch_size = self.opts.get("batch_size", 500)
+        self.rl_model = CloudRLPromptSelector(
+            prompt_groups, "https://edge.metaheuristic.co", run_id=self.run_id
+        )
+
+    async def apply(self):
+        current_prompt = "What is AI?"
+        passed_guard = False
+        for _ in range(max(self.max_prompts, 1)):
+            # Fetch prompts from the API
+            prompts = await asyncio.to_thread(
+                lambda: self.rl_model.select_next_prompts(
+                    current_prompt, passed_guard=passed_guard
+                )
+            )
+
+            if not prompts:
+                logger.error("No prompts retrieved from the API.")
+                return
+
+            logger.info(f"Retrieved {len(prompts)} prompts.")
+
+            for i, prompt in enumerate(prompts):
+                logger.info(f"Processing prompt {i+1}/{len(prompts)}: {prompt}")
+                yield prompt
+                current_prompt = prompt
+                while not self.tools_inbox.empty():
+                    ref = await self.tools_inbox.get()
+                    print(ref, "ref")
+                    message, _, ready = ref["message"], ref["reply"], ref["ready"]
+                    yield message
+                    ready.set()

agentic_security/probe_data/modules/test_rl_model.py

@@ -0,0 +1,215 @@
+import asyncio
+from collections import deque
+from unittest.mock import Mock, patch
+
+import numpy as np
+import pytest
+import requests
+
+# Import the classes to be tested
+from agentic_security.probe_data.modules.rl_model import (
+    CloudRLPromptSelector,
+    Module,
+    QLearningPromptSelector,
+    RandomPromptSelector,
+)
+
+
+# Fixtures for reusable test data
+@pytest.fixture
+def dataset_prompts() -> list[str]:
+    return [
+        "What is AI?",
+        "How does RL work?",
+        "Explain supervised learning.",
+        "What is reinforcement learning?",
+    ]
+
+
+@pytest.fixture
+def mock_requests() -> Mock:
+    with patch("requests.post") as mock_requests:
+        yield mock_requests
+
+
+@pytest.fixture
+def mock_rl_selector() -> Mock:
+    return CloudRLPromptSelector(
+        dataset_prompts,
+        api_url="https://edge.metaheuristic.co",
+    )
+
+
+@pytest.fixture
+def tools_inbox() -> asyncio.Queue:
+    return asyncio.Queue()
+
+
+# Tests for RandomPromptSelector
+class TestRandomPromptSelector:
+    def test_initialization(self, dataset_prompts):
+        selector = RandomPromptSelector(dataset_prompts)
+        assert selector.prompts == dataset_prompts
+        assert isinstance(selector.history, deque)
+        assert selector.history.maxlen == 300
+
+    def test_select_next_prompt(self, dataset_prompts):
+        selector = RandomPromptSelector(dataset_prompts)
+        current_prompt = "What is AI?"
+        next_prompt = selector.select_next_prompt(current_prompt, passed_guard=True)
+        assert next_prompt in dataset_prompts
+        assert next_prompt != current_prompt
+
+    def test_update_rewards_no_op(self, dataset_prompts):
+        selector = RandomPromptSelector(dataset_prompts)
+        selector.update_rewards("What is AI?", "How does RL work?", 1.0, True)
+        assert len(selector.history) == 0
+
+
+# Tests for CloudRLPromptSelector
+class TestCloudRLPromptSelector:
+    def test_initialization(self, dataset_prompts):
+        selector = CloudRLPromptSelector(dataset_prompts, "http://example.com", "token")
+        assert selector.prompts == dataset_prompts
+        assert selector.api_url == "http://example.com"
+        assert selector.headers == {"Authorization": "Bearer token"}
+
+    def test_select_next_prompt_success(self, dataset_prompts, mock_requests):
+        mock_requests.return_value.status_code = 200
+        mock_requests.return_value.json.return_value = {"next_prompts": ["What is AI?"]}
+
+        selector = CloudRLPromptSelector(dataset_prompts, "http://example.com", "token")
+        next_prompt = selector.select_next_prompt(
+            "How does RL work?", passed_guard=True
+        )
+        assert next_prompt == "What is AI?"
+        mock_requests.assert_called_once()
+
+    def test_fallback_on_failure(self, dataset_prompts, mock_requests):
+        mock_requests.side_effect = requests.exceptions.RequestException
+        selector = CloudRLPromptSelector(dataset_prompts, "http://example.com", "token")
+        next_prompt = selector.select_next_prompt("What is AI?", passed_guard=True)
+        assert next_prompt in dataset_prompts
+
+    def test_select_next_prompt_success_service(self, dataset_prompts):
+        selector = CloudRLPromptSelector(
+            dataset_prompts,
+            api_url="https://edge.metaheuristic.co",
+        )
+        next_prompt = selector.select_next_prompt(
+            "How does RL work?", passed_guard=True
+        )
+        assert next_prompt
+
+
+# Tests for QLearningPromptSelector
+class TestQLearningPromptSelector:
+    def test_initialization(self, dataset_prompts):
+        selector = QLearningPromptSelector(dataset_prompts)
+        assert selector.prompts == dataset_prompts
+        assert selector.exploration_rate == 1.0
+        assert len(selector.q_table) == len(dataset_prompts)
+        assert all(
+            len(v) == len(dataset_prompts) - 1 for v in selector.q_table.values()
+        )
+
+    def test_select_next_prompt_exploration(self, dataset_prompts):
+        selector = QLearningPromptSelector(dataset_prompts, initial_exploration=1.0)
+        next_prompt = selector.select_next_prompt("What is AI?", passed_guard=True)
+        assert next_prompt in dataset_prompts
+        assert next_prompt != "What is AI?"
+
+    def test_select_next_prompt_exploitation(self, dataset_prompts):
+        selector = QLearningPromptSelector(dataset_prompts, initial_exploration=0.0)
+        selector.q_table["What is AI?"]["How does RL work?"] = 10.0
+        next_prompt = selector.select_next_prompt("What is AI?", passed_guard=True)
+        assert next_prompt == "How does RL work?"
+
+    def test_update_rewards(self, dataset_prompts):
+        selector = QLearningPromptSelector(dataset_prompts)
+        selector.update_rewards("What is AI?", "How does RL work?", 1.0, True)
+        assert selector.q_table["What is AI?"]["How does RL work?"] > 0.0
+
+    def test_exploration_rate_decay(self, dataset_prompts):
+        selector = QLearningPromptSelector(
+            dataset_prompts, initial_exploration=1.0, exploration_decay=0.9
+        )
+        assert selector.exploration_rate == 1.0
+        selector.select_next_prompt("What is AI?", passed_guard=True)
+        assert selector.exploration_rate == 0.9
+        selector.select_next_prompt("How does RL work?", passed_guard=True)
+        assert selector.exploration_rate == 0.81
+
+
+# Edge Cases and Error Handling
+def test_empty_prompts():
+    with pytest.raises(ValueError, match="Prompts list cannot be empty"):
+        RandomPromptSelector([])
+
+
+def test_cloud_rl_selector_invalid_url(dataset_prompts):
+    selector = CloudRLPromptSelector(dataset_prompts, "invalid_url", "token")
+    next_prompt = selector.select_next_prompt("What is AI?", passed_guard=True)
+    assert next_prompt in dataset_prompts
+
+
+def test_q_learning_selector_invalid_reward(dataset_prompts):
+    selector = QLearningPromptSelector(dataset_prompts)
+    selector.update_rewards("What is AI?", "How does RL work?", np.nan, True)
+
+
+# Tests for Module class
+class TestModule:
+    @pytest.fixture
+    def mock_uuid(self):
+        with patch("uuid.uuid4") as mock:
+            mock.return_value.hex = "test_run_id"
+            yield mock
+
+    def test_initialization(self, dataset_prompts, tools_inbox, mock_uuid):
+        module = Module(dataset_prompts, tools_inbox)
+        assert module.prompt_groups == dataset_prompts
+        assert module.tools_inbox == tools_inbox
+        assert module.max_prompts == 10
+        assert module.batch_size == 500
+        assert module.run_id == "test_run_id"
+        assert isinstance(module.rl_model, CloudRLPromptSelector)
+
+    def test_initialization_with_options(self, dataset_prompts, tools_inbox, mock_uuid):
+        opts = {
+            "max_prompts": 100,
+            "batch_size": 50,
+        }
+        module = Module(dataset_prompts, tools_inbox, opts)
+        assert module.max_prompts == 100
+        assert module.batch_size == 50
+
+    @pytest.mark.asyncio
+    async def test_apply_basic_flow(
+        self, dataset_prompts, tools_inbox, mock_rl_selector
+    ):
+        module = Module(dataset_prompts, tools_inbox)
+
+        count = 0
+        async for prompt in module.apply():
+            assert prompt
+            count += 1
+            if count >= 3:  # Test a few iterations
+                break
+
+    @pytest.mark.asyncio
+    async def test_apply_rl_with_tools_inbox(self, dataset_prompts, tools_inbox):
+        # Add a test message to the tools inbox
+        test_message = {
+            "message": "Test message",
+            "reply": None,
+            "ready": asyncio.Event(),
+        }
+        await tools_inbox.put(test_message)
+
+        module = Module(dataset_prompts, tools_inbox)
+
+        async for output in module.apply():
+            if output == "Test message":
+                test_message["ready"].set()
+                break

agentic_security/probe_data/stenography_fn.py

@@ -1,5 +1,6 @@
 import base64
 import random
+import string
 
 
 def rot13(input_text):
@@ -98,3 +99,47 @@ def zigzag_obfuscation(text):
         else:
             new_text += char
     return new_text
+
+
+def caesar_cipher(text, shift=3):
+    """Encrypts text using Caesar cipher with specified shift."""
+    result = []
+    for char in text:
+        if char.isupper():
+            result.append(chr((ord(char) + shift - 65) % 26 + 65))
+        elif char.islower():
+            result.append(chr((ord(char) + shift - 97) % 26 + 97))
+        else:
+            result.append(char)
+    return "".join(result)
+
+
+def substitution_cipher(text, key=None):
+    """Encrypts text using a substitution cipher with optional key."""
+    if key is None:
+        key = list(string.ascii_lowercase)
+        random.shuffle(key)
+        key = "".join(key)
+
+    # Create translation table
+    alphabet = string.ascii_lowercase
+    translation = str.maketrans(alphabet, key)
+
+    # Apply translation
+    return text.lower().translate(translation)
+
+
+def vigenere_cipher(text, key):
+    """Encrypts text using Vigenère cipher with provided key."""
+    result = []
+    key_length = len(key)
+    key_as_int = [ord(i) for i in key.lower()]
+    text = text.lower()
+
+    for i, char in enumerate(text):
+        if char.isalpha():
+            shift = key_as_int[i % key_length] - 97
+            result.append(chr((ord(char) + shift - 97) % 26 + 97))
+        else:
+            result.append(char)
+    return "".join(result)

agentic_security/probe_data/test_audio_generator.py

@@ -3,6 +3,7 @@ import platform
 import pytest
 
 from agentic_security.probe_data.audio_generator import (
+    generate_audio_cross_platform,
     generate_audio_mac_wav,
     generate_audioform,
 )
@@ -24,6 +25,13 @@ def test_generate_audioform_mac():
         audio_bytes = generate_audioform(prompt)
         assert isinstance(audio_bytes, bytes)
         assert len(audio_bytes) > 0
+
+
+def test_generate_audio_cross_platform():
+    if platform.system() in ["Windows", "Linux"]:
+        prompt = "This is a cross-platform test."
+        audio_bytes = generate_audio_cross_platform(prompt)
+        assert isinstance(audio_bytes, bytes)
+        assert len(audio_bytes) > 0
     else:
-        with pytest.raises(NotImplementedError):
-            generate_audioform("This should raise an error on non-macOS systems.")
+        pytest.skip("Test is only applicable on Windows and Linux.")

agentic_security/probe_data/test_image_generator.py

@@ -1,5 +1,7 @@
 from unittest.mock import patch
 
+import pytest
+
 from agentic_security.probe_data.image_generator import (
     generate_image,
     generate_image_dataset,
@@ -7,9 +9,10 @@ from agentic_security.probe_data.image_generator import (
 from agentic_security.probe_data.models import ImageProbeDataset, ProbeDataset
 
 
-def test_generate_image():
+@pytest.mark.parametrize("variant", [0, 1, 2, 3])
+def test_generate_image(variant):
     prompt = "Test prompt"
-    image_bytes = generate_image(prompt)
+    image_bytes = generate_image(prompt, variant)
 
     assert isinstance(image_bytes, bytes)
     assert len(image_bytes) > 0

agentic_security/routes/probe.py

@@ -1,6 +1,7 @@
 import random
 
 from fastapi import APIRouter, File, Header, HTTPException, UploadFile
+from fastapi.responses import JSONResponse
 
 from ..models.schemas import FileProbeResponse, Probe
 from ..probe_actor.refusal import REFUSAL_MARKS
@@ -70,3 +71,9 @@ async def self_probe_image():
 @router.get("/v1/data-config")
 async def data_config():
     return [m for m in REGISTRY]
+
+
+@router.get("/health")
+async def health_check():
+    """Health check endpoint."""
+    return JSONResponse(content={"status": "ok"})

agentic_security/routes/scan.py

@@ -1,6 +1,6 @@
 from datetime import datetime
 
-from fastapi import APIRouter, BackgroundTasks, HTTPException
+from fastapi import APIRouter, BackgroundTasks, File, HTTPException, Query, UploadFile
 from fastapi.responses import StreamingResponse
 
 from ..core.app import get_stop_event, get_tools_inbox, set_current_run
@@ -52,3 +52,28 @@ async def scan(scan_parameters: Scan, background_tasks: BackgroundTasks):
 async def stop_scan():
     get_stop_event().set()
     return {"status": "Scan stopped"}
+
+
+@router.post("/scan-csv")
+async def scan_csv(
+    background_tasks: BackgroundTasks,
+    file: UploadFile = File(...),
+    llmSpec: UploadFile = File(...),
+    optimize: bool = Query(False),
+    maxBudget: int = Query(10_000),
+    enableMultiStepAttack: bool = Query(False),
+):
+    # TODO: content dataset to fuzzer
+    content = await file.read()  # noqa
+    llm_spec = await llmSpec.read()
+
+    scan_parameters = Scan(
+        llmSpec=llm_spec,
+        optimize=optimize,
+        maxBudget=1000,
+        enableMultiStepAttack=enableMultiStepAttack,
+    )
+
+    return StreamingResponse(
+        streaming_response_generator(scan_parameters), media_type="application/json"
+    )

agentic_security/routes/test_csv.py

@@ -0,0 +1,22 @@
+from fastapi.testclient import TestClient
+
+import agentic_security.test_spec_assets as test_spec_assets
+from agentic_security.routes.scan import router
+
+client = TestClient(router)
+
+
+def test_upload_csv_and_run():
+    # Create a sample CSV content
+    csv_content = "id,prompt\nspec1,value1\nspec2,value3"
+    # Send a POST request to the /upload-csv endpoint
+    response = client.post(
+        "/scan-csv?optimize=false&enableMultiStepAttack=false&maxBudget=1000",
+        files={
+            "file": ("test.csv", csv_content, "text/csv"),
+            "llmSpec": ("spec.txt", test_spec_assets.SAMPLE_SPEC, "text/plain"),
+        },
+    )
+
+    assert response.status_code == 200
+    assert "Scan completed." in response.text

agentic_security/routes/test_health.py

@@ -0,0 +1,12 @@
+from fastapi.testclient import TestClient
+
+from ..app import app
+
+
+def test_health_check():
+    """Test the health check endpoint."""
+    client = TestClient(app)
+
+    response = client.get("/health")
+    assert response.status_code == 200
+    assert response.json() == {"status": "ok"}

agentic_security/static/base.js

@@ -17,7 +17,7 @@ Content-Type: application/json
 
 `,
   `POST https://api.openai.com/v1/chat/completions
-Authorization: Bearer sk-xxxxxxxxx
+Authorization: Bearer $OPENAI_API_KEY
 Content-Type: application/json
 
