watermarks-remover
An agent skill and Python service that strips multi-vendor AI provenance marks, such as invisible watermarking and C2PA or EXIF metadata, from text and files you own.
watermarks-remover is an agent skill paired with a stdlib Python service that strips multi-vendor AI provenance marks from text and files, for privacy and hygiene on content the user owns. The skill itself ships no code: it is a thin client that drives the service over HTTP, so the agent host needs no Python installed.
It handles three layers: deterministic removal of invisible Unicode, exotic spaces, bidirectional marks and tag characters; an agent-assisted rewrite, with an optional scripted hook, for statistical token-sampling text watermarks; and stripping of C2PA, EXIF, XMP and document-property provenance data from a wide range of image, document, audio and video file formats. It targets watermarking schemes from vendors and ecosystems including Claude, Gemini's SynthID-Text, OpenAI's provenance surfaces, and open-LLM green-list and keyed-Gumbel marks.
It installs as a skill for Claude Code, Cowork, Cursor and other supported hosts through a Python 3.10+ installer script with no extra dependencies, and starts the companion HTTP service separately. It is licensed under the MIT license.