guillaumemeyer/watermarks-remover ? reverse-engineered prompt

Build me a Python tool that helps clean AI generated text and files I own, so they no longer carry obvious provenance marks or weird hidden formatting.

I want it to inspect and clean text for invisible Unicode characters, strange spaces, bidi marks, and similar junk, then do a stronger rewrite pass for watermark style text when needed. It should also strip metadata and C2PA style provenance from common file types like PNG, JPEG, SVG, PDF, DOCX, ODT, HTML, and Markdown. For images, it should be able to inspect and clean them, and if possible report a SynthID style confidence score when an external checker is available.

Please make it usable from simple command line scripts, with clear inspect and clean commands, and keep the core code in plain Python so it works without extra setup. If there are optional tools like exiftool or c2patool, use them when present, but do not require them. Add good tests, sensible defaults, and clear output so I can tell what was found and what was removed.