The sanitize_prompt utility function in LLMWorkbook ensures that prompts passed to wrappers or integrators are clean and secure before being processed by an LLM.
This function removes unnecessary special characters, excessive whitespace, and potentially harmful content such as HTML/JavaScript injections. It supports various input types, including:
✔ Removes unnecessary special characters (\, *, _, {}, [], (), #, +, !, $).
✔ Eliminates potential script injections (<script> and javascript:).
✔ Normalizes excessive whitespace to ensure clean formatting.
✔ Supports multiple input formats (str, list, pandas.Series, numpy.ndarray).
✔ Returns sanitized content in the same format as input for easy integration.
def sanitize_prompt(prompt: Union[str, List[str], pd.Series, np.ndarray]) -> Union[str, List[str], pd.Series, np.ndarray]:
"""
A versatile prompt sanitization function that works with various input types.
Args:
prompt: Input that can be a string, list of strings, pandas Series, or numpy array
Returns:
Sanitized version of the input in the same format
"""
from llmworkbook import sanitize_prompt
prompt = " Tell me about AI <script>alert('XSS')</script> with [markdown] formatting! "
clean_prompt = sanitize_prompt(prompt)
print(clean_prompt)
Output:
"Tell me about AI alert('XSS') with markdown formatting!"
🚀 The function removes HTML script tags and normalizes whitespace.
prompt_list = [
"Tell me about {Python}",
" <script>alert('XSS')</script> ",
"Analyze the following data: [1, 2, 3, 4, 5]"
]
clean_list = sanitize_prompt(prompt_list)
print(clean_list)
Output:
['Tell me about Python', "alert('XSS')", 'Analyze the following data: 1, 2, 3, 4, 5']
✔ The function removes curly braces, HTML tags, and cleans formatting.
import pandas as pd
# Sample DataFrame
df = pd.DataFrame({
'user_id': [1, 2, 3],
'prompt': [
"Generate a *summary* of this article",
" <script>malicious code</script> ",
"What is the answer to [4+5]?"
]
})
# Sanitize the 'prompt' column
df['clean_prompt'] = sanitize_prompt(df['prompt'])
print(df)
Output:
user_id prompt clean_prompt
0 1 Generate a *summary* of this article Generate a summary of this article
1 2 <script>malicious code</script> malicious code
2 3 What is the answer to [4+5]? What is the answer to 45?
✔ Markdown special characters, HTML tags, and extra whitespace are removed.
import numpy as np
prompt_array = np.array([
"Calculate 2+2=?",
" What is the ```result```? ",
"<script>alert('XSS')</script>"
])
clean_array = sanitize_prompt(prompt_array)
print(clean_array)
Output:
['Calculate 22=?' 'What is the result?' 'alert(XSS)']
✔ Handles NumPy arrays efficiently with vectorized sanitization.
sanitize_prompt?| Feature | Benefit | |———|———| | ✅ Security | Removes potential JavaScript injections, preventing XSS risks. | | ✅ Consistency | Ensures prompts are properly formatted for better LLM output. | | ✅ Flexibility | Works with strings, lists, pandas Series, and NumPy arrays. | | ✅ Easy Integration | Use before passing prompts to wrappers or integrators. |
Sanitized prompts can be directly passed to LLM wrappers or integrators to ensure safe and clean input.
from llmworkbook import sanitize_prompt, WrapDataFrame
df = pd.DataFrame({
"prompt": ["Tell me about *Python*.", "<script>alert('XSS')</script>", "What is 4+5?"]
})
# Clean the prompts
df["clean_prompt"] = sanitize_prompt(df["prompt"])
# Pass cleaned prompts to the wrapper
wrapper = WrapDataFrame(df, prompt_column="clean_prompt")
wrapped_df = wrapper.wrap()
print(wrapped_df)
✔ Ensures that LLM receives sanitized input, improving response quality and security.
Use sanitize_prompt before passing data to LLM integrators for cleaner, more secure results!