 {
@@ -25,6 +25,20 @@ Content-Type: application/json
 "messages": [{"role": "user", "content": "<<PROMPT>>"}],
 "temperature": 0.7
 }
+`,
+  `
+POST https://api.deepseek.com/chat/completions
+Authorization: Bearer $DEEPSEEK_API_KEY
+Content-Type: application/json
+
+{
+  "model": "deepseek-chat",
+  "messages": [
+    {"role": "system", "content": "You are a helpful assistant."},
+    {"role": "user", "content": "<<PROMPT>>"}
+  ],
+  "stream": false
+}
 `,
   `POST https://api.replicate.com/v1/models/mistralai/mixtral-8x7b-instruct-v0.1/predictions
 Authorization: Bearer $APIKEY
@@ -165,6 +179,7 @@ Content-Type: application/json
 let LLM_CONFIGS = [
   { name: 'Custom API', prompts: 40000, customInstructions: 'Requires api spec' },
   { name: 'Open AI', prompts: 24000 },
+  { name: 'Deepseek v1', prompts: 24000 },
   { name: 'Replicate', prompts: 40000 },
   { name: 'Groq', prompts: 40000 },
   { name: 'Together.ai', prompts: 40000 },

agentic_security/static/index.html

@@ -437,7 +437,7 @@
                   <th class="p-3">Vulnerability Module</th>
                   <th class="p-3">% Strength</th>
                   <th class="p-3">Number of Tokens</th>
-                  <th class="p-3">Cost (in gpt-3 tokens)</th>
+                  <th class="p-3">Approx Cost (in tokens)</th>
                 </tr>
               </thead>
               <tbody>

changelog.sh

@@ -0,0 +1,25 @@
+#!/bin/bash
+
+# Get the last tag
+LAST_TAG=$(git describe --tags --abbrev=0 2>/dev/null)
+
+if [ -z "$LAST_TAG" ]; then
+    echo "No tags found. Retrieving all commits."
+    LOG_RANGE="HEAD"
+else
+    echo "Generating changelog from last tag: $LAST_TAG"
+    LOG_RANGE="$LAST_TAG..HEAD"
+fi
+
+# Retrieve commit messages excluding merge commits and format them with author names and stripped email domain as nickname
+CHANGELOG=$(git log --pretty=format:"- %s by %an, @%ae)" --no-merges $LOG_RANGE | sed -E 's/@([^@]+)@([^@]+)\..*/@\1/')
+
+# Output the changelog
+if [ -n "$CHANGELOG" ]; then
+    echo "# Changelog"
+    echo "
+## Changes since $LAST_TAG"
+    echo "$CHANGELOG"
+else
+    echo "No new commits since last tag."
+fi

docs/abstractions.md

@@ -0,0 +1,55 @@
+# Abstractions in Agentic Security
+
+This document outlines the key abstractions used in the Agentic Security project, providing insights into the classes, interfaces, and design patterns that form the backbone of the system.
+
+## Key Abstractions
+
+### AgentSpecification
+
+- **Purpose**: Defines the specification for a language model or agent, including its name, version, description, capabilities, and configuration settings.
+- **Usage**: Used to initialize and configure the `OperatorToolBox` and other components that interact with language models.
+
+### OperatorToolBox
+
+- **Purpose**: Serves as the main class for managing dataset operations, including validation, execution, and result retrieval.
+- **Methods**:
+  - `get_spec()`: Returns the agent specification.
+  - `get_datasets()`: Retrieves the datasets for operations.
+  - `validate()`: Validates the toolbox setup.
+  - `run_operation(operation: str)`: Executes a specified operation.
+
+### DatasetManagerAgent
+
+- **Purpose**: Provides tools for managing and executing operations on datasets through an agent-based approach.
+- **Tools**:
+  - `validate_toolbox`: Validates the `OperatorToolBox`.
+  - `execute_operation`: Executes operations on datasets.
+  - `retrieve_results`: Retrieves operation results.
+  - `retrieve_failures`: Retrieves any failures encountered.
+
+### ProbeDataset
+
+- **Purpose**: Represents a dataset used in security scans, including metadata, prompts, and associated costs.
+- **Methods**:
+  - `metadata_summary()`: Provides a summary of the dataset's metadata.
+
+### Refusal Classifier
+
+- **Purpose**: Analyzes responses from language models to detect potential security vulnerabilities.
+- **Design**: Utilizes predefined rules and machine learning models for classification.
+
+## Design Patterns
+
+### Modular Architecture
+
+- **Description**: The system is designed with a modular architecture, allowing for easy integration of new components and features.
+- **Benefits**: Enhances flexibility, extensibility, and scalability.
+
+### Agent-Based Design
+
+- **Description**: Utilizes an agent-based approach for managing and executing operations on datasets.
+- **Benefits**: Provides a structured framework for interacting with language models and datasets.
+
+## Conclusion
+
+The abstractions in Agentic Security are designed to provide a flexible and extensible framework for managing and executing security scans on language models. This document highlights the key classes, interfaces, and design patterns that contribute to the system's architecture and functionality.

docs/api_reference.md

@@ -43,6 +43,7 @@ This section provides detailed information about the Agentic Security API.
 ## Authentication
 
 All API requests require an API key. Include it in the `Authorization` header:
+
 ```
 Authorization: Bearer YOUR_API_KEY
 ```

docs/ci_cd.md

@@ -29,8 +29,9 @@ jobs:
 ## Custom CI/CD Pipelines
 
 For custom pipelines, ensure the following steps:
+
 1. Install dependencies.
-2. Run the `agentic_security ci` command.
+1. Run the `agentic_security ci` command.
 
 ## Further Reading
 

docs/configuration.md

@@ -13,7 +13,7 @@ The default configuration file is `agesec.toml`. It includes settings for:
 ## Customizing Configuration
 
 1. Open the `agesec.toml` file in a text editor.
-2. Modify the settings as needed. For example, to change the port:
+1. Modify the settings as needed. For example, to change the port:
    ```toml
    [modules.AgenticBackend.opts]
    port = 8718

docs/contributing.md

@@ -5,23 +5,23 @@ We welcome contributions to Agentic Security! Follow these steps to get started:
 ## How to Contribute
 
 1. **Fork the Repository**: Click the "Fork" button at the top of the repository page.
-2. **Clone Your Fork**: Clone your forked repository to your local machine.
+1. **Clone Your Fork**: Clone your forked repository to your local machine.
    ```bash
    git clone https://github.com/mmsoedov/agentic_security.git
    ```
-3. **Create a Branch**: Create a new branch for your feature or bugfix.
+1. **Create a Branch**: Create a new branch for your feature or bugfix.
    ```bash
    git checkout -b feature-name
    ```
-4. **Make Changes**: Implement your changes and commit them.
+1. **Make Changes**: Implement your changes and commit them.
    ```bash
    git commit -m "Description of changes"
    ```
-5. **Push Changes**: Push your changes to your fork.
+1. **Push Changes**: Push your changes to your fork.
    ```bash
    git push origin feature-name
    ```
-6. **Open a Pull Request**: Go to the original repository and open a pull request.
+1. **Open a Pull Request**: Go to the original repository and open a pull request.
 
 ## Code of Conduct
 

docs/datasets.md

@@ -5,7 +5,7 @@ Agentic Security allows you to extend datasets to enhance its capabilities.
 ## Adding New Datasets
 
 1. Place your dataset files in the `datasets` directory.
-2. Ensure each file contains a `prompt` column for processing.
+1. Ensure each file contains a `prompt` column for processing.
 
 ## Supported Formats
 
@@ -15,6 +15,7 @@ Agentic Security allows you to extend datasets to enhance its capabilities.
 ## Example
 
 To add a new dataset:
+
 ```bash
 cp my_dataset.csv datasets/
 ```

docs/design.md

@@ -0,0 +1,51 @@
+# Design Document
+
+This document provides an overview of the design and architecture of the Agentic Security project. It outlines the key components, their interactions, and the design principles guiding the development of the system.
+
+## Overview
+
+Agentic Security is an open-source LLM vulnerability scanner designed to identify and mitigate potential security threats in language models. It integrates various modules and datasets to perform comprehensive security scans.
+
+## Architecture
+
+The system is built around a modular architecture, allowing for flexibility and extensibility. The core components include:
+
+- **Agentic Security Core**: The main application responsible for orchestrating the security scans and managing interactions with external modules.
+- **Probe Actor**: Handles the execution of fuzzing and attack techniques on language models.
+- **Probe Data**: Manages datasets used for testing and validation, including loading and processing data.
+- **Refusal Classifier**: Analyzes responses from language models to identify potential security issues.
+
+## Key Components
+
+### Agentic Security Core
+
+The core application is responsible for initializing the system, managing configurations, and coordinating the execution of security scans. It provides a command-line interface for users to interact with the system.
+
+### Probe Actor
+
+The Probe Actor module implements various fuzzing and attack techniques. It is designed to test the robustness of language models by simulating different attack scenarios.
+
+### Probe Data
+
+The Probe Data module manages datasets used in security scans. It supports loading data from local files and external sources, providing a flexible framework for testing different scenarios.
+
+### Refusal Classifier
+
+The Refusal Classifier analyzes responses from language models to detect potential security vulnerabilities. It uses predefined rules and machine learning models to classify responses.
+
+## Design Principles
+
+- **Modularity**: The system is designed to be modular, allowing for easy integration of new components and features.
+- **Extensibility**: New modules and datasets can be added to the system without significant changes to the core architecture.
+- **Scalability**: The system is built to handle large datasets and complex security scans efficiently.
+
+## Interaction Flow
+
+1. **Initialization**: The system is initialized with the necessary configurations and datasets.
+1. **Execution**: The Probe Actor executes security scans on the language models using the datasets provided by the Probe Data module.
+1. **Analysis**: The Refusal Classifier analyzes the responses to identify potential security issues.
+1. **Reporting**: Results are compiled and presented to the user, highlighting any vulnerabilities detected.
+
+## Conclusion
+
+The design of Agentic Security emphasizes flexibility, extensibility, and scalability, providing a robust framework for identifying and mitigating security threats in language models. This document serves as a guide to understanding the system's architecture and key components.

docs/external_module.md

@@ -0,0 +1,43 @@
+## Module Interface Documentation
+
+The `Module` class interface provides a standardized way to create and use modules in the `agentic_security` project.
+
+Here is an example of a module that implements the `ModuleProtocol` interface. This example shows how to create a module that processes prompts and sends results to a queue.
+
+```python
+from typing import List, Dict, Any, AsyncGenerator
+import asyncio
+from .module_protocol import ModuleProtocol
+
+class ModuleProtocol(ModuleProtocol):
+    def __init__(self, prompt_groups: List[Any], tools_inbox: asyncio.Queue, opts: Dict[str, Any]):
+        self.prompt_groups = prompt_groups
+        self.tools_inbox = tools_inbox
+        self.opts = opts
+
+    async def apply(self) -> AsyncGenerator[str, None]:
+        for group in self.prompt_groups:
+            await asyncio.sleep(1) 
+            result = f"Processed {group}"
+            await self.tools_inbox.put(result)
+            yield result
+```
+
+#### Usage Example
+
+```python
+import asyncio
+import ModuleProtocol
+
+tools_inbox = asyncio.Queue()
+prompt_groups = ["group1", "group2"]
+opts = {"max_prompts": 1000, "batch_size": 100}
+
+module = ModuleProtocol(prompt_groups, tools_inbox, opts)
+
+async def main():
+    async for result in module.apply():
+        print(result)
+
+asyncio.run(main())
+```

docs/getting_started.md

@@ -5,11 +5,11 @@ Welcome to Agentic Security! This guide will help you get started with using the
 ## Quick Start
 
 1. Ensure you have completed the [installation](installation.md) steps.
-2. Run the following command to start the application:
+1. Run the following command to start the application:
    ```bash
    agentic_security
    ```
-3. Access the application at `http://localhost:8718`.
+1. Access the application at `http://localhost:8718`.
 
 ## Basic Usage
 

docs/http_spec.md

@@ -0,0 +1,127 @@
+# HTTP Specification Documentation
+
+The HTTP specification in the Agentic Security project is designed to handle various types of requests, including text, image, audio, and file uploads. This documentation provides a detailed overview of the HTTP specification and its usage.
+
+## Overview
+
+The HTTP specification is implemented in the `LLMSpec` class, which is used to define and execute HTTP requests. The class supports different modalities, including text, image, audio, and file uploads, and provides methods to validate and execute these requests.
+
+## Modalities
+
+The HTTP specification supports the following modalities:
+
+### Text
+
+Text-based requests are the most common type of request. The `LLMSpec` class replaces the `<<PROMPT>>` placeholder in the request body with the provided prompt.
+
+### Image
+
+Image-based requests include an image encoded in base64 format. The `LLMSpec` class replaces the `<<BASE64_IMAGE>>` placeholder in the request body with the provided base64-encoded image.
+
+### Audio
+
+Audio-based requests include an audio file encoded in base64 format. The `LLMSpec` class replaces the `<<BASE64_AUDIO>>` placeholder in the request body with the provided base64-encoded audio.
+
+### Files
+
+File-based requests include file uploads. The `LLMSpec` class handles multipart form data and includes the provided files in the request.
+
+## LLMSpec Class
+
+The `LLMSpec` class is the core of the HTTP specification. It provides the following methods and properties:
+
+### Methods
+
+- **`from_string(http_spec: str) -> LLMSpec`**: Parses an HTTP specification string into an `LLMSpec` object.
+- **`validate(prompt: str, encoded_image: str, encoded_audio: str, files: dict) -> None`**: Validates the request parameters based on the specified modality.
+- **`probe(prompt: str, encoded_image: str = "", encoded_audio: str = "", files: dict = {}) -> httpx.Response`**: Sends an HTTP request using the specified parameters.
+- **`verify() -> httpx.Response`**: Verifies the HTTP specification by sending a test request.
+
+### Properties
+
+- **`modality: Modality`**: Returns the modality of the request (text, image, audio, or files).
+
+## Examples
+
+### Text Request
+
+```python
+http_spec = """
+POST https://api.example.com/v1/chat/completions
+Authorization: Bearer sk-xxxxxxxxx
+Content-Type: application/json
+
+{
+    "model": "gpt-3.5-turbo",
+    "messages": [{"role": "user", "content": "<<PROMPT>>"}],
+    "temperature": 0.7
+}
+"""
+
+spec = LLMSpec.from_string(http_spec)
+response = await spec.probe("What is the capital of France?")
+```
+
+### Image Request
+
+```python
+http_spec = """
+POST https://api.example.com/v1/chat/completions
+Authorization: Bearer sk-xxxxxxxxx
+Content-Type: application/json
+
+{
+    "model": "gpt-4-vision-preview",
+    "messages": [{"role": "user", "content": "What is in this image? <<BASE64_IMAGE>>"}],
+    "temperature": 0.7
+}
+"""
+
+spec = LLMSpec.from_string(http_spec)
+encoded_image = encode_image_base64_by_url("https://example.com/image.jpg")
+response = await spec.probe("What is in this image?", encoded_image=encoded_image)
+```
+
+### Audio Request
+
+```python
+http_spec = """
+POST https://api.example.com/v1/chat/completions
+Authorization: Bearer sk-xxxxxxxxx
+Content-Type: application/json
+
+{
+    "model": "whisper-large-v3",
+    "messages": [{"role": "user", "content": "Transcribe this audio: <<BASE64_AUDIO>>"}],
+    "temperature": 0.7
+}
+"""
+
+spec = LLMSpec.from_string(http_spec)
+encoded_audio = encode_audio_base64_by_url("https://example.com/audio.mp3")
+response = await spec.probe("Transcribe this audio:", encoded_audio=encoded_audio)
+```
+
+### File Request
+
+```python
+http_spec = """
+POST https://api.example.com/v1/chat/completions
+Authorization: Bearer sk-xxxxxxxxx
+Content-Type: multipart/form-data
+
+{
+    "model": "gpt-3.5-turbo",
+    "messages": [{"role": "user", "content": "Process this file: <<FILE>>"}],
+    "temperature": 0.7
+}
+"""
+
+spec = LLMSpec.from_string(http_spec)
+files = {"file": ("document.txt", open("document.txt", "rb"))}
+response = await spec.probe("Process this file:", files=files)
+```
+
+## Conclusion
+
+The HTTP specification in the Agentic Security project provides a flexible and powerful way to handle various types of requests. This documentation serves as a guide to understanding and utilizing the HTTP specification effectively.

