llm-secret-mcp-decryptor

research decryption tool, for use with llm-secret-mcp

GitHub Stars

3

User Rating

Not Rated

Forks

1

Issues

0

Views

1

Favorites

0

README
LLM-Secrets Decryption Tool

A standalone tool for decrypting files created by the LLM-Secrets system. This tool enables researchers to access and analyze the private thoughts that were encrypted during the LLM interaction process.

This is built to work with llm-secret-MCP

Features
  • Compatible with the encryption method used in the LLM-Secrets project
  • Supports both interactive and command-line modes
  • Provides options for displaying or saving decrypted content
  • Uses the same AES-256 encryption algorithm for seamless decryption
  • Configuration management via settings.json
Setup
Prerequisites
  • Python 3.8 or higher
  • cryptography package
Installation
  1. Install the required packages:
pip install cryptography
  1. The tool is ready to use with the included settings.json file.
Usage

The decryption tool can be used in several ways:

Using Saved Configuration

The easiest way to decrypt files is to use the configuration in settings.json, which already contains the encryption key:

python decrypt.py --config --file ../private/private_thought_20250406170040.enc
Interactive Mode

Run the tool without arguments for interactive mode:

python decrypt.py

The tool will:

  1. Prompt for the encryption key if not specified
  2. Show available encrypted files in the default folder
  3. Let you choose which file to decrypt
  4. Display the decrypted content
Command-Line Options
usage: decrypt.py [-h] [-c] [-k KEY] [-f FILE] [-o OUTPUT] [-d OUTPUT_DIR] [-q]

LLM-Secrets Decryption Tool

optional arguments:
  -h, --help            show this help message and exit
  -c, --config          Use settings from config file
  -k, --key KEY         Base64 encoded encryption key
  -f, --file FILE       Path to encrypted file
  -o, --output OUTPUT   Path to save decrypted output
  -d, --output-dir OUTPUT_DIR
                        Directory to save decrypted output
  -q, --quiet           Suppress console output
Examples
Decrypt and Display Content
python decrypt.py --config --file ../private/private_thought_20250406170040.enc
Decrypt and Save to File
python decrypt.py --config --file ../private/private_thought_20250406170040.enc --output ./my_decrypted_file.txt
Decrypt and Save to Default Output Directory
python decrypt.py --config --file ../private/private_thought_20250406170040.enc --output-dir outputs
Using a Different Key
python decrypt.py --key YOUR_BASE64_KEY --file ../private/private_thought_20250406170040.enc
Quiet Mode (No Console Output)
python decrypt.py --config --file ../private/private_thought_20250406170040.enc --output results.txt --quiet
Configuration File

The tool uses a settings.json file for configuration. The default file includes:

{
    "key": "MXsP1F5wOQk4PsDe/f/RiP6NwwMY+xuCX3qZtTxHnbk=",
    "default_folder": "../private",
    "output_folder": "outputs"
}
  • key: The base64-encoded encryption key
  • default_folder: Directory where encrypted files are stored
  • output_folder: Default directory for saving decrypted output

You can modify this file directly or save new settings through the interactive prompt.

Notes for Researchers

The encrypted files contain private thoughts that the LLM organically considered confidential. When analyzing these decrypted thoughts, you may gain insights into:

  1. What types of information LLMs naturally consider private
  2. How LLMs reason about privacy and confidentiality
  3. Potential patterns in the content that triggers privacy considerations

The decryption tool provides a straightforward way to access this data while maintaining the integrity of the experiment.

Author Information

79

Followers

145

Repositories

1

Gists

7

Total Contributions

Top Contributors

Threads