</>MCP Agents Market
Skill

watermarks-remover

by guillaumemeyer16.1kPythonUpdated 2026-08-19

Strip multi-vendor AI provenance marks: Unicode text hygiene, statistical rewrite hooks, and C2PA/metadata from PNG/JPEG/SVG/PDF/DOCX/HTML/MD

Claude CodeCursorCowork

Watermarks-remover is an agent skill that removes multi-vendor AI provenance marks from text and files. It provides Layer A (deterministic Unicode/metadata stripping) and Layer B (statistical text watermark rewriting) cleaning, plus C2PA and metadata removal from images, documents, and containers. The skill works as a thin HTTP client that calls a Python service, so the agent host needs no dependencies. Developers install it into Claude Code, Cursor, or Cowork to automatically strip watermarks from AI-generated content they own, for privacy and hygiene purposes.

Key Features

Layer A: strips invisible Unicode, bidi marks, exotic spaces, and tag characters from text using deterministic scripts
Layer B: best-effort statistical watermark removal via paraphrase/MLM rewrite strategies targeting Claude, Gemini SynthID, OpenAI, and open-LLM marks
File metadata cleaning: removes C2PA, EXIF, XMP, and document properties from PNG, JPEG, WebP, AVIF, HEIC, SVG, PDF, DOCX, XLSX, PPTX, EPUB, ODT, HTML, Markdown, MP4, WAV, MP3, FLAC
PostToolUse hook for automatic cleaning: deterministically strips marks from files the agent writes, no model cooperation required
HTTP service API: /inspect, /clean, /detect, /watermark endpoints with batch operations and OpenAPI spec
Optional pixel watermark removal via external CtrlRegen and MarkDiffusion backends for SynthID-class, StegaStamp, Tree-Ring image marks
Detection-guided iterative rewriting: stops as soon as MarkLLM, keyed-Gumbel, or stylometry evaluators pass the cleaned output
Multi-client install script: one installer covers Claude Code (personal/project), Cowork bundles, and Cursor with automatic skill validation

Use Cases

  • 01Privacy hygiene: strip AI provenance marks from content you own before publishing or archiving
  • 02Research verification: benchmark how effectively Layer B rewrites clear statistical watermarks using MarkLLM/SynthID harnesses
  • 03Pre-commit gating: automatically detect and optionally clean AI marks in staged files before they enter version control
  • 04File metadata audit: recursively scan directories or websites for C2PA, EXIF, XMP, and AI markers with SARIF export for CI
  • 05Multi-format cleaning: batch-process images, PDFs, Office documents, EPUB, HTML, and Markdown to remove generator metadata and provenance signals
  • 06Automatic cleaning in agent workflows: use the PostToolUse hook so every file the agent writes is deterministically stripped without relying on model instructions

Related Skills

View more

superpowers

An agentic skills framework & software development methodology that works.

Skill294k

skills

Skills for Real Engineers. Straight from my .agents directory.

Skill273.9k

skills

Public repository for Agent Skills

Skill179.3k

ponytail

Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.

Skill150.6k

watermarks-remover — FAQ

What is watermarks-remover?+

Watermarks-remover is an agent skill and Python service that strips multi-vendor AI provenance marks (invisible Unicode, statistical text watermarks, C2PA/EXIF/XMP metadata) from text and files you own. It works in Claude Code, Cursor, Cowork, and as an HTTP API.

How do I install it in Claude Code?+

Run 'python3 install_skill.py --skill remove-ai-marks --target claude-code' or use the plugin marketplace with '/plugin marketplace add guillaumemeyer/watermarks-remover' then '/plugin install watermarks-remover@watermarks-remover'. You must also start the service with 'make serve' (or 'python3 service/scripts/server.py').

Which AI clients does it work with?+

The skill works in Claude Code (personal and project installs), Cowork sessions, claude.ai uploads, Cursor, and Grok Build. The HTTP service can be called from any web app or agent that supports HTTP.

Does it require API keys or external dependencies?+

The core Unicode and metadata cleaners need only Python 3.10+ stdlib. Layer B text rewriting requires a configured LLM backend (Ollama or OpenAI-compatible API, which needs WATERMARKS_REWRITE_API_KEY). Optional tools (c2patool, exiftool, qpdf) enhance cleaning; optional backends (CtrlRegen, MarkLLM, MarkDiffusion, SynthID scorer) require local checkouts.

Is it free?+

Yes, the core skill and service are MIT-licensed and free to use. Layer B rewriting costs depend on your chosen LLM backend (local Ollama is free; cloud APIs charge per token). Optional pixel-removal and verification harnesses are external projects with their own licenses.

Can it guarantee vendor detectors will fail?+

No. Layer A removals (Unicode, metadata) are verifiable. Layer B text rewriting is best-effort and cannot certify that a vendor detector (Claude, Google SynthID, OpenAI) will fail, because vendors do not publish their detection APIs or keys. MarkLLM and keyed-Gumbel detectors are same-config verification harnesses, not oracles.

How do I install watermarks-remover?+

Open the source repository on GitHub and follow its README. watermarks-remover is a skill — MCP Agents Market links you directly to the official repo.

Is watermarks-remover free?+

watermarks-remover is an open-source project hosted on GitHub. Check the repository for its license and any usage requirements.

Related searches

Productivity & Workflow SkillsSecurity SkillsDeveloper Tools SkillsBest MCP ServersBest PluginsBest AgentsTop rated listingsClaude Code toolsBrowse all categories