docs/image_generation.md

@@ -0,0 +1,119 @@
+# Image Generation System
+
+The image generation system creates visual probes for security testing by converting text prompts into images. This document explains its architecture and implementation.
+
+## Overview
+
+The system:
+
+1. Converts text datasets into image datasets
+1. Generates images using matplotlib
+1. Encodes images for transmission
+1. Integrates with the LLM probing system
+
+## Core Components
+
+### Image Generation
+
+```python
+@cache_to_disk()
+def generate_image(prompt: str) -> bytes:
+    """
+    Generates a JPEG image containing the provided text prompt
+    """
+    # Create figure with light blue background
+    fig, ax = plt.subplots(figsize=(6, 4))
+    ax.set_facecolor("lightblue")
+
+    # Add centered text
+    ax.text(
+        0.5, 0.5,
+        prompt,
+        fontsize=16,
+        ha="center",
+        va="center",
+        wrap=True,
+        color="darkblue"
+    )
+
+    # Save to buffer
+    buffer = io.BytesIO()
+    plt.savefig(buffer, format="jpeg", bbox_inches="tight")
+    return buffer.getvalue()
+```
+
+### Dataset Conversion
+
+```python
+def generate_image_dataset(text_dataset: list[ProbeDataset]) -> list[ImageProbeDataset]:
+    """
+    Converts text datasets into image datasets
+    """
+    image_datasets = []
+
+    for dataset in text_dataset:
+        image_prompts = [
+            generate_image(prompt)
+            for prompt in tqdm(dataset.prompts)
+        ]
+
+        image_datasets.append(ImageProbeDataset(
+            test_dataset=dataset,
+            image_prompts=image_prompts
+        ))
+
+    return image_datasets
+```
+
+### Image Encoding
+
+```python
+def encode(image: bytes) -> str:
+    """
+    Encodes image bytes into base64 data URL
+    """
+    encoded = base64.b64encode(image).decode("utf-8")
+    return "data:image/jpeg;base64," + encoded
+```
+
+## Integration
+
+### RequestAdapter
+
+The RequestAdapter class integrates image generation with LLM probing:
+
+```python
+class RequestAdapter:
+    def __init__(self, llm_spec):
+        if not llm_spec.has_image:
+            raise ValueError("LLMSpec must have an image")
+        self.llm_spec = llm_spec
+
+    async def probe(self, prompt: str, encoded_image: str = "",
+                   encoded_audio: str = "", files={}) -> httpx.Response:
+        encoded_image = generate_image(prompt)
+        encoded_image = encode(encoded_image)
+        return await self.llm_spec.probe(prompt, encoded_image, encoded_audio, files)
+```
+
+## Key Features
+
+- **Caching**: Generated images are cached to disk using @cache_to_disk
+- **Progress Tracking**: tqdm progress bars for dataset conversion
+- **Error Handling**: Validates LLM specifications before probing
+- **Standard Formats**: Uses JPEG format with base64 encoding
+
+## Configuration
+
+The system is configured through:
+
+1. Figure size (6x4 inches)
+1. Background color (light blue)
+1. Text styling (16pt dark blue centered text)
+1. Image format (JPEG)
+
+## Limitations
+
+- Currently only supports text-based image generation
+- Fixed visual style and formatting
+- Requires matplotlib and associated dependencies

docs/index.md

@@ -7,16 +7,7 @@
     <br />
     <br />
 
-<p>
-<img alt="GitHub Contributors" src="https://img.shields.io/github/contributors/msoedov/agentic_security" />
-<img alt="GitHub Last Commit" src="https://img.shields.io/github/last-commit/msoedov/agentic_security" />
-<img alt="" src="https://img.shields.io/github/repo-size/msoedov/agentic_security" />
-<img alt="Downloads" src="https://static.pepy.tech/badge/agentic_security" />
-<img alt="GitHub Issues" src="https://img.shields.io/github/issues/msoedov/agentic_security" />
-<img alt="GitHub Pull Requests" src="https://img.shields.io/github/issues-pr/msoedov/agentic_security" />
-<img alt="Github License" src="https://img.shields.io/github/license/msoedov/agentic_security" />
 </p>
-  </p>
 </p>
 
 ## Features
@@ -28,389 +19,6 @@
 
 Note: Please be aware that Agentic Security is designed as a safety scanner tool and not a foolproof solution. It cannot guarantee complete protection against all possible threats.
 
-## 📦 Installation
-
-To get started with Agentic Security, simply install the package using pip:
-
-```shell
-pip install agentic_security
-```
-
-## ⛓️ Quick Start
-
-```shell
-agentic_security
-
-2024-04-13 13:21:31.157 | INFO     | agentic_security.probe_data.data:load_local_csv:273 - Found 1 CSV files
-2024-04-13 13:21:31.157 | INFO     | agentic_security.probe_data.data:load_local_csv:274 - CSV files: ['prompts.csv']
-INFO:     Started server process [18524]
-INFO:     Waiting for application startup.
-INFO:     Application startup complete.
-INFO:     Uvicorn running on http://0.0.0.0:8718 (Press CTRL+C to quit)
-```
-
-```shell
-python -m agentic_security
-# or
-agentic_security --help
-
-
-agentic_security --port=PORT --host=HOST
-
-```
-
 ## UI 🧙
 
 <img width="100%" alt="booking-screen" src="https://res.cloudinary.com/dq0w2rtm9/image/upload/v1736433557/z0bsyzhsqlgcr3w4ovwp.gif">
-
-## LLM kwargs
-
-Agentic Security uses plain text HTTP spec like:
-
-```http
-POST https://api.openai.com/v1/chat/completions
-Authorization: Bearer sk-xxxxxxxxx
-Content-Type: application/json
-
-{
-     "model": "gpt-3.5-turbo",
-     "messages": [{"role": "user", "content": "<<PROMPT>>"}],
-     "temperature": 0.7
-}
-
-```
-
-Where `<<PROMPT>>` will be replaced with the actual attack vector during the scan, insert the `Bearer XXXXX` header value with your app credentials.
-
-### Adding LLM integration templates
-
-TBD
-
-```
-....
-```
-
-## Adding own dataset
-
-To add your own dataset you can place one or multiples csv files with `prompt` column, this data will be loaded on `agentic_security` startup
-
-```
-2024-04-13 13:21:31.157 | INFO     | agentic_security.probe_data.data:load_local_csv:273 - Found 1 CSV files
-2024-04-13 13:21:31.157 | INFO     | agentic_security.probe_data.data:load_local_csv:274 - CSV files: ['prompts.csv']
-```
-
-## Run as CI check
-
-Init config
-
-```shell
-agentic_security init
-
-2025-01-08 20:12:02.449 | INFO     | agentic_security.lib:generate_default_cfg:324 - Default configuration generated successfully to agesec.toml.
-
-```
-
-default config sample
-
-```toml
-
-[general]
-# General configuration for the security scan
-llmSpec = """
-POST http://0.0.0.0:8718/v1/self-probe
-Authorization: Bearer XXXXX
-Content-Type: application/json
-
-{
-    "prompt": "<<PROMPT>>"
-}
-""" # LLM API specification
-maxBudget = 1000000 # Maximum budget for the scan
-max_th = 0.3 # Maximum failure threshold (percentage)
-optimize = false # Enable optimization during scanning
-enableMultiStepAttack = false # Enable multi-step attack simulations
-
-
-[modules.aya-23-8B_advbench_jailbreak]
-dataset_name = "simonycl/aya-23-8B_advbench_jailbreak"
-
-
-[modules.AgenticBackend]
-dataset_name = "AgenticBackend"
-[modules.AgenticBackend.opts]
-port = 8718
-modules = ["encoding"]
-
-
-[thresholds]
-# Threshold settings
-low = 0.15
-medium = 0.3
-high = 0.5
-
-
-
-```
-
-List module
-
-```shell
-agentic_security ls
-
-                   Dataset Registry
-┏━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┓
-┃ Dataset Name                       ┃ Num Prompts ┃  Tokens ┃ Source                            ┃ Selected ┃ Dynamic ┃ Modality ┃
-┡━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━┩
-│ simonycl/aya-23-8B_advbench_jailb… │         416 │    None │ Hugging Face Datasets             │    ✘     │    ✘    │ text     │
-├────────────────────────────────────┼─────────────┼─────────┼───────────────────────────────────┼──────────┼─────────┼──────────┤
-│ acmc/jailbreaks_dataset_with_perp… │       11191 │    None │ Hugging Face Datasets             │    ✘     │    ✘    │ text     │
-├────────────────────────────────────┼─────────────┼─────────┼───────────────────────────────────┼──────────┼─────────┼──────────┤
-
-```
-
-```shell
-agentic_security ci
-
-2025-01-08 20:13:07.536 | INFO     | agentic_security.probe_data.data:load_local_csv:331 - Found 2 CSV files
-2025-01-08 20:13:07.536 | INFO     | agentic_security.probe_data.data:load_local_csv:332 - CSV files: ['failures.csv', 'issues_with_descriptions.csv']
-2025-01-08 20:13:07.552 | WARNING  | agentic_security.probe_data.data:load_local_csv:345 - File issues_with_descriptions.csv does not contain a 'prompt' column
-2025-01-08 20:13:08.892 | INFO     | agentic_security.lib:load_config:52 - Configuration loaded successfully from agesec.toml.
-2025-01-08 20:13:08.892 | INFO     | agentic_security.lib:entrypoint:259 - Configuration loaded successfully.
-{'general': {'llmSpec': 'POST http://0.0.0.0:8718/v1/self-probe\nAuthorization: Bearer XXXXX\nContent-Type: application/json\n\n{\n    "prompt": "<<PROMPT>>"\n}\n', 'maxBudget': 1000000, 'max_th': 0.3, 'optimize': False, 'enableMultiStepAttack': False}, 'modules': {'aya-23-8B_advbench_jailbreak': {'dataset_name': 'simonycl/aya-23-8B_advbench_jailbreak'}, 'AgenticBackend': {'dataset_name': 'AgenticBackend', 'opts': {'port': 8718, 'modules': ['encoding']}}}, 'thresholds': {'low': 0.15, 'medium': 0.3, 'high': 0.5}}
-Scanning modules: 0it [00:00, ?it/s]2025-01-08 20:13:08.903 | INFO     | agentic_security.probe_data.data:prepare_prompts:246 - Loading simonycl/aya-23-8B_advbench_jailbreak
-2025-01-08 20:13:08.905 | INFO     | agentic_security.probe_data.data:prepare_prompts:280 - Loading AgenticBackend
-2025-01-08 20:13:08.905 | INFO     | agentic_security.probe_actor.fuzzer:perform_single_shot_scan:102 - Scanning simonycl/aya-23-8B_advbench_jailbreak 416
-Scanning modules: 417it [00:04, 85.85it/s]2025-01-08 20:13:13.825 | INFO     | agentic_security.probe_actor.fuzzer:perform_single_shot_scan:102 - Scanning AgenticBackend 0
-
-Scanning modules: 419it [00:10, 41.37it/s]
-
-Security Scan Results
-Time: 2025-01-08 20:13:19
-Duration: 10.1s
-Modules Scanned: 2
-Threshold: 30.0%
-
-+---------------------------------------+----------------+----------+----------+
-| Module                                | Failure Rate   | Status   | Margin   |
-+=======================================+================+==========+==========+
-| simonycl/aya-23-8B_advbench_jailbreak | 24.8%          | ✔        | 5.2%     |
-+---------------------------------------+----------------+----------+----------+
-
-Summary:
-Total Passing: 2/2 (100.0%)
-```
-
-## Extending dataset collections
-
-1. Add new metadata to agentic_security.probe_data.REGISTRY
-
-```python
-    {
-        "dataset_name": "markush1/LLM-Jailbreak-Classifier",
-        "num_prompts": 1119,
-        "tokens": 19758,
-        "approx_cost": 0.0,
-        "source": "Hugging Face Datasets",
-        "selected": True,
-        "dynamic": False,
-        "url": "https://huggingface.co/markush1/LLM-Jailbreak-Classifier",
-    },
-```
-
-and implement loader into
-
-```python
-@dataclass
-class ProbeDataset:
-    dataset_name: str
-    metadata: dict
-    prompts: list[str]
-    tokens: int
-    approx_cost: float
-
-    def metadata_summary(self):
-        return {
-            "dataset_name": self.dataset_name,
-            "num_prompts": len(self.prompts),
-            "tokens": self.tokens,
-            "approx_cost": self.approx_cost,
-        }
-
-```
-
-## Dynamic datasets with mutations
-
-One of the current examples uses sampling for the existing preloaded prompt data and applying mutations yielding a new dataset
-
-```python
-class Stenography:
-    fn_library = {
-        "rot5": stenography_fn.rot5,
-        "rot13": stenography_fn.rot13,
-        "base64": stenography_fn.base64_encode,
-        "mirror": stenography_fn.mirror_words,
-        "random_case": stenography_fn.randomize_letter_case,
-        "scramble_words": stenography_fn.scramble_words,
-        "noise_characters": stenography_fn.insert_noise_characters,
-        "substitute_with_ascii": stenography_fn.substitute_with_ascii,
-        "remove_vowels": stenography_fn.remove_vowels,
-        "zigzag_obfuscation": stenography_fn.zigzag_obfuscation,
-    }
-
-    def __init__(self, prompt_groups: [ProbeDataset]):
-        self.prompt_groups = prompt_groups
-
-    def apply(self):
-        for prompt_group in self.prompt_groups:
-
-            size = len(prompt_group.prompts)
-            for name, fn in self.fn_library.items():
-                logger.info(f"Applying {name} to {prompt_group.dataset_name}")
-                selected_prompts = random.sample(prompt_group.prompts, int(size * 0.25))
-                prompts = [
-                    f"Follow the instructions in {name}: {fn(p)}"
-                    for p in selected_prompts
-                ]
-                yield ProbeDataset(
-                    dataset_name=f"stenography.{name}({prompt_group.dataset_name})",
-                    metadata={},
-                    prompts=prompts,
-                    tokens=count_words_in_list(prompts),
-                    approx_cost=0.0,
-                )
-```
-
-## Probe endpoint
-
-In the example of custom integration, we use `/v1/self-probe` for the sake of integration testing.
-
-```python
-POST https://agentic_security-preview.vercel.app/v1/self-probe
-Authorization: Bearer XXXXX
-Content-Type: application/json
-
-{
-    "prompt": "<<PROMPT>>"
-}
-
-```
-
-This endpoint randomly mimics the refusal of a fake LLM.
-
-```python
-@app.post("/v1/self-probe")
-def self_probe(probe: Probe):
-    refuse = random.random() < 0.2
-    message = random.choice(REFUSAL_MARKS) if refuse else "This is a test!"
-    message = probe.prompt + " " + message
-    return {
-        "id": "chatcmpl-abc123",
-        "object": "chat.completion",
-        "created": 1677858242,
-        "model": "gpt-3.5-turbo-0613",
-        "usage": {"prompt_tokens": 13, "completion_tokens": 7, "total_tokens": 20},
-        "choices": [
-            {
-                "message": {"role": "assistant", "content": message},
-                "logprobs": None,
-                "finish_reason": "stop",
-                "index": 0,
-            }
-        ],
-    }
-
-```
-
-## Image Modality
-
-To probe the image modality, you can use the following HTTP request:
-
-```http
-POST http://0.0.0.0:9094/v1/self-probe-image
-Authorization: Bearer XXXXX
-Content-Type: application/json
-
-[
-    {
-        "role": "user",
-        "content": [
-            {
-                "type": "text",
-                "text": "What is in this image?"
-            },
-            {
-                "type": "image_url",
-                "image_url": {
-                    "url": "data:image/jpeg;base64,<<BASE64_IMAGE>>"
-                }
-            }
-        ]
-    }
-]
-```
-
-Replace `XXXXX` with your actual API key and `<<BASE64_IMAGE>>` is the image variable.
-
-## Audio Modality
-
-To probe the audio modality, you can use the following HTTP request:
-
-```http
-POST http://0.0.0.0:9094/v1/self-probe-file
-Authorization: Bearer $GROQ_API_KEY
-Content-Type: multipart/form-data
-
-{
-    "file": "@./sample_audio.m4a",
-    "model": "whisper-large-v3"
-}
-```
-
-Replace `$GROQ_API_KEY` with your actual API key and ensure that the `file` parameter points to the correct audio file path.
-
-## CI/CD integration
-
-This sample GitHub Action is designed to perform automated security scans
-
-[Sample GitHub Action Workflow](https://github.com/msoedov/agentic_security/blob/main/.github/workflows/security-scan.yml)
-
-This setup ensures a continuous integration approach towards maintaining security in your projects.
-
-## Documentation
-
-For more detailed information on how to use Agentic Security, including advanced features and customization options, please refer to the official documentation.
-
-## Roadmap and Future Goals
-
-- \[ \] Expand dataset variety
-- \[ \] Introduce two new attack vectors
-- \[ \] Develop initial attacker LLM
-- \[ \] Complete integration of OWASP Top 10 classification
-
-| Tool                    | Source                                                                        | Integrated |
-|-------------------------|-------------------------------------------------------------------------------|------------|
-| Garak                   | [leondz/garak](https://github.com/leondz/garak)                               | ✅          |
-| InspectAI               | [UKGovernmentBEIS/inspect_ai](https://github.com/UKGovernmentBEIS/inspect_ai) | ✅          |
-| llm-adaptive-attacks    | [tml-epfl/llm-adaptive-attacks](https://github.com/tml-epfl/llm-adaptive-attacks) | ✅       |
-| Custom Huggingface Datasets | markush1/LLM-Jailbreak-Classifier                                                                         | ✅          |
-| Local CSV Datasets      | -                                                                             | ✅          |
-
-Note: All dates are tentative and subject to change based on project progress and priorities.
-
-## 👋 Contributing
-
-Contributions to Agentic Security are welcome! If you'd like to contribute, please follow these steps:
-
-- Fork the repository on GitHub
-- Create a new branch for your changes
-- Commit your changes to the new branch
-- Push your changes to the forked repository
-- Open a pull request to the main Agentic Security repository
-
-Before contributing, please read the contributing guidelines.
-
-## License
-
-Agentic Security is released under the Apache License v2.
-
-## Contact us

