Lemon Juice (Alpha) 作成者: Diego

Lemon Juice holds a web page up to the flame. It reveals SOME hidden text
and prompt-injection payloads: the writing an AI assistant would read but you
can't see, so you can decide before handing the page to ChatGPT, Claude,
Gemini, or a browser agent.

Click the toolbar icon and it scans the active tab for:

• Invisible Unicode characters and bidi/"Trojan Source" control tricks
• ASCII smuggling via the Unicode Tags block
• Visually hidden CSS text (tiny fonts, zero opacity, off-screen positioning)
• Suspicious base64, percent-encoded, and hex-escaped blobs
• Instruction-like phrases that read like commands to an AI

Each finding is highlighted on the page and listed in the popup, ranked by
severity.

This is a detection aid, not a shield. It doesn't block, sanitize, or sit
between the page and your AI assistant; it gives you eyes on content that's
otherwise invisible. A clean scan means "nothing obvious found," not "this
page is safe." Lemon Juice makes no network requests and only inspects the
tab active when you click the icon (activeTab + scripting permissions, no
host permissions).

Source: https://github.com/DiegoFleitas/lemon-juice