watermarks-remover

watermarks-remover is an open-source MIT toolkit for authorized content hygiene. Layer A removes invisible Unicode, bidi tricks, and exotic spaces. Layer B is best-effort rewrite against sampling-style text watermarks. File cleaners strip C2PA, EXIF, XMP, and document props across images, PDF, Office, HTML, Markdown, and common media. Coverage is class-level for Claude, Gemini SynthID-Text, OpenAI provenance surfaces, and open-LLM green-list style marks. The agent skill is markdown-only and drives a stdlib Python HTTP service, so the host does not need to vendor the scrubbers.

Founders ship drafts, assets, and decks that accumulate invisible Unicode, C2PA bags, and model metadata they never asked for. You need a hygiene layer that agents can call, not a one-off hex edit. This turns provenance cleanup into inspect and clean endpoints with a real skill path.

Start the local service with make serve or python3 service/scripts/server.py on loopback. Link skills/remove-ai-marks into your agent skills folder and call /remove-ai-marks or ask to strip AI marks. CLI path is inspect_file.py then clean_file.py on a draft or image. HTTP exposes /health, /inspect, /detect, /clean, and batch variants with base64 payloads and OpenAPI. Optional Docker adds heavier backends. Prefer a non-origin model if you run Layer B rewrite so you do not re-stamp the text.

This is not another coding harness and not anti-slop lint for TypeScript. It is a multi-format provenance stripper with a skill plus HTTP API, honest best-effort wording on statistical marks, and format-safe routing so you do not corrupt Office or binary files by treating them as UTF-8.

• API / SDK surface

• no-ai-slop ↗
• Hallmark ↗