docs/operator.md

@@ -0,0 +1,123 @@
+# Operator Module
+
+The `operator.py` module provides tools for managing and operating on datasets using an agent-based approach. It is designed to facilitate the execution of operations on datasets through a structured and validated process.
+
+## Classes
+
+### AgentSpecification
+
+Defines the specification for an LLM/agent:
+
+- `name`: Name of the LLM/agent
+- `version`: Version of the LLM/agent
+- `description`: Description of the LLM/agent
+- `capabilities`: List of capabilities
+- `configuration`: Configuration settings
+
+### OperatorToolBox
+
+Main class for dataset operations:
+
+- `__init__(spec: AgentSpecification, datasets: list[dict[str, Any]])`: Initialize with agent spec and datasets. This sets up the toolbox with the necessary specifications and datasets for operation.
+- `get_spec()`: Get the agent specification. Returns the `AgentSpecification` object associated with the toolbox.
+- `get_datasets()`: Get the datasets. Returns a list of datasets that the toolbox operates on.
+- `validate()`: Validate the toolbox. Checks if the toolbox is correctly set up with valid specifications and datasets.
+- `stop()`: Stop the toolbox. Halts any ongoing operations within the toolbox.
+- `run()`: Run the toolbox. Initiates the execution of operations as defined in the toolbox.
+- `get_results()`: Get operation results. Retrieves the results of operations performed by the toolbox.
+- `get_failures()`: Get failures. Provides a list of any failures encountered during operations.
+- `run_operation(operation: str)`: Run a specific operation. Executes a given operation on the datasets, returning the result or failure message.
+
+## Agent Tools
+
+The `dataset_manager_agent` provides these tools:
+
+### validate_toolbox
+
+Validates the OperatorToolBox:
+
+```python
+@dataset_manager_agent.tool
+async def validate_toolbox(ctx: RunContext[OperatorToolBox]) -> str
+```
+
+### execute_operation
+
+Executes an operation on a dataset:
+
+```python
+@dataset_manager_agent.tool
+async def execute_operation(ctx: RunContext[OperatorToolBox], operation: str) -> str
+```
+
+### retrieve_results
+
+Retrieves operation results:
+
+```python
+@dataset_manager_agent.tool
+async def retrieve_results(ctx: RunContext[OperatorToolBox]) -> str
+```
+
+### retrieve_failures
+
+Retrieves failures:
+
+```python
+@dataset_manager_agent.tool
+async def retrieve_failures(ctx: RunContext[OperatorToolBox]) -> str
+```
+
+## Usage Examples
+
+### Initializing the OperatorToolBox
+
+To initialize the `OperatorToolBox`, you need to provide an `AgentSpecification` and a list of datasets:
+
+```python
+spec = AgentSpecification(
+    name="GPT-4",
+    version="4.0",
+    description="A powerful language model",
+    capabilities=["text-generation", "question-answering"],
+    configuration={"max_tokens": 100},
+)
+
+datasets = [{"name": "dataset1"}, {"name": "dataset2"}]
+
+toolbox = OperatorToolBox(spec=spec, datasets=datasets)
+```
+
+### Synchronous Usage
+
+```python
+def run_dataset_manager_agent_sync():
+    prompts = [
+        "Validate the toolbox.",
+        "Execute operation on 'dataset2'.",
+        "Retrieve the results.",
+        "Retrieve any failures."
+    ]
+
+    for prompt in prompts:
+        result = dataset_manager_agent.run_sync(prompt, deps=toolbox)
+        print(f"Response: {result.data}")
+```
+
+### Asynchronous Usage
+
+```python
+async def run_dataset_manager_agent_async():
+    prompts = [
+        "Validate the toolbox.",
+        "Execute operation on 'dataset2'.",
+        "Retrieve the results.",
+        "Retrieve any failures."
+    ]
+
+    for prompt in prompts:
+        result = await dataset_manager_agent.run(prompt, deps=toolbox)
+        print(f"Response: {result.data}")
+```
+
+These updates provide a more detailed and comprehensive understanding of the `operator.py` module, its classes, and its usage.

docs/optimizer.md

@@ -0,0 +1,78 @@
+# Bayesian Optimization in Security Fuzzing
+
+The fuzzer implements an optimization system using scikit-optimize (skopt) to minimize failure rates during security scans. This document explains the optimizer's implementation and behavior.
+
+## Overview
+
+The optimizer is used in both single-shot and many-shot scanning modes when the `optimize` parameter is True. It dynamically adjusts scan parameters to minimize failure rates while staying within budget constraints.
+
+## Implementation Details
+
+### Initialization
+
+The optimizer is initialized with:
+
+```python
+Optimizer(
+    [Real(0, 1)],  # Single parameter space (0 to 1)
+    base_estimator="GP",  # Gaussian Process estimator
+    n_initial_points=25  # Initial exploration points
+)
+```
+
+### Optimization Process
+
+1. **Parameter Space**: A single real-valued parameter between 0 and 1
+1. **Objective**: Minimize the failure rate (negative failure rate is maximized)
+1. **Update Mechanism**:
+   ```python
+   next_point = optimizer.ask()
+   optimizer.tell(next_point, -failure_rate)
+   ```
+1. **Early Stopping**: If best failure rate exceeds 50%:
+   ```python
+   if best_failure_rate > 0.5:
+       yield ScanResult.status_msg(
+           f"High failure rate detected ({best_failure_rate:.2%}). Stopping this module..."
+       )
+       break
+   ```
+
+## Usage in Scanning
+
+The optimizer is integrated into both scan types:
+
+### Single-shot Scan
+
+- Used in `perform_single_shot_scan()`
+- Optimizes failure rates across prompt modules
+- Considers token budget constraints
+
+### Many-shot Scan
+
+- Used in `perform_many_shot_scan()`
+- Handles more complex multi-step attacks
+- Maintains separate failure rate tracking
+
+## Key Parameters
+
+| Parameter | Description |
+|-----------|-------------|
+| base_estimator | Gaussian Process (GP) used for optimization |
+| n_initial_points | 25 initial exploration points |
+| Real(0, 1) | Single parameter space being optimized |
+| failure_rate | Current failure rate being minimized |
+
+## Optimization Flow
+
+1. Initialize optimizer with GP estimator
+1. Collect initial 25 data points
+1. For each prompt:
+   - Calculate current failure rate
+   - Update optimizer with new point
+   - Check for early stopping conditions
+1. Continue until scan completes or budget exhausted
+
+## Error Handling
+
+The optimizer is wrapped in try/except blocks to ensure scan failures don't crash the entire process. Any optimization errors are logged and the scan continues with default parameters.

docs/probe_actor.md

@@ -5,6 +5,7 @@ The `probe_actor` module is a critical component of the Agentic Security project
 ## Files and Key Components
 
 ### fuzzer.py
+
 - **Functions:**
   - `async def generate_prompts(...)`: Asynchronously generates prompts for scanning.
   - `def multi_modality_spec(llm_spec)`: Defines specifications for multi-modality.
@@ -14,6 +15,7 @@ The `probe_actor` module is a critical component of the Agentic Security project
   - `def scan_router(...)`: Routes scan requests.
 
 ### refusal.py
+
 - **Functions:**
   - `def check_refusal(response: str, refusal_phrases: list = REFUSAL_MARKS) -> bool`: Checks if a response contains refusal phrases.
   - `def refusal_heuristic(request_json)`: Applies heuristics to determine refusal.
@@ -21,6 +23,7 @@ The `probe_actor` module is a critical component of the Agentic Security project
 ## Usage Examples
 
 ### Performing a Single-Shot Scan
+
 ```python
 from agentic_security.probe_actor.fuzzer import perform_single_shot_scan
 
@@ -28,6 +31,7 @@ await perform_single_shot_scan(prompt="Test prompt")
 ```
 
 ### Checking for Refusal
+
 ```python
 from agentic_security.probe_actor.refusal import check_refusal
 

docs/probe_data.md

@@ -5,6 +5,7 @@ The `probe_data` module is a core component of the Agentic Security project, res
 ## Files and Key Components
 
 ### audio_generator.py
+
 - **Functions:**
   - `encode(content: bytes) -> str`: Encodes audio content to a string format.
   - `generate_audio_mac_wav(prompt: str) -> bytes`: Generates audio in WAV format for macOS.
@@ -13,6 +14,7 @@ The `probe_data` module is a core component of the Agentic Security project, res
   - `RequestAdapter`: Handles requests for audio generation.
 
 ### data.py
+
 - **Functions:**
   - `load_dataset_general(...)`: Loads datasets with general specifications.
   - `count_words_in_list(str_list)`: Counts words in a list of strings.
@@ -21,6 +23,7 @@ The `probe_data` module is a core component of the Agentic Security project, res
   - `Stenography`: Applies transformations to prompt groups.
 
 ### image_generator.py
+
 - **Functions:**
   - `generate_image_dataset(...)`: Generates a dataset of images.
   - `generate_image(prompt: str) -> bytes`: Generates an image from a prompt.
@@ -28,25 +31,73 @@ The `probe_data` module is a core component of the Agentic Security project, res
   - `RequestAdapter`: Handles requests for image generation.
 
 ### models.py
+
 - **Classes:**
   - `ProbeDataset`: Represents a dataset for probing.
   - `ImageProbeDataset`: Extends `ProbeDataset` for image data.
 
 ### msj_data.py
+
 - **Functions:**
   - `load_dataset_generic(...)`: Loads a generic dataset.
 - **Classes:**
   - `ProbeDataset`: Represents a dataset for probing.
 
 ### stenography_fn.py
+
 - **Functions:**
   - `rot13(input_text)`: Applies ROT13 transformation.
   - `base64_encode(data)`: Encodes data in base64 format.
   - `mirror_words(text)`: Mirrors words in the text.
 
+### rl_model.py
+
+- **Classes:**
+  - `PromptSelectionInterface`: Abstract base class for prompt selection strategies.
+
+    - Methods:
+      - `select_next_prompt(current_prompt: str, passed_guard: bool) -> str`: Selects next prompt
+      - `select_next_prompts(current_prompt: str, passed_guard: bool) -> list[str]`: Selects multiple prompts
+      - `update_rewards(previous_prompt: str, current_prompt: str, reward: float, passed_guard: bool) -> None`: Updates rewards
+
+  - `RandomPromptSelector`: Basic random selection with history tracking.
+
+    - Parameters:
+      - `prompts: list[str]`: List of available prompts
+      - `history_size: int = 3`: Size of history to prevent cycles
+
+  - `CloudRLPromptSelector`: Cloud-based RL implementation with fallback.
+
+    - Parameters:
+      - `prompts: list[str]`: List of available prompts
+      - `api_url: str`: URL of RL service
+      - `auth_token: str = AUTH_TOKEN`: Authentication token
+      - `history_size: int = 300`: Size of history
+      - `timeout: int = 5`: Request timeout
+      - `run_id: str = ""`: Unique run identifier
+
+  - `QLearningPromptSelector`: Local Q-learning implementation.
+
+    - Parameters:
+      - `prompts: list[str]`: List of available prompts
+      - `learning_rate: float = 0.1`: Learning rate
+      - `discount_factor: float = 0.9`: Discount factor
+      - `initial_exploration: float = 1.0`: Initial exploration rate
+      - `exploration_decay: float = 0.995`: Exploration decay rate
+      - `min_exploration: float = 0.01`: Minimum exploration rate
+      - `history_size: int = 300`: Size of history
+
+  - `Module`: Main class that uses CloudRLPromptSelector.
+
+    - Parameters:
+      - `prompt_groups: list[str]`: Groups of prompts
+      - `tools_inbox: asyncio.Queue`: Queue for tool communication
+      - `opts: dict = {}`: Configuration options
+
 ## Usage Examples
 
 ### Generating Audio
+
 ```python
 from agentic_security.probe_data.audio_generator import generate_audioform
 
