Implementation:Liu00222 Open Prompt Injection create app

Knowledge Sources	Open-Prompt-Injection
Domains	Prompt_Injection, LLM
Last Updated	2026-02-14 15:00 GMT

Overview

Concrete factory function for creating LLM application instances with optional defense mechanisms, provided by the OpenPromptInjection apps module.

Description

The create_app function instantiates an Application object that wraps a TargetTask and Model with an optional defense strategy. The Application is iterable (yields data samples) and queryable (processes prompts through the defense pipeline and model). It supports 12+ defense strategies including instructional, sandwich, delimiter-based, paraphrasing, retokenization, LLM-based, known-answer, perplexity-based, and response-based defenses.

Usage

Import this function after creating target_task and model objects to assemble them into a queryable application with the desired defense mechanism.

Code Reference

Source Location

• Repository: Open-Prompt-Injection
• File: OpenPromptInjection/apps/__init__.py
• Lines: L5-6

Signature

def create_app(task, model, defense='no'):
    """
    Factory function to create an Application.

    Args:
        task (TargetTask): Target task (must have for_injection() == False).
        model (Model): LLM instance with .query() method.
        defense (str): Defense strategy string (default 'no').
    Returns:
        Application: Queryable application instance.
    """
    return Application(task, model, defense)

Import

import OpenPromptInjection as PI
# or
from OpenPromptInjection import create_app

I/O Contract

Inputs

Name	Type	Required	Description
task	TargetTask	Yes	Target task (must have `for_injection() == False`)
model	Model	Yes	LLM instance with `.query()` method
defense	str	No	Defense strategy (default `'no'`). Options: `'no'`, `'instructional'`, `'sandwich'`, `'random_seq'`, `'delimiters'`, `'xml'`, `'paraphrasing'`, `'retokenization'`, `'llm-based'`, `'known-answer'`, `'ppl-<window>-<threshold>'`, `'response-based'`

Outputs

Name	Type	Description
app	Application	Iterable over `(data_prompt, label)` pairs with `.query(data_prompt) -> str` method

Usage Examples

Creating an Application with Defense

import OpenPromptInjection as PI
from OpenPromptInjection.utils import open_config

target_task = PI.create_task(open_config("configs/task_configs/sst2_config.json"), 100)
model = PI.create_model(open_config("configs/model_configs/gpt_config.json"))

# No defense
app_no_defense = PI.create_app(target_task, model, defense='no')

# With sandwich defense
app_sandwich = PI.create_app(target_task, model, defense='sandwich')

# Query the application
for data_prompt, label in app_no_defense:
    response = app_no_defense.query(data_prompt)
    print(response)
    break

Related Pages

Implements Principle

• Principle:Liu00222_Open_Prompt_Injection_Application_Assembly

Requires Environment

• Environment:Liu00222_Open_Prompt_Injection_Python_Dependencies