watermarks-remover: A privacy-first agent skill and Python service for stripping AI provenance marks from text and files you own.
What problem it solves
AI-generated content can contain provenance marks that remain in text, metadata, or media files even when the content is otherwise ready for use. The project addresses this by inspecting and removing marks across multiple vendor ecosystems and file formats while keeping the agent workflow separate from the processing service.
Key features
Multi-vendor text cleaning — Removes deterministic and statistical provenance marks from text through a layered workflow. Layer A handles invisible Unicode and related characters, while Layer B supports agent rewrites and optional rewrite hooks.
Source: README
Broad file-format coverage — Inspects and cleans metadata and provenance surfaces across images, documents, media, and container formats. Unsupported or unrecognized formats are identified rather than automatically cleaned.
Source: README
Agent skill with HTTP service — Provides a markdown-only agent skill that calls a local stdlib Python service over HTTP. The service exposes inspection, detection, cleaning, watermark generation, and batch operations.
Source: README
Deterministic file hooks — Adds a PostToolUse hook for file-writing tools with check and clean modes. Detection reuses shared inspection logic, while cleaning uses the existing file-cleaning scripts without duplicating logic.
Source: README
Detection and generation endpoints — Separates watermark detection from cleaning and includes endpoints for generating watermarked text for benchmark evaluation and round-trip testing. Detector and generator integrations fail softly when they are unconfigured or unavailable.
Source: README
Quick start
Copied verbatim from the project's README.
python3 install_skill.py --skill remove-ai-marks --target claude-codeWho it may suit
Inference: may suit developers and teams that need to remove provenance marks from content they own while working with agent tools, local scripts, or an HTTP service. Inference: may also suit users who need deterministic checks or cleaning for files written by coding agents.
Topics to discuss
How should privacy-first provenance removal workflows define and communicate their boundaries for content ownership?
Which integrations and rewrite strategies provide the clearest balance between deterministic cleanup and model-driven text changes?
What additional tests, benchmarks, or deployment patterns would help communities adopt the service safely?
Written by ReadmeX AI from the project's README, source and public GitHub data; it may contain mistakes.
Read the original README https://github.com/guillaumemeyer/watermarks-remover#readme
ReadmeX opened this community for a trending project; it is not run by the project. Maintainers can claim it from the community page.
Comments
…