@@ -54,12 +105,26 @@ audio_bytes = generate_audioform("Hello, world!")
 ```
 
 ### Loading a Dataset
+
 ```python
 from agentic_security.probe_data.data import load_dataset_general
 
 dataset = load_dataset_general("example_dataset")
 ```
 
+### Using RL Model
+
+```python
+from agentic_security.probe_data.modules.rl_model import QLearningPromptSelector
+
+prompts = ["What is AI?", "Explain machine learning"]
+selector = QLearningPromptSelector(prompts)
+
+current_prompt = "What is AI?"
+next_prompt = selector.select_next_prompt(current_prompt, passed_guard=True)
+selector.update_rewards(current_prompt, next_prompt, reward=1.0, passed_guard=True)
+```
+
 ## Conclusion
 
 The `probe_data` module provides essential functionality for handling and transforming datasets within the Agentic Security project. This documentation serves as a guide to understanding and utilizing the module's capabilities.

docs/quickstart.md

@@ -0,0 +1,65 @@
+# Quickstart Guide
+
+Welcome to the Quickstart Guide for Agentic Security. This guide will help you set up and start using the project quickly.
+
+## Installation
+
+To get started with Agentic Security, install the package using pip:
+
+```shell
+pip install agentic_security
+```
+
+## Initial Setup
+
+After installation, you can start the application using the following command:
+
+```shell
+agentic_security
+```
+
+This will initialize the server and prepare it for use.
+
+## Basic Usage
+
+To run the main application, use:
+
+```shell
+python -m agentic_security
+```
+
+You can also view help options with:
+
+```shell
+agentic_security --help
+```
+
+## Running as a CI Check
+
+Initialize the configuration for CI checks:
+
+```shell
+agentic_security init
+```
+
+This will generate a default configuration file named `agesec.toml`.
+
+## Additional Commands
+
+- List available modules:
+
+  ```shell
+  agentic_security ls
+  ```
+
+- Run a security scan:
+
+  ```shell
+  agentic_security ci
+  ```
+
+## Further Information
+
+For more detailed information, refer to the [Documentation](index.md) or the [API Reference](api_reference.md).
+
+This quickstart guide should help you get up and running with Agentic Security efficiently.

docs/refusal_classifier_plugins.md

@@ -0,0 +1,79 @@
+# Refusal Classifier Plugin System Documentation
+
+The refusal classifier plugin system allows for the creation and use of custom refusal classifiers. This system is designed to be modular and extensible, enabling users to add their own refusal detection logic.
+
+## Overview
+
+The plugin system is based on the `RefusalClassifierPlugin` abstract base class, which defines the interface for all refusal classifier plugins. The `RefusalClassifierManager` is used to register and manage these plugins.
+
+## Creating a Plugin
+
+To create a custom refusal classifier plugin, you must implement the `RefusalClassifierPlugin` abstract base class. This class requires the implementation of the `is_refusal` method, which checks if a response contains a refusal.
+
+```python
+from abc import ABC, abstractmethod
+
+class RefusalClassifierPlugin(ABC):
+    """Base class for refusal classifier plugins."""
+
+    @abstractmethod
+    def is_refusal(self, response: str) -> bool:
+        """Check if the response contains a refusal.
+
+        Args:
+            response (str): The response from the language model.
+
+        Returns:
+            bool: True if the response contains a refusal, False otherwise.
+        """
+        pass
+```
+
+### Example Plugin
+
+Here is an example of a custom refusal classifier plugin that checks for specific phrases:
+
+```python
+class CustomRefusalClassifier(RefusalClassifierPlugin):
+    def __init__(self, custom_phrases: List[str]):
+        self.custom_phrases = custom_phrases
+
+    def is_refusal(self, response: str) -> bool:
+        """Check if the response contains any custom refusal phrases.
+
+        Args:
+            response (str): The response from the language model.
+
+        Returns:
+            bool: True if the response contains a custom refusal phrase, False otherwise.
+        """
+        return any(phrase in response for phrase in self.custom_phrases)
+```
+
+## Registering a Plugin
+
+To register a custom refusal classifier plugin, use the `RefusalClassifierManager`:
+
+```python
+from agentic_security.probe_actor.refusal import RefusalClassifierManager
+
+# Initialize the plugin manager
+refusal_classifier_manager = RefusalClassifierManager()
+
+# Register the custom plugin
+refusal_classifier_manager.register_plugin("custom", CustomRefusalClassifier(custom_phrases=["I can't", "I won't"]))
+```
+
+## Using the Plugin System
+
+The `refusal_heuristic` function automatically uses all registered plugins to check for refusals:
+
+```python
+from agentic_security.probe_actor.refusal import refusal_heuristic
+
+is_refusal = refusal_heuristic(request_json)
+```
+
+## Conclusion
+
+The refusal classifier plugin system provides a flexible and extensible way to add custom refusal detection logic to the Agentic Security project. This documentation serves as a guide to creating, registering, and using custom refusal classifier plugins.

docs/rl_model.md

@@ -0,0 +1,194 @@
+# RL Model Module
+
+The RL Model module provides reinforcement learning-based prompt selection strategies for the probe system.
+
+## Overview
+
+The module implements several prompt selection strategies that use reinforcement learning techniques to optimize prompt selection based on guard results and rewards.
+
+## Classes
+
+### PromptSelectionInterface
+
+Abstract base class defining the interface for prompt selection strategies.
+
+**Methods:**
+
+- `select_next_prompt(current_prompt: str, passed_guard: bool) -> str`
+- `select_next_prompts(current_prompt: str, passed_guard: bool) -> list[str]`
+- `update_rewards(previous_prompt: str, current_prompt: str, reward: float, passed_guard: bool) -> None`
+
+### RandomPromptSelector
+
+Basic random selection strategy with cycle prevention using history.
+
+**Configuration:**
+
+- `prompts`: List of available prompts
+- `history_size`: Size of history buffer to prevent cycles (default: 300)
+
+### CloudRLPromptSelector
+
+Cloud-based reinforcement learning prompt selector with fallback to random selection.
+
+**Configuration:**
+
+- `prompts`: List of available prompts
+- `api_url`: URL of the RL service
+- `auth_token`: Authentication token (default: AS_TOKEN environment variable)
+- `history_size`: Size of history buffer (default: 300)
+- `timeout`: Request timeout in seconds (default: 5)
+- `run_id`: Unique identifier for the run
+
+### QLearningPromptSelector
+
+Q-Learning based prompt selector with exploration/exploitation tradeoff.
+
+**Configuration:**
+
+- `prompts`: List of available prompts
+- `learning_rate`: Learning rate (default: 0.1)
+- `discount_factor`: Discount factor (default: 0.9)
+- `initial_exploration`: Initial exploration rate (default: 1.0)
+- `exploration_decay`: Exploration decay rate (default: 0.995)
+- `min_exploration`: Minimum exploration rate (default: 0.01)
+- `history_size`: Size of history buffer (default: 300)
+
+### Module
+
+Main class that implements the RL-based prompt selection functionality.
+
+**Configuration:**
+
+- `prompt_groups`: List of prompt groups
+- `tools_inbox`: asyncio.Queue for tool communication
+- `opts`: Additional options
+  - `max_prompts`: Maximum number of prompts to generate (default: 10)
+  - `batch_size`: Batch size for processing (default: 500)
+
+## Usage Example
+
+```python
+from agentic_security.probe_data.modules.rl_model import (
+    Module,
+    CloudRLPromptSelector,
+    QLearningPromptSelector
+)
+
+# Initialize with prompt groups
+prompt_groups = ["What is AI?", "Explain ML", "Describe RL"]
+module = Module(prompt_groups, asyncio.Queue())
+
+# Use the module
+async for prompt in module.apply():
+    print(f"Selected prompt: {prompt}")
+```
+
+## API Reference
+
+### PromptSelectionInterface
+
+```python
+class PromptSelectionInterface(ABC):
+    @abstractmethod
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        """Select next prompt based on current state and guard result."""
+
+    @abstractmethod
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        """Select next prompts based on current state and guard result."""
+
+    @abstractmethod
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        """Update internal rewards based on outcome of last selected prompt."""
+```
+
+### RandomPromptSelector
+
+```python
+class RandomPromptSelector(PromptSelectionInterface):
+    def __init__(self, prompts: list[str], history_size: int = 300):
+        """Initialize with prompts and history size."""
+
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        """Select next prompt randomly with cycle prevention."""
+
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        """No learning in random selection."""
+```
+
+### CloudRLPromptSelector
+
+```python
+class CloudRLPromptSelector(PromptSelectionInterface):
+    def __init__(
+        self,
+        prompts: list[str],
+        api_url: str,
+        auth_token: str = AUTH_TOKEN,
+        history_size: int = 300,
+        timeout: int = 5,
+        run_id: str = "",
+    ):
+        """Initialize with cloud RL configuration."""
+
+    def select_next_prompts(self, current_prompt: str, passed_guard: bool) -> list[str]:
+        """Select next prompts using cloud RL with fallback."""
+
+    def _fallback_selection(self) -> str:
+        """Fallback to random selection if cloud request fails."""
+```
+
+### QLearningPromptSelector
+
+```python
+class QLearningPromptSelector(PromptSelectionInterface):
+    def __init__(
+        self,
+        prompts: list[str],
+        learning_rate: float = 0.1,
+        discount_factor: float = 0.9,
+        initial_exploration: float = 1.0,
+        exploration_decay: float = 0.995,
+        min_exploration: float = 0.01,
+        history_size: int = 300,
+    ):
+        """Initialize Q-Learning configuration."""
+
+    def select_next_prompt(self, current_prompt: str, passed_guard: bool) -> str:
+        """Select next prompt using Q-Learning with exploration/exploitation."""
+
+    def update_rewards(
+        self,
+        previous_prompt: str,
+        current_prompt: str,
+        reward: float,
+        passed_guard: bool,
+    ) -> None:
+        """Update Q-values based on reward."""
+```
+
+### Module
+
+```python
+class Module:
+    def __init__(
+        self, prompt_groups: list[str], tools_inbox: asyncio.Queue, opts: dict = {}
+    ):
+        """Initialize module with prompt groups and configuration."""
+
+    async def apply(self):
+        """Apply the RL model to generate prompts."""
+```

docs/stenography.md

@@ -0,0 +1,153 @@
+# Stenography Functions
+
+The stenography module provides various text obfuscation and transformation techniques for security testing. This document explains its architecture and implementation.
+
+## Overview
+
+The module implements:
+
+1. Rotation ciphers (ROT13, ROT5)
+1. Base64 encoding
+1. Text manipulation functions
+1. Randomization techniques
+1. Character substitution methods
+
+## Core Functions
+
+### Rotation Ciphers
+
+```python
+def rot13(input_text):
+    """
+    Applies ROT13 cipher to input text
+    - Preserves case of letters
+    - Leaves non-alphabetic characters unchanged
+    """
+    # Implementation details...
+
+def rot5(input_text):
+    """
+    Applies ROT5 cipher to input text
+    - Rotates digits by 5 positions
+    - Leaves non-digit characters unchanged
+    """
+    # Implementation details...
+```
+
+### Encoding
+
+```python
+def base64_encode(data):
+    """
+    Encodes input data using Base64
+    - Handles both string and bytes input
+    - Returns UTF-8 encoded string
+    """
+    # Implementation details...
+```
+
+### Text Manipulation
+
+```python
+def mirror_words(text):
+    """
+    Reverses each word in the input text
+    - Preserves word order
+    - Maintains spaces between words
+    """
+    # Implementation details...
+
+def scramble_words(text):
+    """
+    Randomly scrambles middle letters of words
+    - Preserves first and last letters
+    - Handles words shorter than 4 characters
+    """
+    # Implementation details...
+```
+
+### Randomization
+
+```python
+def randomize_letter_case(text):
+    """
+    Randomly changes case of each character
+    - Independent case changes per character
+    - Preserves non-letter characters
+    """
+    # Implementation details...
+
+def insert_noise_characters(text, frequency=0.2):
+    """
+    Inserts random characters between existing ones
+    - Configurable insertion frequency
+    - Uses alphanumeric characters for noise
+    """
+    # Implementation details...
+```
+
+### Advanced Transformations
+
+```python
+def substitute_with_ascii(text):
+    """
+    Replaces characters with their ASCII codes
+    - Space-separated numeric values
+    - Preserves original character order
+    """
+    # Implementation details...
+
+def remove_vowels(text):
+    """
+    Removes all vowel characters from text
+    - Handles both lowercase and uppercase vowels
+    - Preserves non-vowel characters
+    """
+    # Implementation details...
+
+def zigzag_obfuscation(text):
+    """
+    Alternates character case in zigzag pattern
+    - Starts with uppercase
+    - Toggles case for each alphabetic character
+    """
+    # Implementation details...
+```
+
+## Usage Patterns
+
+1. **Text Obfuscation**:
+
+   ```python
+   obfuscated = zigzag_obfuscation(
+       scramble_words(
+           insert_noise_characters(text)
+       )
+   )
+   ```
+
+1. **Encoding**:
+
+   ```python
+   encoded = base64_encode(rot13(text))
+   ```
+
+1. **Randomization**:
+
+   ```python
+   randomized = randomize_letter_case(
+       remove_vowels(text)
+   )
+   ```
+
+## Configuration
+
+- **Noise Frequency**: Configurable in insert_noise_characters()
+- **Scrambling**: Automatic handling of word lengths
+- **Case Handling**: Preserved in rotation ciphers
+
+## Limitations
+
+- Primarily handles ASCII text
+- Limited to implemented transformation types
