A library to analyze text that contains special unicode characters that are unnoticeable by naked eye but are interpretable by LLMs.
Features:
- ⭐️ hush - encode input string into hidden text with zero-width characters
- ⭐️ dehush - reverse the process and discover the hidden text
- ⭐️ sanitize - sanitize the text from hidden text characters
- Support dehush of flag emojis, utf-8 and other multilingual charsets, while the hidden characters are type of flag emoji with custom payload.
example input: "clear text, flag: 🇨🇦, utf-8: א, hushed payload: " ->
example output: "clear text, flag: 🇨🇦, utf-8: א, hushed payload: shhh!"
Demo:
Usage:
via NPX:
encode:
npx husher hush="hello world"
npx husher sanitize="<encoded string>"
via npm/yarn:
yarn install husher
npm install husher
interactive demonstration tool:
You can also play with the online tool which performs hush/dehush operations on client-side.
Credits:
The credit for uncovering the impacts of this method goes to
@goodside. Make sure to follow up.
Scaffolded with 🏗 TS-scaffold