Every character, named and located
Scans for zero-width characters, non-standard spaces, bidirectional overrides, Unicode Tag-block characters and Cyrillic or Greek homoglyphs. Each finding is reported with its Unicode code point, official character name, byte offset and why it matters.
Invisible characters cause real problems independent of any watermark: they break exact-match search, corrupt database keys, defeat code review, and act as a smuggling channel for prompt injection.