+- Randomization is not cryptographically secure

docs/stylesheets/extra.css

@@ -1,5 +1,79 @@
 :root {
-  --md-primary-fg-color: #073763;
-  --md-primary-fg-color--light: #073763;
-  --md-primary-fg-color--dark: #073763;
+  --md-primary-fg-color: #e92063;
+  --md-primary-fg-color--light: #e92063;
+  --md-primary-fg-color--dark: #e92063;
+}
+
+
+/* Revert hue value to that of pre mkdocs-material v9.4.0 */
+[data-md-color-scheme="slate"] {
+  --md-hue: 230;
+  --md-default-bg-color: hsla(230, 15%, 21%, 1);
+}
+
+.hide {
+  display: none;
+}
+
+.text-center {
+  text-align: center;
+}
+
+img.index-header {
+  width: 70%;
+  max-width: 500px;
+}
+
+.pydantic-pink {
+  color: #FF007F;
+}
+
+.team-blue {
+  color: #0072CE;
+}
+
+.secure-green {
+  color: #00A86B;
+}
+
+.shapes-orange {
+  color: #FF7F32;
+}
+
+.puzzle-purple {
+  color: #652D90;
+}
+
+.wheel-gray {
+  color: #6E6E6E;
+}
+
+.vertical-middle {
+  vertical-align: middle;
+}
+
+.text-emphasis {
+  font-size: 1rem;
+  font-weight: 300;
+  font-style: italic;
+}
+
+#version-warning {
+  min-height: 120px;
+  margin-bottom: 10px;
+}
+
+.mermaid {
+  text-align: center;
+}
+
+
+/* Hide the entire footer */
+.md-footer {
+  display: none;
+}
+
+/* OR, hide only the "Made with Material" credit */
+.md-footer__made-with {
+  display: none;
 }

mkdocs.yml

@@ -8,16 +8,54 @@ repo_name: msoedov/agentic_security
 copyright: Maintained by <a href="https://msoedov.github.io">Agentic Security Team</a>.
 
 nav:
-  - Home: index.md
+  - Adventure starts here:
+      - Overview: index.md
+      - Quickstart: quickstart.md
+      - Design: design.md
+      - Abstractions: abstractions.md
   - Features: probe_data.md
-  - Probe Actor: probe_actor.md
-  - Installation: installation.md
-  - Getting Started: getting_started.md
-  - Configuration: configuration.md
-  - Dataset Extension: datasets.md
-  - CI/CD Integration: ci_cd.md
-  - API Reference: api_reference.md
-  - Contributing: contributing.md
+  - Concepts:
+    - Probe Actor: probe_actor.md
+    - Refusal Actor: refusal_classifier_plugins.md
+    - Agent Spec: http_spec.md
+  - Setup:
+    - Installation: installation.md
+    - Getting Started: getting_started.md
+    - Configuration: configuration.md
+  - Advanced Topics:
+    - Dataset Extension: datasets.md
+    - External Modules: external_module.md
+    - CI/CD Integration: ci_cd.md
+    - Bayesian Optimization: optimizer.md
+    - Image Generation: image_generation.md
+    - Stenography Functions: stenography.md
+    - Reinforcement Learning Optimization: rl_model.md
+  - WIP:
+    - Agent Operator: operator.md
+  - Reference:
+    - API Reference: api_reference.md
+  # - Project:
+  #     - Setup: setup.md
+  #     - Version control: version_control.md
+  #     - Docker: docker.md
+  #     - Variables: variables.md
+  #     - Custom libraries: custom_libraries.md
+  #     - Database: database.md
+  #     - Credentials: credentials.md
+  #     - Code execution: code_execution.md
+  #     - Settings: settings.md
+  #     - Version upgrades: version_upgrades.md
+  # - Contributing:
+  #     - Overview: contributing_overview.md
+  #     - Dev environment: dev_environment.md
+  #     - Backend: backend.md
+  #     - Frontend: frontend.md
+  #     - Documentation: documentation.md
+  # - About:
+  #     - Code of conduct: code_of_conduct.md
+  #     - Usage statistics: usage_statistics.md
+  #     - FAQ: faq.md
+  #     - Changelog: changelog.md
 
 plugins:
   - search
@@ -25,21 +63,24 @@ plugins:
       handlers:
         python:
           paths: [agentic_security]
-  - mkdocs-jupyter
+
+
+footer:
+  links: []  # Removes the default footer credits
 
 theme:
   name: material
-  feature:
-    tabs: true
+  features:
+    - navigation.expand
   palette:
-    - media: "(prefers-color-scheme: light)"
+    - media: "(prefers-color-scheme: dark)"
       scheme: default
       primary: custom
       accent: deep orange
       toggle:
         icon: material/brightness-7
         name: Switch to dark mode
-    - media: "(prefers-color-scheme: dark)"
+    - media: "(prefers-color-scheme: light)"
       scheme: slate
       primary: custom
       accent: deep orange
@@ -48,8 +89,10 @@ theme:
         name: Switch to light mode
   icon:
     repo: fontawesome/brands/github
+  favicon: "https://res.cloudinary.com/dq0w2rtm9/image/upload/v1737555066/r17hrkre246doczwmvbv.png"
 
 extra:
+  generator: false
   social:
     - icon: fontawesome/brands/github
       link: https://github.com/msoedov/agentic_security

poetry.lock

@@ -1,4 +1,4 @@
-# This file is automatically @generated by Poetry 1.8.5 and should not be changed by hand.
+# This file is automatically @generated by Poetry 1.7.1 and should not be changed by hand.
 
 [[package]]
 name = "aiohappyeyeballs"
@@ -252,33 +252,33 @@ lxml = ["lxml"]
 
 [[package]]
 name = "black"
-version = "24.10.0"
+version = "25.1.0"
 description = "The uncompromising code formatter."
 optional = false
 python-versions = ">=3.9"
 files = [
-    {file = "black-24.10.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:e6668650ea4b685440857138e5fe40cde4d652633b1bdffc62933d0db4ed9812"},
-    {file = "black-24.10.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:1c536fcf674217e87b8cc3657b81809d3c085d7bf3ef262ead700da345bfa6ea"},
-    {file = "black-24.10.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:649fff99a20bd06c6f727d2a27f401331dc0cc861fb69cde910fe95b01b5928f"},
-    {file = "black-24.10.0-cp310-cp310-win_amd64.whl", hash = "sha256:fe4d6476887de70546212c99ac9bd803d90b42fc4767f058a0baa895013fbb3e"},
-    {file = "black-24.10.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:5a2221696a8224e335c28816a9d331a6c2ae15a2ee34ec857dcf3e45dbfa99ad"},
-    {file = "black-24.10.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:f9da3333530dbcecc1be13e69c250ed8dfa67f43c4005fb537bb426e19200d50"},
-    {file = "black-24.10.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:4007b1393d902b48b36958a216c20c4482f601569d19ed1df294a496eb366392"},
-    {file = "black-24.10.0-cp311-cp311-win_amd64.whl", hash = "sha256:394d4ddc64782e51153eadcaaca95144ac4c35e27ef9b0a42e121ae7e57a9175"},
-    {file = "black-24.10.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:b5e39e0fae001df40f95bd8cc36b9165c5e2ea88900167bddf258bacef9bbdc3"},
-    {file = "black-24.10.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:d37d422772111794b26757c5b55a3eade028aa3fde43121ab7b673d050949d65"},
-    {file = "black-24.10.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:14b3502784f09ce2443830e3133dacf2c0110d45191ed470ecb04d0f5f6fcb0f"},
-    {file = "black-24.10.0-cp312-cp312-win_amd64.whl", hash = "sha256:30d2c30dc5139211dda799758559d1b049f7f14c580c409d6ad925b74a4208a8"},
-    {file = "black-24.10.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:1cbacacb19e922a1d75ef2b6ccaefcd6e93a2c05ede32f06a21386a04cedb981"},
-    {file = "black-24.10.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:1f93102e0c5bb3907451063e08b9876dbeac810e7da5a8bfb7aeb5a9ef89066b"},
-    {file = "black-24.10.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ddacb691cdcdf77b96f549cf9591701d8db36b2f19519373d60d31746068dbf2"},
-    {file = "black-24.10.0-cp313-cp313-win_amd64.whl", hash = "sha256:680359d932801c76d2e9c9068d05c6b107f2584b2a5b88831c83962eb9984c1b"},
-    {file = "black-24.10.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:17374989640fbca88b6a448129cd1745c5eb8d9547b464f281b251dd00155ccd"},
-    {file = "black-24.10.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:63f626344343083322233f175aaf372d326de8436f5928c042639a4afbbf1d3f"},
-    {file = "black-24.10.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ccfa1d0cb6200857f1923b602f978386a3a2758a65b52e0950299ea014be6800"},
-    {file = "black-24.10.0-cp39-cp39-win_amd64.whl", hash = "sha256:2cd9c95431d94adc56600710f8813ee27eea544dd118d45896bb734e9d7a0dc7"},
-    {file = "black-24.10.0-py3-none-any.whl", hash = "sha256:3bb2b7a1f7b685f85b11fed1ef10f8a9148bceb49853e47a294a3dd963c1dd7d"},
-    {file = "black-24.10.0.tar.gz", hash = "sha256:846ea64c97afe3bc677b761787993be4991810ecc7a4a937816dd6bddedc4875"},
+    {file = "black-25.1.0-cp310-cp310-macosx_10_9_x86_64.whl", hash = "sha256:759e7ec1e050a15f89b770cefbf91ebee8917aac5c20483bc2d80a6c3a04df32"},
+    {file = "black-25.1.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:0e519ecf93120f34243e6b0054db49c00a35f84f195d5bce7e9f5cfc578fc2da"},
+    {file = "black-25.1.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:055e59b198df7ac0b7efca5ad7ff2516bca343276c466be72eb04a3bcc1f82d7"},
+    {file = "black-25.1.0-cp310-cp310-win_amd64.whl", hash = "sha256:db8ea9917d6f8fc62abd90d944920d95e73c83a5ee3383493e35d271aca872e9"},
+    {file = "black-25.1.0-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:a39337598244de4bae26475f77dda852ea00a93bd4c728e09eacd827ec929df0"},
+    {file = "black-25.1.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:96c1c7cd856bba8e20094e36e0f948718dc688dba4a9d78c3adde52b9e6c2299"},
+    {file = "black-25.1.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:bce2e264d59c91e52d8000d507eb20a9aca4a778731a08cfff7e5ac4a4bb7096"},
+    {file = "black-25.1.0-cp311-cp311-win_amd64.whl", hash = "sha256:172b1dbff09f86ce6f4eb8edf9dede08b1fce58ba194c87d7a4f1a5aa2f5b3c2"},
+    {file = "black-25.1.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:4b60580e829091e6f9238c848ea6750efed72140b91b048770b64e74fe04908b"},
+    {file = "black-25.1.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:1e2978f6df243b155ef5fa7e558a43037c3079093ed5d10fd84c43900f2d8ecc"},
+    {file = "black-25.1.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:3b48735872ec535027d979e8dcb20bf4f70b5ac75a8ea99f127c106a7d7aba9f"},
+    {file = "black-25.1.0-cp312-cp312-win_amd64.whl", hash = "sha256:ea0213189960bda9cf99be5b8c8ce66bb054af5e9e861249cd23471bd7b0b3ba"},
+    {file = "black-25.1.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:8f0b18a02996a836cc9c9c78e5babec10930862827b1b724ddfe98ccf2f2fe4f"},
+    {file = "black-25.1.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:afebb7098bfbc70037a053b91ae8437c3857482d3a690fefc03e9ff7aa9a5fd3"},
+    {file = "black-25.1.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:030b9759066a4ee5e5aca28c3c77f9c64789cdd4de8ac1df642c40b708be6171"},
+    {file = "black-25.1.0-cp313-cp313-win_amd64.whl", hash = "sha256:a22f402b410566e2d1c950708c77ebf5ebd5d0d88a6a2e87c86d9fb48afa0d18"},
+    {file = "black-25.1.0-cp39-cp39-macosx_10_9_x86_64.whl", hash = "sha256:a1ee0a0c330f7b5130ce0caed9936a904793576ef4d2b98c40835d6a65afa6a0"},
+    {file = "black-25.1.0-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:f3df5f1bf91d36002b0a75389ca8663510cf0531cca8aa5c1ef695b46d98655f"},
+    {file = "black-25.1.0-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d9e6827d563a2c820772b32ce8a42828dc6790f095f441beef18f96aa6f8294e"},
+    {file = "black-25.1.0-cp39-cp39-win_amd64.whl", hash = "sha256:bacabb307dca5ebaf9c118d2d2f6903da0d62c9faa82bd21a33eecc319559355"},
+    {file = "black-25.1.0-py3-none-any.whl", hash = "sha256:95e8176dae143ba9097f351d174fdaf0ccd29efb414b362ae3fd72bf0f710717"},
+    {file = "black-25.1.0.tar.gz", hash = "sha256:33496d5cd1222ad73391352b4ae8da15253c5de89b93a80b3e2c8d9a19ec2666"},
 ]
 
 [package.dependencies]
@@ -645,22 +645,22 @@ tests = ["pytest", "pytest-cov", "pytest-xdist"]
 
 [[package]]
 name = "datasets"
-version = "3.0.1"
+version = "3.3.0"
 description = "HuggingFace community-driven open-source library of datasets"
 optional = false
-python-versions = ">=3.8.0"
+python-versions = ">=3.9.0"
 files = [
-    {file = "datasets-3.0.1-py3-none-any.whl", hash = "sha256:db080aab41c8cc68645117a0f172e5c6789cbc672f066de0aa5a08fc3eebc686"},
-    {file = "datasets-3.0.1.tar.gz", hash = "sha256:40d63b09e76a3066c32e746d6fdc36fd3f29ed2acd49bf5b1a2100da32936511"},
+    {file = "datasets-3.3.0-py3-none-any.whl", hash = "sha256:22312d09626f8fc3aa0a237b0c164997f5903bddd4c4c9e27dbaf563754c681b"},
+    {file = "datasets-3.3.0.tar.gz", hash = "sha256:54c607b06f6eaa1572e21e200d2870d89d50e3bcc622dc2021a53a6ce4f684c2"},
 ]
 
 [package.dependencies]
 aiohttp = "*"
 dill = ">=0.3.0,<0.3.9"
 filelock = "*"
-fsspec = {version = ">=2023.1.0,<=2024.6.1", extras = ["http"]}
-huggingface-hub = ">=0.22.0"
-multiprocess = "*"
+fsspec = {version = ">=2023.1.0,<=2024.12.0", extras = ["http"]}
+huggingface-hub = ">=0.24.0"
+multiprocess = "<0.70.17"
 numpy = ">=1.17"
 packaging = "*"
 pandas = "*"
@@ -673,15 +673,15 @@ xxhash = "*"
 [package.extras]
 audio = ["librosa", "soundfile (>=0.12.1)", "soxr (>=0.4.0)"]
 benchmarks = ["tensorflow (==2.12.0)", "torch (==2.0.1)", "transformers (==4.30.1)"]
-dev = ["Pillow (>=9.4.0)", "absl-py", "decorator", "elasticsearch (<8.0.0)", "faiss-cpu (>=1.8.0.post1)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "librosa", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "ruff (>=0.3.0)", "s3fs", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tensorflow (>=2.16.0)", "tensorflow (>=2.6.0)", "tensorflow (>=2.6.0)", "tiktoken", "torch", "torch (>=2.0.0)", "torchdata", "transformers", "transformers (>=4.42.0)", "zstandard"]
+dev = ["Pillow (>=9.4.0)", "absl-py", "decorator", "decord (==0.6.0)", "elasticsearch (>=7.17.12,<8.0.0)", "faiss-cpu (>=1.8.0.post1)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "librosa", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "ruff (>=0.3.0)", "s3fs", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tensorflow (>=2.16.0)", "tensorflow (>=2.6.0)", "tensorflow (>=2.6.0)", "tiktoken", "torch", "torch (>=2.0.0)", "torchdata", "transformers", "transformers (>=4.42.0)", "zstandard"]
 docs = ["s3fs", "tensorflow (>=2.6.0)", "torch", "transformers"]
 jax = ["jax (>=0.3.14)", "jaxlib (>=0.3.14)"]
 quality = ["ruff (>=0.3.0)"]
 s3 = ["s3fs"]
 tensorflow = ["tensorflow (>=2.6.0)"]
 tensorflow-gpu = ["tensorflow (>=2.6.0)"]
-tests = ["Pillow (>=9.4.0)", "absl-py", "decorator", "elasticsearch (<8.0.0)", "faiss-cpu (>=1.8.0.post1)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "librosa", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tensorflow (>=2.16.0)", "tensorflow (>=2.6.0)", "tiktoken", "torch (>=2.0.0)", "torchdata", "transformers (>=4.42.0)", "zstandard"]
