If you find a security problem, please don't open a public issue for anything sensitive. Instead, open a private security advisory. For low-risk stuff a normal issue is fine.

I'll acknowledge it as soon as I can and keep you in the loop on the fix.

Worth knowing how the extension is built, because it rules out a lot of common concerns:

• It runs entirely in your browser. There are no network requests and nothing you type is sent anywhere.
• It only activates on chatgpt.com, chat.openai.com, claude.ai, and claude.com.
• The tokenizer (js-tiktoken) is bundled in the extension — no remote code is ever fetched or executed.

So the main things worth scrutinizing are the content-script permissions and the bundled tokenizer file. If something there looks off, I'd like to hear about it.