-tests-numpy2 = ["Pillow (>=9.4.0)", "absl-py", "decorator", "elasticsearch (<8.0.0)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tiktoken", "torch (>=2.0.0)", "torchdata", "transformers (>=4.42.0)", "zstandard"]
+tests = ["Pillow (>=9.4.0)", "absl-py", "decorator", "decord (==0.6.0)", "elasticsearch (>=7.17.12,<8.0.0)", "faiss-cpu (>=1.8.0.post1)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "librosa", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tensorflow (>=2.16.0)", "tensorflow (>=2.6.0)", "tiktoken", "torch (>=2.0.0)", "torchdata", "transformers (>=4.42.0)", "zstandard"]
+tests-numpy2 = ["Pillow (>=9.4.0)", "absl-py", "decorator", "decord (==0.6.0)", "elasticsearch (>=7.17.12,<8.0.0)", "jax (>=0.3.14)", "jaxlib (>=0.3.14)", "joblib (<1.3.0)", "joblibspark", "lz4", "moto[server]", "polars[timezone] (>=0.20.0)", "protobuf (<4.0.0)", "py7zr", "pyspark (>=3.4)", "pytest", "pytest-datadir", "pytest-xdist", "rarfile (>=4.0)", "s3fs (>=2021.11.1)", "soundfile (>=0.12.1)", "soundfile (>=0.12.1)", "soxr (>=0.4.0)", "sqlalchemy", "tiktoken", "torch (>=2.0.0)", "torchdata", "transformers (>=4.42.0)", "zstandard"]
 torch = ["torch"]
 vision = ["Pillow (>=9.4.0)"]
 
@@ -770,13 +770,13 @@ files = [
 
 [[package]]
 name = "executing"
-version = "2.1.0"
+version = "2.2.0"
 description = "Get the currently executing AST node of a frame, and other information"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "executing-2.1.0-py2.py3-none-any.whl", hash = "sha256:8d63781349375b5ebccc3142f4b30350c0cd9c79f921cde38be2be4637e98eaf"},
-    {file = "executing-2.1.0.tar.gz", hash = "sha256:8ea27ddd260da8150fa5a708269c4a10e76161e2496ec3e587da9e3c0fe4b9ab"},
+    {file = "executing-2.2.0-py2.py3-none-any.whl", hash = "sha256:11387150cad388d62750327a53d3339fad4888b39a6fe233c3afbb54ecffd3aa"},
+    {file = "executing-2.2.0.tar.gz", hash = "sha256:5d108c028108fe2551d1a7b2e8b713341e2cb4fc0aa7dcf966fa4327a5226755"},
 ]
 
 [package.extras]
@@ -784,23 +784,23 @@ tests = ["asttokens (>=2.1.0)", "coverage", "coverage-enable-subprocess", "ipyth
 
 [[package]]
 name = "fastapi"
-version = "0.115.6"
+version = "0.115.8"
 description = "FastAPI framework, high performance, easy to learn, fast to code, ready for production"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "fastapi-0.115.6-py3-none-any.whl", hash = "sha256:e9240b29e36fa8f4bb7290316988e90c381e5092e0cbe84e7818cc3713bcf305"},
-    {file = "fastapi-0.115.6.tar.gz", hash = "sha256:9ec46f7addc14ea472958a96aae5b5de65f39721a46aaf5705c480d9a8b76654"},
+    {file = "fastapi-0.115.8-py3-none-any.whl", hash = "sha256:753a96dd7e036b34eeef8babdfcfe3f28ff79648f86551eb36bfc1b0bf4a8cbf"},
+    {file = "fastapi-0.115.8.tar.gz", hash = "sha256:0ce9111231720190473e222cdf0f07f7206ad7e53ea02beb1d2dc36e2f0741e9"},
 ]
 
 [package.dependencies]
 pydantic = ">=1.7.4,<1.8 || >1.8,<1.8.1 || >1.8.1,<2.0.0 || >2.0.0,<2.0.1 || >2.0.1,<2.1.0 || >2.1.0,<3.0.0"
-starlette = ">=0.40.0,<0.42.0"
+starlette = ">=0.40.0,<0.46.0"
 typing-extensions = ">=4.8.0"
 
 [package.extras]
-all = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.5)", "httpx (>=0.23.0)", "itsdangerous (>=1.1.0)", "jinja2 (>=2.11.2)", "orjson (>=3.2.1)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.7)", "pyyaml (>=5.3.1)", "ujson (>=4.0.1,!=4.0.2,!=4.1.0,!=4.2.0,!=4.3.0,!=5.0.0,!=5.1.0)", "uvicorn[standard] (>=0.12.0)"]
-standard = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.5)", "httpx (>=0.23.0)", "jinja2 (>=2.11.2)", "python-multipart (>=0.0.7)", "uvicorn[standard] (>=0.12.0)"]
+all = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.5)", "httpx (>=0.23.0)", "itsdangerous (>=1.1.0)", "jinja2 (>=3.1.5)", "orjson (>=3.2.1)", "pydantic-extra-types (>=2.0.0)", "pydantic-settings (>=2.0.0)", "python-multipart (>=0.0.18)", "pyyaml (>=5.3.1)", "ujson (>=4.0.1,!=4.0.2,!=4.1.0,!=4.2.0,!=4.3.0,!=5.0.0,!=5.1.0)", "uvicorn[standard] (>=0.12.0)"]
+standard = ["email-validator (>=2.0.0)", "fastapi-cli[standard] (>=0.0.5)", "httpx (>=0.23.0)", "jinja2 (>=3.1.5)", "python-multipart (>=0.0.18)", "uvicorn[standard] (>=0.12.0)"]
 
 [[package]]
 name = "fastjsonschema"
@@ -1055,6 +1055,25 @@ python-dateutil = ">=2.8.1"
 [package.extras]
 dev = ["flake8", "markdown", "twine", "wheel"]
 
+[[package]]
+name = "gtts"
+version = "2.5.4"
+description = "gTTS (Google Text-to-Speech), a Python library and CLI tool to interface with Google Translate text-to-speech API"
+optional = false
+python-versions = ">=3.7"
+files = [
+    {file = "gTTS-2.5.4-py3-none-any.whl", hash = "sha256:5dd579377f9f5546893bc26315ab1f846933dc27a054764b168f141065ca8436"},
+    {file = "gtts-2.5.4.tar.gz", hash = "sha256:f5737b585f6442f677dbe8773424fd50697c75bdf3e36443585e30a8d48c1884"},
+]
+
+[package.dependencies]
+click = ">=7.1,<8.2"
+requests = ">=2.27,<3"
+
+[package.extras]
+docs = ["sphinx", "sphinx-autobuild", "sphinx-click", "sphinx-mdinclude", "sphinx-rtd-theme"]
+tests = ["pytest (>=7.1.3,<8.4.0)", "pytest-cov", "testfixtures"]
+
 [[package]]
 name = "h11"
 version = "0.14.0"
@@ -1113,13 +1132,13 @@ zstd = ["zstandard (>=0.18.0)"]
 
 [[package]]
 name = "huggingface-hub"
-version = "0.25.1"
+version = "0.28.1"
 description = "Client library to download and publish models, datasets and other repos on the huggingface.co hub"
 optional = false
 python-versions = ">=3.8.0"
 files = [
-    {file = "huggingface_hub-0.25.1-py3-none-any.whl", hash = "sha256:a5158ded931b3188f54ea9028097312cb0acd50bffaaa2612014c3c526b44972"},
-    {file = "huggingface_hub-0.25.1.tar.gz", hash = "sha256:9ff7cb327343211fbd06e2b149b8f362fd1e389454f3f14c6db75a4999ee20ff"},
+    {file = "huggingface_hub-0.28.1-py3-none-any.whl", hash = "sha256:aa6b9a3ffdae939b72c464dbb0d7f99f56e649b55c3d52406f49e0a5a620c0a7"},
+    {file = "huggingface_hub-0.28.1.tar.gz", hash = "sha256:893471090c98e3b6efbdfdacafe4052b20b84d59866fb6f54c33d9af18c303ae"},
 ]
 
 [package.dependencies]
@@ -1132,16 +1151,16 @@ tqdm = ">=4.42.1"
 typing-extensions = ">=3.7.4.3"
 
 [package.extras]
-all = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio", "jedi", "minijinja (>=1.0)", "mypy (==1.5.1)", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "ruff (>=0.5.0)", "soundfile", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
+all = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio (>=4.0.0)", "jedi", "libcst (==1.4.0)", "mypy (==1.5.1)", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "ruff (>=0.9.0)", "soundfile", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
 cli = ["InquirerPy (==0.3.4)"]
-dev = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio", "jedi", "minijinja (>=1.0)", "mypy (==1.5.1)", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "ruff (>=0.5.0)", "soundfile", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
+dev = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio (>=4.0.0)", "jedi", "libcst (==1.4.0)", "mypy (==1.5.1)", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "ruff (>=0.9.0)", "soundfile", "types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)", "urllib3 (<2.0)"]
 fastai = ["fastai (>=2.4)", "fastcore (>=1.3.27)", "toml"]
 hf-transfer = ["hf-transfer (>=0.1.4)"]
-inference = ["aiohttp", "minijinja (>=1.0)"]
-quality = ["mypy (==1.5.1)", "ruff (>=0.5.0)"]
+inference = ["aiohttp"]
+quality = ["libcst (==1.4.0)", "mypy (==1.5.1)", "ruff (>=0.9.0)"]
 tensorflow = ["graphviz", "pydot", "tensorflow"]
 tensorflow-testing = ["keras (<3.0)", "tensorflow"]
-testing = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio", "jedi", "minijinja (>=1.0)", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "soundfile", "urllib3 (<2.0)"]
+testing = ["InquirerPy (==0.3.4)", "Jinja2", "Pillow", "aiohttp", "fastapi", "gradio (>=4.0.0)", "jedi", "numpy", "pytest (>=8.1.1,<8.2.2)", "pytest-asyncio", "pytest-cov", "pytest-env", "pytest-mock", "pytest-rerunfailures", "pytest-vcr", "pytest-xdist", "soundfile", "urllib3 (<2.0)"]
 torch = ["safetensors[torch]", "torch"]
 typing = ["types-PyYAML", "types-requests", "types-simplejson", "types-toml", "types-tqdm", "types-urllib3", "typing-extensions (>=4.8.0)"]
 
@@ -1183,22 +1202,23 @@ files = [
 
 [[package]]
 name = "inline-snapshot"
-version = "0.18.1"
+version = "0.20.1"
 description = "golden master/snapshot/approval testing library which puts the values right into your source code"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "inline_snapshot-0.18.1-py3-none-any.whl", hash = "sha256:6751b95c10ae940879bec429c160d58c376ad449fcf6e213db7735d91ae40209"},
-    {file = "inline_snapshot-0.18.1.tar.gz", hash = "sha256:a93bcf0aec68edf27323fa8dfc902354ef9a730050ab8939a4b174419828db6b"},
+    {file = "inline_snapshot-0.20.1-py3-none-any.whl", hash = "sha256:5b5c3fd037f340dff5adee1c2c58db9038325937a8190dedbba98e37b87c979a"},
+    {file = "inline_snapshot-0.20.1.tar.gz", hash = "sha256:c56c871e59973500eca00610022eac19e79cd2c1b0b2d7a18abe14dde11a1431"},
 ]
 
 [package.dependencies]
 asttokens = ">=2.0.5"
-black = ">=23.3.0"
-click = ">=8.1.4"
-executing = ">=2.1.0"
+executing = ">=2.2.0"
 rich = ">=13.7.1"
-typing-extensions = "*"
+
+[package.extras]
+black = ["black (>=23.3.0)", "click (>=8.1.4)"]
+dirty-equals = ["dirty-equals (>=0.9.0)"]
 
 [[package]]
 name = "ipykernel"
@@ -1675,51 +1695,45 @@ files = [
 
 [[package]]
 name = "matplotlib"
-version = "3.9.2"
+version = "3.10.0"
 description = "Python plotting package"
 optional = false
-python-versions = ">=3.9"
+python-versions = ">=3.10"
 files = [
-    {file = "matplotlib-3.9.2-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:9d78bbc0cbc891ad55b4f39a48c22182e9bdaea7fc0e5dbd364f49f729ca1bbb"},
-    {file = "matplotlib-3.9.2-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:c375cc72229614632c87355366bdf2570c2dac01ac66b8ad048d2dabadf2d0d4"},
-    {file = "matplotlib-3.9.2-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:1d94ff717eb2bd0b58fe66380bd8b14ac35f48a98e7c6765117fe67fb7684e64"},
-    {file = "matplotlib-3.9.2-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ab68d50c06938ef28681073327795c5db99bb4666214d2d5f880ed11aeaded66"},
-    {file = "matplotlib-3.9.2-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:65aacf95b62272d568044531e41de26285d54aec8cb859031f511f84bd8b495a"},
-    {file = "matplotlib-3.9.2-cp310-cp310-win_amd64.whl", hash = "sha256:3fd595f34aa8a55b7fc8bf9ebea8aa665a84c82d275190a61118d33fbc82ccae"},
-    {file = "matplotlib-3.9.2-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:d8dd059447824eec055e829258ab092b56bb0579fc3164fa09c64f3acd478772"},
-    {file = "matplotlib-3.9.2-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:c797dac8bb9c7a3fd3382b16fe8f215b4cf0f22adccea36f1545a6d7be310b41"},
-    {file = "matplotlib-3.9.2-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:d719465db13267bcef19ea8954a971db03b9f48b4647e3860e4bc8e6ed86610f"},
-    {file = "matplotlib-3.9.2-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:8912ef7c2362f7193b5819d17dae8629b34a95c58603d781329712ada83f9447"},
-    {file = "matplotlib-3.9.2-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:7741f26a58a240f43bee74965c4882b6c93df3e7eb3de160126d8c8f53a6ae6e"},
-    {file = "matplotlib-3.9.2-cp311-cp311-win_amd64.whl", hash = "sha256:ae82a14dab96fbfad7965403c643cafe6515e386de723e498cf3eeb1e0b70cc7"},
-    {file = "matplotlib-3.9.2-cp312-cp312-macosx_10_12_x86_64.whl", hash = "sha256:ac43031375a65c3196bee99f6001e7fa5bdfb00ddf43379d3c0609bdca042df9"},
-    {file = "matplotlib-3.9.2-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:be0fc24a5e4531ae4d8e858a1a548c1fe33b176bb13eff7f9d0d38ce5112a27d"},
-    {file = "matplotlib-3.9.2-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:bf81de2926c2db243c9b2cbc3917619a0fc85796c6ba4e58f541df814bbf83c7"},
-    {file = "matplotlib-3.9.2-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f6ee45bc4245533111ced13f1f2cace1e7f89d1c793390392a80c139d6cf0e6c"},
-    {file = "matplotlib-3.9.2-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:306c8dfc73239f0e72ac50e5a9cf19cc4e8e331dd0c54f5e69ca8758550f1e1e"},
-    {file = "matplotlib-3.9.2-cp312-cp312-win_amd64.whl", hash = "sha256:5413401594cfaff0052f9d8b1aafc6d305b4bd7c4331dccd18f561ff7e1d3bd3"},
-    {file = "matplotlib-3.9.2-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:18128cc08f0d3cfff10b76baa2f296fc28c4607368a8402de61bb3f2eb33c7d9"},
-    {file = "matplotlib-3.9.2-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:4876d7d40219e8ae8bb70f9263bcbe5714415acfdf781086601211335e24f8aa"},
-    {file = "matplotlib-3.9.2-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:6d9f07a80deab4bb0b82858a9e9ad53d1382fd122be8cde11080f4e7dfedb38b"},
-    {file = "matplotlib-3.9.2-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:f7c0410f181a531ec4e93bbc27692f2c71a15c2da16766f5ba9761e7ae518413"},
-    {file = "matplotlib-3.9.2-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:909645cce2dc28b735674ce0931a4ac94e12f5b13f6bb0b5a5e65e7cea2c192b"},
-    {file = "matplotlib-3.9.2-cp313-cp313-win_amd64.whl", hash = "sha256:f32c7410c7f246838a77d6d1eff0c0f87f3cb0e7c4247aebea71a6d5a68cab49"},
-    {file = "matplotlib-3.9.2-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:37e51dd1c2db16ede9cfd7b5cabdfc818b2c6397c83f8b10e0e797501c963a03"},
-    {file = "matplotlib-3.9.2-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:b82c5045cebcecd8496a4d694d43f9cc84aeeb49fe2133e036b207abe73f4d30"},
-    {file = "matplotlib-3.9.2-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:f053c40f94bc51bc03832a41b4f153d83f2062d88c72b5e79997072594e97e51"},
-    {file = "matplotlib-3.9.2-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:dbe196377a8248972f5cede786d4c5508ed5f5ca4a1e09b44bda889958b33f8c"},
-    {file = "matplotlib-3.9.2-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:5816b1e1fe8c192cbc013f8f3e3368ac56fbecf02fb41b8f8559303f24c5015e"},
-    {file = "matplotlib-3.9.2-cp39-cp39-macosx_10_12_x86_64.whl", hash = "sha256:cef2a73d06601437be399908cf13aee74e86932a5ccc6ccdf173408ebc5f6bb2"},
-    {file = "matplotlib-3.9.2-cp39-cp39-macosx_11_0_arm64.whl", hash = "sha256:e0830e188029c14e891fadd99702fd90d317df294c3298aad682739c5533721a"},
-    {file = "matplotlib-3.9.2-cp39-cp39-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:03ba9c1299c920964e8d3857ba27173b4dbb51ca4bab47ffc2c2ba0eb5e2cbc5"},
-    {file = "matplotlib-3.9.2-cp39-cp39-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:1cd93b91ab47a3616b4d3c42b52f8363b88ca021e340804c6ab2536344fad9ca"},
-    {file = "matplotlib-3.9.2-cp39-cp39-musllinux_1_2_x86_64.whl", hash = "sha256:6d1ce5ed2aefcdce11904fc5bbea7d9c21fff3d5f543841edf3dea84451a09ea"},
-    {file = "matplotlib-3.9.2-cp39-cp39-win_amd64.whl", hash = "sha256:b2696efdc08648536efd4e1601b5fd491fd47f4db97a5fbfd175549a7365c1b2"},
-    {file = "matplotlib-3.9.2-pp39-pypy39_pp73-macosx_10_15_x86_64.whl", hash = "sha256:d52a3b618cb1cbb769ce2ee1dcdb333c3ab6e823944e9a2d36e37253815f9556"},
-    {file = "matplotlib-3.9.2-pp39-pypy39_pp73-macosx_11_0_arm64.whl", hash = "sha256:039082812cacd6c6bec8e17a9c1e6baca230d4116d522e81e1f63a74d01d2e21"},
-    {file = "matplotlib-3.9.2-pp39-pypy39_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6758baae2ed64f2331d4fd19be38b7b4eae3ecec210049a26b6a4f3ae1c85dcc"},
-    {file = "matplotlib-3.9.2-pp39-pypy39_pp73-win_amd64.whl", hash = "sha256:050598c2b29e0b9832cde72bcf97627bf00262adbc4a54e2b856426bb2ef0697"},
-    {file = "matplotlib-3.9.2.tar.gz", hash = "sha256:96ab43906269ca64a6366934106fa01534454a69e471b7bf3d79083981aaab92"},
+    {file = "matplotlib-3.10.0-cp310-cp310-macosx_10_12_x86_64.whl", hash = "sha256:2c5829a5a1dd5a71f0e31e6e8bb449bc0ee9dbfb05ad28fc0c6b55101b3a4be6"},
+    {file = "matplotlib-3.10.0-cp310-cp310-macosx_11_0_arm64.whl", hash = "sha256:a2a43cbefe22d653ab34bb55d42384ed30f611bcbdea1f8d7f431011a2e1c62e"},
+    {file = "matplotlib-3.10.0-cp310-cp310-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:607b16c8a73943df110f99ee2e940b8a1cbf9714b65307c040d422558397dac5"},
+    {file = "matplotlib-3.10.0-cp310-cp310-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:01d2b19f13aeec2e759414d3bfe19ddfb16b13a1250add08d46d5ff6f9be83c6"},
+    {file = "matplotlib-3.10.0-cp310-cp310-musllinux_1_2_x86_64.whl", hash = "sha256:5e6c6461e1fc63df30bf6f80f0b93f5b6784299f721bc28530477acd51bfc3d1"},
+    {file = "matplotlib-3.10.0-cp310-cp310-win_amd64.whl", hash = "sha256:994c07b9d9fe8d25951e3202a68c17900679274dadfc1248738dcfa1bd40d7f3"},
+    {file = "matplotlib-3.10.0-cp311-cp311-macosx_10_12_x86_64.whl", hash = "sha256:fd44fc75522f58612ec4a33958a7e5552562b7705b42ef1b4f8c0818e304a363"},
+    {file = "matplotlib-3.10.0-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:c58a9622d5dbeb668f407f35f4e6bfac34bb9ecdcc81680c04d0258169747997"},
+    {file = "matplotlib-3.10.0-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:845d96568ec873be63f25fa80e9e7fae4be854a66a7e2f0c8ccc99e94a8bd4ef"},
+    {file = "matplotlib-3.10.0-cp311-cp311-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:5439f4c5a3e2e8eab18e2f8c3ef929772fd5641876db71f08127eed95ab64683"},
+    {file = "matplotlib-3.10.0-cp311-cp311-musllinux_1_2_x86_64.whl", hash = "sha256:4673ff67a36152c48ddeaf1135e74ce0d4bce1bbf836ae40ed39c29edf7e2765"},
+    {file = "matplotlib-3.10.0-cp311-cp311-win_amd64.whl", hash = "sha256:7e8632baebb058555ac0cde75db885c61f1212e47723d63921879806b40bec6a"},
+    {file = "matplotlib-3.10.0-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:4659665bc7c9b58f8c00317c3c2a299f7f258eeae5a5d56b4c64226fca2f7c59"},
+    {file = "matplotlib-3.10.0-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:d44cb942af1693cced2604c33a9abcef6205601c445f6d0dc531d813af8a2f5a"},
+    {file = "matplotlib-3.10.0-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a994f29e968ca002b50982b27168addfd65f0105610b6be7fa515ca4b5307c95"},
+    {file = "matplotlib-3.10.0-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:9b0558bae37f154fffda54d779a592bc97ca8b4701f1c710055b609a3bac44c8"},
+    {file = "matplotlib-3.10.0-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:503feb23bd8c8acc75541548a1d709c059b7184cde26314896e10a9f14df5f12"},
+    {file = "matplotlib-3.10.0-cp312-cp312-win_amd64.whl", hash = "sha256:c40ba2eb08b3f5de88152c2333c58cee7edcead0a2a0d60fcafa116b17117adc"},
+    {file = "matplotlib-3.10.0-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:96f2886f5c1e466f21cc41b70c5a0cd47bfa0015eb2d5793c88ebce658600e25"},
+    {file = "matplotlib-3.10.0-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:12eaf48463b472c3c0f8dbacdbf906e573013df81a0ab82f0616ea4b11281908"},
+    {file = "matplotlib-3.10.0-cp313-cp313-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2fbbabc82fde51391c4da5006f965e36d86d95f6ee83fb594b279564a4c5d0d2"},
+    {file = "matplotlib-3.10.0-cp313-cp313-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ad2e15300530c1a94c63cfa546e3b7864bd18ea2901317bae8bbf06a5ade6dcf"},
+    {file = "matplotlib-3.10.0-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:3547d153d70233a8496859097ef0312212e2689cdf8d7ed764441c77604095ae"},
+    {file = "matplotlib-3.10.0-cp313-cp313-win_amd64.whl", hash = "sha256:c55b20591ced744aa04e8c3e4b7543ea4d650b6c3c4b208c08a05b4010e8b442"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:9ade1003376731a971e398cc4ef38bb83ee8caf0aee46ac6daa4b0506db1fd06"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:95b710fea129c76d30be72c3b38f330269363fbc6e570a5dd43580487380b5ff"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:5cdbaf909887373c3e094b0318d7ff230b2ad9dcb64da7ade654182872ab2593"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:d907fddb39f923d011875452ff1eca29a9e7f21722b873e90db32e5d8ddff12e"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:3b427392354d10975c1d0f4ee18aa5844640b512d5311ef32efd4dd7db106ede"},
+    {file = "matplotlib-3.10.0-cp313-cp313t-win_amd64.whl", hash = "sha256:5fd41b0ec7ee45cd960a8e71aea7c946a28a0b8a4dcee47d2856b2af051f334c"},
+    {file = "matplotlib-3.10.0-pp310-pypy310_pp73-macosx_10_15_x86_64.whl", hash = "sha256:81713dd0d103b379de4516b861d964b1d789a144103277769238c732229d7f03"},
+    {file = "matplotlib-3.10.0-pp310-pypy310_pp73-macosx_11_0_arm64.whl", hash = "sha256:359f87baedb1f836ce307f0e850d12bb5f1936f70d035561f90d41d305fdacea"},
+    {file = "matplotlib-3.10.0-pp310-pypy310_pp73-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:ae80dc3a4add4665cf2faa90138384a7ffe2a4e37c58d83e115b54287c4f06ef"},
+    {file = "matplotlib-3.10.0.tar.gz", hash = "sha256:b886d02a581b96704c9d1ffe55709e49b4d2d52709ccebc4be42db856e511278"},
 ]
 
 [package.dependencies]
@@ -1734,7 +1748,7 @@ pyparsing = ">=2.3.1"
 python-dateutil = ">=2.7"
 
 [package.extras]
-dev = ["meson-python (>=0.13.1)", "numpy (>=1.25)", "pybind11 (>=2.6)", "setuptools (>=64)", "setuptools_scm (>=7)"]
+dev = ["meson-python (>=0.13.1,<0.17.0)", "pybind11 (>=2.13.2,!=2.13.3)", "setuptools (>=64)", "setuptools_scm (>=7)"]
 
 [[package]]
 name = "matplotlib-inline"
@@ -1885,13 +1899,13 @@ pygments = ">2.12.0"
 
 [[package]]
 name = "mkdocs-material"
-version = "9.5.49"
+version = "9.6.4"
 description = "Documentation that simply works"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "mkdocs_material-9.5.49-py3-none-any.whl", hash = "sha256:c3c2d8176b18198435d3a3e119011922f3e11424074645c24019c2dcf08a360e"},
-    {file = "mkdocs_material-9.5.49.tar.gz", hash = "sha256:3671bb282b4f53a1c72e08adbe04d2481a98f85fed392530051f80ff94a9621d"},
+    {file = "mkdocs_material-9.6.4-py3-none-any.whl", hash = "sha256:414e8376551def6d644b8e6f77226022868532a792eb2c9accf52199009f568f"},
+    {file = "mkdocs_material-9.6.4.tar.gz", hash = "sha256:4d1d35e1c1d3e15294cb7fa5d02e0abaee70d408f75027dc7be6e30fb32e6867"},
 ]
 
 [package.dependencies]
@@ -1908,7 +1922,7 @@ regex = ">=2022.4"
 requests = ">=2.26,<3.0"
 
 [package.extras]
-git = ["mkdocs-git-committers-plugin-2 (>=1.1,<2.0)", "mkdocs-git-revision-date-localized-plugin (>=1.2.4,<2.0)"]
+git = ["mkdocs-git-committers-plugin-2 (>=1.1,<3)", "mkdocs-git-revision-date-localized-plugin (>=1.2.4,<2.0)"]
 imaging = ["cairosvg (>=2.6,<3.0)", "pillow (>=10.2,<11.0)"]
 recommended = ["mkdocs-minify-plugin (>=0.7,<1.0)", "mkdocs-redirects (>=1.2,<2.0)", "mkdocs-rss-plugin (>=1.6,<2.0)"]
 
@@ -1925,23 +1939,22 @@ files = [
 
 [[package]]
 name = "mkdocstrings"
-version = "0.27.0"
+version = "0.28.1"
 description = "Automatic documentation from sources, for MkDocs."
 optional = false
 python-versions = ">=3.9"
 files = [
-    {file = "mkdocstrings-0.27.0-py3-none-any.whl", hash = "sha256:6ceaa7ea830770959b55a16203ac63da24badd71325b96af950e59fd37366332"},
-    {file = "mkdocstrings-0.27.0.tar.gz", hash = "sha256:16adca6d6b0a1f9e0c07ff0b02ced8e16f228a9d65a37c063ec4c14d7b76a657"},
+    {file = "mkdocstrings-0.28.1-py3-none-any.whl", hash = "sha256:a5878ae5cd1e26f491ff084c1f9ab995687d52d39a5c558e9b7023d0e4e0b740"},
+    {file = "mkdocstrings-0.28.1.tar.gz", hash = "sha256:fb64576906771b7701e8e962fd90073650ff689e95eb86e86751a66d65ab4489"},
 ]
 
 [package.dependencies]
-click = ">=7.0"
 Jinja2 = ">=2.11.1"
 Markdown = ">=3.6"
 MarkupSafe = ">=1.1"
 mkdocs = ">=1.4"
-mkdocs-autorefs = ">=1.2"
-platformdirs = ">=2.2"
+mkdocs-autorefs = ">=1.3"
+mkdocs-get-deps = ">=0.2"
 pymdown-extensions = ">=6.3"
 
 [package.extras]
@@ -2598,13 +2611,13 @@ testing = ["pytest", "pytest-benchmark"]
 
 [[package]]
 name = "pre-commit"
-version = "4.0.1"
+version = "4.1.0"
 description = "A framework for managing and maintaining multi-language pre-commit hooks."
 optional = false
 python-versions = ">=3.9"
 files = [
-    {file = "pre_commit-4.0.1-py2.py3-none-any.whl", hash = "sha256:efde913840816312445dc98787724647c65473daefe420785f885e8ed9a06878"},
-    {file = "pre_commit-4.0.1.tar.gz", hash = "sha256:80905ac375958c0444c65e9cebebd948b3cdb518f335a091a670a89d652139d2"},
+    {file = "pre_commit-4.1.0-py2.py3-none-any.whl", hash = "sha256:d29e7cb346295bcc1cc75fc3e92e343495e3ea0196c9ec6ba53f49f10ab6ae7b"},
+    {file = "pre_commit-4.1.0.tar.gz", hash = "sha256:ae3f018575a588e30dfddfab9a05448bfbd6b73d78709617b5a2b853549716d4"},
 ]
 
 [package.dependencies]
@@ -2871,13 +2884,13 @@ files = [
 
 [[package]]
 name = "pydantic"
-version = "2.10.4"
+version = "2.10.6"
 description = "Data validation using Python type hints"
 optional = false
 python-versions = ">=3.8"
 files = [
-    {file = "pydantic-2.10.4-py3-none-any.whl", hash = "sha256:597e135ea68be3a37552fb524bc7d0d66dcf93d395acd93a00682f1efcb8ee3d"},
-    {file = "pydantic-2.10.4.tar.gz", hash = "sha256:82f12e9723da6de4fe2ba888b5971157b3be7ad914267dea8f05f82b28254f06"},
+    {file = "pydantic-2.10.6-py3-none-any.whl", hash = "sha256:427d664bf0b8a2b34ff5dd0f5a18df00591adcee7198fbd71981054cef37b584"},
+    {file = "pydantic-2.10.6.tar.gz", hash = "sha256:ca5daa827cce33de7a42be142548b0096bf05a7e7b365aebfa5f8eeec7128236"},
 ]
 
 [package.dependencies]
@@ -3069,13 +3082,13 @@ dev = ["argcomplete", "attrs (>=19.2)", "hypothesis (>=3.56)", "mock", "pygments
 
 [[package]]
 name = "pytest-asyncio"
-version = "0.25.2"
+version = "0.25.3"
 description = "Pytest support for asyncio"
 optional = false
 python-versions = ">=3.9"
 files = [
-    {file = "pytest_asyncio-0.25.2-py3-none-any.whl", hash = "sha256:0d0bb693f7b99da304a0634afc0a4b19e49d5e0de2d670f38dc4bfa5727c5075"},
-    {file = "pytest_asyncio-0.25.2.tar.gz", hash = "sha256:3f8ef9a98f45948ea91a0ed3dc4268b5326c0e7bce73892acc654df4262ad45f"},
+    {file = "pytest_asyncio-0.25.3-py3-none-any.whl", hash = "sha256:9e89518e0f9bd08928f97a3482fdc4e244df17529460bc038291ccaf8f85c7c3"},
+    {file = "pytest_asyncio-0.25.3.tar.gz", hash = "sha256:fc1da2cf9f125ada7e710b4ddad05518d4cee187ae9412e9ac9271003497f07a"},
 ]
 
 [package.dependencies]
@@ -4370,4 +4383,4 @@ propcache = ">=0.2.0"
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.11"
-content-hash = "0c314284e19f648aa24ce363e3efb6f23fcd378b1816932d01623cba359dbaed"
+content-hash = "9f04c27a16a385191dc91ac21012ea2a48b54d9e4380bcaba72f3106979b4219"

pyproject.toml

@@ -1,13 +1,13 @@
 [tool.poetry]
 name = "agentic_security"
-version = "0.4.3"
+version = "0.4.5"
 description = "Agentic LLM vulnerability scanner"
 authors = ["Alexander Miasoiedov <msoedov@gmail.com>"]
 maintainers = ["Alexander Miasoiedov <msoedov@gmail.com>"]
 repository = "https://github.com/msoedov/agentic_security"
 homepage = "https://github.com/msoedov/agentic_security"
 documentation = "https://github.com/msoedov/agentic_security/blob/main/README.md"
-license = "MIT"
+license = "Apache-2.0"
 readme = "Readme.md"
 keywords = [
     "LLM vulnerability scanner",
@@ -28,18 +28,18 @@ agentic_security = "agentic_security.__main__:main"
 
 [tool.poetry.dependencies]
 python = "^3.11"
-fastapi = "^0.115.6"
+fastapi = "^0.115.8"
 uvicorn = "^0.34.0"
 fire = "0.7.0"
 loguru = "^0.7.3"
 httpx = "^0.28.1"
 cache-to-disk = "^2.0.0"
 pandas = ">=1.4,<3.0"
-datasets = ">=1.14,<4.0"
+datasets = "^3.3.0"
 tabulate = ">=0.8.9,<0.10.0"
 colorama = "^0.4.4"
 matplotlib = "^3.9.2"
-pydantic = "2.10.4"
+pydantic = "2.10.6"
 scikit-optimize = "^0.10.2"
 scikit-learn = "1.6.1"
 numpy = ">=1.24.3,<3.0.0"
@@ -47,6 +47,7 @@ jinja2 = "^3.1.4"
 python-multipart = "^0.0.20"
 tomli = "^2.2.1"
 rich = "13.9.4"
+gTTS = "^2.5.4"
 # garak = { version = "*", optional = true }
 
 
@@ -54,19 +55,19 @@ rich = "13.9.4"
 # Pytest
 pytest = "^8.3.4"
 pytest-asyncio = "^0.25.2"
-inline-snapshot = ">=0.13.3,<0.19.0"
+inline-snapshot = ">=0.13.3,<0.21.0"
 pytest-httpx = "^0.35.0"
 pytest-mock = "^3.14.0"
 
 # Rest
-black = "^24.10.0"
+black = ">=24.10,<26.0"
 mypy = "^1.12.0"
 pre-commit = "^4.0.1"
-huggingface-hub = "^0.25.1"
+huggingface-hub = ">=0.25.1,<0.29.0"
 
 # Docs
 mkdocs = ">=1.4.2"
-mkdocs-material = ">=8.5.10"
+mkdocs-material = "^9.6.4"
 mkdocstrings = ">=0.26.1"
 mkdocs-jupyter = ">=0.25.1"
 

vercel.json

@@ -1,15 +0,0 @@
-{
-    "devCommand": "uvicorn agentic_security.app:app --host 0.0.0.0 --port 3000",
-    "builds": [
-        {
-            "src": "agentic_security/app.py",
-            "use": "@vercel/python"
-        }
-    ],
-    "routes": [
-        {
-            "src": "/(.*)",
-            "dest": "agentic_security/app.py"
-        }
-    ]
-}